Mine Planning and Equipment

MINE PLANNING AND EQUIPMENT SELECTION – MPES 2010

Edited by: E Topal and M Kuruppu

1 - 3 DECEMBER 2010 FREMANTLE, WESTERN AUSTRALIA

The Australasian Institute of Mining and Metallurgy Publication Series No 11/2010

Published by: The Australasian Institute of Mining and Metallurgy Level 3, 15 - 31 Pelham Street, Carlton Victoria 3053, Australia

© The Australasian Institute of Mining and Metallurgy 2010 All papers published in this volume were refereed prior to publication. The Institute is not responsible as a body for the facts and opinions advanced in any of its publications.

ISBN 978 1 921522 35 2

Desktop published by: Kylie McShane and Olivia Tet Fong The Australasian Institute of Mining and Metallurgy Compiled on CD ROM by: Visual Image Processing Pty Ltd PO Box 3180 Doncaster East VIC 3109

CONTENTS Keynote Address Open Pit Production Scheduling Optimisation – Why the Mining Industry Needs a New Phase Design Approach

K Dagdelen

3

Steel and Iron Ore Scenario – An Indian Perspective

V S Rao

5

Optimisation Challenges in Underground Mine Planning and Design

H Rubinstein, M Brazil, P Grossman, D Lee, D Thomas and N Wormald

11

Economics of Mine Planning and Equipment Selection

I C Runge

13

More than Coal Mining

J Tourek

23

The Next Challenge in Optimising Mining Operations

J Whittle

25

Advances in Surface Mine Planning Design and Computer Application A Simulation Approach for the Comparison of In-Pit Crushing and Conveying and Truck-Shovel Mining Methods

R Bearman and S Munro

33

Production Capacity Optimisation in the Presence of Grade Uncertainty

T Elkington and F Grobler

43

Robust versus Flexible Open Pit Mine Design

B Groeneveld, E Topal and B Leenders

55

The Role of Cost Estimating in Mine Planning and Equipment Selection

J B Leinart and O L Schumacher

69

Ultimate Pit Limit of Iron Ore Mines using Maximum Flow Algorithms

M Osanloo, M Rahmanpour and A Sadri

81

A New Model to Improve Ore Grade Reconciliation Between the Exploration Model and the Mine

A Parhizkar, M Ataee, P Moarefvand, V Rasouli and A H Bangian

89

A New Genetic Algorithm for Open Pit Design – The Two-Dimensional Case

J Saavedra-Rosas

95

Enterprise Optimisation

G Whittle

105

Advances in Underground Mine Planning Design and Computer Application Optimisation Improvements in Whittle Using Stope Optimisation Software

S Keane

121

A Review of the Theory and Application of Multi-Criteria Decision Analysis Techniques in Mine Planning

C Musingwini

129

An Investigation to Integrate Optimum Long-Term Planning with Short Planning in Underground Mine Production Scheduling

M Nehring, E Topal, M Kizil and P Knights

141

Application of Genetic Algorithms for Reliability Assessment of Two Mine Hoisting Systems

N Vayenas, X Wu and S Peng

155

Blast Vibration Monitoring and Elastic Wave Reflection Models to Assess Blast-Induced Damage to Mine Infrastructure – An Underground Case Study

K G Fleetwood and E Villaescusa

165

Role of Physico-Mechanical Properties in Cutting Performance of Diamond Wire Saw in Marble Quarrying Operation

S C Jain and S S Rathore

179

Never Touch a Running System? Longwall Cutting Sequences and their Potential

K Nienhaus and S Hetzel

191

Developing Algorithms and a Computer Program to Predict the Peak Particle Velocity After Blasting in Open Pit Mines and Quarries

 Topal, B Elevli and K Erarslan

197

Numerical Modelling of Monorail Support Requirements in Decline Development

B Besa, M Kuruppu and E K Chanda

209

Case Study – The Mogiana Quarry Reclamation

A Curi and O Quaglio

225

Study on Utilisation of Flyash for Barrier Layer/Buffer Materials for Radioactive Waste Disposal

T Sasaoka, H Shimada and K Matsui

233

Fundamental Study of Acid Drainage Control using Flyash

H Shimada, T Sasaoka, K Matsui, G J Kusuma, J Oya, H Takamoto, S Kramadibrata and B Sulistianto

247

Simulation for Dependable Mining Automation

N Hillier and J Ryde

257

Coal Bed Boundary Detection using Infrared Technology for Longwall Shearer Automation

K Nienhaus, F Mavroudis and M Warcholik

267

Advances in Continuous Miner Automation

J C Ralston, M T Dunn, C Hargrave and D C Reid

275

Path Planning and Position Control for Autonomous Loading Operation by Wheel Loader

S Sarata and N Koyachi

293

Drilling Blasting and Excavation Engineering

Environmental Issues in Surface and Underground Mining

Mine Automation

Mine Equipment Selection, Utilisation and Maintenance A Non-Linear Programming Model for Open Pit Mine Equipment Selection

A Aghajani Bazzazi, M Osanloo and B Karimi

303

Equipment Selection – What is the Issue? What About the Application of Simulation Packages?

M Campbell

317

Improvement of Mechanical Tool Performance by Waterjet Assistance

R Ciccu and B Grosso

323

Optimum Equipment Management Through Life Cycle Costing

B Hall

333

Precision Surface Mining – Choosing the Best Tool for the Job

J Hutchins

341

Software Tools for Open Pit Trucking Studies

S Law

355

Infrared Imager as a Tool for Monitoring of Power Shovels, Gearcase Bearings and Electric Motors

V Sergeev, M Zhuravlev and R Y Poderni

365

The EMESRT Approach to Equipment Design – Bringing End Users into the Picture

G von Horsten

369

Ranking of Belt Conveyor Systems for Prefeasibility Study

E Y Baafi and D A Bedward

373

BATMIN – Business Analysis Tools for Mining

D Heugh

379

Preliminary Study on the Improvement of Afforestation Techniques Applied to Aggregate Mines in Japan

Y Sakai, Y Sato, K Matsui, H Shimada and T Sasaoka

399

Reopening and Production Planning of Kırdar – Yoncalı Quarry, Turkey

 Topal, B Elevli, H Akçakoca, K Oul and M M Güleç

411

Assessment of Occupational Exposure to Whole-Body Vibration for Drivers of Mining Equipment

V Dentoni, G Massacci and L Piras

419

‘She’ll be Right Mate’ – Culture and Safety

P Milnes, N S Melkoumian, D Mather, T Milnes, A Stewart and L Tan

427

Using Haulage Accidents and Incident Reports to Identify the Impact of Substandard Haul Road Design on Operational Safety

R J Thompson

439

Prediction of Near-Field Rock Damage Due to Production Blasting in Limestone Quarries, Japan

S Wahyudi, H Shimada, T Sasaoka, S Kubota, Y Ogata and K Matsui

453

Mine Feasibility Case Studies

Mine Health, Safety and Environment

Mine Planning and Block Modelling Hierarchical Clustering Algorithm for Block Aggregation in Open Pit Mines

H Askari-Nasab, M Tabesh, M M Badiozamani and H Eivazy

469

Operative Expert Estimation of Innovative Policy of Mining Enterprise Technological Development using Sustainable Development Principles

S Zh Galiyev, A A Boyandinova and K K Zhussupov

481

Investigating the Sensitivity of Different Estimation Techniques over Block Dimensions

A Hekmat and M Osanloo

487

A New Effective Labour-Planning Model for Mines

L D Meyer

499

Mass Mining using Room and Pillar Methods at Estonia Oil Shale Mine

R Freeman and O Nikitin

515

A Tool for Benchmark Study of Roadway Development Performances

G Gibson and E Y Baafi

525

Planning the Kotre Basantpur-Pachmo Coking Coal Mine

M D O’Brien, R Srikanth, A L Vidale and G M Springbett

533

Mining the LG6 Chromitite Layer – A Comparison of Two Approaches

S M Rupprecht

547

Computer Simulation of Roadway Development to Support Longwall Mining

E Y Baafi and I Porter

561

Information and Methodical Base of Simulation Modelling of Excavator-Truck-Conveyor Complex Operation

S Zh Galiyev, A A Boyandinova and Zh A Adilkhanova

567

Simulation and Animation Model for the Millerton Coal Mine (New Zealand)

T O’Connell and J R Sturgul

577

Performance Evaluation of Machinery used in Dimensional Stone Mining and Processing

S S Rathore and S C Jain

581

A Comparison of Rebar and Spiral Bolts using Laboratory Pull-Out Tests

S S Kang, D K Je, A Hirata and D Kim

591

Determination of Fracture Toughness of Anisotropic Rocks Under Water Vapour Pressure by Semi-Circular Bend Test

M Kuruppu, Y Obara and M Kataoka

599

Mining Methods and Planning

Mining System Monitoring and Simulation

Rock Mechanics, Geotechnical Applications and Mine Environment

Performance of the Coal Mine Openings and Pillars in Anisotropic Stress Field

Ö Yılmaz and T Ünlü

611

Fundamental Study on Fracturing Grouting for Shaft Lining Reinforcement at Deep Alluvium Seam in Eastern China

Q Yu, H Shimada, T Sasaoka and K Matsui

623

The Application of Fuzzy Analytical Hierarchy Processing to Define Optimal Post Mining Land use for Pit Area to Recognise Reclamation Costs in Open Pit Mining

A H Bangian, M Ataei, A Sayadi and A Gholinejad

637

The Effect of the Australian Carbon Trading Scheme on a Large Scale Open Pit Mining Operation

E Chanda, F Ball, J Dunstan, H Maier, P Mumford and C Shaw

653

Energy Conservation in the Indian Mining Industry – An Overview

G K Pradhan

669

System Optimisation of Design and Scheduling of Open Mining Operations

B Bekmurzayev, D Bukeikhanov and Y Zulkarnayev

689

Mining Environmental Impacts Evaluation in a Karstic Zone using Geographical Information Systems

A Curi and V C Garcia

695

Defining Optimal Post-Mining Land use for Different Sections of Mined Land Through Different Multi-Attribute Decision-Making Techniques in Open Pit Mining

A Gholinejad and A H Bangian

703

Application of Multi-Slicing Highwall Mining System with Stowing at Surface Coal Mine in Thailand

A Hamanaka, T Sasaoka, H Shimada, K Matsui, H Takamoto, P Meechumna and P Laowattanabandit

719

Adaptation of Hydraulic Shovels for Arctic Temperature of Yakutia Region

R Y Poderni and H R Koelsh

729

An Innovative Scaled Model to Simulate the Perimeter Hole Blasting in Tunnel

H-S Yang, H-D Jang, W-B Kim and P Rai

731

Systematic Monitoring of Operation of Mining Companies and Branches

 Zharmenov, S Galiev,  Lisenkov and S Zhumabekova

737

Role of Ore Preparation During Chrysotile Manufacturing

K K Zhusupov and S Y Punenkov

743

Sustainable Development of Mining Resources

Poster Papers

Author Index

749

Open Pit Production Scheduling Optimisation – Why the Mining Industry Needs a New Phase Design Approach K Dagdelen1 ABSTRACT The long-term production schedule of open pit mines that maximises the net present value (NPV) of a given project relies on the proper design of phases (pushbacks) within the ultimate pit limits. There is not a phase design algorithm applied in the mining industry today that gives yearly production schedules that are optimum. The majority of phase design algorithms currently in use by the mining industry aims at grouping the blocks that fall within the ultimate pit limits into phases such that the average value of the blocks within each phase decrease as the size of the phases increase. Current phase design methods are typically based on a pit limit optimisation algorithm, originally introduced by Lerchs and Grossmann (LG) in 1965. To obtain multiple pit shells to be used for the phase design, the LG pit limit algorithm successively applied to economic block models by changing the price of the commodity or some other factor that properly parameterises the size of the pit. These modified economic block models are developed by predetermining the destination for each block in the model based on a breakeven analysis and highest economic outcome between alternative destinations. Phase designs determined by these existing algorithms may be suboptimal if a given project utilises multiple processing facilities for metal recovery and particularly if these facilities have restrictions (constraints) that require blending of attributes in the feed. To overcome these problems in obtaining phases that optimise overall production schedules, a new algorithm is developed at the Colorado School of Mines. The algorithm determines the phases of an open pit mine by taking into account multiple process options for the blocks and period by period blending requirements. This algorithm has been implemented within a software program (the software) and is being prototyped successfully in the design of improved pushbacks at a copper mine in South America and on a gold mine in Nevada. This presentation will demonstrate the shortcomings of the traditional phase design methodologies based on comparison of results coming from the new approach versus the traditional.

1. Professor and Head, Mining Engineering Department, Colorado School of Mines, Colorado 80401-1843, USA. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

3

Steel and Iron Ore Scenario – An Indian Perspective V S Rao1 ABSTRACT India is on fast economic growth. Even in the turbulent years of global crisis, it could achieve a growth of over eight per cent. In recent past, Indian Government has been contemplating a faster growth in all the sectors of economy so as to make the nation an economic power in the world. Steel is a versatile material. It has its role in most of the major industrial activities, such as infrastructure, automobiles, urbanisation, consumer durables and other. In other words, steel is a part and parcel of industrial expansion, urban and rural development. It is rightly admitted that steel acts as pillar of any development model. Volume of steel consumption is the barometer which measures economic progress and/or industrial status of any country. India and China are the two countries in Asia which have the significant impact on the global steel industry. Of course, China is much ahead of India in consumption level of steel.

INTRODUCTION In 2005, India released its National Steel Policy, highlighting the steps to realise target steel production and consumption in different terminal years. By 2020, it has projected a target of 110 Mt. But this projection gets changing with the changing dynamics influencing the speed of industrialisation. I thus chose the above subject – steel and iron ore scenario with Indian perspective in the mirror of global steel production scene. It is an endeavour to present a realistic picture of India iron and steel industry and its strategy to facilitate the expansion plan of Indian steel industry by ensuring the availability of the principal raw material, ie iron ore. TABLE 1 Crude steel production – country-wise. Country

2004

2005

2006

2007

2008

2009

2010*

China

280.5

355.8

422.7

489.2

502

567.8

323

Japan

112.7

112.5

116.2

120.2

118.7

87.5

54.6

USA

99.7

94.9

98.6

9782

91.5

Russia

65.6

66.1

70.8

72.4

68.5

India

32.6

45.8

49.5

53.1

55.1

South Korea

47.5

47.8

48.5

51.5

53.5

Germany

46.4

44.5

47.2

48.6

45.8

Ukraine

38.7

38.6

40.9

42.8

37.1

Brazil

32.9

31.6

30.9

33.8

33.7

Italy

28.6

29.3

31.6

31.5

30.5

Rest of the World

283.7

279.6

294.3

302.9

293.3

World Total

1068.9

1146.5

1250.2

1344.2

1329.7

7.4

7.8

7.9

7.9

7.6

Australia

41 32.6 62.8

32.5

1224

706

* First six months January - June.

1. Director, VBC Industries Ltd, Flat 302, Whitehouse, Road No 13, Banjara Hills, Hyderabad, Andhra Pradesh 500034, India. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

5

V S RAO

TABLE 2 World iron ore production (country-wise). Country

2004

2005

2006

2007

2008

2009

145.7

200.3

276

332

345

395

Brazil

270.5

280

318

355

355

380

Australia

234.7

262

275

299

342

370

India

120.6

140

181

206

220

220

Russia

97

97

102

105

100

85

Ukraine

65.5

69

74

78

73

56

USA

54.7

54.3

53

52

54

26

World total

1184

1320

1488

1633

1693

1588

China

*

* China’s production reduced to 40 per cent of total production to equate its grade with the global average grade.

TABLE 3 World seaborne iron ore trade. Country

2004

2005

2006

2007

2008

World Seaborne Iron Ore Trade

587

658

720

781

833

Major importing countries and their volume: China

208

275

326

383

444

Japan

135

132

134

139

140

South Korea

43

44

43

44

49

Major exporting countries and their volume: Australia

210

240

247

266

310

Brazil

201

224

243

269

282

India

63

89

93

104

106

TABLE 4 Indian iron ore production – state-wise. State

2007 - 2008

2008 - 2009

2009 - 2010(P)

Orissa

68.5

74.13

80.28

Karnataka

45.6

45.9

42

Chhattissgarh

30.6

30.09

24.78

Goa

29.2

32.3

30.4

Jharkhand

20.9

21.21

22.28

Others

11.40+

11.76

08.26+

Total

213.25

215.44

220

STEEL Global steel The economic crisis of 2007 badly affected the steel industry throughout the world. For two consecutive years, 2008 and 2009, the global steel production dipped to a substantial low. After attaining a peak production of 1344 Mt in 2007, it scaled down to 1329 Mt in 2008 and 1224 Mt in 2009. 2010 appears to be favourable and may bring good days for the steel industry. In the first six months of 2010, ie from January to June, the global production achieved a level of 706 Mt, a 27.9 per cent MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

6

STEEL AND IRON ORE SCENARIO – AN INDIAN PERSPECTIVE

TABLE 5 Statewise and gradewise iron ore production (2009 - 2010). State

Lumps

Fines

Total

Orissa

41.66

38.62

80.28

Karnataka

15.58

26.42

42

Chhattissgarh

10.57

14.21

24.78

Goa

7.17

23.23

30.4

Jharkhand

10.35

11.93

22.28

Others

4.11

3.89

8.00+

Total

90.00+

120.00+

220

TABLE 6 Indian production – grade-wise. Particulars

2005 - 2006

2006 - 2007

2007 - 2008

2008 - 2009

2009 - 2010(P)

Lump

68.3

81.28

95.96

95.57

90.00+

Fines

93.3

98.69

116.67

119.22

120.40+

Concentrate

3.6

0.95

0.61

165.23

180.92

213.25

Total

0.60 +

0.6

215.44

220

TABLE 7 Public versus private sector iron ore production. Particulars

2001 - 2002

2002 - 2003

2003 - 2004

2004 - 2005

2005 - 2006

2006 - 2007

2007 - 2008

Public sector

45.1

49.7

57.5

57

61.2

60.4

68.24

Private sector

41.1

49.4

65.3

88.9

104

120.5

145

Total

86.2

99.1

122.8

145.9

165.2

180.9

213.25

2005 - 2006

2006 - 2007

2007 - 2008

TABLE 8 Captive versus non-captive iron ore production. Particulars

2001 - 2002

2002 - 2003

2003 - 2004

2004 - 2005

Captive

28.0

29.9

33.5

35.2

35.1

36.5

38.38

Non-captive

58.2

69.1

89.3

110.7

130.1

144.4

174.86

Total

86.2

99.1

122.8

145.9

165.2

180.9

213.25

TABLE 9 Iron ore consumption in India.

Consumption

2004 - 2005

2005 - 2006

2006 - 2007

2007 - 2008

2008 - 2009

CAGR %

48.15

52.52

66.9

85

90

0.17

higher from the same period of 2009. All the regions of the world showed increased production during the first half of 2009. Again, comparing with the first six months of 2007, most of the world has not recovered to the precrisis level except Asia and the Middle East. Crude steel production in Europe, CIS, United States and Canada is still more than 15 per cent below the 2007 level. However, with the 706 Mt production of the first six months of 2010, the annualised production for the whole year comes to about little over 1400 Mt, ie 15 per cent increase over the previous year. It is a good sign and the world can look forward for a conspicuous growth trend for the future of steel industry.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

7

V S RAO

TABLE 10 Country-wise export. Importing Country of Indian Iron Ore

2005 - 2006

2006 - 2007

2007 - 2008

2008 - 2009

China

74.1

80.2

91.98

97.8

Japan

10.3

8.6

7.7

5.4

South Korea

1.3

1.9

1.8

0.99

Taiwan

0.1

-

-

-

Europe

2.1

2.1

1.6

0.76

Others

1.2

1

1.2

0.84

Total

89.3

93.8

104.3

105.9

TABLE 11 Grade-wise export. Type of Ore

2004 - 2005

2005 - 2006

2006 - 2007

2007 - 2008

Lump ore

13.54

14.28

15.3

14.87

+0.64

7.77

5.47

3.06

4.86

62 - 64%

2.99

4.02

5.09

4.47

-0.62

2.78

4.79

7.15

5.54

Fine ore

64.6

74.99

78.49

89.4

+0.64

12.2

12.6

10.46

7.29

62 - 64%

19.33

20.99

26.01

20.82

-0.62

33.07

41.4

42.02

61.29

Total

78.14

89.27

93.79

104.27

+0.64

19.97

18.07

13.52

12.15

62 - 64%

22.32

25.01

31.1

25.29

-0.62

35.85

46.19

49.17

66.83

Note: Data pertaining to the year 2009 - 2010 is not available.

Indian steel The Indian steel industry maintained a phenomenal rise even during the time of global economic crunch when global steel production fell in most of the regions of the world, except China and the Middle East. In fact the global crisis did affect India in making its forward move to its expansion program. Many of its Green field projects and brown field expansion programs slowed down. Moreover, because of special domestic economic condition, which is expanding due to huge demand from infrastructure sector, Indian steel sector is likely to be in a growth path. If the infrastructure spending and development in overall industry is sustained, the growth is likely to be even higher. Moreover, about 194 MoUs have been signed with various State Governments covering a new capacity of 243 Mt to be added in India. A substantial brown field expansion of existing plant of SAIL, TATA Steel, Essar Steel, JSW and Ispat has also been projected. In a nutshell, India has made strategic plan to achieve a steel production to a level of over 200 Mt by 2020.

IRON ORE Global Iron ore is the principal raw material for production of steel, pig iron and sponge iron. The world iron ore reserve and reserve base are estimated by USGS to be at 150 and 350 billion tons respectively.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

8

STEEL AND IRON ORE SCENARIO – AN INDIAN PERSPECTIVE

The global iron ore production dipped by 6.2 per cent in 2009 to 1588 Mt. It was the first time in the preceding ten years that production was lowered. Vale, Rio Tinto and BHP Billiton no doubt would continue to dominate with production of about 620 Mt (45 per cent) of the world total.

Sea borne trade Iron ore trade has reached a new record level of 950 (P) Mt in 2009. China emerged as the largest importer with 628 Mt, followed by Japan and South Korea. Australia and Brazil, as usual, collectively contribute more than two thirds of the world’s iron ore trade. Major increase in demand has been only from China whereas demand from rest of the world has been stagnant.

Future outlook The three large companies, Vale, Rio Tinto and BHP Billiton, control about 70 per cent of the world’s iron ore trade. New capacities have been added over the last two to three years. The iron ore market presently appears to be in an over supply situation. But the way production of steel picked up from the beginning of 2010, the situation may reverse or demand-supply equation may find balanced for some time. More capacities are required to be added in the next five years time to avoid short supply position in decades ahead.

Indian scene Presently, India is producing about 215 - 220 Mt, of which about 50 per cent is exported. Projected steel production by 2020 is about 180 Mt which requires about 288 Mt of iron ore. Then, a minimum export of 100 Mt can be a possibility for trading. Thus the requirement of iron ore by 2020 would be about 400 Mt. Further growth would be matching with the growth pattern of steel production and international situation on trade.

Resource availability Haematite and magnetite are the two exploitable iron ores. Indian resource accounts for 13 per cent of global resource. Haematite resource is estimated to be at 14.6 billion tons which is 58 per cent of the total resource of the country. Magnetite ore is estimated at 10.6 billion tonnes. Most of the magnetite deposits are located in the ecological fragile zone of western Ghats. Under the prevailing ecological consciousness, the magnetite ore is not being considered for mining on account of its environmentally sensitive location. The closure of ‘Kudremukh Mine’ was a biggest causality in recent past on ecological consideration. Unless some good mining practices are adopted fully imbibed with environmental integrity, it may be difficult to open and operate a mine in such a zone. So India is left with 14.6 billion tonnes of haematite ore to sustain domestic industry and export compulsion. With a 400 Mt demand by 2020, the haematite resource would likely to exhaust in 36 - 37 years from the year 2020. Under this alarming condition, India is relooking its assessment by lowering the cut-off percentage of Fe from 55 to 45 per cent. This would add some more resource substantially. Then, BHQ/BMQ, the associated underlying rocks, also contain appreciable Fe content of about 25 - 30 per cent. A lab scale beneficiation test has been conducted and it is found that 30 per cent Fe content can be upgraded to +64 per cent Fe. In this way, BHA/BMQ can also be considered as ore in the distant future.

Production and supply The public sector and private sector both are operating iron ore mines. The share of private sector in production is on continuous rise and currently it is about 68 per cent public sector contribute only about 32 per cent. Both the sectors have captive and non-captive mines. The majority of production comes from non-captive mines, which is about 82 per cent of the total production, while the remaining 18 per cent was accounted for by captive production. The captive production grew at a rate of five per cent while non-captive about 20 per cent. The incremental increase in production with each passing year comes from a non-captive mine.

Consumption Domestic Consumption in domestic industry has grown at an annual rate of 17 per cent over the years. Domestic

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

9

V S RAO

industry consumes more lump, ie about 85 per cent. Utilisation of fines is less and thus it finds an export outlet. Export India is the third largest exporter of iron ore after Australia and Brazil. Export has grown at a rate of 5.5 per cent in the first five years. Moreover, the exports decreased to all the other countries except China. The growth in exports has been mainly due to increase in export of 62 per cent Fe grade fine ore.

CONCLUSION x Public sector operates about 40 mines and private sector operates over 200 mines – large and small. Several action plans are being scrutinised to persuade the SME sector to adopt mechanised production process so as to contain ecological degradation and mitigate environmental threats. x Access of land for mining is major constraint to open new mines. It is a conflict between mineral bearing land and forest land. The government is attempting to address this issue. x Out of total exports, about 60 Mt is of less than 62 per cent Fe content. A policy is being drawn which would encourage for beneficiation and value addition in fines, even the value added product and pellets need to be exported.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

10

Optimisation Challenges in Underground Mine Planning and Design H Rubinstein1, M Brazil2, P Grossman3, D Lee4, D Thomas5 and N Wormald6 ABSTRACT Underground mine planning is characterised by a series of designs, starting with prefeasibility and culminating in operational. Information on the mineralisation and geotechnical conditions is available in increasing levels of detail, as development occurs. In the recently completed AMIRA Planning and Rapid Integrated Mine Optimisation (PRIMO) project, a methodology for producing good strategic and then tactical designs using new optimisation tools, was developed. At the strategic level, fundamental decisions must be made such as those relating to cut-off grade, bench heights, mining methods, access and haulage systems, properties of the mill, etc. The inputs are the block and financial models, geotechnical data, infrastructure and production costs, etc. Stope definition for strategic designs includes bench RLs (relative levels), orebody envelopes, and design and placement of amalgamated or simplified stopes. If multiple mining methods are under consideration, then stopes need to be associated with these. Access design must take into account decisions such as declines versus shafts, a template for level layouts and the topology of the access network. Some of the geotechnical and physical constraints are relaxed for strategic design stages. Strategic scheduling is high level, from a month to a year at a time. In the PRIMO project, software tools have been developed enabling the mine planner to rapidly produce a large number of such strategic designs. After the designs have been evaluated, the highest value designs are chosen and a small number of these can then be further refined at a tactical level. For these designs, the first step is determination of all stopes meeting physical and geotechnical requirements and constraints, allowing for complex stope shapes. Next, detailed layout of the mine access network is obtained, satisfying all operational and geotechnical constraints. Finally, more detailed scheduling is performed, over shorter time periods. The Melbourne University group has focused on optimal access design. In particular, the access network is optimised for minimum cost, taking into account the cost of haulage and development over the lifetime of the mine. Costs vary over the different parts of the network depending on the quantity of haulage and so a weighted network optimisation problem has to be solved. We have developed two software tools for this task, decline optimisation tool (DOT) and planar underground network optimiser (PUNO). DOT produces an optimal network of declines and crosscuts connecting access points on the levels to a portal or break-out point from existing infrastructure. PUNO designs the level layout, including footwall and hanging wall drives, development drives and connections to vent raises. A survey will be given of the capabilities of DOT and PUNO when used on some recent case studies. Our current goals include incorporating some aspects of ventilation design and modelling the development costs when additional ground support is required.

1. Professor of Mathematics, Department of Mathematics and Statistics, University of Melbourne, Vic 3052. Email: [email protected] 2. Senior Lecturer, Department of Electric and Electronic Engineering, University of Melbourne, Vic 3052. Email: [email protected] 3. Research Fellow, Department of Electric and Electronic Engineering, University of Melbourne, Vic 3052. Email: [email protected] 4. Emeritus Professor, University of South Australia, SA 5001. Email: [email protected] 5. Head, Department of Mechanical Engineering, Deptartment of Mechanical Engineering, University of Melbourne, Vic 3010. Email: [email protected] 6. Canadian Research Professor, University of Waterloo, Ontario, Canada. Email: [email protected] MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

11

Economics of Mine Planning and Equipment Selection I C Runge1 ABSTRACT This paper examines some of the fundamental economics of mining – in particular, for optimisation studies, for production rate determination and for reserve estimates. It includes introductory discussion on the principles as to how economics drives mining investment decision-making, applicable to the optimising of pit shapes and production, or truck fleets, or any other part of the mine. The main part of this paper looks at the decision to start-up a mine and why the ‘optimum’ production rate to start-up a mine is not the point of lowest cost according to traditional discounted cash flow (DCF) analysis. The optimum production rate is usually substantially less than this rate. The paper recognises the difference between typical mining investments and investments in other industries that do not have the unique characteristics of mining and the problem of uncertainty resolution. It examines the marginal revenue effects, the marginal cost of capital and the value of a real option to expand pending uncertainty resolution as determinants of optimum initial production rates. The final part of this paper looks at applying these principles to renewable and non-renewable resources. It examines mine life and reserve estimates from an economic perspective.

INTRODUCTION This paper follows a similar theme to the last paper I presented to the ‘International Mine Planning and Equipment Selection’ Conference – in Calgary, in 1988. The title of this previous paper was ‘Strip Ratio – An Outdated Indicator of Economic Value’ and the message in the paper was that it is ‘economics’ which drives our decision-making – or ought to drive our decision-making. The paper highlighted that many of the well accepted terms used in industry – terms like: ‘strip ratio’ and ‘cutoff grade’ – were really just simple proxies for some underlying economic-based number. In the precomputer era such simple proxies for the real economic number were necessary; however, with the computer tools now available the actual economic number can be determined and used without relying on the less-precise proxies. Several papers at this conference follow and extend this theme. The application of economics in day-to-day mine planning and equipment selection is clearly much more widespread now than it was 22 years ago. Understanding the fundamental economics is not difficult. Yet when the work is done inside a computer program the basic economic calculations and data inputs can easily become buried and used without due consideration. The robustness of results is at risk. The paper revisits the economics of mining for optimisation studies, for production rate determination and for reserve estimates. It questions the robustness of ‘optimum’ production rates determined from pit optimisation studies. The paper has been framed in three parts: 1. It revisits some very basic and fundamental economic principles as to how economics drives mine decision-making – be that optimising pit shapes or truck fleets, or any other part of the mine. 2. It looks at the decision to start-up a mine and why the ‘optimum’ production rate to start-up a mine is not the point of lowest cost according to traditional discounted cash flow (DCF) analysis. This part of the paper looks at how we choose production rates and pit sizes. It also explains why 1. FAusIMM, Founder, Runge Limited, Brisbane, Australia. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

13

I C RUNGE

mines almost always expand – even immediately after production start-up – and why the initial mine plans should provide for this. 3. The paper looks at applying these principles to another issue that has a contemporary political flavour –the debate about renewable and non-renewable resources. It examines mine life and reserve estimates from an economic perspective. In recent decades the mining industry has been on the defensive against the broad community view favouring recycling and use of renewable resources, over newly-mined commodities. This part of the paper challenges the notion that if you manufacture steel (for example) by recycling that you hold some moral high ground over someone who makes steel out of ‘non-renewable’ iron ore; or that if you generate electricity from non-renewable coal or uranium you are in a morally less-defensible position than someone who generates his electricity from, for example, solar, the wind, or tides. It isn’t so.

MARGINAL COST YET AGAIN In my book Mining Economics and Strategy (Runge, 1998) the notion of marginal cost gets frequent mention. It is one of the most important concepts in economics. To operate mines efficiently – optimally – then ‘marginal cost’ and its partner, ‘marginal revenue’ have to take centre stage in our calculations. There is no ‘optimum’ mine plan without consideration of marginal costs. Equipment selection cannot be ‘optimised’ without consideration of marginal costs. This next section reviews a simple marginal cost example and provides the foundation for the following section which extends the concept to look at whole-of-mine production rate and mine development decisions.

Simple marginal cost example Every production process involves some fixed costs and some variable costs. As production expands, the fixed costs are unchanged, so the average per-unit cost of production attributable to this component declines. If this were the only trend, then the highest production case would be the lowest overall cost of production. However, few production processes work this way. The fixed parts of the process can only service a limited range of variable parts. Trucks interfere with each other if there are too many of them in a confined space. Bigger mines are also deeper and each increment of production means greater haulage distance. As production expands the efficiency of the system declines. Each increment of production incurs variable costs that are more than the variable costs incurred in the previous increment. Figure 1 shows idealised trends associated with economic assessment of this style of production process. The figure does not relate to any particular mine, but the numbers are somewhat in line $150 $140 $130

$ per unit of Production

Marginal Cost $120 $110 $100 $90

Average Cost $80 $70 $60 $50 $40 $30

0

10

20

30

40

50

60

70

Production (Mt/a)

FIG 1 - Average and marginal costs. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

14

ECONOMICS OF MINE PLANNING AND EQUIPMENT SELECTION

with some of the new large scale mines being proposed in the Pilbara area of Western Australia or in the Galilee Basin of central Queensland. Costs per unit of production include return on capital using discount rates at the weighted average cost of capital. The average cost of production is initially high at low levels of production because of the cost of servicing the capital associated with port and rail infrastructure and in servicing the initial acquisition and development costs of the mine. Each increment of production initially has low costs, representing just the operating cost of the equipment and little more. But with increasing production these costs also increase. If the marginal cost is less than the average cost, the average cost declines with increases in production. The production rate that yields the lowest average unit cost of production occurs where the marginal cost curve crosses the average cost curve (40 Mt/a in Figure 1). Although the lowest average unit cost of production is a desirable objective, usually the objective is to maximise profits or net present value (NPV). If the selling price is $100, for example, production can be expanded to 48 Mt/a and the additional production still yields a return higher than the marginal cost. Indeed, this is the rule: Expand production until the marginal cost equates to the marginal revenue (selling price). If in this pricing scenario the production was expanded to (say) 60 Mt/a, with an average cost of about $91/tonne, then the mine would still be profitable. However, the production from the first 48 Mt/a would be subsidising the last 12 Mt/a. Profits cannot be improved by increases in production rate where the marginal cost exceeds the marginal revenue. The same style of calculation applies whether you are optimising a truck/loader fleet or a pit design. From some initial starting position, the marginal cost for some expanded production is compared with the marginal revenue. The theory is simple, but the task is difficult, as anyone who has ever tried to undertake these sorts of calculations can attest. The marginal cost is not just the extra cost of some expanded production scenario. The marginal cost is the change in total cost. That is: it is necessary to design and calculate a complete case for each production scenario. For whole-of-mine studies, the cost curves are hardly ever continuous smooth curved or up-ward sloping lines like shown in this figure. In practice there will be a style of mining and development scenario that is well suited to a certain production rate and then using some other style of mining and development scenario there will be a different but relatively fixed production rate suited to that method. So whole-of-mine analysis will likely have step functions, not smooth curves. Nevertheless the principles still apply, albeit not so easily applied for most cases. Marginal costs have to be calculated, they cannot be measured. Take, for example, a simple truck/loader marginal cost/productivity calculation. Assume that a loader is currently working with a five-truck fleet and that each truck hauls (say) 15 loads in a shift for a total of 75 loads per shift. Now let’s assign a sixth truck to the fleet. Production improves by 12 per cent for a total of 84 loads in the shift, or 14 loads per truck each shift. Adding the extra truck only resulted in an additional nine loads in the shift: the productivity of the additional truck – the marginal productivity – is only 60 per cent of the productivity of the fleet before the change. But this marginal productivity does not appear anywhere in the production records! All trucks, including the extra truck, will achieve similar production (in this case, 14 loads per shift). There is no way of knowing just from the production records what the productivity would have been if a different fleet had been selected. The marginal cost calculation is a planning tool, not a tool for operational decision-making. It is the mine planner who must examine what the expected costs and revenues will be under one scenario and compare them with the expected costs and revenues for the alternate scenario. The choice is then based on the change in total productivity or change in total cost as the case may be. Operations personnel, production statisticians and mine accountants can only record average production and average costs, not marginal cost.

DECISIONS ON MINE INVESTMENTS This next section looks at a proposed investment in a new mine development using the same average and marginal cost model used in the previous section. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

15

I C RUNGE

Any new proposed mine starts with a feasibility study. Most feasibility studies conclude with a financial analysis and discounted cash flow (DCF) followed by a series of sensitivity analysis. Prior to any final feasibility study most technical work is complete and, at least to a non-technical person, subject to little uncertainty. The selling price throughout the life of the mine is also one of the inputs to the DCF, but because this is always an unknown most studies use sensitivity analysis to examine its impact on the project return. Indeed, the selling price, internal rate of return (IRR) and NPV are linked: a nominated selling price and rate of return will yield the project NPV; if the rate of return is set, the selling price necessary for project viability (NPV set to zero or greater) can be calculated; or, the IRR can be calculated for any selling price by calculating whichever rate results in an NPV equal to zero. Most DCF analyses use the company weighted average cost of capital as the discount rate and undertake sensitivity analysis on the results from this starting point. Figure 1 has been prepared using this form. It works out, for each production rate, what selling price yields a zero NPV when discounted at the cost of capital. The value of this form of presentation is that it allows a lot of analysis of a project without necessarily requiring knowledge of the selling price in advance. Figure 1 shows a set of results from a range of production scenarios ranging from less than 10 Mt/a to more than 60 Mt/a. At a selling price of $100 per tonne, the mine is viable at annual production rates ranging from 16 Mt/a to 48 Mt/a. A ‘pit optimisation’ would suggest the mine be developed at 48 Mt/a for maximum NPV. Yet mines are seldom developed this way. They are almost always developed at production rates substantially less than these ‘optimum’ rates. There are good economic reasons why this is the case. They suggest that if you are using computer models for pit optimisation, additional considerations beyond the narrow classical DCF analysis is warranted. The three main reasons concern marginal revenue, marginal costs of capital and the value of options to address uncertainty.

Marginal revenue In the example shown in Figure 1 the ‘optimum’ production rate yielding the greatest NPV (48 Mt/a) was based on a selling price that was assumed constant and independent of the mine itself. This assumption is a very common one in economics, because in most markets with large numbers of buyers and sellers, any one producer affects the price very little. Demand curves are flat. In mining, except for perhaps gold and silver producers, this is rarely the case. There are standardised prices (‘benchmark’ prices, or prices from the London Metal Exchange (LME)) but these are not the prices that a new producer can rely on receiving for new products on offer in the market. For bulk commodities like iron ore and coal, new mine production has to be explicitly sold to individual customers and with variations in mineral content and impurities, not to mention that individual customers might already have contracts in place with other suppliers, only a limited number of new customers will be in a position to use the products directly and pay the ‘market’ price. Over time, assuming competitive market pricing, other customers will adapt, but at the time prior to project commitment the products can be sold to this broader set of customers only at a discount. The demand curve slopes downwards. The optimum production occurs not when the marginal cost equates to average selling price, but when it equates to the marginal revenue. At the time of project commitment, this marginal revenue may be substantially less than the average selling price. Even for base metals that are marketed on the basis of LME pricing, the ‘marginal revenue’ impact is relevant. The LME price is a reference price, but the trading of base metals doesn’t necessarily require products to conform to these LME specifications. Nickel, for example, is frequently sold with Fe impurities, but this ‘impurity’ is a bonus when used in stainless steel production because the nickel is mixed with Fe anyway. So new nickel production sold into the stainless steel market may attract the full LME pricing for its Ni content, but the same nickel placed in some other market where the Fe impurity isn’t a benefit may be sold only at a substantially reduced ‘marginal’ revenue. The slope of the demand curve and this marginal revenue effect, is time dependent. Initial operations are commonly faced with a steeply declining demand curve, but, depending upon the speed at which the marketplace can adapt, the slope of the curve reduces over time. Thus, although the marginal revenue effect constrains initial production rates, over time the economic forces favour expansion. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

16

ECONOMICS OF MINE PLANNING AND EQUIPMENT SELECTION

Two outcomes flow from the marginal revenue effect: 1. Pit optimisation studies should not just consider ‘average’ prices, but should consider marginal revenues in determining initial production rates from new mines; and 2. Sensitivity analysis is an inadequate tool for understanding this marginal revenue effect – it tells you the impact of price changes on a mines ROI, but it provides no guidance for changes in the mine plan to accommodate the effect.

Marginal cost of capital This next section examines the second reason why mines aren’t and shouldn’t be, developed at production rates suggested by classical DCF optimums. It has to do with the marginal cost of capital. In the calculations described up to this point, discounting has used the ‘weighted average’ cost of capital. This term is the give-away. ‘Optimisation’ of anything has to be based on marginal costs, not average costs and this applies to capital as anything else. There are two components of this marginal cost of capital calculation: a risk element and a size element.

Marginal cost of capital from a risk perspective The prime element of the weighted average cost of capital for any company is the risk characteristics of its existing business. The risk characteristics of any new mine may be substantially different to the risk characteristics of a mining company’s portfolio of existing mines. Risk associated with new mines will be higher until the mine is in full production. Once a mine is in operation and its risk characteristics are understood with greater certainty then the cost of capital notionally applicable to the new mine will reduce. However, this is something that occurs only after the mine is in operation (hopefully), not prior to commitment. Of course, every effort is usually made in the feasibility study to address known risk elements, but some of these are unknown until the mine commences operation. Market demand is one of these elements – potential customers might say they will buy your product and perhaps provide a ‘letter of intent’, but it is in their best interest to encourage greater supply to the market, so this intent cannot necessarily be relied upon. No amount of assessment prior to and during the feasibility study will necessarily reduce the uncertainty associated with this sort of input. The private operators of toll tunnels in Australia’s eastern states capitals can attest to this patronage uncertainty very well. Thus the risk and with it, the marginal cost of capital for a new mine at the time of commitment is almost certainly higher than the weighted average cost of capital for the company itself. Does the size of the company make any difference? A statement commonly heard is that such-andsuch company (a large mining company) can develop projects that a small company cannot develop because they have a lower cost of capital. From this ‘risk perspective’ this isn’t necessarily true. A large company with a low weighted average cost of capital can still have a high marginal cost of capital if the new mine involves geographical locations or application of technology that they are unfamiliar with. A smaller enterprise with recognised expertise in the geographical area or recognised expertise in the technology can still have a lower marginal cost of capital even while having a higher average cost of capital.

Marginal cost of capital from a size-of-investment perspective This second component of the ‘marginal cost of capital’ involves the size of the investment in relation to the size of the company. In Figure 1, the curves for all production rate scenarios were prepared using the (weighted average) cost of capital. This assumes that the extra investment necessary for a mine developed at the 48 Mt/a annual production rate can be sourced at the same cost as the smaller investment for a much smaller 16 Mt/a mine. If the size of the investment is small compared to the size of the company then this increase in funding cost may be negligible. But for most companies as the amount of funding increases so does its cost. The cost of the extra funding (the marginal cost of capital) needed to finance the mine development at the higher production rate might make higher production scenarios uneconomic. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

17

I C RUNGE

Does the size of the company make any difference from this ‘size of investment’ perspective? Yes. Small companies wishing to grow big through developing a large deposit are particularly impacted. To compete with larger companies in similar circumstances they have to find deposits that are profitable at lower-than-optimum production rates because they have higher marginal costs of capital with project size compared to large companies. Two outcomes flow from these marginal cost of capital effects: 1. Pit optimisation studies should not just consider ‘weighted average’ costs of capital in assessing costs of production, but should consider a rising marginal cost of capital in determining initial production rates from new mines. The outcome will be lower ‘optimum’ initial production rates than is the case with classical DCF analysis. 2. Sensitivity analysis is an inadequate tool for understanding this effect. Sensitivity analysis tells you the impact of cost-of-capital and discount-rate changes on a mines viability, but it provides no guidance for changes in the mine plan to accommodate this impact. The risk issues which partially give rise to the higher marginal costs of capital will often resolve themselves once the mine is in production. Thus, although the marginal cost of capital effect constrains initial production rates, over time the economic forces again favour expansion.

Option value of uncertainty resolution In the introduction to this section, classical feasibility studies and financial analysis were understood to be based on little uncertainty in orebody definition. Moreover, they assumed a relatively constant if not-well-known selling price and a simple (one number) weighted average cost of capital across the range of scenarios. The section ‘marginal revenue’ relaxed the assumption of the relatively constant selling price in favour of marginal revenue. The section ‘marginal cost of capital’ relaxed the assumption of the relatively constant costs of capital. Now this final section examines the issue of uncertainty. There is always uncertainty in the orebody – if not in the definition of what is in the ground, then in uncertainty in the cost and efficiency in its metallurgical treatment, or in the cost and efficiency impacts on the customer processes. Every industry, including mining, has to face uncertainty prior to the investment decision; it is the nature of the uncertainty that sets mining apart. A tourism investment, like a mining development, for example, might be subject to the vagaries of the weather, but no amount of study beforehand will necessarily resolve the uncertainty. What sets mining investments apart from most other investments is that the economics of resolving the uncertainty are both significant and endogenous to the investment process. In a mining study you can resolve the uncertainty through more drilling and metallurgical testing, but at some point the reduction in risk associated with additional work doesn’t yield benefits to offset the additional cost. Moreover, once a mine is in production the costs of resolving the uncertainty may be much less than these costs in the preproduction phase. Underground orebodies can be drilled out from a drill location deep underground at lower cost than the same orebody drilled out from the surface. The decision regarding uncertainty resolution is this – to incur high costs now to resolve more of the uncertainty now, or start-up the mine and resolve the uncertainty then at much lower cost then. The option to expand values this second alternative. Such a strategy requires that the mine still be profitable at the initial lower production rate. This is where the uniqueness of mining investment enters the analysis, because in few other industries can such a plan be entertained. There must be unique and variable inputs to the production process giving rise to exploitable short-term revenue or cost-saving effects. Orebody delineation costs (particularly in an underground mine) may be substantially lower than the costs of the same task if undertaken prior to mine commencement. This style of uncertainty will often be resolved within a very short time after the mine is in production and like the marginal revenue effects and the marginal cost of capital effects, the economic forces favour expansion almost immediately upon its resolution. Real options such as this ‘option to expand’ are highly valued by industry practitioners. The option has value because it is associated with resolution of one or other form of uncertainty. However, MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

18

ECONOMICS OF MINE PLANNING AND EQUIPMENT SELECTION

this value does not show up in the financial analysis if the underlying assumption in the feasibility study assumes no uncertainty. Sensitivity analysis, based on a base case that has no uncertainty is an inadequate tool for understanding this option value. If an orebody turns out differently to what is initially expected sensitivity analysis can potentially tell you the impact of this on the return on investment (ROI), but it provides no guidance for how differently the mine should be planned to accommodate the effect.

BAD GUYS IN A COMMODITY-CONSTRAINED WORLD This final part of the paper looks at the economics of mining and exploitation of non-renewable resources. It poses the question: If you mine and burn coal to make electricity are you in a less-defensible moral position than someone who generates electricity from solar or wind? Are you unfairly exploiting the world’s resources to the detriment of future generations? Whilst this notion is quite a contemporary one, it has actually been around for some time. The earliest scholarly mention of it can probably be attributed to William Stanley Jevons, an English economist, who published a paper in 1865 titled The Coal Question (Jevons, 1865). In the paper, Jevons explored the implications of Britain’s reliance on coal and, given that coal was a finite, non-renewable energy resource, raised the question of sustainability. His central thesis was that the UK’s supremacy over global affairs was transitory and when the coal ran out so too would the British standard of living. He advocated constraint. Suffice it to say that Britain did not follow his recommendations. And while the UK’s supremacy over global affairs was transitory, this undoubtedly was due to the rise of the American economy and to the impact of two world wars rather than to the run-down in coal production. Indeed, even now with the run-down in the British coal industry since the 1980s – 120 years after Jevons’ alarm – the standard of living in Britain hasn’t collapsed. There are several reasons why history has proved Jevons concerns unjustified. The reasons also apply today for coal, oil, gas and any other commodity. They involve the economics of reserve estimation, how effectively commodities are used in society and the changes in technology that make some commodity that is important today far less important tomorrow. This paper addresses just the economics of finding and delineating reserves, supported by the empirical evidence from history.

Economics of reserve estimation It is undeniable that the earth’s natural resources are finite and if used continuously they will eventually run out. The world’s reserves of copper, for example, equate to about 35 years of mining at current production rates, suggesting that this day of reckoning is not too far into the future. Most of the important mineral commodities used in the world have a similar limited future from this perspective, with some precious metals such as silver, for example, having reserves of less than 15 years at current production rates. Yet the same trends have existed since records started to be kept. Table 1 shows the reserves of four important minerals in 1950 and 1980, compared to the production during the same period (after Repetto, 1987). TABLE 1 World reserves and cumulative production of selected minerals (millions of tonnes of metal content).

Mineral

Reserves (1950)

Production (1950 - 1980)

Reserves (1980)

Aluminium

1400

1346

5200

Copper

100

156

494

Iron

19 000

11 040

93 466

Lead

40

85

127

Thus, from a 1950s perspective, even though production in the ensuing 30 years exceeded, or nearly exceeded, the reserves available at that time, the reserves at the conclusion of period far outweighed the reserves at the start. Indeed this trend applies for most commodities not just in total terms, but MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

19

I C RUNGE

also in ‘years of production’ terms. This is indeed comforting, but is it a trend that those outside of our industry can rely on us to replicate into the future and if so, how far into the future? The reason this trend exists and why it can be relied upon, comes down to simple economics. Exploring for and proving up reserves is costly. It is uneconomic to prove up reserves too far into the future, because the return from exploiting them only materialises when they are mined. The world would certainly take comfort in the knowledge that oil reserves, for example, are good for 60 or more years into the future, but what company can justify spending even $5 per barrel to prove up more oil now, in the knowledge that the oil won’t be exploited until 2070 or later? It is just uneconomical. How far into the future reserves are known is largely a function of how costly it is to prove up the reserves as a proportion of the costs of extraction. Silver has limited reserves (in comparison to annual production rates) because orebody delineation is a relatively high cost in typical silver deposits. In contrast, the delineation cost for most coal deposits, for example, is relatively low – accounting for the current ‘healthy’ reserve position for coal at more than 100 years at current production rates. As the cost of proving up reserves increases in proportion to the total production cost, the reserve position in terms of ‘years into the future’ will decline, but this says nothing about the total size of reserve or annual production rate. A second concern commonly expressed involves supply and demand. In a market economy if there is a shortage of any commodity, the price will rise until supply matches demand. At this point the economics will favour additional exploration and deployment of technologies to exploit deposits hitherto considered uneconomic. There is no doubt that that this has been an important factor in spurring new mining technologies throughout history. However, it suggests that the world can rely on such an economic trend only at the cost of an everincreasing real price of the scarce commodity. Much of the anxiety in the world today concerning oil rests on this assumption. Yet history suggests that this should also not be an issue. Supply and demand fluctuations are important economic triggers to spur new technology, but having done so, the long term price impact appears to be negligible. Baumol and Blackman (1993) quote several studies that find long term real price declines of the majority of mineral commodities over time. Even energy commodities such as coal, natural gas and oil, though experiencing substantial fluctuations, show negligible upward trend in real (inflationadjusted) price: The price history of nonfuel minerals is even more striking. Some, like iron, have experienced a very slow rise over the last 100 years or so. The price of others, like lead, have remained stable. And for some, including aluminium and magnesium, real prices today are far lower than they were 70 years ago. The prices of about half of the mineral resources investigated actually fell after correction for inflation. If reserves of important mineral commodities were indeed likely to be exhausted in 15 years, or even 30 years, then all of us, not just environmental activists, would see cause for alarm. We in the industry certainly cannot guarantee that as known reserves are exhausted each year we will be successful in finding the same or more new reserves. And like the impact of the Organisation of Petroleum Exporting countries (OPEC) cartel 30 years ago, there is always the scope for non-market forces to impact the supply/demand balance of specific commodities. However, with reliance on market forces there are two conclusions that can be drawn from this assessment: 1. The current reserve positions of most important mineral commodities are strong, are consistent with historical trends and the reserves are as extensive or as limited as they are based on sound economics. 2. the trends from at least 150 years show constant or reducing real costs of production, even with substantial and continual annual increases in production. This trend provides high confidence that future demand for these commodities can be met. Our critics who suggest otherwise have to date been largely left unchallenged. Yet economics and the weight of history in our industry suggest it is upon them that the burden of proof should rest. Let solar electricity generation compete. Let wind electricity generation compete. Let recycling compete. A level economic playing field for all technologies, including the extraction of nonrenewable resources, is the best guarantee that these resources will be responsibly exploited. But let us not concede any advantage – moral or otherwise – to these competitors based only on the notion that the products we make are subject ultimately to exhaustion. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

20

ECONOMICS OF MINE PLANNING AND EQUIPMENT SELECTION

REFERENCES Baumol, W J and Blackman, S B B, 1993. Natural resources, in The Fortune Encyclopaedia of Economics (ed: D R Henderson), pp 40-43 (Warner Books, Inc: New York). Jevons, W S, 1865. The Coal Question; An Inquiry Concerning the Progress of the Nation, and the Probable Exhaustion of our Coal Mines (MacMillan and Co: London). Repetto, R, 1987. Population, resources, environment: An uncertain future, Population Bureau, 42:2, quoted in Baumol and Blackman (1993). Runge, I C, 1998. Mining Economics and Strategy, pp 40-43 (The Society for Mining, Metallurgy, and Exploration, Inc: Littleton).

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

21

More than Coal Mining J Tourek1 ABSTRACT The energy supply in the 21st century is still secured by coal. According to estimates of the International Energy Agency (IEA), coal will have the same importance as oil for the world-wide supply of energy until 2030. Coal is also an important element of EU policy, its future use is closely linked to effectively addressing the environmental challenges it poses. Hard and brown coal represent approximately 80 per cent of European Union (EU) reserves of fossil fuels and coal utilisation can co-exist with the development of power generation based on renewables. Europe is the world’s third largest consumer of coal behind China and the United States. The Czech Republic is poor in energy resources except for coal. Brown coal is relatively abundant (mineable reserves estimated at 1800 Mt, output 48 Mt/a) and significant quantities of hard coal (mineable reserves 900 Mt, 13 Mt/a), much of it suitable for coke production, are also present. Severoceske Doly (SD Company), the biggest Czech coal mining company (22 Mt/a, 48.6 per cent Czech market share) and one of the biggest in Europe, reinforces despite the last deep crisis in the Czech and European economies its long-standing stable and strong market and financial position. From 2009, the SD Company co-owns, with partners, a brown-coal mining company MIBRAG (21 Mt/a). SD Company invests in large mining equipment (eg new giant spreaders and bucket-wheel excavators, extension of overland belt conveyors), rebuilds and upgrades mining and processing installations, auxiliary and other machinery. We should see the investment we make today in tomorrow’s mining and reclamation as an opportunity and not as a burden. They enable to make our position more resistant in the longer term. Projects focused on the environment and operational safety went hand in hand with these efforts. R&D activities relate to technical innovations, improvements of mining equipment, research and evaluation of conditions for extracting coal seam and/or overburden and mitigation of environmental impacts, utilisation of coal in new products. The company spends considerable amounts on reclaiming affected land and creating a new post-mining landscape. Our dialogue with influential stakeholders in the region where we operate contributes to the long-term improvement in the quality of life of local residents. Such dialogue with communities in the last ten years has been instrumental for confidence building. SD Company helps to develop many projects of towns and villages in its vicinity. The stable and transparent relations with our neighbours are in accordance with our basic principles of corporate social responsibility.

1. Chancellor of the CEO Office, Severoceske Doly (SD Company), North Bohemian Coal Mines, Bozeny 5359, 43001 Chomutov, Czech Republic. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

23

The Next Challenge in Optimising Mining Operations J Whittle1 ABSTRACT This paper describes the past, normal and advanced practice of optimisation in mining. It then discusses what needs to happen for the mining industry to make full use of what is now available.

INTRODUCTION When developing a long-term plan for the operation of a mine, it is important to be very clear about what our main aim is. If our aim is to maximise shareholder value, then our aim is not to produce as much product as possible, to maximise reserves or to maximise mine life. Our aim is not to minimise mining cost per tonne or to maximise recovery. To maximise shareholder wealth, we must maximise net present value (NPV). I fully understand that there are other measures of mine value, and I will return to this later. For now, let us stick with NPV. In developing a long-term plan, we have to design the pit and the mining phases. We have to plan how fast to mine. We have to decide on what plant to install and at what throughput and recovery to operate it. There are many such planning decisions. For each planning decision there is, by definition, a best plan, the plan that gives the highest NPV. Most decisions are about size or speed and there will usually be figures that are too small and figures that are too large, with a value somewhere in between that gives the best result. However, all these planning decisions interact. Change one and all the others may change. In other words it does not make sense to optimise one decision without considering its effect on the other decisions.

HISTORY Roughly 25 years ago, software became widely available that would find the pit outline with the maximum total cash flow. It was a start, but total cash flow is not NPV, because it values a dollar received in ten years time the same as a dollar received today, and that is not realistic. Within a couple of years there was software that would generate a series of pit shells based on a range of product prices, and which would calculate an NPV by simulating a crude mining schedule using mining phases based on particular pit shells. Over time the scheduling was improved, but you still had to design the pit and fix the phases before optimising the schedule. Later, software to optimise cut-offs became available, but it assumed a pit design, a phase design and a schedule. Also, when the cut-offs were optimised, the schedule generally proved to be wrong, so some iterative work was required. About a dozen years ago, software was developed that would optimise a number of planning aspects simultaneously, and we now have a lot of experience in its use. That experience, and further software development, has led to some large gains in NPV.

WHAT CAN NOW BE DONE Given a pit design and a set of phases, it is now possible to simultaneously optimise the mining and processing schedule, the cut-offs and stockpiles, the mill throughput/recovery set-up, the logistics and the capital expenditure. With manual iteration of the ultimate pit outline and the phases, the

1. FAusIMM, Whittle Consulting Pty Ltd, Suite 13, 333 Canterbury Road, Canterbury Vic 3126. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

25

J WHITTLE

whole long-term plan can be optimised. This can be done for a single pit or for a complex of dozens of open pits and underground mines with multiple processing options and products.

EXAMPLE In order to be able to illustrate this, we set up a simple case study based on an artificial orebody. Many of you will be familiar with the ‘Marvin’ copper/gold orebody. We initially assumed x single copper/gold pit with four phases x 60 Mt/a mining x a maximum of eight benches per year advance in a phase x 20 Mt/a crush/grind/float producing 28 per cent copper concentrate x 88 per cent copper recovery, 60 per cent gold recovery x 70 km 600 Ktpa pipeline to port x offshore smelter/refinery x gold price 900/oz x copper price 2.50/lb declining to 1.50 in the first five years and x initial capital expenditure of M582. ou will find more detail of the optimisations we did in the paper by Gerald Whittle in this proceedings, but here I will just give an overview. We started from a manual design and schedule. We then did pit limit, phase, skin and schedule optimisation using the Gemcom-Whittle software. These were optimised separately, but they were iterated manually. The NPV increased by 18 per cent from M1597 to M1885. This can be regarded as ‘normal modern’ planning. We next added cut-off and stockpile optimisation, which is available in Gemcom-Whittle, but seldom used. Again the different components were optimised separately but iterated manually. This raised the NPV to M2201, an increase of 17 per cent on the normal modern plan. Next we used different software to do schedule, cut-off and stockpile optimisation simultaneously, with manual iteration of the pit limit and phases. This added a further ten per cent. In other words, just optimising these things simultaneously rather than separately gave us a ten per cent increase. This facility has recently become available in Gemcom-Whittle. Finally we added simultaneous optimisation of the mill throughput/recovery calibration, the concentrate percentage, and the capital expenditure on the mining fleet and the pipeline. This gave an NPV of M2775. This final NPV was 74 per cent higher than the manual plan and 47 per cent higher than the normal modern plan. Those are the gains. What about the implementation x Are there difficulties es. x Are there impossibilities No. x Will your life be more comfortable No. x Will the mine be much more profitable es, by any measure What did this plan look like Its most obvious feature is that every year is different. Let us look at a couple of aspects of the plan. Figure 1 shows the mining rate for the fully optimised plan. The optimiser spent more capital to increase the mining capacity from 60Mt/a to 83Mt/a but only used the whole capacity in three of the years. There are significant periods with idle equipment. ‘But we have to keep the equipment busy’, says the Mining Manager ‘otherwise we increase the average cost per tonne of mining and that is against my key performance indicator’. Now we are getting near to the real problem. The plan is not the problem, the key performance indicator is the problem. Our aim is to maximise NPV, not to minimise mining cost. Figure 2 shows the mill throughput for the fully optimised plan. Remember that the nominal throughput was 20Mt/a. We have sacrificed recovery for throughput for most of the life of the mine. A KPI could cause problems here too. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

26

THE NEXT CHALLENGE IN OPTIMISING MINING OPERATIONS

There are other radical changes but, in summary, we have x reduced the size of the ultimate pit and hence the reserves, x increased the mining cost per tonne, x reduced the recovery in the plant for much of the project life, x increased the capital expenditure by five per cent, and x reduced the life of the operation by three years. Note that each of these changes is counter-intuitive and would have been resisted by the relevant manager. Also, if any of them were to be made in isolation, the result could be disastrous. The changes must be part of a carefully coordinated plan.

MillionsofTonnesperYear

FullyOptimizedMiningRate 100 80 60 40 20 0 0

2

4

6

8

10

12

14

16

Year

 FIG 1 - The annual mining rate for the fully optimised plan.

MillionsofTonnesperYear

FullyOptimizedProcessingRate 30 25 20 15 10 0

2

4

6

8

10

12

14

16

Year

 FIG 2 - The annual mill throughput for the fully optimised plan.

THE REAL CHALLENGE The real challenge is a human one. We need to change the way mines are managed. Currently some decisions, like the mining and processing rates, are made and ‘cast in bronze’, far too early and on too little evidence. It then becomes difficult, though not impossible, to change them. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

27

J WHITTLE

The reserves are announced to the share market to support the share price before thorough optimisation has been done. For simplicity, the running of the mine is divided into different silos and over-all targets are set for each. To keep their lives simple, the various silo managers concentrate on their own area and have little interaction with other managers in relation to planning. We are paying a lot for this simplicity. Management will protest that, if we reduce the reserves, the share price will go down Well it won’t if we tell the market that we now plan to pay back the whole of the capital cost in eight and a half months rather than seventeen months, and that the net revenue for the first three years will be four times the capital cost. Figures 3, 4 and 5 show the net cash flows for the manual, the ‘normal modern’ and the fully optimised plans. Most investors would greatly prefer Figure 5 to Figure 3 or Figure 4. Figure 5 offers more growth opportunities and/or dividends even after allowing for taxation and any royalties. The point is that the gain is so substantial that it improves the mine value however you measure it.

ManualNetCashFlow

MillionsofDollars

1000 800 600 400 200 0 0

2

4

6

8

10

12

14

16

Year

 FIG 3 - The annual net cash flow for the manual plan.

NormalModernNetCashFlow

MillionsofDollars

1000 800 600 400 200 0 0

2

4

6

8

10

12

14

16

Year

 FIG 4 - The annual net cash flow for the ‘normal modern’ plan. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

28

THE NEXT CHALLENGE IN OPTIMISING MINING OPERATIONS

FullyOptimizedNetCashFlow

MillionsofDollars

1000 800 600 400 200 0 0

2

4

6

8

10

12

14

16

Year

 FIG 5 - The annual net cash flow for the fully optimised plan.

IS YOUR CURRENT PLAN OPTIMAL? How can you tell if your existing plan is optimal There are a few obvious things to check. If you plan to mine the same amount every year, your plan is unlikely to be optimal. After all, the ore you mine isn’t the same every year, why should your plan be the same every year If you plan to use the same cut-offs every year, the plan is unlikely to be optimal. In fact, if almost any major parameter, other than the primary bottleneck, is constant over time, your plan is almost certainly not optimal.

WHAT SHOULD BE DONE? Let us assume that you are operating a mine along traditional lines and not using the techniques I have described. First you need a champion. Hopefully it is the Chief Executive Officer (CEO) or Chief Operating Officer (COO). An internal task force needs to be set up with senior representatives from geology, mining, processing, logistics, sales, finance and human resources, and the focus should be primarily on NPV. For an effective enterprise optimisation, decision-makers from all aspects of the enterprise need to be involved. They have a vested interest in the outcomes and will be implementing those outcomes. This is not a job that can be handled by the mining department alone. This is enterprise optimisation, not mining optimisation. Many of the gains are made by modifying the operation of more than one silo at once. The task force meetings should be chaired by the CEO or COO and mediated by someone with skills in this type of optimisation. ou need access to suitable software and a skilled operator to run it. This software cannot be treated as a black box that you put a block model into and get a finely tuned plan out of. For example, the study I have described required 58 different runs of the optimiser. The example is a single-pit single-plant operation, but it is worth noting that the larger and more complicated the operation, the greater the opportunity for improvement. If you have a dozen pits with two or three different plants in different locations, simultaneous optimisation of the whole complex can produce great gains. In our practice, we have ceased to be surprised when our studies reveal a way of increasing the NPV by a billion dollars It has happened several times.

CONCLUSIONS The next challenge in optimising mining operations is not to develop new software. The next challenge is to change the way mines are managed and to change the way management reports to the stock exchange. Isolated silos are just not good enough, and we need to report more than just reserves. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

29

J WHITTLE

Twenty five years ago, when my wife and I were trying to sell optimisation software in the United States, I had a guy stand in front of me and say ‘My granddaddy showed me how to design a mine, and that’s good enough for me’. The mining industry has come a long way since then, but there is still much more to do.

ACKNOWLEDGEMENTS Although I wrote the software we use, much of the credit for these developments should go to our son Gerald Whittle. He saw possibilities that I had not dreamed of, and he ran with them.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

30

A Simulation Approach for the Comparison of In-Pit Crushing and Conveying and Truck-Shovel Mining Methods R Bearman1 and S Munro2 ABSTRACT In-pit crushing and conveying (IPCC) is being increasingly considered as an alternative to traditional truck-shovel mining methods. The move towards IPCC is driven by a number of factors including the potential to reduce greenhouse gas emissions, the availability of skilled labour and the need to increase extraction rates to levels that present significant challenges to traditional methods. When considering the relative performance of IPCC and truck-shovel methods, the outcome can be inadvertently impacted by either the use of different simulation approaches, or techniques that do not adequately model the processes in sufficient detail. A simulation approach is proposed that models both IPCC and truck-shovel using the same software platform and is capable of functioning at a wide range of resolutions from life-of-mine scale through to second-by-second asset tracking level. The simulation framework also holds models of processing equipment and hence the mining simulations can be extended to gauge the impact on downstream plant. The approach, known as mine-process operational simulation (MPOS), can be tailored to the level of information available and hence the type of study required, ie order of magnitude, prefeasibility, feasibility, or detailed design. Using information from block models and a range of detailed equipment and process models, MPOS can span major system discontinuities, typified by the mine-plant interface, to allow the estimation of final mine and plant product specifications.

INTRODUCTION While the use of in-pit crushing and conveying (IPCC) in open pit mining is not a new concept, the number of studies considering IPCC as an alternative to traditional truck-shovel mining methods has increased dramatically over the last ten years. The increased level of interest in IPCC has been driven by a variety of factors, such as: x the contribution of diesel-powered heavy mobile equipment emissions to greenhouse gas (GHG) footprint, x the availability and cost of labour to operate and maintain truck fleets, x the potential for mine production efficiency gains through the use of IPCC, and x a potentially reduced total cost of ownership for IPCC versus truck-shovel.

Greenhouse gas element Inevitably, the mining industry is seeking to reduce the GHG footprint of operations. IPCC has attracted a great deal of attention on the basis that an IPCC system coupled with face shovels will predominantly use electrical power, compared to truck fleets that are diesel powered. The debate is still ongoing in regard to IPCC versus truck-shovel and, as will be illustrated later in this paper, the conclusion is contextual. The contextual element relates to two main points, namely: x What is the source of the electrical power? ie hydro, coal, nuclear, etc and therefore what is the embedded emissions in a unit of electrical power? 1. MAusIMM, Director, Bear Rock Solutions Pty Ltd, PO Box 150, Melville WA 6956. Email: [email protected] 2. MAusIMM, Director, Met Dynamics Pty Ltd, PO Box 150, Melville WA 6956. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

33

R BEARMAN AND S MUNRO

x What is the basis for the comparison and how have the battery limits been applied? ie if power generation emissions are included in the electrical power used for IPCC, are the process emissions for diesel fuel production included in the assessment of GHG for trucks? In terms of the first point above, the wide range of emissions can be illustrated simply by examining a single country. Aube (2001) studied various aspects in relation to the calculation of equivalent CO2 emissions. As part of the study he assembled a table that stated the CO2 emissions for the various provinces of Canada for 1998 (Table 1). Although the data is now somewhat dated, it illustrates how the case for IPCC could be influenced just by small variations in location. TABLE 1 CO2 equivalent emission factors for electricity, by Canadian province, 1998 (Aube, 2001). Province

t/MWhr

Newfoundland and Labrador

0.034

Prince Edward Island

1.012

Nova Scotia

1.984

New Brunswick

1.012

Quebec

0.009

Ontario

0.236

Manitoba

0.030

Saskatchewan

0.844

Alberta

0.915

British Columbia

0.027

Territories

0.368

Average for Canada

0.222

For a fair comparison of IPCC and alternative methods, the battery limits must be chosen to be equivalent and appropriate for the study being undertaken. Any inconsistencies in this element can grossly distort the outcome of the study. Any comparison of IPCC versus truck-shovel mining needs to answer these questions.

Availability and cost of labour There is little doubt that IPCC systems are less labour-intensive than equivalent truck-based mining methods. Given the scale of some of the newer IPCC systems, where it is possible to achieve 10 000 t/h as an average production figure, the sheer number of trucks required to move such quantities demand a significant workforce to operate and maintain the fleet. The cost of employment of personnel is variable in terms of geography, but certainly in remote areas, as seen in parts of Australia, the cost of personnel is dramatically increased due to fly-in, fly-out operations. Costs include air charters, site accommodation, power, water, infrastructure and training, in addition to normal remuneration packages. Such associated expenses can approximately double employment costs and if this is coupled with low employee retention rates, the ongoing re-training cost can also be considerable. In addition, mining in many countries is failing to attract personnel willing to operate trucks, as it is not seen as an attractive career.

Efficiency and cost of in-pit crushing and conveying IPCC, in terms of material movement rates, is raising the bar for all mining methods. Equipment requirements, labour, energy and resulting costs per tonne of material, are often claimed to be considerably lower than for truck-based alternative methods. Schroder (2003), states that the specific cost (inclusive of capital) for a 20 Mtpa coal conveying and stockpiling system is US$0.334 per tonne compared to US$0.925 per tonne for a truck-based mining method. Other papers claim similar levels of saving for IPCC (Zimmermann and Kruse, 2006). MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

34

A SIMULATION APPROACH FOR THE COMPARISON OF IPCC AND TRUCK-SHOVEL MINING METHODS

There is also a scale effect associated with the increasing truck payloads that are required to match the production rates achievable with IPCC. Bozorgebrahimi et al (2003) examined how increasing truck payload size impacts maintenance costs. Although not conclusive, one finding was that productivity increases expected, due to larger payloads, did not match those found in reality. When maintenance costs and productivity were factored into the equation, it was found that the larger trucks were ten per cent higher on a cost per tonne basis. If the types of cost advantage stated above for IPCC can be replicated under different conditions and in various applications, then the pursuit of IPCC will be wholly justified. It is essential to ensure that assessments of IPCC deployment evaluate this aspect and also take into account changing truck fleets. Any assessment needs to do so in a clear and appropriate manner otherwise this could impact the credibility of IPCC. This aspect was one of the main drivers behind the development of the system at the heart of this paper.

DEVELOPMENTS IN IN-PIT CRUSHING AND CONVEYING EQUIPMENT Developments in IPCC have mainly centred around the mobility of the crusher stations and the improved use of conveyor networks. Mobility of crushing plant spans the entire spectrum from fixed, single point installations all the way to crawler-mounted fully mobile units. The mobility of crushing equipment is only restricted by the ingenuity of engineers. Even some of the largest gyratory crushers in the world can be moved from one static installation to another. A good example of this type of ingenuity is the movement of the TKF 63-114 crushers at the PT Freeport Indonesia, Grasberg pit. Figure 1 shows two such crushers in the process of relocation.

 FIG 1 - Relocating 63-114 primary gyratory crushers at the Grasberg Operation (ThyssenKrupp, 2010).

Although a significant feat of engineering, the mobility displayed in Figure 1 is still towards the static end of the mobility spectrum. Traditional true-IPCC equipment was predominantly based on the large ‘walking’ style gyratory crushers. Although mobile, the mobility is limited and hence the number of moves is usually restricted to the absolute minimum necessary. However, even with this limited mobility, several major operations found that such equipment offered flexibility and options that were not available through the usual pit-rim, or concentrator based static crusher installations. The walking type mobile crushers have been successfully used in a number of mining and aggregates operations. Figure 2a shows the walking gyratory crusher installed at the Bardon Torr Quarry (formerly Foster Yeoman), whilst Figure 2b shows a machine treating limestone using a mobile double shaft impact crusher. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

35

R BEARMAN AND S MUNRO

(A)

(B)

FIG 2 - (A) Walking gyratory crusher in operation at the Torr Quarry; (B) Walking double shaft impact crusher at Deuna Zement GmbH (ThyssenKrupp Materials Handling, 2010).

The tendency has been to demand more mobility to the point whereby the crusher can be fed directly from the face-shovel, or excavator. To achieve this, several manufacturers have designed tracked systems based around high throughput, smaller footprint crushers such as sizers and double rolls crushers. There are a number of suppliers of this type of equipment, predominantly coming from MMD, ThyssenKrupp, FLSmidth and P&H. The applications for these types of fully mobile IPCC are mainly limited to overburden stripping, oil sands, coal and low-moderately abrasive materials. This restriction is not a function of the mobility, but more the types of crushers installed. One of the first fully-mobile tracked sizer stations from 2002 is shown in Figure 3.

FIG 3 - Typical mobile sizer station (Sinclair Knight Merz, 2010).

Since 1994, many other manufacturers have been pursuing alternative designs. Some of the latest innovative designs include machines from Rahco, Sandvik and MMD. The Rahco Dual Truck Mobile Sizer is a development aimed at maintaining the fully mobile nature of the station, yet allowing it to be loaded directly from trucks. Sandvik’s new PF300 is another innovation that offers the ability to be loaded directly from a face-shovel, without the need for external stabilising supports. MMD have now also updated their mobile sizer station with the introduction of the low profile sizer station with slewing discharge conveyor. A review of many of the latest developments in IPCC is provided by International Mining (2009). Such mobile stations work alongside the face shovel or excavator and are integrated with a face conveyor, passing material onto intermediate conveyors for transport to a final destination. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

36

A SIMULATION APPROACH FOR THE COMPARISON OF IPCC AND TRUCK-SHOVEL MINING METHODS

The destination usually being a spreader for overburden, or a coarse ore stockpile for subsequent processing. The network of conveyors changes to meet the position of the feed point and the eventual discharge point. The movement of the mobile crushing front-ends and the changing nature of the conveyor network throughout the mine life presents a different set of operational parameters to the usual mining methods. Given that truly mobile IPCC systems are widely available for a range of applications, the assessment of their performance as a system compared to alternative approaches, ie truck shovel mining, has become a major discussion point.

Simulation techniques A variety of simulation tools are available to determine the best open pit mining method. These tools can be broken down into a number of types, namely: x fleet selection: determining the type and numbers of equipment required to meet a mining schedule; x pit optimisation: design of the pit in terms of geological constraints, grade issues and the equipment to be deployed; and x system performance: how the mining method selected will perform in terms of financial elements, power usage and GHG footprint. All the above systems are well represented in the market for truck-shovel mining and a full assessment can be undertaken to cover all of the main points. In terms of IPCC, fleet selection is a less complex issue, simply due to the small numbers of mobile crushing units and the use of a conveyer network to replace the truck haulage element for mining. For IPCC, pit optimisation still relies on the geological, geotechnical, grade and mining schedule requirements common to the truck-shovel approach. The main difference between the evaluation of IPCC and truck-shovel mining is in the assessment of the total system performance. In truck-shovel mining the mainstay of evaluations are discrete event simulation packages. These packages are essentially frameworks whereby the user can program equipment movements versus time. In doing so, it soon becomes clear if the fleet selected can meet production targets, the cost of the material movement is tolerable and whether other operational parameters may impact viability. For IPCC the total system performance is not ideally suited to simulation using discrete event simulation, as the system is essentially continuous. To address this, the authors looked to utilise a simulation platform that would be more appropriate for IPCC. The framework developed is now termed mine-process operational simulation (MPOS).

Mine-process operational simulation In seeking a more appropriate representation for IPCC based mining, the authors also wanted to understand how the more continuous nature of IPCC impacted the downstream process plants. It is well known that interruptions to material presentation to process plants can have a major impact, which can be considerably out of proportion to the duration of the feed interruption. Figure 4 shows a typical case, where once interruptions have started, it can take up to one hour for the process plant to regain its full throughput. To cover both the IPCC mining and the process plant elements, a dynamic process simulation package was selected as the basis for MPOS. Using the commercial package, MPOS was populated with dynamic models of shovels, crushers, conveyors, screens, bins, mills and other key process equipment. Supporting the measures above, MPOS can be used in the following ways to ensure that the extraction of the resource is optimised: x tracking of ore/waste from extraction through to the point of delivery; x predicting the processing characteristics and recovered value of delivered ore; x scheduling and movement of IPCC equipment and/or haul routes in response to changing orebody characteristics and plant performance; MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

37

R BEARMAN AND S MUNRO

Consistenttruck &plantoperation

Erratictruckarrivals 1TruckTips

3500

0.9 Plant 0.8 Design Rate(tph) 0.7

3000

PlantTPH

2500

1500

0.6 Average 0.5 Rate (tph) 0.4

1000

0.3

2000

0.2 500

0.1

0

0

Seconds PlantTPH

TruckTips

FIG 4 - The effect of feed interruptions on plant throughput (blue diamonds indicate truck tips).

x variable time frame analysis – real time, accelerated time, life-of-mine; and x variable level of application – order of magnitude, prefeasibility, feasibility, detailed design, brownfield engineering.

Analysis using mine-process operational simulation Where possible, a common set of inputs for the IPCC and truck-shovel systems are used by the MPOS analysis to improve the basis of comparison between the two methods. The type of inputs used to characterise the application include: x mine plan – block sequence extracted from XPAC, or similar; x fragmentation – calculated from fragmentation model; x material characteristics – strength, density; x IPCC specifics such as energy source, crusher (size/type), shovel (type/size), sizer station, belt wagons, conveyers (type, motor, length, elevation); x truck-shovel specifics, such as shovel (type/size), trucks (size), haul routes (length, elevation); and x availability – planned and unplanned downtime events by equipment item, for each system. The simulations are configured using these inputs and are performed over the desired timespan and at an appropriate level of data resolution. Increases to the complexity of the simulated system, the simulated timespan and the intervals between data measurements, increase the overall run-time of the simulation, which is performed on a standard desktop PC. MPOS can be programmed to report a wide range of simulation parameters, including: x power/energy consumption; x GHG equivalent emissions; x capital and operating costs; x utilisation of equipment, operating delays; x achieved tonnages and payloads (instantaneous and averaged); x particle size distributions; and x ore and product grades (including full size-by-grade distributions). These parameters are output into a spreadsheet format, which allows the generation of a variety of graphs for individual pieces of equipment, or for the total system. Figures 5 - 8 show some

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

38

A SIMULATION APPROACH FOR THE COMPARISON OF IPCC AND TRUCK-SHOVEL MINING METHODS

typical graphs generated from various case studies. Figure 5 shows an instantaneous loading on key conveyors in the network, along with costs and carbon dioxide equivalent emissions. MPOS continually recalculates this instantaneous state throughout the changing simulation. MobileSizerStation

Block Inventory,Mining Sequence

15,000t

AWT

10,000t

BWT 5,000t

BIF

t 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

EnergyCost Equiv.CO2

Massload Power Energy

CV01Profile

4,000t/h

Power Energy

3,000t/h

743 0.42

kW PJ

9.29 MUS$p.a. 29,570 tCO2 p.a.

538 587 0.37

t kW PJ

2,000t/h

EnergyCost Equiv.CO2

1,000t/h 0t/h 50m

300m

550m

800m

1050m 1300m 1550m 1800m 2050m 2300m

Massload Power Energy

CV02Profile

4,000t/h

8.31 MUS$p.a. 26,457 tCO2 p.a.

3,000t/h

235 220 0.15

t kW PJ

2,000t/h

EnergyCost Equiv.CO2

1,000t/h 0t/h 20m

120m

220m

320m

420m

520m

620m

720m

820m

920m

Massload Power Energy

CV03Profile

4,000t/h

3.29 MUS$p.a. 10,470 tCO2 p.a.

3,000t/h

620 1,145 0.74

t kW PJ

2,000t/h

EnergyCost Equiv.CO2

1,000t/h 0t/h 100m

600m

16.47 MUS$p.a. 52,420 tCO2 p.a.

1100m 1600m 2100m 2600m 3100m 3600m 4100m 4600m

FIG 5 - Instantaneous tonnage profiles on key conveyors, with operational calculations for power, energy, cost and equivalent carbon dioxide emissions.

Figure 6 shows a graph of energy use from a direct comparison of truck-shovel and an IPCC scenario.

20 18

Sizer

MJpertonne

16 14 12

HaulTruck

10

Conveyors

8

Sizer

6 4 2

Loader Blasting

Loader Blasting

0

TruckShovel

FullyMobileSizer

FIG 6 - Truck-shovel and in-pit crushing and conveying energy comparison graph.

The subsequent conversion from energy use to GHG emissions depends on the battery limits applied to the calculation and the source of the electrical power for the IPCC system. From alternative case studies, Figure 7 shows that depending on the cleanliness of the power generation and the battery limits, the carbon dioxide footprint can be significantly different, thus highlighting the contextual nature of IPCC suitability. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

39

R BEARMAN AND S MUNRO

6,000,000

1,800,000 1,600,000

5,000,000

1,400,000 Conveyors

4,000,000

TonnesCO2

TonnesCO2

1,200,000 1,000,000 800,000 Trucks 600,000

Trucks 3,000,000

MSS

2,000,000

Shovel

1,000,000

Conveyors

400,000 200,000

MSS Shovel

Shovel

Shovel 0

0 TruckShovel

TruckShovel

IPCC

(a)

IPCC

(b)

FIG 7 - Comparison of greenhouse gas generation for and in-pit crushing and conveying and truck-shovel methods: (A) greenhouse gas based on equipment consumption and high carbon dioxide power generation; (B) greenhouse gas including production, refining and delivery of diesel and low carbon dioxide power generation.

As MPOS captures the dynamic behaviour of the entire system, it is able to identify production benefits associated with the system configuration and equipment, as well as initial and ongoing capital and operating expenditure as time progresses. The MPOS outputs are coupled with a financial analysis to support the business case for the mining method. A comparison often examined for IPCC is the capital and operating cost profiles against those for traditional methods. It is generally of great interest as IPCC has a significantly different profile to that of truck-shovel mining. The main difference is that the truck based mining fleet can be introduced over a period of time to match the required mine production and schedule. For IPCC a majority of plant is purchased early in the life of the operation. In terms of the capital cost profile, IPCC therefore has a much higher capital spend in the early years. Once the operation is underway the operating cost associated with IPCC tends to be lower than truck shovel and as such the cost differential between the two mining methods is eroded. The rate of the erosion and the point in the mine life where the costs equalise is keenly sought. In MPOS a net present cost (NPC) calculation is undertaken for operations such as overburden stripping, where cost is the driver. Figure 8 shows a typical graph showing the NPC over time.  Cum.NPCUS$ Millions

350 300 250 200 150 100 50 0

IPCC

TruckShovel



FIG 8 - Net present cost comparison for a typical in-pit crushing and conveying and truck-shovel mining study. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

40

A SIMULATION APPROACH FOR THE COMPARISON OF IPCC AND TRUCK-SHOVEL MINING METHODS

The type of output illustrated in Figure 8 is simply an example. Overall, the range of analysis possible using MPOS can be customised to the needs of the client, which is one of the advantages of an open system that reports data to a generic spreadsheet format.

CONCLUSIONS With the IPCC mining approach being studied for an increasing number of applications, it is believed the MPOS evaluation tool is useful to engineers and those involved in the evaluation. One of the features of MPOS is that is can be customised to cover a range of IPCC scenarios, up to and including, a full blast to final product study. In addition the study can be run with a varying time base (years to seconds) and to different levels of detail, dependent of the stage of the study. It is believed that the ability to analyse IPCC to the same degree as the usual mining methods will provide a fuller picture of the total value proposition for IPCC and an improved idea of the true comparison against more traditional mining methods.

REFERENCES Aube, F, 2001. Guide to computing CO2 emissions related to energy use [online]. Available from: . Bozorgebrahimi, E, Hall, R A, Blackwell, G and Scoble, M, 2003. Economies of scale in surface mining: A case study of the impact of haul truck size on maintenance cost, SME Annual meeting, Cincinnati, Ohio. International Mining, 2009. IPCC Innovations, report, pp 8-14, June. Schroder, D, 2003. The use of in-pit crushing and conveying methods to significantly reduce transportation costs by truck, in Proceedings CoalTrans Asia, pp 1-12 (CoalTrans International). Sinclair Knight Merz, 2010. Using continuous mining equipment to develop an open cut mining operation [online]. Available from: . Zimmermann, E and Kruse, W, 2006. Mobile crushing and conveying in quarries – a chance for better and cheaper production, in Proceedings Eighth International Symposium on Continuous Surface Mining (ISCSM), pp 1-7 (RWTH Aachen University: Aachen).

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

41

Production Capacity Optimisation in the Presence of Grade Uncertainty T Elkington1,2 and F Grobler3,4 ABSTRACT Grade uncertainty can have a number of important implications for a mining operation. One of the most significant risks in not considering grade uncertainty in planning is the over or under allocation of processing or mining capacity when compared to realised grades. This could lead to wasted capital expenditure or missed opportunities to add value. This paper demonstrates a method that optimises production capacity and mining schedules in the presence of grade uncertainty. Applying this method to the case study of a nickel deposit recommended a reduction in processing capacity, an increase in mining capacity and more extensive utilisation of stockpiles than optimisations completed with traditional methods. As a result, average NPV improvements of between seven per cent and 27 per cent were gained.

INTRODUCTION Strategic mine planning involves making decisions about key drivers of project success. Much of recent research effort in the field of strategic mine planning has focused upon cut-off grade and scheduling decisions. Both of these decisions are dependent on the capacities (mining, processing and distribution) imposed on it. Production capacities therefore become a root driver of project outcomes and should receive greater focus during decision-making. Capacity is a large contributor to project capital and fixed costs. An incorrect decision on capacity can lead to misallocated financial resources. Additionally, changing capacity is difficult, expensive and takes a long time to reverse. It is therefore essential to identify the most appropriate production capacity option from the outset through appropriate analysis and informed decision-making. However, production capacity is often an arbitrary decision made based on rules of thumb (Taylor, 1978) or past experience, without being optimised. Improvements in the schedule and cut-off grade optimisation approach could be gained by introducing production capacities as a decision variable. This was demonstrated in a mixed integer programming (MIP) approach in Elkington and Durham (2010). An important outcome of this paper was the ability to fully utilise both mining and processing capacity to maximise benefit from capital expenditure. Lower than expected processing capacities were recommended from the outcomes of the model, attributed to the diminishing returns offered by processing lower grade material. Instead, the emphasis is to stockpile this material for later processing. This approach did not consider grade uncertainty. Grade uncertainty has been a rapidly advancing area of research in recent years. Dimitrakopoulos, Farrelly and Godoy (2002) demonstrated that grade uncertainty could impact on economic mine life. Dimitrakopoulos, Martinez and Ramizan (2007) suggested that grade uncertainty can influence pit design and proposed a risk based procedure to select pit designs given a number of options. Leite and Dimitrakopoulos (2009) developed a simulated annealing approach to pit design and scheduling in the presence of grade uncertainty. This work indicated that schedules can be used to ‘blend’ grade uncertainty to achieve smoother ore profiles. 1. Consultant, Snowden Mining Industry Consultants, 87 Colin Street, West Perth WA 6005. Email: [email protected] 2. Adjunct Assistant Professor, School of Civil and Resource Engineering, The University of Western Australia, 35 Stirling Highway, Crawley WA 6009. 3. Principal Consultant, Snowden Mining Industry Consultants, 87 Colin Street, West Perth WA 6005. Email: [email protected] 4. Senior Adjunct Research Fellow, Western Australian Centre of Excellence in Industrial Optimisation, Curtin University of Technology, Kent Street, Bentley WA 6102.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

43

T ELKINGTON AND F GROBLER

These papers have imposed a marginal cut-off grade restriction and focused on manipulating pushback designs as a mechanism to plan for uncertainty. Menabde et al (2007) demonstrated that a variable cut-off grade is a mechanism to achieve more robust schedules in the presence of grade uncertainty, while fixing pushback design. This paper integrates the approaches of Menabde et al (2007) and Elkington and Durham (2010). The approach considers a fixed pushback design, but provides the ability to adjust production capacities, raise cut-off grade and stockpile material as a mechanism to manage risk and sometimes benefit from grade uncertainty. A case study is used to demonstrate the outcomes of this approach.

METHOD The method is an extension on the MIP formulation provided in Elkington and Durham (2010). A visual representation of the process is shown in Figure 1.

FIG 1 - Process schematic (Elkington and Durham, 2010).

A bench is a vertical unit of advance. A pit shell is the mining outline of an open pit that maximises undiscounted cash flows for a given set of slope constraints, and revenue and cost parameters. A parcel is an aggregation of blocks which share common bench b and incremental pit shell h (ie blocks in pit shell h but not in pit shell h-1) properties. A pushback is an incremental expansion of a pit outline, formed by an aggregation of parcels to generate a practical mining width. A panel is the set of one or more parcels allocated to the same pushback (two in the above example), located on the same bench. Each parcel within a panel is assumed to be extracted at the same rate. Finally, every block within each panel is allocated to a grade group. It is assumed that material in the same grade group, in the same panel, will be sent to the same destination(s) in the same proportion(s). The properties of the blocks within each grade group in each panel are aggregated and averaged as appropriate.

Concept This paper does not propose a new method of ultimate pit and pushback optimisation. The optimisation assumes a fixed set of pushbacks, previously determined by appropriate methods. The scope of the optimisation is in scheduling, cut-off grade, stockpiling and mining and processing capacity. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

44

PRODUCTION CAPACITY OPTIMISATION IN THE PRESENCE OF GRADE UNCERTAINTY

The unique aspect of this method is the incorporation of grade uncertainty within the schedule optimisation, where production capacity (mining and processing) is treated as a decision variable. Grade uncertainty is modelled by multiple grade realisations generated through conditional simulation. Each realisation has its own unique tonnes and grade properties for each grade group. This enables each realisation to have a unique processing, cut-off grade and stockpiling schedule. The freedom for each of the grade realisations is constrained by a common mining schedule and common processing constraints. Thus, the optimal solution represents the mining schedule and processing capacity that maximises the average of the NPVs of schedules for all grade realisations. The problem is formulated as a MIP and solved with CPLEX software.

Key assumptions and limitations The key assumption made in this formulation is that companies align their strategy with maximising the probability-weighted average NPV, when considering multiple equally probable grade realisations. The optimised strategy may be different when acting upon another objective, or a combination of objectives. This section outlines the key assumptions associated with optimisation of each aspect of the mine plan, defining areas where the method is most applicable and the limitations introduced by these assumptions.

Grade uncertainty Conditional simulation is used to model grade uncertainty in this model. All grade realisations have been re-blocked to the parent block size. It is assumed that while none of the grade realisations will represent the ultimate ‘truth’ for the resource, each realisation estimates the variability and distribution characteristics of the truth. Viewed individually, the grade realisations are unsuitable for design purposes due to their speculative nature. However, when many realisations are considered together they give an indication of the range of outcomes expected from the deposit.

Ultimate pit and pushback selection Pushback selection is determined prior to optimisation within this process using any valid technique. This method could be improved by incorporating ultimate pit and pushback selection into the optimisation.

Scheduling The schedule is defined by the portion of each panel that is mined in each period. It is assumed that each block within a panel is extracted at the same rate, regardless of the grade realisation. It is assumed that mining will not change order or speed depending on the grade realised. Instead, if grade is higher than expected, surplus material is stockpiled. If grade is lower than expected, there will be a shortfall in the processing schedule (unless there is material on the stockpile for the grade realisation that can be drawn from at that time). There are several limitations to this type of scheduling approach. Progression of mining within a panel is not optimised or planned. Additionally, the panel sequencing constraints are applied strictly between time periods, but not within them. If time periods are too large (ie when there are few panels mined in each time period) these two issues may combine to produce impractical schedules. This must be managed with careful schedule verification and appropriate time period length selection.

Cut-off grade Cut-off grade is not applied explicitly in this method. Cut-off grade is implied by the material contained within the grade groups selected for processing. In a single element operation with no blending constraints, the cut-off grade can be estimated as the lowest grade block contained within material selected for each process. If these conditions are not present, the definition of cut-off grade is multi-dimensional. The accuracy of the cut-off grade estimation is dependent on the number of grade groups applied. An increased number of grade groups will likely result in tighter bounds for each grade group and a more accurate estimation of cut-off grade. This must be reconciled with increased problem size associated with more grade groups and resultant increases in solution time. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

45

T ELKINGTON AND F GROBLER

When multiple grade realisations are incorporated into the optimisation, each realisation will have its own unique cut-off grade schedule. This evolves naturally through the assumption that the processing plant will take the highest grade material possible within the material mined in that period. The remainder is either stockpiled or treated as waste.

Stockpiling From a modelling perspective, stockpiles apply a time delay between mining a panel and when material within the panel is processed. It is assumed that material can be removed from the stockpile in any period after it has been placed. There are no explicit restrictions on the sequence of stockpile depletion. This method assumes the mining operation is able to carry out such a plan and have control over the movement of material on stockpiles. This can be achieved by either grouping stockpiled material into grade bins, or by designing stockpiles based on the optimisation schedule, ie place material that is removed from the stockpile in the same period in the same bin. It is acknowledged that this may not be possible for all operations and high stockpiling costs have been assumed in the case study to reflect the difficulty of stockpiling in such a manner. Each grade realisation will have its own stockpile addition/depletion schedule, closely linked to the cut-off grade schedule. If grade is higher than expected, more material will be stockpiled. If grade is lower than expected, less material will be stockpiled. This effect becomes important in the final periods of the project when the stockpile is depleted.

Production capacity Production capacities (mining and processing) are included in the formulation as free decision variables. Mining capacity covers only material mined from the pit. Processing capacity includes both material processed directly from the mine and material reclaimed from the stockpile. Each unit of mining and processing capacity must be purchased at a given unit cost. It is assumed that the relationship between production capacity and capital cost is linear, although it is acknowledged that this may not be accurate in all cases. It is possible to also model staged increments of capital cost with the addition of more integer variables. The addition of these integer variables would also allow fixed costs to be modelled (not accommodated in this formulation), but also increase solution times. This is an area for future development. Production capacities are held constant over the life of the operation, ie they are established from the outset of the project5. It is assumed that capacity cannot be sold. Therefore, each unit of additional capacity must be utilised profitably in present value terms. Capital costs are applied in year zero.

Mixed integer programming formulation Decision variables There are seven main types of decision variables used in this model. All variables are continuous between zero and one unless stated otherwise. mcbt portion of bench b in pushback c to be mined in period t pcbgst portion of grade group g of grade realisation s within pushback c and bench b to be processed in period t xcbsgt portion of grade group g of grade realisation s within pushback c and bench b to be added to the stockpile in period t ycbsgt portion of grade group g of grade realisation s within pushback c and bench b to be reclaimed from the stockpile in period t dcbt (binary) indicating whether bench b in pushback c can be extracted in period t (1) or not (0) mc (non-negative continuous) mining capacity pc (non-negative continuous) processing capacity

5. The model can be simply adjusted to allow staged capacity increases over time. See Elkington and Durham (2010) for details.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

46

PRODUCTION CAPACITY OPTIMISATION IN THE PRESENCE OF GRADE UNCERTAINTY

Coefficients N Mcbt Pcbsgt Xcbsgt Ycbsgt

MC PC Ocbsg Gcbsg Qcb Scsg Bctmax

number of grade realisations discounted cost of mining bench b in pushback c in period t discounted net revenue (revenue minus processing cost) of processing grade group g of grade realisation s within pushback c/bench b in period t discounted cost of stockpiling grade group g of grade realisation s within pushback c/bench b in period t discounted net revenue (revenue minus processing cost and reclaiming cost) of reclaiming from the stockpile and processing grade group g of grade realisation s within pushback c/ bench b in period t discounted cost of purchasing an additional unit of mining capacity discounted cost of purchasing an additional unit of processing capacity ore tonnage in grade group g of grade realisation s within pushback c in bench b average available grade in grade group g of grade realisation s of pushback c in bench b total tonnage of pushback c in bench b total recovered product in grade group g of grade realisation s of pushback c in bench b maximum number of benches to be completed within pushback c in period t

Objective function The objective of the optimisation model is to maximise the average net present value (NPV) reported by each of the grade realisations (Equation 1). This incorporates an average of the discounted net revenue from material that is directly processed (Pcbsgt) and processed from the stockpile (Ycbsgt), as well as costs from material added to a stockpile (Xcbsgt), for each grade realisation. Mining cost (Mcbt) and capital costs associated with mining (MC) and processing capacity (PC) purchased, are common to all grade realisations and are thus do not need to be averaged. Maximise: 1 / Pp + Yy - Xx - / Mcbt mcbt - MCmc - PCmc E N ; c,b,s, g ,t c,b, t

(1)

Constraints Equations 2 and 3 relate to the capacity of mining and processing in each period, respectively. These capacities are established from the outset and are assumed to be constant over the life of the operation. Each additional unit of capacity incurs a capital establishment cost, reflected in the objective function.

/Qcb mcbt # mc

6t

(2)

c,b

/

c,b, g

6s, t

Ocbsg ^ pcbsgt + ycbsgth # pc

(3)

Equation 4 enforces a constraint on the maximum number of benches completed within a pushback in each period.

/ mcbt # Bctmax

6c, t

b

(4)

It is important that material flow equilibrium is achieved throughout the optimisation from period to period. The amount of material processed directly and added to the stockpile from a panel progresses at the same rate as the mining of the panel (Equation 5). Material not directly processed or stockpiled is assumed to be sent to a universal waste dump. Secondly, material cannot be removed from the stockpile until it has been placed there (Equation 6). pcbsgt + xcbsgt - mcbt # 0

6c,b, s, g,t

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

(5)

47

T ELKINGTON AND F GROBLER

t

/^ycbsgq - xcbsgqh # 0

(6)

6c, b, s, g, t

q=1

In order to optimise the sequencing of panels, a series of constraints are used to ensure that practical scheduling requirements are met. Equation 7 enables mining in a panel only when it becomes available. This availability is dependent on the bench directly above in the same pushback (Equation 8), as well as the same bench in the previous pushback (Equation 9) being completed. Equation 8 assumes that bench numbering increases with depth. Finally, Equation 10 ensures that each panel is only mined once. t

/ mcba - dcbt # 0

6c, b, t

(7)

6c, b, t

(8)

6c, b, t

(9)

a=1

dcbt -

t

/ mc(b - 1)a # 0

a=1

dcbt -

t

/ m(c-1)ba # 0

a=1

/ mcbt # 1

(10)

6c, b

t

CASE STUDY A hypothetical nickel deposit is used as a case study. Ten grade realisations were generated using conditional simulation (CS) techniques. Three of these grade realisations are shown in Figure 2. An ordinary kriged (OK) model was also developed and shown alongside a model which is the average of the grade realisations from the conditional simulation (Figure 3)

FIG 2 - Section view of three grade realisations (coloured by grade). (A)

(B)

FIG 3 - (A) Section view of the kriged; and (b) average grade models (coloured by grade). MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

48

PRODUCTION CAPACITY OPTIMISATION IN THE PRESENCE OF GRADE UNCERTAINTY

The pit was designed with three pushbacks, developed using conventional pit optimisation and general mine planning software. The parameters applied for the optimisation are as follows: x price = $18 000/t Ni, x discount rate = ten per cent, x metallurgical recovery = 70 per cent, x processing cost = $20/t, x mining cost = $(0.700 + 0.002 × depth)/t, x stockpile addition cost = $1/t, x stockpile reclamation cost = $1/t, x mining capital cost = $(mining capacity) × 2, x processing capital cost = $(processing capacity) × 25, and x maximum vertical advance = 120 m per year.

Production capacity optimisation without grade uncertainty A base case schedule was completed based on the OK model for comparative purposes. In this case (referred to as ‘kriged marginal’, or ‘KM’), stockpiles were excluded and a marginal cut-off grade was enforced. This means that any grade group that earns more revenue than its costs to process and sell will be processed. This case was selected because, if production capacity optimisation is carried out at all (as opposed to applying rules of thumb), it would usually be completed on this basis. Implementation of this within the MIP formulation required altering Equation 5 to Equation 11 for such grade groups. pcbsgt - mcbt = 0

6c, b, s, g, t

(11)

The maximum NPV for this was found to be $66 million with a processing capacity of 1.94 Mta and a mining capacity of 18.79 Mta. The schedule for this case is shown in Figure 4.

FIG 4 - Production schedule for kriged marginal case.

A more sophisticated production capacity optimisation might compare scenarios without the marginal cut-off grade constraint, thus allowing cut-off grades to be optimised and stockpiles to be used. However, the optimisation would typically still be completed with inputs from the OK resource model. This case will be referred to as ‘kriged optimisation’, or ‘KO’. For this case, the NPV increases to $77 million; a 17 per cent improvement over the KM result. The improvement is derived from a much lower processing capacity of 1.46 Mta and a similar mining capacity of 19.04 Mta as demonstrated by the KM case. The schedule is shown in Figure 5. The drop in processing capacity reduces the capital cost by around $12 million, only processing higher grade material and allowing marginal material to be stockpiled for later processing without eroding value. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

49

T ELKINGTON AND F GROBLER

FIG 5 - Production schedule for kriged optimisation case.

Production capacity optimisation with grade uncertainty The final case considered, which is the approach proposed in this paper, optimised the schedule subject to the conditionally simulated grade realisations without a marginal cut-off grade restriction. This case will be referred to as ‘conditional simulation optimisation’, or ‘CSO’. This optimisation found the optimal processing capacity to be similar to the KO case at 1.49 Mta, but the mining capacity to be significantly higher at 27.36 Mta. The average NPV for this scenario is $238 million. This is a significant increase over the KM and KO results. Part of this increase is attributable to the greater variability in the grade realisations developed with conditional simulation, with generally higher grades over smaller volumes. In order to provide a fair comparison to the KM and KO cases, the KM and KO cases must also be reported against the same grade realisations (having previously been considered with the OK model). To do this the previously determined mining schedules and processing capacity are fixed and the optimisation is repeated using the same grade realisations. When this is done, the KM case reports an average NPV of $188 million and the KO reports $222 million (Table 1). Thus, the CSO approach represents an improvement of between 7 per cent and 27 per cent when considering grade uncertainty. TABLE 1 Summary of key results.

Ave NPV ($ million)

Kriged marginal

Kriged optimisation

Conditional simulation optimisation

188

222

238

Processing capacity (Mtpa)

1.94

1.46

1.49

Mining capacity (Mtpa)

18.79

19.04

27.36

Max stockpile size (Mt)

-

1.48

3.26

The source of this value improvement becomes apparent when comparing the ore processing schedules of each of the optimisations, reported for all ten grade realisations (Figure 6). Whilst the KM case has the highest processing capacity (1.94 Mta), it is unable to fill this capacity in any period; often struggling to achieve 75 per cent utilisation. Effectively, 500 ktpa of processing capacity (or around $12 million of capital cost) is wasted in this plan. This shortfall occurs because of the smoothing effects present in the OK model. The smoothing creates more mid-grade material over larger volumes, thus suggesting a higher processing capacity. When infill drilling is completed to define the resource at the level of selectivity demonstrated in the conditional simulated realisations, higher and lower grades within this resource will be defined. The lower grade portion can be treated as waste. Therefore the grade MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

50

PRODUCTION CAPACITY OPTIMISATION IN THE PRESENCE OF GRADE UNCERTAINTY

FIG 6 - Ore schedule comparison.

realisations from conditional simulation will tend to produce less tonnes, but at higher grade than the kriged model (Figure 7)6. So, while the value increases for the KM model with the conditionally simulated realisations due to higher grades, some of the installed processing capacity is wasted. The KO case performs considerably better than the KM model because it recommends a lower processing capacity. This was done on the basis that marginal material would not justify additional capacity and could be deferred until the end of the mine life by stockpiling. This high grading and stockpiling recommendation provids ‘insulation’ against the lower tonnages realised and leads to a schedule that could fill the process plant in most periods after year 3.

FIG 7 - Tonnage and grade comparison between ordinary kriged and conditionally simulated grade realisations.

6. The difference between tonnes and grade of smoothed and more variable grade models depends on the amount of drilling data available, block size, and cut-off grade applied. The

more drilling, the more similar they will be. The relationship shown in this paper is due to a combination of low drilling density and low cut-off grade. At higher cut-off grades the more variable model will tend to report higher grade and higher tonnage than the smoothed model. This is demonstrated in Elkington and Durham (2010). MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

51

T ELKINGTON AND F GROBLER

The CSO case was able to recognise the likely grade distribution and plan for it, by installing additional mining capacity and using stockpiles as a buffer7. This capacity ensured that the processing plant could be kept full for all periods (with the exception of one realisation in year 2). By keeping the process plant full in the presence of uncertainty, revenue potential is maximised. It can also be observed that the CSO case postponed the effects of uncertainty until the end of the mine life, where there is uncertainty in the size of the stockpile to be depleted at the end of the project (Figure 8). If the nickel grade is worse than expected, this manifests in a shorter stockpile depletion period, rather than affecting the processing schedule at the beginning of the schedule when high cash flows are vital.

FIG 8 - Stockpile size schedule.

The increased mining capacity of the CSO case, results in bringing forward high-grade material by a year in comparison with the KO case and four years in the KM case (Figure 9). Thus, revenues are brought forward with resultant benefits in NPV.

FIG 9 - Average processed grade schedule.

From a production forecasting perspective, this type of analysis is useful in defining the uncertainty in metal production by period (Figure 10). While the CSO case reports a very high grade in year 4, there is a large amount of uncertainty in the grade achieved. 7. This optimisation recommends stockpiles as a method to deal with grade uncertainty for this case. This is despite the application of high cost penalties on stockpiling addition and

reclamation. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

52

PRODUCTION CAPACITY OPTIMISATION IN THE PRESENCE OF GRADE UNCERTAINTY

FIG 10 - Recovered metal schedule.

CONCLUSIONS This paper has demonstrated a method for optimising production capacity in the presence of grade uncertainty. The results of the case study indicated that pursuing lower processing capacity, higher mining capacity and greater utilisation of stockpiles led to an improvement of between seven per cent and 27 per cent in the average realised NPV when compared to conventional approaches to production capacity without considering grade uncertainty. Further research is ongoing to identify whether this outcome is common for other types of projects.

ACKOWLEDGEMENTS The authors would like to thank Matt Cotterell from Snowden for reviewing the paper.

REFERENCES Dimitrakopoulos, R, Farrelly, C and Godoy, M, 2002. Moving forward from traditional optimization: Grade uncertainty and risk effects in open-pit design, Transactions of the Institutions of Mining and Metallurgy, Mining Technology, 111:A82-A88. Dimitrakopoulos, R, Martinez, L S and Ramazan, S, 2007. A maximum upside/minimum downside approach to the traditional optimisationof open pit mine design, Journal of Mining Science, 43:73-82. Elkington, T and Durham, R, 2010. Integrated open pit pushback selection and production capacity optimization, Journal of Mining Science (accepted). Leite, A and Dimitrakopoulos, R, 2007. A stochastic optimisation model for open pit mine planning: Application and risk analysis at a copper deposit, Transactions of the Institutions of Mining and Metallurgy, Mining Technology, 116:A109-A118. Menabde, M, Froyland, G, Stone, P and Yeates, G, 2007. Mining schedule optimisation for conditionally simulated orebodies, in Proceedings Orebody Modelling and Strategic Mine Planning, second edition (ed: R Dimitrakopoulos), pp 379-384 (The Australasian Institute of Mining and Metallurgy: Melbourne). Taylor, H, 1978. Valuation and feasibility studies, Mineral Industry Costs (ed: J Hoskins and W Green), Northwest Mining Association, Washington.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

53

Robust versus Flexible Open Pit Mine Design B Groeneveld1, E Topal2 and B Leenders3 ABSTRACT Strategic planning in mining is an important value accretive process. One of the most important aspects during the planning phase is determining the correct mine and plant design. The traditional mine site design (mine, plant, stockpiles, dumps) develops a fixed system for one set of conditions or expected values. A robust design is a fixed system that is designed to deal with a large range of conditions without changing the system design. A flexible design changes the system dynamically in response to changes in uncertainties. It is hypothesised that a robust design generates less value than a flexible design as it fails to realise the value of actively managing the operation. One of the most critical aspects in creating strategic value from a project is embedded in the design of the system. A flexible design has components that allow the design of the system to change over time. An operational plan is proposed as a hybrid of a robust and flexible design. This paper seeks to compare these different design methodologies. An analysis of the same case study is undertaken to compare these approaches, showing an a traditional net present value (NPV) of 1678 M, robust design NPV of 1787 M and an operational plan NPV of 1867 M – an increase of 170 M.

INTRODUCTION Decision-making in mining operations can take many years because the capital cost is large and the payback period lengthy. During the evaluation period many uncertainties can unfold and multiple economic cycles may occur. Making decisions based on single point estimates of the future can lead to premature foreclosure of an operation as the company has limited flexibility to react to changing conditions. Improving a company’s success is obtained by having the opportunity to capture value by designing for changing circumstances. The dominant decision-making technique used across the mining industry is the use of discounted cash flow analysis (DCF). The traditional DCF model includes a fixed mine design (production rates, plant capacities, processing routes, etc), a fixed schedule(s) based on the fixed design and probably some ability to undertake sensitivity analysis. Sensitivity analysis is generally carried out by applying a percentage change to a key input variable to determine the effect on net present value (NPV). Whilst this may capture, to some degree, the impact of key variables on the value of an operation, it will overstate the negative impact and grossly understate the positive impact. Why If an operation is actively managed then in situations where things turn ‘bad’, managers can limit production, even to the extent of stopping production (Monita, 2009 Le May, 2009), thus reducing the downside risk. On the contrary, if things turn ‘good’, managers can choose to de-bottleneck or expand the operation (Trounson, 2007, 2008 Chamber, 2008), in turn increasing the expected value of the project. In a traditional DCF analysis, this flexibility is not considered or valued. Consequently a DCF will provide an ‘answer’ which is vastly different from the true potential value of the project. Whilst a decision maker may gain some confidence from a traditional DCF analysis, generally they will miss opportunities to design for change, hence reducing the potential value of the project from the outset. Building flexibility into an operation provides the company with an ability to quickly respond to change however this flexibility comes at a cost. For example, an operation building a crusher to feed a processing plant with an initial plan to produce at 6 Mta, will design the crusher to meet this capacity. One year into the operation, the sale price of the product doubles while all other variables hold. In this environment it is considered favourable to expand the operation however a 1. GAusIMM, PhD Candidate, Mining Engineering Department, Western Australia School of Mines, Curtin University of Technology and Underground Mining Engineer, Telfer Gold Mine.

Email: [email protected] 2. MAusIMM, Associate Professor and Head of Mining Engineering Department, Western Australia School of Mines, Curtin University of Technology, Kalgoorlie WA 6433.

Email: [email protected] 3. MAusIMM, Principal Advisor – Strategic Development, Rio Tinto Iron Ore, Perth WA 6000. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

55

B GROENEVELD, E TOPAL AND B LEENDERS

new crusher is required to achieve the required production rate at a significant capital cost and time to build. Alternatively, had the original design included flexibility to easily expand production to 8 Mta (for example by increase the size of the foundations and footings to allow physical room for larger equipment), then production could be increased at a lower cost, shorter time and at a lower production impact. Current decision-making tools do not give the planner the ability to justify this upfront flexibility, however advances in a technique broadly know as real options ‘in’ projects is addressing this gap (Wang and de Neufville, 2005, 2006 Cardin, de Neufville and Kazakidis, 2008 de Neufville 2005, 2006 Groeneveld, Topal and Leenders, 2009). Real options ‘in’ projects (ROIP) have been proposed as a methodology to justify an increase in system flexibility under uncertainty. This method is located midway between financial real options analysis (which does not deal with engineering system flexibility) and traditional engineering approaches (which do not deal with financial flexibility). An analysis done using ROIP methods determines the effect of allowing the underlying system to adjust in response to change. Having this information allows the decision maker to make a more informed decision on whether the additional cost is warranted. Previous papers using ROIP methods on different problems clearly show the value of this technique (Cardin, de Neufville and Kazakidis, 2008 Groeneveld, Topal and Leenders, 2009, 2010). Cardin, de Neufville and Kazakidis (2008) implemented this technique for mining projects with a Chilean mine in the ‘Cluster Toki’ region. Groeneveld, Topal and Leenders (2009) outlined a methodology for a flexible mine design that showed how design options could dynamically be included to represent mining, plant and port constraints. The associated model was processed multiple times, with each run representing a different ‘state of the world’. Value-at-risk graphs showed a significant increase in value available by using a flexible design methodology when compared to a fixed base design. The base case for comparing the flexible design was based on a traditional deterministic solution for one fixed set of uncertainties. This paper extends this concept by preparing a single robust design derived using multiple sets of uncertainties. This allows for an even better base case design from which to make the comparison point for justifying the inclusion of flexibility. An important clarification is differentiating between a robust design, as presented here and what is known as a reliable design. These two aspects are very different. A reliable design is one that delivers high utilisation for a piece of equipment or project. A robust design is one that handles changing uncertainties best performs best under all ‘states-of-the-world’. Reliable designs come from the field of reliability engineering (Bergman, Mare and Svensson, 2009) which has a different objective to real option ‘in’ projects. In this paper, we develop a methodology for determining a robust design under all potential ‘states-of-the-world’. This paper outlines a methodology to determine a robust design under uncertainty, using mixed integer programming (MIP) and Monte Carlo simulation (MCS). The methodology involves determining a single design based on numerous design options (mine, stockpile, plant and port) and multiple uncertainties (price, capital cost, operating cost, recoveries and utilisation). This design is compared with the results of a flexible design produced using previously developed models (Groeneveld, Topal and Leenders, 2009). An application of this methodology to a copper-gold deposit will be undertaken to show the differences in the two methodologies and demonstrate the need for a combination of both analysis techniques.

METHODOLOGY A methodology to evaluate flexibility in strategic mine design in order to determine beneficial options to execute is proposed by using a combination of MCS and MIP. Uncertainties (or stochastic parameters) are simulated using MCS to generate inputs to a MIP model. The MIP model allows for ‘go’ or ‘no go’ decisions to be modelled for the optimal execution timing under a set of uncertainties.

Design options Four categories of options are available in the initial design phase that are incorporated dynamically into the model. These four categories are 1. mine options, 2. preprocessing stockpile options, 3. processing plants options, and 4. capacity constraint options. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

56

ROBUST VERSUS FLEXIBLE OPEN PIT MINE DESIGN

The objective is to determine the types of options used and the timing of their execution. Examining this will allow the decision maker to determine the best set of options to incorporate – when and at what capacity – in a mine plan. The four types of options have different characteristics which need to be explained further. Mine options represent the physical extraction capacity that is required to move material from the ground. This constraint may be an annual tonnage constraint or an effective flat haul constraint which considers the time required to move material (useful in highly variable haul distance scenarios). Preprocessing stockpiles are stores of material after extraction from the ground, either for long term low grade scenarios, fluctuating demand scenarios or for waste material storage. Processing plant options represent the physical and/or chemical process that is undertaken to ‘recover’ ore from the gangue material. Processing plants may contain multiple different circuits which the material may pass through. These circuits may have different beneficiation characteristics. Capacity constraint options represent physical constraints which may need to be modelled in the design. These may represent attributes such as port capacity, loading facilities, crusher capacities or conveyor capacities. Any of these design options may be incorporated in the network.

Resource representation The representation of the resource in the model is carried out by parcels of material. A parcel of material can be defined as a quantity of material with an average grade determined by the weighted average of grade bins contained within the parcel. A parcel may be made up of one or more grade bins. A grade bin represents a quantity of material at a specified grade. These grade bins provide a higher resolution of data to the model, whilst minimising the number of integer variables needed to provide this information. These parcels are designed to represent a physical constraint on the resource, such that they must be fully mined before mining a parcel lower in the physical sequence.

Flow paths A flexible mine design is created based on a set of options that are incorporated in the model through a network structure. Different routes through the network are termed ‘flow paths’. A flow path is a singular route through the network that material could travel along to explain this concept further consider Figure 1. Examples of flow paths in Figure 1 include x the path from the resource (R) to mine 1 (M1) to stockpile 1 (S1) to plant 1 (P1) through circuit 1 (C1) to product A which would be RM1S1P1C1A, x the path from the resource (R) to mine 1 (M1) to waste stockpile 1 (W1) which would be RM1W1,

FIG 1 - Example design option network showing numerous flow paths. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

57

B GROENEVELD, E TOPAL AND B LEENDERS

x the path from the resource (R) to mine 1 (M1) to stockpile 1 (S1) to plant 4 (P4) through circuit 2 (C2) to product B (B) which would be RM1S1P4C2B, and x the path from the resource to mine 3 (M3) to stockpile 1 (S1) to plant 3 (P3) through circuit 1 (C1) to product A (A) which would be RM3S1P3C1A. This is only a small number of the potential paths through the network, in reality there are numerous flow paths. An important aspect of the model formulation is that unique mine designs can be generated. That is capacity constraints can be incorporated anywhere along the network, multiple processing plants/ routes can be included and products can be generated at any point in the network (helping to model options of selling the ore at the mine gate).

Stockpiling Stockpiling is used in mine operations for many reasons including blending of material, storage of excess mine production, storage of waste material and storage of low grade ore for future production. When material is stockpiled the grade and tonnage of the material is known. However, as the material is mixed on the stockpile the grade and the tonnage become unknown. Since the quantity of material in the stockpile is unknown prior to the optimisation, this gives rise to a nonlinear constraint. To solve this problem virtual grade bins are created in the stockpile. These grade bins have a maximum and minimum grade of material which can enter the bin. The model assumes taking the average grade from the bin. Many grade bins can be created without adversely affecting the performance of the model which limits the averaging effect. When removing material the average grade is taken from the grade bin or alternatively, the maximum or minimum grade limit of the bin can be used.

Stochastic parameters The model incorporates uncertainty around the input parameters by Monte Carlo simulation. Each simulation of these values represents a ‘state of the world’ that is equally probable in the future. Various parameters can be incorporated in the model including price, capital cost, operating cost, equipment utilisation, recovery and time to build. Running a set of simulations is intended to give a representative sample of the future ‘states-of-the-world’.

DESIGN MODELS Three different models have been proposed to determine a flexible design, a robust design and an operational design. All models use Monte Carlo simulation and mixed integer programming (MIP) techniques to determine a system design. The fundamental difference between the models is that under a robust design multiple ‘states-of-the-world’ are considered together, whilst a flexible design considers just one ‘state-of-the-world’ at a time. An operational design is developed by determining a fixed design for the first couple of periods and having a flexible design after this period. The flexible design model has been published previously in (Groeneveld, Topal and Leenders, 2009 Groeneveld and Topal, 2010). The robust model is outlined in this paper and the operational plan is a new hybrid of these two models to be published at a later date.

Flexible design Groeneveld, Topal and Leenders (2009) and Groeneveld and Topal (2010) have previously outlined a new methodology for undertaking flexible mine design. The basis of these models was to optimise a design for a given single ‘state-of-the-world’. This was achieved by dynamically including mining, capacity constraint and plant design options in the system through a MIP model. Monte Carlo simulation was used to generate the different ‘states-of-the-world’. This methodology assumes that a decision maker makes optimal decisions based on the knowledge of all states of the project over time (ie what price and costs occurred over time). In reality, forecasting the final state of a project is always difficult. The proposed methodology provides information and insight that can be used by the decision maker in conjunction with other tools to make timely, informed and value adding decisions.

Robust design A new design methodology is proposed in this paper that seeks to develop a design that is robust for all ‘states-of-the-world’ hopefully generating a more realistic solution. It uses the same concepts and assumption developed in the flexible design model. A robust design is achieved by solving one ‘large’ mixed integer programming model that generates one design from multiple possible options, but MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

58

ROBUST VERSUS FLEXIBLE OPEN PIT MINE DESIGN

uses a different schedule for each ‘state-of-the-world’ given a fixed extraction sequence. In essence, this design is the one which handles a range of conditions best out of all possible designs.

Model formulation The developed MIP model optimises the system design for a risk neutral investor for the simulated ‘states-of-the-world’. Each design option can impact capital commitment, revenue generated and operating expenses. The optimisation process seeks to determine the design with the highest weighted net present value for the given financial and technical conditions. Within each pushback further grade bins are defined which provide further detail on the resource model. An outline of the mathematical formulation is provided below.

Notation Indices b c d e f g j k l m n p r t y T

a grade bin of material within a pushback circuit within a processing plant product type dependent options flow path of material through the design network grade element of material within a resource bin of material within a stockpile, this bin will have a maximum and minimum grade of material which can enter component of a product design options mining options within the set of design options simulated ‘state-of-the-world’ pushback of material required rate of return on the project time period step (periods do not need to be equal) tolerance factor for the deviation of the mining of a bin within pushback maximum number of time periods evaluated

Parameters Cl,t,n DT Dd,k,t,n Dl,t,n Ep,b,l,n FDl,t,n Fl,t,n GLg,d,k GLj GUg,d,k GUj Gp,b Gg,k,c,n Gg,k,s,j Hl,t,n Kl,t,n

the capital cost of option l in time t for trial n the lag time between these relationships (ie build option two, three periods after option one) the capacity of product d component k in time t for trial n the disposal cost of option l in time t for trial n the EFH required to move pushback p bin b to location l for trial n the fixed cost saved from not operating option l from time t for trial n to the end of the project life T the fixed cost of operating option l from time t for trial n to the end of the project life T the lower grade limit of grade g product d component k the lower grade limit of bin j the upper grade limit of grade g product d component k the upper grade limit of bin j the grade of pushback p bin b the grade g to component k through circuit c for trial n the calculated average, maximum or minimum metal units of grade g for component k in stockpile s in bin j the EFH limit on option l in time t for trial n the capacity of option l in time t for trial n

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

59

B GROENEVELD, E TOPAL AND B LEENDERS

Kl,c,t,n Ks,t,n Mp,b,l,t,n Pd,k,t,n Rk,c Rk,s,j Rp,b Rp Rp+1 Vl,t,n

the capacity of option l circuit c in time t for trial n the stockpile capacity of stockpile s in time t for trial n the mining cost from pushback p to bin b through mine option l in time t for trial n the sale price of product d component k in time t for trial n (in /metal unit) the recovery of material component k through circuit c the calculated average, maximum or minimum recovery for all material of component k in stockpile s bin j the available resource of pushback p bin b the available resource of pushback p the available resource of the successor pushback p + 1 the variable cost of option l in time t for trial n

Variables Gg,k,f,t,n IDl,t Sd,k,t,n XIs,j,t,n Xl,c,t,n Xl,t,n XRk,f,t,n XRk,f,t,n XOs,j,t,n Xp,b,f,l,t,n Xp,b,f,t,n Xp,b,t,n Xp,t,n Xp+1,b,t,n Xs,j,f,t,n Ye,t

the metal units of grade g produced from component k through flow path f in time t for trial n 0, if Option l is not disposed in time t ) 3 otherwise number of times disposed. the sale quantity (tonnage or metal units) of product d component k in time t for trial n the flow in from stockpile s bin j in time t for trial n the tonnage processed through plant option l circuit c in time t for trial n the tonnage through option l in time t for trial n the recovered tonnage to component k through flow path f in time t for trial n the recovered tonnage to component k through flow path f in time t for trial n the flow out from stockpile s bin j in time t for trial n the tonnage from pushback p bin b through flow path f which contains mine option l in time t for trial n the tonnage from pushback p bin b through flow path f in time t for trial n the tonnage mined from pushback p bin b in time t for trial n the tonnage mined from pushback p in time t for trial n the tonnage mined from the successor pushback p + 1 bin b in time t for trial n the tonnage of material sent from stockpile s bin j through flow path f in time t for trial n the dependent option e of Yl,t 1, if Option l is executed in time t ) 3 0, otherwise 1, if pushback p is fully mined in time t ) 3 0, otherwise

Yl,t Yp,t

Formulation Objective function The objective function seeks to maximise the equally weighted before tax net present value (NPV) for all simulated ‘states-of-the-world’ N T 1 = D,/K / 1e/ t n 1 N t 1 ^1 + rh d 1, k L

l

1

L

/ Cl, t, n Yl, t - / Fl, t Yl, t 1

l

1

P, B, L

L

Pd, k, t, n Sd, k, t, n - / Vl, t, n Xl, t, n l

L

l

/

1 t!1

Dl, t, nIDl, t +

1

L

l

/

1 t!1

p

1, b

/

1, l

1 ; lef ; lem

M p, b, l, t, nX p, b, f, t, n -

FDl, t, nIDl, t G o

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

60

ROBUST VERSUS FLEXIBLE OPEN PIT MINE DESIGN

The constraints in the model can be divided into five categories production, mining, stockpiling, processing and port constraints.

Production constraints Resource constraint This constraint makes sure the total amount of material extracted from a mining pit has an upper bound based on the resource. This constraint is applied at a pushback and bin level in the model T

t

/ X p, b, t, n - R p, b # c

6p, b, n

1

Sequencing constraint one This constraint in conjunction with the next constraint forces the binary value to be one in the period the pushback is fully mined. This then allows the model to mine any successor pushback B, t

/

b

1,tt

1

X p, b, tt, n $ R p * Yp, t

6p, t, n

Sequencing constraint two This constraint ensures that a pushback’s predecessor is mined before the successor is mined t

X p + 1, b, t, n # R p + 1 * / Yp, tt, n tt

6p, t, n

1

Set packing constraint This constraint forces a pushback to only be mined once T

t

/ Yp, t, n # 1

6p, n

1

Equal mining constraint This constraint ensures the equal mining of a pushback in each period within a given tolerance value ( per cent) 1 X # c% 6p, b, t, n - 1 X R p p, t, n R p, b p, b, t, n 1 X $ - c% - 1 X R p p, t, n R p, b p, b, t, n

6p, b, t, n

Recovered tonnage constraint This constraint calculates the recovered tonnage through a flow path. P, B

XRk, f, t, n

p

1, b

/

1;sbf

Rk, c X p, b, f, t, n

6k, f, t, n ; c e f

Grade units constraint This constraint calculates the grade units produced through a flow path. Gg, k, f, t, n

P, B p

/

1, b

1 sbf

Gg, k, c, nRk, c X p, b, f, t, n

6g, k, f, t, n ; c e f

Option constraints Option capacity constraint This constraint applies the upper capacity limit for each option. t

t-1

Xl, t, n - / Kl, t, nYl, tt + / Kl, t, nIDl, tt # 0 tt

1

tt

1

6l, t, n

Plant circuit capacity constraint This constraint applies a limit on the circuit capacity for the plant option t

t-1

Xl, c, t, n - / Kl, c, t, nYl, tt + / Kl, c, t, nIDl, tt # 0 tt

1

tt

1

6l, t, n

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

61

B GROENEVELD, E TOPAL AND B LEENDERS

Mining capacity constraint This constraint calculates the amount of capacity used for each mining option on an effective flat haul kilometres (EFH) basis P, B, F p

1, b

/

1, f

t

1 ; lef

t-1

E p, b, l, nX p, b, f, t, n - / Hl, t, nYl, tt + / Hl, t, nIDl, tt # 0 tt

1

tt

1

6l, t, n

Disposal constraint This constraint ensures that an option can only be disposed if it has previously been built. t-1

IDl, t - / Yl, tt # 0 tt

6l, t ! 1

1

t

t-1

/ IDl, tt - / Yl, tt # 0

tt

2

tt

6l, t ! 1

1

Option dependency constraint This constraint ensures the dependent relationship between options exists. Yl, t -

t - DT

/

1 ; eel

tt

Ye, tt # 0

6l, t

Stockpiling constraints Stockpile capacity constraint This constraint ensures that the maximum stockpiling capacity is not exceeded in any time period. t

t

Ks, t, n $ / XIs, j, tt, n - / XOs, j, tt, n tt

1

tt

6s, j, t, n

2

Stockpile flow out constraint This constraint ensures that the amount of material flowing out of a stockpile is less than or equal to what has flown in and flown out in previous periods. t

t-1

XOs, j, t, n # / XIs, j, tt, n - / XOs, j, tt, n tt

1

tt

6s, j, t, n

2

Stockpile recovery constraint This constraint determines the recovered material from a stockpile processed through flow path f. J

XRk, f, t, n

j

/

1 ; sef ; jef

6k, f, n, t ! 1

Rk, s, j X f, t, n

Stockpile grade constraint This constraint determines the grade produced from stockpiled material through flow path f. Gg, k, f, t, n

J

j

/

1 ; sef ; jef

Gg, k, s, j Rk, s, j X f, t, n

6g, k, f, n, t ! 1

Product constraints Maximum product component capacity This constraint sets an upper bound on the amount of product that can be produced to a particular component. Sd, k, t, n # Dd, k, t, n

6d, k, n, t

Product grade limit constraint This constraint ensures the grade limits for products are satisfied. F

f

/

1 ; def

Gg, k, f, t, n - GUg, d, k

F

f

/

1 ; def

XRk, f, t, n # 0

6g, d, k, t, n

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

62

ROBUST VERSUS FLEXIBLE OPEN PIT MINE DESIGN

F

/

1 ; def

f

F

Gg, k, f, t, n - GLg, d, k

/

XRk, f, t, n $ 0

1 ; def

f

6g, d, k, t, n

Flow balance constraints These constraints ensure what is flowing into a node equals what is coming out of a node. They have been implemented at various stages of the model and are seen below. Total pushback tonnage mined equals total mining from bins within that pushback. B

/ X p, b, t, n

X p, t, n

b

6p, t, n

1

Total mining from a bin equals the total flow through all paths. F

/ X p, b, f, t, n

X p, b, t, n

f

6p, b, t, n

1

Total flow through a path equals the total flow through all pushbacks and bins. P, B

X f, t, n

p

/

1;sbf;jbf

1, b

6f, t, n

X p, b, f, t, n

Total tonnage through each circuit equals the sum for all flow paths through that plant and circuit. F

Xl, c, t, n

/

1 ; lef ; cef

f

6l, c, t, n

X f, t, n

The tonnage processed through an option equals the sum of all flow paths which pass through. F

Xl, t, n

/

1 ; lef

f

6l, t, n

X f, t, n

Total tonnage flowing into a stockpile equals the sum from all flow paths which pass through. XIs, j, t, n

P, B, F p

1, b

/

1, f

6s, j, t, n ; G p, b $ GL j and G p, b 1 GU j

X p, b, f, t, n

1 ; sef ; jef

Total tonnage out of the stockpile equals the tonnage from each flow path out of the stockpile. F

XOs, j, t, n

/

1 ; jef ; sef

f

X f, t, n

6s, j, t, n

Total product component sale quantity may equal the total recovered component tonnage from flow paths. F

Sd, k, t, n

f

/

1 ; def

XRk, f, t, n

6d, k, t, n

or the metal units recovered from flow paths Sd, k, t, n

F

f

/

1 ; def ; g

0

Gg,k, f, t, n

6d, k, t, n

Non-negativity and integrality constraint This constraint enforces non-negativity and integrality of the variables, as appropriate. X p, b, f, t, n, X p, b, t, n, X f, t, n, XRk, f, t, n, XIs, j, t, n, XOs, j, t, n, Xl, c, t, n, Xl, t, n, El, t, n, Xl, c, t, n, Sd, k, t, n $ 0 6p, b, f, s, j, l, c, d, k, t MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

63

B GROENEVELD, E TOPAL AND B LEENDERS

IDl, t Integer 6 l, t and Yl, t Yp, t binary 6 p, l, t

CASE STUDY This case study examines the use of several different mining capacity and plant capacity options for the deposit using both a flexible design and robust design methodology side by side. Data from a copper-gold deposit is used to implement the methodology. It is assumed that the deposit will be mined by a single open cut operation. The deposit was divided into four pushbacks generated by a single deterministic optimisation. Whilst this may be considered to be removing the optimality from the model upfront, it was primarily used as a starting point for pushback selection. Likewise, the purpose of this case study is to examine the execution of mining and plant options more than generate an optimal schedule and/or sequence of extraction. Each pushback defines a scheduling constraint in the model, ie pushback one must be mined before pushback two. Table 1 provides a summary of the deposit used in this case study. The case study uses a single resource model, however multiple stochastic models can be included in the analysis. If desired each simulation would choose one resource model at random and adjust the grade and tonnage of the pushback accordingly. TABLE 1 Summary of pushbacks in the resource model. Pushback

Bins

Type

Tonnage (Mtpa)

Copper (%)

Gold (g/t)

1

37

Ore

36.9

1.3

0.7

1

1

Waste

40.9

-

-

2

29

Ore

28.8

1.4

0.7

2

1

Waste

63.6

-

-

3

9

Ore

7.3

1.4

0.8

3

1

Waste

21.7

-

-

4

18

Ore

17.0

1.2

0.8

4

1

Waste

66.5

-

-

Several options were included in this case study to undertake an analysis of the deposit. For this run of the model four mining options, four processing options and two stockpiling options were examined. A summary of the options included in the model is outlined in Table 2. The case study uses a commodity price for gold and copper, capital cost, operating cost and plant utilisation. No detailed analysis of the underlying nature of the stochastic variables has been carried out, as detailed research in other papers cover this aspect (Dimitrakopoulos and Abdel Sabour, 2007 Godoy and Dimitrakopoulos, 2004 Topal, 2008 Shafiee and Topal, 2010). TABLE 2 Summary of options included in case study. Option type

Capacity (Mtpa)

Capital ($M)

Fixed cost ($/yr)

Variable cost ($/t)

Disposal cost ($M)

Mine A

10.0

38.5

2.60

1.22

3.5

Mine B

12.5

42.5

3.25

1.21

4.0

Mine C

15.0

45.0

3.90

1.12

4.5

Mine D

25.0

55.0

6.50

1.12

5.5

Processing A

7.5

57.5

13.0

1.75

11.0

Processing B

10.0

70.0

16.8

1.31

7.0

Processing C

12.5

85.0

20.0

1.05

8.5

Processing D

15.0

97.5

22.5

0.87

9.5

Stockpile Waste

360.0

-

-

0.20

-

Stockpile Low Grade

20.0

-

1.00

0.60

-

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

64

ROBUST VERSUS FLEXIBLE OPEN PIT MINE DESIGN

The following distributions are used x gold price lognormal distribution with a mean of 1000/oz, standard deviation of 350/oz and a correlation of 0.20 between periods x copper price lognormal distribution with a mean of 4500/t, standard deviation of 2500/t and a correlation of 0.25 between periods x capital cost multiple lognormal distribution with a mean of 3.0 per cent, standard deviation of 7.0 per cent and a correlation of 0.10 between periods x operating cost multiple triangular distribution with a lower limit of -40.0 per cent, midpoint of 5.0 per cent, upper limit of 35.0 per cent and a correlation of 0.05 between periods and x plant utilisation triangular distribution with a lower limit of 30 per cent, midpoint of 80 per cent and upper limit of 95 per cent. For the case study both the robust design methodology and the flexible design methodology were run for comparison. A sample set of 50 simulations or ‘states-of-the-world’ were used for the case study. A summary of the ranges of values for each variable is shown in Table 3. TABLE 3 Summary of ‘states-of-the-world’ variable ranges. Variable

Mean

Median

Maximum

Minimum

Copper ($/t)

4514.47

3867.65

14 265.88

767.88

Gold ($/oz)

1011.48

956.68

3131.68

308.58

Capital (%)

2.8%

2.7%

25.9%

-13.8%

Operating (%)

-2.0%

1.5%

32.9%

-39.2%

Utilisation

68.7%

70.7%

94.2%

30.7%

Traditional plan A traditional plan for operating a mine was generated from current manual approaches with limited optimisation. For this case study the design used was to build plant B, plant D, mine B and mine D in period one and to build mine B in period two. Using the mean values as the uncertainties inputs this design had an NPV of 1697 M. For comparison purposes this plan was processed under each ‘state-of-the-world’ to determine the payoff profile for this fixed design. The average NPV was 1678 M, with a minimum of 720 M and a maximum of 3603 M.

Robust plan The robust model produced an average NPV of 1787 M, with a minimum of 721 M and maximum of 3885 M. It generated a robust design which entailed building mine option C and D (40 Mta), plant options A, B, C and D in period one. Also, it can be determined from the schedule that mining of pushback 1 finished in period two, followed by pushback 2 in period three and pushback 3 in period four. This suggests that either the mining options in the model had capacities that were too large or that the model prefers to extract ore at a high rate. Further scenarios could be generated by the decision maker to test each hypothesis. It should be noted that processing of the model with over 50 ‘states-of-the-world’ is slow. This is because the number of linear variables present in the model increases linearly with the number of simulations. For the model with 50 ‘states-of-the-world’ there was 1.3 million variables. Gurobi solves this model in two hours on a Quad core 2.66 Ghz Ubuntu Server.

Flexible plan The flexible model produced an average NPV of 2010 M, minimum of 818 M and maximum of 3969 M for the 50 ‘states-of-the-world’. In this model the design changes with each simulation so many mine and plant option combinations are selected. In order to analyse the options used in the model the frequency of execution graphs are generated to show how often a particular option is used (number of simulations option is used divided by number of simulations). The frequency of execution for the flexible model is shown in Figure 2. The available number of times an option could be executed was 500, which is 50 simulations multiplied by ten periods. Further to this graph, Table 4 MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

65

B GROENEVELD, E TOPAL AND B LEENDERS

FIG 2 - Frequency of execution for flexible design model. TABLE 4 Summary of frequency of execution for each period in flexible design model. Option

All Periods

1

2

3

4

5

6

7

8

9

10

Mine A

15%

26%

52%

22%

18%

18%

6%

12%

0%

0%

0%

Mine B

16%

26%

14%

34%

20%

18%

14%

18%

12%

4%

0%

Mine C

23%

50%

32%

30%

22%

28%

28%

18%

10%

8%

0%

Mine D

24%

80%

48%

28%

26%

28%

16%

8%

2%

0%

0%

Processing A

3%

8%

8%

4%

2%

2%

2%

2%

0%

0%

0%

Processing B

8%

22%

22%

12%

2%

10%

6%

6%

2%

0%

0%

Processing C

11%

44%

20%

14%

10%

10%

6%

8%

0%

0%

0%

Processing D

20%

76%

42%

20%

20%

20%

8%

8%

2%

2%

0%

shows the frequency of execution per period. It is important to note that the robust design is not the same as the flexible design for each ‘state-of-the-world’.

Operational plan An operational model was developed as a hybrid between the robust and flexible model. The design was fixed for the first two periods by choosing the design for these periods from the robust model. For period three onwards flexibility was available so the model could turn on and off design options. The model formulation for this was the same as the flexible model with the only difference being that the binary values for the design options where fixed in period one and two. A similar approach as the flexible model was used to generate a solution for each ‘state-of-the-world’. The average NPV for this model was 1867 M, with a minimum of 721 M and maximum of 3885 M. It is proposed that a further refinement of the initial two year design can be achieved by analysing the results in further detail. One approach for this is to compare the frequency of execution for the operational plan detailed in Table 5 with the frequency of execution for the flexible model in Table 4.

Comparison In comparing the models it can be seen that the flexible model produces a higher expected value than any other model. This is as expected since the flexible model produces an optimal design for each MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

66

ROBUST VERSUS FLEXIBLE OPEN PIT MINE DESIGN

TABLE 5 Summary of frequency of execution for each period in operational design model. Option

All Periods

1

2

3

4

5

6

7

8

9

10

Mine A

1%

0%

0%

12%

0%

0%

0%

0%

0%

0%

0%

Mine B

1%

0%

0%

8%

0%

0%

0%

0%

0%

0%

0%

Mine C

12%

100%

0%

10%

0%

8%

2%

4%

0%

0%

0%

Mine D

25%

100%

100%

22%

12%

8%

6%

6%

0%

0%

0%

Processing A

0%

0%

0%

2%

0%

0%

0%

0%

0%

0%

2%

Processing B

2%

0%

0%

8%

4%

4%

0%

4%

2%

0%

2%

Processing C

13%

100%

0%

8%

2%

4%

4%

4%

2%

4%

0%

Processing D

24%

100%

100%

12%

2%

8%

6%

6%

0%

2%

2%

‘state-of-the-world.’ Such a flexible design may be impractical in reality so fixing the first couple of design periods is proposed as an alternative. The operational plan outlined fixed the first two years of the design and left the remainder flexible which on average increased the expected value by 11 per cent. A value of risk graph is produced to highlight this, as shown in Figure 3, which allows for easy comparison of the expected values from different design approaches. It can be seen that the flexible design curve shows the highest expected value.

FIG 3 - Value at risk graph comparing design models.

The operational plan design has a lower expected value then the fully flexible design due to the reduced flexibility. The robust design curve shows a value lower than the operating plan curve as no flexibility is incorporated in the design, thus prohibiting the design from reacting to change. A traditional design produces the lowest expected value overall since it has no flexibility and is not optimised for a range of conditions. The differences between the expected values of each design approach can be attributed to two key aspects. First of all, actively managing the operation and allowing a flexible design (one that changes over time) will contribute significant additional value. The second component that contributes to additional value is being able to develop a robust fixed design which can handle a range of conditions. When comparing the expected value of the operational plan with the traditional model NPV an increase of 170 M was achieved. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

67

B GROENEVELD, E TOPAL AND B LEENDERS

Further analysis of the results using such techniques as data mining may be able to provide decision rules to help guide the decision maker. Refinement of the initial design chosen for the operational plan may lead to the discovery of plans with a higher expected value.

CONCLUSIONS The paper has demonstrated the concept of a robust design methodology further extending the field of real options ‘in’ design. It has been demonstrated that a fixed design (as is the case with the robust design) produces a lower project value than a flexible design. However, knowing the robust design will help to further guide the decision-making process. Finally, it has been shown with clarity the impact of adopting a build and leave approach to operations. Flexible operations produce the greatest project value, thus actively managing an operation is imperative. Further research continues in the following areas x incorporating conditionally simulated models to handle grade uncertainty, x use of data mining techniques to further understand the best flexible design practice, x application of the model to a real dataset to prove up the power of this new technique, and x application of this technique to an underground mine as the first step in building a model that can handle the full array of design options regardless of mining method.

REFERENCES Bergman, B, Mare, J and Svensson, T, 2009. Robust Design Methodology for Reliability Exploring the Effects of Variation and Uncertainty (John Wiley Sons Oxford). Cardin, M A, de Neufville, R and Kazakidis, V, 2008. A process to improve expected value of mining operations, Transactions of the Institutions of Mining and Metallurgy, Mining Technology, 117(2) A65-A70. Chambers, M, 2008. Rio to study Weipa expansion to cash in on aluminium boom, The Australian, 13 June. de Neufville, R, 2006. Analysis methodology for the design of complex systems in uncertain environment Application to mining industry (unpublished) Engineering Systems Division, Massachusetts Institute of Technology, Cambridge, Massachusetts. de Neufville, R, Scholtes, S and Wang, T, 2005. Real options by spreadsheet Parking garage case example, Journal of Infrastructure Systems, 12(2) 107-111. Dimitrakopoulos, R G and Abdel Sabour, S A 2007. Evaluating mine plans under uncertainty Can the real options make a difference Resources Policy, 32(3) 116-125. Godoy, M and Dimitrakopoulos, R 2004. Managing risk and waste mining in long-term production scheduling, SME Transactions, 316 43-50. Groeneveld, B and Topal, E 2010. Flexible open-pit mine design under uncertainty, Journal of Mining Science (in press). Groeneveld, B, Topal, E and Leenders, B 2009. A new methodology for flexible mine design, in Proceedings Orebody Modelling and Strategic Mine Planning (ed R Dimitrakopoulos), pp 109-118 (The Australasian Institute of Mining and Metallurgy Melbourne). Le May, R, 2009. Mincor to keep Miitel mine in WA closed, The Sydney Morning Herald, 3 August 2009. Monita, G, 2009. BHP Billiton to close Ravensthorpe nickel mine, Metal Markets, 22 January. Shafiee, S and Topal, E, 2010. An overview of global gold market and gold price forecasting, Resource Policy Journal, 35(3) 178-189 Topal, E, 2008. Evaluation of a mining project using discounted cash flow analysis, decision tree analysis, Monte Carlo simulation and real option using an example, International Journal of Mining and Mineral Engineering, 1 62-76. Trounson, A, 2007. Rio ponders 50pc Pilbara expansion, The Australian, 12 June. Trounson, A, 2008. BHP plans 2.3bn Worsley expansion, The Australian, 2 May. Wang, T and de Neufville, R, 2005. Real options ‘in’ projects, paper presented to Ninth Real Options Annual International Conference, Paris. Wang, T and de Neufville, R, 2006. Identification of real options ‘in’ projects, paper presented to 16th Annual International Symposium of the International Council on Systems Engineering (INCOSE), Orlando.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

68

The Role of Cost Estimating in Mine Planning and Equipment Selection J B Leinart1 and O L Schumacher2 ABSTRACT The objective of a mine planning exercise is to plan a safe mine that maximises net present value (NPV) for the project. Two primary elements are key to design of a mine plan optimised to maximise profit selection of an optimised production rate and selection of an optimised equipment fleet. Both depend upon accurate estimations of capital and operating costs for the equipment and for the life cycle of the mining operation. Once a trial production rate is determined, a primary equipment fleet, ie excavators, haulers, and blasthole drills, can be determined based on material characteristics and haulage distances and gradients. An auxiliary fleet, dozers, road maintainers, service trucks, etc can be determined using less rigid productivity criteria. With an equipment fleet determined, operating and support personnel can be assigned based on shift requirements and equipment operating needs and material and supply requirements can be estimated. Knowing the areal dimensions of the major equipment items, road widths and repair shop dimensions can be determined. Personnel numbers and supply requirements determine the square metre sizes of offices, change rooms, and storage facilities. Reliable sources of capital and operating unit prices and costs for all these items are referenced in the paper. Once preliminary capital and operating cost estimates are calculated by applying unit costs to equipment, personnel, material and supply needs, potential profit can be tested using DCF techniques. An optimised mine plan can then be determined by testing the potential profitability of multiple mine production rates. Because of the economies of scale, full capital and operating cost estimates must be completed for each production rate in order to produce meaningful comparisons.

INTRODUCTION Absent any external restraints such as market limitations or permit restrictions, the objective of a mine planning exercise is to plan a safe mine that maximises net present value (NPV) for the project. Critical to this process are the selection of a production rate and an equipment fleet, each optimised to maximise NPV as determined through standard discounted cash flow (DCF) analysis techniques. In the simplest of terms, all of the geological and mining parameters that we consider in an NPV exercise contribute either to revenues, the plus arrows on our project timeline, or to costs, the negative arrows on our project timeline (Figure 1). Fundamental to the process of DCF analysis for the purpose of selecting a production rate and an equipment fleet is the completion of multiple DCF analyses that assume multiple production rates and various equipment fleets. However, it is dangerous to think of this process simply in terms of a statistical sensitivity analysis. With appropriate software, it is a simple matter to assign a variable for analysis and then ask the software to show how the NPV responds to changes in the value of the variable. For some variables, production rate and equipment costs in particular, this can produce very misleading results. The reason being, that changes in production rates or equipment fleets produce changes in the size and distribution of many of the plus arrows and the minus arrows on our project timeline. A basic statistical sensitivity analysis would not account for these changes. The changes can only be predicted by completing a new or revised capital and operating cost estimate for each of the production rates or equipment fleets analysed. Furthermore these cost estimates must consider all of the parameters that vary with each model. For instance, a change in the size 1. President, InfoMine USA Inc, CostMine Division, 1120 N Mullan Road, Suite 100, Spokane Valley, Washington 99206, USA. Email: [email protected] 2. Director, InfoMine USA Inc, CostMine Division, 1120 N Mullan Road, Suite 100, Spokane Valley, Washington 99206, USA. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

69

J B LEINART AND O L SCHUMACHER

Revenue Items Reserves Grades Production Recoveries Commodity Prices Etc.

+$

1

Years

2

Project Timeline

-$ Operating Cost Items Haul Distances Personnel Wages/Salaries Supplies Equipment Operation Etc

Capital Cost Items Overburden Removal Haul Road Construction Equipment Purchase Facilities Construction

FIG 1 - Project timeline depicting revenue and cost data items.

and number of equipment units brings about changes not only in equipment costs, but in personnel requirements and costs, haul road size and maintenance requirements, repair shop requirements, etc. A change in production rate brings about changes in virtually all revenue and cost items. Hence, an exercise to optimise production rate or equipment fleet selection must first emphasise completion of cost estimates for each model considered. Early in the feasibility process, maximum attention is focused on the revenue items, especially ore reserve amount and grade, usually with little attention paid to the cost items. In the fortuitous circumstance that exploration activities give way to mine planning efforts, attention shifts to the cost items. A prefeasibility level mine planning and cost estimating effort should be initiated as early as possible, as soon as a reasonable estimate of reserves can be made and a preliminary conceptual mine plan can be developed. This early cost estimate serves as an exploration tool in itself, in that once we move from focusing mainly on geologic concepts such as alteration zones and ore grades, and begin thinking of our orebody as an actual chunk of the earth’s crust that we are going to excavate and move, we begin to ask questions that help guide further exploration activities and perhaps jump start the mine planning, equipment selection, and mine permitting processes. Questions start us thinking about our project in a new way ‘What is the configuration of the orebody boundary ’, ‘Where are the high grade zones and the waste zones ’ ‘Where will the waste dump and mill be located ’. A number of systems have been developed in attempts to shortcut the cost estimating process, such as O’Hara’s uick Guides to the E aluation of Orebodies ( O’Hara, 1980, 1981 O’Hara and Suboleski, 1992) Mular s Mining and Mineral Processing E uipment Costs and Preliminary Capital Cost Estimations ( Mular, 1982), which was updated in a publication called CAPCOSTS A Handbook for Estimating Mining and Mineral Processing E uipment Costs and Capital Expenditures and Aiding Mineral Project E aluations (Mular and Powlin, 1998) and the CES system developed by the US Bureau of Mines (Camm, 1991, 1994). These are all parametric systems that relate the cost of a unit or process to some parameter of the unit or process, for instance the cost of a truck may be related to the capacity of a truck or the cost of sinking a shaft may be related to the cross-sectional area of the shaft. The relationship may be expressed in a graph or a formula such as the following C

f(P)

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

70

THE ROLE OF COST ESTIMATING IN MINE PLANNING AND EQUIPMENT SELECTION

where C the cost of the unit or process P some parameter of the unit or process Unfortunately all of these systems are now out of date and are not being supported, and none of them are suitable for use in selecting and optimising an equipment fleet because they do not allow the estimator to analyse mine operating parameters in sufficient detail to aid in equipment selection. In the opinion of these authors, the only method suitable for purposes of equipment selection and optimisation is a traditional itemised method, which requires the estimator to use basic engineering principles to determine the size and number of equipment units required and the hours per day, month or year that each will operate. This procedure is described in the paragraphs that follow. The first step necessary to initiate a cost estimate is to develop a conceptual mine plan that will provide the following basic information x proposed production rates for ore and waste x basic geometry of the ore deposit the estimate does not require a detailed pit plan, but a waste to ore ratio must be determined or assumed, and an approximate centre of mass for ore and waste should be determined as starting points for material haulage x mill, leach pad, or stockpile location the mill, leach pad, or stockpile location is necessary in order to identify the destination point for ore haulage x waste dump location the waste dump location is necessary to identify the destination point for waste haulage and x haul routes for ore and waste proposed haul routes to the mill or stockpile site and to the waste dump should be plotted on a plan map in sufficient detail to estimate road gradients and lengths (Figure 2). The plan can be prepared to any level of detail desired or possible, but the cost estimate cannot proceed until these essential items have been determined or at least assumed.

FIG 2 - Basic information to be derived from conceptual mine plan (photo of Pipeline Mine in Nevada, USA, courtesy of Dave Schumacher).

Proposed daily production rate Proposed daily production rate, the first item to be determined, is commonly used to express the ‘size’ of a mine, eg a 10 000 ton per day mine, or a 50 000 ton per day mine. Indeed, the production rate immediately tells us something about the relative size of the equipment fleet, the size of the required working area, the size of the workforce, as well as the size of the mill. It follows that the selection of the daily production rate is an important decision that must be made before we can begin to estimate costs, no matter what method we choose to use for our estimate. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

71

J B LEINART AND O L SCHUMACHER

For metals and other commodities for which there is a large market that is insensitive to the amount of production from the proposed mine, the question becomes one of practicality and internal economic analysis. In the absence of external constraints, the proposed production rate should be an optimised rate based on maximisation of NPV, constrained only by practical considerations such as safety, environmental considerations, working area availability, etc but to find the optimum rate requires the completion of multiple economic analyses based on multiple cost estimates for a range of possible production rates, followed by economic (DCF) analyses for each. Time constraints placed upon the planner usually preclude such complete analyses at the prefeasibility level, although software packages produced by Adventurine Engineering (Stebbins, 2010a, 2010b, 2010c) and cost data available through InfoMine USA Inc (2009a, 2009b, 2010a, 2010b) make it possible to perform multiple analyses quickly and efficiently. In the absence of such tools, the planner typically is left to his own experiences and knowledge of existing mines to select a reasonable production rate upon which to base the initial cost estimate.

Equipment selection Once the conceptual mine plan and proposed production rate are established for an open pit mine, the underpinning of a successful mine and the basis for the entire cost estimating process to follow is selection of an optimised excavator and hauler fleet. This fleet will be the major cost centre during the operation of the mine, and as we shall see, the design of most of our mine facilities, the selection of our auxiliary equipment fleet, our personnel requirements, and most of our costs are influenced at least to some degree by the number, size and type of equipment in this fleet.

Excavator selection The initial step in the fleet selection process is to select an appropriate size and type of excavator or excavators. The choice of cable shovel, hydraulic shovel, wheel loader, dragline, or bucketwheel excavator depends on the size of the operation, required breakout strength, regularity of the ore/ waste boundary and other mining parameters. Once the type of excavator is selected, cycle time calculations must be completed to determine the expected productivity of the excavator under the expected mining conditions. Equipment specifications provided by the manufacturer are important to this process. Rock density, swell factor, bucket fill factor and operating conditions all impact excavator productivity. One or more excavators must be selected that are capable of maintaining the desired productivity at the assumed mine production rate.

Hauling equipment selection Once an initial size and type of excavator is assumed, a matching hauler fleet must be selected. The hauler fleet selection process is described in detail here to illustrate the basic principles involved in the selection process for most types of mining equipment. The most common choices for hauling equipment include x conveyors, x scrapers, x rigid frame rear dump trucks, x articulated rear dump trucks, and x bottom dump trucks. For purposes of this example, truck haulage will be assumed. Once the type of hauler is decided upon, the number and size of hauler units must be determined. For truck haulers, this is accomplished through a productivity analysis similar to that done for the excavator. Material characteristics important to this analysis are in situ rock density, swell factor, truck fill factor, and haul road profile. The productivity analysis for truck haulers is accomplished in the following steps 1. Select a reasonably sized truck as a starting point for the analysis. If this size turns out to be an inappropriate selection, the process must be repeated using other truck sizes until a satisfactory size is found. 2. Determine the truck cycle time. The time required for a truck to complete one cycle consists of the time required to x manoeuvre the truck into position for loading, x load the truck, MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

72

THE ROLE OF COST ESTIMATING IN MINE PLANNING AND EQUIPMENT SELECTION

x travel from the loading point to the dumping point, x turn and dump the load, and x travel from the dumping point to the loading point. Manoeu ring time truck manoeuvring times in the loading area typically run 0.6 to 0.8 minutes. This assumes that adequate space is available for unrestricted manoeuvring, and the truck does not have to be backed into position a significant distance, such as down an initial drop cut ramp. Similarly for manoeuvring at the dump point, manoeuvre and dump times of 1.0 to 1.2 minutes are typical. Loading time loading time depends on the excavator, and is determined by the previously completed excavator cycle time calculation. Tra el time truck travel time is determined by estimating the time it will take the truck to travel through each of the road gradient segments. The times for each segment are accumulated to determine the total time, first for the loaded truck travelling from the loading point to the dump point, and then again for the empty truck travelling in the return direction. Segment by segment travel times are estimated using speed and retarder curves supplied by the manufacturer of the selected truck. These ponderous looking charts depict the design speed of the vehicle when travelling against the total of the gradient resistance plus or minus the rolling resistance. Speed curves are used for uphill travel. Retarder curves are used for downhill travel. Rolling resistance is a measure of the force that must be overcome to roll a wheel over the ground. It is usually expressed as a percent gradient, in recognition of the fact that certain road and tire conditions impact speed in the same way that an uphill road gradient does. Hence, it is added to the road gradient when estimating uphill travel speed, and it is subtracted from the road gradient when estimating downhill travel speed. For preliminary estimating purposes, a rolling resistance of about three per cent can be used, assuming the roadway will be fairly well maintained (Figure 3).

FIG 3 - A typical truck speed curve (courtesy of Caterpillar Company for a 777F truck).

Figure 3 (Caterpillar Inc, 2010) illustrates a typical speed curve used to determine the speed at which a hauler will travel on an uphill road segment. To use the curve, read down from the gross vehicle weight (either empty or full as appropriate) to the appropriate line for total effective resistance (grade plus rolling). From this point read horizontally, either right or left to the curve with the highest obtainable speed. The number on this line represents the gear in which the truck will be working. From this point, read down to the estimated speed. For high altitude situations, the estimated speed should be adjusted by an altitude deration factor suggested by the truck manufacturer. Caterpillar lists deration factors for its equipment on its website and in the Caterpillar Performance Handbook. A retarder curve designed for each truck model provides a means of estimating the truck’s speed on downhill segments. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

73

J B LEINART AND O L SCHUMACHER

The speed for each segment is then converted to travel time by the formula T

60

D/S

where T travel time in minutes D distance in feet or metres S speed in feet or metres per hour When the travel times for all the road segments have been determined, they should be added together to determine the total travel time loaded, and the total return travel time empty. So now we have the total cycle time required for a truck to position itself to load, travel to the dump point, dump, return, and position itself for loading again. Four questions remain to be answered to complete the truck analysis 1. How much material can one truck haul in each cycle 2. How much material can one truck haul in one hour 3. How many trucks are required to achieve the desired amount of production per shift 4. How many hours must the haul trucks operate each shift to achieve the desired level of production To determine how much material one truck can haul in each cycle, the manufacturer’s model specifications are referenced to determine the target payload in tons for which the truck is designed. The volume of the target payload is then calculated, after the bank density of the material is converted to its loose density, to assure that it does not exceed the volume of the truck bed. If it does, the payload per haul is limited by the volume of material the truck can hold. If not, the design payload weight for the truck is usually used as its capacity for purposes of preliminary analysis. Manufacturers commonly list a struck capacity and a heaped capacity for their trucks. The struck capacity is the volume of the truck bed to the top of the sides. The heaped capacity is the struck capacity plus an amount equal to a pile over the top of the sides with 2 1, 3 1, or 4 1 slopes. The struck capacity adjusted by a fill factor is commonly used for preliminary estimates. A fill factor can be applied to provide a means for the estimator to account for the fact that the excavator does not provide a perfectly full load with each cycle. The amount of material one truck can haul in one hour is the number of cycles that can be completed in one hour, multiplied by the anticipated amount per load, adjusted by an ‘efficiency factor’. The efficiency factor is applied in recognition of the fact that the truck is not always operated with 100 per cent efficiency, ie the driver may stop occasionally for a restroom break, a conversation with his supervisor, a traffic delay, etc. Commonly an 83 per cent efficiency factor is applied for preliminary estimates of routine production work. This is sometimes called the ‘50 minute hour’. An ‘availability’ factor can also be applied, on the assumption that each truck may be out of service from time to time for mechanical reasons. Once the adjusted hourly production per truck is determined, the number of trucks required to achieve the desired production level per shift and the total number of hours the trucks will operate per shift are determined. The numbers should be reviewed to assure that the planned fleet is practical, manageable, and appropriate for the situation. To be avoided is a very large number of small trucks or a very small number of large trucks. Whether or not to include spare trucks in the equipment fleet is a judgment call based on a number of factors, such as the need to assure uninterrupted haulage, the breakdown history of the fleet, capital availability, etc. Adding spares to the fleet will of course add to capital costs, but will have no effect on estimated operating costs because the estimate assumes that no operating costs are involved when a truck is idle. Three critical pieces of information are carried forward from this haulage analysis 1. the size of the trucks required, 2. the number of trucks required, and 3. the hours per shift these trucks will operate. These three items are the basis for determining the capital cost for haulage equipment, the number of hauler operators required, and the cost per day for haulage equipment operation, as well as for a number of other seemingly extraneous factors such as those listed in the next paragraph. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

74

THE ROLE OF COST ESTIMATING IN MINE PLANNING AND EQUIPMENT SELECTION

Optimisation of primary equipment fleet Upon completion of the excavator and hauler productivity analyses, it may become obvious that more than one combination of hauler and excavator sizes are capable of accomplishing the proposed production at the desired daily rate, eg a large number of small units or fewer larger units. For the initial estimate, the planner may simply choose the most reasonable combination based on experience and judgment. Ultimately, the excavator and hauler fleets should be optimised to maximise NPV. To do so requires consideration of both capital and operating costs analysed using DCF techniques. The DCF analyses must be based on multiple complete cost estimates that give appropriate consideration to the complex relationships among various mining parameters that depend upon the size and number of equipment units, for instance x the widths of the haul roads depend upon the size of the haul trucks x the sizes and operating hours of the road maintainers and watering trucks depend upon the width and length of the haul roads x the size of the mechanical shop depends upon the size and number of the equipment units to be serviced x the number of operating personnel and mechanics depend upon the number of units in the fleet x the number of supervisory personnel depends upon the number of operating personnel and x the square metre sizes of offices, change rooms and storage facilities are determined by the personnel numbers and supply requirements.

Drill selection After devoting considerable attention to the selection of excavators and haulers, the next most important type of equipment to select for cost estimating purposes is blasthole drilling equipment. The selection process can be completed in a number of ways depending upon how much is known about blasthole pattern design, bench height, rock hardness, etc but basically, the size and number of drills must be selected that are capable of drilling hole volume sufficient to contain enough explosives to blast the amount of material needed to meet daily production requirements. This procedure is explained more fully in Schumacher (2010). Specification information provided by drill manufacturers can be helpful as well. As with the truck analysis, three critical pieces of information are carried forward from this analysis, the number of drills required, the size of the drills, and the hours per day the drills must operate. Additional items that will come out of this analysis are the daily consumption rates for drill bits and steel, primers, caps and explosives.

Selection of auxiliary equipment Considerably less effort is commonly applied to the selection of an auxiliary equipment fleet for preliminary cost estimating purposes. Various assumptions or rules of thumb can be applied to the selection process, eg the number and size of road maintainers should be selected capable of maintaining the total haul road lengths a certain number of times per day or week, similarly for watering tankers bulk explosive trucks should be selected capable of loading the required amount of explosives per day, etc. Cost models in Mining Cost Service (InfoMine USA Inc, 2010a, 2010b) provide convenient reference lists for auxiliary equipment requirements at mines of various sizes. Table one lists items typically included in an auxiliary equipment fleet (Table 1). TABLE 1 Typical list of items that may be included in the auxiliary equipment fleet. Road maintainers

Pumps

Bulk powder trucks

Water tank truck

Dozers

Portable lighting plants

Service trucks

Pickup trucks

Buildings and structures Buildings and structures pose a special problem, in that until detailed design work has been completed, one should not attempt to do a standard itemised cost estimate based on design take-off amounts. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

75

J B LEINART AND O L SCHUMACHER

Instead, building sizes should be assumed, and references such as RSMeans Building Construction Cost Data Book (RSMeans, 2009) should be consulted for costs per square foot or square metre. Convenient rules of thumb for building sizes are discussed in Schumacher (2010).

Haul roads Considerable attention is paid to the design, construction and maintenance of haul roads at most large mining operations, because of the abusive, heavy traffic to which they are subjected, and because a poorly constructed or maintained road can have considerable negative impact on productivity and on equipment and tire wear. In the US, Mine Safety and Health Administration (MSHA) guidelines recommend a road width sufficient to provide clearance of half the vehicle width on each side of the vehicle, ie for two-way traffic, the road width should be 3.5 times the vehicle width (two truck widths for the traffic lanes plus half truck width twice for outside clearance, and truck width for centre clearance between trucks Haul road construction involves a series of tasks such as surveying, cutting and filling, compacting, watering for dust control, etc. Costs per kilometre for all these tasks can be estimated using equipmentbased productivity techniques similar to those described for the truck analysis but usually this approach is far too time consuming for most feasibility studies, particularly at the prefeasibility level. A more common approach is to utilise reliable precalculated costs per metre or kilometre of road from published references.

Unit costs In the preceding paragraphs we discussed methods for determining the amounts of materials, number of personnel, equipment requirements and daily hours of usage, and certain development needs. Once these items have been estimated, it is time to apply unit pricing to each item in order to calculate total cost. The mathematics of all this is very simple, basically involving little more than multiplying the daily usage of an item by the unit price for the item, and accumulating the results on a spreadsheet. The difficult part is obtaining reliable unit prices and costs to use for the estimate. Possible sources for unit costs are many and varied, some reliable, some not so reliable. Listed below are available sources for unit cost information. Equipment capital and operating unit costs: x Mine and Mill Equipment Cost Guide (InfoMine USA Inc, updated annually) x Mining Cost Service (InfoMine USA Inc, updated annually) x Equipment Cost Calculator (InfoMine USA Inc, updated annually) x manufacturers/distributors and x company catalogues. Unit costs for supply items: x Mining Cost Service (InfoMine USA Inc, updated annually) x manufacturers/distributors and x power company rate schedules. Wages, salaries and benefits for employees: x Mining Cost Service (InfoMine USA Inc, updated annually) x wage and salary reports (InfoMine USA Inc, updated annually) and x local surveys available from government agencies and consultants. Buildings and structures: x Mining Cost Service (InfoMine USA Inc, updated annually) and x building construction cost data (RSMeans, updated annually). Haul road construction: x Mining Cost Service (InfoMine USA Inc, updated annually).

Cost summary spreadsheets The final step in the cost estimating procedure is to complete the simple math equations to convert the numbers of units and the prices per unit to costs per day (or month or year) and cost per tonne. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

76

THE ROLE OF COST ESTIMATING IN MINE PLANNING AND EQUIPMENT SELECTION

This is best done on a spreadsheet that provides a means to complete the mathematics efficiently and to summarise the costs in a concise manner. Table two is an example spreadsheet summarising operating costs for a proposed mine, and table three is an example spreadsheet showing capital costs for a proposed mine (Table 2 and Table 3). TABLE 2 Mine operating cost summary: example mine – 20 000 tonnes per day ore, 40 000 tonnes per day waste (2:1 stripping ratio waste:ore). Date of costs: May 2009. Labour Overtime hourly wage @ 150% ($/hr)

Regular hours/day (hrs/day) each

Overtime hours/day (hrs/day) each

Total wages per day (US$/ day)

Payroll loading 30% (US$/day)

Total Cost per daily tonne payroll ore cost (US$/tonne) (US$/day)

Hourly personnel

Number of personnel

Hourly wage ($/hr)

Wheel loader operator

2

23.51

35.27

8

2

517.22

155.17

672.39

0.03

Front shovel operator

2

23.51

35.27

8

2

517.22

155.17

672.39

0.03

Haul truck drivers

30

22.22

33.33

8

2

7332.60

2199.78

9532.38

0.48

Drillers

6

21.04

31.56

8

2

1388.64

416.59

1805.23

0.09

Dozer operators

8

23.51

35.27

8

2

2068.88

620.66

2689.54

0.13

Road maintainer operator

1

23.51

35.27

8

2

258.61

77.58

336.19

0.02

Water truck driver

1

22.22

33.33

8

2

244.42

73.33

317.75

0.02

(N/A)

-

-

-

-

-

-

-

-

Blasters

4

21.04

31.56

8

2

925.76

277.73

1203.49

0.06

Labourers/ Maintenance/helpers

26

18.00

27.00

8

2

5148.00

1544.40

6692.40

0.33

Subtotal hourly personnel

80

18 401.35

5520.41

23 921.76

1.20

Mechanics

Number of personnel

Annual salary (US$/yr)

Annual hours (hrs/yr)

Total salaries per day (US$/day)

Payroll loading 30% (US$/day)

Total daily payroll cost (US$/day)

Cost per tonne ore (US$/day)

0.5

124 000

2080

238.46

71.54

310.00

0.02

Superintendent

1

76 400

2080

293.85

88.15

382.00

0.02

Foreman

2

68 600

2080

527.69

158.31

686.00

0.03

Chief engineer

1

87 200

2080

335.38

100.62

436.00

0.02

Engineer

3

64 600

2080

745.38

223.62

969.00

0.05

Chief geologist

1

86 800

2080

333.85

100.15

434.00

0.02

Geologist

1

61 100

2080

235.00

70.50

305.50

0.02

Shift supervisor

4

62 000

2080

953.85

286.15

1240.00

0.06

Technician

9

50 000

2080

1730.77

519.23

2250.00

0.11

Accountant

2

64 700

2080

497.69

149.31

647.00

0.03

Clerk

4

37 800

2080

581.54

174.46

756.00

0.04

HR personnel manager

1

86 200

2080

331.54

99.46

431.00

0.02

HR personnel staff

2

45 000

2080

346.15

103.85

450.00

0.02

Security/safety manager

1

68 000

2080

261.54

78.46

340.00

0.02

Security/safety staff

2

50 000

2080

384.62

115.38

500.00

0.03

Subtotal salaried personnel

34.5

-

-

5724.23

1717.27

7441.50

0.37

Salaried personnel Manager

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

77

J B LEINART AND O L SCHUMACHER

TABLE 2 CONT... Mine operating cost summary: example mine – 20 000 tonnes per day ore, 40 000 tonnes per day waste (2:1 stripping ratio waste:ore). Date of costs: May 2009 Equipment operation No of Units

Cost/unit/operating hour (US$/hr)

Operating hours/ day/unit (hr/day)

Cost/dayall units (US$/day)

Cost/tonne ore-all units (US$/tonne)

Wheel loader

1

191.53

16.5

3160.25

0.16

Front shovel

1

444.63

17.3

7692.10

0.38

Haul trucks

15

139.45

17.5

36 605.63

1.83

Rotary blasthole drills

3

116.36

17.2

6004.18

0.30

Road maintainer

1

54.15

10.0

541.50

0.03

Dozers

4

87.73

11.0

3860.12

0.19

Bulk explosives truck

1

31.20

20.0

624.00

0.03

Lighting plants

4

2.81

9.0

101.16

0.01

Fuel/Lube truck

2

13.78

20.0

551.20

0.03

Mechanic’s truck

2

14.33

8.0

229.28

0.01

Tire service truck

1

18.60

8.0

148.80

0.01

Centrifugal pump

1

11.34

24.0

272.16

0.01

Water truck

1

147.90

12.0

1774.80

0.09

Pickup trucks

13

7.33

4.0

381.16

0.02

61 946.33

3.10

Total equipment operating cost Supplies

Units

Cost/unit (US$)

Number of units/day (ea)

Cost/day (US$/day)

Cost/tonne ore (US$/tonne)

4400

308.00

0.02

Fuel/lube (included in eq op costs)

NA

Tires (included in eq op costs)

NA

Repair parts (included in eq op costs)

NA

Electricity

kwh

0.07

Drill bits

ea

2463.00

3

6403.80

0.32

Drill rods

ea

508.00

0

203.20

0.01

ANFO

kg

0.97

12 737

12 355.18

0.62

Primers

ea

4.40

26

114.40

0.01

metres

0.71

786

558.06

0.03

ea

20.00

26

520.00

0.03

Total supplies

20 462.64

1.02

Miscellaneous items (10%)

11 377.22

0.57

2502.99

0.13

Detonation cord Assays

General and administrative (2%) Total operating costs

127 652.43

Total operating costs per tonne of ore

6.38

CONCLUSIONS We have gone to considerable detail in some areas of the estimate, while we have seriously shortcutted some others. If we were to carry this estimate on to a complete feasibility study, a number of additional cost items may have to be considered, such as permitting costs, taxes, home office overhead, and of course, milling and smelting costs. All in all we have come up with a reasonable estimate in sufficient detail to set the stage for optimisation studies for the mine operation itself. Optimisation to maximise NPV is an important step in selecting a mine production rate and a primary equipment fleet. The optimisation process should be based on carefully prepared estimates MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

78

THE ROLE OF COST ESTIMATING IN MINE PLANNING AND EQUIPMENT SELECTION

TABLE 3 Capital cost summary: example mine – 20 000 tonnes per day ore, 40 000 tonnes per day waste (2:1 stripping ratio waste:ore). Date of costs: May 2009 Equipment

Specifications

Units required

Units

Cost/unit (US$)

Extended cost (US$)

Wheel loader

800 hp, 11.5 m3 bucket

1

ea

1 807 100

1 807 100

3

Front shovel

1550 hp, 17.0m bucket

1

ea

4 396 000

4 396 000

Haul trucks

938 hp, 90 tonne, rigid frame

15

ea

1 128 000

16 920 000

Rotary blasthole drills

15.2 cm hole diam, 52.7 m hole depth approx 27 000 kg pulldown, 500 hp

3

ea

670 000

2 010 000

Road maintainer

4.3 blade, 270 hp

1

ea

445 000

445 000

Dozers

4.5 blade, 350 hp

4

ea

731 000

2 924 000

459 kg/min

1

ea

74 800

74 800

16 kw

Bulk explosives truck

4

ea

21 900

87 600

Fuel/Lube truck

Lighting plants

2

ea

55 400

110 800

Mechanic’s truck

2

ea

67 000

134 000

Tire service truck

1

ea

158 000

158 000

65 hp, 3028 lmp, 45 m head

1

ea

23 670

23 670

Water truck

53 000 litre

1

ea

744 000

744 000

Pickup trucks

3/4 ton, four wheel drive

13

ea

23 600

306 800

Centrifugal pump

Total equipment capital cost

Capital cost (US$)

30 141 770 Preproduction development costs

Preproduction stripping Road construction

5 000 000

tonne

2.13

10 650.000

7983

metres

800.00

6 386 400

Total preproduction deveopment costs

17 036 400 Buildings and structures

Office building

770

sq m

1345

1 035 650

Dry

510

sq m

1011

515 610

Mechanical shop

1010

sq m

1119

1 130 190

Warehouse

905

sq m

635

574 675

Explosive storage facility

130

sq m

700

91 000

100 000

100 000

Fuel storage/dispensing facility

1

Total buildings and structures

3 447 125

Engineering, procurement and construction management (10%)

5 062 530 Other capital costs

End of project reclamation

2 000 000

Sustaining capital

15 000 000

Working capital

7 656 000

Total other capital costs

24 656 000

Contingency (20%)

16 068 765

Total capital cost

96 412 589

of capital and operating costs that include productivity analyses of the primary equipment fleets. Separate cost estimates should be completed for each production level and each equipment fleet to be tested. Itemised cost estimating techniques should be used that give due consideration to the impacts that changes in equipment size, number and type have on seemingly extraneous cost centres such as haul road and facilities design and construction, personnel requirements, and supervisory needs. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

79

J B LEINART AND O L SCHUMACHER

Time constraints placed upon the estimator may make it difficult to complete multiple estimates in such detail, unless appropriate software is utilised. Sherpa software (Stebbins, 2010b) incorporates itemised cost estimating methodologies similar to those described above, thereby providing the necessary detail, while considerably reducing the time required to complete an estimate.

REFERENCES Camm, T W, 1991. Simplified cost models for prefeasibility mineral e aluations, USBM Information Circular 9298. Camm, T W, 1994. Simplified cost models for prefeasibility mineral e aluations, Mining Engineering, 46(6) 559-562. Caterpillar n , 2007. Caterpillar Performance Handbook online , 452 p. Available from ebookee.com/Caterpillar-Performance-Handbook-Edition-36- 682132.html . Caterpillar n , 2010. F off-high ay truck specifications online . Available from com/servlet/ImageServlet imageId C611389 Accessed 15 October 2010 . nfoMine USA Inc, 2009a. E uipment cost calculator online . Available from com .

http //www.

http //xml.catmms. http //www.costmine.

nfoMine USA Inc, 2009b. Mine and mill e uipment costs, an estimator s guide online . Available from http //www.costmine.com . nfoMine USA Inc, 2010a. Coal cost guide, Spokane Valley online . Available from com .

http //www.costmine.

nfoMine USA Inc, 2010b. Mining cost ser ice, two volumes online . Available from com .

http //www.costmine.

Mular, A L, 1982. Mining and Mineral Processing E uipment Costs and Preliminary Capital Cost Estimations, special volume 25, 265 p (Canadian Institute of Mining and Metallurgy Montreal). Mular, A L and R, Poulin, 1998. CAPCOSTS A Handbook for Estimating Mining and Mineral Processing E uipment Costs and Capital Expenditures and Aiding Mineral Project E aluations, special volume 47 (Canadian Institute of Mining and Metallurgy Montreal). ara, T A, 1980. Quick guides to the evaluation of orebodies, CIM Bulletin, 73(2) 87-99. ara, T A, 1981. Mine evaluation, Mineral Industry Costs, chapter 6 pp 88-99 (Northwest Mining Association Washington). ara, T A and Suboleski, S C, 1992. Costs and cost estimation, Mining Engineering Handbook, Chapter 6.3, second edition, volume 1 (ed H L Hartman), pp 405-424 (SME Littleton). SMeans, 2009. Building construction reedconstructiondata.com/ .

cost

data

online .

Available

S huma her, O L, 2010. Estimating the cost of mining online . Available from

from

http //rsmeans.

www.edumine.com .

Stebbins, S A, 2010a. Apex economic analysis software for mining projects online . Available from http //www.aventurineengineering.com . Stebbins, S A, 2010b. Sherpa cost estimating software for surface mines online . Available from www.aventurineengineering.com .

http //

Stebbins, S A, 2010c. Sherpa cost estimating software for underground mines online . Available from http //www.aventurineengineering.com .

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

80

Ultimate Pit Limit of Iron Ore Mines using Maximum Flow Algorithms M Osanloo1, M Rahmanpour2 and A Sadri3 ABSTRACT A major problem in open pit mine planning is to determine the optimum pit limit of the mine. The optimum pit limit is the one with maximum discounted net profit and satisfies technical conditions. A number of optimisation techniques have been proposed for determination of pit limits. Among these techniques, those based on graph theory are suitable to solve the three-dimensional (3D) problem. In this paper, two maximum flow methods named, Dinic’s algorithm and Push-Relabel algorithm have been used to find the optimum pit limit of an iron ore mine. This method has been implemented in Gol-e-Gohar iron mine in south east of Iran in order to find the final pit limit. Issues, like properties of the mine and orebody, are studied. These properties include wall slope, sulfur and phosphor grade restrictions. The result shows that these properties have significant effect on the determined pit limit.

INTRODUCTION Iron ores are rocks and minerals from which metallic iron can be economically extracted. Sometimes iron ore contains some other minerals such as sulfur and phosphor which make problems in processing the ore. Phosphor (P) has four major effects on iron: increased hardness and strength, lower solidus temperature, increased fluidity and cold shortness. Depending on the use intended for the iron, these effects are either good or bad. Sulfur dissolves readily in both liquid and solid iron at the temperatures present in iron smelting. The effects of even small amounts of sulfur are immediate and serious. Sulfur causes iron to be red or hot short (Gordon, 1996). Mine planning problem is to determine which block to remove at which time. The orebody is divided into fixed size rectangular part which is called a ‘block’ and the orebody will be called ‘Geological block model’. The size of blocks, depends on the exploration drilling pattern, orebody geology and mine equipment size. After determining the block model dimensions, geological characteristics of each block are assigned using inverse distance, Kriging or other techniques. Using financial data (selling price and costs), the economic value of each block is calculated. It should be noted that the cost of accessing the block is not included in calculating the block value. Mine planning is to maximise the net present value of the pit and is a large scale mathematical optimisation problem that can not be solved using available commercial soft wares. Determination of final pit outline is one of the major parts of mine planning problem. The solution to the final pit limits problem is usually used to justify the project economically and it is also a guide to the mine planner to locate the mine site facilities. Based on the physical and operational constraints, the ultimate pit limits represent the set of blocks in the pit that maximises the total profit of the mine. Determining the pit limits is based on the net present value of each block. The net present value of each block, let’s call it block value, is the difference between the total value of the extracted mineral block and the cost of extracting that mineral from the mine and processing the blocks. Block value is calculated through this formula: Block value = (metal content) × price × recovery - processing costs - mining costs

(1)

1. Professor, Amirkabir University of Technology, Department of Mining and Metallurgical Engineering, Hafez Street, Tehran. Email: [email protected] 2. Msc Student, Amirkabir University of Technology, Department of Mining and Metallurgical Engineering, Hafez Street, Tehran. Email: [email protected] 3. Lecturer, Amirkabir University of Technology, Department of Mining and Metallurgical Engineering, Hafez Street, Tehran. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

81

M OSANLOO, M RAHMANPOUR AND A SADRI

If metal content of a block is zero or not enough to extract then block value will be negative, so it is a waste block otherwise it has a value larger than zero and it is an ore block. This block value is calculated for the present time and time value of money is not considered. When the ultimate pit limit is determined it means that the value of ore blocks can pay for the extraction of overlaying waste blocks. Inside the pit limit, mining operations is economical and outside the pit limit, open pit mining is not likely to be profitable. There are different methods to determine the optimum pit limit and these methods are divided into Heuristic and Rigorous methods (Kim, 1979). Heuristics methods use short cuts to solve the ultimate pit limits problem and heuristics such as moving cone technique are not likely to determine the optimum solution of the ultimate pit limit problem. These methods are easy to implement but it should be noted that they lack rigorous mathematical proof. Pana introduced the floating cone technique. This technique creates a cone for every ore block and searches for any cone with positive value. Korobov attempted to improve the Pana’s moving cone method (Kim, 1979). Lerchs and Grossman used the concept of dynamic programming and introduced their well known two-dimensional (2D) algorithm in order to find an optimum pit limit (Lerchs and Grossman, 1965). Later, Johnson and Sharp (1971) used this idea in order to design a 3D pit. Koenigsberg (1982) used dynamic programming in order to find the optimum pit limit in 3D format. It was also Lerchs and Grossman (1965) who introduced the use of Graph theory in optimum pit limit design. They modeled the block model of a mine as a graph and determine the ultimate pit limits by solving for the maximum closure of the graph. Zhao and Kim (1992) used this idea with some modifications and they claim that their method is simpler and faster than the Lerchs and Grossman. Tolwinski and Underwood (1998) used the concept of graph theory and mathematical programming and solve the ultimate pit limit problem as a dual simplex one. The concept of integer programming in pit limit design problem was first used by Meyer (1969). He made so many hypotheses in his work which results in a non optimal solution. Huttagosol and Cameron (1992) dealt with pit design as a transportation problem. Pit parameterisation is also an innovative method by Matheron (1976). In this method they search for a parameterisation function instead of directly searching for the pit limit. Denby and Schofield (1994) combine the pit limit determination and production planning by using genetic algorithm. Another method of solving the ultimate pit limits is by using network flow algorithm. Johnson (1969), Picard (1976), Giannini (1991), Yegulalp et al (1992) and Hochbaum and Chen (2000) used this algorithm to determine the pit limit. They solve the pit limit problem as a maximum flow problem on a network. In the next section, the basics of network flow theory will be discussed.

MAXIMUM FLOW THEORY The open pit mining problem can also be solved by any of maximum flow algorithms. Maximum flow algorithms that are exceptionally efficient have been developed over the last years. The maximum flow problem is one of the most fundamental problems in network flow theory and has been investigated extensively. This problem was first formulated by Fulkerson and Dantzig (1955) and Dantzig and Fulkerson (1956), and solved by Ford and Fulkerson (1956), using their well known augmenting path algorithm. Since then, a number of algorithms have been developed for this problem. Maximum flow theory is one of the most widely used methods in operation research topics and one of its applications is to determine the optimum pit limits. In order to use maximum flow theory to determine the pit limit, the main step is to convert the block model into a network. Each block is shown as a node in the network and will be connected to its overlaying blocks. The arcs of the network are created using technical constraints (Figure 1). It means that, if we want to extract block (v), we must first extract its overlaying blocks a, b and c. These constraints can also be used in order to create a 3D network. Consider an orebody defined by means of ore and waste blocks of given dimensions and each block is assigned a value. The following steps should be undertaken to convert the block model into a network (Yegulalp and Arias, 1992): 1. Connect all nodes (blocks) to the nodes (blocks) with respect to the mining constrains. Assign infinite flow capacity to these arcs. This step will convert the block model into a directed graph. 2. Add two nodes named source and terminal (sink) to the directed graph. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

82

ULTIMATE PIT LIMIT OF IRON ORE MINES USING MAXIMUM FLOW ALGORITHMS

FIG 1 - Technical constraints in order to extract the block v.

3. Connect the source node to each node corresponding to (+) valued blocks with an arc whose maximum flow capacity is equal to the value assigned to that block. 4. Connect all nodes corresponding to (-) or zero valued blocks to the sink with an arc whose maximum flow capacity is equal to the corresponding (-) or zero valued blocks. After these steps, the block model will be converted into a network. As an example consider the 2D block model shown in Figure 2. The resulting network is shown Figure 3.

FIG 2 - An 4 × 7 block model.

FIG 3 - The network describing the block model shown in Figure 2.

This creates a network through which we seek to push the maximum amount of flow from the source to the terminal (sink). All arcs are directed arcs permitting flow only to the direction of the sink. As mentioned above, there are many algorithms to determine the maximum flow in a network. These algorithms are divided into two major groups (Ahuja, Magnanti and Orlin, 1993): MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

83

M OSANLOO, M RAHMANPOUR AND A SADRI

1. augmentation of flow along a path, and 2. preflow push algorithms. In group 1, there is a mass balance on every node. It means that the amount of flow entering a node is equal to the flow exiting that node. Dinic’s algorithm is a good example of this group. In group 2, there is no need for mass balance at the initiation of these algorithms and every node reach a mass balance at the termination of the algorithm. The generic pre flow push algorithm is a member of this group. Here we use these two methods to find the final pit of the Gol-e-Gohar iron ore mine 2.

Dinic’s algorithm Dinic’s algorithm is one of the fastest algorithms in calculation of maximum flow problem and is based on the following definition (Dinic 1970): given a network G and an s-t-flow of f, the level graph GL of G is a graph in which: ^V^Gh, "^e

= ^ x,yhh e E^G f h: distGf ^s,x h+1 = distGf ^s,yh,h

(2)

The steps of Dinic’s algorithm, are as follows: 1. Set the flow in every arc to zero. 2. Construct the level graph GL of G. 3. Find a blocking s-t-flow f’ in GL. If f’ = 0 then stop. 4. Augment f by f’ and go to step 2. In the given network (G), an s-t-flow of f is called ‘blocking’ if the network (V(G), {e  E(Gf): f(e) < u(e)}) contains no s-t-path. Dinic’s algorithm programming is simple in comparison with other algorithms. The main step in implementation of this algorithm is to define the layered network (GL). The steps to make a layered network are: 1. V0 = {source} and i = 0. 2. Construct Vj. If u  Vi, i  j then Vj contains arc u-v. 3. If Vj is empty, then flow is maximum. 4. If Vj contains ‘sink’, then Vj+1 = {sink}. 5. Increase i to i+1 and go to step 2. Implementing this algorithm will result in a layered network which describes the blocks that are in the final pit.

The generic preflow push algorithm In this algorithm, instead of augmenting flow along an s-t-path, it pushes excess flow away from active nodes. Active node is a node in which the flow entering the node is larger than the flow exiting that node. This algorithm pushes flow along the arcs called admissible arcs. Admissible arc is the arc (i, j) if these two conditions are true (Ahuja, Magnanti and Orlin, 1993): 1. capacity of the arcs are positive ri j >0, and 2. d(i) = d(j) + 1. The generic preflow-push algorithm steps are as follows: x begin, x preprocess, and x while the network contains ‘active node’ do: x begin, x select an active node i, x push- Relabel node i, and x end; x end. In ‘preprocess procedure’, some initiation is made to start the algorithm and in this step the following will be done: x preprocess procedure, x begin, MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

84

ULTIMATE PIT LIMIT OF IRON ORE MINES USING MAXIMUM FLOW ALGORITHMS

x x x x

compute distance label d(i), for all arcs coming out of source node put xs,i = cs,i, d(s) = n, and end. In ‘preflow-push procedure’, every active node will be analysed in order to push away the excess flow in the node. This procedure is also called node examination and its steps are as follow: x push-Relabel procedure, x begin, x if the network contains an admissible arc (i, j) then: x push delta amount of flow on the arc: delta= minimum (e(i), ri j); x else replace d(i) by minimum (d(i) + 1: (i, j) ∈A (i), ri j > 0), and x end The other preflow-push algorithms are based on the generic algorithm described above. The main step in using maximum flow theory is to create the network.

BLOCK MODEL DESCRIPTION In this paper, we implemented the algorithms to determine the pit limits of Gol-e-Gohar iron mine Number 2 in south-east of Iran. The block model designed for this mine has got up to 588 000 blocks. In Table 1, we brought up the block model properties (Kosha-Madan, 2008). Each block contains information about its sulfur, phosphor and Fe content.

TABLE 1 Block model properties. Number of positive or ore blocks

8259

Block dimensions

Total block

metre

Number of blocks

X axis

10

210

Y axis

10

140

Z axis

15

20

58 8000

DETERMINING THE FINAL PIT Constructing the network is the essential part of determining pit limit using maximum flow theory. Every network is consisted of nodes and arcs. According to geotechnical considerations, the pit slope is decided to be 45° in this mine. Arcs are generated in order to produce the pit slope and also to show the technical constrains. Arc generation pattern is produced using the minimum search pattern (Caccetta, 1988). This pattern is named 1 - 53 and it means that every node in the network is connected to 53 nodes above it. This pattern guaranties the wall slope of approximately 45°. After constructing the network, we used the Dinic’s and ‘Push-Relabel’ algorithms in order to determine the final pit limits of Gol-e-Gohar iron ore mine. The programs were developed in C++ language and were run on a Dell laptop set4. The pit produced by these methods are the same and we name this pit as pit number 1, but the difference is the running time of these algorithms. According to Table 2, running time of push-relabel algorithm is faster than Dinic’s algorithm. The trend in determination of pit limit is based on the blocks undiscounted value but in here we want to find a pit limit based on the blocks discounted value. To consider the effect of time value of money in determination of final pit, we should first calculate the sequence of block extraction. The sequence shows the time of extracting each block and using this time we can calculate discounted value of each block. The sequence of extraction is calculated using NPV-S software. In determination of block sequence, blending optimisation is considered, so the amount of sulfur and phosphor will be restricted and it will lead to a product with the best quality. 4. Dell Inspiron 1501- AMD processor +3500, 512 Mb of RAM. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

85

M OSANLOO, M RAHMANPOUR AND A SADRI

TABLE 2 Running time of algorithms. Running time (s) Dinic

Push-Relabel

131

95

As the time of extracting each block is revealed, the discounted value of each block can be calculated. Using these values and network flow theories, will result in a new pit which is smaller and inside the pit number 1 and we name this pit as pit number 2. In Figure 4, these two pits are shown. Pit 1 contains 8222 ore blocks and pit 2 contains 8200 ore blocks. The total number of blocks in pit number 1 is 61861 and in pit number 2 there is 60 864 blocks. The tonnage difference between these two pits is about 3 Mt.

FIG 4 - Plan view of pit 1 and pit 2 are shown together.

The coloured blocks in this figure, indicates those blocks which are not included in pit 2.

CONCLUSIONS Network flow theory is a great tool for determination of optimum pit limit and it is superior in solving the ultimate pit limit not only in running time but also in the simplicity of its programming. In the other hand, for solving the ultimate pit limit problem in 3D, using a network flow method could be more helpful and applicable. The trend described here is capable of determining the optimum pit based on blocks discounted values. The resulting pit, is a exact and real pit.

REFERENCES Ahuja, R K, Magnanti, T L and Orlin, J B, 1993. Network Flows: Theory, Algorithm and Applications, 864 p (Prentice-Hall Inc: Englewood Cliffs). Bongarcon, D F and Marechal, A, 1976. A new method for open pit design: Parameterization of the final pit contour, in Proceedings 14th International APCOM (ed: R V Ramani), pp 573-583 (Society of Mining Engineers of American Ins of Mining, Metallurgical, and Petroleum Engineers Inc: New York) Caccetta, L and Giannini, L M, 1988. The generation of minimum search pattern in the optimum design of open pit mines, The AusIMM Bulletin, 293(5):57-61 (The Australasian Institute of Mining and Metallurgy: Melbourne). Dantzig, G B and Fulkerson, D R, 1956. On the max-flow min-cut theorem of networks, Linear Inequalities and Related Systems, Annual of Mathematics study 38 (eds: H W Kuhn and A W Tucker), pp 215-221 (Princeton University Press: Princeton). MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

86

ULTIMATE PIT LIMIT OF IRON ORE MINES USING MAXIMUM FLOW ALGORITHMS

Denby, B and Schofield, D, 1994. Open pit design and scheduling by use of genetic algorithm, Transactions of the Institutes of Mining and Metallurgy, 103:A21-A26. Fulkerson, D R and Dantzig, G B, 1955. Computation of maximum flow in networks, Naval Res Log Quart, 2:277-283. Ford, L R and Fulkerson, D R, 1955. Maximal flow through network, Canadian Journal of Mathematics, 8:399-404. Giannini, L M, Caccetta, L, Kelsey, P and Carras, S, 1991. PITOPTIM: A new high speed network flow technique for optimum pit design facilitating rapid sensitivity analysis, The AusIMM Proceedings, 2:57-62 (The Australasian Institute of Mining and Metallurgy: Melbourne). Gordon, R B, 1996. American Iron 1607-1900, 344 p (The Johns Hopkins University Press). Hochbaum, D, S and Chen, A, 2000. Performance analysis and best implementation of old and new algorithms for the open pit mining problem, Operation Research, 48(6):894-914. Huttagosol, P and Cameron, R, 1992. A computer design of ultimate pit limit by using transportation algorithm, in Proceedings 23rd International APCOM, (ed: Y C Kim), pp 443-460 (SME: Tucson) Johnson, T B, 1968. Optimum open pit mine production scheduling, PhD thesis (unpublished), University of California, Berkeley. Johnson, T B and Sharp, W R, 1971. A 3-D dynamic programming method for optimal ultimate open pit design, US Bureau of Mines, report of investigation 7553. Kim, Y C, 1979. Open pit limits analysis: Technical overview, Computer Methods for the 80’s in the Mineral Industry (ed: A Weiss), pp 297-303 (AIME: New York). Koenigsberg, E, 1982. The optimum contours of an open pit mine: An application of dynamic programming, in Proceedings 17th International APCOM (eds: T B Johnson and R J Barnes), pp 274-287 (Society of Mining Engineers of American Ins. of Mining, Metallurgical, and Petroleum Engineers Inc: New York). Kosha-Madan, 2008. Mine planning report, internal report, Gol-e-Gohar, pp 80-87. Lersch, H and Grossman I F, 1965. Optimum design of open pit mines, CIM Bulletin, volume 58, no 633, pp 47-54. Meyer, M, 1969. Applying linear programming to the design of ultimate pit limits, Management Science, 16:B121-B135. Picard, J C, 1976. Maximal closure of a graph and application to combinatorial problems, Management Science, 22:1268-1272. Underwood, R and Tolwinski, B, 1998. A mathematical programming viewpoint for solving the ultimate pit problem, European J of Operational Research, 107:96-107. Yegulalp, T M and Arias, A J, 1992. A fast algorithm to solve the ultimate pit limit problem, in Proceedings 23rd International APCOM (ed: Y C Kim), pp 391-397 (SME: Tucson). Zhao, Y and Kim, Y C, 1992. A new optimum pit limit design algorithm, in Proceedings 23rd International APCOM (ed: Y C Kim), pp 423-434 (SME: Tucson).

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

87

A New Model to Improve Ore Grade Reconciliation Between the Exploration Model and the Mine A Parhizkar1, M Ataee2, P Moarefvand3, V Rasouli4 and A H Bangian5 ABSTRACT The poor reconciliation between the values of estimated and actual grades can cause major economic losses to the mining industry. In a mining operation grade reconciliation is comparing the values of estimated grade calculated in exploration stage with actual grade obtained from blastholes’ data. Many different factors affect the degree of reconciliation in a mining operation. In this paper we investigate the factors related to estimated grade which affects the reconciliation process in the exploration stage of the orebody. These factors are the sources of uncertainty in the rest of mine life. In this area the most important factors are inherent variability, statistical uncertainty and systematic uncertainty. A probabilistic model was presented to model each source of uncertainty. The implementation of proposed model was checked and its validity was investigated using real data in an iron open pit mine in Iran.

INTRODUCTION Ore grade reconciliation is a complicated process in a mining operation. A good reconciliation indicates that the selected method for estimation of ore grade has predicted ore grade successfully in exploration stage. The degree of reconciliation affects the accuracy of recoverable resource estimation and grade control procedure. These factors are the foundation of a successful mining venture and affect long, medium and short term production planning in a mining operation (Thomas and Snowden, 1990). Many different mines failed to estimate the ore grade accurately it means the procedure of grade estimation in exploration stage has been unsuccessful (Burmeister, 1988 Knoll, 1989 Clow, 1991). Rossi and Parker (1993) have shown 20 mines out of 39 mines failed to estimate the ore grade accurately. Similar reports have been published showing the impact of poor grade reconciliation on the economic condition of a mining company (Baker and Giacomo, 1998 Carrasco, Carrasco and Jara, 2004).Unfortunately large differences between reserve estimation and actual production are not unknown in the mining industry (Schofield, 2001). Schofield (2001), Morley (2003) and Noppe (2004) have highlighted the importance of accurate and precise estimation of ore grade for different types of orebodies. The aim of this paper is to develop a model to improve reconciliation between the exploration model and the mine.

TYPES OF ERROR IN EXPLORATION STAGE OF THE OREBODY As mentioned earlier, three types of uncertainty related to exploration stage affecting the reconciliation process are inherent variability, random uncertainty and systematic uncertainty, respectively. The natural variability of the orebody affects the estimation process of the ore grade. The nugget effect, in a variogram function, indicates the natural or inherent variability of the orebody. The random or statistical uncertainty usually exists due to the limited number of samples. The amount of this uncertainty decreases with increasing the number of samples. The influence of statistical uncertainty is on the precision of estimated grade. Limited number of samples causes the random 1. Academic Member, Department of Mining Engineering, Lahijan Branch, Islamic Azad University, Lahijan, Iran. Email: [email protected] 2. Academic Member, Department of Mining Engineering, Shahrood University, Shahrood, Iran. Email: [email protected] 3. Academic Member, Department of Mining and Metallurgical Engineering, Amirkabir University of Technology, Tehran, Iran. Email: [email protected] 4. Academic Member, Department of Petroleum Engineering, Curtin University, Perth WA 6102. Email: [email protected] 5. Academic Member, Department of Mining Engineering, Tehran South Branch, Islamic Azad University, Tehran, Iran. Email: [email protected] MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

89

A PARHIZKAR et al

uncertainty but there is another type of error which has no relationship with the number of samples. These errors exist because of differences between real (in situ) and laboratory conditions due to factors such as scale effect and anisotropy. The discrepancy between blasthole and borehole samples cannot be resolved by increasing the number of samples. This type of error is called systematic error.

THE PROBABILISTIC MODEL Statistically, the best estimate of an unknown parameter is the mean or expected value and the uncertainties can be expressed in the form of variance or standard deviation or coefficient of variation (COV). If a grade estimation method such as kriging, which considers the inherent variability of the grade parameter, used to estimate the ore grade then two correction factors will be required to reconcile the estimated grade with actual grade. This is written in the form of Ga

C r Cs Ge

(1)

where Ga and Ge

represent the actual and estimated grade, respectively and the natural variability is accounted for in estimated grade Cr and Cs are the correction factors applied to rectify statistical and systematic errors, respectively Using first order uncertainty analysis model the mean value of actual grade can be calculated as Ga

Cr CsGe

(2)

where the bars show the mean value of each parameter. Also the uncertainty can be described as coefficient of variation which is the proportion of the standard deviation to the mean value and hence the COV is dimensionless. This parameter can be defined for actual grade as CVGa ,

2 2 2 CVGe + CVCr + CVCs

(3)

In this equation, the CVs are the COV corresponding to each parameter. The uncertainty derived from inherent variability can be expressed as SG e (4) CVG e Ge where SGg is the standard deviation of the estimated grade The standard error of estimated grade in independent observations can be expressed as SG e (5) SEG e n The statistical uncertainty depends on statistical error which decreases with increasing number of samples. The COV for the correction factor of Cr can be calculated as below SEG e (6) CVC r Ge The combination of Equations 5 and 6 can be expressed as CVC

r

CVG

e

n

(7)

The mean value of Cr, C r , is taken as one because only random statistical error is considered. The scale effect and anisotropy are the two major factors causing systematic errors affecting reconciliation process. The mean value of correction factor accounting for systematic uncertainty, C s , can be calculated as MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

90

A NEW MODEL TO IMPROVE ORE GRADE RECONCILIATION BETWEEN THE EXPLORATION MODEL AND THE MINE

C s % im= 1C i

(8)

where is the mean value of Cis which are correction factors accounting for the ith systematic error Ci The COV of Cs, CVCs, can be quantified as CVCs

2 / im 1CVCi =

(9)

where CVCi is the COV of Ci There are two systematic errors affecting the reconciliation process and hence two correction factors would be required as C1 and C to account for scale effect and anisotropy, respectively. So Equations 8 and 9 can be rewritten as (10)

C s C 1C 2 CVC

s

CVC2 CVC2 1

(11)

2

The overall uncertainty of actual grade in the reconciliation process can be expressed as follows CVG b a

SG2

e 2 Ge

+

CVG2

e

n

+ CVC2 + CVC2 1

(12)

2

MODEL IMPLEMENTATION In this section the implementation of the proposed model in a real case study is presented. Chadormalou is a large iron deposit in Iran with 320 Mt of mineable reserve which is being mined using open pit mining method. In order to investigate the reconciliation condition in Chadormalou iron ore mine the reserve block model which was constructed using kriging estimator was used to compare the estimated grade with actual grade gathered from boreholes. The block model dimension is 25 m 25 m 15 m but the blastholes’ pattern is 6 m 7 m. Considering 3 m of subdrilling the blastholes’ depth is 18 m. In order to investigate the reconciliation condition in Chadormalou iron ore mine the reserve block model was used to compare the estimated grade with actual grade gathered from boreholes. The normal Q-Q plot of estimated and actual grades which is shown in Figure 1 shows the different structure of estimated grade and actual grade in Chadormalou iron mine. Figure 2 shows a comparison between estimated and actual grade of Chadormalou iron ore mine. This figure demonstrates the poor reconciliation between estimated grade, calculated using kriging estimator, and the actual grade, comes from blastholes data. The mining process suffers from such low degree of reconciliation especially for the purpose of a precise mine planning and design, production planning. Applying a cut-off grade of 55 per cent to each axis the misclassification of exploited materials to ore and waste can be determined. This is shown in Figure 2 where most proportion of misclassification is seen to belong to ore materials sent to waste dump mistakenly. The mean value and standard deviation of estimated grade, `G e and SGgj are equal to 50.69 per cent and 6.06 per cent, respectively. The COV of estimated grade equals 6.06 ( ) CVG 0.12 50.69 ( ) e The number of borehole samples used in estimation process is 2232. The statistical uncertainty which depends on the number of samples can be calculated from Equation 7 as CVCr

01

# 10-3

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

91

A PARHIZKAR et al

 FIG 1 - Q-Q plots of estimated grade and actual grade.

 FIG 2 - Grade reconciliation condition in Chadormalou mine. The straight line indicates 1:1 slope. The dash lines show a cut off grade of 55 per cent for each axis.

To quantify the systematic uncertainty it is necessary to determine the anisotropy exists in different area of the mine and then apply the required correction factors to remedy the sources of each systematic error such as scale effect and anisotropy. To calculate correction factors corresponding to scale effect one could consider an area with no anisotropy. This discards the anisotropy effect and therefore allows the scale effect on reconciliation process to be determined. Variogram, which represents the dispersion of variables, is a powerful function for studying anisotropy. The variogram maps were computed to distinguish the anisotropic areas of the Chadormalou deposit. Table 1 shows the statistical parameters of estimated grade, actual grade and the correction factors for anisotropic levels of the mine. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

92

A NEW MODEL TO IMPROVE ORE GRADE RECONCILIATION BETWEEN THE EXPLORATION MODEL AND THE MINE

The correction factor in Table 1 accounts for both scale effect and anisotropy. It is now necessary to find the proportion of anisotropy in correction factor which was cited in Table 1. Equation 10 can be rewritten as TABLE 1 The statistical parameters related to anisotropic area of the mine. Parameter

Estimated grade

Actual grade

Correction factor (Cs)

Mean

54.80

52.07

1.01

Variance

4.49

54.04

0.02

COV

25.86

7.08

7.14

C2

Cs C1

(13)

The mean value of C accounting for anisotropy can be calculated from Equation 13 as C2

1.01 1.18

0.86

The variance of C can be expressed as SC2

2

2 S2 2 2 2S + 12 SC2 + C s 1 + 2 C s Coy c 1 , Cs m + 1 SCs + ;Coy c 1 , Cs mE C C C C C1 C1 s 1 1 1 1

(14)

SC2 can be calculated using Equation 14 discarding small numbers as s

SC2

2

0.02 + (1.012 # 0.01) + 2 # 1.01 # 2.6 # 10-3 1.18 1.182

0.03

Hence The COV of C2 can be calculated as SC

0.03 0.17

2

Now the uncertainty of actual grade can be calculated using Equation 12. Replacing numerical values of parameters into Equation 12 gives the COV of actual grade as CVGa ,

01

+ 0 10 + 0 0 +

0 0

0

APPLYING THE CORRECTION FACTORS TO IMPROVE RECONCILIATION In order to analyse the impact of correction factors on improvement of grade estimation and hence the reconciliation process, each area with determined actual grade was considered. Variograghy analysis was down and for isotropic areas only a correction factor ^C 1h and for anisotropic areas two required correction factors ^C 1 and C 2h were applied to the estimated grade in order to improve the reconciliation process due to scale effect and anisotropy. Figure 3 is an overlay scatter to show the effect of correction factors on reconciliation process and the degree of misclassified materials to ore and waste in Chadormalou iron ore mine.

CONCLUSIONS A simple probabilistic model was presented in this paper to improve reconciliation between estimated and actual grade. This model can be applied to any type of ore deposits. In this study the developed model was applied in data taken from Chadormalou iron ore of Iran. The natural variability of grade parameter was considered in grade estimator but for each type of other uncertainties a correction factor was considered to apply to estimated grade values to reconcile them with the actual grade values. The amount of reconciliation was improved by applying appropriate correction factors to the estimated grade values for Chadormalou iron ore mine. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

93

A PARHIZKAR et al

 FIG 3 - The overlay scatter diagram of two pairs of data. The blue and red markers indicate actual-estimated grade and actual-modified estimated grade pairs respectively.

REFERENCES Baker, C T and Giacomo, S M, 1998. Resource and Reserve Their uses and abuses by the equity markets, Ore reserve and Finance A Joint Seminar between Australasian Institute of Mining and Metallurgy and AS , Sydney. Burmeister, B, 1988. From Resource to Reality A Critical Re ie of the Achie ements of e Australian Gold Mining Projects During the Period January 1 to September 1 (Macquarie University). Carras o, P, Carrasco, P and Jara, E, 2004. The economic impact of correct sampling and analysis practices in the copper mining industry, Chemometrics and Intelligent Laboratory Systems Journal, 74(1) 209-213. Clo , G, 1991. Why gold mines fail,

orthern Miner Maga ine, 2(4) 31-34.

noll, K, 1989. And now the bad news,

orthern Miner Maga ine, 4(6) 48-52.

Morley, C, 2003. Beyond Reconciliation – A Proactive Approach to using Mining Data, in Proceedings Fifth Large Open Pit Mining Conference, pp 185-191 (The Australasian Institute of Mining and Metallurgy Melbourne). Noppe, M, 2004. Reconciliation Importance of good sampling and data QA-QC, Mining and Resource Geology Symposium. ossi, M E and Parker, H M, 1993. Estimating recoverable reserves Is it hopeless Presented at the Geostatistics for the Next Century Forum, Montreal, Quebec, Canada, June 3 - 5, peer reviewed. S hofield, N A, 2001. The myth of mine reconciliation, Mineral Resource and Ore Reser e Estimation The AusIMM Guide to Good Practice (ed A C Edwards), pp 601-610 (The Australasian Institute of Mining and Metallurgy Melbourne). Thomas, M and Snowden, V, 1990. Improving reconciliation and grade control by statistical and geostatistical analysis Strategies for grade control, AIG Bulletin, 10 49-59.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

94

A New Genetic Algorithm for Open Pit Design – The Two-Dimensional Case J Saavedra-Rosas1 ABSTRACT The design of a mine should aim to optimise the market value by maximising revenues whilst minimising costs. In open pit mining, the nature of the extraction technique translates this objective into the maximum recovery of reserves while minimising waste extraction subject to technical, economical and safety constraints. Since the introduction by Lerchs and Grossman, in 1965, of a model that is now considered a classic in the area, open pit design has reached a level of maturity and acceptance by industry. These models are being used on a regular basis by designers and planners. This paper proposes a new alternative model for open pit design. It is based on genetic algorithms (GA) and exploits a reduction of the number of decision variables of the more traditional techniques. The novel approach provides more flexibility than traditional models; it can also be extended to consider planning problems that traditional open pit design models are not capable of handling. The algorithm has been designed for two dimensional problems and tested in a simple instance. The results obtained in this simple test are encouraging and are a good supporting argument for the development of a GA model capable of handling three-dimensional (3D) instances. The model presented in this paper is just a first step towards a full GA to solve the pit design problem and it is believed that novel tools like the one presented herein could become great allies in tackling challenging problems in open pit design, an illustration of such a challenging problem is presented and discussed.

INTRODUCTION In open pit mining, pit optimisation is the heart of the mine investment. The final pit outline or optimised pit shell must in the final analysis reflect the profitability of the mine investment (Songolo, 2010). Since the introduction by Lerchs and Grossman (1965) of a model that is now considered classical in the subject, open pit design has reached a level of maturity and acceptance by industry becoming the de facto standard. Open pit optimisation is the determination of the ultimate pit or optimal pit limit for a given deposit under given set of mining and economic constraints (Schofield and Denby, 1993). In this paper, a model based in a genetic algorithm (GA) is introduced. It is mainly based on a simplification of the open pit planning problem by considering pillars or columns rather than a block model. The approach allows for a more natural expression of the slope angle constraint and also reduces the number of variables to use. The price to pay for such simplification is the use of a tool that does not incorporate a certificate of optimality as an ending criterion. The model presented in this paper has been designed and tested in two dimensional instances; the rationale behind this decision is that of keeping things simple to see if they work rather than investing time of developing a much more challenging model that could not necessarily work. The results obtained in the simple test performed are encouraging to continue with the research. It is important to note that the model presented in this paper is just a first step towards a full GA to solve the pit design problem; obviously further work has to be conducted to extend the GA model to three dimensions but under the light of the experience already gained is not deemed an impossible task. Surprisingly enough, an approach similar to the one used in this paper has been proposed a long time ago by Meyer (1969). In his work he proposes the use of a mathematical model that uses 1. Lecturer, Western Australian School of Mines, Curtin University of Technology, Locked Bag 30, Kalgoorlie WA 6433. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

95

J SAAVEDRA-ROSAS

variables representing a maximum length of a pillar to be mined. The Slope constraints are specified in a manner similar to what is proposed in the current article. More recently Goodwin et al (2006) go back to the same idea but slightly modified. As GAs are not a common use tool they require a proper introduction and this is done in the upcoming section. The rest of the paper is organised as follows; a section that review the use of GAs in open pit planning, basically the work done by Schofield and Denby (1993) and Denby and Schofield (1994, 1996), a section that introduces the proposed GA model for open pit planning, a section containing a simple numerical experiment to validate the model, a section describing some ideas related to the extension of the model to three dimensions and finally the conclusions which also contain a discussion of potential research avenues and some applications of a model like the one introduced here are discussed.

GENETIC ALGORITHMS Genetic algorithms were introduced by John Holland (Holland, 1975). They are inspired in Darwin’s mechanisms of natural selection. Such mechanisms establish that an individual is generated as a mixture of the genes from his parents by means of crossover, added to this process of mixture there is a process called mutation (change in some segments of genetic material). The evolution which usually starts from a population of randomly selected individuals happens in generations. In each generation, the fitness of every individual in the population is evaluated; individuals are selected from the current population based on their fitness values and recombined by means of operators to form a new population. The new population is then used in the next iteration of the algorithm until an optimal solution is obtained or some stopping criterion is reached (Songolo, 2010). Finally, the adaptation to the medium makes that some individuals survive and inherit their genes to his sons. The general form of an Evolutionary Algorithm is presented in Algorithm 1 (inspired in Bodenhofer, (2000) but with modifications). Algorithm 1 Genetic algorithm Require: Fitness function, crossover probability, mutation probability Ensure: A candidate optimal solution if exists Compute initial population B0 while Termination criteria is not reached do Select individuals for reproduction Create offspring by crossing individuals Occasionally mutate some individuals Compute new generation end while The main aspects to have in mind when implementing GAs are chromosomal representation, fitness function and crossover, and mutation operators. Koza (1992) and Michalewicz (1996) prove to be fundamental and provide an in depth coverage of the topic. In the remaining part of the section we will illustrate the genetic algorithm for a simple case so understanding can be gained. What follows in this section is partially based in Saavedra (2009). The simplest representation for solutions is based in what is called binary encoding. Each potential solution is represented as a string of binary digits. Representation of solutions by means of a specific encoding is called chromosome. The selection of individuals for reproduction can be accomplished in several ways; the most common being roulette selection. In this selection procedure, each individual of the population is given a weight based on his fitness value (the ability of the solution to solve the problem). For example, if a fitness function f() is being maximised and a population of potential solutions is denoted by {x } i d I ; then the weight for the individual k can be defined as: wk: =

f {xk} / f {xi}

idI

provided that / f (xk ) ! 0. idI

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

96

A NEW GENETIC ALGORITHM FOR OPEN PIT DESIGN – THE TWO-DIMENSIONAL CASE

The crossover operation is a procedure that interchanges genetic material between chromosomes and generates new individuals. There are several variants available. We will illustrate the procedure by using the simplest one. It has to be noted that crossover is not applied indiscriminately but just on a certain percentage of the cases (which is another parameter of the operation of the GA). If two parents are effectively selected for crossover, the procedure is then: x Select at random a position in the interval {1, …, n} (with n being the size of the individual), suppose that position p is selected; p is referred as to the crossover point. x Decompose the mating individuals as: x = (x1, ..., x p ; x p+1, ..., xn) and, y = (y1, ..., y p ; y p+1, ..., yn) . x Form the new individuals: a = (x1, ..., x p ; y p+1, ..., yn) and, b = (y1, ..., y p ; x p+1, ..., xn) . Mutation is an unary operator that modifies portions of a given chromosome. It must be applied following the same rules for applicability of crossover by using the mutation probability for this end. There are several variants of it the most common being the one point mutation: x Select at random a position in the interval {1, …, n} (with n being the size of the individual), suppose that position p is selected; p is referred as to the mutation point. x Flip the bit at position p, ie for the individual x if the value at position p is one then we change this to zero and vice-versa. For computing the next generation there are two basic common mechanisms. The first one puts the parents and the offspring together and selects the best |I| individuals (with I being the set representing the indexes for denoting elements of the population). The other method eliminates the parents and keeps only the offspring. More elaborate generation forming mechanisms can be defined. The termination criterion is usually specified in terms of fitness improvement between generations; usually we can decide to stop if no improvement has been appreciated during the last q iterations of the algorithm. Another alternative is to simply stop after a fixed number of iterations. As we can see GA can be highly customised but fine tuning of the operative parameters must be decided in order to avoid premature stopping or an excessive number of iterations. There is a trade-off between the quality of the solutions and the number of iterations required for getting them.

GENETIC ALGORITHMS FOR OPEN PIT PLANNING This section discusses the known papers for the area of application of GA into open pit lay out, little work has been done which certainly shows the predominance of more traditional techniques for open pit layout. Schofield and Denby (1993) propose the first application of GA for open pit planning. In their work, a relatively large population of random schedules – also including a small number of highly constrained schedules – was used. According to the authors, the size of the population has an effect on the performance of the algorithm. The proposed GA was named GO-PIT. As per fitness function, the proposed GA carries out a discounted cash flow calculation for the simulated pit on the basis of a set of user definable functions that linked a number of factors such as grade, cut-off grade, recovery factors, extraction timing, mining and processing costs, discount factors and mineral market values. Noticeably, the fitness function is totally independent of the genetic algorithm operators, which is one of the key advantages of the genetic algorithm approach. Thus there is a possibility of having either simple cost or revenue functions or highly complex fitness MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

97

J SAAVEDRA-ROSAS

functions without modifying the basic optimising system and on this basis the GO-PIT system has the potential to easily be customised for different situations. In terms of reproduction, it has been found out by the researchers that the scheduling constraints get violated in the system, for example, production tonnages during a scheduling period become greater than feasible or block precedence relationships being broken. This being a major difficulty that arises when pits are produced or modified by one of the genetic operators, reproduction in GOPIT is based on a linear normalisation technique. This ensures that the pits are ranked in order of fitness and their probability of reproduction is mapped onto a linearly decreasing scale from high to low probability. Thus, it is less likely in this way to have a good schedule that dominates the population quickly and reduce the chance of the system from converging quickly to a false optimum. As regards risk incorporation, Schofield and Denby (1993) stresses that the fitness function is decoupled from the optimising mechanism making it one of the highly promising features of a genetically based optimising system for pit design and scheduling. All this entails is that the measure of the fitness function can be modified extensively without the principal parts of the genetic algorithm requiring any alteration. Denby and Schofield (1994) present an updated discussion of their 1993 paper improving their numerical results for the numerical examples considered the year before. The basics of the reasoning and the algorithm remain mostly unchanged. Is important to note that the underlying model of the 1993 and 1994 papers if for two-dimensional deposits only. In Denby and Schofield (1996), an extension to 3D deposits is presented. Much of the discussion remains along the same lines of what has been presented before with the addition of the ability to handle variable internal slopes angles. Still in this version there is a constant generation of infeasible individuals after crossover and because of that normalisation is still widely used. It is unfortunate that no details are given on any of these papers as to how the individuals were generated or how the genetic operators were applied. The discussion is always in general terms and the only clue provided mentions that the chromosomal representation involved discarding the binary string concept changing it to a multidimensional representation capable of storing the 3D spatial data of the problem. It is believed that the representation used was consequently a 3D array.

PROPOSED GENETIC MODEL FOR OPEN PIT LAYOUT Chromosome structure The proposed approach abstracts the block model as a set of columns, a similar idea is found in Goodwin et al (2006). This abstraction reduces the size of the problem as it actually converts it to a one-dimensional (1D) one instead of a two-dimensional (2D) problem. The classical slope angle constraints have been specified by taking the differences between two neighbouring columns, for example: ; xi (t) - xi+1 (t) ; # f

6t

indicates that the difference in depth between columns i and column i + 1 (denoted by x•(t)) cannot exceed ε (tolerance) at any given time. For the pit layout problem, the final depth of each column has to be determined. Consequently, the genetic algorithm encoding will consist of a real-valued vector of columns where each column indicates the depth up to which the column will be extracted (or an equivalent measure of it). More specifically, for each column i the variable xi will be a number between zero and one that will specify the percentage of the column extracted. To illustrate the encoding, consider the following chromosome: [0, 0.1, 0.2, 0.3, 0.2, 0.3, 0.2, 0.3, 0.2, 0.1] If the depth is 250 m then this chromosome will translate into the following depths of extraction: [0 m, 25 m, 50 m, 75 m, 50 m, 75 m, 50 m, 75 m, 50 m, 25 m] The last chromosome has an open pit counterpart that is illustrated in Figure 1. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

98

A NEW GENETIC ALGORITHM FOR OPEN PIT DESIGN – THE TWO-DIMENSIONAL CASE

 FIG 1 - Open pit representation of the chromosome [0 m, 25 m, 50 m, 75 m, 50 m, 75 m, 50 m, 75 m, 50 m, 25 m].

The chosen an approach has the additional benefit of providing independence of the problem formulation from block model geometry. Classical models such as Lerchs and Grossmann decide at a block level if extraction for that block will proceed or not. This conduces to layouts that are mainly defined in terms of the block model geometry dimensions forcing bench sizes based on the block model or on the contrary potentially forcing the block model to follow the bench design. Traditional models usually use slope constraints of 45° and the way to implement different slope angle constraints are based on considering blocks that are larger in the vertical axis. With the approach taken on this paper the slope constraints can be enforced for any slope angle (not just 45°) being that decision independent of the block model. The block model then can be established based on geostatistical and geological considerations rather than exploitation/design constraints thus avoiding undesirable effects such as volume-variance effect.

Fitness function The fitness function calculates the income generated by the ore extracted in each column subtracting the cost incurred to extract that ore. It has to be noted that if there is extraction at the border of the bounding box for the mineralisation then the pit design has to be extended by using waste as illustrated in Figure 2. The formula used for the calculation of the fitness function is given by: n

f^ x h = / xi bi - C (WLV + WRV + EOV) i =1

where: bi is the benefit gained by exploiting the column i C is the cost for extraction

 FIG 2 - Elements used in the calculation of the fitness function. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

99

J SAAVEDRA-ROSAS

WLV WRV EOV

is the volume of waste to the left is the volume of waste to the right is the volume of extracted ore

Crossover The crossover operator defined for the chromosomal representation take two individuals, selects a crossover point at random and interchanges the genetic material after the crossover point. Care has to be put of ensuring the feasibility of the resulting children, for such reason a reparation procedure has to be implemented after the exchange of genetic material. The reparation process basically makes the existing portion to be compatible with the new portion that has been added. The reparation process considers the next column to the right of the insertion point and checks whether the difference in height between columns is compatible (ie satisfying the slope constraint), if they are not then a new height within the bounds of what should be permissible is draw at random for the offending column, ie: xi + 1 ~ U[xi - , xi + ] with U being a uniform random variable. The same process is carried forward for all the columns to the right taking them one by one. Because of the characteristics of this process it has been named propagation. A graphical example of a crossover applied to two individuals is presented in Figure 3.

FIG 3 - Example of the application of the proposed crossover operator, on the left column the original parents are presented, in the middle column the crossover point is applied separating both parents in two pieces, finally in the last column the resulting children are shown.

Mutation operator The mutation operator simply chooses a column at random and changes the height of extraction for that column assigning to it a value at random. The columns to the left and to the right of the mutated column have to be repaired to make the mutated individual compatible. The procedure used is similar to the one described in the previous subsection with the difference that in the case of this operator the propagation process is bi-directional (right and left of the mutation point have to be fixed). For a graphical illustration of the process consider Figure 4.

NUMERICAL EXPERIMENT The proposed GA was programmed in using the Python language. Python (Van Rossum and Drake, 2009) is an easy to learn, powerful programming language which has efficient high-level data structures and a simple but effective approach to object-oriented programming. Python is an interpreted, interactive object-oriented programming language suitable (amongst other uses) for distributed application development, scripting, numeric computing and system testing. Because Python’s elegant syntax and dynamic typing, together with its interpreted nature, makes it an ideal MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

100

A NEW GENETIC ALGORITHM FOR OPEN PIT DESIGN – THE TWO-DIMENSIONAL CASE

FIG 4 - Example of the application of the proposed mutation operator. In the leftmost figure a column is selected for mutation (shown in black), in the central figure the column value has been changed and it is noted that the column to the left of it is inconsistent with the slope constraint (column depicted in grey), finally in the rightmost figure the inconsistency is solved by changing the value of the offending column. All the other columns are consistent so no further changes are needed.

language for scripting and rapid application development in many areas on most platforms, certainly its use has eased the enormous work required to produce an open pit optimiser based genetic algorithm. The example used for the study is based on a simple block model consisting in eight columns of 10 m having a total height of ten blocks of 10 m each. The distribution of grades is presented in Figure 5.

FIG 5 - Grades for the example.

The optimiser was run for different prices values. For a low price the natural answer is that no extraction would occur. For a high price the algorithm should be giving an answer that tries to get as much as possible from the orebody. The pits obtained for several different prices can be appreciated in Figure 6. We can see that as the price increases so does the pit. This answer is consistent with the expectations for a model for open pit design; in fact it is the basis for pushback design techniques in software packages such as Datamine. It is noted that the solution for price zero (total absence of income) is not exactly zero extraction but close to it. The algorithm should reach to the optimum solution provided there is enough time to iterate. Unfortunately this is one of the aspects that contribute to create a negative image of the use of genetic algorithms because they cannot guarantee an optimal solution, even allowing them to iterate for a long time period. On the other hand, starting from a good initial solution, a GA has the ability to evolve that solution towards an improved solution. It is believed that in a mining context this characteristic is0 beneficial; the miners can put their experience developing a design after which an automated tool based on genetic algorithms can improve that starting answer to have a much better design incorporating the knowledge of the mine planners and designers. For the simple example presented the solution times are low. It obviously depends on the number of iterations that the GA will run; in the current test it is still within some seconds because not MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

101

J SAAVEDRA-ROSAS

FIG 6 - Results obtained for different prices in the test problem.

too many iterations were used. Traditional approaches are certainly faster when compared to our approach but as mentioned before traditional approaches are more inflexible too. This simple experiment it is just an initial validation for the model and the results are satisfactory. They clearly show an extraction profile which is compatible with slope constraints where there are not abrupt changes between consecutive columns and exhibiting expected behaviour with changing conditions.

CONCLUSIONS The proposed model is certainly a novel approach to open pit planning. As mentioned before there are some advantages in the use of such kind of models, particularly when it comes to gain independence from the underlying block model. Another advantage of such a model comes in the form of the ability (by means of the use of genetic algorithms) to incorporate uncertainty into the decision process (Saavedra, 2009) and it is believed that this is a promising research avenue for the future. The flexibility provided by GAs can be exploited to easily extend the proposed optimisation model to other related problems. For example, Songolo (2010) proposes the use of the model for pushback design with capacity constraints. Songolo’s work extends the model presented here to consider capacity constrains for pushback design. The approach used is to consider a penalised version of a problem with a pushback capacity constraint. The approach is still in its infancy yet the penalisation MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

102

A NEW GENETIC ALGORITHM FOR OPEN PIT DESIGN – THE TWO-DIMENSIONAL CASE

factor has proved to be useful in enforcing the capacity constraint for the pushbacks but still more work has to be done in order to determine the correct size for this factor in order to produce pushbacks of the correct size. On the other hand, Saavedra (2009) has outlined the potential use of the model to produce a robust pit design and some results are expected to come out in the short-term. Generally speaking, GAs allow the use of general objective functions, eliminating in this way the usual requirements for linearity or convexity that the mathematical programming approaches require. The price to pay is the lack of an optimality certificate for the procedure. With this in mind it is easy to think in objective functions that simulate a schedule, or that simulate haul road performance, etc. This kind of techniques could open a new class of developments in the area and some really fructiferous research is expected for the future. For the model itself the work that has to be done is to move the current two-dimensional model to a 3D version. The purpose of the model presented on this paper is to show the practicality of the approach and to illustrate the potential benefits before moving to a fully fledged model for the real world applications. Provided the experiences performed in two dimensions it is anticipated that the design of the crossover and mutation operations will have to be careful because production of non feasible individuals is expected to increase with the increase of the dimension parameter. The numerical results are consistent with what should be expected and are encouraging. Several tests need to be performed in order to determine the proper population size, the crossover and mutation probabilities, the number of iterations required to reach an optimum (in average), etc. Also, it is important to develop insight of the dependence of the GA model on the size of the problem to solve (ie solution time versus number of columns). A final test is required for the two-dimensional model before further progress into the 3D arena can be performed: a thorough comparison with a solution of a model solved using Lerchs and Grossman. Preliminary tests have shown a decent performance of the genetic model but still more fine tuning of the genetic algorithm parameters are required before conducting more definitive tests.

REFERENCES Bodenhofer, U, 2000. Genetic algorithms, theory and applications [online]. Available from: [Accessed: 22 October 2010]. Denby, B and Schofield, D, 1994. Open pit design and scheduling by use of genetic algorithms, Transactions of the IMM, 103:A21-A26. Denby, B and Schofield, D, 1996. Genetic algorithms for open pit scheduling – Extension into 3-dimensions, in Proceedings MPES 1996 (eds: L A Ayres da Silva and A P Chaves) (Balkema: Rotterdam). Goodwin, G, Seron, M, Middleton, R, Zhang, M, Hennessy, B and Stone, P, 2006. Receding horizon control applied to optimal mine planning, Automatica, 42:1337-1342. Holland, J H, 1975. Adaptation in Natural and Artificial Systems (University of Michigan Press: Ann Arbor). Koza, J R, 1992. Genetic Programming: On the Programming of Computers by Means of Natural Selection (The MIT Press: Cambridge). Lerchs, H and Grosmman, I, 1965. Optimum design for open pit mines, CIM Bulletin, 58:47-54. Meyer, M, 1969. Applying linear programming to the design of ultimate pit limits, Management Science, 16(2):B121-B135. Michalewicz, Z, 1996. Genetic Algorithms + Data Structures = Evolution Programs (Springer). Saavedra-Rosas, J F, 2004. An Evolutionary Model for Underground Mining Planning, in Proceedings MassMin 2004 (eds: A Karzulovic and M Alfaro). Saavedra-Rosas, J F, 2009. A genetic optimizer for stochastic problems with applications to orebody uncertainty in mine planning, PhD thesis (unpublished), Laurentian University, Ontario. Schofield, D and Denby, B, 1993. Genetic Algorithms: A New Approach to Pit Optimisation, in Proceedings Symposium on Application of Computers & Operations Research in Mineral Industry, pp 126-133 (Canadian Institute of Mining and Metallurgy: Montréal). Songolo, M W, 2010. Pushback Design Using Genetic Algorithms, MSc thesis (unpublished), Curtin University of Technology, Perth. Van Rossum, G and Drake, F L, 2010. Python tutorial, release 2.7 [online]. Available from: [Accessed: 4 July 2010]. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

103

Enterprise Optimisation G Whittle1 ABSTRACT Enterprise Optimisation is a methodology for increasing the value of mining and mineral processing operations through better long-term planning decisions. It involves a combination of ten mechanisms which deal with decisions at different stages of the value chain. A decision made at any point in the chain potentially affects the decision for all other points in the chain. The key, therefore, is to optimise them simultaneously. The Enterprise Optimisation approach can be shown to produce increases of five per cent to 35 per cent or more in net present value (NPV), even when several optimisation methods have already been applied. This paper steps through an example case study, illustrating the effect of each mechanism and how they can work together to develop greatly improved results. This example results in a series of counter-intuitive outcomes, which challenges conventional management practices and typical organisational objectives based on organisational silos.

INTRODUCTION Enterprise Optimisation is a methodology for increasing the economic value of mining and mineral processing operations through better long-term planning decisions. It involves a combination of ten mechanisms, which deal with decisions at different stages of the value chain. A decision made at any point in the system potentially affects the optimal decision for all other points in the chain. The key, therefore, is to optimise them simultaneously. Enterprise Optimisation involves simultaneously optimising: x All steps in the value chain (see Figure 1). It is clear that a decision on one step in the value chain can affect all the others, eg the plant constraint will affect the optimum cut-off grade, a change in cut-off grade will affect the mining schedule, and a change in schedule affects which pit shells should be selected as mine phases. A change in the metal price affects everything, right back to the pit design.

FIG 1 - The steps in the value chain.

x All assets in the enterprise portfolio. A decision affecting one component of an asset portfolio (many mines, plants, products) can affect the optimal operation of the others. x All periods together. We are mining a depleting resource, so a decision for one period affects our options for the other periods. What is mined determines the surface for the next period, what will be available from stockpiles, etc. You cannot just optimise one period and then consider the next. Apart from the analytical and computational challenges this raises, organisation barriers exist in the form of departmental and divisional silos which compound the problem. Organisation-wide participation in an Enterprise Optimisation study is essential to ensure that the analysis is correct, and that the outcomes are accepted and understood by the individuals who must implement them. 1. MAusIMM, Managing Director, Whittle Consulting Pty Ltd, Suite 13, 333 Canterbury Road, Canterbury Vic 3126. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

105

G WHITTLE

To simplify the planning process many decisions are fixed, when they should be dynamic (plant configuration, head grade, product specification, production rate, etc) and many decisions are made too early, before the options are analysed (size of the resource, product specification, capital, plant size/type, mine life, etc). These simplifications prevent the full value of a project being realised. Enterprise Optimisation overcomes the analytical challenge in dealing with these issues, but we must all cooperate to overcome the organisational challenges for simultaneous optimisation to be achieved effectively. Money has a time value. Net present value has therefore been used as the basic measure of economic value.

CASE MODEL This is a fictitious but realistic case, developed so that we can clearly demonstrate the mechanisms involved without client confidentiality issues. The settings have been made very simple to make the case easier to follow. Do not interpret this as a limitation on the methodology which can account for full details of real-life complications in a real case. x single copper/gold pit with four phases; x 60 Mt/a mining, maximum eight benches pa vertical advancement (ie 160 m); x plant is 20 Mt/a crush/grind/float; x 88 per cent copper recovery, 60 per cent gold recovery, fixed; x producing 28 per cent copper concentrate, with gold contained; x concentrate is transported 70 km by a 600 Kt/a pipeline to the port; x the concentrate is then shipped offshore to a smelter/refinery; x gold price US$900/oz; x copper price US$2.50/lb declining to US$1.50/lb in the first five years; and x ten per cent discount rate for NPV calculation.

CASE 0 – MANUAL CASE The starting point is a manual pit and phase design, scheduled with five bench fixed-lead between phases. All economic material is processed, and there is no stockpiling. It took three frustrating days to develop the manual pit and phase shapes, and the designer had a guide as he had already seen the optimised pit. Considering the waste stripping required to gain access and the contribution of other economic material accessed on the way, it was a question of judgement of whether to include pockets of high-grade material. The manual plan produced an NPV of US$1598 M. That is the sum of the discounted cash flows after US$592 M in capital expenditure. This is a good result, but it can be improved. Although Enterprise Optimisation is about simultaneous optimisation, in this example we will turn on one mechanism at a time so that the effect of each can be discussed.

STEP 1 – PIT Pit designs can be optimised using Lerchs-Grossmann (1965). Lerchs-Grossmann is effective at determining the economic three-dimensional shape of the pit considering block grades, pit slopes, costs, recoveries, and metal prices. In this step, pit shells 4, 8, 10 and 16 with roughly equal tonnes were arbitrarily selected as intermediate phases from the 16 nested pits produced by varying the revenue factor. The optimised pit does not look significantly different from the manual pit but it does vary in detail in certain parts. You cannot beat the computer. NPV is increased by 7.2 per cent. The Lerchs-Grossmann optimisation is based on the valuation of each block in the model. The cost modelling applied and the degree to which material can be characterised for its processing throughput, cost and recovery can have a significant impact on the resulting pits and phases.

STEP 2 – PHASES Phases can be selected using Whittle auto-pushback chooser and skin analysis (choice of ultimate shell). Selecting 4 phases from 16 nested shells generated by varying the revenue factor in Lerchs-Grossmann, involves testing 1820 combinations. With simple scheduling assumptions this can be computed quite quickly. This leads to a smaller ultimate pit with 18 per cent less rock but only nine per cent less ore. Shells selected for phases are 2, 3, 8, and 12 giving better early access to ore, but reduced overall mined resource. Only a small proportion of the NPV increase is due to reducing the size of the ultimate pit, MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

106

ENTERPRISE OPTIMISATION

but this change causes much discussion and controversy. In most pits, although the outer shells may be cash positive (which is why Lerchs-Grossmann included them), NPV may be negative due to the time delay between the waste mining near the surface and the eventual revenues derived from the deeper ore – the extra discounting of the later periods in the NPV calculation causing this effect. The bulk of the value, however, is produced by designing the best first phase with early access to high grade material with a low stripping ratio. NPV is increased by 6.4 per cent.

STEP 3 – SCHEDULING So far simple five bench fixed-lead scheduling has been applied. Applying the Whittle ‘Milawa’ algorithm produces a different bench lead for each phase for each period. The optimiser is able to delay waste and bring forward high grade material within the phases presented. Mine scheduling is mathematically challenging as the orebody is not linear. Conventional mathematics would indicate the use of mixed integer programming for this purpose. Milawa applies a search algorithm (not mixed integer) to solve the mine scheduling challenge. NPV is increased by 4.4 per cent.

STEP 4 – CUT-OFF GRADE Inspired by Ken Lane (1988), the Whittle cut-off grade optimiser raises early cut-off to increase early metal production, even if positive margin material is discarded and the mine life shortened. This is not ‘high-grading’ but ‘right-grading’. The magnitude of the effect may be a surprise to many but not to those familiar with this mechanism, which is sadly not universally practised. Extra mining capacity is required to raise the cut-off grade – if mining rate variations have already been flattened by applying a limit before this step, then little or no gain will be observed which would be an opportunity missed. Cut-off grade optimisation produces the best results in deposits with lots of grade variation, ie a wide grade-tonnage curve. It is common for geologists and mining engineers to be uncomfortable with the proposition of discarding economic material – this is an arguably irrational reaction. NPV is increased by 15.1 per cent in this case.

STEP 5 – STOCKPILES Stockpiles were limited to 60 Mt in this case. Rather than discard low value material, it may as well be stockpiled and processed later in the life of the operation. Reclaiming low grade material later recoups extra value and allows the early cut-off grade to be raised even further. Rehandling costs and possibly different recoveries for the additional weathering of the material must be taken into account. The mine life and overall resource recovery is almost returned which is comforting to many. NPV is increased by 4.6 per cent. Note: This is how far you can go with the widely used Gemcom Whittle software – the shades on individual steps 1 to 5 on the graph in Figure 2 are different as the mechanisms are sequential and isolated, but nevertheless sum to a total increase in value of 37.7 per cent. In late 2010, Whittle version 4.4 will integrate steps 3 to 5 (schedule, cut-off/blending, stockpile), which gives most of the benefit of step 6.

STEP 6 – SIMULTANEOUS OPTIMISATION We now switch to Whittle Consulting’s proprietary ‘Prober’ software described by Whittle (2009). This more recent development is capable of performing the previous steps 1 to 5 simultaneously from a set of nested shells produced by Lerchs-Grossmann. Same settings, same problem, simply a better optimisation by having the mechanisms work together. NPV is increased by a further 14.1 per cent (Note: approximately two per cent of this was due to reselecting the phases to shells 3, 5, 9, and 13). Components 1 to 6 cover what can be referred to as Mining Optimisation – everything the mining department is responsible for (see Figure 2). From step 6 onwards the graph covers all previous steps as these are reassessed simultaneously. Now we must consider the rest of the value chain.

STEP 7 – PROCESSING CALIBRATION All the steps so far presumed that the plant will be run at its nameplate 20 Mt/a with fixed recoveries of copper 88 per cent and gold 60 per cent. The fact is that the plant could be run at a range of throughput rates, with consequences on the recovery. Less grind time and residence can increase plant throughput but recoveries suffer significantly, and conversely slightly higher recoveries can be achieved if the plant is slowed down. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

107

G WHITTLE

Accumulated Value $3,000 $2,800

NPV $Millions

$2,600 $2,400 $226

$2,200

$74 $241

$2,000

$2,426 $70

$1,800

$102 $115

$1,600 $1,598

7.2%

6.4%

4.4%

15.1%

4.6%

51.8%

14.1%

tal To

ou s

ile

ne

kp 6S

im ul t a

toc

utOf f

5S

4C

3S

ch

ed ule

se s

it

ha

1P

2P

Ma n

ua

l

$1,400

FIG 2 - The steps in mining optimisation.

Please note that we are not expanding the plant with capital in this step, just changing the residence time and therefore altering the throughput and recovery. Looking at Figure 3 in isolation is unlikely to result in a metallurgist or process manager selecting the high throughput solution. If the plant is only to be run at one speed then 20 Mt is possibly the right choice. However, if the optimiser is in control it uses the whole range during the life of the operation – a dynamic mechanism. Given throughput flexibility, the optimiser often sacrifices recovery to increase throughput. It is a complex trade-off between extra immediate cash flows, extra mining and processing cost, and the value of the metal lost and the impact that it has on the longevity of the operation and future cash flows. At the end of the mine life it enjoys the benefit of higher recovery, with less throughput (see Figures 4 and 5). Allowing the optimiser to use this approach adds a further 4.4 per cent to the NPV. This figure would have been closer to eight per cent if not for the fact that we are now regularly hitting the concentrate pipeline limit of 600 Ktpa – a downstream bottleneck. This mechanism has some similarities with cut-off optimisation in that it is trading off overall resource recovery against more rapid delivery of metal to market early in the life of the operation to increase early cash flows.

R eco very

Recovery vs Throughput 90.0% 85.0% 80.0% 75.0% 70.0% 65.0% 60.0% 55.0% 50.0% 18

20

22 Au Recovery

24

26

Cu Recovery

Throughput (Mill tpa)

FIG 3 - Recovery versus throughput. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

108

ENTERPRISE OPTIMISATION



FIG 4 - Throughput optimised.

Recoveries 100% 90% 80% Cubase

70%

Curecov

60%

Aubase 50% Aurecov 40% 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Year

FIG 5 - Recovery optimised.

STEP 8 – PRODUCT SPECIFICATION The base case presumes that a 28 per cent copper concentrate is produced, with recovered gold contained. Figure 6 shows how a range of concentrates could be produced with significant impact on recovery, with an effect on the transportation of a more or less bulky concentrate product and different percentage of metal paid by the smelter to the mine for its concentrate. With a spreadsheet to calculate net smelter return (NSR) it is easy to show that: x when the copper price is $2.50/lb, the best solution is a 24 per cent concentrate as the benefit of extra recovery outweighs the extra transport cost of the bulky product; and Cu Recovery 95.0%

Recovery

90.0% 85.0%

24%Con 26%Con

80.0%

28%Con 30%Con

75.0%

32%Con

70.0% 18

20

22

24

26

Throughput (Mill tpa)

FIG 6 - The effect of concentrate percentage on recovery. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

109

G WHITTLE

x when the copper price is $1.50/lb (as per long-term in this case), the best solution is the 28 per cent concentrate as used in the base case. This approach is pursuing ‘margin’. It is more difficult, however, to calculate the benefit of throughput. The optimiser has used this flexibility, at great expense in terms of metal recovery at certain times, to get more metal to market through the restrictive pipeline, which at this point is the major constraint in the system. A metallurgist would be unlikely to recommend producing a 32 per cent concentrate involving an eight per cent lower recovery by looking at Figure 6 in isolation, yet the optimiser has shown this to be the best decision for the business under many circumstances. In Figures 7, 8 and 9 it can be seen that when the pipeline is active as a bottleneck, the optimiser moves to a higher grade concentrate to get more metal to market, even at a significant cost in terms of lower recovery. Allowing the optimiser to use this approach adds a further 5.5 per cent to the NPV.

Case 8 - Product: Copper Con Tonnes 700,000

600,000

Concentrate Tonnes

500,000

400,000

Truck Pipe

300,000

Limit

200,000

100,000

1

2

3

4

5

6

7

8

9

10 11 Year

12

13

14

15

16

17

18

19

20



FIG 7 - The concentrate production reaches the pipeline limit.

ConCuGrade

33.0% 32.0% 31.0% 30.0% 29.0% 28.0% 27.0% 26.0% 25.0% 24.0%

Default Optimized

1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17 18 19 20 Year



FIG 8 - Concentrate copper grade raised to increase metal production.

CuRecovery

90.0% 88.0% 86.0% 84.0% 82.0% 80.0% 78.0% 76.0% 74.0% 72.0% 70.0%

Default Optimized

1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17 18 19 20 Year



FIG 9 - Copper recovery losses – a consequence. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

110

ENTERPRISE OPTIMISATION

STEP 9 – LOGISTICS Additional trucking of concentrate is possible at a cost of $30/tonne. This cost is high because it is a short-term strategy – we will not make a long-term contract, nor invest in handling facilities to make this process less costly, and because the road needs significant maintenance not having been designed for this level of heavy traffic. The additional capacity allows the previous mechanism (processing and product specification, supported by all the mining mechanisms) to pursue margin again, not throughput. More costly logistics but better value overall. Allowing the optimiser to use this flexibility adds a further 7.1 per cent to the NPV. See Figure 10.

Case 9 - Logistics: Copper Con Tonnes 1,000,000 900,000

Concentrate Tonnes

800,000 700,000 600,000 Truck

500,000

Pipe Limit

400,000 300,000 200,000 100,000 1

2

3

4

5

6

7

8

9

10 11 12 13 14 15 16 17 18 19 20 Year

FIG 10 - Movement of the concentrate with additional trucking.

STEP 10 – CAPITAL Rather than taking such drastic measures to manage the bottlenecks in the system, why not size the components of the system correctly in the first place. As long as this analysis is done during the feasibility study and not after construction of the plant and infrastructure, then rescaling is feasible. In the manual base case, there was ample mining fleet capacity and pipeline capacity; in fact the finance manager may have taken steps to save some capital in these areas. However, now that we have simultaneously optimised the pits, phases, schedule, cut-off, stockpiles, then raised the plant throughput, we have both the mining fleet and the pipeline fully utilised. This raises the question of whether they should be expanded. (Increasing the plant size should also be considered, but that has been left out of the scope of this example). We can add extra mining capacity at $1.25 per tonne per year and extra pipeline capacity at $20 per tonne per year – these increases are only allowed in period 0 for simplicity of this example. Making a single capital decision can be done by design trial and error iterations – say five, increasing the capacity progressively and seeing if the NPV increases more than the capital cost of the expansion. With two capital decisions that affect each other, that would be 5 × 5 = 25 iterations of a complete life-of-mine plan with nine active mechanisms. If we did not increase the pipeline capacity then we would not need so much mining capacity, and if we did not increase the mining capacity then we would not need such a large pipeline. These decisions must be made simultaneously, yet it is typical to delegate these decisions to the mining manager and the logistics manager respectively, without any dialogue or cooperation between the two. This would lead to a suboptimal outcome. In practice, there are dozens of capital decisions that would require thousands of life-of-mine plan iterations. This is not possible, so it does not get done. The optimiser can determine how much capital is worth spending on each constraint – simultaneously, and rebalance the pit and phase selection, mine schedule, cut-off, stockpile, processing, product, logistics at the same time (steps 1 to 9). MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

111

G WHITTLE

Mining capacity goes up to 83 Mt/a, the pipeline to 1 Mt/a. Trucking no longer required. The shells used for phases are now 4, 7, 9, and 14. An extra $37 M of capital spend but NPV after the extra capital increases by a further 4.9 per cent.

SUMMARY Figure 11 summarises the impact of all ten components described above. The recent ones have been focused on dealing with downstream bottlenecks and although each one introduced a new mechanism, it was critical to assess the impact it had on the previous mechanisms to calibrate it properly. The benefits of downstream modifications must be assessed with consideration of what the mine can deliver – due to its internal limits the head grade is likely to reduce as mining tonnage increases. Simultaneous optimisation is essential to get the balance of these effects right. Accumulated Value $3,000 $2,800

$78 $113

NPV $Millions

$2,600

$88 $70

$2,400 $226

$2,200

$74 $2,775 $241

$2,000 $70

$1,800

$102 $115

$1,600 $1,598

7.2%

6.4%

4.4%

15.1%

4.6%

14.1%

4.4%

5.5%

7.1%

4.9%

73.7%

tal

l pit a

To

cs

Ca

ist i og

9L

10

uc t

g

ro d

es ro c 7P

8P

sin

ou s

ile

ne

kp 6S

im ul t a

toc

utOf f

5S

4C

3S

ch

ed ule

se s

it 1P

ha 2P

Ma n

ua

l

$1,400

FIG 11 - The ten components of Enterprise Optimisation combine to increase net present value by 73.7 per cent in this case.

OVERALL RESULT NPV has been increased from US$1598 M in the base case, to US$2775 M in case 10 with all mechanisms working together – an increase in value of 73.7 per cent. The optimised plan is undoubtedly more complicated than the manual plan as can be seen from Figures 12 and 13, but the higher NPV justifies this. Mining has increased, significant amounts are stockpiled, plant throughput is up for many of the early years, and most importantly the early feed grades to the plant are approximately doubled. The combination of these significantly increases metal production in early years when the time value of money is highest and the copper metal price is high. Please consider what has been done in the various cases: x ultimate pit has been reduced, along with reserves; x mining, processing and logistics costs have been increased significantly; x plant recovery has been significantly decreased, due to higher throughput and higher concentrate grade in some cases; x mining equipment has been made idle, two thirds of the fleet for several years; x capital expenditure has increased; and x life of the operation has reduced by several years. All these outcomes are counter-intuitive and in a typical organisation would have been resisted by the individual managers concerned, and by the Chief Executive Officer. However, together they have transformed the economic performance of the business. If any of these changes were made in isolation it would be disastrous, so a coordinated approach to the analysis and implementation is essential. The value comes from looking at how the different steps in the value chain can affect each other and can work together, rather than looking at them in isolation. Much of this new value is hidden ‘between’ the steps managed by the organisational silos, and so a different mindset is required to identify it. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

112

ENTERPRISE OPTIMISATION

Case 0. Manual - Schedule 1.40

140

1.30 120

1.20 1.10

100

1.00

0.70 60

0.60 0.50

40

Au g/t

0.80

Cu %

Tonnes Millions

0.90 80

0.40 0.30

20

0.20 0.10

0

0.00 1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

Year

Stockpile Bal

Waste

Stockpiled

Processed from Mine

Cu grade Processed

Au Grade Processed

Processed from Stockpile

FIG 12 - Schedule for the manual design.

Case 10. Capital - Schedule 1.40

140

1.30 1.20

120

1.10 100

1.00

0.70 60

0.60 0.50

40

Au g/t

0.80

Cu %

Tonnes Millions

0.90 80

0.40 0.30

20

0.20 0.10

0

0.00 1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

Year

Stockpile Bal

Waste

Stockpiled

Processed from Mine

Cu grade Processed

Au Grade Processed

Processed from Stockpile

FIG 13 - Schedule for full Enterprise Optimisation plan.

Figure 14 sums up the overall change in the cash flow profile due to the ten mechanisms applied. Basically US$1.5 billion of cash flow has been moved from the last few years of the life of the operation to the first three years. This provides business options and security for the project, and greatly increases the likelihood that it will achieve financial support and approval. The possible negatives of reducing reserves and reducing employment longevity can be compensated for by reinvesting some of the early proceeds into exploration or other worthwhile projects. See Tables 1 and 2 for details of the manual case and the fully optimised life of mine schedules. Please note that some of the benefit in this case is derived from increasing copper production in the early years when the copper price is high. Repeating the entire case study with flat copper price at $1.50/lb still generates and overall increase in NPV of 53.9 per cent (rather than 73.7 per cent). The amount of the improvement possible depends entirely on the configuration of the base case, and the orebody itself. Applying the mechanisms in a different order would also affect the quantification of the effect of each step. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

113

G WHITTLE

Net Cash $1,200 $1,000 $800

Manual

$400

Optimized

Millions

$600

$200 $0 1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

-$200 -$400 -$600 -$800

Year



FIG 14 - Net cash flow before and after Enterprise Optimisation.

ACTUAL PROJECTS The example above involves just one pit. The Enterprise Optimisation method has been applied to situations involving a significant number of pits, underground stopes, elements, processing plants, products, etc and all their attendant interactions. Portfolio Enterprise Optimisation studies performed to date include: x 86 mines, smelter and leach; x 38 pit and underground mines, five concentrators, smelter, leach and refinery, blending requirements, 15 product options; x 18 deposits with 130 pits, pressure and heap leach and refinery; x 40 underground mines, three pits, two product streams with shared infrastructure; x 80 deposits country-wide, several wash plants, logistics options to central smelter; and x three pits with direct feed and flotation option, negotiable oxygen ratios with throughput and recovery implications. Actual Enterprise Optimisation studies completed this year have identified opportunities for NPV increase of between 25 per cent and 85 per cent. In each of these cases conventional optimisation approaches (including the use of Whittle software) had already been applied.

CONCLUSION Management performance measurement involving key performance indicators like minimising cost, maximising recoveries, maximising reserves and life of mine, may be leading us away from significant value enhancing opportunities. Any business plan which has the same setting for any decision (mining rate, stripping ratio, cut-off grade, plant head grade, throughput, production, product specification) year after year cannot be optimal. The orebody is not the same each year and even if it was, the time value of money would lead us to different trade-off decisions in each period. It is therefore relatively easy to spot where opportunity for improvement by this technique may exist. Enterprise Optimisation unleashes value by dynamically harmonising the flexibility in all parts of the operation. It relies on a combination of philosophy, methodology and sophisticated software to achieve the result. We have shown that there are very considerable gains to be made by optimising a number of mechanisms that are not usually considered by many organisations, and by optimising them simultaneously. Complexity and variation are opportunity – embracing this with a structured approach can realise significant economic benefits. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

114

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

TABLE 1 Case 0: Manual base case. Case 0: Manual Base Case Total mined Ktonne

Waste Ktonne

To Stkpl Ktonne

To plant Ktonne

From Stkpl Ktonne

Ore proc Ktonne

Cu %

Au g/t

Conc Ktonne

Con Cu %

Con Au g/t

Revenue $M

Min cost $M

Reclaim $M

Proc cost $M

Sell cost $M

Net cash $M

DCF 10% $M

1

27 609

7609

0

20 000

0

20 000

0.36

0.52

226

28.0

27.5

531

$40

$-

$136

$83

$273

$248

2

20 146

146

0

20 000

0

20 000

0.57

0.70

358

28.0

23.3

742

$29

$-

$136

$131

$446

$369

3

28 468

8 468

0

20 000

0

20 000

0.70

0.81

440

28.0

22.1

827

$41

$-

$136

$160

$490

$368

4

30 723

10 723

0

20 000

0

20 000

0.65

0.64

409

28.0

18.9

665

$45

$-

$136

$148

$336

$230

5

31 677

11 677

0

20 000

0

20 000

0.67

0.52

421

28.0

14.8

570

$46

$-

$136

$152

$236

$146

6

55 488

35 488

0

20 000

0

20 000

0.68

0.47

427

28.0

13.2

556

$80

$-

$136

$153

$187

$105

7

53 747

33 747

0

20 000

0

20 000

0.86

0.48

541

28.0

10.7

668

$78

$-

$136

$194

$259

$133

8

60 000

43 899

0

16 101

0

16 101

0.86

0.50

435

28.0

11.0

542

$87

$-

$115

$157

$184

$86

9

60 000

45 080

0

14 920

0

14 920

0.78

0.45

366

28.0

10.9

454

$87

$-

$108

$131

$127

$54

10

60 000

41 869

0

18 131

0

18 131

0.75

0.42

427

28.0

10.8

527

$87

$-

$126

$153

$161

$62

11

52 297

32 297

0

20 000

0

20 000

0.73

0.45

459

28.0

11.7

583

$76

$-

$136

$166

$205

$72

12

48 999

28 999

0

20 000

0

20 000

0.70

0.39

440

28.0

10.6

544

$71

$-

$136

$159

$178

$57

13

36 904

16 904

0

20 000

0

20 000

0.64

0.31

402

28.0

9.4

483

$54

$-

$136

$146

$148

$43

14

27 011

7 011

0

20 000

0

20 000

0.74

0.41

465

28.0

10.5

572

$39

$-

$136

$167

$229

$60

15

23 308

3 308

0

20 000

0

20 000

0.75

0.50

471

28.0

12.6

611

$34

$-

$136

$171

$270

$65

16

21 585

1 585

0

20 000

0

20 000

0.72

0.53

453

28.0

14.2

604

$31

$-

$136

$163

$274

$60

17

10 654

327

0

10 327

0

10 327

0.77

0.58

250

28.0

14.3

335

$15

$-

$70

$90

$159

$32

Year

0

0

0

0

0

0

0.00

0.00

0

-

$-

$-

$-

$-

$-

$-

0

0

0

0

0

0

0.00

0.00

0

-

$-

$-

$-

$-

$-

$-

20

0

0

0

0

0

0

0.00

0.00

0

-

$-

$-

$-

$-

$-

$-

648 616

329 138

0

319 479

0

319 479

0.70

0.51

6 991

9814

$940

$-

$2187

$2525

$4162

$2190

Less capital

$592

NPV

$1598

Total

28.0

14.0

115

ENTERPRISE OPTIMISATION

18 19

G WHITTLE

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

TABLE 2 Case 10: Fully optimised. Case 10: Fully Optimised Year

Total mined Ktonne

Waste Ktonne

To Stkpl Ktonne

To plant Ktonne

From Stkpl Ktonne

Ore proc Ktonne

Cu %

Au g/t

Conc Ktonne

Con Cu %

Con Au g/t

Revenue $M

Min cost $M

Reclaim $M

Proc cost $M

Sell cost $M

Net cash $M

DCF 10% $M

1

83 259

20 628

36 630

26 001

0

26 001

0.83

0.93

799

24.0

16.8

1447

$121

$-

$177

$277

$873

$793

2

77 809

29 624

22 183

26 001

0

26 001

1.11

0.77

968

26.0

11.5

1570

$113

$-

$177

$341

$939

$776

3

68 257

41 853

820

25 584

0

25 584

1.15

0.73

999

26.0

10.4

1446

$99

$-

$174

$352

$821

$617

4

83 259

57 142

16 863

9254

16 496

25 750

0.73

0.61

638

26.0

13.7

892

$121

$5

$175

$225

$366

$250

5

26 974

3042

6955

16 976

6955

23 931

0.81

0.51

601

28.0

11.8

762

$39

$2

$163

$217

$341

$212

6

23 470

244

2167

21 060

2167

23 227

0.95

0.57

689

28.0

11.1

859

$34

$1

$158

$248

$418

$236

7

21 120

146

740

20 234

1740

21 974

0.81

0.70

559

28.0

16.3

780

$31

$1

$149

$202

$398

$204

8

83 259

81 689

1570

0

22 425

22 425

0.52

0.43

362

28.0

15.6

500

$121

$7

$152

$131

$89

$41

9

73 363

36 163

16 191

21 009

1175

22 184

0.66

0.34

460

28.0

9.6

553

$106

$0

$151

$165

$130

$55

10

28 154

608

6839

20 707

0

20 707

0.77

0.55

501

28.0

13.5

659

$41

$-

$141

$181

$297

$114

11

4865

1189

302

3375

16 384

19 759

0.50

0.37

312

28.0

14.2

417

$7

$5

$134

$113

$158

$55

12

0

0

0

0

18 000

18 000

0.40

0.28

226

28.0

13.7

299

$-

$5

$122

$82

$90

$29

13

0

0

0

0

18 000

18 000

0.36

0.25

203

28.0

13.5

267

$-

$5

$122

$73

$66

$19

14

0

0

0

0

7 918

7918

0.28

0.16

70

28.0

11.0

87

$-

$2

$54

$25

$6

$1

15

0

0

0

0

0

0

0.00

0.00

0

-

$-

$-

$-

$-

$-

$-

16

0

0

0

0

0

0

0.00

0.00

0

-

$-

$-

$-

$-

$-

$-

17

0

0

0

0

0

0

0.00

0.00

0

-

$-

$-

$-

$-

$-

$-

18

0

0

0

0

0

0

0.00

0.00

0

-

$-

$-

$-

$-

$-

$-

19

0

0

0

0

0

0

0.00

0.00

0

-

$-

$-

$-

$-

$-

$-

20

0

0

0

0

0

0

0.00

0.00

0

-

$-

$-

$-

$-

$-

$-

573 789

272 328

111 261

190 201

111 261

301 462

0.75

0.55

7 387

10 539

$832

$33

$2050

$2632

$4992

$3404

Less capital

$629

NPV

$2775

Total

26.9

12.9

116

ENTERPRISE OPTIMISATION

REFERENCES Lane, K F, 1988. The Economic Definition of Ore (Mining Journal Books: London). Lerchs, H and Grossmann, L, 1965. Optimum design of open-pit mines, Trans CIM, LXVII:17-24. Whittle, J, 2009. The Global Optimizer works – What next?, in Proceedings Orebody Modelling and Strategic Mine Planning Conference, pp 3-6 (The Australasian Institute of Mining an Metallurgy: Melbourne).

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

117

Optimisation Improvements in Whittle using Stope Optimisation Software S Keane1 ABSTRACT Mining One (M1) is often challenged with the task of evaluating deposits considering open cut and/or underground (UG) mining methods. In many near surface deposits a combination of both methods results in the optimum economic means of exploitation. As common as it is to have open pit mines eventually develop into underground mines, theory associated with determining the optimum underground/open pit interface location continues to be subject to refinement. With recent developments in underground software packages, the generation of rudimentary underground mining stopes from basic economic inputs is not only fast, it provides a practical underground mining solution. The determination of the open pit/underground interface is not new to Gemcom’s Whittle 4 (Whittle). However, its solution has only ever claimed to be indicative, mainly due the rudimentary nature of how Whittle represents any potential underground mining. The efficient generation of the aforementioned underground mining solution has provided mining engineers with a unique opportunity. Underground mining can now be represented within the Whittle environment to allow the open pit/underground mining scenario to be modelled with increased accuracy. Providing Whittle with the ability to now consider mineable underground stopes combined with the consideration of time value of money (discounting), has delivered a solution with higher credibility than ever before. This paper will elaborate on the potential value gains associated with improved accuracies in the modelling of any potential underground mining in the Whittle environment. Furthermore, it will explain the importance of incorporating time value of money in the optimisation process to provide a consistent solution through to pit selection and final analysis.

BACKGROUND ON WHITTLE AND MINEABLE SHAPE OPTIMISER Whittle Gemcom Whittle (‘Whittle’) is a pit optimisation program. It provides several indicative open pits, based upon an input block model, and costs, slopes, profits, etc input by the user. The basic theory behind how Whittle works, is it calculates the value for all the individual blocks in the model, based upon mining cost, processing cost, ore value, etc. Whittle then attempts to make shells that provide the theoretically most economic pits.

Datamine Studio 3 and Mineable Shape Optimiser Datamine Studio 3 is, amongst other things, a mine design and block modelling software package made by Datamine. Mineable Shape Optimiser (MSO) is an optional add-on to Datamine Studio 3, which calculates theoretical economic stope shapes in a block model based upon input parameters by the user (minimum and maximum stope widths, cut-off grade, etc). MSO Mineable Shape Optimiser is a true underground optimisation tool that maximises the alue of the resource, gi en stope geometry and design rules MSO computes the optimal si e, shape and location of stopes for an underground mine using an input block model containing grades or alues The Shape Optimi er searches for the optimal mineable 1. Mining Engineer, Mining One, Level 9, 50 Market Street, Melbourne Vic 3000. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

121

S KEANE

shapes and takes into account the orebody geometry Constraints can be applied on the dip and strike of the final stope shape Multiple stopes can be generated across the strike of the orebody ith minimum pillar dimensions enforced (Datamine, 2010).

Optimisation without Mineable Shape Optimiser stopes (traditional method) if both abo e-ground and underground methods are to be used, this can affect the design of the open pit, and hittle can take account of this hen considering this issue, it is important that, if a particular block can be mined by either abo e-ground methods or underground methods, then the correct alue to gi e it during open pit optimi ation is the difference bet een its alue hen mined abo e-ground and its alue hen mined underground Generally, when Whittle open pit/underground (OP/UG) optimisations are run, Whittle will optimise every block of specified rock types (sometimes oxide is excluded, or low-JORC-confidence material, as defined by the user). Theoretically, some of the blocks that Whittle values as more economic if mined via UG methods actually can’t form an economic stope (eg very thin high grade veins create high value areas for UG mining in Whittle, but when practical mining widths are introduced the mining dilution reduces the grade significantly, resulting in less attractive economics).

Optimisation with Mineable Shape Optimiser stopes Due to the nature of the OP/UG interface programming in Whittle, there is an inherent bias towards over-optimistic evaluations of the UG mining potential, since Whittle considers all blocks for UG mining no matter how impractical mining those block may be. There is potential for increasing the accuracy of the optimisation by restricting Whittle to only consider potentially feasible stopes for UG extraction. Before exporting the block model from Datamine to Whittle, an MSO run is conducted on it. MSO creates shapes around any material that may be economic and practical to mine. These shapes are saved as a Datamine Studio wireframe. Wireframes are then filled with empty blocks. These blocks are assigned a new field within the block model (eg ‘MSO 1’). This newly created block model is combined with the original block model (via an ADDMOD process). Then an E TRA process is done on the model, with the intent of slightly changing the type field names to be able to identify in Whittle blocks that are contained in the MSO stope areas. Only the areas that MSO identified are considered for UG extraction by Whittle. Excluding uneconomic areas from UG mining in the Whittle optimisation can only serve to increase the accuracy and practicality of the resultant potential open pits.

EXAMPLE CALCULATIONS The usual way … Preparing the model Each user has a slightly different way of preparing models. To prepare a model for Whittle, several manual processes should be run. These can be put into a macro. For this model, there are two rock types – fresh, and oxide. Fresh and oxide material is denoted by the fields starting with ‘F ’ and ‘O ’ respectively.

Whittle optimisation Please see Appendix 2 for the full range of values used in the Whittle optimisations (see Figures 1 and 2). This optimisation resulted in the most economic pit being 11 493 000 t in size, with 697 000 t of ore. Note that Figure 2 shows just one main area that the pit targets, at the northern end of the resource.

The Mineable Shape Optimiser way … Running Mineable Shape Optimiser Depending upon ground conditions, available equipment, mining rate, etc MSO can be used to run a range of sensitivities if required. For this example, only one case will be done. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

122

OPTIMISATION IMPROVEMENTS IN WHITTLE USING STOPE OPTIMISATION SOFTWARE



FIG 1 - Image of the pit shells/processing tab in Whittle showing all rock types for processing.

FIG 2 - Initial Whittle optimisation best pit.

The full range of MSO parameters used can be found in Appendix 1 (see Figures 3 and 4). It can be seen from Figures 3 and 4 that MSO did not identify many potentially viable stopes in the southern area of the deposit.

Preparing the model The difference with preparing the model after an MSO run is the requirement to ‘flag’ potential stopes in the block model. The resultant wireframes from the MSO run need to be incorporated into the block model that Whittle will optimise. This is done by the following processes in addition to those briefly explained in the non-MSO section above. TRIFIL is done on the MSO wireframe, to fill it with ‘empty blocks’ (ie blocks with no density, grade, type, etc). A new field is added to all blocks called ‘MSO ’ and the value is set to 1 (Figure 5). MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

123

S KEANE



FIG 3 - An example of the mineable shape optimiser inputs.



FIG 4 - Potential stopes derived from mineable shape optimiser (orange). Note that the potential stopes target the high grade areas (block model is filtered on AU >0.1).

An ADDMOD process is used to combine the blocks in the potential stopes with the original block model containing grade, density, etc. When running the extra/output macro the blocks in potential stope areas need to be assigned a different rock type – this is done by changing the type field to show a ‘U’ (ie if type was ‘O 3 ’ and is now contained in a potential stope, type field will now have the value ‘O 3U’). This can be seen in Figure 6.

Whittle optimisation Please see Appendix 2 for the full range of values used in the Whittle optimisations. For these Whittle optimisations, only the rock types ‘ U’ (blocks inside MSO potential stopes) are considered for mining via both open pit and underground methods (see Figures 7 and 8). MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

124

OPTIMISATION IMPROVEMENTS IN WHITTLE USING STOPE OPTIMISATION SOFTWARE



FIG 5 - Mineable shape optimiser shapes filled with blocks.



FIG 6 - Part of the macro used to show the mineable shape optimiser areas in the type field.

This optimisation resulted in the most economic pit being 15 737 000 t in size, with 922 000 t of ore (see Figure 8).

Traditional method compared to the Mineable Shape Optimiser method Comparison It can be seen from the results of the Whittle optimisations that there is a significant difference in pit size between the standard OP/UG interface calculations and utilising MSO. The most optimum pit for the traditional method pit has a total size of 11 493 000 t, with 697 000 t of ore, which is 37 per cent larger in total tonnage than the MSO enabled Whittle optimisation which found the most economic pit was 15 737 000 t in size, with 922 000 t of ore (see Figure 9). It can be seen from Figure 9 that the southern pit area (on the left side of the image) is not present if the usual methods of UG interface optimisations are used in Whittle. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

125

S KEANE



FIG 7 - Only rock types ending in ‘***U’ are considered for potential underground mining.

FIG 8 - Most economic pit optimisation with the mineable shape optimiser model.

It can be concluded that the difference in shells is partly attributable to the fact that the southern area does not form many theoretically economic stopes. Therefore the potential for underground mining in this area is smaller. When the underground interface in Whittle is unrestricted, the southern area presents a larger advantage to the company to be mined via underground methods. However, with the lack of economic stopes in this area, it can be seen that there is minimal practical UG potential in the southern area. Therefore it is more economic in this scenario to mine the southern area via open cut methods.

CONCLUSIONS When faced with evaluating a deposit that may be mineable by a combination of open pit and underground methods, it can be seen that there is a potential for increased accuracy by utilising MSO to identify potential practical and economic stopes before pit optimisations with Whittle. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

126

OPTIMISATION IMPROVEMENTS IN WHITTLE USING STOPE OPTIMISATION SOFTWARE

FIG 9 - Comparison between the normal method (purple) and the mineable shape optimiser method (light brown). Mineable shape optimiser created potential stopes that are green.

A more accurate optimisation has a large potential to significantly increase the value of mining a deposit. In the example presented in this paper, the difference in methods resulted in significantly different open pit sizes – approximately 37 per cent larger pit (by tonnage) with 32 per cent more ore.

REFERENCES Datamine, 2010. MSO – Shaping underground mining online . Available from com.au/mso/ .

http //products.datamine.

APPENDIX 1 Values used in the Mineable Shape Optimiser calculations Cut-off grade Minimum stope width Maximum width Minimum waste pillar width Near wall dilution Far wall dilution Minimum dip angle Maximum dip angle Maximum strike angle Maximum strike angle change

2.5 g/t 5m 100 m 10 m 2m 2m 60 120 45 20

APPENDIX 2 Values used in the Whittle optimisations Overall slope angles in oxide Overall slope angles in fresh Mining cost Mining recovery Mining dilution Processing cost

30 40 3.40/t 95 per cent 10 per cent 25/t

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

127

S KEANE

Processing recovery UG processing cost UG processing recovery Selling price

95 per cent 60/t (note that his includes the UG mining cost) 80 per cent (note that this includes the UG mining recovery) At the time of writing this report (early July 2010) the spot gold price is approximately US 1250/oz and the exchange rate is approximately A 1 US 0.85. Therefore the gold price used is 1470.59/oz. Assuming total royalties of four per cent, the effective selling price is calculated to be 1411.76/oz, or 45.39/g. Discount rate per period ten per cent per annum Mining rate 5 Mt/a Assume all ore mined can be processed straight away (no mill limit).

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

128

A Review of the Theory and Application of Multi-Criteria Decision Analysis Techniques in Mine Planning C Musingwini1 ABSTRACT Mine planning includes equipment selection as a subprocess because aging equipment must be replaced to maintain or improve productivity levels. The criteria used for decision-making in mine planning and equipment selection are often inextricably linked and inherently contradictory. For example, financial wisdom demands that development, which is an expense and locks up capital, be deferred as far into the future as possible yet on the contrary technical knowledge suggests that developing well ahead of stoping is practically desirable because it generates additional geological information required to improve planning of the unmined portion of the orebody and create better operating flexibility. Consequently, mine planning professionals have to delicately balance all decision criteria when executing mine planning in order to achieve optimal mineral extraction or when selecting equipment, so that capital is optimally deployed. These two processes can therefore be characterised as belonging to the multi-criteria decision analysis (MCDA) type of problems. MCDA problems are alternatively referred to as multiple attribute decision-making (MADM) problems and belong to a broader group of problems called multi-criteria decision-making (MCDM). There is now a gradual recognition and application of MCDA techniques in the minerals industry. This paper explores the theory, current and potential application of these techniques in decision-making in the minerals industry.

INTRODUCTION The mine planning process is not a one-off event, but an integral part of the routine management of a well-run mine because mine plans need to be continually re-optimised to account for firstly, new information about the orebody and secondly, changes in the technological, economic, environmental, legal and social frameworks. Equipment selection is also part of the planning process and is occasionally done because aging equipment has to be replaced in order to maintain or improve productivity levels. The criteria used for decision-making in these processes are often inextricably linked and inherently contradictory. For example, in mine planning, financial wisdom demands that development, which is an expense and locks up capital, be deferred as far into the future as possible yet on the contrary, technical knowledge suggests that developing well ahead of stoping is practically desirable because it generates additional geological and geotechnical information required to improve planning of the remainder of the unmined orebody, thus creating better operating flexibility. Consequently, mine planning professionals are faced with the challenge of delicately balancing the decision criteria when executing mine planning, in order to achieve optimal mineral extraction or when selecting equipment, so that capital is optimally deployed. Such contradictions were noted by McCarthy (2002) in the scheduling of open pit mines. In addition to the above challenge, there is also the challenge that the decision-making criteria are measured in different units of measure but must be integrated through a trade-off, to arrive at the overall decision. The difficulty of having to consider different units of measure makes MCDA problems intrinsically hard to solve. It becomes even more difficult to configure how to achieve an optimal trade-off among the competing criteria, unless the importance attached to each criterion is known. The trade-offs can also be complicated further if the decision-making criteria have nonlinear relationships between them. For example it is difficult to configure an optimal trade-off between two criteria if say one is varying logarithmically while other one is varying following a quadratic function with respect to time or cost. 1. Senior Lecturer, School of Mining Engineering, University of Witwatersrand, Private Bag 3, Johannesburg WITS 2050, South Africa. Email: [email protected] MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

129

C MUSINGWINI

Another challenge is that the human brain can easily configure an optimal decision such as deriving maximum benefit or minimum loss when faced with a two-dimensional (2D) problem expressed as a quadratic function in an x-y Cartesian plane, or when the decision problem is three-dimensional (3D) expressed as a surface in 3D x-y-z space. Ballington et al (2005) alternatively referred to such a 3D surface as a ‘Hill-of-Value’. When optimisation decisions involve decision criteria that exceed three dimensions, humans have to rely on abstract thinking or attempt to simplify the problem back to 2D or 3D for easier configuration. However, as Saaty and Ozdemir (2003), Yavuz (2007), Yavuz and Pillay (2007a), Yavuz and Pillay (2007b) and Saaty (2008) noted, there are general limitations on human performance on abstract thinking. Noting the human limitation on abstract thinking, this paper therefore explores the theory and examples of application of MCDA techniques in decision-making in the minerals industry, with more focus on mine planning and equipment selection as these are integral parts of the routine management of a well-run mine. The review is not intended to be exhaustive, but seeks to layout the structure of the MCDA models and typical application areas, but also noting the potential for future application of these techniques in decision-making in the minerals industry.

GENERAL OVERVIEW OF THE MULTI-CRITERIA DECISION ANALYSIS (MCDA) PROCESS Since several criteria have to be evaluated simultaneously in mine planning and equipment selection, these two processes can therefore be characterised as multi-criteria decision analysis (MCDA), alternatively called multiple attribute decision-making (MADM). Some authors have argued that MADM or MCDA belongs to a broader group of techniques called multi-criteria decision-making (MCDM) which has multi-objective decision-making (MODM) techniques as another subset (Zimmermann, 1991; Triantaphyllou et al, 1998; Mendoza and Martins, 2006). The argument is that a distinction can be made between MCDA or MADM and MODM techniques. MADM or MCDA techniques are designed to solve problems in which discrete alternatives can be selected from a predetermined set of alternatives. On the other hand MODM techniques are designed to solve problems in which the decision space is continuous because a theoretically infinite number of alternatives exists. The basic structure of a generic MCDA problem (Table 1) is premised on requiring a decision maker or a group of decision makers, to select an alternative, Ai, from a set of alternatives, A = {A1, A2, …, Am}, such that Ai gives the best trade-off among decision criteria defined by a set, C = {C1, C2, …, Cn}. In total there are m alternatives and n criteria. The efficiency of alternative i against criterion j, is expressed as the outcome Oij. TABLE 1 The structure of a generic multi-criteria decision analysis problem. Criteria C1 Alternatives

C2

…

Cj

…

Cn

A1

O11

…

…

…

…

…

A2

…

…

…

…

…

…

…

…

…

…

…

…

…

Ai

…

…

…

Oij

…

…

…

…

…

…

…

…

…

Am

…

…

…

…

…

Omn

The generic MCDA process begins with defining objectives, mapping them into decision criteria, assigning weights to the criteria to indicate the importance of each criterion to the overall objective, identifying all possible alternatives, measuring the efficiency of each alternative against each criterion, synthesising the performance of each alternative for all the criteria and lastly selecting the best alternative. These stages which are briefly discussed in the next subsections can be condensed into basically four steps: 1. determining objectives and mapping them into criteria, 2. assigning weights to criteria, 3. aggregating the Oij values using the weights, and 4. executing the decision. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

130

A REVIEW OF THE THEORY AND APPLICATION OF MULTI-CRITERIA DECISION ANALYSIS TECHNIQUES IN MINE PLANNING

Mapping objectives into criteria There are three kinds of criteria namely, natural criteria, constructed criteria and proxy criteria (Keeney, 1992). Natural criteria can be measured directly and physically. For example the objective, ‘maximise production rate’ can be mapped into the criterion, ‘production rate’ and measured in ‘tpa’. Constructed criteria cannot be measured directly and physically but are measured through a derived index. For example the objective, ‘maximise operating flexibility’ can be mapped into the criterion ‘Flexibility index (FI)’, using a dimensionless scale. Proxy criteria are indirect measures of an objective when it is difficult to identify a natural or constructed criterion for that objective. For example, if the objective is to, ‘minimise the amount of acid mine drainage (AMD) released by a tailings dam’ then water draining from the dam is measured for the proxy criteria acidity and Iron (III) Hydroxide, which are the products of AMD formation. Acidity is measured on a pH scale while Iron (III) Hydroxide is measured in mg/l of drainage water and both are then used in conjunction to indicate the degree of AMD. Irrespective of the category that a criterion falls into, it must be measured. Criteria can be measured qualitatively (as non-numerical data or linguistic data), or quantitatively as numerical data. Linguistic scales are usually converted to equivalent numeral scales to facilitate analysis. For example a linguistic scale can be assigned numerical values on a scale of one to ten, such that ‘low’ is equivalent to one to three, ‘medium’ to four to six and ‘good’ to seven to ten. The numerical data on efficiency of alternatives when measured against decision criteria can be further divided into any of the following three broad classes (Chen, 2006): 1. Cardinal data: Data is cardinal if the outcome Oij is a real number; 2. Ordinal data: Data is ordinal when it is reported using linguistic scales; and 3. Interval data, Probabilistic data, Fuzzy data: This type of data accounts for uncertainty and is expressed as a probability function.

Assigning weights to criteria The importance of each decision criterion is measured by assigning a weight to the criterion. There are two broad categories of criteria weights namely, trade-off based weights and non-trade-off weights (Belton and Stewart, 2002). Trade-off based weights require the pair-wise comparison of criteria, thus creating some kind of ‘compensation’ across criteria. Non-trade-off based weights do not require trade-offs to be made across criteria. The raw data for trade-offs is usually obtained from questionnaire surveys or by techniques such as the Delphi technique or the Vicekry-Groves-Clarke method. Weights measure the relative importance of a criterion to the overall objective. The weight of a criterion Cj is denoted wj where wj ∈R (the set of all real numbers) and wj >0 for all criteria. Weights are normalised to sum up to 1 as shown in Equation 1 in order to assist decision makers to interpret the relative importance of each criterion: n /wj = 1 (1) j=1 The weight vector for each alternative is then defined as w = (w1, w2, …, wj …,wn) since the number of weights should be equal to the number of criteria.

Aggregate the weights and Oij values The outcomes of the efficiency of alternatives against criteria are normalised using a value function so that each Oij score corresponds to a dimensionless value v(Ai). Normalising is necessary to bring all values into a dimensionless space in order to avoid aggregating or adding ‘apples with oranges’. The weights of the criteria and the outcomes of the performance of alternatives are then aggregated as a linear additive value function, V(Ai), defined by Equation 2: V(Ai) = / wj.v (Ai)

(2)

Execute decision The aggregate values are then used to derive the specific decision required by the MCDA problem. This is when all the alternatives are ranked in terms of preference and the best or optimal alternative selected. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

131

C MUSINGWINI

COMMONLY USED MULTI-CRITERIA DECISION ANALYSIS TECHNIQUES Five MCDA techniques can be identified to have been used several times in other sectors and also found application in the minerals industry. The first technique is the French version, Elimination Et Choix Traduisant la Réalité (ELECTRE), which was translated into the English version Elimination and Choice Translating the Reality. The second technique is the Preference Ranking Organisation Method for Enrichment Evaluation (PROMETHEE). The third technique is the Multiple-Attribute Utility (MAUT). The fourth technique is the Analytic Hierarchy Process (AHP) and its generalisation the Analytic Network Process (ANP). The fifth and last technique is the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS). The techniques are classified according to the type of information available to the decision maker or group of decision makers and its salient features depending on whether it is ordinal or cardinal scale information (Geldermann and Rentz, 2005). MAUT and AHP techniques are most often applied when the information is cardinal while ELECTRE, PROMETHEE and TOPSIS techniques are applied to mostly ordinal scale information (Geldermann and Rentz, 2005; Triantaphyllou et al, 1998). The ELECTRE and PROMETHEE techniques are founded on the outranking procedure. Outranking is done to account for the fact that preferences are not constant in time, are not ambiguous and are not independent of the process of analysis (Geldermann and Rentz, 2005). Saaty (2008) concurred with the argument that human preferences are dynamic because, ‘people, then, not only have different feelings about the same situation, but their feelings change or can be changed by discussion, new evidence and interaction with other experienced people’. The outranking argument is that an alternative Ai outranks, dominates or is superior to alterative Aj if the decision maker strongly perceives Ai to be at least as good as Aj. This is expressed mathematically as Ai.Aj. A comparison of two alternatives is called a pair-wise comparison. A brief description of each technique is discussed in the next subsections.

Elimination Et Choix Traduisant la Réalité (ELECTRE) technique The ELECTRE group of techniques comprises of the versions ELECTRE I, ELECTRE II and ELECTRE III. The fundamental concept of ELECTRE techniques is that in the outranking relationship, Ai.Aj (read as Aj is dominated by Ai), implies that Aj is dominated by Ai under one or more criteria but is equally as good as Ai under the remaining decision criteria. The ELECTRE techniques initially make pair-wise comparisons of alternatives against each criterion, after converting the efficiency of each alternative against each criterion to a common physical or monetary value, defined as v(Ai) and v(Aj) for alternatives Ai and Aj, respectively. Threshold levels are then set for the difference v(Ai) - v(Aj). The decision maker then compares the difference against the threshold to decide if: x he/she is indifferent to the two alternatives, Ai and Aj; x he/she has a weak preference or strict preference of one alternative over the other; or x he/she is unable to express any perceived preference of one alternative over the other. This step of the process makes ELECTRE techniques difficult to explain to decision makers in industry because they work on threshold levels that usually have no realistic meaning (Geldermann and Rentz, 2005). Additionally, subtle differences among pair-wise comparisons usually complicate the ELECTRE decision-making process. Consequently, the set of outranking relationships may be complete or incomplete. The decision maker must then assign weights to the criteria to reflect their relative importance in the decision-making process. A series of consecutive assessments of the outranking relationships aggregated with the weights produces a concordance index, which is an indication of the degree to which Ai outranks Aj. The indices help to identify the core set of preferred alternatives since the system of outranking relationships is never necessarily complete as mentioned earlier.

Preference Ranking Organisation Method for Enrichment Evaluation (PROMETHEE) technique The PROMETHEE techniques comprise of PROMETHEE I and PROMETHEE II. The PROMETHEE techniques were developed to overcome the main problem associated with ELECTRE methods, that of nuances in the pair-wise comparisons (Geldermann and Rentz, 2005). The fundamental mathematical model underlying the PROMETHEE methods is that when comparing two alternatives Ai and Aj for each criterion, k, a preference function Pk can be defined as indicated by Equation 3: Pk(fk(Ai) – fk(Aj)) = Pk(d) ∈ [0,1]

(3)

where Pk(d), is the difference in the degree of preference for alternative Ai over Aj and varies from Pk(d) = 0, representing indifference in preference through a zone of weak preference, then a zone of MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

132

A REVIEW OF THE THEORY AND APPLICATION OF MULTI-CRITERIA DECISION ANALYSIS TECHNIQUES IN MINE PLANNING

strong preference up to Pk(d) = 1, representing strict preference. The PROMETHEE algorithm can be summarised into six steps as outlined below (Geldermann and Rentz, 2005): 1. For each criterion, k, specify a generalised preference function, Pk(d). The preference function can take any of six possible forms of function distributions which are the criterion distribution; quasicriterion distribution; criterion with linear preference distribution; level criterion distribution; linear and indifference area distribution; and Gaussian distribution. 2. Define a vector of weights that indicate the relative importance of each criterion given by, wT = [w1, …,wk] as expressed by the decision maker. 3. Define the outranking-relation,π, for all alternatives A1, An ∈ A as indicted by Equation 4: k r (Ai,A j) = /Wk # Pk (fk (A j)) (4) k=1 4. Calculate the leaving flow Φ+(Aj), defined by Equation 5: T z+(Ai) = 1 / r (Ai,A j) T j 1 = 5. Calculate the entering flow Φ-(Aj), defined by Equation 6: T z-(Ai) = 1 / r (A j,A i) T j 1 =

(5)

(6)

6. Perform a graphical evaluation of the outranking relation. Generally, the higher the leaving flow and the lower the entering flow, the better the alternative. The PROMETHEE methods use graphical output to show the partial preordering of the alternatives represented as nodes and the outranking relations depicted as arcs.

Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) technique The TOPSIS technique was developed by Hwang and Yoon (1981), also as an alternative technique to the ELCETRE technique. Triantaphyllou et al (1998) described the condensed form of the TOPSIS technique. The TOPSIS technique assumes that the decision maker or group of decision makers can adequately configure where the ideal and negative-ideal solutions would lie in a Euclidean space. The ideal solution is one that maximises desirable benefits at the same time minimising undesirable outcomes. The converse is true for a negative-ideal solution. The technique starts by converting dimensions of the attributes against which the alternatives are measured, into a dimensionless space through a normalisation process. The decision-maker then applies a set of weights to the normalised attributes to produce a normalised weighted matrix. The N-dimensional Euclidean distance method is then used to calculate the separation distance of each alternative from the ideal and negative-ideal solutions. The best alternative then, is the one with the shortest distance from the ideal solution and the furthest distance from the negative-ideal solution in Euclidean space.

Multiple-Attribute Utility (MAUT) technique The MAUT technique is based on utility, a concept that evolved from the branch of economics. In economic terms utility (U) is simply, satisfaction. Common among individuals or individual groups of people is the need to maximise utility by maximising desirable outcomes and minimising undesirable outcomes. The MAUT technique structures the problem as a hierarchy with the primary objective occupying the pinnacle of the hierarchy and having first-layer and second-layer objectives below the primary objective, arranged in terms of hierarchical importance. The objective is measured using attributes or criteria. Feasible alternatives are represented by, ai, attributes are represented by xj, the trade-off weights of attributes are represented by wij and pij is the most likely probability of attaining a predetermined value of efficiency measure which alternative ai scores against attribute xi. The overall relative utility (Ui) of an alternative, ai, is given by Equation 7. n (7) Ui = / pijwij j=1 MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

133

C MUSINGWINI

Analytic Hierarchy Process (AHP) technique Saaty (1980) developed the AHP methodology. Matrix and vector algebra form the basis of the mathematical framework of the AHP methodology, enabling AHP calculations to be easily performed in Microsoft Excel®. Additionally, there are generic off-the-shelf software such as Expert Choice® and DecisionLens® that can be used to solve AHP problems. The books, The Analytic Hierarchy Process, by Saaty (1980) and Decision-making for Leaders: The Analytic Hierarchy Process for Decisions in a Complex World, by Saaty (2008) comprehensively explain the AHP mathematical theory. The next paragraphs provide a brief discussion of the AHP theory. The AHP methodology is premised on four main axioms which are (Saaty, 1986; Harker and Vargas, 1987): x Axiom 1 (the reciprocity axiom): which says that given any two criteria Ci and Cj, the degree of preference of Ci over Cj is an inverse of the complementary preference decision of Cj over Ci. x Axiom 2 (the homogeneity axiom): which says that when comparing two alternatives or two criteria, the scale of the ratio of comparison is bounded (ie alternative/criteria one cannot be infinitely better than alternative/criteria two). x Axiom 3 (the dependence axiom): which says that the set of alternatives is dependent on the set of criteria if a fundamental scale can be defined to measure each alternative against each criterion (ie the decision problem can be formulated as a hierarchy). x Axiom 4 (the expectations axiom): which says that all alternatives and criteria which impact a decision-making problem are represented in a hierarchy and assigned priorities compatible with the expectations. It is not the intention in this paper to prove these axioms as such proof can be found in Saaty (1986) and Harker and Vargas (1987). Rather the intention is to recognise that the decision problems that can be solved using the AHP must satisfy all the four axioms. The mathematical procedure starts with a pair-wise comparison of the relative weight or importance of each criterion over another using the reciprocity axiom. The relative weight of Ci over Cj is denoted by wij such that, wij = 1 , 6i ! j , and wij=1, ∀i=j, since a criterion is as important as itself. Due to the wij

reciprocity of weights the matrix of weights is a square matrix, W = (wij), of order n, corresponding to the number of criteria. The matrix, W, is termed a reciprocal matrix because the inverse of the weight of one criterion over another is equal to the weight of the second criterion over the first one. For example if capital costs are twice as important as operating costs in choosing a mining method, then the logic is that operating costs will be expected to be half as important as capital costs. The matrix of weights, W, is then evaluated for transitivity. A relationship is transitive if the relative importance is multiplicative. For example, if criterion C2 is twice as important as criterion C1 and criterion C3 is three times as important as C2, then logically criterion C3 should be six times as important as C1. A matrix satisfying the transitive axiom represents consistent judgements. Typical human judgements are characteristically inconsistent to a greater or lesser degree and cannot satisfy the transitive axiom. The AHP methodology provides a way of measuring the degree of inconsistency in judgements. The transitive relationship between weights can be expressed mathematically as wik = wijwjk, ∀i,j,k. A vector, w, of order n can be established such that Ww = λw. The vector, w, is called an eigenvector of the matrix W and the constant λ is its corresponding eigen value. If the matrix, W, is consistent then λ = n. For inconsistent human judgements, the eigenvector, w, cannot satisfy the earlier condition but will satisfy the condition Ww = λmaxw such that λmax n. The difference between λmax and n indicates that there is some inconsistency in the judgements but, if λmax = n then logically, the judgements were consistent. Several methods are available for estimating the eigenvector. Of these, a close approximation of the eigenvector is obtained when geometric means are used to estimate the eigenvector elements. The rationale for geometric means is simple. If a typical scale of one to ten is used to denote the relative weights, then from the reciprocity axiom, the reciprocal weights 0.1 and ten will differ by an order of magnitude of 100. Costa (2007) indicated that geometric means are meaningful when evaluating data that differs by several orders of magnitude, the minimum order being three (ie the MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

134

A REVIEW OF THE THEORY AND APPLICATION OF MULTI-CRITERIA DECISION ANALYSIS TECHNIQUES IN MINE PLANNING

largest number is three times as big as the smallest number in the data set). The geometric mean is useful for such data because unlike the arithmetic mean, it tends to dampen the effect of very high or low values, which could bias the mean if an arithmetic mean were calculated (Costa, 2007). A Consistency Index, CI, is then calculated from λmax and n using the relationship defined by Equation 8: CI = ^

m max - nh ^n - 1h

(8)

In order to determine if judgements are reasonably consistent a Consistency Ratio, CR, is calculated by assessing the calculated CI against judgements that are made completely at random. Saaty (1980) simulated large samples of random matrices of increasing order and calculated their corresponding CIs which are random indices, RIs. For matrices of order between one and 15, Saaty (1980) established the corresponding RIs as shown in Table 2. TABLE 2 Random index (RI) for n-ordered matrix (source: Saaty, 1980). Matrix order RI

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

0.00

0.00

0.58

0.90

1.12

1.24

1.32

1.41

1.45

1.49

1.51

1.48

1.56

1.57

1.59

The CR is obtained by dividing the CI by its corresponding RI. Saaty (1980) suggests that if the CR exceeds 0.1 then the judgements are likely to be too inconsistent to be reliable and the assignment of weights to criteria should be redone. The threshold ratio of 0.1 can be interpreted to mean that the judgments are approximately ten per cent random and a ratio of 1.0 would therefore mean that the judgements are completely too random to be trusted. A CR ratio of 0 therefore implies that judgements are perfectly consistent (ie not random at all). In practice CRs of more than 0.1 are sometimes accepted provided there is adequate justification for their acceptance (Coyle, 2004). If the degree of inconsistency in judgements is acceptable, the efficiencies of all alternatives on a criterion, Oij, are then normalised to eliminate the effect of different units of measure for each criterion. For m alternatives on a criterion, the normalised Oij values denoted by, ONij, are derived as shown in Equation 9: O O Nij = m ij (9)

/ Oij

i=1

The matrix of normalised efficiency outcomes is finally multiplied by the eigenvector to obtain the aggregated AHP priority score. The decision is then made based on the logic that the higher the AHP priority score for an alternative, then the more preferable the alternative. There are three main limitations of the AHP methodology. Firstly, the AHP only works if the matrix for the criteria weights is a positive reciprocal matrix (Coyle, 2004). Positive reciprocity is satisfied if criterion Ci is x times more important than criterion Cj and correspondingly Cj is 1 times as x important relative to criterion Ci. Secondly, when the scale for measuring the relative importance of criteria with respect to each other is changed, say from a scale of one to ten to a scale of one to 20, the weight vector will also change, in some cases affecting the final decision (Coyle, 2004). Lastly, as the number of criteria to be compared increases, the number of pair-wise comparisons increases rapidly following a power function as shown in Table 3, thus clouding judgement and rendering the calculations more complex. For example, for the recommended maximum number of criteria of nine, a total of 36 comparisons have to be made.

Summary differences of multi-criteria decision analysis techniques The previous sections have discussed each MCDA technique individually but did not provide a comparison of the techniques. A summary of the structural comparison of the four broad categories of MCDA methodologies is shown in Table 4. The major difference among the techniques can be seen firstly, in whether or not the technique produces a complete preference order of alternatives and secondly, in whether the relative importance or weights of criteria are treated on a trade-off or non MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

135

C MUSINGWINI

TABLE 3 Relationship between number of criteria and comparisons (source: Kardi, 2006). Number of criteria

1

2

3

4

5

6

7

n

Number of comparisons

0

1

3

6

10

15

21

n(n - 1) 2

TABLE 4 Comparison of the four multi-criteria decision analysis methodology categories (adapted from Geldermann and Rentz, 2005; Chen, 2006). MAUT

AHP

ELECTRE

PROMETHEE

TOPSIS

Foundation

Classical MCDA approach

Classical MCDA but hierarchical approach

Outranking procedure

Outranking procedure

Classical MCDA approach

Theoretical basis

Utility function additive model

Pair-wise comparison (weighted eigenvector evaluation)

Pair-wise comparison (concordance analysis)

Pair-wise comparison (preference function)

Dimensionless Euclidean space evaluation model

Measurement of criteria

Numerical (nonnumerical data must be converted to numerical scale)

Numerical (nonnumerical data must be converted to numerical scale)

Numerical (nonnumerical data must be converted to numerical scale)

Numerical (nonnumerical data must be converted to numerical scale)

Numerical (nonnumerical data must be converted to a dimensionless numerical scale)

Determination of weights of criteria

Trade-off based weights (generate weights using swing, direct-ratio, or Eigenvector methods)

Trade-off (generate weights using Saaty’s Eigenvector and geometric mean)

Non-trade-off (does not provide procedure to obtain weights)

Non-trade-off (does not provide procedure to obtain weights)

Trade-off based weights

Result

Relative preference order

Relative preference order

A set of nonPartial and complete dominated alternatives ranking order

Relative preference order

trade-off basis. These differences are necessitated by the nature of the multi-criteria problem, the decision required and the type of information available for the decision-making process.

EXAMPLES OF SUCCESSFUL APPLICATION OF MULTI-CRITERIA DECISION ANALYSIS TECHNIQUES IN THE MINERALS INDUSTRY MCDA techniques are appropriate tools for analysing complex mineral industry decision-making problems because they have two unique desirable features that match the level of complexity of the problems to be solved. Firstly, the techniques can be applied to solve problems that have both quantitative and qualitative data, even including expert opinions. Expert opinion is usually sought in the minerals industry because data pertaining to and knowledge about the orebody, technical planning parameters, or commodity prices is neither never known with certainty nor fully understood. Secondly, the structure of MCDA techniques is such that it creates a decision-making environment that enables participation and collaboration by experts from different disciplines that have different objectives and stakeholders with different expectations, who consequently will have multiple decision-making criteria that most often are conflicting across disciplines. These factors increase the complexity of decision-making problems in the minerals industry. The overview in Table 5 presents a wide range of examples of case studies of application of MCDA techniques in the minerals industry for the period 1995 - 2010. Table 5 indicates a diversity of applications of MCDA techniques to decision-making in the minerals industry with the AHP appearing to be the most widely used MCDA technique. Most of the examples that were identified are post-2000, indicating a gradual recognition of the use of MCDA techniques in solving minerals industry problems. The main areas of application include mine planning, equipment or technology selection, location of a mining facility and strategy formulation. There are also potential areas of application of MCDA techniques in the minerals industry. Musingwini and Minnitt (2008) identified some of these areas to include performance evaluation of line managers for promotion, performance evaluation of operating shafts, ranking of projects competing for funding, measuring company performance on mining score card in meeting the requirements of the Mining MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

136

A REVIEW OF THE THEORY AND APPLICATION OF MULTI-CRITERIA DECISION ANALYSIS TECHNIQUES IN MINE PLANNING

TABLE 5 Examples of minerals industry problems solved using multi-criteria decision analysis techniques between 1995 and 2010. Source

Multi-criteria decision analysis decision problem

Owusu-Mensah (2010)

Used AHP to successfully recommend the optimal ore transport system for Obuasi gold mine in Ghana, to meet expected increase in production. Four transport options were evaluated against three broad criteria. The three criteria had a total of 15 sub-criteria linked to them.

Musingwini (2009)

Used the AHP to aid in the successful optimisation of level and raise spacing for a conventional platinum reef mining layout in a typical Bushveld Complex mining environment. A total of 15 different layout alternatives were compared against 12 techno-economic criteria.

Aghajani-Bazzazi, Osanloo and Karimi (2009)

Used a combination of the AHP and VIKOR methods to select the optimal equipment suite for overburden removal in surface iron ore and coal mines. Five alternative equipment suites were compared on the basis of seven selection criteria. This approach was used because the criteria were measured in a mixture of units that included deterministic data, fuzzy numbers, interval numbers and linguistic terms.

Musingwini and Minnitt (2008)

Used AHP to rank four mining methods to mine the UG2 reef on the Bushveld Complex on the basis of eight evaluation criteria.

Yavuz (2008)

Compared fuzzy multiple attribute decision making (FMADM) and AHP to successfully select the best site for the location of a natural stone processing plant. Four alternatives were compared on the basis of four main criteria that had 15 sub-criteria.

Karadogan, Kahriman and Ozer (2008)

Used AHP based fuzzy multiple attribute decision-making methodology to select the most suitable underground mining method for the Ciftalan Lignite Mine in Turkey. Five possible mining methods were compared on the basis of 18 criteria.

Wu et al (2007)

Successfully used AHP to advise the board of directors of Wugang Mining Cooperation in strategy formulation based on the order in which the company was weakest in terms of core competence for each of the four products (iron concentrates; pellets; copper and sulphur concentrates; and non-metallic concentrates). The products were compared on the basis of eight criteria clustered into three broad criteria.

Uysal and Demirci (2006)

Used a hierarchical multi-dimensional objective system similar to AHP to select the more suitable mining method for the ELI and GLI coalfields in Turkey. Two mining methods were compared on the basis of 19 criteria.

de Almeida, Alencar and Miranda (2005)

PROMETHEE II used to select the mining method for ornamental rocks that best satisfies a set of evaluation criteria. Six mining methods were compared on the basis of five criteria.

Ataei (2005)

Used AHP to select the best location of an alumina-cement plant in Iran. Five possible locations were compared on the basis of five criteria.

Vieira (2003; 2004; 2005)

Successfully used MAUT to select the best mining method from four possible methods to mine ultra-deep tabular gold deposits of the Witwatersrand Basin. Four mining methods were compared on the basis of 49 attributes clustered into five decision criteria.

Kazakidis, Mayer and Scoble Used AHP based Expert Choice® software to model mining scenarios for selecting the: (2004) • best rockbolt support system from 14 possible rockbolt support systems on the basis of ten criteria; • best option from five operational options to improve tunnelling advance rates based on seven criteria; and • mine with the highest risk to mine production performance arising from ground problems, from a set of eight mines in a mining company, based on four criteria. Bitarafan and Ataei (2004)

Used two methods, an AHP based fuzzy multiple attribute decision-making method and fuzzy dominance method, to select the optimal mining method for extracting the no 3 Anomaly at the Gol-Gohar iron mine in Iran. Seven mining methods were compared on the basis of 15 criteria.

Elevli and Demirci (2004)

PROMETHEE I and PROMETHEE II were used to select the most suitable underground ore transport system for a chromite mine in Turkey. Five alternative transportation systems were compared on the basis of six criteria.

Dessureault and Scoble (2000)

Used AHP to assist a mine to decide whether to purchase new drill-monitoring technology, maintain status quo, or retrain drillers and surveyors but require them to work more productively and safely. The three alternatives were compared on the basis of six criteria.

Liquin et al (1995)

Used AHP to select an optimal mining plan from a set of possible mining plans for a generic multi-criteria decision-making problem.

Charter in South Africa, comparison of different ore haulage systems and the evaluation of different support systems for production stopes. Other areas are tender selection, candidate selection from job interviews and ranking research project proposals for possible funding.

CONCLUSION This paper has demonstrated through examples of successful application of MCDA techniques that most decision-making in mine planning and equipment selection are multi-criteria in nature and MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

137

C MUSINGWINI

can be solved using MCDA techniques. The successful application of MCDA techniques has helped to eliminate ‘gut-feel’ and empirical decisions from being made due to limited or uncertain information by laying out a systematic, logical and transparent decision-making process that is defendable and repeatable by other decision-makers. This is of paramount importance in today’s modern world characterised by strong demands for good corporate governance. There is now a gradual recognition of and application of MCDA techniques in the minerals industry, particularly the AHP technique.

REFERENCES Aghajani-Bazzazi, A, Osanloo, M and Karimi, B, 2009. Overburden removal equipment selection through MADM methods: Application of modified VIKOR method, in Proceedings Eighteenth International Symposium on Mine Planning and Equipment Selection (MPES 2009) and Eleventh International Symposium on Environmental Issues and Waste Management in Energy and Mineral Production (SWEMP 2009), pp 134-141. Ataei, M, 2005. Multicriteria selection for an alumina-cement plant location in East Azerbaijan province of Iran, The Journal of The South African Institute of Mining and Metallurgy, 105(8):507-513. Ballington, I, Smith, G L, Lane, G R and Hudson, J, 2005. The systemic interdependency of closure decisions at shaft level, in Proceedings First International Seminar on Strategic Versus Tactical Approaches in Mining, pp 153-164 (The Southern African Institute of Mining and Metallurgy: Johannesburg). Belton, V and Stewart, T J, 2002. Multiple Criteria Decision Analysis: An Integrated Approach, (Kluwer Academic Publishers: Massachusetts). Bitarafan, M R and Ataei, M, 2004. Mining method selection by multiple criteria decision-making tools, The Journal of The South African Institute of Mining and Metallurgy, 104(9):493-498. Canadian Conservation Institute (CCI), 2005. Tools: Analytical hierarchy process (AHP) program [online]. Available from: [Accessed: 26 September 2009]. Chen, Y, 2006. Multiple criteria decision analysis: Classification problems and solutions [online], PhD thesis (unpublished), University of Waterloo, Canada. Available from: [Accessed: 4 February 2009]. Costa, J, 2007. Calculating geometric means, Buzzards Bay national estuary program [online]. Available from: [Accessed: 19 February 2008]. Coyle, G, 2004. The analytic hierarchy process (AHP), in Practical Strategy: Structured Tools and Techniques [online] (ed: G Coyle), Open Access Material, Pearson Education Ltd: Glasgow. Available from: [Accessed: 7 January 2008]. de Almeida, A T, Alencar, L H and de Miranda, C M G, 2005. Mining method selection based on multicriteria models, in Proceedings Application of Computers and Operations Research in the Mineral Industry (APCOM 2005) (eds: S D Dessureault, R Ganguli, V Kecojevic and J G Dwyer), pp 19-24. Dessureault, S and Scoble, M J, 2000. Capital investment appraisal for the integration of new technology into mining systems, Transactions of the Institution of Mining and Metallurgy: Section A, Mining Technology, 109:A30-A40. Elevli, B, Demirci, A and Dayi, O, 2002. Underground haulage selection: Shaft or ramp for a small-scale underground mine, The Journal of The South African Institute of Mining and Metallurgy, 102(5):255-260. Geldermann, J and Rentz, O, 2005. Multi-criteria analysis for technique assessment: Case study from industrial coating [online], in Journal of Industrial Ecology, 9(3):127-142. Available from: [Accessed: 19 February 2008]. Harker, P T and Vargas, L G, 1987. The theory of ratio scale estimation: Saaty’s analytic hierarchy process [online], in Management Science, 33(11):1383-1403. Available from: [Accessed: 16 October 2007]. Hwang, C L and Yoon, K, 1981. Multiple Attribute Decision-making: Methods and Applications (SpringerVerlag: New York). Karadogan, A, Kahriman, A and Ozer, U, 2008. Application of fuzzy set theory in the selection of underground mining method, The Journal of the Southern African Institute of Mining and Metallurgy, 108(2):73-79. Kardi, T, 2006. Analytic hierarchy process (AHP) tutorial [online]. Available from: [Accessed: 15 February 2008]. Kazakidis, V N, Mayer, Z and Scoble, M J, 2004. Decision-making using the analytic hierarchy process in mining engineering, Transactions of the Institution of Mining and Metallurgy: Mining Technology, 113:A30-A42. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

138

A REVIEW OF THE THEORY AND APPLICATION OF MULTI-CRITERIA DECISION ANALYSIS TECHNIQUES IN MINE PLANNING

Keeney, R L, 1992. Value Focused Thinking: A Path to Creative Decision-making, (Harvard University Press: Cambridge). Liquin, Z, Shihui, L, Lianfu, Z and Lianming, J, 1995. The analysis and practice of multi-objective decisionmaking technique for selecting a mining plan, in Proceedings APCOM XXV Conference, pp 255 - 259 (The Australasian Institute of Mining and Metallurgy: Melbourne). McCarthy, P L, 2002. Production scheduling [online]. Available from: [Accessed: 31 March 2006]. Mendoza, G A and Martins, H, 2006. Multi-criteria decision analysis in natural resource management: A critical review of methods and new modelling paradigms, Forest Ecology and Management, 230:1-22. Musingwini, C, 2009. Techno-economic optimisation of level and raise spacing range in planning a Bushveld Complex platinum reef conventional breast mining layout [online], PhD thesis (unpublished), University of Witwatersrand, Johannesburg. Available from: . Musingwini, C and Minnitt, R C A, 2008. Ranking the efficiency of selected platinum mining methods using the analytic hierarchy process (AHP), in Proceedings Third International Platinum Conference ‘Platinum in Transformation’, pp 319-326 (The Southern African Institute of Mining and Metallurgy: Johannesburg). Owusu-Mensah, F, 2010. Evaluation of transport options from KMS shaft to the mill at Obuasi mine [online], Anglogold Ashanti, MSc research report (unpublished), University of Witwatersrand, Johannesburg, South Africa. Available from: . Saaty, T L, 1980. The Analytic Hierarchy Process (McGraw Hill International: New York). Saaty, T L, 2008. Decision-making for Leaders: The Analytic Hierarchy Process for Decisions in a Complex World (RWS Publications: Pittsburg). Saaty, T L and Ozdemir, M S, 2003. Why the magic number seven plus or minus two, Mathematical and Computer Modelling, 38:233-244. Triantaphyllou, E, Shu, B, Nieto Sanchez, S and Ray, T, 1998. Multi-criteria decision-making: An operations research approach, Encyclopedia of Electrical and Electronics Engineering (ed: J G Webster), 15:175-186. Uysal, Ö and Demirci, A, 2006. Shortwall stoping versus sublevel longwall caving-retreat in Eli coal Fields, The Journal of the Southern African Institute of Mining and Metallurgy, 106(6):425-432. Vieira, F M C C, 2003. Utility-based framework for optimal mine layout selection, subject to multipleattribute decision criteria, in Proceedings 31st International Symposium on Application of Computers and Operations Research in the Minerals Industries, pp 133-149. Vieira, F M C C, 2004. Rock engineering-based evaluation of mining layouts applicable to ultra-deep, gold bearing, tabular deposits, PhD thesis (unpublished), University of the Witwatersrand, Johannesburg. Vieira, F M C C, 2005. An integrated, multi-disciplinary evaluation of ultra-deep layouts, in Proceedings Application of Computers and Operations Research in the Mineral Industry (APCOM 2005), Tucson, USA, pp 655-665. Wu, H, Yuan, J, Zhang, Y and Song, S, 2007. The evaluation of the core competition of the Wugang Mining 2 Cooperation using the Analytic Hierarchy Process, The International Journal of Mineral Resources Engineering, 1(2):119-126. Yavuz, M, 2007. Analytical hierarchy process for selection of roadheaders, The Journal of the Southern African Institute of Mining and Metallurgy, 106(8):569-575. Yavuz, M and Pillay, S, 2007a. Mining method selection by multiple criteria decision-making tool, The Journal of the Southern African Institute of Mining and Metallurgy, 104(9):493-498. Yavuz, M and Pillay, S, 2007b. EQS: A computer software using fuzzy logic for equipment selection in mining engineering, The Journal of the Southern African Institute of Mining and Metallurgy, 106(1):63-70. Yavuz, M, 2008. Selection of plant location in the natural stone industry using the fuzzy multiple attribute decision-making method, The Journal of the Southern African Institute of Mining and Metallurgy, 108(10):641-649. Zimmermann, H J, 1991. Fuzzy Set Theory and Its Applications, second edition, (Kluwer Academic Publishers: Boston, Massachusetts).

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

139

An Investigation to Integrate Optimum Long-Term Planning with Short Planning in Underground Mine Production Scheduling M Nehring1, E Topal2, M Kizil3 and P Knights4 ABSTRACT Generating short- and long-term mine production schedules in isolation and independently from each other means that only a local optimum can be achieved when each scheduling phase is carried out. The globally optimal solution however, can be achieved when integrating scheduling phases and accounting for the interaction between short-term and long-term activities simultaneously. This paper addresses the task of integrating short- and long-term production plans by combining the short-term objective of minimising deviation to targeted mill feed grade with the long-term objective of maximising net present value (NPV) into a single mathematical model. A case study is presented which compares the present values of the same stope data set scheduled via separate long and shortterm models and than by the proposed integrated model which applies a predetermined penalty for each tonne of metal deviation to target for the respective ore tonnage being processed. This allows an operation to obtain a globally optimal scheduling regime when taking into consideration the cost and loss of value resulting from fluctuations in grade to the mill over the short-term.

INTRODUCTION The traditional mine scheduling process is carried out in a number of phases. Over the medium to long-term an operation will seek to develop a schedule on a monthly or quarterly basis that maximises its NPV while adhering to various timing, resource and sequencing constraints inherent to that particular site. The results of this phase will generally form the inputs for the development of a short-term schedule. At this level of planning the focus turns to maintaining a feed supply to the mill of consistent grade. While some gradual re-balancing of the process route may take place the main throughput/recovery equilibrium is generally set based on the expected average grade of the material feed into the plant. An operation may therefore not be obtaining full value by feeding material into its plant with high fluctuations in grade ultimately resulting in reduced efficiency and recovery over the short-term. This may also be case with ore feed containing a certain impurity, such as arsenic in a copper ore. A plant will generally have the process balances in place for the targeted average impurity blend. Deviation from this target may also alter the balance and result in reduced efficiency and recovery. For the most part, traditional approaches deal with each phase separately whereby the solution for one forms the starting point for the next. As such, the segregation of these phases has meant that only a local optimum is possible. For a globally optimal result to be achieved, the implications of a long-term decision on short-term objectives must be considered. These interactions must be explicitly defined when attempting to integrate the two phases. A number of authors have advanced the development of mathematical programming tools which are capable of modelling the complex underground mine scheduling problem. However, these tend 1. GAusIMM, Student, School of Mechanical and Mining Engineering, CRC Mining, The University of Queensland, St Lucia Qld 4072. Email: [email protected] 2. MAusIMM, Associate Professor and Head of Mining Engineering Department, Western Australia School of Mines, Curtin University of Technology, Kalgoorlie WA 6433.

Email: [email protected] 3. MAusIMM, Senior Lecturer, MEA and Mining Program Leader – Director, Teaching and Learning School of Mechanical and Mining Engineering, The University of Queensland,

St Lucia Qld 4072. Email: [email protected] 4. MAusIMM, Professor, Executive Director – Mining Education Australia, BMA Chair and Head of Division of Mining Engineering, School of Mechanical and Mining Engineering,

The University of Queensland, St Lucia Qld 4072. Email: [email protected] MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

141

M NEHRING et al

to either focus on the long-term scheduling problem or the short-term scheduling problem with progression lacking in regard to the development of a tool which combines both. Another drawback associated with existing tools and what largely work against their more widespread implementation are their excess solution times. As such, there are a number of general improvements still to be made in regard to the development of mathematical programming tools seeking to optimise the mine production scheduling problem. This paper focuses on the integration of scheduling plans into a single model so as to obtain simultaneous short- and long-term solutions that satisfy the objectives of both scheduling horizons. It is widely accepted that mathematical programming techniques such as mixed integer programming (MIP) has the capability to efficiently model complex scenarios such as the production scheduling problem. A long-term production scheduling model using MIP was formulated by Trout (1995) and tested on 55 stopes in the 1100 orebody at Mount Isa over a three year period. The model used five binary and two linear variables to represent the four phases of stope production. The model was written using C++ code and solved in CPLEX on a dual 75 MHz processor Sun SparcStation 10/512 with 128 MB RAM. The solution process had to be interrupted prior to proof of optimality after 209 hours with the reported integer solution being obtained after 1.6 hours. When compared to an existing manually generated schedule the MIPPS model increased before tax NPV by 23 per cent from $273M to $337M (Trout, 1995). The schedule however was not implemented into the operation and is thus open to a number of improvements. Using Trout’s (1995) model as a basis, both Nehring and Topal (2007) and Little, Nehring and Topal (2008) added further improvements. Nehring and Topal (2007) formulated an additional function to limit simultaneous fill-mass exposure to a single side while limiting simultaneous stope exposure to no more than two sides. This new formulation was successfully trialled on a conceptual nine stope example. Following on from this Little, Nehring and Topal (2008) develops theories relating to natural sequence and natural commencement which allowed various production phases to be defined as a function of the first extraction commencement variable. Using the same software on a similar machine to produce the same production schedule and NPV the new formulation produced an 80 per cent reduction in the number of variables resulting in a 92 per cent reduction in solution time of just 23 minutes and 14 seconds. Topal (2003) developed a MIP model to minimise deviation to target production of three products produced from two main ore types at the Kiruna mine. The author defined ‘machine placements’, which were representative of loader capacities and contained a number of production blocks with an associated ore quantity of a certain quality in each time period. This significantly increased the efficiency of model by effectively scheduling machine placements (of which there are far fewer) as opposed to individual production blocks as well as eliminating the need for an additional reserve constraints. Early and late start algorithms further reduced the number of variables by fixing those that were currently in production thereby ensuring a smooth transition into the new production schedule as well as removing machine placements from the model that are known not to be active during certain time periods (Topal, 2008). The combination of these activities significantly reduced the number of variables in the model resulting in an optimal solution in under 100 seconds when implemented over 36 monthly time periods on a Sun Ultra 10 machine with 256 MB RAM. This solution improved deviation from targeted production to just six per cent down from 10 - 20 per cent compared to the manual schedule with no constraint violation. The model was ultimately implemented into Kirunas’ main stream scheduling process. Production scheduling on a short-term level tends to be more dynamic due to the uncertainty surrounding equipment failure, geotechnical issues and individual operator ability. Frequent re-running of the short-term scheduling model is often required in order to provide reassignments when necessary. Few literature items deal with the underground mine environment however in open pit mining, real-time machine allocation to cater for these dynamic circumstances has traditionally been handled by fleet management and dispatch systems. Most notably, White, Olson and Vohnout (1993) present the algorithms used by Modular Mining Systems’ computer based DISPATCH system which is reported to be fully implemented at more than thirty open pit mining operations around the world. The DISPATCH system itself comprises of three subsystems. Using raw mine topology of locations, elevations and roads and distance data a Best Path (BP) algorithm generates the shortest MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

142

AN INVESTIGATION TO INTEGRATE OPTIMUM LONG-TERM PLANNING WITH SHORT PLANNING

path between all pairs of locations in the mine road network. Taking travel time and optimal routing data as well as pit configuration, available trucks and shovels, blending requirements and machine priority data from the BP, a linear programming (LP) subsystem model is then used to generate optimal path flow rates in tonnes per hour to minimise haulage requirements. Given a list of trucks needing an assignment and current travel times and distances a dynamic programming (DP) subsystem model finally provides assignments for each truck in real-time. Significant productivity improvements of between seven per cent and 20 per cent have been reported in open literature at 13 open pit mining operation after installation of the system. In the work of Osanloo and Saidy (1999) a simple semi-dispatch system using linear programming is generated with the objective of minimising the total number of trucks required to meet production targets. Variables in this case relate to the number of trucks assigned to each individual haulage path. The model was trialled on the Sarcheshmeh open pit copper mine in Iran which was required to produce 80 000 tonne of ore and waste daily over three production shifts. LINDO software was used to solve the final problem which showed daily production could be achieved with just 30 ×120 tonne capacity trucks as opposed to 37 × 120 tonne capacity trucks. Knights and Li (2006) introduce an algorithm to optimise shovel extraction sequences with the objective of minimising ore tonnage deviation as well as minimising grade deviation from predefined targets. The formulation is tested and validated on a hypothetical ore deposit over a ‘rolling horizon ‘seven day period at daily intervals with reserves and shovel performance updated after each run. Tests showed that short-term shovel sequencing was successfully achieved. The authors noted that mining blocks in a single period were sometimes not connected. As such, further development will need to focus on maintaining equipment movement and relocations within a single period to an appropriate distance. Beaulieu and Ganache (2004) propose an enumeration algorithm based on dynamic programming for solving the fleet management problem in underground mines. From an initial starting position for each machine, the objective is to find the best route and schedule for each machine such that their destination is reached in the shortest possible time. The solution approach is tested on 60 instances on three networks comprising of 20 instances per network with four machines in operation in each instance. The authors go into extensive detail about routing and displacement of machines throughout the network of underground drives and haul-routes in a concerted attempt to remain in a conflict free state. The reality is that the underground mine environment is very rigid. As such, in most cases there is only a single route a machine can feasibly take in the working of an ore-movement. Unlike open pit dispatching, instances where multiple production machines are simultaneously operating on the same route are generally avoided, thus providing little justification for such a stringent focus on routing. In the work of Tsomondo (1999) an Underground Active Dispatch Model (UADM) based on linear and goal programming is developed to optimally allocate machines at the start of a shift based on static operating conditions. The active program allocates machines to a single trip job after which it is free to request another job. A total of six dispatching polices are implemented into the UACM, with each able to be interchanged at will. These include shortest travel time, earliest expected service time, minimum server-client slack time, maximum product quality, maximum quality with minimum slack time and critical work site ratio. While not all policies may be applicable in practise the author compares the productivity of each policy as LHD fleet size is increased concluding that each policy has a unique optimal fleet size for a given mine layout.

MODEL FORMULATION The proposed model utilises MIP to model the integrated short- and long-term production scheduling problem. For the purposes of this paper only items dealing specifically with the extraction of each stope and its subsequent ore-movements is presented to the scheduling model. As such, the inclusion of all external development, internal development, production drilling and backfilling activities as well as machine assignments and capacities, ore flow progression and ore-movement lengths are outside the scope of this paper. A shaft capacity will therefore be the single limiting resource constraint in scheduling stope or ore-movement production in each monthly time period and short-term interval. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

143

M NEHRING et al

All extraction related items for both the short- and long-term phases are presented in full, along with the formulations and constraints that tie the two previously separate models together.

Indices The model is defined in general terms using the following subscript notation: t long-term schedule time period: t = 1, 2, 3 … T s long-term stope identification: s = 1, 2, 3 … S f existing fillmass: f = 1, 2, 3 … F o short-term ore-movement identification: o = 1, 2, 3 …, O i short-term scheduling interval: i = 1, 2, 3 …, I

Sets Several sets are defined which aid in the formulation of constraints: adjs set of all stopes that are adjacent to and share a boundary with stope s badjf set of all stopes that are adjacent to and share a boundary with each existing fillmass f tpbt set of time periods that include all periods up to and including the current period t po set of intervals in which ore-movement o is available for production npo zt binti

set of intervals in which ore-movement o is unavailable for production set of short-term intervals i corresponding to each long-term period t set of short-term intervals that include all short-term intervals up to and including the current interval

pbo

the ore-movement o+1 that must proceed ore-movement o in order to maintain the natural stope extraction sequence

sos

set of ore-movements o associated with the production of each stope s

Parameters These parameter items represent the numeric inputs and conditions: rs extraction quantity for each stope s sct shaft/LHD/truck fleet capacity for each long-term time period t es earliest start time for stope s ls nt cfs gus gls reso grao llim M msci sopso

pena penb

latest start time for stope s present value discount factor associated with each long-term time period t the undiscounted cash flow generated by each stope s difference between targeted upper ore feed head grade and stope grade of stope s difference between targeted lower ore feed head grade and stope grade of stope s ore reserve contained within ore-movement o difference between target ore feed and grade of ore-movement o lower ore production limit large number, M = 10 000 000 ore tonnage capacity of haulage shaft in short-term interval i link between the ore-movements associated with the production of each stope sos and the long-term periods each ore-movement is expected to be available once production on the stope is initiated. Predefined as: if available = 1, if not = 0 penalty factor ($) applied to each tonne of metal deviation above target mill feed grade for the respective ore tonnage being processed across all short-term intervals penalty factor ($) applied to each tonne of metal deviation below target mill feed grade for the respective ore tonnage being processed across all short-term intervals

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

144

AN INVESTIGATION TO INTEGRATE OPTIMUM LONG-TERM PLANNING WITH SHORT PLANNING

Decision variables A total of three linear and two binary variables were required to reflect operating conditions and ultimately perform the scheduling task: wst 1 if production from stope s is scheduled for time period t 0 xoi

otherwise

ore tonnes extracted from ore-movement o in interval i 1 if extraction of ore-movement o takes place in interval i 0 otherwise

x1oi ai

metal tonnes produced above the predefined mill feed grade target for the respective ore tonnage being processed interval i

bi

metal tonnes produced below the predefined mill feed grade target for the respective ore tonnage being processed interval i

Objective function Maximize: / nt # cfs # wst + / ai # pena + bi # penb s,t i

Constraints

/ xoi # grao - ai + bi = 0

6i

o

(1)

/ xoi # reso tepo

60

(2)

/ xoi # msci

6i

(3)

o

/

ilebinti

xoil - reso # e

/ x1o'i o $ 0 olepbo

6o,i

(4)

/ wst # 1

6 s ls 2 T

(5)

/ wst = 1

6 s ls # T

(6)

t

t

/ rs # wst # sct

6t

s

(7)

/ gls # rs # wst $ 0

6t

(8)

/ gus # rs # wst # 0

6t

(9)

s

s

wst + w s't # 1

/

tletpbt

wst' +

6 s,t s'e adjs

/ ws't # 2 s'eadjs

/ ws't # 1 s'ebadjf

(10)

6s,t

6f,t

/ xo'i = reso' # wst # sopso' iezt'

(11) (12)

6s,t, o' o' e sos t = 1..2

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

(13)

145

M NEHRING et al

x1oi # llim - xoi # 0 x1oi # M - xoi $ 0

/ x1oi = 0 ienpo

6o,i 6o,i

60

xoi' x1oi $ 0 x1oi = binary integer wst = binary integer

(14) (15) (16)

(17)

The objective function seeks to maximise the present value of all activities under consideration by determining the optimal schedule within which to progress each stope through production. A cash penalty is applied to each tonne of metal produced above and below target for the ore tonnage being processed. This cash penalty is equivalent to the operational efficiencies and reduced recoveries from deviations in mill feed grade. Equation 1 ensures that the tonnage of metal produced across all ore-movements in each short-term interval balances with the target for the ore tonnage being processed (accounting for all over-production ai and under-production bi). Equation 2 ensures that ore production from each ore-movement does not exceed its reserve. Ore-movement production in any short-term interval cannot exceed the haulage shaft capacity as enforced by Equation 3. Equation 4 ensures full reserve production from each oremovement before production from the proceeding ore-movement can commence, thus maintaining the natural sequential transition into production from one ore-movement to the next. Equation 5 ensures that commencement of stope production is initiated no more than once during the long-term scheduling horizon if its late start date occurs beyond the maximum time horizon. Equation 6 requires that stope production commences at some point during the long-term scheduling horizon if its late start date falls within the long-term scheduling horizon. Equation 7 limits ore production from all stope extraction activities from exceeding the shaft/LHD/truck fleet capacity in any long-term time period. Equation 8 restricts the combined mill feed ore grade from all stope extraction activities in any long-term time period from exceeding a lower grade limit. Equation 9 restricts the combined mill feed ore grade from all stope extraction activities in any long-term time period from exceeding an upper grade limit. Equation 10 ensures that simultaneous production between all stopes that share a boundary does not occur. Equation 11 enforces stability of all stopes by limiting simultaneous production to two adjacent sides before itself commencing production and to a single adjacent side once having completed production to become a fillmass. Equation 12 ensures ongoing fillmass stability of all existing fillmasses by limiting simultaneous exposure to a single side. Equation 13 performs two crucial roles. It firstly ensures that if an ore-movement is selected for production, its entire reserve is extracted during its production availability period. Secondly, it also provides the critical link between short- and long-term variables by essentially forcing the simultaneous long and short-term production scheduling of a stope and all its respective oremovements if it is selected to be part of the schedule. Depending on how many months worth of longterm production is to be presented for short-term scheduling, this will alter how many months the constraint is applied. In this case the first two months are offered for short-term interval scheduling. Equation 14 ensures that the minimum amount of ore produced from an ore-movement in any interval must equal at least the lowest feasible movement limit. This constraint also ensures that if xoi = 0, x1oi = 0. Equation 15 ensures that if xoi is utilised, x1oi must equal one. Short-term interruptions to the production availability of ore-movements are able to be taken into account courtesy of Equation 16. Finally, Equation 17 enforces non-negativity and ensures the appropriate variables maintain binary values.

IMPLEMENTATION OF INTEGRATED MIXED INTEGER PROGRAMMING MODEL For the purpose of comparing the integrated scheduling model against traditional segregated long and short-term scheduling a small conceptual and marginal operation utilising the sublevel stoping MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

146

AN INVESTIGATION TO INTEGRATE OPTIMUM LONG-TERM PLANNING WITH SHORT PLANNING

technique with little capacity for stockpiling and having its ore toll treated at a nearby facility is used as a trial case. As shown in the plan view contained in Figure 1, the operation comprises of 14 stopes, three of which having already completed the extraction process to become a fully consolidated fill mass, leaving the remaining 11 stopes available for production. In scheduling production from this operation all adjacency interactions for all stopes as well as existing and future fill masses are enforced.

FIG 1 - Plan view of 14 stope operation.

The 11 remaining stopes contain a combined reserve of 375 553 tonne of ore grading 1.92 per cent Cu for a total of 7200 tonne of copper metal. An ore tonnage and grade breakdown for each stope is contained in Figure 2.

FIG 2 - Expected stope ore tonnage and grade.

Breaking down the data contained in Figure 2 and utilising existing stope production profiles for stopes with similar characteristics including size, shape and location the expected monthly drawdown down rates at the respective grades for each stope over its entire production life once production is initiated, is presented in Table 1. These predetermined production rates form the basis for long-term scheduling at monthly intervals. Following on from this, the expected cash flows generated from each stope when taking the predetermined production profile, metal prices and fixed and variable extraction costs into account, are calculated and presented in Figure 3. It is assumed that all stopes are available for extraction commencement from the outset. During each month of production the operations’ ore handling capacity is restricted to 30 000 tonnes at a MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

147

M NEHRING et al

TABLE 1 Expected monthly stope ore tonnages and grades once in production. Month 1 Stope

Month 2

Month 3

Month 4

Month 5

Tonnes

Grade (% Cu)

Tonnes

Grade (% Cu)

Tonnes

Grade (% Cu)

Tonnes

Grade (% Cu)

Tonnes

Grade (% Cu)

L280

6699

2.50

7602

2.30

6609

2.40

7071

2.40

6874

2.50

L281

8834

1.95

8096

2.10

8659

2.00

8759

2.00

M279

7645

2.40

7428

2.30

6971

2.50

7567

2.40

M280

8446

1.95

7983

2.05

8341

1.95

M282

9412

1.60

12 160

1.60

11 080

1.50

11 273

1.70

10 634

1.65

M283

10 502

1.60

9859

1.60

10 250

1.50

9854

1.70

N279

7629

2.30

6815

2.40

6375

2.50

N280

7451

2.00

7042

2.00

8056

1.90

7514

2.10

7980

1.95

N281

8627

2.00

9211

1.90

8050

2.10

N283

9642

1.65

10 859

1.60

9839

1.70

O282

9833

1.70

10 400

1.60

11 080

1.50

10 726

1.60

FIG 3 - Expected cash flows from individual stopes.

combined ore blend of no less than 1.7 per cent Cu and no greater than 2.2 per cent Cu, with a shortterm mill feed blend target of 1.9 per cent Cu. For the purpose of short-term scheduling it becomes necessary to break down each stope into smaller more definable blocks in order to more appropriately reflect the grade fluctuations that naturally occur within the stope and thus provide ore allocations that aid the short-term objective of minimising deviation to mill feed grade in each interval. As a result of this, the tabulated long-term production data for each stope contained in Table 1 is reworked further into its individual ore-movement availability before being presented to the mathematical model for evaluation. This is an activity that should be carried out in any case by any operation seeking reliable grade controls even under a manual scheduling regime. In this case, the first two months is scheduled for short-term purposes in which each month is divided into five equal intervals of approximately six days resulting in a total short-term scheduling period of ten intervals. Table 2 presents the ore-movement production tonnages and grades as well as the availabilities for selected stopes. As shown, the expected initial 6699 tonnes at 2.5 per cent Cu produced from stope L280 (the first stope in Table 4) for its first month in production in reality comprises of two individual ore-movements, MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

148

AN INVESTIGATION TO INTEGRATE OPTIMUM LONG-TERM PLANNING WITH SHORT PLANNING

TABLE 2 Ore-movement availabilities for selected stope.

Stope

Ore movement

Month 1

Month 2

Interval availability

Tonnes

Grade (% Cu)

Tonnes

Grade (% Cu)

6699

2.5

7602

2.3

L280a

3115

2.6

1 - 10

L280b

3584

2.4

3 - 10

L280

L280c

3727

L280d L281

2.2

6 - 10

3875

2.4

8 - 10

8096

2.1

8834

1.95

L281a

4579

2.0

1 - 10

L281b

4684

1.9

3 - 10

L281c

4550

2.0

6 - 10

L281d

3545

2.2

8 - 10

10 859

1.60

N283

9642

1.65

N283a

5201

1.7

N283b

4441

1.6

1 - 10 3 - 10

N283c

4949

1.5

6 - 10

N283d

5910

1.7

8 - 10

FIG 4 - Long-term scheduling results (tonnes produced from each stope in each month).

including 3115 tonnes at 2.6 per cent Cu for L280a and 3584 tonnes at 2.4 per cent Cu for L280b. Determination of all ore-movement availabilities has thus increased the initial 11 stopes into a total of 44 ore-movements for consideration across the ten intervals. Anticipating the timeframe that each ore-movement associated with each stope for a particular operation will be available to a high level of accuracy is not an easy task. The further away a particular oremovement is from production the more difficult this becomes. This is therefore an activity that relies on the judgement of an experienced person. Long-term scheduling will take place across a 12 month time frame. The cash flows associated with each stope are discounted at a rate of ten per cent per annum. Throughout each short-term interval a 6000 tonne ore production limit is enforced. A mill feed grade target of 1.9 per cent Cu is applied across all ten short-term intervals. A stringent feed grade must be maintained under toll treatment MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

149

M NEHRING et al

contracts. In this case, each tonne of copper production above target for the respective ore tonnage being processed in each interval is penalised at $500. Each tonne of copper production below target for the respective ore tonnage being processed in each interval is penalised at $1000.

COMPARISON OF RESULTS Construction of all models took place using a mathematical programming language (AMPL) which was then solved using CPLEX 10.3 on a standard office computer. Having established all long and short-term scheduling parameters the traditional segregated scheduling approach is firstly used. This involved generating a long-term production schedule spanning 12 months with the objective of maximising NPV. The results of this are displayed in Figure 4. As shown, commencing production with stopes L280, N280 and N283 which rank as first, equal sixth and eighth highest cash flow stopes respectively (on a cash flow per month in production basis) was determined to be locally optimal, producing a preliminary NPV (before accounting for grade deviation penalties) of $719 821.1M. Even though stopes M280 and N281 were mid range cash flow stopes, these were completely avoided. This is most likely due to their proximity to other stopes and the resulting delay in production from surrounding stopes due to the more complex fillmass interactions associated with these stopes. Following on from this, the first two months production results could be taken for input into the short-term production scheduling process. The ore-movement production data contained in Table 2 for stopes L280, N280 and N283 was therefore presented to the short-term mathematical production scheduling model with the objective of minimising deviation to the targeted mill feed grade of 1.9 per cent Cu across all ten short-term intervals. A $500 penalty for each tonne of over production and a $1000 penalty for each tonne of under production for the respective ore tonnage being processed are applied. The results of this are displayed in Table 3 which shows the ore and metal tonnes to be produced from each ore-movement across the ten intervals. TABLE 3 Segregated short-term scheduling results (ore tonnes from each ore-movement).

L280a

int1

int2

int3

765.9

1333.3

1015.8

L280b

int4

int5

10

3573

L280c

int6

int7

2367.4

1574.1

int10

586.6

1714.3

3737

N280b

3714

N280c

3095

N280d

3947

N283a

4666.7

N283b

534.3 2014

2427

N283c

493

1547.5

N283d Ore (t)

int9

1359.6

L280d N280a

int8

2908.4 157.9

1666.4

4285.7

4502.9

6000.0

3564.1

3724.0

6000.0

2860.5

4642.5

6000.0

6000.0

6000.0

Grade (%)

2.02

1.90

1.90

2.10

2.08

2.08

1.90

1.90

1.90

1.90

Metal (t)

90.9

114.0

67.7

78.2

124.6

59.5

88.2

114.0

114.0

114.0

Deviation (t)

5.4

0.0

0.0

7.5

10.6

5.1

0.0

0.0

0.0

0.0

Penalty ($)

2700.0

0.0

0.0

3750.0

5300.0

2550.0

0.0

0.0

0.0

0.0

Across all ten intervals a total production of 49 294 tonne of ore at an average grade of 1.96 per cent Cu containing 964 tonne of copper is achieved. The schedule in this case exceeds target mill feed grade which translates into an excess copper metal production for the respective quantity of ore MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

150

AN INVESTIGATION TO INTEGRATE OPTIMUM LONG-TERM PLANNING WITH SHORT PLANNING

being processed in intervals 1, 4, 5 and 6 of 5.4, 7.5, 10.6 and 5.1 tonnes respectively. This results in a total deviation of 28.6 tonnes of copper resulting in a total applicable penalty of $14 300. When subtracting this amount from the preliminary NPV generated by the long-term schedule, this results in a final NPV of $705 521.1 ($719 821.1M - $14 300). When considering that these penalties occur in the first two months of production which contribute $140 904.3 to the 12th month preliminary NPV, these penalties reduce the first two months contribution to NPV by 10.1 per cent to $126 604.3. The same data set under the same parameters is presented to the proposed integrated mathematical scheduling model. Table 4 compares the NPVs and corresponding month of commencement of each stope between the segregated and integrated scheduling approaches respectively. TABLE 4 Long-term schedule comparison (month of production commencement). Stope

Segregated scheduling

Ingegrated approach

L280

1

1

L281

6

6

M279

9

9

M282

12

12

M283

4

1

N279

6

6

N280

1

1

N283

1

5

O282

8

8

Long-term contribution ($)

719 821.1

719 721.8

Applicable penalities ($)

14 300.0

11 500.0

Final NPV ($)

705 521.1

708 221.8

M280

N281

It is apparent that some significant changes in the schedule have taken place, with stope M283 in the integrated schedule commencing in the first period as opposed to the fourth period. Stope N283 which commenced in the first period under the segregated approach has been pushed back in the integrated schedule to commence in the fifth period. Stopes M280 and N281 again been completely left out of either schedule. Table 5 presents the corresponding optimal short-term schedule associated with the new long-term schedule under the integrated approach. Across all ten intervals a total production of 49 154 tonne of ore at an average grade of 1.95 per cent Cu containing 957 tonne of copper is achieved in the integrated short-term schedule. This represents a slight decrease of 140 tonne in ore production over the segregated short-term results. As expected, average copper grades move toward the target mill feed grade, reducing by 0.01 per cent Cu to 1.95 per cent Cu when compared to the segregated short-term result. The integrated short-term schedule also exceeded target mill feed grade in intervals 1, 3, 5 and 10 with 4.2, 3.2, 8.2 and 7.4 tonnes of copper being produced above what is expected for the respective quantity of ore being processed. This results in a total deviation across the ten intervals of 23.0 tonnes of copper resulting in a total applicable penalty of $11 500. When subtracting this from the long-term schedules’ contribution to NPV, this results in a final NPV of $708 221.8 ($719 721.8M - $11 500). In this case, the applicable penalties of $11 500 reduces the contribution of the first two months to the 12 month preliminary NPV of $139 256.8 by 8.2 per cent to $127 756.8. Integrating both short- and long-term scheduling horizons into a single model has improved final NPV by $2700.1 to $708 221.8M. While this difference only represents a 0.4 per cent improvement over the final 12 month segregated scheduling approach, it should again be noted that only the first MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

151

M NEHRING et al

TABLE 5 Integrated short-term scheduling results (ore tonnes from each ore-movement).

L280a

int1

int2

761.7

1333.3

int3

int4

int5

int6

int7

int8

1844.9

2030.1

3583.0

L280c

2553.9

1173.1

L281d 568.3

4666.7

M283b

3850.0

2417.0

M283c

1915.4

1547.5

879.9

M283d N280a

3737.0

N280b

1584.0

457.3 3697.8

1361.2

2130.0

N280c

3095.0

N280d Ore (t)

int10

1020.0

L280b

M283a

int9

3947.0 1333.0

6000.0

5321.0

6000.0

6000.0

4469.3

4642.5

6000.0

6000.0

3391.3

Grade (%)

2.22

1.90

1.96

1.90

2.04

1.90

1.90

1.90

1.90

2.12

Metal (t)

29.5

114.0

104.3

114.0

122.2

84.9

88.2

114.0

114.0

71.9

Deviation (t)

4.2

0.0

3.2

0.0

8.2

0.0

0.0

0.0

0.0

7.4

Penalty ($)

2100.0

0.0

1600.0

0.0

4100.0

0.0

0.0

0.0

0.0

3700.0

two months (of the 12 month period) were evaluated for short-term scheduling purposes. In this case, the final NPV contribution of the first two months of the integrated result of $127 756.8 compared to that of the segregated result of $126 604.3, achieves an improvement of $1152, representing a gain of almost one per cent on this small example. The remaining difference in NPV can be attributed to increased cashflows over the remaining ten months due to the delay of higher cashflow stopes in earlier months as mill feed grade is a more prevalent value contributor in the earlier months. Even though the contribution of the first two months to the preliminary long-term NPV of the integrated schedule falls short- of the segregated long-term schedule by $1647.5 ($139 256.8 - $140 904.3) or 1.2 per cent, it more than makes up for this through a reduction in the applicable penalties by $2800 ($14 300 - $11 500), representing a 19.5 per cent decrease.

CONCLUSION While traditional approaches may find the locally optimal solution of each planning and scheduling phase, it is only a proper integrated approach that is able to yield the globally optimal solution. This is because all actions in one phase and the subsequent implications on another are simultaneously evaluated. As a result of this, extra value may be achieved that was previously hidden. Through the application of the new proposed integrated scheduling tool on a small conceptual underground copper operation an extra one per cent increase in NPV was achieved during the overlap between short- and long schedules. The operation under consideration had limited stockpiling capacity and was required to meet stringent mill feed grades of 1.9 per cent Cu or be charged penalties for deviating from this target. During the over lapping period between short- and longterm scheduling the integrated approach found that rescheduling long-term production for a 1.2 per cent drop in long-term value could be more than offset by a 19.5 per cent decrease in the applicable penalties for deviating from target mill feed grade. The application of this approach is also highly amenable to minimising mill feed deviation of an impurities which may also adversely affect process efficiencies and recoveries. Essentially, whatever the long and short-term objects may be an integrated approach should be used to achieve the globally optimal result. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

152

AN INVESTIGATION TO INTEGRATE OPTIMUM LONG-TERM PLANNING WITH SHORT PLANNING

FUTURE DEVELOPMENTS The use of a penalty rate for deviation to target mill feed grade as used in this paper lends itself to a number of alternative strategies which can be incorporated into the integrated model depending on the scheduling and operational objective. Placing an emphasis on production smoothing in order to avoid spikes in deviation from one interval to the next may also provide benefits to an operation. Assigning penalties which increase with an increase in deviation may be one way of addressing this issue. The next natural progression in the further development of the proposed integrated model is to apply it to a full data set which from a long-term perspective also incorporates external development, internal development, production drilling and backfilling activities. From the short-term perspective this would also incorporate machine assignments and the accompanying necessary information including, ore-movement lengths and ore flow progression throughout the ore pass system.

REFERENCES Beaulieu, M and Gamache, M, 2004. An enumeration algorithm for solving the fleet management problem in underground mines, Computers & Operations Research, pp 1606-1624. Knights, P F and Li, S, 2006. Short-term sequence optimisation – problem definition and initial solution, in Proceedings Australian Mining Technology Conference (CRC Mining), pp 385-394 (The Australasian Institute of Mining and Metallurgy: Melbourne). Little, J, Nehring, M and Topal, E, 2008. A new mixed-integer programming model for mine production scheduling optimisation in sublevel stope mining, in Proceedings Australian Mining Technology Conference (CRC Mining), 157-172 (The Australasian Institute of Mining and Metallurgy: Melbourne). Nehring, M and Topal, E, 2007. Production schedule optimisation in underground hard rock mining using mixed integer programming, in Proceedings Project Evaluation 2007, pp 169-175 (The Australasian Institute of Mining and Metallurgy: Melbourne). Osanloo, M and Saidy, S H, 1999. The possibility of using semi-dispatching systems in Sarcheshmeh copper mine of Iran, Computer Applications in the Minerals Industries, pp 243-252 (Colorado School of Mines, Golden). Topal, E, 2003. Advanced underground mine scheduling using mixed integer programming, PhD thesis (unpublished), Colorado School of Mines, Colorado. Topal, E, 2008. Early start and late start algorithms to improve the solution time for long term underground mine scheduling, Journal of the Southern African Institute of Mining and Metallurgy, pp 99-107. Trout, L P, 1995. Underground mine production scheduling using mixed integer programming, in Proceedings 25th International APCOM Symposium, pp 395-400 (The Australasian Institute of Mining and Metallurgy: Melbourne). Tsomondo, C M, 1999. A flexible underground mining system with active dispatching model – A solution for the next millennium, Computer Applications in the Minerals Industries, pp 463-473 (Colorado School of Mines: Golden). White, J W, Olson, J P and Vohnout, S I, 1993. On improving truck/shovel productivity in open pit mines, CIM Bulletin, pp 43-49.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

153

Application of Genetic Algorithms for Reliability Assessment of Two Mine Hoisting Systems N Vayenas1, X Wu2 and S Peng3 ABSTRACT This paper discusses the application of a computerised model based on genetic algorithms (GAs), called GenRel, for reliability assessment of underground mine hoisting systems. The purpose of this paper is to investigate whether GenRel can be applied to predict future failure data of a mine hoisting system based on historical records of failures. The incentive of selecting the GAs for prediction of failures is that GAs are a class of evolutionary algorithms which imitate the biological evolution procedures such as reproduction, selection, crossover and mutation. The reliability of mining equipment changes over time due to an array of factors (eg equipment age, the operating environment, number and quality of repair). These factors affect the equipment’s failure patterns and have complex impacts on the equipment’s reliability characteristics. The failure patterns are assumed to follow a biological evolution process, and thus GAs can be considered applicable in the modelling process. To conduct the reliability assessment, a study was carried out with historical failure data of hoists in the time period from January to December 2007. Two failure data sets of two mine hoisting systems were collected from two typical underground mines in Ontario, Canada, which are denoted as the NA mine and the SB mine. The failure data sets were prepared in the format of time between failures (TBF). Then, these data sets were entered in GenRel to generate predicted failure data sets for the period of January to March 2008. The paper discusses the statistical similarity of the actual failure data sets for the period of January to March 2008 with the predicted failure data set generated by GenRel in the same time period.

INTRODUCTION GenRel version 1.12 is a computerised model developed in MS-Excel using Visual Basic for reliability assessment of mobile mining equipment based on genetic algorithms (GAs). Past research using GenRel was based on the application of two probability distribution functions, the Exponential and the Lognormal distribution functions (Vayenas and Yuriy, 2007). Application of GenRel in real life data was focused on load-haul-dump (LHD) vehicles, (Nuziale and Vayenas, 2000; Yuriy and Vayenas, 2008; Vayenas and Wu, 2009). This paper refers to an improved algorithmic logic based on the availability of several probability distribution functions (Palisade Corporation, 2002). Then this logic is tested for its applicability in assessing mine hoisting systems. Overall, the main steps in the design principles of GenRel are as follows: x An adequate size of historical input data must be gathered (eg six months of failure data per piece of equipment). The data set must be divided into a set of raw input data and into a set of raw evaluation data. Failure data includes time between failures and type of failure per piece of mining equipment. x The above two sets of data are used to verify whether GenRel can be used as a predictor of the reliability of the mining equipment under study. If the verification process is successful, then the hypothesis that GenRel can be applied to predict future equipment failures based on these input data sets is acceptable. After the verification process is successfully completed, the raw input data set and the raw evaluation data set are merged to a single input data set. 1. Professor, Head of Laurentian University Mining Automation Laboratory, Laurentian University, 935 Ramsey Lake Road, Sudbury, Ontario P3E2C6, Canada. Email: [email protected] 2. Student, Laurentian University Mining Automation Laboratory, Laurentian University, 935 Ramsey Lake Road, Sudbury, Ontario P3E2C6, Canada. Email: [email protected] 3. Student, Laurentian University Mining Automation Laboratory, Laurentian University, 935 Ramsey Lake Road, Sudbury, Ontario P3E2C6, Canada. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

155

N VAYENAS, X WU AND S PENG

x In this step, GenRel is used to predict future failure data. After a distribution function is acquired to fit the raw input data set, the inverse transform technique is used to generate random numbers as offspring data sets (Law and Kelton, 2000). These offspring data sets start the iteration of cross-over and mutation with a probability of mutation entered as an input by the end user of GenRel. When the model is applied for prediction of future failure data, criteria for convergence are considered such as the maximum number of iterations and the numerical difference of the probability parameters of the generated data in comparison with the probability parameters of the input data. The algorithm is terminated when either maximum number of iterations or a convergence limit preset by the end user is reached. If these criteria are not met, it is then considered that the model cannot be used with sufficient statistical confidence and thus, other reliability methods for failure prediction should be examined. For a detailed discussion about the logic in GenRel, see Vayenas and Wu (2009).

DATA CLASSIFICATION AND MANIPULATION In this study, the first step is to collect time between failures (TBF) data from both mine hoisting systems. In this paper, we focus only on the skip hoist data. The work time of the mines per day is from 7.00 am to 6.00 am, and all delay codes applicable to the operation of the hoists under study are shown in Table 1. For the data manipulation, assumptions are made. The four main assumptions are as follows: 1. The operating hours of the skip hoist includes the hours of hoisting waste and the hours of hoisting ore. 2. In each day, the summation of the machine delay hours and the machine working hours equals to 24 hours. TABLE 1 Delay codes description and stand-by/failure classification for both mines.

Code

Explanation

Stand-by or failure

Available: Planned hang time

Stand-by

DD

Dest full: Surface bin is full

Stand-by

DE

Other equipment: Cage work that affects skip hoist

Stand-by

DM

Source empty: No muck feed from pocket

Stand-by

DP

Utilities down: No power, no air, etc

Stand-by

DY

High fines

Stand-by

DZ

High water

Stand-by

ME

Electrical down

Failure

MH

Hydraulic down

Failure

AV

MI

Instrument down

Failure

MM

Mechanical down

Failure

MO

Maintenance out of plan: Planned maintenance that has been rescheduled or has taken longer than planned

Failure

MP

Planned maintenance: Any kind of planned maintenance work or inspection – daily mechanical, weekly electrical/mechanical, etc

Failure

NT

Travel time: Shift change

Stand-by

NX

Not reported: Unaccounted for delay time

Stand-by

OI

Operator planned inspection: Any kind of planned operations work or inspection – shaft inspection, guide change, etc

Failure

OP

Operator repairs: Any kind of unplanned work/inspections done by operations – burst pipeline, shaft inspection due to stray bell, etc

Failure

RF

Preop: Hoist run-ins

Stand-by

Note: Failures include scheduled maintenance and unscheduled downtime.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

156

APPLICATION OF GENETIC ALGORITHMS FOR RELIABILITY ASSESSMENT OF TWO MINE HOISTING SYSTEMS

3. In the original data sheet, if a data cell is blank, it is assumed that there is an unaccounted delay time and the data cell is filled with delay code ‘NX’. Table 1 shows the description of all the delay codes. 4. The delay types of the skip hoist are divided into types of machine stand-bys and types of machine failures. Machine stand-bys represent the delays of the skip hoist which are due to failures or delays caused by other sources than the machine itself. Machine failures represent the delays of the skip hoist which are due to the machine itself. Table 1 shows the classification list in regards to the delay codes. After data manipulation, the failure data of both mine hoisting systems in the time period from January to December 2007 was transformed into a failure database as shown in Table 2. Then, the time between failures (TBF) data was entered into GenRel to generate new failure data. TABLE 2 An example of the failure database after data manipulation.

Time between failures

Type of failures

Time to repair

4.75

MP

14

8

MP

2

7.5

OI

9

0

MP

1

0

OI

9

16

MP

10

12.25

MP

10

10.25

MP

25

11.75

MP

1

0

MO

8

0

MP

1

13.25

OI

9

0

MP

1

0

OI

9

VERIFICATION OF THE PREDICTABILITY OF GENREL For the purpose of the statistical analysis of the failure data based on probability distribution fitting, the failure data should be tested for trends and serial correlations (Vayenas, Runciman and Clement, 1997). The purpose of these tests is to verify the assumption that the failure data is independent and identically distributed (IID). In our study, the trend test presents almost a linear relation between cumulative failure number and cumulative TBFs, which indicates that the TBF data is independently and identically distributed. The serial correlation test shows a scattered pattern, which indicates there is no serial correlation in the TBF data. Initially, the historical TBF data of the skip hoists of the NA mine and the SB mine in the time period from January to December 2007 was entered into GenRel to verify the model’s predictability. This was done separately for each of the two mines. For the NA mine, the failure data set, containing 444 data points, was divided into the raw input date set and the raw evaluation data set to verify the predictability of GenRel. The raw input data set consists of the first 222 data points of the original failure data set, and the raw evaluation data set consists of the last 222 data points. For the SB mine which contains in total 640 data points, the raw input data set consists of the first 320 data points of the original failure data set, and the raw evaluation data set consists of the remaining 320 data points. Figure 1 shows a snapshot of the GenRel’s input menu. Key model parameters are set as follows: x maximum number of GAs iterations: ten, x convergence limit: one per cent, and x probability of mutation: six per cent. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

157

N VAYENAS, X WU AND S PENG

 FIG 1 - A snapshot of the input menu of GenRel V.1.12.

Here, the convergence limit parameter of the input menu was used to define the user permitted difference of the probability distribution parameters between the input data set and the generated data set. In this case, assuming, for example, that the best fitting probability distribution function is the normal probability distribution, if the difference of the probability distribution parameters between the input data set and the generated data set D = |input data set - generated data set |+|σ input data set σ generated data set | is smaller than one per cent × (|input data set|+| σ input data set |), where  is the mean and σ is the standard deviation of the respective data set, then, the verification procedure is considered to be satisfactory. The probability of mutation parameter is used to constraint how often mutation operations are performed, see (Haupt and Haupt, 2004). In our case, it was found that ten iterations sufficed for the verification process. Model parameters in the prediction procedure were set to be identical to the model parameters used in the verification procedure.

GENERATION OF PREDICTED DATA SETS After the successful completion of the verification process for the input data from both hoists for the period from January to December 2007, GenRel can now be considered applicable for the prediction of future failure data, for example, for the three months period of January to March 2008 based on the past three months period of October to December 2007. After several model iterations based on historical data in the time period from October to December 2007, the parameters of the generated data sets in the period from January to March 2008 for both mines were within the convergence limit of one per cent which was set as an input. The generated failure data set of 78 data points for the NA mine follows the Weibull Distribution with a shape parameter  = 1.7603, a scale parameter  = 8.3888, and a location parameter  = 0.62203. The generated failure data set of 186 data points for the SB mine follows the exponential distribution with a mean  = 4.461 and a location parameter  = 0.47568. To test the statistical similarity between the predicted data set and the actual data set in the time period from January to March 2008 at both mine sites, the t-test and F-test were applied (Kanji, 1999). The t-test investigates the significance of the difference between the mean of the generated MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

158

APPLICATION OF GENETIC ALGORITHMS FOR RELIABILITY ASSESSMENT OF TWO MINE HOISTING SYSTEMS

data and the mean of the actual data (from January to March 2008). If the t-test statistic (at a level of significance  = 0.05) is less than a specific critical value, then there is no significant difference in the population means at the given level of significance. The F-test investigates the significance of difference between the variance of the generated data and the variance of the actual data (from January to March 2008). If the F-test statistic (at a level of significance  = 0.05) is less than a specific critical value, then, there is no significant difference in the population variances at the given level of significance. Table 3 and Table 4 show the results for both the NA mine and the SB mine. Table 3 indicates that the predicted data set from NA mine passes both the t-test and the F-test. Figure 2 shows a comparative graph between the generated failure data and actual failure data (from January to March 2008) for the NA mine with data in an ascending order. TABLE 3 Comparison between the predicted data set and the actual failure data set for the NA mine in the time period from January to March 2008.

Predicted failure data set by GenRel

Actual failure data set

Population

78

92

Mean

8.02

9.83

Variance

18.70

72.97

t-test statistic

0.26

t-test critical value

1.645

F-test statistic

1.05

F-test critical value

1.35

(Source: Kanji, 1999.)



$FW XDO 'DW D

*HQHU DW HG 'DW D



7%)

                          1XPEHU  RI  )DL O XU HV

FIG 2 - Graph of generated failure data by GenRel and actual failure data (from January to March 2008) for the NA mine with data in ascending order.

Table 4 indicates that the predicted data set from SB mine passes both the t-test and the F-test. Figure 3 shows a comparative graph between generated failure data and actual failure data (from January to March 2008) for the SB mine with data in an ascending order.

FINAL REMARKS The study of failure data for both the SB mine and NA mine proves the applicability of GenRel as a software tool to predict future failure data based on historical failure data in hoisting systems in underground mines. The comparison between the predicted data set and the actual data set using the MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

159

N VAYENAS, X WU AND S PENG

TABLE 4 Comparison between the predicted data set and the actual failure data set for the SB mine for the period of January to March 2008.

Predicted failure data set by GenRel

Actual failure data set

Population

186

182

Mean

4.96

5.68

Variance

18.83

27.30

t-test statistic

0.07

t-test critical value

1.645

F-test statistic

1.13

F-test critical value

1.35

(Source: Kanji, 1999.)

$FW XDO 'DW D



*HQHU DW HG 'DW D

 

7%)

      

                1XPEHU  RI  )DL O XU HV 

FIG 3 - Graph of generated failure data by GenRel and actual failure data (from January to March 2008) for the SB mine with data in ascending order.

t-test and F-test shows no significant difference between them. In our study, it was also observed that the number of iterations in GenRel appeared to vary due to the convergence limit or the probability of mutation. Therefore, further investigation is required to assess the impact of the variability of key model parameters such as the convergence limit or the probability of mutation.

ACKNOWLEDGEMENTS The authors wish to thank the Centre for Excellence in Mining Innovation (CEMI) and Vale Inco Ltd for their support to this research study.

REFERENCES Haupt, R L and Haupt, S E, 2004. Practical Genetic Algorithms (John Wiley and Sons: New Jersey). Kanji, G, 1999. 100 Statistical Tests, 224 p (Sage Publications). Law, A M and Kelton, W D, 2000. Simulation Modeling and Analysis, Third Edition (McGraw-Hill). Nuziale, T and Vayenas, N, 2000. A software architecture for reliability analysis of mining equipment, International Journal of Mining, Reclamation and Environment, 14(1):19-34. Palisade Corporation, 2002. A Concise Summary of @risk Probability Distribution Functions (Palisade Corporation: Newfield). MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

160

APPLICATION OF GENETIC ALGORITHMS FOR RELIABILITY ASSESSMENT OF TWO MINE HOISTING SYSTEMS

Vayenas, N, Runciman, N and Clement, S R, 1997. A Methodology for Maintenance Analysis of Mining Equipment, International Journal of Mining Reclamation and Environment, 11(1):33-40. Vayenas, N and Wu, X, 2009. GenRel: A computerised model for reliability prediction of mining machinery, International Journal of Mining, Reclamation and Environment, 1:365-382. Vayenas, N and Yuriy, G, 2007. Using Genrel for Reliability Assessment Of Mining Equipment, Journal of Quality in Maintenance Engineering, 13(1):62-74. Yuriy, G and Vayenas, N, 2008. Discrete-event simulation of mine equipment systems combined with a reliability assessment model based on genetic algorithms, International Journal of Mining, Reclamation and Environment, 22(1):70-83.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

161

Blast Vibration Monitoring and Elastic Wave Reflection Models to Assess Blast-Induced Damage to Mine Infrastructure – An Underground Case Study K G Fleetwood1 and E Villaescusa2 ABSTRACT During underground mining, rock mass changes can occur as a function of the mining sequence and excavation processes. Mining-induced rock mass damage can occur due to redistribution of static stresses and the application of dynamic stresses from blasting and seismic activity. These rock mass changes can result in a degradation of strength of the bulk rock mass and the remaining rock surrounding existing excavations. Rock mass damage near existing excavations which lead to excavation-scale instabilities such as rock fall, excavation closure, or visible rock mass fracturing can detrimentally affect the safety and profitability of mining operations. The focus of the current work was the prediction and assessment of damage to mine infrastructure occurring in relation to the blasting of a sublevel open stope. The study case involved a 60kt stope located in a highly-competent rock mass at a large Australian sublevel open stoping mine. The mine infrastructure of interest for investigation included local stope access development, critical access and haulage development crossing a regional-scale fault, an explosives storage magazine and a ventilation return air raise. The methods used to predict or assess the blast-induced damage to each infrastructure element in response to stope extraction were x intact rock strength testing, x discontinuity mapping, x linear-elastic static stress modelling, x blast-damage mapping, x near-field blast vibration monitoring and analysis, x geophysical assessment using inferred stiffness and wave velocity methods, and x post-extraction void assessment. The results of the detailed blast damage prediction and assessment program at the mine site have been discussed by the primary author elsewhere (Fleetwood, 2010). The current work will address blast-induced damage to mine infrastructure as a result of extraneous blasting vibrations from stope production blasting. A method for predicting the amplification of excavation surface motions from reflecting body waves will be discussed as well as the results of damage mapping around the study stope. The results of the damage mapping program identified blast-induced damage that occurred to the surface of the explosives storage magazine and local stope access development. No damage was observed at the main fault crossing or return air raise where higher vibration amplitudes were measured or predicted.

INTRODUCTION The ability to predict or assess damage to mine infrastructure resulting from nearby blasting can aid in establishing guidelines for limited access to mine areas during stope extraction. These predictions can also identify areas of mine development or other infrastructure in which additional surface support may be installed to help prevent rock fall or other damage from blasting vibrations. Existing research has identified peak particle velocity (PPV)-related excavation damage thresholds ranging from less than 100 mm/s up to 500 mm/s (eg u, 1980 Page, 1987). Therefore, assessment

1. Senior Research Fellow, Western Australian School of Mines, CRC Mining, PMB 22, Kalgoorlie WA 6430. Email: [email protected] 2. Professor of Mining Geomechanics, Western Australian School of Mines, CRC Mining, PMB 22, Kalgoorlie WA 6430. Email: [email protected] MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

165

K G FLEETWOOD AND E VILLAESCUSA

of existing vibration-induced damage becomes a planning tool for geotechnical engineers for scheduling of rehabilitation or installation of secondary surface support prior to blasting. The following sections detail a blast damage study conducted in conjunction with extraction of a stope at an Australian sublevel open stoping mine. Similar damage investigations were performed for a total of four stopes at the mine site over the course of four years. Each damage investigation included rock mass characterisation, near-field blast vibration monitoring, damage mapping, geophysical investigation of the rock masses surrounding each stope and post-extraction void analysis (Fleetwood, 2010). Details of the blast vibration-induced damage investigation to nearby development, a main access way fault crossing and an explosives storage magazine in the proximity of one of the study stopes will be discussed. In addition, a proposed method for predicting excavation surface response amplifications using elastic wave reflection relationships will be discussed and applied to mapped damage locations around the stope.

STUDY STOPE SUMMARY The 60 kt, single lift study stope selected for the analysis presented in this paper was adjacent to a number of previously-mined and backfilled stopes and was located in proximity to critical mine infrastructure elements (Figure 1). These elements included x localised stope access development x main underground explosives storage magazine x primary ore haulage-way and travel-way crossing the regional-scale fault and x 4 m diameter, raise-bored unsupported ventilation return air raise (RAR).

FIG 1 - Isometric view of the study stope and the neighbouring mine infrastructure (looking south-southwest).

The stope dimensions were approximately 20 m 20 m footprint and 40m height. The stope was mined from the 220 mLv level to the 180 mLv using a combination of downholes and upholes. The stope was extracted using a total of 11 production firings.

BLAST VIBRATION MONITORING AND MODELLING Near-field ( 10 m) and intermediate-field ( 30 m) blast vibration monitoring around the study stope was performed for three different wave propagation orientations. The initial concern for monitoring of stope blasting was the potential for significant damage to the adjacent ventilation return air raise (RAR). This raise was one of two raises in the mining block and represented critical mine infrastructure supporting mining of the ore block over the following years. The instrumentation program for the stope consisted of four tri-axial transducers cement-grouted into boreholes along specified orientations. A 500 g tri-axial accelerometer and a tri-axial geophone sonde were placed in an array at distances of 5.2 m and 15 m from the northwest stope boundary to monitor the blasting vibrations within the pillar containing the RAR. Each transducer was located on either side of the RAR such that the vibrations at the RAR surface could be interpolated from the MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

166

BLAST VIBRATION MONITORING AND ELASTIC WAVE REFLECTION MODELS

two sets of data. The additional two tri-axial geophone sondes were located perpendicular to the northeast stope wall at 12.1 m from the stope and below the stope at a distance of 16.4 m, near the main access fault crossing on the 240 mLv. The results of the monitoring program were used to predict the average or maximum vibration amplitudes at the locations of concern and to develop orientation-specific attenuation relationships. The predicted values of vector sum peak particle velocity at each damage location or infrastructure element will be discussed in the following sections.

MAPPING OF BLAST-INDUCED DAMAGE TO EXCAVATIONS Visual inspections were carried out after each main stope firing as a method of assessing the blastingrelated damage to mine infrastructure in the proximity of the study stope. The inspections were generally performed one shift after the stope firing upon re-entry such that any freshly-formed large fractures or fall-off from the development or RAR could be detected and mapped prior to disturbance. Inspection of the RAR was limited to observation through the ventilation windows located at the 180 mLv, 200 mLv and 220 mLv development levels. Prior to initiation of blasting in the study stope, the 240 mLv magazine was visually inspected and existing cracks in the shotcrete on the walls and backs were recorded photographically.

Mapped damage locations from stope firings Blast-induced damage events to the development drives of the upper stope access level (180 mLv) were mapped following stope firing events as part of the damage investigation. The damage events were located on a map and photographed such that records of the ground support conditions, locations and types of damage could be assessed. Blast-induced damage to the development generally occurred in areas where there was no surface support or the existing surface support was damaged or was installed poorly in the case of fibrecrete. Standard ground support and surface support in the development of the 180 mLv consisted of split-set rock bolts and steel-weld mesh installed from gradeline to gradeline (1.5 m from the floor). A majority of observed damage to development followed one of two classifications. The first type of damage was fall of preformed blocks, which were detached along existing geologic structures. This type of damage was characterised by geometrically regular and oxidised failure surfaces, and has been referred to as discrete block falls. The second type of observed damage was spalling damage, typically indicated by shallow or thin rock sheets and fresh, irregular fracture surfaces. At several damage locations, the discrete block fall consisted of shotcrete that detached from the rock mass and fell away from the excavation. This type of behaviour was observed in the damage recorded in the explosives storage magazine. The locations of all mapped blast-related damage events are shown in Figure 2 and characteristics of the mapped damage locations are listed in Table 1.

FIG 2 - Plan view of the 180 mLv stope access development and 240 mLv magazine mapped damage occurring with all firings of the study stope (after Fleetwood, 2010). MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

167

K G FLEETWOOD AND E VILLAESCUSA

TABLE 1 Distance from source point, associated blast, ground support conditions at excavation damage site and predicted peak particle velocity for all mapped damage events. Damage site

Type of damage

Blast associated with damage

3D distance from blast site (m)

Surface support at damage location

1

Medium area spall shoulder

South mass firing

25.8

Damaged steel-weld mesh

2

Medium block fall gradeline

North main rings

26.0

N/A

3

Medium spall gradeline

South mass firing

25.6

Damaged steel-weld mesh

4

Medium block fall < gradeline

Cut-off slot

24.4

N/A

5

Medium spall < gradeline

Cleaner ring

20.1

N/A

6

Medium spall < gradeline

Cut-off slot

17.9

N/A

7

Medium block fall < gradeline

Cut-off slot

11.0

N/A

8

Large area spall < gradeline

Cut-off slot

11.0

N/A

9

Large block fall < gradeline

South mass firing

11.8

N/A

10

Shotcrete spall

North main rings

63.2

25 mm shotcrete

11

Shotcrete spall

North main rings

69.4

25 mm shotcrete

The sizes of failures observed for discrete block falls were typically of the order of 20 kg up to several tonnes. Spalling failures generally occurred from the wall of the drive below the mesh grade line. Mapped spall damage was typically less than 100 mm thickness and covered areas of 1 m2 to 10 m2. Figures 3 and 4 show examples of discrete block falls, spall damage and detached shotcrete. Although most of the blasting-related damage events did not pose significant risk to personnel or equipment, additional mine services were required to perform spot scaling where damage was viewed as representing a significant risk to charge-up personnel.

(A)

(B)

FIG 3 - (A) Photograph of a discrete block fall (event 9) from 1.2 m height; and (B) spall failure (event 8) from floor to grade line; (B) in stope access development after cot-off slot firings. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

168

BLAST VIBRATION MONITORING AND ELASTIC WAVE REFLECTION MODELS

(A)

(B)

FIG 4 - (A) Pictures of spalled shotcrete (event 10); (B) (event 11) in the 240 mLv explosives magazine occurring with the north main rings firing within the study stope.

PREDICTION OF VIBRATIONS ASSOCIATED WITH DAMAGE EVENTS The results of blast vibration analysis and elastic wave reflection relationships were used to predict values of the aggregate surface-normal particle velocities associated with the mapped damage events. This prediction process involved several steps that are not typically considered in classical blast vibration analysis. The unique components of this analysis process were consideration of the orientation-specific attenuation anisotropies on the full-field wave amplitudes, the partitioning of coupled wave types at the peak amplitudes and determination of wave reflection amplification factors at the excavation boundaries.

Full-field wave attenuation relationships from blast monitoring The first step in the prediction of the surface normal aggregate particle velocities responsible for the recorded damage events was determination of the vibration attenuation relationships derived from the blast monitoring program. The method of multi-variable nonlinear blast vibration analysis used in this study has been detailed previously by the authors (eg Fleetwood et al, 2009 Fleetwood, 2010). The analysis process reports the regression constants for the general form of the scaled distance relationship for predicting peak particle velocity (PPV) based on the charge weight ( ) and the distance from the source midpoint to the point of interest (D). The nonlinear scaled distance regression approach has been chosen due to the improved prediction accuracies when compared with alternative near-field approaches. The attenuation relationships for each stope firing type were determined from the analysis of each transducer data set collected for the associated firings to increase the accuracy of the prediction. For example, the PPV at a damage location associated with a cut-off slot firing would be predicted using only the data measured from the cut-off slot firings. The analysis performed for each monitoring orientation could then be integrated to define the full-field vibration contours using elliptical calculations of points of equal vibration amplitude. The roughly perpendicular nature of the vibration transducer orientations and observed rock mass anisotropy orientations from other analysis allowed this type of calculation approach. The coordinates of the points of equal vibration amplitude and the MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

169

K G FLEETWOOD AND E VILLAESCUSA

value of predicted vector sum peak particle velocity (VSPPV) could then be contoured to visualise the wave propagation fields. Figure 5 illustrates the predicted vibration contours around a blast point for the data measured from the cut-off slot firings.

FIG 5 - Plan view of elliptical vibration amplitude contours of vector sum peak particle velocity from a representative cut-off slot firing in the study stope.

The source locations for each blast were determined as the average of the three-dimensional (3D) coordinates of each charge in the firing. The charge weight used for estimation of the vibration contours was the average of all charge weights in the blast. Therefore, the predicted PPV could be considered to be an average value based on all blastholes in the blast as the actual charge contributing to the damage could not be determined. An alternate approach would be to determine the maximum PPV from the combination of the largest charge weight and minimum distance. This approach has been used for predictions of VSPPV in the assessment of the return air raise and the main access fault crossing. Variable locations and charge weights of the different blastholes could not be integrated into the orientation-specific PPV contouring approach.

Partitioning of radial and shear components at vector sum peak particle velocity from measured blast vibrations The fundamental behaviour of longitudinal (P-) and distortional (S-) waves differ greatly when reflected from a free surface. Therefore, the characteristics of each wave type present in the complex coupled blast wave interacting with an excavation should be considered to determine the aggregate effect of wave reflection on the amplitudes at the surface. This consideration is complicated, as information on the partitioning of wave types within a complex, blast-induced body wave is scarce in the existing research. Analysis of vibration data collected from 30 similarly-charged blastholes at different elevations within the stope has revealed information on the partitioning of wave types dependent on the angle between the monitoring point and the blasthole axis. The analysis procedure involved rotation of the waveforms into the principal wave motion directions (radial, vertical shear and transverse shear) from each charge, and determining the amplitudes of the three components at the peak vector sum. The transverse and vertical shear components were then combined into the vector sum in-plane shear amplitude. The radial and vector sum shear amplitudes were then compared with the peak vector sum amplitude to determine the ratio of the component to the peak vector sum. This ratio for each wave type was then plotted against the angle of the transducer from the blasthole axis. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

170

BLAST VIBRATION MONITORING AND ELASTIC WAVE REFLECTION MODELS

The partitioning of radial components (P-wave) and the vector sum or transverse and vertical components (in-plane SV- and SH-waves) was found to be largely independent of the peak amplitude or any other blast-related factors such as charge weight or distance for the coupled blast waves. The geometric factor was strongly correlated, where the angle between the transducer and the charge axis varied from 45 to 85 . At angles close to 45 , expected behaviours were observed based on mathematical elastic wave models containing shear waves (eg Heelan, 1953 Hazebroek, 1966). The relationship between the component PPVs and the VSPPVs in relation to the angle from the blasthole axis is shown in Figure 6. For angles less than 45 , an axis of symmetry was assumed for prediction of partitioning from angles of 0 to 45 . The sum of the ratios of the two components at each angle to the blasthole in Figure 6 would not be expected to equal one, as the ratios reflect the relationship between the single vector amplitude and the vector sum amplitude. Therefore, only the relative magnitude of the ratio gives an indication of the component dominance at each angle. The vibration contours for each blast and the relationships in Figure 6 were used to determine the predicted amplitudes of the radial and shear components of the blast waves at each mapped damage location. These results are listed in Table 2.

FIG 6 - Radial and shear components as a function of angle from the charge axis with a point of symmetry at 45°.

TABLE 2 Predicted radial and shear peak particle velocity components from predicted vector sum peak particle velocity and angle from charge axis to damage location (θcharge). Predicted VS PPV (mm/s)

θcharge (degrees)

PPV Radial ratio

Predicted PPVShear ratio

PPVShear ratio

Predicted PPV Shear (mm/s)

1

100

59

0.59

59

0.74

74

2

175

67

0.75

132

0.58

101

3

38

67

0.75

29

0.58

22

4

75

65

0.71

54

0.62

46

5

135

0

1.00

135

0.10

14

6

85

53

0.47

40

0.85

72

7

175

17

0.85

149

0.45

79

8

225

17

0.85

191

0.45

101

9

1000

24

0.73

725

0.57

570

10

62

149

0.57

35

0.72

45

11

38

142

0.44

17

0.86

33

Damage site

Angle of incidence and surface normal motion amplification The next step in the process of predicting the vibration amplitudes responsible for the damage events was addressing the amplitude amplification factors from wave reflection at the excavation boundary. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

171

K G FLEETWOOD AND E VILLAESCUSA

This process relied on the relationships between the amplitudes of incident P- and S-waves and the reflected amplitudes at various angles of incidence. Once these amplitude ratios were calculated from elastic wave reflection equations, surface velocity conditions could be assessed using conservation of motion relationships.

Effect of angle of incidence on surface particle velocities from wave reflection The basic relationship in elastic wave propagation in linear-elastic isotropic homogeneous materials under reflection of a normal incident P-wave would suggest a surface amplification factor of two (Kolsky, 1953). In reality, the waves produced by cylindrical charges that interact with existing excavations can have variable wave types (P- and S-waves) and varying angles of incidence. Any prediction of the surface excitation from measurement of body waves therefore becomes complicated for any situation outside of the normal incident P-wave. Additionally, the interpretation of surface measurements as representative of in situ vibration values from body waves is not a simple one to one relationship. To investigate the effect of angle of incidence and wave types on the surface excitation from reflected blast waves, some solutions for the amplification factor from reflected Pand S-waves have been proposed using fundamental equations of wave reflection at a zero-stress boundary (free surface). Fundamental Rayleigh wave conversions were not considered, as numeric solutions for conversion of body waves to Rayleigh waves are not well documented. Instead, shear waves at angles of incidence greater than the critical angle have been assumed to be converted entirely into Rayleigh waves and the surface amplitudes have been assumed to be the entire shear wave amplitude. The effects of phase changes, differing wave propagation velocities, or existing damage to the excavation on surface amplification have not been considered explicitly. Surface amplification factors researched by Jiang (1993) using numerical modelling has identified vertical (surface normal) amplification factor ranging from two to six with increasing angle of incidence due to limitation of waves to the shallow layer near the surface from Rayleigh wave conversion. Existing research on the interaction of seismic waves with excavations under jointed or damaged rock mass conditions have identified amplification factors up to 12 (Milev and Spottiswood, 2005).

Aggregate amplification factor based on angle of incidence Reflection of P- and S-waves at free boundaries based on the angles of incidence of each wave type has been investigated using equations listed by Achenbach (1973). The reference equations (Equations 1 to 5) apply to reflection of P- and vertically-polarised shear waves (SV-wave) in two dimensions. The shear waves described in the suggested model based on measured data were the vector sum shear wave resulting from the vertical and transverse components. The single vector amplitude representation of the two shear components located in the plane perpendicular to the direction of wave propagation was assumed to be valid in the reflection model for SV-waves. The amplitude ratios for the reflected P- and SV-waves from an incident P-wave are specified by Equations 1 and 2. The amplitude ratios for the reflected P- and SV-waves from an incident SV-wave is given by Equations 3 and 4. AP - P' = sin 2iPi sin 2iP - SV' - l2 cos2 2iP - SV' APi sin 2iPi sin2iP - SV' + l2 cos2 2iP - SV'

(1)

AP - SV' = 2l sin 2iPi cos 2iP - SV' APi sin 2iPi sin 2iP - SV' + l2 cos2 2iP - SV'

(2)

l sin 4iSVi ASV - P' = ASVi sin 2iSVi sin 2iSV - P' + l2 cos2 2iSVi

(3)

ASV - SV' = sin 2iSVi sin 2iSV - P' - l2 cos2 2iSVi ASVi sin 2iSVi sin 2iSV - P' + l2 cos2 2iSVi

(4)

and l=

2^1 - h ^1 - 2 h

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

(5) 172

BLAST VIBRATION MONITORING AND ELASTIC WAVE REFLECTION MODELS

where APi AP-P AP-SV ASVi ASV-P ASV-SV Pi P-SV SVi SV-P

amplitude of incident P-wave amplitude of reflected P-wave from incident P-wave amplitude of reflected SV waves from incident P-wave amplitude of incident SV wave amplitude of reflected P-wave from incident SV-wave amplitude of reflected SV-wave from incident SV-wave incident angle of incident P-wave reflection angle of reflected SV-waves from incident P-wave incident angle of incident SV-wave reflection angle of reflected P-wave from incident SV-wave material Poisson’s ratio The curves of incident and reflected amplitude ratios using Equations 1 to 5 were generated for the Poisson’s Ratio of 0.29, which was determined from rock testing for the mine site (AMC, 1998 WASM, 2009). Figure 7 shows the two incident wave situations and the associated reflected amplitudes versus the angle of incidence from the reflection surface normal. (A)

(B)

FIG 7 - (A) Ratio of reflected amplitude to incident amplitude for incident P-waves; and (B) incident S-waves.

The motions induced in the surface of the excavation could be estimated through conservation of motion at the point of reflection. The assumption of zero boundary losses was required for the case of conservation, which would be a specialised case for linear-elastic homogeneous material with a tractionless boundary. In reality, significant boundary effects would be expected due to dispersion from the irregular surface boundary or as plastic or permanent deformations were induced. Additionally, changes in the material properties from dynamic loading conditions and the existing damage from blasting or stress effects were not considered. These effects can significantly influence MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

173

K G FLEETWOOD AND E VILLAESCUSA

wave amplification from changes in material elastic moduli and wave propagation velocities as measured and modelled by previous researchers (eg Hildyard, 2001 Saiang, 2008). Consideration of all variables would necessitate high-order numerical modelling, which was outside the scope of the current work. The amplification factors for surface motions were estimated using the balance of particle velocity amplitudes for the incident wave (Ai) and the reflected P-wave (AP ) and shear wave (ASV ) the at a point on the excavation such that Ai

AP + ASV

(6)

At the time of reflection, the surface of the excavation would experience an aggregate amplitude of Ai - AP - ASV for each wave type. The negative amplitude values noted for the reflected amplitude ratios in Figure 7 represent a change of polarity. This change of polarity would allow numeric summation of the values without additional consideration of the sign of the amplitude. For example, an incident compressive longitudinal wave (P-wave) would reflect as a tensile wave of negative amplitude. Therefore, the aggregate motion at the surface would be in the same direction for the incident and reflected waves and the two amplitudes could be summed. The assumption was made that the detachment of blocks or spalled layers would be a result of the surface normal amplitude. The components of amplitude for each wave type were therefore resolved into the surface normal component using the angles of reflection. To simplify the calculations of surface normal amplitudes from actual vibration data, an aggregate amplification factor for each situation was calculated based on the relative incident amplitude of 1 mm/s. The actual amplitudes could be calculated by applying the surface normal amplification factors shown in Figure 8 to the predicted amplitudes listed in Table 2. This process required two steps due to the variations in P- and S-wave amplitude distributions with the angle from the blasthole axis at the assumed representative blast locations. One important note in the application of Equations 3 and 4 is identification of the critical angle at which the SV-waves no longer reflect. This angle is unique to the material Poisson’s Ratio, and any angle of incidence greater than the critical value would result in complete conversion of the shear waves into surface waves (Rayleigh or Love waves). In the model shown in Figure 8, the aggregate curve of amplification factors for the incident S-wave takes this angle into account. At all angles of incidence greater than the critical angle of 32.95 , where the reflected amplitudes asymptote to 0, no reflected P-wave was assumed to be present. Only the incident S-wave amplitude was then assumed to contribute to the surface normal amplitude as a Rayleigh Wave. The aggregate surface normal amplification factors were applied to the predicted component amplitudes displayed in Table 2 at each damage location. Each surface normal amplitude value and the aggregate surface normal amplitude at each damage location are listed in Table 3.

FIG 8 - Aggregate amplitude amplification factors for surface normal particle velocity. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

174

BLAST VIBRATION MONITORING AND ELASTIC WAVE REFLECTION MODELS

TABLE 3 Surface normal amplification factors and aggregate surface normal peak particle velocity at each damage location.

θ1 (°)

Predicted VS PPV (mm/s)

Surface normal amplification factor

Surface normal PPV Radial (mm/s)

Surface normal amplification factor

Surface normal PPV Shear (mm/s)

Aggregate surface normal amplitude (mm/s)

1

31

100

1.3

78

1.5

108

186

2

23

175

1.6

208

1.7

172

380

3

23

38

1.6

45

1.7

37

82

4

25

75

1.5

81

1.7

77

158

5

32

135

1.3

174

1.4

20

193

6

36

85

1.1

46

1.6

115

161

7

73

175

0.6

96

2.0

154

251

8

73

225

0.6

124

2.0

198

322

9

66

1000

0.6

438

1.9

1089

1527

Damage site

10

59

62

0.6

22

1.9

83

105

11

52

38

0.7

12

1.8

58

70

The ratios of the predicted aggregate surface normal amplitudes versus the predicted VSPPV of the body waves at each damage location varied between 1.4 and 2.2 and were highly influenced by both the angle of incidence with the excavation and the angle of the damage location with the blasthole axes. The values of predicted surface normal PPV associated with each damage event fell within the ranges published by previous researchers, and as such could be viewed as valid in producing some degree of damage under the surface support conditions. The actual values of surface normal PPVs related to damage events could have been several times larger than the predicted value due to Rayleigh wave activity or site effects as discussed by Cichowicz, Milev and Durrheim (2000), Hildyard (2001) and Milev and Spottiswood (2005). In absence of data to identify the additional location-specific amplification factors for the existing excavation conditions, the values from the wave reflection solution have provided a predictive tool. This tool has allowed determination of the ratio of the surface normal aggregate PPVs to the predicted VSPPVs of the incident body waves versus the angle of incidence at each damage location. This relationship is shown in Figure 9.

FIG 9 - Ratio of predicted surface normal amplitude versus predicted vector sum peak particle velocity of body wave. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

175

K G FLEETWOOD AND E VILLAESCUSA

PREDICTED VIBRATION AMPLITUDES AT LOCATIONS OF NO OBSERVED DAMAGE Main fault crossing The maximum surface normal PPVs were predicted for the backs and nearest sidewalls of the main fault crossing using the relationships developed above. The predicted surface normal amplitudes were 151 mm/s for the backs and 116 mm/s for the nearest sidewall. No damage was recorded near the fault crossing for these vibration conditions. The absence of observed damage has been attributed to the heavy surface support at the location, which included two layers of fibre-reinforced shotcrete and steel-weld mesh in addition to sprayed fibrecrete arches through the fault intersection zone (Figure 10).

FIG 10 - Photo of main access fault crossing heavy surface support with fibre-reinforced shotcrete and steel-weld mesh in addition to sprayed fibrecrete arches.

Return air raise The maximum VSPPV at the leading edge of the RAR nearest the stope was predicted using the results of near-field vibration monitoring along the transducer array installed on either side of the RAR. The resulting Scaled Distance prediction equation for vibrations along the orientation of the pillar containing the raise was PPVmax

0

D

-1

(7)

All blastholes for all firings were then evaluated against the PPVmax equation to predict the potential maximum particle velocities of incident waves interacting with the raise. The maximum predicted VSPPV was 3172 mm/s. According to traditional PPV-based damage criteria, this level of vibration would have been sufficient to cause significant damage through fresh fracture, slabbing or block fall (eg Holmberg and Persson, 1978 Forsyth, 1993 Rorke and Milev, 1999 Singh and Narendrula, 2004). No such damage was observed at any observation point within the raise. The circular shape of the raise and the virtually undamaged condition of the rock mass from raise boring excavation method were believed to account for the lack of damage.

OTHER WAVE EFFECTS ON SURFACE DISPLACEMENTS One factor that has not been explicitly considered in the model is the effect of the incident and surface wave frequencies, which would ultimately influence the wave numbers and therefore the displacement potentials upon reflection and transmission. Site effects observed by Cichowicz, Milev MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

176

BLAST VIBRATION MONITORING AND ELASTIC WAVE REFLECTION MODELS

and Durrheim (2000) identified a general relationship where lower wave frequencies reduced the amplification of measured surface velocities. This factor may be related to the depth of activity of the Rayleigh waves, which is commonly accepted to be bound to a layer near the surface of approximate thickness of 1.5 times the wavelength (Kolsky, 1953). Wavelength ( R) is determined by the propagation velocity (VR) and frequency (fR) of the Rayleigh wave such that R VR fR. Near the point of Rayleigh wave conversion, a higher source frequency would be expected to result in a lower wavelength and thus the energy in the wave would be bound by a layer of reduced thickness. Conservation of energy density or energy flux would therefore require greater surface displacements for the same energy input when compared with waves bound to a larger volume of material from a greater depth of activity. Additionally, the reduced velocity of the Rayleigh wave in comparison to the group velocity of the incident body waves would be expected to cause significant interaction between the time-function oscillations. This effect could possibly lead to increased peak displacements and particle velocities if constructive interference were to occur. Determination of wave interaction with excavations is further complicated when considering the dimensions and properties of the excavation surface and the effects of fatigue damage through repeated post-peak loading. The excavation shape and dimensions influence the resonant frequencies of the exposures, and the frequency of the propagating surface waves. During propagation, high frequencies would be expected to attenuate at a higher rate until the natural frequency of the Rayleigh wave for the material properties is reached. If this frequency nears the resonant frequency of the excavation wall, further surface displacements may be expected. Accumulated damage from repeated subcritical loading can also lead to failure events. It is therefore unlikely that the mapped damage events in the case study have been caused by a single vibration event of given amplitude.

DISCUSSION AND CONCLUSIONS Extraneous blast vibrations can cause damage to mine infrastructure in proximity to blasting events, given the amplitudes of the induced vibrations are sufficient to detach blocks or cause fresh damage to surfaces. In the presence of properly installed and functioning surface support, these damage events can be minimised or contained. Containment of damage through application of surface support can lead to fewer risks associated with falls of ground or loosening of blocks that may become unstable during future blasting or mining activities. The case study has attempted to predict the surface normal peak vibration amplitudes using nearfield blast vibration monitoring and the properties of reflecting elastic waves at zero stress boundaries. The method has identified a significant influence of the angle of incidence on the amplification factors for each wave type and has addressed the partitioning of body wave components (P- and S-waves) from a series of monitored blastholes. Some of the fundamental relationships between radial and shear components noted in mathematical solutions were supported. These results led to a model for approximation of the wave component amplitudes for the predicted VSPPV at the recorded damage locations. Blast vibration monitoring along multiple orientations allowed for prediction of VSPPV values from elliptical wave propagation patterns through integration of multiple attenuation equations. This approach is not common in traditional blast vibration monitoring, analysis and interpretation, even though field measurements have highlighted the behaviour. Therefore, integration of multiorientation monitoring results should be considered when attempting to predict the full-field vibrations from an explosive source. The interaction of blast-induced body waves with existing excavations is an extremely complex issue involving many factors. Some factors include x damage and stress conditions of the rock mass at the excavation boundary x the partitioning of wave types, energies, and frequencies of the incident waves x interaction of existing surface waves with reflecting body waves x the shape of the excavation x existing surface support and x other dispersion or amplification effects. No analytical or numerical model can account for all of these factors and as such there is no standard solution for the amplification factor when a body wave encounters an excavation boundary. The MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

177

K G FLEETWOOD AND E VILLAESCUSA

converse of this relationship identifies that there is no standard solution when attempting to relate vibrations measurements taken on the excavation surface to the waves produced near blastholes in the confined rock mass. Under these circumstances, the solutions proposed in this study may aid in extrapolating surface measurements to in situ confined rock mass vibrations by decomposing the surface amplified waves through damping factors.

ACKNOWLEDGEMENTS The authors would like to thank the BHP Billiton Cannington Mine engineering, charge-up and geotechnical departments for financial and logistical support over the four years of the study. The support of CRC Mining and the WASM Rock Mechanics Research Group is also gratefully acknowledged.

REFERENCES henba h, J D, 1973. Netherlands).

a e Propagation in Elastic Solids, 425 p (North-Holland Publishing Company

ustralian Mining Consultants, 1998. Cannington stress measurement 68 C, 450 mLv, report from Australian Mining Consultants to BHP World Minerals – Cannington Project, report 196066. Ci ho i , A, Milev, A M and Durrheim, R J, 2000. Rock mass behaviour under seismic loading in a deep mine environment Implications for stope support, Journal of the SAIMM, 100(2) 121-128. leet ood, K G, 2010. Near-field blast vibration monitoring and analysis for prediction of blast damage in sublevel open stoping, PhD thesis (unpublished), Curtin University, Western Australian School of Mines. leet ood, K G, Villaescusa, E, Li, J and Varden, R, 2009. Comparison of traditional near-field vibration prediction models with three-dimensional vibration scaling and blast wave energy, in Proceedings ineth International Symposium on Rock Fragmentation by Blasting, Fragblast , pp 579-588 (CRC Press Granada). orsyth, W W, 1993. A discussion of blast-induced overbreak around underground excavations, Rock Fragmentation by Blasting, pp 161-166 (Balkema Rotterdam). a ebroek, P, 1966. Elastic waves from a finite line source, in Proceedings Royal Society of London Series A, Mathematical and Physical Sciences, 294(1436) 38-65. eelan, P A, 1953. Radiation from a cylindrical source of finite length, Geophysics, 18(3) 685-696. ildyard, M W, 2001. Wave interaction with underground openings in fractured rock, PhD thesis (unpublished), University of Liverpool. olmberg, R and Persson, P A, 1978. The Swedish approach to contour blasting, in Proceedings Annual Conference on Explosi es and Blasting Techni ue, pp 113-126 (International Society of Explosives Engineers Cleveland). iang, J, 1993. Vibrations due to a buried explosive source, PhD thesis (unpublished), Curtin University, Western Australian School of Mines. olsky, H, 1953. Stress

a es in Solids, 211 p (Clarendon Press Great Britain).

Milev, A M and Spottiswood, S M, 2005. Strong ground motion and site response in deep South African mines, Journal of the SAIMM, 105 515-524. age, C H, 1987. Controlled blasting for underground mining, in Proceedings Annual Conference on Explosi es and Blasting Techni ue, pp 33-46 (International Society of Explosives Engineers Cleveland). orke, A J and Milev, A M, 1999. Near field vibration monitoring and associated rock damage, in Proceedings International Symposium on Rock Fragmentation by Blasting, Fragblast , pp 19-22 (SAIMM Johannesburg). Saiang, D, 2008. Behaviour of blast induced damaged zone around underground excavations in hard rock mass, PhD thesis (unpublished), Lule University of Technology. Singh, S P and Narendrula, R, 2004. Assessment and prediction of rock mass damage by blast vibrations, in Proceedings Mine Planning and E uipment Selection MPES , pp 317-322 (Taylor and Francis Group London). Western Australian School of Mines (WASM), 2009. Report on intact rock properties for BHP Billiton Cannington, Rock testing report submitted to BHP Billiton Cannington Mine, March. u, T R, 1980. Ground control at Kidd Creek, underground rock engineering, in Proceedings 1 th Canadian Rock Mechanics Symposium, pp 73-79 (CIM Toronto). MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

178

Role of Physico-Mechanical Properties in Cutting Performance of Diamond Wire Saw in Marble Quarrying Operation S C Jain1 and S S Rathore2 ABSTRACT Marble deposits of dolomitic variety are available in large quantities in India and for extracting these deposits diamond wire saw cutting machine is mainly used. Cutting performance of the diamond wire saw machine depends on rock parameters, machine parameters and operational parameters. Rock parameters related to physico-mechanical properties, such as, compressive strength, tensile strength, shear strength, modulus of elasticity and abrasion resistance of marble play role in determining cutting rate and beads wear rate. Thus, this study was undertaken to determine the relationship between different physico-mechanical properties and cutting performance. This work also aimed at enhancing the cutting performance of diamond wire saw in dolomitic marble deposit. Study was conducted in the field at three different zones having deposit of soft, medium hard and hard dolomitic marble. During the study, cutting rates and wearing of diamond beads were recorded in the field. Physico-mechanical properties were determined in the laboratory. The results obtained were subsequently evaluated by using statistical analysis. Study revealed that these properties of marble play a significant role in the performance of diamond wire saw machine. Predictive equations for the wearing of diamond beads and cutting rate were developed based on regression models. Results of this study will be of ample use for operators of wire saw to achieve maximum cutting performance and for manufacturers in designing the diamond beads.

INTRODUCTION Marble reserves in India are estimated at 2216 Mt with Rajasthan accounting for 51.6 per cent of the reserves ie 1144 Mt (Gupta and Singh, 2007). Out of 1144 Mt, about 800 Mt is of dolomitic variety. Dolomitic marble stones variety is mainly found in Rajsamand district of Rajasthan state. The marble stone industry in India has now graduated to the use of modern technology with widespread use of cutting technology such as diamond wire saw, diamond belt saw and chain saw cutter (Singh, 2003). Presently, diamond wire saw cutting is a widely used method in getting marble block production. During cutting action by diamond wire saw machine primarily involves the pulling of continuous loops of spinning diamond wire around the marble bench. Vertical cut is performed after flushing of two holes one into another at 90° angle and diamond wire threaded through these holes. The diamond wire is simply a steel cable on which diamond impregnated beads are mounted at a regular interval with the help of spring spacer placed between the beads for cutting marble. Wear of diamond beads during cutting operation is one of the main parameters in the economics of marble cutting. The harder rock gives more wear of the diamond beads and lower productivity of diamond wire saw machine. When the wear on diamond beads increases, cutting rate and diamond wire efficiency decrease and production cost increases. Univer (1996) had developed predictive equations for the estimation of specific wear and cutting force in rock sawing. Moreover, some researchers have attempted to investigate the effect of textural and mineralogical properties of rocks on wearing of diamond beads and cutting rate in diamond wire cutting method. The physico-mechanical properties of marble affect the sawability of stones and mainly depend on mineralogy and texture of rocks (Tugrul and Zarif, 1999). Wear on beads is affected by the engineering 1. Assistant Professor, Maharana Pratap University of Agriculture and Technology, Udaipur 313001, India. Email:[email protected] 2. Associate Professor, Maharana Pratap University of Agriculture and Technology Udaipur 313001, India. Email:[email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

179

S C JAIN AND S S RATHORE

properties like mechanical, textural, mineralogical and petrographical (Ozcelik, Kulaksiz and Cetin, 2000). Textural characteristics of rocks are most important factors in determining the mechanical behavior and prediction of performance of cutting. According to Prikryl (2001), the variations in mechanical strength of rocks obtained from laboratory tests can be explained by various factors, such as, mineralogical composition, density, porosity, texture, moisture content, sample size and test conditions. With impregnated tools, continuous and efficient cutting can only be facilitated by compatible wear of the diamond particles and their bonding matrix (Wright, Tagg and Davis, 2000; Wright and Engels, 2003). The absence of quartz in carbonated rock generally gives high cut productivity with lower wire wear. Stellin et al (2001) found a correlation between the cutting rate of the helical wire and the metallic wear of the wire. The most important subject in rock cutting with diamond wire is the determination of operating conditions at which the wear on diamond beads must be lowest and wire efficiency becomes highest. The main factors to be taken into account while researching cutting technology and optimisation of cutting process are tool wear, cutting forces and the support stability (Deliormanli and Pamukcu, 2001). The relationship among the texture coefficient and wear on diamond beads and cutting rate were investigated, and significant relationship between these values was found (Ozcelik et al, 2004). However, no detailed studies were carried out to determine the effect physico-mechanical properties on cutting performance of diamond wire saw machine. Thus, this study was conducted at three different zones of dolomitic marble: soft, medium-hard and hard at Morwad mines of M/s R K Marble Ltd located in Rajsamand district of Rajasthan State, India. Total of 12 vertical cuts were observed in selected area of deposit having soft, mediumhard and hard dolomitic marble. During the study, cutting rates and wearing of diamond beads were recorded in the field. Physico-mechanical properties were determined in the laboratory at Department of Mining Engineering, Maharana Pratap University of Agriculture and Technology, Udaipur, India. The results obtained were subsequently evaluated by using statistical analysis. This study has revealed the relationship between physico-mechanical properties and cutting performance of diamond wire saw in marble quarrying operations.

DESCRIPTION OF STUDY AREA The study area is situated near Morwad village at a distance of 15 km in western direction from district headquarter Rajsamand of Rajasthan State of India. This area is situated between latitude 24° 46’ to 26° 01’ N and Longitudes 73° 28’ to 74° 18’ E. Geology of the study area and marble extraction technique are described below.

Geology Morwad area exhibits a huge concentration of dolomitic marble of soft, medium-hard and hard variety. Reserves of dolomitic marble in study region are about 800 Mt of fine to coarse grained in nature (Jain, Rathore and Wahy, 2007). Grain size varies from few mm to 1.5 cm. Silica is also present, mainly in the form of quartz, which reduces the cutting performance of diamond wire saw as well as life of diamond beads. Other forms of silica are also present in the marble like tremolite and actionolite needles, which deteriorates the quality of marble and decreases the strength. The silica percentage in dolomitic marble varies from 3.5 per cent to 40 per cent. Marble formation also contains a number of amphibolite intrusive veins. The bands in the gneiss are marked by dark streaks and rich in ferromagnesian mineral, mainly biotite, and alternating with light coloured quartz feldspar layers. The hardness, on the basis of Moh’s scale of dolomitic marble, varies from 2.70 to 2.85. Soft dolomitic marble deposit has coarse grained texture of calcite with subhedral grains of quartz, augite, pyroxene and muscovite. Straight contacts between calcite grains are well visible and also isolated grains of sphene are present. The medium to coarse grained texture of calcite was observed in medium-hard dolomitic marble deposit. Here, small grains of quartz and fibrous tremolite are present as mineral impurities and quartz grains showing undolose extension with interlocking boundaries. The hard dolomitic marble deposit has fine to medium grained of calcite and high amount of quartz grains. The quartz grains are subangular with sutured contact showing undolose extension. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

180

ROLE OF PHYSICO-MECHANICAL PROPERTIES IN CUTTING PERFORMANCE OF DIAMOND WIRE SAW

Extraction method The mining of dolomitic marble at Morwad area is carried out by opencast mining by forming separate benches of poor quality marble with waste rock and blockable marbles. First, overburden of 8 m to 15 m is removed by forming benches using controlled blasting technique after providing a horizontal cut with diamond wire saw above the blockable marble. This cut separates the blockable marble from the overburden and protects the damages due to blasting vibrations. Subsequently, benches of 10 m height are formed and blockable marble bench of about 1000 to 2000 tonnes is separated by providing horizontal and vertical cuts with the help of diamond wire saw cutting machine (Figure 1), after flushing the holes of 115 mm in desired plane.

FIG 1 - Schematic diagram of horizontal and vertical cuts with diamond wire saw.

The big blocks so formed are toppled by using hydraulic jacks/ water bags/air bag, after providing the proper cushion on floor. After toppling the block from its position, the work of marking and sizing is carried out to ascertain the maximum recovery of saleable blocks of size 3 m × 1.5 m × 1.5 m. The crack free saleable blocks are identified and, thereafter, loaded into the truck with the help of derrick cranes/mobile crane. The recovery of saleable product in block form of marble varies from 30 per cent to 50 per cent.

DESIGN OF EXPERIMENTS The cutting performance of diamond wire saw machine depends on rock parameters, machine parameters and operational parameters as given in Table 1. TABLE 1 Parameters affects cutting performance of diamond wire saw machine in marble. Rock parameters

Machine parameters

Operational parameters

• Physico-mechanical properties

• Power

• Dimension of block (bench cut size)

• Mineralogical composition

• Peripheral speed

• Types of cut (machine position)

• Texture

• Thrust

• Operator’s skill

• Discontinuities

• Construction of diamond wire

• Quantity of water used

• Water content

• Beads structure

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

181

S C JAIN AND S S RATHORE

The purpose of this study was to determine the effect of physico-mechanical properties on cutting performance of diamond wire saw machine. First, different marble quarries of soft, medium-hard and hard dolomitic marble deposits were selected in Morwad area for field experiments. Design of experiment have both laboratory and field investigations. The laboratory investigations were planned to determine physico-mechanical properties of soft, medium-hard and hard dolomitic marble. Field investigations were designed for vertical cuts as these cuts are made in maximum number than other types of cuts in marble quarrying operation. We have planned four vertical cuts in each type of dolomitic marble : S-1, S-2, S-3 and S-4 cuts in soft; M-1, M-2, M-3 and M-4 cuts in medium-hard and H-1, H-2, H-3 and H-4 in hard. To measure cutting rate and beads wear rate, cutting area, cutting time and beads diameter were measured before and after completion of cut. Figure 2 shows the investigative path followed in this study.

FIG 2 - Flowchart of study.

LABORATORY AND FIELD INVESTIGATIONS The laboratory investigations were conducted to determine the various physico-mechanical properties of samples collected from the field. Field investigations were carried out to determine cutting and beads wear rates in soft, medium-hard and hard zones of dolomitic marble deposit. The details of both investigations are described in the next section.

Laboratory investigations Marble block cube of size 30 cm was taken as sample from field location for each cut for laboratory investigations. These samples were transported to the departmental laboratory. In the laboratory, samples were prepared according to International Society of Rock Mechanics standards norms (Brown, 1981) for determining different physico-mechanical properties with the assistance of core drilling and core cutting and grinding machine. Thereafter, physico-mechanical properties, such as, compressive strength, tensile strength, shear strength, modulus of elasticity, ultrasonic p-wave velocity and abrasion resistance were determined as per standard test procedure. Figure 3 shows tensile test with electronic universal testing machine (UTM). The compressive strength varies for soft, medium-hard and hard zone of dolomitic marble from 45.15 - 49.90 MPa, 51.23 - 64.62 MPa and 67.95 - 86.84 MPa, respectively. Similar trend was MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

182

ROLE OF PHYSICO-MECHANICAL PROPERTIES IN CUTTING PERFORMANCE OF DIAMOND WIRE SAW

FIG 3 - Tensile strength test with UTM (Brazilian method).

observed for other physico-mechanical properties except abrasion resistance. Abrasion resistance value percentage was decreasing from soft to hard dolomitic marble.

Field investigations Field experiments were conducted in the study area having quarries of soft, medium-hard and hard dolomitic marble deposits. Before commencing the field experiments for making vertical cuts, some of the machine parameters were fixed in order to achieve more consistent results (see Table 2). TABLE 2 Fixed parameters of the machine taken employed in the field study. Parameters No of beads/meter

Dimensions 33 (sintered type)

Power of machine: 1. Main motor (ac)

44.7 kw

2. Feed motor (dc)

0.75 kw

Diamond wire peripheral speed Voltage

27.2 m/s 440 V

Pulley diameter

800 mm

Pull back force (thrust)

930.14 N

Diamond beads

Sintered type (0.63 carat/bead)

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

183

S C JAIN AND S S RATHORE

During each experiment, observations of cutting area and cutting time were recorded. Diamond beads diameters were measured before and after completion of cut with digital caliper. Thereafter, cutting rates, in m2/h, and beads wear rate, in mm/m2, were determined and are shown in Table 3 corresponding to physic-mechanical properties of marble. TABLE 3 Physico-mechanical properties and cutting performance results of different cuts. Cut no

Uni compressive Tensile strength Shear strength Modulus of (strength (MPa) (MPa) (MPa) elasticity (GPa)

Abrasion resistance %

Cutting performance results Cutting rate (m2/h)

Beads wear rate (μm/m2)

S-1

45.15

4.38

20.15

5.40

34.02

5.41

1.03

S-2

49.43

4.59

22.40

9.93

32.33

5.32

1.04

S-3

49.70

4.83

22.54

10.87

31.34

5.27

1.05

S-4

49.90

4.85

22.67

11.03

29.53

5.07

1.06

M-5

51.23

4.99

23.73

12.21

22.92

4.95

1.12

M-6

54.06

5.35

24.80

13.36

21.71

4.77

1.14

M-7

56.41

5.58

25.41

14.08

19.34

4.74

1.16

M-8

64.62

5.85

26.54

14.60

18.52

4.57

1.20

H-9

67.95

6.35

28.61

15.60

13.41

4.55

1.26

H-10

73.52

6.84

29.74

17.81

12.52

4.27

1.28

H-11

83.24

6.97

30.10

20.22

11.76

4.16

1.30

H-12

86.84

7.64

30.75

23.44

10.66

4.13

1.32

The results obtained for cutting rates in soft zone, medium-zone and hard zone of dolomitic marble were between 5.07 - 5.41 m2/h, 4.57- 4.95 m2/h and 4.13 - 4.55 m2/h, respectively. These results indicate decreasing trend from soft to hard dolomitic marble whereas beads wear rate showed increasing trend from soft zone to hard zone.

EVALUATION OF RESULTS Results obtained for cutting performance of diamond wire saw machine were evaluated with respect to different dolomitic marble properties, such as, uniaxial compressive strength, tensile strength, shear strength, modulus of elasticity and abrasion resistance. In the first step of evaluation, the relationship between properties and cutting rate was investigated. In the next step, relationships between properties and diamond beads wear rate were investigated graphically.

Relationships between cutting rate and physico-mechanical properties Relationships between cutting rate and uniaxial compressive strength (UCS) are shown in Figure 4a. The trend of cutting rate observed was reduced as compressive strength increased from soft to hard 2 dolomitic marble. The regression line obtained, Y = -0.0301x + 6.6058; R = 0.8980, shows that cutting rate have significant relationships with compressive strength. The cutting rate decreased by 23.65 per cent as uniaxial compressive strength increased by 92.33 per cent. Figure 4b shows the relationship between tensile strength and cutting rate, showing similar trend of reduction in cutting rate with increased tensile strength of marble. The trend show reduction of cutting rate with increased shear strength from soft to hard marble, as shown in Figure 4c. The regression line y = -0.1234x + 5.5695 2 with a coefficient R = 0.9872 indicates the sound relationship between cutting rate and shear strength. Similarly, relationship of modulus of elasticity for soft to hard marble with cutting rate is shown in Figure 4d. The regression lines as shown in Table 4 indicate significant relationships between these properties and cutting rate. Figure 4e shows the relationship between the cutting rate and abrasion resistance of marble. The trend shows an increase in cutting rate with an increase in 2 abrasion resistance percentage and its regression line y = 0.0511 + 3.6685 with R = 0.9585 indicates significant linear relationship. As the abrasive resistance percentage decreased from soft to hard dolomitic marble from 34.02 per cent to 10.66 per cent, the corresponding cutting rate was reduced from 5.41 m2/h to 4.13 m2/h. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

184

ROLE OF PHYSICO-MECHANICAL PROPERTIES IN CUTTING PERFORMANCE OF DIAMOND WIRE SAW

y = -0.0301x + 6.6058 R2 = 0.898

6 5 4 3 2 20

30

40 50

60

70

80 90 100

Cutting rate (m2/h)

(B)

Cutting rate (m2/h)

(A)

5 4 3 2 3

Uniaxial com pressive strength (MPa)

5

7

9

Tensile strength (MPa)

y = -0.1234x + 5.5695 R2 = 0.9872

6 4 2 0 0

5

10

15

Cutting rate (m2/h)

(D) Cutting rate (m2/h)

(C)

y = -0.4118x + 7.1085 R2 = 0.939

6

y = -0.0971x + 6.1552 R2 = 0.9158

6 5 4 3 2 0

10

20

30

Modulas of elasticity (GPa)

Shear strength (MPa)

Cutting rate (m2/h)

(E) 6 4 y = 0.0511x + 3.6685 R2 = 0.9585

2 0 0

10

20

30

40

Abrasion resistance

FIG 4 - Relationships between physic-mechanical properties of and cutting rate.

Relationships between diamond beads wear rate and physico-mechanical properties Figure 5a shows the relationships between diamond beads wear rate and the uniaxial compressive strength (UCS). Wear rate observed was increasing with an increase in compressive strength from 2 soft to hard marble stones and the regression line Y1 = 7E-06x + 0.0007; R = 0.9190, indicates significant relationships. The diamond beads wear rates increased by 28.15 per cent as the uniaxial compressive strength increased by 92.33 per cent. The significant relationship was observed between tensile strength and wear rate, as shown in Figure 5b. The similar trend was observed for the wear rate versus marble tensile strength. Figure 5c shows that the wear rate was increasing as shear 2 strength increased from soft to hard marble. The regression line Y1 = 3E-05x + 0.0001; R = 0.9754 indicates significant relationship between wear rate and shear strength. The relationship of modulus of elasticity versus soft to hard marble with diamond beads wear rate is shown in Figure 5d. The significant relationships observed by regression models are depicted in Table 4 between these properties and wear rate. Figure 5e shows the relationships between wear rates and the marble abrasion resistance. The trend shows reduction in wear rate with an increase 2 in abrasion resistance percentage with the regression line Y1 = -1E- 0 5x + 0.0014; R = 0.9725 indicating significant linear relationship. It was observed that shear strength and abrasion resistance properties were closely correlated with wear rate. The regression models shown in Table 4 may be applied to other marble stones based on their physico-mechanical properties. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

185

S C JAIN AND S S RATHORE

(B) 0.0014 0.0012 0.001

y = 7E-06x + 0.0007 R2 = 0.919

0.0008 0.0006

20

30

40

50

60

70

80

wear rate (mm/m2)

wear rate (mm/m2)

(A)

y = 1E-04x + 0.0006 R2 = 0.9597

0.0014 0.0012 0.001 0.0008 0.0006 3

90 100

4

5

6

7

8

Tensile strength (MPa)

Uniaxial com pressive strength (MPa)

0.0014 0.0012 0.001 0.0008 0.0006 0.0004 0.0002 0

y = 3E-05x + 0.001 R2 = 0.9754

0

5

10

wear rate (mm/m2)

(D)

wear rate (mm/m2)

(C)

y = 2E-05x + 0.0008 R2 = 0.9049

0.0015 0.001 0.0005 0 0

15

5

10

15

20

25

Modulas of elasticity (GPa)

Shear strength (MPa)

wear rate (mm/m2)

(E) y = -1E-05x + 0.0014 R2 = 0.9725

0.0014 0.0012 0.001 0.0008 0.0006 0.0004 0

10

20

30

40

Abrasion resistance

FIG 5 - Relationships between properties of dolomitic marble and diamond beads wear rate. TABLE 4 Regression models between cutting rate (Y), wear rate (Y1) and physico- mechanical properties. Serial no 1

Physico-Mechanical Proporties (x)

Regression models for cutting rate (Y)

Regression models for wear rate (Y1)

Compressive strength

Y = -0.0301x + 6.6058; R2 = 0.8980

Y1 = 7E-06x + 0.0007; R2 = 0.9190

2

2

Tensile strength

Y = -0.4118x + 7.1085; R = 0.9390

Y1 = 1E-04x + 0.0006; R2 = 0.9597

3

Shear strength

Y = -0.1234x + 5.5695; R2 = 0.9872

Y1 = 3E-05x + 0.0001; R2 = 0.9754

4

Modulus of elasticity

Y = -0.0971x + 6.1552; R2 = 0.9158

Y1 = 2E-05x + 0.0008; R2 = 0.9049

5

Abrasion resistance

Y = -0.0511x + 3.6685; R2 = 0.9585

Y1 = -1E-05x + 0.0014; R2 = 0.9725

Relationship between cutting rate and diamond beads wear rate It was observed from regression analysis that the cutting rate and diamond beads wear rate also have 2 significant linear relationship (R = 0.9661). Figure 6 show that cutting rate was reduced as diamond beads wear rate increased from soft to hard marble. As diamond beads wear rates increased by 28.15 per cent, cutting rate decreased by 23.65 per cent.

Multivariate regression analysis (MVRA) The complexity of most scientific mechanisms is such that in order to be able to predict an important response, a well known multiple regression models is needed. For MVRA, we have considered MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

186

Wear rate (mm/m2)

ROLE OF PHYSICO-MECHANICAL PROPERTIES IN CUTTING PERFORMANCE OF DIAMOND WIRE SAW

y = -0.0002x + 0.0023 R2 = 0.9661

0.0016 0.0014 0.0012 0.001 0.0008 0.0006 0.0004 0.0002 0 0

2

4

6

8

Cutting rate (m 2/h)

FIG 6 - Relationship between cutting rate and diamond beads wear rate.

physico-mechanical properties of marble as independent variables, ie compressive strength (X1), tensile strength (X2), shear strength (X3), Modulus of elasticity (X4) and abrasion resistance (X5) for each dependent variable, ie cutting rate (Y1) and wear rate (Y2). On the basis of observations given in Tables 3, regression models were derived using SYSTAT 8.0 program for vertical cutting and are given in Table 5. The factor coefficients show the effect of parameters on cutting rate and diamond beads wearing. TABLE 5 Regression models derived for cutting rate and diamond beads wear rate. S No

Models

Standard error

Explanation

1

Y1 = 4.187 - 0.010X1 + 0.023X2 + 0.019X3 - 0.017X4 + 0.038X5

0.085

0.98

2

Y2 = 1.056 + 0.003X1 + 0.021X2 + 0.003X3 - 0.006X4 - 0.008X5

0.003

0.99

2

The R value of 0.98 for cutting rate indicates that all five independent variables put together accounted for about 98 per cent variation of the cutting rate of diamond wire saw machine and, 2 similarly, R value of 0.99 for wear rate indicates that independent variables contributed 99 per cent variation in diamond wire wear rate. The statistical validity of any regression model can be tested by variance analysis method. However, different approaches can also be used for this purpose. One of these methods can be used to demonstrate the differences between measures and predicted values with scattered graphs. Figures 7a and 7b show the significant correlation between the results obtained from field measurement and those predicted from the linear regression model for cutting rate and wear rate, respectively. These models can be used to predict the cutting rate and wear rate for various marble and can also be used to select appropriate diamond wire saw machine.

CONCLUSIONS In this study, the relationships of various physico-mechanical properties, such as, uniaxial compressive strength, tensile strength, shear strength, modulus of elasticity and abrasion resistance with diamond beads wear rate and cutting rate were investigated. As a result, the following conclusions were drawn: x The regression analysis between physico-mechanical properties and cutting rate of diamond wire saw indicates significant relationships. Shear strength has more correlation with cutting rate. It was observed that cutting rate decreased as the values of physico-mechanical properties increased except for abrasion resistance from soft to hard dolomitic marble. x Diamond beads wear rate have significant relationships with physico-mechanical properties of dolomitic marble. The wear rate was increased as the value of physico-mechanical properties MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

187

S C JAIN AND S S RATHORE

(A)

(B)

FIG 7 - Validity of the models for (A) cutting rate; and (B) wear rate.

increased from soft to hard dolomitic marble, except for abrasion resistance, which has shown a reverse trend. Beads consumption per m2 cutting in soft, medium-hard and hard dolomitic marble was on average 0.00105 mm, 0.00116 mm and 0.00129 mm, respectively. x A significant relationship was observed between cutting rate and beads wear rate. Diamond beads wear rates increased by 28.15 per cent as cutting rate were reduced by 23.65 per cent from soft to hard dolomitic marble. x An accurate relationship and prediction of wear rate and cutting rate with physico-mechanical properties will be helpful to diamond beads manufactures in planning of appropriate design of diamond beads, which will increase the cutting rate and reduce the wear rate in marble. Mine operators will also benefit from selecting appropriate diamond beads according to physicomechanical properties of marble for optimum production at minimum cost.

ACKNOWLEDGEMENT The authors are very grateful to the management of M/s R K Marble (P) Ltd, Udaipur, Rajasthan, India for permitting to carry out field research study at their mines. The authors are also thankful to the International Marble Institute, Napoli, Italy for sponsoring this project and providing all support in carrying out the study.

REFERENCES Brown, E T, 1981. Rock Characterisation, Testing and Monitoring: ISRM Suggested Methods, 211 p (Pergamon Press: Oxford). MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

188

ROLE OF PHYSICO-MECHANICAL PROPERTIES IN CUTTING PERFORMANCE OF DIAMOND WIRE SAW

Deliormanli, A H and Pamukcu, C, 2001. Optimization of stone sawing operation, in Proceedings Tenth Mine Planning and Equipment Selection Conference, pp 215-221, New Delhi. Gupta, R K and Singh, K, 2007. Indian stone industry – An insight, ‘Stonedge’ Journal of Indian Dimensional Stone Industry, February, pp 18-22. Jain, S C, Rathore, S S and Wahy, S N, 2007. Performance of new innovative cutting machines in dolomitic marble mining, Journal of Rock Mechanics and Tunneling Technology, 13(1):29-39. Ozcelik, Y, Kulaksiz, S and Cetin, M C, 2000. Assessment of the wear of diamond beads in the cutting of different rock types by the ridge regression, Journal of Materials Processing Technology, 127:392-400. Ozcelik, Y, Ploat, E, Bayram F and Ay, A M, 2004. Investigation of the effect of textural properties on marble cutting with diamond wire, Int Journal Rock Mechanics Mining Sciences, 41(3):1-7 Prikryl, R, 2001. Some microstructural aspets of strength variations in rocks, Int Journal of Rock Mechanics and Mining Sciences, 38:671-682. Singh, V S, 2003. Mathematical modeling of wire cutting technology, in Proceedings National Seminar on Recent Development in Machinery and Equipment for Dimensional Stone Mining, pp 139-146 (Maharana Pratap Agriculture University: Udaipur). Stellin, A, Hennies, W T, Soares, L and Fujimura, F, 2001. Dimensional stone block extraction by steel wire, in Proceedings Tenth International Symposium on Mine Planning and Equipment Selection, pp 215-221, New Delhi. Tugrul, A and Zarif, I H, 1999. Correlation of mineralogical and textural characteristics with engineering properties of selected granitic rocks from Turkey, Engineering Geology, 51:03-317. Univer, B, 1996. A statistical method for practical assessment of sawability of rocks, in Proceedings International Symposium ‘Eurock 96’, pp 59-65 (Balkema: Rotterdam). Wright, D N and Engels, J A, 2003. The environmental and cost benefits of using diamond wire for quarrying and processing of natural stone, Industrial Diamond Review, Lamda Publicity Ltd, England, 4/03, pp 16-24. Wright, D N, Tagg W R J and Davis, P R, 2000. The development of a rock classification system for use with diamond tools, in Proceeding Conference Stonemart, Jaipur, India, pp 109-128.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

189

Never Touch a Running System? Longwall Cutting Sequences and their Potential K Nienhaus1 and S Hetzel2 ABSTRACT The optimisation and validation of shearer cutting methods is a time consuming engineering task. Many different parameters, of which not all can be controlled, can vary. The easiest solution, to increase the installed power and the longwall length, is not always the best. A different cutting method can increase the capacity without the need of new machines. The Department for Mining and Metallurgy Machinery at RWTH Aachen University therefore developed ProSL, the simulation tool for shearer and longwall operations. ProSL enables the engineer to create his own cutting methods and compare it with other methods. Machine-data can easily be varied to see the impact on the method. The research shows that sophisticated longwall methods increase the productivity of the system. The paper presents the results of the calculation with ProSL and shows how higher efficiency and lower costs can be achieved.

INTRODUCTION The question which cutting sequence is used for a longwall is a complex and time consuming task. Most mines are using their preferred sequence for a long time and do not try to change it. In other cases the geological conditions force the mine to use a particular method (Bahr, 2005). The standard for improving output is to increase the cutting and haulage power from shearer and AFC and to increase the length of the longwalls. This will increase the total output and probably the output per man-hour, but it does not ask for the most efficient way to produce coal. To take the load of the planning engineers, the Institute for Mining and Metallurgy Machinery developed the ‘Production Simulation for Shearer and Longwall Operations’ (ProSL) (Nienhaus and Lange, 2005a). A recent overview of the Australian longwalls shows (Table 1), that most of the mines (65 per cent) are still using the bi-directional cutting sequence (bi-di). The uni-directional cutting sequence (Uni-di) is applied at 42 per cent and variable-web methods are only applied at ten per cent of the mines. According to Rutherford (Bahr, 2005) the longwall length should be over 230 m for bi-di cutting. Lange (Nienhaus and Lange, 2005b) shows that in the range between 200 m and 400 m bi-di does not always is the most efficient cutting method. Bi-di wins efficiency with length and installed cutting motor power. A direct comparison between the three commonly used methods is done (Table 1). TABLE 1 Usage of cutting methods at Australian mines†. Cutting sequence

†

Mines

Percentage (%)

Uni-di

13

42

Bi-di

20

65

Variable-web

3

10

Created from Longwalls.com, 2010.

1. Full Professor, Mining and Metallurgy Machinery (IMR), RWTH Aachen University, 2 Wuellnerstrasse, Aachen 52062, Germany. Email: [email protected] 2. Research Engineer, Mining and Metallurgy Machinery (IMR), RWTH Aachen University, 2 Wuellnerstrasse, Aachen 52062, Germany. Email: [email protected] MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

191

K NIENHAUS AND S HETZEL

CUTTING SEQUENCES The cutting sequences are divided into five different groups by Peng (Peng and Chiang, 1984): 1. Number of cutting passes per cycle: the number of cutting passes per cycle determines the method. Cutting sequences with one cutting cycle per pass are referred to as uni-directional. Sequences with two cutting cycles per pass are referred to as bi-directional. 2. Sump into face: there are two different methods generally used to sump into the face, either a box cut or a wedged cut. The bi-di sequence uses two box cuts, one at each to gate, while the variableweb sequence uses a wedge cut. The sumping operation has a high influence on the unproductive times of a sequence. 3. Cutting depth: in the so called variable-web sequences the coal is cut with only a part of the drums depth. The remaining coal to the full depth of the drum is cut on the return trip. 4. Cutting height: depending on the seam height and geological conditions the cutting height per cycle can be varied. Methods utilising two cuts to cut the whole seam height are called partial opening sequences. According to the mentioned common cutting sequences in Australia, this paper will focus on the bidirectional (Figure 1), the uni-directional (Figure 2) and the variable-web cutting method (Figure 3). The variable-web will be realised with a ratio of 60/40 for the web on the uphill/downhill run.

FIG 1 - Bi-directional cutting sequence.

FIG 2 - Uni-directional cutting sequence.

FIG 3 - Variable-web cutting sequence.

Bi-directional cutting sequence The bi-directional cutting sequence is characterised by its two cutting passes per cycle, one on the uphill- and one on the downhill run. Two full cutting passes also means that this sequence advances faster than the other mentioned sequences with two full web depths per recycle. The shearer sumps into the face with a box cut at each of the gates (Figure 1). The box cuts are the least productive time of the sequence and reduce the average production capacity due to the slow speed. As a reaction to that the length of longwalls using bi-di is increasing, to lower the percentage of face end operation time on the total cycle time. The leading drum is always cutting with full height while the trailing drum cuts the remaining height. In case that the total height of the seam is higher than the height of both drums together, MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

192

NEVER TOUCH A RUNNING SYSTEM? LONGWALL CUTTING SEQUENCES AND THEIR POTENTIAL

uni-di sequences are used instead, especially if the conveyor capacity is not enough. During the main cutting cycle both drums are cutting at full depth.

Uni-directional cutting sequence The uni-directional cutting sequence uses only one cutting pass per cycle and advances with one web per cycle. For this cutting run the sumping takes place at the face end (Figure 2). The second run is used for cleaning or floor control if problems with weak floor are encountered. During the main cut both drums are at full web and the leading drum is cutting at full height. The change between cutting and cleaning is also affecting the AFC, which is loaded unsteadily.

Variable-web cutting sequence The variable-web cutting sequence utilises two cutting passes per cycle and advances at one full web per cycle. The reduced web is partially compensated by the possibility to use higher shearer speeds for cutting. While the drums are cutting with full height, the cutting depth is varying between the runs. The so called ‘Half-Web’ sequence uses a 50/50 pattern for the uphill and the downhill run. Commonly also 60/40 and 70/30 sequences are used. The benefit of using variable-web cutting sequences comes from the high utilisation of the shearer. Face end operations are held short and the shearer speed is high due to the fact that it is only cutting at full depth at the very end of the face (Figure 3).

CAPACITY CALCULATIONS The capacity calculations start with the creation of the base data set. This case is then modified to show an automated longwall with a higher availability. In the third case the manual speed restriction is replaced by the restriction of the sequence. For all cases the same configuration of shearer and AFC are used. The selected Shearer is an Eickhoff SL 300 with 480 kW cutter motors. More detailed information on shearer and AFC configuration can be found in Table 2. TABLE 2 Information on used shearer and AFC. Shearer

SL 300

Power haulage motors (kW)

Power cutting motors (kW)

Machine’s weight (t)

Drum diameter (m)

Clearance area (m²)

AFC area (m²)

Power AFC motor (kW)

AFC speed (m/s)

2 x 80

2 x 480

60

1.8

0.44

0.67

3 x 800

1.8

Like mentioned in chapter 1, the three relevant cutting sequences are the bi-di-, the uni-di- and the variable-web cutting sequence. The uni-di sequence is used with a backward snake at the main gate.

Geological setup The main geological parameters for this paper are the following: x A varying face length of 150 m and 450 m. x A seam height of 2.4 m. x The coal density is 1.36 t/m³ with 0.3 kWh/m³ required specific cutting energy and 130 kN/m² specific cutting forces. x A maximum web of 1 m. x The whole longwall has got an inclination of 5 t. x The used AFC is a PF4 from Bucyrus with a width of 1332 mm. The geology is assumed to be stable and there is no restriction for a cutting sequence due to it.

Possible restrictions During the cutting process the shearer speed can be restricted by various different factors. The face end operations are mostly uncritical because the speed is generally lower than for production runs. On the production run the shearer speed can be restricted by: MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

193

K NIENHAUS AND S HETZEL

x x x x x x

haulage motor power: restrict the advancing speed of the shearer, cutting motor power: restricts the advance rate because of coal hardness, conveyor motor power: restricts the transportable load on the AFC, conveyor cross section: restricts the loadable amount of coal, shearer clearance area: restricts the loadable amount of coal, and stage loader capacity: restricts the loadable amount of coal. To increase the production these restricting factors have to be considered.

The base case In the base case the shearer is controlled manually. The maximum speed for the shearer is set to 13 m/min to enable the driver to walk beside the machine. The average speed during the face end operations is set to 7 m/min. Due to the forced speed limit, both the uni-di and the variable-web method cannot use their advantage against the bi-di sequence (Figure 4). The uni-di sequence is producing the same amount of coal on the uphill run as the bi-di, but it only has got one production run instead of two. The shearer sumps in only once per pass, so the total loss against the bi-di sequence is 27 per cent.

FIG 4 - Capacities for the different methods.

Variable-Web benefits from the quick sump in and the low amount of unproductive times. Forced to use the same speed but with a lower cutting depth it losses against the bi-di sequence a full web per pass. The high amount of unproductive time of the bi-di sequence on the other hand is reducing the loss to 24 per cent. With these settings it is best to use the bi-directional sequence. Increasing the longwall length to reduce the percentage of unproductive time works well in this case. An increase of 200 m to a total length of 500 m will improve the capacity by 24 per cent for bi-di. Uni-di can only achieve a ten per cent increase and the Variable-Web sequence increases by nine per cent. All results for this and the following case can be found in Figure 4.

Improved availability The second case simulates the first step of integrating an automation system. A miner will still walk with shearer in case anything unexpected happens and so the maximum moving speed is still limited to 13 m/min. The automation system is increasing the total longwall availability from 55 per cent to 65 per cent. It is expected that this increase will improve the capacity also by ten per cent. Instead the capacity is raised by 18.2 per cent. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

194

NEVER TOUCH A RUNNING SYSTEM? LONGWALL CUTTING SEQUENCES AND THEIR POTENTIAL

FIG 5 - Production capacities with sequence restrictions.

Even when compared to the lower availability, uni-di and Variable-Web cannot increase the production capacity over the capacity of bi-di with less availability. An overview about the capacities for both cases, the lower and the improved availability, can be found in Figure 4.

Higher shearer speeds At the time the shearer is automated, it is possible to let it operate at higher speeds as if an operator is involved. In this case the shearer uses the speed limitation that comes from the sequential prerequisites. A manual speed limitation of 23 m/min is used for safety reasons. Face end operation speed is increased to 10 m/min in average. The new resulting speed limitations for the sequences are as follows: x bi-di: as well on the uphill run as on the downhill run the speed is limited by the cutter motor. In both cases at least one drum is cutting its full diameter and full web, which results in a high cutting resistance. x uni-di: the limitation for the cutting run is the same as for bi-di, since this part of the sequence is equal. The flitting run on the other hand is only restricted by the maximum possible speed of the shearer or, if the loading takes place by the loading capacity of the drums. x variable-web: on the uphill run, where 70 per cent of the web is cut, the speed limitation result from the cutter drive, but is significantly higher than the limitation for full web cutting. On the downhill run the conveyor would probably restrict the shearer speed, but in this case the manual limit is lower. Without the speed limit the variable-web cutting sequence can claim to be the most productive of the three cutting methods (Figure 5). Its high speed and the short face end times create a capacity that is 6.5 per cent higher than the bi-di for this longwall. The uni-di sequence can also improve, but due to its lack of a second production run it cannot compete against the bi-di. Production capacity is increased by 86 per cent, but is still 17 per cent behind bi-di. The production capacity increase shows a direct proportional dependency of the cutting speeds of the shearer. The increases can be found in Table 3. Bi-di has the lowest increase in production, but profits the most when comparing the production increase to the average speed increase. The increase in production is twice the increase in speed. For MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

195

K NIENHAUS AND S HETZEL

TABLE 3 Speed and production increase. Method

Increase uphill run

Increase downhill run

Increase in production

Increase to base Bi-Di

Bi-di

30%

30%

63%

93%

Uni-di

30%

130%

86%

60%

Variable-web

118%

130%

128%

105%

uni-di and Variable-Web the increase in speed and the increase in production are proportional. The remaining difference in production increase is due to the increased face end operation speed.

CONCLUSIONS The selection of the cutting method is a task that has to incorporate more than the pure production capacity. To be able to react to changes in geology appropriately everyone working with the shearer should know the different possible cutting sequences. The flexibility ‘is one of the keys to get most out of a cutting sequence’ (Bahr, 2005). Without spending money it is possible to increase the production capacity. A change of the cutting sequence or modifications to the used sequence can already improve the performance. In a next step the question should be where the bottlenecks are, before just increasing the size of shearer and longwall. The comparison shows that the production capacity could be nearly doubled without changing many parameters. For bi-di an increase of 93 per cent is possible by increasing the speed of the shearer. With a change of the cutting sequence the capacity can be increased by another 12 per cent. If necessary the base case could also be switched to an uni-di sequence that still increases the production by 60 per cent.

REFERENCES Bahr, A, 2005. Cutting edge, Australian Longwall Magazine, June, pp 13-16. Longwalls.com, 2010. International Longwall News [online]. Available from: [Accessed: July 2010]. Nienhaus, K and Lange, U, 2005a. Big is not always beautiful, American Longwall Magazine, May, pp 21-23. Nienhaus, K and Lange, U, 2005b. Can David compete against Goliath, in Proceedings Mine Planning and Equipment Selection 2005, pp 935-950 (Banff). Peng, S and Chiang, H S, 1984. Coal extraction – The shearer, Longwall Mining, pp 335-399 (Wiley: New York).

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

196

Developing Algorithms and a Computer Program to Predict the Peak Particle Velocity After Blasting in Open Pit Mines and Quarries İ Topal1, B Elevli2 and K Erarslan3 ABSTRACT Drill blasting is a widely used method for rock fragmentation in mining, quarrying and construction industries. Environmental effects such as air shocks, dust, flying rocks and ground vibration are unavoidable results and undesired effects of blasting. Among these ground vibrations can have a damaging effect and may influence on long distances. Therefore various researchers have developed mathematical models to predict ground vibration after blasting. These models take into account the amount of explosive, distance and several coefficients related to the field under investigation. The developed models to predict ground vibration may yield different results for the same case. Selection of the most suitable model requires engineering judgment and evaluation of models. At this point, a computerised system helps the engineer to evaluate and select the most suitable model in order to design and plan a blasting pattern. The safe charge of explosive and magnitude of vibration after blasting should be predicted beforehand. Therefore a computer program namely ‘prediction of blast induced vibration’ (P-BIV) has been developed. This paper introduces and presents a computer program (P-BIV) which includes the set of models developed to estimate ground vibration to provide blasting engineers a tool to determine the amount of explosive for safe blasting.

INTRODUCTION Blasting is still widely used method for fragmentation of rocks in open-pit mines as well as in civil construction works. Engineers continuously investigate drill blasting patterns for each and every geological and geomechanical condition to obtain desired fragmentation size. However, environmental effects of blasting are also needed to be investigated. The undesired environmental effects are dust, air-shock, flying rocks and ground vibrations (Kecajevic and Radomsky, 2005). Among these effects, ground vibrations can have a damaging effect and may influence long distances. This situation sometimes causes mining companies to indemnify damages occurring at surrounding residential areas (Salzano and Cozzani, 2006). Ground vibrations is directly related to the amount of explosive used and distance between blasting site and the point of interest as well as geological and geotechnical conditions of the rock. Therefore, various researchers have developed mathematical models to predict ground vibration after blasting (Lu, 2005 Morin and Ficarozzo, 2006 Ambraseys, 1968 Nicholls, Johnson and Duvall, 1971 Langefors and Kihlstrom, 1973 Indian Standards Institute, 1973 Birch and Chaffer, 1983 Attewell, Farmer and Halsam, 1965 Davies, Farmer and Attewel, 1964 Ghosh, 1983 Gupta, Roy and Singh, 1987 Gupta et al, 1987 Gupta, Roy and Singh, 1988 CMRS, 1990 Arpaz, 2000). These models takes into account amount of explosive, distance and several coefficients related to the field under investigations. Therefore every blasting site should be considered as a particular case. The developed models to predict ground vibration may yield different results for the same case. The selections of the most suitable model require engineering judgment and evaluation of models. At this point, a computerised system helps engineer to evaluate and select the most suitable models in order to design and plan a blasting pattern. This paper introduces and present a computer program 1. Assistant Professor, Dumlupinar University, Department of Mining Engineering, Kütahya, Turkey, Email: [email protected] 2. Professor, Ondokuzmayıs University, Department of Industrial Engineering, Samsun, Turkey. Email: [email protected] 3. Professor, Dumlupinar University, Department of Mining Engineering, Kütahya, Turkey. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

197

I TOPAL, B ELEVLI AND K ERARSLAN

(P-BIV) which includes the set of models developed to estimate ground vibration to provide blasting engineers a tool to determine the amount of explosive for safe blasting.

ESTIMATION OF VIBRATION OCCURS AFTER BLASTING Research has carried out different studies and equations have been derived for predicting and estimating the vibration of blasting. These equations are unexceptionally function of distance, amount of explosives and geological parameters (Roy 2004 Sing and Roy, 1993). The relation between distance and amount of explosive is stated as scaled distance (SD) and given as follows (Konya and Walter, 1990). SD

R/W1/m

(1)

where SD is scaled distance R is distance in metre from blasting point to point of interest (metre) W is the maximum amount of explosive (kg) m is constant (mostly two or three) Ground vibration is known as peak particle velocity and widely accepted empirical relation between scaled distance and peak particle velocity is given below (Siskind et al, 1980) PPV

Ks

(SD)-

(2)

where PPV is the peak particle velocity (mm/s), Ks and  are the site constant which can be determined by multiple regression analysis (Barlow, 2000), by using recorded data of about 30 blasting. Based on Equation 1 and 2, different researcher have been developed various prediction equation to estimate PPV. These equations are given in Table 1. TABLE 1 Vibration prediction models. No

Researcher

Equations

No

Researcher

Equations PPV = K s c 3 R m .e- a.R

b

b

1

Ambraseys and Hendron (1968)

PPV = Ks c 3 R m W

7

Ghosh ve Daemen (1983) (II)

2

Nicholls, Johnson ve Duvall (1971)

b PPV = K s c R m W

8

Gupta, Roy and Singh (1987) (I)

3

Langefors ve Kihlstrom (1973)

W b 3 o R2

9

Gupta et al (1987) (I)

PPV = K s e W 2 o .e- a.R R

4

Indian Standarts Institute (1973)

PPV = K s e W 3 o R

10

Gupta, Roy and Singh (1988)

b R PPV = K s c R m .e - a.W W

5

Birch (1983) Attewel (1965) Davies (1964)

PPV = Ks .Ra .W b

11

CMRS (1990), Roy (1991)

-1 PPV = N + K s c R m W

6

Ghosh ve Daemen (1983) (I)

PPV = K s e

W

PPV = K s e

W b .e- a.R 3 o R2 3 b

2 b

b PPV = K s c R m .e- a.R W

The equations given in Table 1 were derived from the results of different studies carried out in different locations and formations. Although all equations seem to differ from each other, their basic logic is the same. They use scaled distance term with different representation. It is not fair to say they are inadequate or incorrect. They are just different. However, the question is that which equation should be used in order to predict ground vibration. In order to compare these equations, fast and easy way is needed. Therefore a computer program has been developed. This program determines site constant by utilising regression analysis methods based on the recorded blasting data. Then, by using these site constants and widely accepted 11 equations, the program will estimate expected ground vibrations for each blast. By doing so, all of the equations will be compared and the best suitable equations for the blasting site will be decided. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

198

DEVELOPING ALGORITHMS AND A COMPUTER PROGRAM TO PREDICT THE PEAK PARTICLE VELOCITY AFTER BLASTING

The 11 equations were re-organised in order to develop proposed program. The five formulas were transferred to dependant and one dependent regression model six of the formulas were transferred to dependent and two independent regression models since they have three unborn variables. The re-organisations are given in following sections.

RE-ORGANISATION OF THE EQUATIONS In order to utilise equations given in Table 1, , , Ks, and n site coefficients should be determined. In the developed system, in order to simplify the equations, several parameters have been substituted by an artificial variable and exponential statements are turned into linear form by logarithmic conversions. In order to do this, scaled distance SD of each equation has been substituted with the term of SDi as given below SD1 = c 3 R m, SD2 = c R m, SD3 = e

3 R2

o, SD4 = e

2 3

R

o, SD5 = c R m, SD6 = c

R

m

(3)

After substituting SDi terms in the equations, logarithmic conversion was applied to equations and results are given in Table 2. Then, the converted equations are used in linear regression analysis. Finally, the equations are converted to regression models in the form of one dependent-one independent and one dependent-two independent variables. TABLE 2 The results of logarithmic conversation. No

Equation b

2

PPV = Ks ^ SD1h => In PPV = InK s + b.InSD1 PPV = Ks ^ SD2hb => In PPV = InKs + b.InSD2

3

PPV = Ks ^ SD3hb => In PPV = InK s + b.InSD3

4

PPV = Ks ^ SD4hb => In PPV = InK s + b.InSD 4

5

PPV = Ks .Rb .Wa => In PPV = InK s + b.InR + a.InW PPV = Ks ^ SD2h- b .eaR => In PPV = InKs - b.InSD2 + aR

1

6 7

PPV = Ks ^ SD1h- b .eaR => In PPV = InKs - b.InSD1 + aR

8

PPV = Ks ^ SD3h- b .eaR => In PPV = InKs - b.InSD3 + aR

9

PPV = Ks ^ SD4h- b .eaR => In PPV = InKs - b.InSD 4 + aR

10

PPV = Ks ^ SD2h- b .eaSD5 => In PPV = InK s - b.InSD2 + aSD5 PPV = n + Ks ^ SD6h => PPV = n + Ks ^SD6h

11

REGRESSION AND CORRELATION ANALYSIS Equations 1, 2, 3, 4 and 11 are one dependent-one independent variable type regression model, while the Equations 5, 6, 7, 8, 9, and 10 are one dependent-two independent type regression model. Computations for the Equation 1 and the Equation 6 are given as example below. Computation of Equation 1 as example The equation InPPV InK1 + 1 ln(SD1) is equivalent to regression relation of Y a + b X. When is replaced with InPPV, a is replaced with InK1, b is replaced with 1 and is replaced with InSD1. Hence, the site coefficient can be calculated as given in (Spiegel, 1996), where 1 and InK1 are obtained from the normal equations n n / lnPPVi = ^lnK1h.n + b1 / lnSDli (4) i=1 i=1 n n n / lnSDli .lnPPVi = lnK1 / lnSDli + b1 /^lnSDlih2 (5) i=1 I i=1 i=1 which yield

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

199

I TOPAL, B ELEVLI AND K ERARSLAN

n b1 =

n

n n lnSDli .lnPPVi - f / lnSDli p. f / lnPPVi p i=1 i=1 i=1 2 n n n / ^ lnSDlih2 - f / lnSDli p i=1 i=1

(6)

/

n

n

i 1

i 1

n

n

i 1

i 1

2

e / lnPPVi o . e / ^ lnSDlih o - e / lnSDli o . e / lnSDli .lnPPVi o

lnK1 =

(7)

n/ ^ lnSDlih2 - ^/ lnSDlih2

K1 = elnK 1

(8)

Correlation coefficient for the Equation 1 is as follows (Konuk and nder, 1999) n

n

/ lnSDli . / lnPPVi

n

/ lnSDli lnPPVi - i

1

i 1

n

i 1

R1

R n n 2 VR 2V S lnSDli o WS lnPPVi o W e e S n WS n W S WS lnPPVi2 - i 1 W lnSDli2 - i 1 n n Si 1 WSi 1 W T XT X

/

/

(9)

/

/

Computation of Equation 6 as example When Y a + b X1 + c X regression model is adapted to equation of ln PPV where,

lnK6, B

lnPPV, a

6, C

6,

1

ln SD2,

R

2

lnK -  lnSD + R (10)

Hereby, new sixth model equations are n

/^InPPVih

i 1 n

/ ^InPPVih^InSD2ih

i 1 n

/ ^InPPVih^Rih

i 1

n

n

i 1

i 1

n.InK6 + b6 / ^InSD2ih + a6 /^Rih

(11)

n

n

n

i 1

i 1

i 1

InK6 /^InSD2ih + b6 / ^InSD2ih 2 + a6 /^InSD2ih^Rih n

n

n

i 1

i 1

i 1

InK6 /^Rih + b6 / ^InSD2ih^Rih + a6 /^Rih2

(12) (13)

6, 61, 62 and 63 matrixes are

V R n n W Sn ^ lnSD2ih ^ Rih W S 1 1 i i W S n n W S n 2 ^ lnSD2ih ^ lnSD2ih^ Rih W D6 S ^ lnSD2ih i 1 i 1 W Si 1 W S n n n W S ^ Rih ^ lnSD2ih^ Rih ^ Rih2 W Si 1 i 1 i 1 X TR V n n n W S ^ lnPPVih ^ lnSD2ih ^ Rih W Si 1 1 1 i i W S n n n W S ^ lnSD2ih2 ^ lnSD2ih^ Rih W D61 S ^ lnPPVih^ lnSD2ih i 1 i 1 W Si 1 W S n n n 2 W S ^ lnPPVih^ Rih ^ lnSD2ih^ Rih ^ Rih W Si 1 i 1 i 1 X TR V n n W S ^ lnPPVih ^ Rih n W Si 1 i 1 W S n n W S n ^ lnSD2ih ^ lnSD2ih^ Rih W D62 S ^ lnPPVih^ lnSD2ih i 1 i 1 W Si 1 W S n n n W S ^ lnPPVih^ Rih ^ Rih ^ Rih2 W Si 1 i 1 i 1 X T

/

/

/

/

/

/

/

/

/

/

/

/

/

/

/

/

/

/

(14)

(15)

/

/

/

/

/

/

/

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

(16)

200

DEVELOPING ALGORITHMS AND A COMPUTER PROGRAM TO PREDICT THE PEAK PARTICLE VELOCITY AFTER BLASTING

D63

Rn S ^ lnPPVih Si 1 S S n S ^ lnPPVih^ lnSD2ih Si 1 S n S ^ lnPPVih^ Rih Si 1 T

V W W i 1 W n W ^ lnSD2ih2 W i 1 W W n ^ lnSD2ih^ Rih W W i 1 X n

/

n

/^lnSD2ih

/

/ ^lnSD2ih

n

/

i 1 n

/

/^ Rih

(17)

/

i 1

hence D6

n

n

2

2

n

n

n

i 1

i 1

i 1

n. / ^InSD2ih : /^Rih + /^InSD2ih : /^InSD2ih^Rih : /^Rih + i 1

i 1

n

n

n

i 1

i 1

n

n

n

2

/^Rih : /^InSD2ih : /^InSD2ih^Rih - /^Rih :/ ^InSD2ih : /^Rih i 1

i 1

n

n

n

i 1

i 1

i 1

i 1

(18)

i 1

2 n

n

/^InSD2ih^Rih : /^InSD2ih^Rih.n - /^Rih :/ ^InSD2ih : /^InSD2ih i 1

i 1

Below equalities are handled after solving the matrices by Gauss method D61

n

n

n

n

n

i 1

i 1

i 1

/^InPPVih : / ^InSD2ih2 : /^Rih2 + /^InPPVih^InSD2ih : /^InSD2ih^Rih

i 1

n

n

i 1

i 1

i 1

n

n

n

: /^Rih + /^InPPVih^Rih : /^InSD2ih : /^InSD2ih^Rih - /^Rih : i 1

i 1

i 1

n

n

n

n

i 1

i 1

i 1

i 1

(19)

/^InSD2ih2 : /^InPPVih^Rih - /^InSD2ih^Rih : /^InSD2ih^Rih : n

n

n

i 1

i 1

i 1

/^InPPVih - R^Rih2 : /^InSD2ih : /^InPPVih^InSD2ih

D62

n

n

n

n

n

i 1

i 1

i 1

i 1

i 1

/^InPPVih : / ^InSD2ih : /^Rih2 + /^InPPVih^InSD2ih : /^Rih : n

n

i 1

i 1

n

n

n

i 1

i 1

/^Rih + /^InPPVih^Rih : n : /^lnSD2ih^Rih - /^Rih : /^InSD2ih : i 1

n

n

n

n

n

i 1

i 1

i 1

i 1

i 1

(20)

/^InPPVih^Rih - /^InSD2ih^Rih : /^Rih : /^InPPVih - /^Rih2 : n : n

/^InPPVih^InSD2ih

i 1

D63

n

n

n

i 1

i 1

i 1

n

n

/^InPPVih : / ^InSD2ih : /^InSD2ih^Rih + /^InPPVih^InSD2ih : /^Rih : n

n

i 1

i 1

i 1

n

i 1

n

n

/^InSD2ih + /^InPPVih^Rih : n : /^InSD2ih2 - /^InSD2ih : /^InSD2ih : i 1

i 1

i 1

n

n

n

n

n

i 1

i 1

i 1

i 1

i 1

/^InPPVih^Rih - /^InSD2ih2 : /^Rih : /^InPPVih - /^InSD2ih^Rih : n :

(21)

n

/^InPPVih^InSD2ih

i 1

K6, 6, 6 field coefficients can be derived from these equations

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

201

I TOPAL, B ELEVLI AND K ERARSLAN

InK6

D61 , D6

e InK , a6

K6

D63 , D6

(22)

D62 D6

b6

Equations 2, 3, 4 and 11 can be solved by the same manner. Correlation coefficient for the Equation 6 is (Spiegel and Stephens, 2000) R6

2 r6212 + r623 - 2r612 r613 r623 2 1 - r623

(23)

where n

n

n

n / InPPVi : InSD2i - e / InPPVi o : e / InSD2i o r612

i 1

n

n

i 1

i 1

i 1 2

i 1

n

n

(24)

2

2 >n / InPPVi - e / InPPVi o H : >n / InSD i - e / InSD2i o H 2

i 1

n

n

i 1

n

n / InPPVi : Ri - e / InPPVi o : e / Ri o r613

i 1

n

i 1

i 1

2

n

n

n

i 1

i 1

(25)

2

2 >n / InPPVi - e / InPPVi o H : >n / Ri - e / Ri o H 2

i 1

i 1

n

n

n

n / In2i : Ri - e / InSD2i o : e / Ri o

r623

i 1

i 1 2 n n 2 2 >n InSD2i - e InSD2i o H : >n Ri - e Ri o H i 1 i 1 i 1 i 1 n

/

i 1

n

/

2

/

/

(26)

Equations 5, 7, 8, 9, and 10 can be solved as Equation 6. The developed program carries out regression and correlation analysis for each equation. The result of these analysis provides site constant for each equation. The algorithm of program is given in the following sections.

THE PROGRAM FOR PREDICTION OF BLAST INDUCED VIBRATION (P-BIV) P-BIV (prediction of blast induced vibration) is the computer program developed by using Borland C++ Builder 5. The algorithm of the software is based on the derived equations presented in section 3. Main function of the program is to predict ground vibration at any distance to explosion centre and estimate the required explosive amount for defined distances. In order to use the formulas, field coefficients should be known. P-BIV has a utility to estimate the coefficient. Thereafter, blast induced vibration is predicted by referring to these coefficients. From programming point of view, P-BIV has three major units running under main menu (Figure 1). These are ‘field record unit’, ‘ground-vibration prediction unit’ and ‘calculation of safe amount explosive unit’. The flowchart of the program is given in Figure 2.



FIG 1 - Main menu of P-BIV. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

202

DEVELOPING ALGORITHMS AND A COMPUTER PROGRAM TO PREDICT THE PEAK PARTICLE VELOCITY AFTER BLASTING



FIG 2 - General flow chart of P-BIV.

Initially, a database is formed in field record unit including distance, explosive amount and recorded PPV (for each blast) (Figure 3). Once the field measurements are recorded, regression and correlation analysis can also be performed in this unit to determine field coefficients.



FIG 3 - Database recording, regression and correlation analysis unit. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

203

I TOPAL, B ELEVLI AND K ERARSLAN

Blast induced vibration prediction unit enables blasting engineer to examine PPV at any distance by using 11 equations (Figure 4). In order to find PPV and regression analysis parameters R and S, the system uses the previously explained approaches. Furthermore, the system can also give a report for probable damage by regarding the approaches of Edwards and Northwood (1959) and Langefors, Khilstr m and Westerberg (1948).



FIG 4 - Unit for prediction of blast induced vibration.

The third unit is developed to estimate the amount of explosive for safe blasting. Here, dependent parameters are vibration and distance to explosion centre. After determination of calculation method, PPV and distance, maximum weight of explosives, the system calculates optimum amount of explosive (Figure 5). The developed computerised system provides a practical and comprehensive tool for blast designers.

 FIG 5 - Estimation of explosive amount for an expected peak particle velocity and distance. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

204

DEVELOPING ALGORITHMS AND A COMPUTER PROGRAM TO PREDICT THE PEAK PARTICLE VELOCITY AFTER BLASTING

CONCLUSIONS In this study, a computer system (P-BIV) for blast induced vibration prediction has been developed. The main utilities of P-BIV are database generation, regression and correlation analysis to determine site constants, vibration prediction and estimation of explosive amount for particular peak particle velocity. It involves a set of 11 equations for vibration prediction used worldwide. Because geomechanical conditions are varying from field to field, the vibration approaches yield different results. The system enables the user to consider and compare all approaches and decide to use one of them quickly. Additionally, P-BIV can also estimate the amount of explosives that will yield the predetermined PPV at a particular distance. With its substantiality and features, the system presents a practical and comprehensive evaluation and examination tool for blast designers.

REFERENCES mbraseys, N R and Hendron, A J, 1968. Dynamic behavior of rock masses, Rock Mechanics in Engineering Practice, pp 203-207 (John Wiley and Sons London). rpa , E, 2000. Monitoring and evaluation of vibrations arising from some of the open pit blasting Turkey, PhD thesis, Cumhuriyet University, Sivas, Turkey. tte ell, P B, Farmer, I W and Halsam, D, 1965. Prediction of ground vibration parameters from major quarry blasts, Mining and Minerals Engineering, December, pp 621-626. Barlo , S J, 2000. A geotechnical evaluation and statistical investigation of blasting regulations in Malden Massachusetts, Boston College dissertations and theses, paper AAI1398191, Master Science (Boston College Massachusetts). Bir h, W J and Chaffer, R, 1983. Prediction of ground vibration from blasting on opencast sites, Trans Inst Min Metal, A102-A107 (Sect A, Min Ind). CM S, 1990. International project report no EC/43/90, September, 33 p. Davies, B, Farmer, L W and Attewel, P B, 1964. Ground vibration from shallow subsurface blasts, Engineer, (217) 553-559. d ards, A T and Northwood, T N, 1959. Experimental blasting studies, Engineer, 210 538-546. Ghosh, A and Daemen, J K, 1983. A simple new blast vibration predictor (based on wave propagation laws), in Proceedings th US Symposium on Rock Mechanics, pp 151-161 (University of Arizona). Gupta, R N, Roy, P P, Bagchi, A and Singh, B, 1987. Dynamic effects in various rock mass and their predictions, Journal of Mines, Metals and Fuels, pp 455-462. Gupta, R N, Roy, P P and Singh, B, 1987. Internal Project rep rtst EC 97685, MT/6/84, AC/19/84, EC/49/84, MT/PF/27/83,AC/55/83, EC/28/83, EC/61/83, MT/PF/15/84, Blasting Department, CMRS. Gupta, R N, Roy, P P and Singh, B, 1988. On a blast induced vibration predictor for efficient blasting Safety in mines research, in Proceedings nd International Conference of Safety in Mines Research Institutes , pp 1015-1021 (Center Mining Research Station Dhambad). ndian Standards nstitute, 1973. Criteria for safety and design of structures subjected to underground blast, ISI Bulletin, (IS-6922). e a evi , V and Radomsky, M, 2005. Flyrock phenomena and area security in blasting-related accidents, Safety Science, 43 739-750 (Elsevier Pennsylvania). onuk, A and nder, S, 1999. Mining Statistics, Osmangazi University, Faculty of Engineering, Mining Engineering Department (Eski ehir/Turkey). onya, C J and Walter, E J, 1990. Surface Blast Design, 534 p (Prentice Hall New Jersey). Langefors, U and Kihlstr m, B, 1973. Rock Blasting, 405 p (Wiley New ork). Langefors, U, Khilstr m, B and Westerberg, H, 1948. Ground Vibrations in Blasting (Water Power). Lu, , 2005, Underground blast induced ground shock and its modelling using artificial neural network, Computer and Geotechnics, 32 164-178 (Elsevier Singapore). Morin, M A and Ficarozzo, F, 2006. Monte Carlo simulation as a tool to predict blasting fragmentation based on the Kuz-Ram model, Computers and Geosciences, 32 352-359 (Elsevier). Ni holls, H R, Johnson C F and Duvall, W I, 1971. Blasting vibrations and their effects on structures, US Bureau Mines Bulletin, 656 105. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

205

I TOPAL, B ELEVLI AND K ERARSLAN

oy, P P, 2004. Vibration control in an opencast mine based on improved blast vibration predictors, Mining Science and Technology, 12 157-165 (Elsevier). Sal ano, E and Cozzani, V, 2006. A fuzzy analysis to estimate loss intensity following blast wave interaction with process equipment, Journal of Loss Pre ention in the Process Industries, 19 343-352 (Elsevier). Sing, B and Roy, P P, 1993. Blasting in Ground Exca ations and Mines (Balkema Rotterdam). Siskind, M S, Stagg, J W, Kopp, J W and Dowding, C H, 1980. Structure response and damage produced by ground vibration from surface mine blasting, Bureau of Mines Report of In estigations, RI 0 Spiegel, R M, 1996. Statistics Second Edition , ISBN 0-07-060234-4 (McGraw-Hill). Spiegel, R M and Stephens, L J, 2000. Statistic

ith Problems and Theory (McGraw-Hill Nebraska).

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

206

Numerical Modelling of Monorail Support Requirements in Decline Development B Besa1, M Kuruppu2 and E K Chanda3 ABSTRACT This paper discusses support requirements for the proposed monorail system to be used in decline development. The monorail drilling and loading systems are systems that move on the rail (monorail) installed in the roof of the decline and supported by roof bolts, suspension chains and steel supports. However, due to the weight of the components of the two systems, it is imperative that the force in each roof bolt, suspension chain and steel support capable of suspending the weight of the heaviest component is determined. Numerical models that relate the weight of the monorail drilling and loading components to the required strength in the support system have been developed. Using these developed models, numerical values of the forces in each roof bolt, suspension chain and steel support, required to suspend the weight of the heaviest component of the monorail drilling and loading systems are determined.

INTRODUCTION The monorail drilling and loading systems are systems that move on the rail (monorail) installed in the roof of the decline and supported by roof bolts, suspension chains and steel supports (Figure 1). The monorail consists of a track of jointed section rails, which can easily be extended to the desired length. Monorails are made of an I-profile rail, which completely prevents any derailment of the train. The monorail train, together with containers or carriages, hangs by its wheels on the bottom flange of the track. The train is powered by electric motors. Depending on the transportation task, the monorail system can be equipped with manriding cabins, material container and bottom discharge hoppers (Guse and Weibezhn, 1997). With a load carrying capacity of up to 30 tonnes and the ability to negotiate gradients of up to 36°, the monorail system can make transport in decline development considerably more efficient than conventional truck haulage system. Considering the requirements of typical underground mines, the monorail system is designed to negotiate horizontal and vertical curves with a minimum radius of 4 m and 10 m respectively. The system also shares many of the advantages of floor mounted rail, but overcomes the bulk of that system’s limitations (Scharf, 2007; Chanda and Besa, 2008). Other advantages include: x reduction in size of excavations leading to improved stability of underground excavations; x small excavations also means reduced ventilation and need for air conditioning; x reduced haulage costs per tonne per kilometre because of less power consumption; x less fire hazards compared to truck haulage system; x environmentally friendly technology – no diesel fumes; x multi-purpose haulage system for men, material and rock; x small and medium sized orebodies can be mined with less initial capital investment; x the system has potential for automation – reduction in personnel; and x low system operating costs – a must for narrow vein high grade orebodies, hence improved profitability for such ore deposits. 1. Research Fellow, Curtin University of Technology, Mining Engineering and Surveying Department, Western Australian School of Mines, Locked Bag 22, Kalgoorlie WA 6433.

Email: [email protected] 2. MAusIMM, Senior Lecturer, Curtin University of Technology, Mining Engineering and Surveying Department, Western Australian School of Mines, Locked Bag 22, Kalgoorlie WA 6433.

Email: [email protected] 3. MAusIMM, Associate Professor and Program Leader in Mining Engineering, The University of Adelaide, School of Civil and Environmental and Mining Engineering, Adelaide SA 5005.

Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

209

B BESA, M KURUPPU AND E K CHANDA

Considering the above and many more advantages that the monorail system offers, Chanda and Besa (2008; 2009) conceptualised the drilling and loading systems based on the monorail technology. The conceptual monorail drilling and loading systems will be used in decline development for face drilling and cleaning, respectively. Figure 2 shows the conceptual monorail drilling and loading systems in a decline.

FIG 2 - Conceptual monorail drilling and loading system (Chanda and Besa, 2009).

The monorail train is fitted with two independent drilling booms that are used in decline development. The system has its own power supply attached to it. It has also two horizontal and two vertical hydraulic stabilisers to act as supports during drilling operations. The loading system consists of an incline suction pipe that is connected to the hopper. The high pressure pump/fan connected to a storage hopper creates negative pressure inside the hopper that enables transport

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

210

NUMERICAL MODELLING OF MONORAIL SUPPORT REQUIREMENTS IN DECLINE DEVELOPMENT

of blasted rock fragments from the development face into the hopper to take place (Chanda and Besa, 2009). The monorail loading system will also serve as a means of ore and waste transport system from underground to surface. Men and material will also be transported using the system by, respectively, connecting man riding and material carriers to the system. According to manufacturers of the monorail train, Scharf, the system has a self-weight of 92 tonnes and carries up to six containers with total payload up to 30 tonnes including the weight of the container. The aim of this paper is to determine the minimum required strength of the roof bolt, suspension chain and steel supports for suspending and supporting the monorail drilling and loading systems components during operations. This is in order to avoid failure of the two systems from the support systems as well as to overcome dynamic forces. The methodology employed to achieve this objective is to develop numerical models that relate the weight of the monorail drilling and loading systems components to the required strength in each roof bolt, suspension chain and steel support. Using the developed models, numerical values of the minimum required strength in each roof bolt, suspension chain and steel supports to suspend and support the components of the two systems are determined.

NUMERICAL MODELLING In this section, numerical models that relate the weight of monorail system components with required support system at various sections of the decline (ie in inclines as well as at vertical and horizontal curves) are presented.

Equilibrium of forces in the loading system The force required in each roof bolt and suspension chain to support and suspend the monorail drilling and loading system components in an incline is significant in ensuring the components of the two systems remain suspended under load. To avoid failure of roof bolts and/or suspension chains due to the weight of monorail system components, high strength roof bolts and suspension chains must be installed. It is, therefore, important that the minimum required force in each support system necessary to suspend the weight of monorail drilling and loading system components is determined. In this section, models that determine the required force in each roof bolt and suspension chain in an incline based on the heaviest monorail drilling and loading system components are established.

Weight of monorail loading system components versus required support system The relationship between the weight of monorail loading system components and required force in each roof bolt and suspension chain in an incline is shown in Figure 3. Roof bolts

Suspension Chain

Powerpackwith driveunits

Lpart Lpart

FMS Lpart A



FC

Fpart F Cos part

 

Driver's cabin Z

Fpart

Loaded containers

Fpart F Cos part



FpartCos

Y

Lpart is length of monorail loading system component Fpart is weight of monorail loading system component Rs is roof bolt spacing FMS is force in each roof bolt (forces suspending monorail system) FC is force in each suspension chain  is decline gradient

FIG 3 - Forces in roof bolts and suspension chains for the monorail loading system components in an incline.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

211

B BESA, M KURUPPU AND E K CHANDA

Taking equilibrium of forces in Z-direction at point A, the following equation that relates the force in each roof bolt and suspension chain is established. FMS = FCCos

(1)

However, the monorail loading system component remains in equilibrium (in Z direction) if the total upward force (ie total forces in suspension chains installed within length Lpart occupying the system component is equal to the total downward force (ie weight of the heaviest monorail loading system component). In these calculations, the weight of the rail, chains and bolts is neglected.

Total force in suspension chains within Lpart The total force in suspension chains depends on the number of suspension chains installed within the span Lpart and the force in each chain. Since the roof bolt spacing (Rs) is known (which is also equal to suspension chain spacing), the number of suspension chains installed within Lpart is determined as follows: Total number of suspension chains with L part = e

L part o Rs

(2)

Thus, the total force in suspension chains within the span Lpart is the product of the total number of suspension chains installed and the force in each chain (FC) as indicated below: Total force in suspension chains within Lpart = FC e

Lpart o Rs

Z component of the total force in suspension chains = FC e

(3) Lpart o Cos a Rs

(4)

Weight of monorail loading system component In determining the weight of the heaviest monorail loading system component (Fpart) of length Lpart, the weight of the driver’s cabin, loaded containers and the power pack (with drive units) and their respective lengths were considered in this analysis. Figure 4 shows the monorail loading system components with respective lengths and weights. ͷͳ

ʹǤ͸

Ͷ

͵Ǥͷ

ƒ„‹Ȁ–”ƒ‹ ȋͳͲȌ

’‘™‡”’ƒ… ȋ͵ͺȌ

‘– ȋͷͲȌ

ʹǤ͸

ƒ„‹Ȁ–”ƒ‹ ȋͳͲȌ Fcont is weight of loaded monorail container Fpower pack is weight of one power pack (with drive unit) Fcabin/train is weight of driver’s cabin or train

FIG 4 - Schematic diagram showing lengths and weights of monorail loading system components.

As shown in Figure 4, the monorail loading system consists of components of different lengths and weights (Scharf, 2007). The heaviest component is the loaded monorail container, which has a total weight of 50 kN. Therefore, the length (Lpart) and weight (Fpart) of the heaviest component of the monorail loading system used in the analysis is 3.5 m and 50 kN, respectively. The weight of the heaviest component of the monorail loading system is: Weight of heaviest component of monorail loading system = Fpart

(5)

Z component of the heaviest monorail loading system = Fpart Cos

(6)

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

212

NUMERICAL MODELLING OF MONORAIL SUPPORT REQUIREMENTS IN DECLINE DEVELOPMENT

Required strength of suspension chains As shown in Figures 3 and 4, the heaviest monorail loading system component remains in equilibrium (in Z-direction) if its weight and total force in the suspension chains are equal. Therefore, the relationship between the weight of the heaviest monorail loading system component and the required force in each suspension chain just before failure is determined by equating Equations 4 and 6 to yield: FC = Rs Fpart Lpart

(7)

However, for the heaviest monorail loading system component to remain in equilibrium requires that the total force in the suspension chains within Lpart be equal to the total weight of that component. Since the allowable load in suspension chains just before failure is known (from Equation 7), the strength of the suspension chain should be more than the allowable load. However, the classical approach used in designing engineering structures is to increase the capacity (ultimate load) of the system in comparison with the allowable load. It should also be noted that suspension chain failure occurs if the weight of the heaviest monorail loading system component (allowable load) is more than the capacity (ultimate load) of the chains within Lpart occupying the heaviest component. Since the allowable force in each suspension chain is the same as the required force just before failure, a factor of safety is applied to increase the loading capacity of the chains. In this study, a factor of safety of 2.0 is assumed. Therefore, applying a factor of safety to Equation 7 yields: FC, max = 2Rs Fpart Lpart

(8)

Since Rs and Lpart are constants, the required strength of suspension chains, therefore, depends on the weight of the loaded monorail containers. Alternatively, Rs can be determined if the strength of the suspension chain is known.

Required strength of roof bolts The required strength of installed roof bolts within the span occupying the heaviest monorail loading system component Lpart is determined using the relationship in Equation 1. Therefore, substituting Equation 8 into Equation 1 gives the required strength in each roof bolt as: FMS, max = 2Rs Fpart Cosa Lpart

(9)

Equilibrium of forces for the drilling system The minimum strength required in each roof bolt and suspension chain to suspend the monorail drilling system components depends on the weight of the drilling system components (ie the weight of monorail train together with the two drilling booms and the weight of power pack with drive units). Figure 5 is used to determine the required strength in each roof bolt and suspension chains based on the weight of monorail drilling system components. It should be noted that this analysis is limited to the case of drilling system in transit. The system is further supported when drilling a face.

Total forces in suspension chains within Ldpart An analysis similar to that shown in the previous section yields: Z component of total force suspension chain FC e

Ldpart o Cos a Rs

(10)

Weight of monorail drilling system component Figure 6, is used to determine the weight of the heaviest component to be used in the analysis. Figure 6 shows the monorail drilling system consisting of two components, ie the driver’s cabin with two drilling booms and the power pack each with different length and weight. Thus, to determine the strength of the roof bolts and suspension chains the heaviest component of the drilling system is MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

213

B BESA, M KURUPPU AND E K CHANDA

Roof bolts

Suspension Chain

FMS A FC

Ldpart Fdpart

Drilling booms

FdpartCos 

Z Y dpart is length of monorail drilling s stem component Fdpart is weight of monorail drilling s stem component s is roof bolt spacing F is force in each roof bolts (forces suspending monorail s stem) F is force in each suspension chain  is decline gradient

FIG 5 - Forces in roof bolts and suspension chains for the monorail drilling system components in an incline.

used in the analysis. Figure 6 shows that the heaviest component is the driver’s cabin together with the two drilling booms which has a weight of 56 kN (ie weight of driver’s cabin is 10 kN and the two drilling booms were assumed to weigh 46 kN (Chanda, Besa and Kuruppu, 2008). Therefore, the length (Ldpart) and the weight (Fdpart) of the heaviest monorail drilling system component used in the analysis are 2.6 m (suspended length only) and 56 kN, respectively. The heaviest drilling system component in Z direction is written as: Z component of the heaviest drilling system component = FdpartCos

ʹǤ͸

‘‘”ƒ‹Ž –”ƒ‹™‹–Š –™‘„‘‘•

Ͷ

(11)

‘™‡”’ƒ…

’‘™‡” ’ƒ… ȋ͵ͺȌ

…ƒ„‹ Ȁ–”ƒ‹ ȋͷ͸Ȍ

Fpower pack is weight of power pack with drive unit Fcabin / train is weight of driver’s cabin together with two drilling booms FIG 6 - Schematic diagram showing lengths and weights of monorail drilling system components.

Required strength of suspension chains The required strength (with factor of safety of 2.0) in each suspension chain is determined by equating Equations 10 and 11 to yield: FC, max = 2Rs Fdpart Ldpart MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

(12) 214

NUMERICAL MODELLING OF MONORAIL SUPPORT REQUIREMENTS IN DECLINE DEVELOPMENT

Required strength of roof bolts The required strength in each roof bolt is determined by substituting Equation 12 into Equation 1 to yield: FMS, max = 2Rs Fdpart Cosa Ldpart

(13)

Strength of support system at horizontal and vertical curves During monorail installation at vertical and horizontal curves, the required support system must be sufficient to overcome the dynamic effects and to avoid system failure. Also, it should be noted that during monorail installation process, the monorail components are rigidly fixed using steel supports at vertical curves while suspension chains are used at horizontal curves. It is, therefore, necessary to determine the strength of the required support systems that are used to suspend the monorail components at vertical and horizontal curves. As highlighted in the introduction, the monorail system can negotiate horizontal and vertical curve radii of 4 m and 10 m respectively. However, the curve lengths that result from these radii are small to accommodate the whole length of the monorail drilling and loading systems. Therefore, the weight of heaviest monorail drilling and loading systems components passing the vertical and horizontal curve is used. In this section, models that determine the strength of roof bolts, steel supports and suspension chains at vertical and horizontal curves based on the dynamic forces of the heaviest monorail drilling and loading system components are presented.

Strength of steel supports at vertical curves based on weight of monorail loading system components Taking equilibrium of forces at point A (Figure 7), the following equation that relates the force in each roof bolt and steel support is established: FMS = FM

(14) rv Rs Lpart Steel supports

Roof bolts FMS A FM

FD

B Fpart



 

Z Y



FM

FD B Fpart Cos(+) (+) Fpart Sin(+)

F part

is length of monorail loading s stem component is roof bolt spacing F is force re uired in each roof bolts (forces suspending monorail s stem) Fpart is weight of monorail loading s stem component F is force in each steel support F is net driving (propulsion) force  is decline gradient  is angle change at vertical curve rv is vertical curve radius part s

FIG 7 - Schematic longitudinal-section view of required support system at vertical curve based on the weight of monorail loading system components. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

215

B BESA, M KURUPPU AND E K CHANDA

During motion of the monorail loading system at a curve a centrifugal force FS (Equation 15) directed towards the centre of the curve is needed to make the monorail train or any component attached to it undergo motion at a vertical curve (Alan, 2003; Lawrence, 1997): FS =

mpart v2 rv

(15)

where: FS is the centrifugal force needed to make the monorail train or its components undergo uniform motion at a curve v is velocity of the monorail train or its component as it moves along the curve rv is vertical radius of the curve around which the monorail loading system or its components is moving mpart is mass of the monorail loading system component negotiating the curve Total force in steel supports at vertical curve The total force in steel supports at vertical curves depends on the number of steel supports installed within the length, Lpart, occupying the monorail loading system component and the force in each steel support. Since the roof bolt spacing, Rs, is known, the number of steel supports occupying the monorail loading system component of length Lpart, at a vertical curve is determined as: Total number of steel supports within Lpart = e

Lpart o Rs

(16)

where: Lpart is the length of monorail loading system component Thus, the total force in steel supports within the length Lpart at a vertical curve is the product of the total number of steel supports installed within the length Lpart and the force in each steel support (FM) as: Total force in steel support within Lpart = FM e

Lpart o Rs

(17)

It is assumed that the support distance Lpart is small and the variation of angle can be ignored. Using Equation 15 and taking equilibrium of forces in Z direction at point B (Figure 7), the resultant force of the monorail loading system component at a curve is determined as: FM e

Lpart mpart v2 o - Fpart Cos^a + Dbh = rv Rs

(18)

where: 00  700 Similarly, the net propulsion force (FD) of the monorail train at a curve is determined using Equation 19: FD = FpartSin ( +  )

(19)

From Equation 18, the strength in each steel support (with a factor of safety of 2.0) is determined as: 2

m v FM, max = 2Rs # e part + Fpart Cos^a + Dbho rv Lpart

(20)

The strength of steel supports at vertical curves is determined based on the heaviest component of the monorail loading system as discussed in previous section. Required strength of steel supports at vertical curves The required strength of each steel support at a vertical curve is determined using Equation 20. It should be noted that the maximum force in steel supports occurs when  = 0. Therefore, with this condition, Equation 20 can be written as: 2

m v FM, max = 2Rs # e part + Fpart Cosao rv Lpart MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

(21) 216

NUMERICAL MODELLING OF MONORAIL SUPPORT REQUIREMENTS IN DECLINE DEVELOPMENT

Required strength of roof bolts at vertical curves According to Equation 14, the force in each roof bolt at vertical curve is equal to the force in each steel support. Therefore, using Equations 14 and 21, the required strength in each roof bolt is determined as: 2

m v FMS, max = 2Rs # e part + Fpart Cosao rv Lpart

(22)

Strength of steel supports at vertical curves based on weight of monorail drilling system component The strength of required roof bolts and steel supports at vertical curves based on the weight of the monorail drilling system components is determined using Figure 8 (configuration is the same as loading system in Figure 7). rv Rs Ldpart Steel supports

Roof bolts

FMS A

FM

FD

B Fc

Fdpart



 

Z Y



FM

FD B Fdpart Cos(+) (+) Fdpart Sin(+)

F dpart

part is length of monorail drilling s stem component sis roof bolt spacing F is force re uired in each roof bolts (forces suspending monorail s stem) Fdpart is weight of an monorail drilling s stem component F is force in each steel support  is decline gradient  is angle change at vertical curve rv is vertical curve radius

FIG 8 - Schematic longitudinal-section view of required support system at vertical curve based on weight of monorail drilling system components.

Total force in steel supports at vertical curves Using similar analysis as for the monorail loading system yields: Total force in steel supports within curve Ldpart = FM e

Ldpart o Rs

(23)

The net pushing force (FD) of the monorail drilling system at vertical curve is determined as: FD = FdpartSin ( +  )

(24)

The force in each steel support (with a factor of safety of 2.0) is also determined as: 2

m v FM, max = 2Rs # e dpart + Fdpart Cos^a + Dbho r v Ldpart

(25)

Weight of monorail drilling system component The weight of the heaviest monorail drilling system component (Fdpart) of length Ldpart, is determined as outlined in previous section. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

217

B BESA, M KURUPPU AND E K CHANDA

Required strength of steel supports at vertical curve Since the maximum force in steel supports occurs when  = 0, the ultimate force in steel supports at a vertical curve is determined as: 2

m v FM, max = 2Rs # e dpart + Fdpart Cosao rv Ldpart

(26)

Required strength in roof bolts at vertical curve Using Equations 14 and 26, the ultimate force in each roof bolt is determined as: 2

m v FMS, max = 2Rs # e dpart + Fdpart Cosao rv Ldpart

(27)

Strength of suspension chains at horizontal curves based on monorail loading system Force and displacement of suspension chains at horizontal curves As the monorail loading systems negotiates a horizontal curve, suspension chains are displaced from the vertical position due to dynamic forces resulting from the motion of the systems. Figures 9 and 10 show the forces and displacement of suspension chain at a horizontal curve. As shown in Figures 9 and 10, centrifugal force, FS, results as the monorail loading system moves along the curve as given in Equation 28. FS =

m part v2 rh

(28)

where: rh is horizontal radius of the curve around which the monorail loading system or its components is moving FD

FS

VerticalSteel supports

rh F is centrifugal force e erted on moving monorail loading s stem F is net driving (propulsion) force rh is hori ontal curve radius

FIG 9 - Plan view of forces at the horizontal curve.

As shown in Figure 10, as the monorail drilling and loading systems negotiate the horizontal curve, the suspension chains are displaced from vertical positions by the angle, , and horizontal distance, X. It is important to determine these two parameters and the force carried by the suspension chains so as to determine whether the chains will fail or the systems will crash (as the chain is displaced) into the sidewall of the underground opening at a horizontal curve so that control measures are put in place. Using Figure 10, the following equation that relates the force in each roof bolt and the force in suspension chains at horizontal curve is established. FMS = FCSin MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

(29) 218

NUMERICAL MODELLING OF MONORAIL SUPPORT REQUIREMENTS IN DECLINE DEVELOPMENT

‘‘ˆ„‘Ž– ȋ Ȍ A

ƒ‰‹‰™ƒŽŽ ”‘‘ˆ

—•’‡•‹‘ Šƒ‹





FC   B  ‰ FIG 10 - Displacement of suspension chain from vertical position at horizontal curve.

Angular displacement (δ) of suspension chains due to monorail loading system The angular displacement of suspension chains from the vertical position is determined by resolving forces at point B to yield: Horizontal force balance: FC Cosi = FS =

mpart v2 rh

(30)

Vertical force balance: FCSin = mpartg

(31)

Dividing Equation 31 by Equation 30 yields: g r Tani = c #2 h m v

(32)

Therefore, the angular displacement of suspension chains is determined as: g r d = 90 - Tan-1 c #2 h m v

(33)

Equation 33 indicates that the maximum angular displacement of suspension chains depends on the radius of curvature of the horizontal curve and the velocity of the monorail system component at the curve. Thus, an increase in the radius of the horizontal curve results in smaller angular displacement and vice versa. Equation 33 also reveals that an increase in the velocity of the monorail system component at a curve results in an increase in angular displacement of suspension chains and vice versa. Horizontal displacement (X) of suspension chains due to monorail loading system Using trigonometry, the horizontal displacement by which suspension chains are displaced from the vertical position due to dynamic forces is found using Equation 34: 2

X = eL # v o rh # g MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

(34)

219

B BESA, M KURUPPU AND E K CHANDA

Equation 34 shows that horizontal displacement depends on the length of suspension chains, velocity of the monorail loading system components and the radius of curvature of the horizontal curve. The length of suspension chains and square of the velocity of the monorail loading system component varies directly with the horizontal displacement while the radius of curvature is inversely related with X. Force in suspension chains at horizontal curves Having found as per Equation 32, Equations 30 gives the value of the force (with a factor of safety of 2.0) in suspension chains FC due to dynamic force of the system as: FC,max =

2m part v2 # 1 rh Cosi

(35)

Using trigonometry: v2 ^ v + g2 r h2h Replacing Equation 36 into Equation 35 gives:

Cosi =

FC,max =

4

2mpart rh #

^ v4 + g2 r h2h

(36)

(37)

Force in roof bolts at horizontal curves The force in roof bolts at horizontal curves is obtained using Equation 32 and 37 to yield: FMS, max = 2mpartg

(38)

Strength of suspension chains at horizontal curves based on monorail drilling system Force and displacement of suspension chains at horizontal curves Base on similar analysis as in previous section, centrifugal force, FS, which results as the monorail drilling system moves along the curve is: FS =

mdpartv2 rh

(39)

Using Figure 10, the angular displacement of suspension chains from the vertical position is given as follows: Horizontal force balance: FC Cosi = FS =

mdpartv2 rh

Vertical force balance: FC Sini = mdpartg

(40) (41)

Required strength of suspension chains at horizontal curves The required strength of suspension chains (with a factor of safety of 2.0) at horizontal curves for the monorail drilling system is: FC,max =

2mdpart rh #

^ v4 + g2 r h2h

(42)

Force in roof bolts at horizontal curves The force in roof bolts at horizontal curves is obtained by using Equation 32 and 42 to yield: FMS, max = 2mdpartg

(43)

Variation of support system strength with change in decline gradient In this section, the variation of support system strength with changes in decline gradient is established. As the decline gradient changes, there is a corresponding change in the required force MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

220

NUMERICAL MODELLING OF MONORAIL SUPPORT REQUIREMENTS IN DECLINE DEVELOPMENT

in each support system. The developed models are used to establish this variation. Table 1 show the data used during the determination. The data is based on information from manufacturers of the monorail train, Scharf, and publications on monorail drilling and loading systems (Chanda and Besa, 2008, 2009; Chanda, Besa and Kuruppu, 2008b). TABLE 1 Parameters of the monorail system. Unit

Value

Comment

Lpart

Parameter

m

3.5

Manufacturer supplied

Ldpart

m

2.6

Manufacturer supplied

mpart

kg

5.1

Manufacturer supplied

mdpart

kg

5.7

Manufacturer supplied

Fpart

kN

50

Manufacturer supplied

Fdpart

kN

56

Manufacturer supplied

Rs

m

3

Manufacturer supplied

α

degrees

20

Assumed (Chanda et al, 2008; Chanda and Besa, 2009)

rv

m

10

Manufacturer supplied

rh

m

4

Manufacturer supplied

v

m/s

3.5

Manufacturer supplied

L

m

0.6

Manufacturer supplied

9.81

Constant

g

m/s

2

Numerical values of forces required in the support systems Using the developed models, the strength of the required support system with changes in decline gradient for the monorail drilling and loading systems is determined as of Figures 11 and 12. Results shown in Figure 11 and Figure 12 indicate that the force required to suspend the monorail drilling system components is higher than that needed to suspend the loading system components. According to the results, the force in suspension chains in an incline, horizontal curves and in roof bolts at horizontal curves remains constant with changes in decline gradient. However, in an incline and at vertical curves, the force in the roof bolts varies inversely with change in decline gradient, ie as the decline gradient increases the required force in the roof bolts reduces. Similarly, the force in steel supports at vertical curves varies inversely with decline gradient. ͳ͸Ͳ ͳͶͲ ͳʹͲ

Force (kN)

ͳͲͲ ͺͲ ͸Ͳ ͶͲ ʹͲ Ͳ Ͳ

ͷ

ͳͲ

ͳͷ

ʹͲ

ʹͷ

͵Ͳ

͵ͷ

ͶͲ

Declinegradient(degrees) Suspensionchainforce(FC)inanIncline Suspensionchainforce(FC)athorizontalcurve Roofboltforce(FMS)inaninline Roofboltforce(FMS)andsteelsupport force(FM)atverticalcurve Roofboltforce(FMS)athorizontalcurve

FIG 11 - Variation of force in support system with change in decline gradient for the monorail drilling system. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

221

B BESA, M KURUPPU AND E K CHANDA

ͳʹͲ

ͳͲͲ

Force (kN)

ͺͲ

͸Ͳ

ͶͲ

ʹͲ

Ͳ Ͳ

ͷ

ͳͲ

ͳͷ

ʹͲ

ʹͷ

͵Ͳ

͵ͷ

ͶͲ

Declinegradient(degrees) —•’‡•‹‘…Šƒ‹ˆ‘”…‡ȋ Ȍ‹ƒ‹…Ž‹‡ —•’‡•‹‘…Šƒ‹ˆ‘”…‡ȋ Ȍƒ–Š‘”‹œ‘–ƒŽ…—”˜‡ ‘‘ˆ„‘Ž–ˆ‘”…‡ȋ Ȍ‹ƒ‹Ž‹‡ ‘‘ˆ„‘Ž–ˆ‘”…‡ȋ Ȍƒ†•–‡‡Ž•—’’‘”–ˆ‘”…‡ȋ Ȍƒ–˜‡”–‹…ƒŽ…—”˜‡ ‘‘ˆ„‘Ž–ˆ‘”…‡ȋ Ȍƒ–Š‘”‹œ‘–ƒŽ…—”˜‡

FIG 12 - Variation of force in support system with change in decline gradient for the monorail loading system.

Strength of support system at 20° decline gradient Chanda and Besa (2008) presented a mine design case study in which a decline gradient of 20° was used. Using this case study, numerical values of the required support system strength at that gradient have been determined. Figure 13 shows the numerical values for each system while Table 2 shows the displacements of suspension chains at horizontal curves for 20° gradient. 11.4

Requiredstrengthofroofboltsathorizontalcurves(FMS)

10.2 12.0

Requiredstrengthofsuspensionchainsathorizontalcurves(FC)

10.7 14.0

Requiredstrengthofroofboltsatverticalcurves(FMS)

9.3 14.0

Requiredstrengthofsteelsupportsatverticalcurves(FM)

9.3 12.4

Requiredstrengthofroofboltsinanincline(FMS)

8.2 13.2

Requiredstrengthofsuspensionchainsinanincline(FC)

8.7

0.0

Supportsystem

2.0

4.0

6.0

8.0

10.0

12.0

14.0

16.0

Tonnes Monoraildrillingsystem

Monorailloadingsystem



FIG 13 - Strength of support system at 20° decline gradient. TABLE 2 Displacement of suspension chains at horizontal curves. Parameter Angular displacement (δ) Horizontal displacement (X)

Unit

Monorail loading system

Monorail drilling system

degrees

17.3

17.3

cm

18.7

18.7

High strength roof bolts, suspension chains and steel supports are required to suspend and support the monorail drilling system components more than that required for the monorail loading system. In comparison with the roof bolts (namely Hilti OneStep® roof bolts) and suspension chains currently being used which have an ultimate strength of 320 kN (32 tonnes) and 250 kN (25 tonnes), respectively, it is clear that the roof bolts and suspension chains have adequate strength to suspend and support the components of the two systems. Analysis of variation of decline gradient with strength of support system shows that the higher the decline gradient, the lower is the force in the support system. In terms of suspension chain displacements at horizontal curves, results have shown that both systems would give the same angular and horizontal displacement of 17.3° and 18.7 cm, respectively. These displacements can MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

222

NUMERICAL MODELLING OF MONORAIL SUPPORT REQUIREMENTS IN DECLINE DEVELOPMENT

be minimised by reducing the velocity of the monorail systems at horizontal curves or increasing the radius of the curve. Since both systems move on the same rail, Table 3 shows minimum numerical values that have been recommended. TABLE 3 Required strength of the support system.

Strength of parameter

Recommended value kN

Tonnes

Suspension chains in an incline (FC)

129.2

13.1

Roof bolts in an incline (FMS)

121.4

12.3

Steel supports at vertical curves (FSS)

137.6

14.0

Roof bolts at vertical curves (FMS)

137.6

14.0

Suspension chains at horizontal curves (FC)

117.3

12.0

Roof bolts at horizontal curves (FMS)

112.0

11.4

DISCUSSION AND CONCLUSION In spite of the advantages of the monorail system, one of the major risks is the potential of failure of the support system due to inadequate strength of roof bolt, suspension chain and steel support. Therefore, to avoid system failure, adequate strength of roof bolt, suspension chain and steel support that can support the proposed monorail system needs to be installed. This paper has demonstrated that to avoid roof bolt, suspension chain and steel support failure due to additional stresses from weight of the monorail drilling and loading systems, high strength roof bolts, suspension chains and steel supports to support the two systems must be installed. In comparison with the roof bolts currently in use, the models developed have demonstrated that the support system has adequate strength to support and suspend the two systems. It has also been established that the required strength of roof bolts varies inversely with the decline gradient. However, the strength of suspension chains in the decline and at horizontal curves as well as the strength of roof bolts at horizontal curves remains constant. To reduce or minimise displacements of suspension chains, it is recommended that the velocity of the monorail system at horizontal curves be reduced during motion.

ACKNOWLEDGEMENTS The authors would like to thank Minerals and Energy Research Institute of Western Australia (MERIWA) and Newmont for financial support of the monorail research project at Western Australian School of Mines (WASM) and Scharf Mining Solutions for providing technical data on the Electric Monorail Transport System (EMTS).

REFERENCES Alan, D, 2003. Mechanical Engineering: BTEC National Option Units, 432 p (Newnes). Chanda, E K and Besa, B, 2008. Monorail technology – A rapid and cost effective method of decline development, in Proceedings Narrow Vein Mining Conference, pp 129-141 (The Australasian Institute of Mining and Metallurgy: Melbourne). Chanda, E K and Besa, B, 2009. Design of pneumatic loading system for monorail application, Int J Mining and Mineral Engineering, 1(2):181-203. Chanda, E K, Besa B and Kuruppu, M, 2008. Design of a continuous monorail drilling system for decline development, in Proceedings The First International Future Mining Conference, pp 101-111 (The Australasian Institute of Mining and Metallurgy: Melbourne). Guse, A and Weibezhn, K, 1997. Continuous transport in hard rock mining, Colloquium – Underground Lateral Transport, p 1-6 (South African Institute of Mining and Metallurgy: Johannesburg). Lawrence, S L, 1997. Physics for Scientists and Engineers, 128 p (Jones and Bartlett: Boston). Scharf, 2007. Electrical monorail transport system (EMTS), brochure, DBT Maschinenfabrik, Scharf, GmbH.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

223

Case Study – The Mogiana Quarry Reclamation A Curi1 and O Quaglio2 ABSTRACT This article presents a proposal for the rehabilitation of the mined area of the Mogiana quarry. The Mogiana quarry produces material used in civil construction and is situated in the Brazilian State of São Paulo in the kilometre six of the Mogí Mirim - Itapira highway. The mineral exploitation in the Mogiana quarry provokes modifications in the environment such as alterations in the relief, vegetation, quality of the air, water and landscape. Through previous study of the area it was verified that after the closure of the exploitation the mine should be occupied mainly by pasturage. At the closing of the mine the exploitation benches, berms and slopes in general as well as the surrounding areas should be covered with a layer of soil and vegetated with native species in order to mitigate the negative impacts caused by the mine. The rehabilitation project of the area is based on the probable future use of the area, located between agricultural cultivations and cattle creation. The great alteration of the relief and soil would make future use of the area for agriculture difficult. It is more suitable t0 animal creation, mainly cattle. The high and thick rock slopes originated by the mining should be covered by vegetation, where possible, to reduce the negative visual impact. Obviously several measures will be necessary to reduce the impacts on the environment during mining. The final and definitive reclamation will only be possible after abandonment of the area. The main aspects of this reclamation process will be discussed in this paper.

INTRODUCTION The area of the mine presents a relief characteristic of the rocks of the Brazilian Paraná Basin. The outcrops in the sedimentary rocks area are characterised by sandy sediments. The magmatic rocks are represented by an intrusive body of diabase. The sandy rocks belong to the carboniferous period. The basic intrusive body presents fine granulation and is constituted of feldspar, plagioclase, pyroxene and magnetite. The diabasic rocks belong to the Cretaceous period. The diabase, locally, is an impermeable rock and the percolation of waters is only possible through the fractures system. In the pit area there is no evidence of water. The climate of the region is classified as subtropical, warm and humid (Stezer, 1966; Nascimento, 1988). According to studies on natural resources performed by the project RADAM Brazil (1983), the quarry is in a bordering area between savannah and seasonal forest. The savannah is represented by dense arboreal formation. The savannah, denominated cerrados in Brazil is characterised by the vast occurrence in the dominant stratum of the micro-fhanerophytas Sucupira (locust tree) (Pterodon sp), Angico (Piptadenia sp), Pequi (Caryocar sp) and Barbatimão (Stryphnodendron sp). The predominant species in the seasonal forest are the macro-fhanerophytas Peroba (Aspidosperma sp), Jequitibá (Cariniana sp), Cedro (Cedar) (Cedrela sp) and the meso-fhanerophytas copaiba (Copaifera sp) and canelas (Octotea sp and Nectandra sp). The natural vegetation of the region has been altered, frequently, by the agriculture and pastures. Citric cultures and another annual cultures such as corn and cotton also exist in the region. According to environmental studies executed by the project RADAM Brazil (1983), some of these areas are in an advanced process of degradation as a result of the human action. The rehabilitation of these areas will not be done naturally. It will be necessary to plant new species adapted to the environment. In the quarry adjacencies predominates corn and cotton cultivation. There is a small forest at the pit vicinity. 1. Professor, University Federal Ouro Preto, Escola de Minas – Demin, Campus Universitário, Ouro Preto MG 35400-000, Brazil. Email: [email protected] 2. Professor, Universidade Federal de Alfenas-MG, Departamento de Ciência e Tecnologia, Campus Avançado de Poços de Caldas, Rua Corumbá, 72 Jardim dos Estados,

Poços de Caldas MG 37701-070, Brazil. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

225

A CURI AND O QUAGLIO

ENVIRONMENTAL IMPACTS IN THE MINE Although the mining of diabase is only a temporary occupier of the land surface, it causes serious environmental impacts associated with overburden, hauling roads, unvegetated surfaces, coarse rejects, topsoil and stockpiles. The quality of the air, water and land is affected. Dust adversely affects air quality. The physical quality and quantity of surface water and groundwater can be affected if reclamation is not practised. Watercourses may be disturbed and flow rates altered. Erosion may be excessive and the surface and groundwater may become mineralised. The topography, drainage, vegetation, appearance and surface texture of the mine site may all be seriously impacted (Williams and Morris, 1993). The impacts of the mining also include the destruction of the landscape, degradation of the visual environment and destruction of agricultural and forest lands. The original relief was significantly altered, existing in the pit area a general debasement of 30 m taking as reference the original ground elevation. There are cuts in the terrain projected for construction of workshops, offices and goods storage. These alterations interfere in the landscape and modify the morphology of the terrain. The soil removed from the exploitation was positioned, erratically, in two different locations, forming large piles. In these piles the soil was mixed with waste rock of different sizes and types. All material was deposited in a disorderly manner and without criterion. A nearby location was selected for the future soil and waste disposition, which will be subdivided in three sectors (RB, OS, S2). In these sites will be stored, respectively, the rock blocks (Sector RB), the superior level of organic soil (OS) and the remaining soil (S2), considering the other levels of the soil profile. The soil should be stored in approximately horizontal layers. The total volume of soil and waste to be stored during the mine useful life was estimated in 355 600 m3. The estimate was made considering the thickness of the overburden and the area of the final pit.

GUIDELINES FOR RECLAMATION OF THE DEGRADED AREAS Soil reclamation According to Reichmann (cited by Griffith, 1986), a vegetation with seeds is the most economic and effective method in terms of soil recovery and slope stabilisation. Bradshaw and Chadwichv (cited by Griffith, 1986) recommend a maximal slope inclination of 18° to maintain aesthetics, to allow a mechanisation in the erosion control and application of soil correctives, planting and fertilisation. Curtis (cited by Griffith, 1986) recognises vegetation as the unique long-term measure for an effective control of soil erosion. Soil acidity inhibits vegetation establishment and affects nutrients availability and the biological processes of the plants. Czapowskyj and Sowa (cited by Griffith, 1986) concluded that application of calcareous was essential for pastures establishment and growth in coal mining. Some special measures of protection, besides the control of the acidity, are necessary for preservation of the soil quality stored in heaps. The improvement of the soil’s chemical and physical condition depends on the implementation of measures for appropriate storage, such as limestone addition and organic and inorganic manures (Farmer et al, cited by Griffith, 1986). The efficiency of this manuring depends on the fertility of the soil. In poor soils nitrogen manuring can jeopardise pastures. The manuring with phosphates is recommended by Primavesi (1986). Then, plowing of the terrain and a sowing of gramineous with a manure distributor or manually is recommended.

Pedologic and hydrologic measures The mine waste was stored in two big piles inside the pit. These piles were exposed to erosion and slidings. These deposits should receive a vegetable coverage for contention of the soil avoiding an earthwork formation in the stream situated in the proximities. The soil superficial organic layer of 25 cm should be stored, separately, to be used in posterior reclamation. The organic soil should be disposed in layers of at least 1.5 m of height, 3 to 4 m of width and a necessary length. The site used for storage must have a flat format to avoid the transportation of the fine particles, eventually, subject to the erosion. The operations of movement and deposition of the soils should be completed in drought time. In each soil layer disposed should be done, sporadically, the planting of leguminosae. The assimilation of the leguminosae will prosecute after a brief period of growth (100 - 120 days). The planting should receive manure for faster growth of the vegetation, better coverage of the soil and increase of the fertility of the waste that will be used after in the berms of the pit. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

226

CASE STUDY – THE MOGIANA QUARRY RECLAMATION

The inclination of the pit slopes should be softened and the same ones also should be vegetated to avoid erosion and sliding. For slopes coverage the species of gramineous denominated Brachiaria decumbens was selected due to its descending habit (creeper). This creeper provide an efficient coverage of the slopes surfaces. A month before the gramineous planting should be completed a manuring with phosphates and a sowing should be applied at the beginning of the rainfalls (November - December). The sowing can be done manually. The areas with risen slopes angles, in the neighbourhood of the stream, including the adjacent woods, should receive a vegetal cover according to curves of levelling. The proposal is to form a dense gramineous strip to prevent torrents formation and to retain water in the soil. Where the capacity of water absorption of the soil is low it will be necessary to maintain a small gradient in the lateral slope (0.5 per cent to one per cent ) to water drainage. The drainage water system must be vegetated, where possible. For reclamation of the pit area it is possible to use the waste generated in mining. This material can be used in the base of the layer constructed for vegetation in the berms of the pit and in the bottom of the pit. The berms or platforms to be vegetated should have at least 4 m of width and 10 m of height and should circle all the rock wall inside the pit. In the pit limits or in more critical regions, dykes formed of overlapped rocks up to 30 m of height should be constructed. These berms or platforms will receive the stored soil and improved through the techniques cited previously. In the top around the pit a strip of 10 m of vegetation, with the same species used in the berms should be planted. Around the pit channels for water capitation and flow should be made to avoid the berms flooding. The channels should be vegetated with gramineous to cut off the water speed and not let the soil be exposed to erosion. It was planned to construct a soccer field on one of the waste piles, close to the quarry office.

Reclamation plan The waste piles should be disposed, if possible, inside of the mine pit exploited, or close to the areas already degraded. It should be avoided to dispose the material in valleys, mainly, with larger inclination than 18°; natural surfaces drainages, water courses; areas of permanent preservation; unstable areas, flooded lands; areas with exuberant native vegetation; areas with fertile soils. In terms of external and internal geometry of the pile, the following limits and cares should be observed: maximum height of banks of 10 m; minimum width of berms of 6 m; maximum height of the pile of 200 m; existence of maintenance accesses; to reduce the angle among banks, for values less than the natural angle of the waste; berms with minimal longitudinal and traverse gradient of one per cent and five per cent, respectively; pile drainage (ABNT, 1993). Other parameters are: size classification of the materials to be disposed of to take advantage of the maximum resistance and drainage characteristics of each material; size classification of the material; execution of the pile in an ascending way; protection of the slopes with vegetation; correct disposition of the organic soil of the pile for future use; interns, superficial and outlying drainage system; a system for retention of sediments; a system for environment control or management (ABNT, 1993). For the establishment of the vegetation of the area some species are suggested, selected due to their characteristics of acclimatisation in the region: fast growth, tolerance in terms of nutrients deficit and hydrologic deficit. The recommended species are listed in Table 1 (Nicola, 1992). For vegetation, fruitful species of fast growth were selected, with the objectives of feed the fauna, promote soil coverage and protect stream banks. Candiúba and Calabura will be planted first. After a growth period of the pioneer species other native species of slower development will be planted. Additionally, the mulberry tree (Morus nigra) will be used, which, although exotic, was chosen for protection of the stream banks and supply feed for the animals. The indicated spacing between plants is 2 - 3 m, planted at random, blending the individuals of the three species. A second planting will be realised 18 to 24 months later. The objective of the second planting is to promote an enrichment of the vegetation through the introduction of new species, such as Ingá, Genipapo, Pitanga (Surinam cherry tree), Uvaia, Jaboticabeira and Goiabeira (guava tree), using a spacing of 4 - 5 m. The indicated spacing between plants is 4 - 5 m, planted at random, blending the individuals of the several species selected. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

227

A CURI AND O QUAGLIO

TABLE 1 Species selected for reclamation in the Mogiana quarry (Nicola,1992). Scientific name

Popular name

Size

Location

Piptaenia macrocarpa

Angico

A/B

L-S-P-B

Dalberoia nigra

Cabiúna

A/B

L-S-P-B

Guapuruvu

A/B

L-S-P-B

Havenia dulcis

Schizolobium parahibum

Uva-japonesa

A/B

L-S-P-B

Bauhinia sp.

Casco-de-vaca

A/M

L-S-P-B

Tibouchina sp.

Quaresmeira

A/M

L-S-P-B

Sibipiruna

A/A

L-S-P-B

Manacá-da-serra

A/S

L-S-P-B

Amoreira

A/S

L-S-P-B

Goiaba

A/S

L-S-P-B

Caesalpinia peltophoroides Tibouchina pulchra Morus nigra Psidium guajava Carica papaya

Mamoeiro

A/S

L-S-P-B

Eugenia sp

Pitangueira

A/M

L-S-P-B

Eugenia sp

Jambeiro

A/S

L-S-P-B

Eugenia sp

Uvaia

A/M

L-S-P-B

Carambola

A/S

L-S-P-B

Ingá

A/M

L-S-P-B

Alamandra

C

L-S-P-B

Glicínia

C

L-S-P-B

Averhoa carambola Inga sp. Allamanda cathartica Wisteria sinensis Mandevilla sanderi

Dipladênia

C

L-S-P-B

Trema micrantha

Candiúba

Sh

L-S-P-B

Muntingia calabura

Calabura

Sh

L-S-P-B

Genipa americana

Jenipapo

Sh

L-S-P-B

Mirciaria cauliflora

Jaboticabeira

Sh/M

L-S-P-B

Legend: A – arboreal; Sh – shrubby; C - creeper; (B – big, S – small, M – medium); L – limits of the property; P – strip of 10 m around the pit; S – banks of the stream; B – berms.

Planting in the berms In this planting observe the size of the species used. Close to the rock wall plant medium size species, where the soil is more profound; in the intermediary strip plant underbrush and in the berms extremities use creepers species. The creepers project their foliages downwards, hiding the rock wall. The spacings between individuals of the same species should be the next: x the distance between creepers (Alamanda, Glicínia, Dipladênia) should be 1 m and these ones can be planted up to 0.5 m of the contention dyke, x the distance between medium size trees should be maintained at 3 m and these ones should be planted at least 1 m in distance from the wall of the berms, and x the distance between underbrush should be 2 m.

Enrichment of the stream adjacent wood The forests and other forms of natural vegetation along the rivers or any water course are considered of permanent preservation, according to the Brazilian Forests Code. The wood preservation in the banks of the brooks is important for the contention of the soils, protection against erosion and to avoid a filling with land of the water course, influencing the environmental quality. Besides giving support to the wild fauna, the banks are of great value to the ichthyofauna. The preservation of existing wood and planting of fruitful species should be accomplished in a strip of 10 m along both sides of the stream. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

228

CASE STUDY – THE MOGIANA QUARRY RECLAMATION

Pastures planting All the area used by mining will be afterwards transformed into pastures. A selection of species adapted to condition of low fertility and high acidity is necessary for the formation of the pasturage. These characteristics are usual in degraded areas. Plowing of the superficial layer of the soil compacted by mining operations will be required. The height of the superficial layer compacted should be determined through drilling in the terrain. The plowing must be done in drought time (July to August). The fertilisation through mineral manures, as well as the need of limestone addition, will be based on the results of the soil analyses. The maximal quantity of calcareous applied in a year should not exceed three tons per hectare. Above this level, the availability of nutrients for the plants will be affected. The correction of the acidity should also be realised in drought time, after plow and at least a month before the planting of the gramineous. A manuring with phosphates can be accomplished little time before the sowing. The quantity of the manures to be applied must be be determined through soil analysis. Three or four months after the planting is time to proceed a nitrous manuring to guarantee a better resistance of the plant in the winter and better coverage of the soil in the drought times. It can be accomplished at the end of the rainfalls (March or April).

Pastures planting in the service area The species of gramineous denominated Brachiaria decumbens will be used in this area because it is more tolerant to poor soils and has a habit of descending growth, providing faster soil coverage. The quantity of seeds necessary is 10 - 15 kg per hectare, depending on the quality of the soil. Before the planting of gramineous the soil should be recovered through measures to improve its physical structure and fertility. With the suppression of the superficial layer of the soil in the service areas exposing the subsoil it is necessary to incorporate great quantities of organic matter and fertilisers for the adequate development of introduced plants. After plowing and preparation of these areas up to 20 tons per hectare of cane husks should be added to the soil using iron fence and at the same time limestone should be added for correction of the acidity. In October the sowing of the Crotalária (Crotalaria juncea) and the Brachiaria together with a manuring with phosphates should be done. After the sowing perform a passage with a land roller of little weight to increase the contact of the seed with the soil and to increase its humidity. After a growth period of four to five months the division of the area in sectors, using pickets should be realised. This division will be based on the production of green mass or organic material, observing the need for water supply, addition of salts and other chemical products or supplements. The trees planting for enrichment of the soil and protection of the cattle should be done mostly in the areas with deficient drainage and accentuated slope. All the sectors, delimitated by the pickets, should have trees, with a rarefied distribution in order to not provide many shadows by excessive foliages. The recommended species are: Erythrina sp and Casesalpinia peltophoroides (Sibipiruna). Cutting planting from 2 - 3 m height that will grow quickly and will not be susceptible to damage caused by cattle is recommended. After the planting and when the pasture is formed it is suggested to protect trees with a fence before introducing cattle to the area.

Pastures planting in the pit The surface reclamation of the open pit will be executed using the overburden generated in the mining operations. The structure and general characteristics of this material was modified. There are a loss of the structure and natural profile of the soil and it becomes more susceptible to the erosion and aggregation of the particles. In such cases a different handling of the pastures will be imperious. The equipment, including tractors, must be used with moderation, doing the limestone addition, manuring with phosphates and sowing only. The plow and the movement of the soil can be dispensed to avoid excessive aggregation of the particles of the soil. Moreover the waste should be in good condition, having been stored according to the previous recommendations. The soil placement in the pit should be done in drought time. Soon after the correction of the acidity and the manuring with phosphates should be done. The sowing of the gramineous selected for pasture is recommended at the beginning of the next rainfall period. The species more indicated in this case is the grass (Brachiaria mutica), which is resistant to flooding and poor soils. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

229

A CURI AND O QUAGLIO

To guarantee that the soil deposited in the pit stabilises without flooding risk an efficient system of drainage should be adopted. The fund of the pit should present a gradient in direction to the exit of water. A drainage system should be executed in every sector of the pit. The bottom of the pit should be covered by a layer formed by small pieces of stones so that the drainage will be facilitated. A drainage trench of larger dimensions projected in direction to the adjacent stream is necessary to allow the flowing of the water from the pit bottom. The trees planting with a system of radial roots benefits soil drainage and must be encouraged. These tress should be planted in groups inside the pit. The species to use can be the same ones cited previously for pasture in the service area. Good drainage is essential for the correct development of the pastures. Flooding of the soil for long periods creates a reducer ambient, neutralising the effect of the limestone. This area should be managed carefully, supporting just a few animals and short periods of cattle pasture. The division of the pasture in sectors should be accomplished following the previous recommendations for the pasture in the service area. Some auxiliary facilities for the handling of the cattle will be necessary. It is possible to adapt some old mine constructions for this objective and the others ones should be destroyed.

WASTE DISPOSITION IN THE PIT The soil placement in the berms will be processed by the top of the pit, pouring the material, directly, using the trucks to deposit the waste in the platforms below. The average height of soil in the berms will be of 0.75 m, which means that about 6.900 m3 of waste will be necessary to fill this part of the pit. In the bottom of the pit the minimum height of soil should be of 0.80 m, 40.000 m3 being necessary for the coverage of this area. The waste should be placed at drought time to allow better disposition to the soil and to avoid damages to the physical structure of the soil.

MANAGEMENT The development of the species chosen for vegetation, both arboreal and gramineous, for formation of the pasture should be observed, evaluating the efficiency of the coverage and adaptation of the species to the existing condition in the area. This handling allows alterations of the initial proposal aiming best results. The planting should be accomplished at the beginning of the rainy station. The handling should evaluate the number of individual for area, its height, diameter of the stem, percentage of soil covered by vegetation, vigour, vegetative and reproductive status, presence of plagues or diseases, possible failures in the vegetation. Relating to trees consider the projection of the treetop, diameter for the height of 1.5 m (DHP), total height and percentage of failures for each species. Every two months it is convenient to accomplish observations about: x occurrence of laminar erosion and in furrow; x formation of the vegetable coverage in the soil; x falling of leaves, bloom, seeds and fruits liberation; x animals footprints, animals faeces, birds feathers, animals coat, animals food residues; x natural regeneration of plants, pioneering and secondary trees initials, herbs and underbrush; x chemical and physical analyses of the soil,which must be done annually; x control of herbivorous animals; x irrigation of the location when necessary; x fertilisation with phosphates to guarantee a good roots formation; x inspection of the area in order to detect plagues or diseases attacks; x protection of the area against the fire; x accomplishing of the planting in the trees failures; and x evolution of the growth of trees and dry vegetal cover in the soil. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

230

CASE STUDY – THE MOGIANA QUARRY RECLAMATION

The main activities of the final reclamation plan are foreseen below, in Table 2. The beginning of the recovery plan should be accomplished before the closure of the quarry. The management should be done during the implantation phase of the plan and every two months after the planting and sowing of the gramineous, until the vegetation is developed and stable. TABLE 2 Reclamation schedule. Activities

Period of the year

Soil analysis

First month

Soil preparation (plow and incorporation of organic material, addition of calcareous)

July - August

Waste disposition in the berms and bottom of the pit, correction of the acidity

July - August

Manuring of the pastures areas with phosphates

September - October

Sowing of leguminosae in the areas destined to pastures

October - November

Planting of calabura , candiúba and mulberry tree in the berms and adjacent woods Incorporation of the organic matter of the leguminosae New correction of the acidity based on results of soil analyses

November - December March - April (second year) July - August

Manuring based on analysis of the soil

September - October

Sowing of the gramineous for pastures formation

November - December

Planting of the arboreal species in the pastures

November - December

Manuring with nitrogen of the pastures

March - April

Demarcation of the area with pickets

March - April

Planting of slow growth plants in berms, around the pit and adjacent wood

November - December (third year)

ACKNOWLEDGEMENTS The authors would like to thanks the Fapemig – Fundação de Amparo à Pesquisa do Estado de Minas Gerais and the CNPQ – Conselho Nacional de Desenvolvimento Científico e Técnológico for the financial support to prepare and present this work.

REFERENCES Associação Brasileira de Normas Técnicas (ABNT), 1993. Coletânea de Normas de Mineração e Meio Ambiente, Norma ABNT 13029, Elaboração e apresentação de projeto de disposição de estéril, em pilha, em mineração, Rio de Janeiro. Griffith, J J, 1986. Política Ambiental e métodos de Reabilitação para Mineração de cassiterita na Floresta Nacional do Jamari, IBDF Instituto Brasileiro de Desenvolvimento Florestal, Sociedade de Investigações Florestais, Departamento de Engenharia Floresta, Universidade Federal de Viçosa, Brasil. Nascimento, C M, 1988. Atlas Climatológico do Estado de São Paulo, 93 p (Fundação Cargil: Campinas). Nicola, J P, 1992. Resumo do plano de reconstituição ambiental da Pedreira Mogiana, Pedreira Mogi Mirim, internal report, São Paulo. Primavesi, A, 1986. Manejo Ecológico de Pastagens, 184 p (ed: Nobel) (São Paulo). Projeto RADAM-BRASIL, 1983. Levantamento de Recursos Naturais, vol 32 (Ministério das Minas e Energia: Rio de Janeiro). Stezer, J, 1966. Atlas Climático e Ecológico do Estado de São Paulo, 61 p (Comissão Interestadual da Bacia Paraná-Uruguai e CESP: São Paulo). Williams, D J, Wu, Y and Morris, P H, 1993. Systems analysis of engineered mine site rehabilitation, in Proceedings Fourth International Conference on Tailings and Mine Waste, Fort Collins, Colorado (A A Balkema: Rotterdam).

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

231

Study on Utilisation of Flyash for Barrier Layer/Buffer Materials for Radioactive Waste Disposal T Sasaoka1, H Shimada2 and K Matsui3 ABSTRACT About 15 per cent of total electricity (400 MkW) was generated by nuclear power plants in the world (2009) and about 36 000 tons of uranium has been mined in uranium mines annually. Radioactive waste material is produced from uranium mines and nuclear power plants. The wastes must be disposed or stored safely for long-term. If they leak and/or move from disposal or storage sites due to air/groundwater flow, then a serious environmental pollution can occur. Hence, multi-layer system has been proposed and employed in order to seal off these radioactive waste materials from biosphere. Basically, bentonite is now used for establishing absorbing and sealing layers in this system. However, the amount of high quality betonite is very limited and in some cases it is hard to be obtained. On the other hand, a great deal of refuse from coal burning plants is produced every year and the amount of it is expected to be higher each year, especially in developing countries. More than half of flyash is utilised and the remaining is disposed at the disposal sites. However, the life of the disposal site is limited and it is difficult to find a new disposal site. It is requested that the percentage of the utilisation of the flyash be increased in every field. From the above two points of view, a flyash-based barrier layer/cover system is considered in this research and this paper discusses the applicability of flyash as a content of barrier layer/cover material based on the results of a series of laboratory tests.

INTRODUCTION Considering the situation of energy demand in the world, nuclear power generation might be growing up from now on. About 15 per cent of total electricity (400 MkW) was generated by nuclear power plants in the world (2009) and about 36 000 tons of uranium has been mined in uranium mines annually. Radioactive waste material such as overburden, waste rocks and tailing materials is produced in uranium mines due to the mining operation, milling and uranium refinement. Radioactive waste is also a by-product from nuclear reactors, fuel processing plants and institutions such as hospitals and research facilities (United States Nuclear Regulatory Commission, 2010). Since the only way radioactive wastes finally become harmless is through decay, which for some isotopes contained in high-level wastes can take hundreds of thousands of years, the wastes must be stored in a way that provides adequate protection for very long times. Because if they leak and/or move from disposal or storage sites due to air/groundwater flow, then a serious environmental pollution can occur. Hence, a multi-layer system has been proposed and employed in order to seal off these radioactive waste materials from biosphere (Berlin and Stanton, 1989). Basically, bentonite is now used for establishing one of absorbing and sealing layers in this system. However, the amount of high quality betonite is very limited and in some case it is hard to be obtained. On the other hand, a great deal of refuse from coal burning plants is produced every year and the amount of it is expected to be higher each year especially in developing countries. More than half of flyash is utilised and the remaining is disposed at the disposal sites. However, the life of the disposal site is limited and it is difficult to find a new disposal site. It is requested that the percentage of the utilisation of the flyash be increased in every field. 1. Assistant Professor, Department of Earth Resources Engineering, Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan. Email: [email protected] 2. Associate Professor, Department of Earth Resources Engineering, Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan. Email: [email protected] 3. Professor, Department of Earth Resources Engineering, Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

233

T SASAOKA, H SHIMADA AND K MATSUI

From the above points of view, a flyash-based barrier layer/cover system instead of bentoniteonly one is proposed in this research. This paper describes the current system and technology for radioactive waste disposal and then proposes and discusses the applicability of flyash as a content of barrier layer/cover material based on the results of a series of laboratory tests.

RADIOACTIVE WASTE MATERIAL AND DISPOSAL SYSTEM The Nuclear Regulatory Commission separates wastes into two broad classifications: high-level or low-level waste (United States Nuclear Regulatory Commission, 2010). High-level radioactive waste results primarily from the fuel used by reactors to produce electricity. Low-level radioactive waste results from uranium mine, reactor operations and from medical, academic, industrial and other commercial uses.

High-level radioactive waste management High-level radioactive wastes are the highly radioactive materials produced as a by-product of the reactions that occur inside nuclear reactors (Byalko, 1994). Reprocessing extracts isotopes from spent fuel can be used again as reactor fuel. Because of their highly radioactive fission products, high-level waste and spent fuel must be handled and stored with care. Since the only way radioactive waste finally becomes harmless is through decay, which for high-level wastes can take hundreds of thousands of years, the wastes must be stored and finally disposed of in a way that provides adequate protection of the public for a very long time. High-level waste will be disposed of in a stable geological formation at a depth of more than 300 m. The vitrified waste in fabrication canisters will be encapsulated in strong metal containers (overpacks) and, once placed in the repository, will be surrounded by a clay/bentonite buffer material. The canisters, overpacks and clay/bentonite buffer material are referred to as the engineered barrier system. The geological environment, which isolates the waste for long time periods, is termed the natural barrier. The multi-barrier system used for safe waste disposal is a combination of engineered and natural barrier. Research and development on the multi-barrier system will continue with a view to building confidence in this concept (Nuclear Waste Management Organisation of Japan, 2010). Figure 1 shows the schematic of high-level radioactive waste disposal facility.

FIG 1 - Schematic of high-level radioactive waste disposal facility.

Low-level radioactive waste management Low-level waste includes items that have become contaminated with radioactive material or have become radioactive through exposure to neutron radiation. This waste typically consists of contaminated protective shoe covers and clothing, wiping rags, mops, filters, reactor water treatment MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

234

STUDY ON UTILISATION OF FLYASH FOR BARRIER LAYER/BUFFER MATERIALS FOR RADIOACTIVE WASTE DISPOSAL

residues, equipment and tools, luminous dials, medical tubes, swabs, injection needles, syringes and laboratory animal carcasses and tissues. The radioactivity can range from just above background levels found in nature to very highly radioactive in certain cases such as parts from inside the reactor vessel in a nuclear power plant. Low-level waste is typically stored onsite by licensees, either until it has decayed away and can be disposed of as ordinary trash, or until amounts are large enough for shipment to a low-level waste disposal site in containers. Figure 2 illustrates the schematic of lowlevel radioactive waste disposal facility.

FIG 2 - Schematic of low-level radioactive waste disposal facility.

Uranium mill tailings Uranium mill tailings are primarily the sandy process waste material from a conventional uranium mill (Brookins, 1984). This ore residue contains the radioactive decay products from the uranium chains (mainly the U-238 chain) and heavy metals. The tailings or wastes produced by the extraction or concentration of uranium or thorium from any ore processed primarily for its source material content is by-product material. This includes discrete surface waste resulting from uranium solution extraction processes, such as in situ recovery, heap leach, and ion-exchange. By-product material does not include underground orebodies depleted by solution extraction. The wastes from these solution extraction facilities are transported to a mill tailings impoundment for disposal. Thick earthen covers is constructed in order to protect it by rock and designed to prevent seepage into ground water, over the waste. Earthen covers also effectively limit radon emissions and gamma radiation and, in conjunction with the rock covers, serve to stabilise the piles to prevent dispersion of the tailings through erosion or intrusion. In some cases, piles may be moved to safer locations. Figure 3 illustrates the schematic of dumping site in uranium mine.

UTILISATION OF FLYASH Flyash utilisation in Japan Considering the expansion of coal utilisation, it is necessary to promote the development of highly efficient coal and flyash utilisation technologies (Centre for Coal Utilisation, 2003). In Japan, flyash production has been increasing and in FY2006, about 10 Mt of flyash was produced. However, the MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

235

T SASAOKA, H SHIMADA AND K MATSUI

FIG 3 - Schematic of dumping site in uranium mine.

issue is not only the level of flyash produced but also the effective availability of flyash utilisation that gradually increases every year. This means that effective flyash utilisation technology has steadily improved in Japan over the past decade. Total flyash utilisation in FY2007 was 10 Mt, and the average ash utilisation ratio was 85 per cent (Yoshida et al, 2009). The amount of flyash utilisation has steadily increased while disposal has steadily decreased. Currently flyash utilisation is approximately doubled compared with that in the early 1990s, and the amount of flyash disposal approximately reduced by half. However, the capacity of the landfill site is approaching its limit year by year. The promotion of flyash utilisation must be discussed seriously.

Characterisation of flyash The different shapes of flyash generally include spheres for that with a low fuse temperature point and irregular shapes for that with a high fuse temperature point. The average particle diameter of flyash produced by combustion of pulverised coal is approximately 25 m, coarser than clay and finer than granular sand, which is equal to silt in terms of soil quality. The main chemical components in flyash are silica and aluminium oxide, which is close to pit soil (SiO2: 60 - 70 per cent, Al2O3: 10 - 25 per cent). Flyash produced by fluidised bed combustion has a higher CaO content than that produced by combustion of pulverised coal. Since the coal used in Japan is imported from different countries, its physical properties vary significantly. The chemical and physical properties of flyash produced by combustion of pulverised coal are shown in Table 1 and those resulting from fluidised bed combustion are shown in Table 2.

Utilisation of flyash in various fields The utilisation of flyash in each sector is shown in Figure 4. The amount of flyash used in the cement industry, which is one of Japan’s major areas of utilisation, accounts for 75 per cent of the total, of which nearly 6.0 Mt of ash was used in the raw material for cement manufacturing. Limestone, clay and iron oxide are used as raw materials for cement, among which clay generally accounts for 15 per cent of the total. The use of flyash as a substitute for clay accounts for a large part of the current utilisation of flyash. Flyash is particularly expected to be used as a material for cement/ concrete admixtures or in public works where there is a high potential for large-scale utilisation. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

236

STUDY ON UTILISATION OF FLYASH FOR BARRIER LAYER/BUFFER MATERIALS FOR RADIOACTIVE WASTE DISPOSAL

TABLE 1 Main chemical and physical properties of pulverised coal combustion flyash.

Field

Cement

Electric utilities

Description

General industries

Total

Amount used

%

Amount used

%

Amount used

%

Cement raw material (substitute for clay)

4354

71.32

1522

66.90

5876

70.12

Cement admixture

149

2.44

159

6.99

308

3.68

Concrete admixture

95

1.56

48

2.11

143

1.71

4598

75.32

1729

76.00

6327

75.50

Material for ground improvement

138

2.26

104

4.57

242

2.89

Pubric works

103

1.69

25

1.10

128

1.53

Power supply construction

79

1.29

0

0.00

79

0.94

Road base material

50

0.82

110

4.84

160

1.91

Asphalt filler

9

0.15

0

0.00

9

0.11

Coal mine filler

204

3.34

0

0.00

204

2.43

583

9.55

239

10.51

822

9.81

Building material board

213

3.49

164

7.21

377

4.50

Artificial lightweight aggregate

0

0.00

0

0.00

0

0.00

Concrete products

18

0.29

1

0.04

19

0.23

231

3.78

165

7.25

396

4.73

Fertiliser

53

0.87

26

1.14

79

0.94

Material for soil improvement

11

0.18

82

3.60

93

1.11

64

1.05

108

4.75

172

2.05

Sewage disposal agent

4

0.07

1

0.04

5

0.06

Steel manufacturing

13

0.21

8

0.35

21

0.25

Others

612

10.02

25

1.10

637

7.60

Total

629

10.30

34

1.49

663

7.91

Evective use total

6105

100.00

2275

100.00

8380

100.00

Total Pubric Works

Total Constraction

Total Agriculture, forestry and fisheries Total Others

TABLE 2 Main chemical and physical properties of fluidised coal combustion flyash.

Element/characteristics

Flyash class 1

Flyash class 2

Flyash class 3

Silicon dioxide %

Min 45.0

Moisture content %

Max 1.0

Loss on ignition %

Max 3.0

Max 5.0

3

Specific gravity g/cm

Fineness

Max 8.0

Max 5.0

Min 1.95 45 μm residue on sieve % (Mesh sieve method)

Max 10

Max 40

Max 40

Max 70

Specific surface area cm2/g (Blaine’s method)

Min 5000

Min 2500

Min 2500

Min 1500

Min 105

Min 95

Min 85

Min 75

28 days

Min 90

Min 80

Min 80

Min 60

91 days

Min 100

Min 90

Min 90

Min 70

Percent flow % Activity index %

Flyash class 4

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

237

T SASAOKA, H SHIMADA AND K MATSUI

FIG 4 - Coal ash utilisation by sector in 2006.

REQUIRED PROPERTIES OF BENTONITE BARRIER LAYER (CEG, 2010) The extent and orientation of research into bentonites is given by the unique requirements set by the area of its exploitation. Bentonite, or a bentonite-based material, will be used as the main composition of an engineered barrier in the underground repository, preventing any potential leakage of radio nuclides from the container with high-level radioactive waste into a natural barrier and further into the biosphere. The engineered barrier must retain this capability for a period of up to hundreds of thousands of years. Within the engineered barrier, bentonite-based mixture will fulfil absorbing, filling and sealing functions. Therefore, there are basic geotechnical requirements for the bentonitebased barrier.

Impermeability (filtration coefficient k = 10 - 10 and 10 - 14 m/s) The design of a bentonite-based material (mix), which will fulfil the required non-permeability parameters, does not represent the biggest problem. The material itself can fulfil this requirement without problems. The hazard of radio nuclide leakage, however, rapidly increases with the appearance of any discontinuity interface. Discontinuity interfaces are a potential source of formation of paths for the spread of dangerous radioactive substances in any state. Different types of interfaces may be distinguished, namely in relation to the way of their formation: x discontinuity interfaces arising during preparation of multi-barrier system, x discontinuity interfaces arising during multi-barrier system’s construction, or x discontinuity interfaces arising during long-term operation of underground repository.

Swelling capacity Swelling capacity of the used material is important namely due to the necessity of sealing discontinuity interfaces and/or cracks in their contact with groundwater self-sealing. Swelling capacity, described in geotechnics by the value of swelling pressure, should be optimised by admixtures. Swelling pressure must not negatively affect the function of the container, the function of individual structural units of the engineered barrier or the function of the natural barrier.

Thermal conductivity The bentonite-based barrier material must be designed in such a way to facilitate easy removal of the heat radiated by the container further into the natural barrier. Thermal conductivity grows with the growing volume density and material moisture content. It also shows a slight increase with growing temperature. In order to facilitate heat removal from the container, bentonite mixture is treated by adding graphite. Groundwater leaking from the natural barrier into the engineered barrier gradually saturates part of it, which increases the thermal conductivity in the saturated medium. Thermal conductivity of the material within the barrier body will show changes in time. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

238

STUDY ON UTILISATION OF FLYASH FOR BARRIER LAYER/BUFFER MATERIALS FOR RADIOACTIVE WASTE DISPOSAL

Extremely long-term unchangeability of bentonite-based barriers behaviour This requirement forms the most difficult part of research objectives. It is, however, evident that implementation of this research requires a multi-disciplinary approach with the use of all available methods. Such methods include namely experimental research, physical and mathematical modelling and a study into natural analogues. Input parameters for mathematical models may be obtained namely by using laboratory testing, onsite tests and field measurements, research in underground laboratories. In this respect it should be mentioned that the accuracy of obtained results requires a practice in which the tests and experiments are carried out under the conditions corresponding to actual conditions. This means, for example, that strength tests of prefabricates should be performed at temperatures of 70 - 140°C, the material should be subjected to long-term loading with this temperature before testing, or the thermal conductivity coefficient should be measured at this temperature and on a material saturated with water of a specific chemical composition under the conditions when it cannot change its volume, etc.

LABORATORY TESTS This research investigates how much impact of different substitute ratio of flyash for bentonite on the characteristics of bentonite-based barrier layer/buffer in order to discuss the applicability of flyash as the content of bentonite-based barrier layer/buffer for radioactive waste disposal. A series of laboratory tests were conducted as follows.

Strength test In the case of high-level radioactive waste disposal, the depth of is more than 300 m deep from the surface. At least 8.1 MPa in vertical direction and 6.5 MPa in horizontal direction stresses are affected as the ground pressure. Even though several kinds of support systems are installed or measures such as a grouting are conducted, the strength of bentonite-based mixtures itself has to be in some extents. Therefore, the strength test under different flyash-bentonite contents has been conducted in order to evaluate the applicability of flyash and investigate the appropriate mix content for applying barrier layer/buffer for radioactive waste disposal. The specimens were made with bentonite, flyash and water. Two different sizes of flyash were employed. Before moulding, all contents were dried and water was sprayed on them and then they were left for 36 hours. A cylindrical mould, 50 mm in diameter and 100 mm in length, was used for specimens. Mixtures which volume is 1/3 of total volume of mould are put into in mould and then was pounded by falling heavy weight by 20 times. This procedure is repeated three times. After that, the specimens were removed from moulds and shapes of both sides were restored. Figure 5 shows the specimens for this test. Uniaxial compressive test and needle penetration test were performed for each specimen. Table 3 shows the pattern of mixtures contents for specimens.

FIG 5 - Specimens for uniaxial compressive test. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

239

T SASAOKA, H SHIMADA AND K MATSUI

TABLE 3 Patterns of mixtures contents for specimens. No

Bentonite (wt%)

Flyash (wt%)

Water Content (%)

1. Bentonite only

100

0

5

2. Bentonite + Flyash (raw)

80

20

5

3. Bentonite + Flyash (raw)

70

30

5

4. Bentonite + Flyash (raw)

60

40

5

5. Bentonite + Flyash (crushing)

80

20

5

6. Bentonite + Flyash (crushing)

70

30

5

7. Bentonite + Flyash (crushing)

60

40

5

Figure 6 shows the uniaxial compressive strengths under different mixture contents. It can be seen that the substitution of flyash for bentonite improve the strength of bentonite-based mixtures and the strength of bentonite-flyash mixtures increases with increasing its substitution ratio. The strength of bentonite-flyash mixtures are from 0.8 MPa to 2.2 MPa and this range is almost the same as the soil around 300 m deep from the surface and meets the required properties. Moreover, depending on the site conditions, the strength of bentonite-based mixtures can be controlled as the required level by substituting flyash for bentonite. In additions, as the particle size/distribution of flyash has no obvious impact on the mechanical properties especially UCS of bentonite-flyash mixtures, it can also be said that the original flyash can be used and cost can be saved.

FIG 6 - Uniaxial compressive strengths under different mixture contents.

Falling head permeability test The characteristics of permeability of bentonite-flyash mixtures is also one of the important key for barrier layer/buffer for radioactive waste disposal in order to prevent immersed water to overpack from surrounding soil/groundwater and leak the radioactive materials from its inside. The falling head permeability test was conducted. Figure 7 shows the equipment of falling head permeability test. Based on the results of strength tests, two different contents of bentonite-flyash mixtures were selected and tested. Table 4 shows the results of this permeability test. It can be said from this table that the permeability of bentonite-flyash mixtures increases with increasing the substitution ratio of flyash for bentonite. In other words, its barrier/sealing function decreases with increasing the substitution ratio of flyash for bentonite. However, even if the substitution ratio of flyash for MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

240

STUDY ON UTILISATION OF FLYASH FOR BARRIER LAYER/BUFFER MATERIALS FOR RADIOACTIVE WASTE DISPOSAL

FIG 7 - Device of falling head permeability test. TABLE 4 Results of permeability test. No

Permeability k15 m/3

1. Bentonite only

7.4 x 10-10

2. Bentonite: Flyash (raw) = 1:1

2.0 x 10-8

3. Bentonite: Flyash (crushing) = 1:1

1.0 x 10-8

bentonite is 50 per cent, the permeability of bentonite-flyash mixtures is still around 1.0 × 10-8 m/sec and this value can be considered as impermeable in practically from the engineering point of view. Moreover, it can be expected that the permeability of bentonito-flyash mixture at 300 m deep from the surface is lower than the value obtained from this test due to the consolidation of bentonite-flyash mixtures by large ground pressure. Hence, it can be expected that the benonite-based mixture meets the required impermeability even though a bentonite is substituted with flyash in some extents.

Swelling test As mentioned above, swelling capacity is important due to the necessity of sealing discontinuity interfaces and/or cracks in their contact with groundwater self-sealing. Here, swelling capacity described by the value of swelling volume. Figure 8 shows the equipment of swelling test. Table 5 shows the results of swelling test. It can be said from this table that the swelling volume decreases with increasing substitution ratio of flyash for bentonite. Moreover, the particle size of flyash has no impact on the swelling characteristics of bentonite-flyash mixtures. This is because only bentonite has swelling characteristics and flyash does not have. Hence, it can be said that the barrier layer/ buffer material has to contain bentonite in some extent in order to have the swelling capacity.

pH value and electro conductivity measurement In the case of high-level radioactive waste disposal, bentonite-flyash buffer contacts directly with the canisters and surrounding rock, not only its sealing characteristics of radioactive waste materials but also the impact/effect of buffer itself on surrounding rock/environment has to be investigated. For example, the chemical reaction between buffer and surrounding rock and the leakage of contaminated material through the buffer by underground water flow, etc. Hence, pH value and MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

241

T SASAOKA, H SHIMADA AND K MATSUI

FIG 8 - Device of swelling test. TABLE 5 Results of swelling test. Specimen no

Original height (mm)

Vertical displacement (mm)

Swelling ratio (%)

Bentonite only

36

5.64

15.7

Bentonite: Flyash (raw) = 1:1

46

4.63

10.1

Bentonite: Flyash (crushing) = 1:1

37.5

3.88

10.3

Bentonite: Flyash (raw) = 1:2

39

3.93

10.1

Bentonite: Flyash (raw) = 1:5

40

4.21

10.5

Bentonite: Flyash (crushing) = 1:5

38

4.02

10.6

electro-conductivity of bentonite-flyash mixtures were measured under different substitution ratio of flyash for bentonite. Test procedures are as follows: the 100 ml of distilled water and 2.0 g of each sample put into a beaker. Then the top clear layer liquid was sampled and a couple of drop was put on the both sensors of pH and electro-conductivity. Table 6 shows the composition of test specimens. Figures 9 and 10 show the relationship between the pH value/electro-conductivity and elapsed time, respectively. The suspension of bentonite itself is classified as weak alkaline and the pH value of that of flyash is 12 - 13 and this is classified as alkaline. Compared with these results, no obvious change of pH value and electoro-conductivity due to the chemical reaction among bentonite, flyash and water was observed under different composition ratios. However, as all of these samples represents alkaline, if underground flow has any impact of storage area, the application of neutraliser or other measures should be considered in order to restrain the impact of alkaline on the surrounding environment. TABLE 6 Compositions of test specimens. 1

Bentonite only

2g

2

Flyash (raw)

2g

3

Flyash (crushing)

2g

4

Bentonite: Flyash (raw) = 1:1

1 g, 1 g

5

Bentonite: Flyash (crushing) = 1:1

1 g,1 g

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

242

STUDY ON UTILISATION OF FLYASH FOR BARRIER LAYER/BUFFER MATERIALS FOR RADIOACTIVE WASTE DISPOSAL

FIG 9 - Relationship between pH value and elapsed time.

FIG 10 - Relationship between electro conductivity and elapsed time.

Thermal conductivity test After the radioactive waste materials are stored, they generate heat and then the temperature of its inside rises gradually. Under these situations, the deterioration of buffer can be expected due to high temperature. In order to prevent this deterioration, the thermal conductivity of bentonite-flyash mixtures has to be investigated. Low thermal conductivity prevents the diffusion of heat generated by radioactive waste materials to outside effectively and as a result the characteristics of buffer such as impermeability, thermal transfer, radionuclide transport, stress relaxation may be weaken due to the high temperature. Hence, the thermal conductivity is also one of the important characteristics of buffer material. In this research, a thermal conductivity of each specimen was measured by using the thermal conductivity metre QTM-500 (see Figure 11) and the impact of the different composition of flyash-bentonite on the thermal conductivity was discussed. Table 7 shows the compositions of test specimens. Figure 12 shows the results of this test. It can be seen that the thermal conductivity increases with increasing substitution ratio of flyash for bentonite. Moreover, the range of thermal conductivity is 0.3 ~1.5 W/mK and all of these values meet a required conditions proposed as the buffer for high-level radioactive waste. Hence, the substitution of flyash for bentonite improves the thermal conductivity of barrier layer/buffer. However, in this test, the impact of rising temperature due to the MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

243

T SASAOKA, H SHIMADA AND K MATSUI

FIG 11 -Thermal conductivity metre QTM-500.

(A)

(B)

FIG 12 - Thermal conductivity for each sample ((A) three days later; (B) ten days later).

heat generated by radioactive wastes on the thermal conductivity of bentonite-flyash mixtures was not taken into account. The effect of high temperatures and elapsed long time should be investigated in the next.

CONCLUSIONS The applicability of flyash to the contents of barrier layer/buffer for radioactive wastes was investigated in this research. From the results of a series of laboratory tests, it can be concluded that MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

244

STUDY ON UTILISATION OF FLYASH FOR BARRIER LAYER/BUFFER MATERIALS FOR RADIOACTIVE WASTE DISPOSAL

TABLE 7 Compositions of test specimens. No

Bentonite (wt%)

Flyash (wt%)

Water content (%)

1. Bentonite only

100

0

5

2. Bentonite + flyash (raw)

80

20

5

3. Bentonite + flyash (raw)

70

30

5

4. Bentonite + flyash (raw)

60

40

5

5. Bentonite + flyash (crushing)

80

20

5

6. Bentonite + flyash (crushing)

70

30

5

7. Bentonite + flyash (crushing)

60

40

5

flyash has a great potential for use in them as the bentonite is substituted. However, in order to prove the ability and estimate the appropriate mixture contents, more research has to be conducted, such as colloid filtration effect and long-term stability/unchangeability.

REFERENCES Berlin, R E and Stanton, C C, 1989. Radioactive Waste Management, 444 p (Wiley-Interscience). Brookins, D G, 1984. Geotechnical Aspects of Radioactive Waste Disposal, 321 p (Springer: Verlag). Byalko, A V, 1994. Nuclear Waste Disposal: Geophysical Safety, 281 p (CRC Press). Centre of Experimental Geotechnics (CEG), 2010. Research [online]. Available from:. Nuclear Waste Management Organisation of Japan, 2010. Available from:. United States Nuclear Regulatory Commission, 2010. News releases and speeches [online]. Available from:. Yoshida, Y, Shimada, H, Sasaoka, T, Matsui, K, Nakagawa, H, Sakai, Y and Gottfried, J, 2009. Consumption of coal and utilisation of coal ash in Japan, in Proceedings 13th Conference on Environment and Mineral Processing, 1:291-301.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

245

Fundamental Study of Acid Drainage Control using Flyash H Shimada1, T Sasaoka2, K Matsui3, G J Kusuma4, J Oya5, H Takamoto6, S Kramadibrata7 and B Sulistianto8 ABSTRACT Japan is the largest coal importer in the world, importing from Australia, China, Indonesia, Canada, Vietnam, etc. In Indonesia, most of the produced coal is extracted from open cut mines. More open cut coal mines will be developed to fill the demand for coal in Indonesia and the world. Many surveys and projects for developing new coal mines and transferring mining technology are being conducted in tropical counties such as Indonesia and Vietnam with the aid of the Japanese Government. It has, however, been pointed out that excessive and precarious development of mines in tropical areas will damage the environments around the mines including cultivated lands, and reduce the precious rainforest. This would undoubtedly exacerbate the instability of the global climate and the food crisis in the world. This paper describes the effects of flyash on the control of acid mine drainage in order to avoid potential negative impacts on the environment caused from open cut coal mines.

INTRODUCTION About 200 Mt of coal is imported into Japan annually and Indonesia was its second largest exporter, accounting for about 30 Mt. Indonesia produced about 200 Mt of coal, about 75 per cent of which was exported in 2008. Over 99 per cent of the coal produced in Indonesia comes from open cut mines. With this, Japan is the largest coal importer in the world. It is projected that more open cut coal mines will be created to fill the demand for coal in Indonesia and the world. Many surveys and projects for developing new coal mines and transferring mining technology are being carried out in Indonesia with the aid of the Japanese Government. It has, however, been pointed out that excessive and precarious development of mines in tropical areas will damage the environments around the mines and reduce the precious rainforest, which would undoubtedly lead to further instability of the global climate. In addition, a large amount of waste from the preparation plant is produced. The properties of this waste vary depending on the mineralogical contents of the mother rock in which the coal is embedded. The waste quality depends on the method of mining and cleaning. The waste mainly consists of clay, quartz, carbonaceous materials, mica, pyrite, and so on. Now, about 85 per cent of flyash is utilised and the remains are disposed of at designated disposal sites, however, the life of these disposal sites is limited. Moreover, it is difficult to find new disposal sites. Therefore, it is requested that the percentage of coal ash being utilised be increased in every field. This paper describes some methods to generate acid mine drainage and discusses the effects of flyash on the control of acid mine drainage in order to avoid potential environmental damage caused by open cut coal mines in tropical areas. 1. Associate Professor, Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan. Email: [email protected] 2. Assistant Professor, Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan. Email: [email protected] 3. Professor, Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan. Email: [email protected] 4. Doctoral Student, Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan. Email: [email protected] 5. Mining Engineer, MMI Coal Tech Co Ltd, 1-7-7 Hachobori, Chuo-ku, Tokyo 104-0032, Japan. Email: [email protected] 6. Mining Engineer, MMI Coal Tech Co Ltd, 1-7-7 Hachobori, Chuo-ku, Tokyo 104-0032, Japan. Email: [email protected] 7. Senior Lecturer, Institute Technology Bandung, Jl, Ganesha 10, Bandung 40132, Indonesia. Email: [email protected] 8. Lecturer, Institute Technology Bandung, Jl Ganesha 10, Bandung 40132, Indonesia. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

247

H SHIMADA et al

CONTROL OF ACID MINE DRAINAGE Acid mine drainage Acid mine drainage (AMD) is low pH water with dissolved metals and sulfate that drains from pyriterich rocks which have been exposed to oxidation. Pyrite is commonly associated with coal deposits, and its oxidation upon mining, results in the production of sulfuric acid, a significant environmental issue. Discharge of acid drainage into streams, rivers and lakes can adversely impact downstream aquatic ecosystems, and as such is an environmental threat. The time required for AMD to develop can vary from a matter of days to many years, and it may last for many years as well. Unlike industrial pollution, AMD has the potential to continue indefinitely, well past the cessation of mining. AMD is quite possibly the most significant environmental management issue linked to open cut coal mine. Coal mining produces rock by-products such as claystone, siltstone, shale and mudstone, which contain variable amounts of naturally occurring sulfides, mainly in the form of pyrite. If sulfides are exposed to air (oxygen) and water, they will oxidise, producing acid. Within undisturbed rock, this process occurs very slowly. However, within a waste rock dump, the rate of such a reaction can be much faster due to the greatly increased surface area of the broken rock. Under these circumstances, any water leaching from the waste rock dump will be acid, and may cause environmental damage. A slaking behaviour of some coal by-product rocks accelerates oxidisation. Apparently, even strong rocks show severe slaking behaviour once becoming wet. If oxidisation is not successfully controlled, AMD from a waste dump may persist for an excess of years. The best approach to minimise the risk of AMD from waste rock dumps is to control the entry of air into the dumps, in effect reducing the oxidation rate of the sulfides. This is achieved by constructing a cover over the dump made from either clay or non-acid forming (NAF) waste rock as shown in Figures 1a and 1b.

(A)

(B)

FIG 1 - The function of a waste dump cover to control acid mine drainage. (A) The conceptual view of sulfuric waste dump without a cover. Water and oxygen easily enter the dumping materials and react with sulfides (pyrite). Resulting in acid mine drainage. (B) The conceptual view of a sulfuric waste dump with a cover. The cover can limit the entry of oxygen and water and reduce the rate of the acid mine drainage.

Net acid generation (NAG) test The different types of waste rock in current and proposed pit areas are identified through drill holes using downhole geophysics and logging of core or drill cuttings. At Kaltim Prima Coal (KPC) in East Kalimantan, Indonesia, NAG testing has identified two main groups of AMD types for the waste rock being mined 1. non-acid forming (NAF), and 2. potentially acid forming (PAF). Identifying where the waste is in the pit and when it will be mined is the critical issue for effective waste management. Figure 2 shows the results of the NAG test using cuttings from blast holes at Pit J (Ikematsu et al, 2005). From these test results on Pit J, it is estimated that about 80 per cent of overburden is NAF (types 1 and 2) and about 20 per cent is PAF (types 3a, 3b and 4) at Pit J. Therefore, DC03 method is employed as a dump cover in this area. The dump cover system will be discussed later. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

248

FUNDAMENTAL STUDY OF ACID DRAINAGE CONTROL USING FLYASH

FIG 2 - Results of the net acid generation test at Kaltim Prima Coal.

Slaking behaviour Some coal by-product rocks such as claystone, siltstone, shale, and mudstone show slaking behaviour (Ichinose, 1987 Sadisun, 2004). Disintegration due to slaking increases the oxidisation process of pyrite because of increased specific area of fragmented debris. Wet and dry cyclic conditions as those found in tropical climates, severely accelerate slaking behaviour. Cracks due to slaking, increases permeability in dump and softened rocks, reducing the stability of the dump wall and increasing damage caused by erosion at rehabilitated areas. To investigate the possibility of slaking, the quantitative test of equivalent smectite was carried out on the KPC waste rocks using the methylene-blue adsorption test (Ichinose, 1987). Ichinose concluded that if smectite is present in excess of six per cent, the slaking behaviour becomes severe. Figure 3 shows the results of the test using borehole cuttings (Ikematsu et al, 2005). All values for the rock samples have over six per cent smectite meaning that these rocks are capable of severe slaking behaviour.

FIG 3 - Amount of smectite in the borehole cuttings.

Waste dump construction The current cover option adopted at KPC is the placement of PAF waste in the bulk of the dump and construction of a cover using NAF waste material utilising one of the three following cover design options MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

249

H SHIMADA et al

1. 1 m compacted clay cover (DC01) – a 1 m thick layer of compacted clay overlain by a layer (usually 2 m) of loose NAF rocks, 2. 2 m compacted NAF cover (DC02) – a 2 m thick layer of compacted NAF rock overlain by a layer (usually 2 m) of loose NAF rock, and 3. Loose NAF cover (DC03) – a thick layer (10 m to 20 m) of uncompacted NAF rock. The loose NAF cover is preferred in principle, since it is more cost-effective, has low risk of erosion, and is immune to dump settlement, which may crack a compacted cover. However, it requires a large amount of run-of-mine NAF rocks (KPC, 2000). In DC01 and DC02 covers, the function of the compacted layer is to form a barrier preventing oxygen diffusion, and thus generation of AMD. The overlain, uncompacted waste rock layer protects the compacted layer from erosion and stores water allowing for revegetation. Controlling acid mine drainage really depends on dump construction. It is necessary to specify the steps for placing NAF waste and PAF waste within the dump so that the potential for acid mine drainage is minimised. However, if the PAF waste deteriorates excessively by slaking, seepage water from rainfall permeates to a PAF layer, and there remains a possibility that AMD may be generated as a result. In particular, coal by-product rocks in Indonesia show remarkable slaking behaviour, as mentioned before (Matsui et al, 2004). From this perspective, the use of a column test elicits the effects of flyash on the control of AMD in order to avoid potential environmental damage caused by open cut coal mines.

EFFECT OF FLYASH ON CONTROL OF ACID MINE DRAINAGE Samples and procedures of column test It is integral that the acid components leached from pyrite are finally neutralised by their reaction to flyash. Therefore, it is necessary to investigate the optimum combination and the filling method of pyrite and flyash. With this in mind, a water quality analysis of eluted water was conducted by a column test using the samples of mixed pyrite and flyash. Figure 4 illustrates the results of the column test. In this test, a cylindrical pipe 100 bmm in diameter and 1000 mm in length was filled with the sample. Different combinations and filling methods of pyrite and flyash were prepared as listed in Table 1. Cases 1 and 4 were solely consistent of pyrite and flyash, respectively. Cases 2 and 3 were adopted as a different kind of filling method. Two thousand, two hundred and fifty millilitres of ion exchange water was poured from the upper part of the column for 12 hours at a rate of 3 mL/min, then, seepage water was collected in the container at the lower part of the column. After the seepage water was finished discharging from the column, the sample was kept as it was for 12 hours because the sample was exposed to the air. Next, 2250 mL

Flyash

Ion exchange water

Pyrite Mixture with Flyash and Pyrite

Case 1

Case 2

Case 3

Case 4

FIG 4 - Illustration of the column test. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

250

FUNDAMENTAL STUDY OF ACID DRAINAGE CONTROL USING FLYASH

TABLE 1 Samples and filling methods. Pyrite (g)

Flyash (g)

The filling method

Case 1

1500

0

Case 2

1500

1500

Stratified

Case 3

1500

1500

Homogeneous mixing

Case 4

0

1500 TABLE 2 Chemical components of flyash.

(%)

SiO2

AI2O3

Fe2O3

CaO

MgO

K2O

Na2O

66.25

18.11

4.87

3.96

0.051

0.27

0.92

of ion exchange water was poured into the column again where the sample reacted to the water for 12 hours. In this case, the sample was not replaced. Again, the sample was kept as it was for 12 hours. These procedures were carried out a maximum of nine times. The pyrite was a product of Furukusa, Aichi Prefecture, Japan, and was crushed to 0.35 - 2.0 mm. The flyash was a product from Matsuura power plant, Saga Prefecture, Japan, and was crushed to the grain size of 2.0 - 5.0 mm. Table 2 shows the chemical components of flyash. The pH and the qualities of iron and aluminum were measured by pH meter and an inductively coupled plasma atomic emission spectrometer (ICP-AES), respectively. Moreover, the quantities of Na+, K+, Ca2+, Mg2+, SO42-, and Cl- were measured by the ion chromatograph method.

pH – concentration and precipitates Table 3 lists the results obtained from the water quality analysis. The colour of the seepage water from the sample of case 1 at the first discharging was light green and there was a black precipitate that was FeS at the bottom of the column. The seepage water in case 1 at the second to the final discharges, was colourless. The colour of the seepage water from case 2 at the first discharging also showed a light green hue, and there was FeS sediment at the bottom of the column. The colour at the third discharging was colourless and there was white sediment at the bottom of the column. With case 3, dark brown water was discharged from the column at the first seepage and there was dark brown precipitate, specifically Fe(OH)3 at the bottom of a column. Conversely, colourless water was discharged from the sample of case 4 at the first seepage. Although 7 mg/L of chloride ion was only detected in the seepage water of the first discharging of cases 2 and 3, as compared with other components, the concentration of the chloride ion was very low and was not detected in the seepage water of the second discharging or subsequent discharges. It is thought that the influence of the pH on the chloride ion in the seepage water can be disregarded. In case 1 with only pyrite, the main component of the seepage water was the sulfate ion, due to the elution by oxidation of the pyrite. Since the pH of the seepage water was 3.4, it is considered that the component which exhibited acidity in the seepage water was the sulfate ion. As the number of times discharging was increased, all ion concentrations in the seepage water had correspondingly decreased in case 2. The main components of the first and the fifth discharging were iron and sulfate ions, respectively, but the sulfate iron was only remnant in the seepage water of the eighth discharging. In case 3, ion concentrations were greater than those in case 2. Since each ion concentration decreased with an increase in the number of discharges, the same trend as in case 2 was observed. In case 4 containing only flyash, each ion concentration with the exception of the magnesium ion, was comparable to the other cases. Moreover, the pH was 11.2 and it is hypothesised that MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

251

H SHIMADA et al

TABLE 3 Ion concentrations obtained by water quality analysis (unit: mg/L). Case 1 The fifth time

Case 2 The first time

Case 2 The fifth time

Case 2 The eighth time

3.4

1.6

10.3

10.0

Na

0.4

52

5.8

2.1

+

Less than 0.4

140

25

8.0

Ca2+

0.8

300

57

31

0.34

180

0.58

Less than 0.02

Less than 1.0

7.0

Less than 1.0

Less than 1.0

SO4

85

46 000

360

110

Fe

34

24 000

140

0.80

Al

0.07

70

1.0

0.63

Case 3 The first time

Case 3 The fifth time

Case 3 The nineth time

Case 4 The fifth time

3.9

5.9

6.2

11.2

Na

74

8.0

2.3

4.8

+

200

9.0

2.3

4.8

470

440

320

8.9

170

36

13

0.04

7.0

Less than 1.0

Less than 1.0

Less than 1.0

Item pH +

K

Mg

2+

-

CI

2-

Item pH +

K

2+

Ca

2+

Mg CI

2-

SO4

11 000

1800

840

27

Fe

4600

340

41

2.5

Al

9.9

1.3

1.1

1.4

the components which presented the alkalinity eluted from the flyash ash were mainly sodium, potassium, and calcium ions.

Neutralising mechanism of acid water From the results obtained from the water quality analysis, the acid neutralising mechanism of case 2 and case 3 was discussed. First, in the seepage water of case 2, since the sample was filled up with the flyash and the pyrite in layers, it was predicted that the elution of various ions would occur separately in each layer. As shown in Table 4, when comparing with the sum of the various ions of cases 1 and 4 referring to the fifth discharge and the various ions of case 2, it turns out that both are extremely similar. Therefore, when filling up in layers, like in case 2, the hydroxide ion which was eluted from the flyash layer permeated the pyrite layer. After which, the hydroxide ion was neutralised by the hydrogen ion eluted from the pyrite layer. Conversely, in case 3, even TABLE 4 Ion concentrations (the fifth time) (unit: mg/L). Item

Case 1

Case 4

Case 1+4

Case 2

Na

0.4

4.8

5.2

5.8

+

8.0

17

25

25

+

K

2+

Ca

0.8

8.9

9.7

25

Mg2+

0.34

0.04

0.38

0.58

SO42+

85

27

112

360

Fe

34

2.5

36.5

140

Al

0.07

1.4

1.47

1.0

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

252

FUNDAMENTAL STUDY OF ACID DRAINAGE CONTROL USING FLYASH

if various ion concentrations increased the number of discharges greatly, as that with case 2, the iron and sulfate ions of the pyrite origin still exist. Moreover, since iron hydroxide precipitate was observed in the seepage water, the interaction by homogeneous mixing with the flyash and the pyrite was recognised. That is, the Fe2+ eluted from the pyrite as shown in Equation 1, oxidised to Fe3+ as a result of contact with the flyash and dissolved oxygen as seen in Equation 2. Moreover, oxidation of the pyrite was promoted as is seen in Equation 3, by existence of Fe3+. Therefore, the acid neutralising effect of acid water was reduced due to increasing the hydrogen ion within the column by repeating the reaction of Equations 1 - 3. FeS2 + 7/2O2 + H2O " Fe2+ + 2SO42- + 2H+

(1)

Fe2+ + 1/4O2 + H+ " Fe3+ + 1/2H2O

(2)

FeS2 + 14Fe3+ + 8H2O " 15Fe2+ + 2SO42- + 16H+

(3)

In order to utilise flyash as the neutralising material in rehabilitation, it is suggested that the construction method which covers the flyash is best suited for the upper part of coal by-product rock containing pyrite.

Optimum condition of flyash cover From the results obtained from the water quality analysis, it is suggested that the construction method which covers the flyash is best suited for the upper part of coal by-product rock containing pyrite. The central and regional governments have released the regulations regarding with water quality for coal mining activity in Indonesia. They are Decree of the State Minister for the Environment Number 113/2003 about Quality Standard of Wastewater for Coal Mining Business and or Activities and Decree of the Government of East Kalimantan Number 26/2002 regarding Liquid Waste Standard for Industrial Activity and Other Activity in the Province of East Kalimantan. Table 5 shows the threshold limit values for mining industry. The differences of the threshold limit value between central and regional regulations are caused by determination of the usage of water receiving environment that more specific considered at regional government. From this table, the optimum mixing percentage of flyash with pyrite was discussed that pH 6 - 9 of the seepage water can be discharged. TABLE 5 Threshold limit value. Threshold limit value Parameter

Unit

pH

Decree of the State Minister for the Environment Number 113/2003

Decree of the Government of East Kalimantan Number 26/2002

6-9

6-9

Total suspended solid

mg/L

400

400

Fe

mg/L

7

10

Mn

mg/L

4

5

Figure 5 shows the relation between the mixing percentage of flyash and pH. From this figure, if 30 - 40 per cent of flyash is covered onto pyrite, the seepage water can be controlled the extent of pH 6 - 9. Moreover, as the percentage of pyrite in the waste rock is four per cent or less, smaller amount of flyash can be controlled AMD.

CONCLUSIONS In this paper, the effects of flyash on the control of acid mine drainage in order to avoid potential environmental damage caused by open cut coal mines in tropical areas were discussed. In order to utilise flyash as the neutralising material in rehabilitation, it is suggested that the construction method which covers the flyash is best suited for the upper part of coal by-product rock containing pyrite. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

253

H SHIMADA et al

FIG 5 - Relation between mixing percentage of flyash and pH.

REFERENCES hinose, M, 1987. A study of the effect of water content on mechanical properties of coal measures rocks, Doctor of Engineering Dissertation, Kyushu University, Japan, 162 p (in Japanese). kematsu, K, Matsui, K, Shimada, H, Ichinose, M, Gottfried, J, Kramadibrata, S and Nugraha, C, 2005. Geochemical test of coal measures rocks at KPC, Indonesia, in Proceedings Annual Meeting of MMIJ, pp 33-34 (in Japanese). altim rima Coal, 2000. Rehabilitation specifications version 2, internal report, Department Enviro, PT Kaltim Prima Coal, 52 p. Matsui, K, Shimada, H, Itoi, R, Ueda, T, abuki, A, Sadisun, I A, Sulistianto, B, Kramadibrata, S and Rai, M A, 2004. Development of opencut coal mines considering Indonesian environments, in Proceedings Eighth International Symposium on En ironmental Issues and aste Management in Energy and Mineral Production, Antalya, Turkey, pp 157-162. Sadisun, I A, 2004. Comprehensive study on slaking of argillaceous rocks by means of static slaking tests and its roles in engineering design, Doctor of Engineering Dissertation, Kyushu University, Japan, 142 p.

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

254

Simulation for Dependable Mining Automation N Hillier1 and J Ryde2 ABSTRACT A discussion on simulation tools to aid the development of robotics platforms is presented. The focus of the work is towards mining robotics platforms where operations are typically in unstructured, unpredictable environments and a high degree of dependability in their operations is required. It is argued that a test-based development approach leads to more robust solutions and faster development cycles than traditional single-track field-only development and that the key enabling technology for this development paradigm is the judicious use of simulation tools. The future direction of simulation for use in online applications such as active planning and as a means to assist tele-operation in high latency communication applications is also briefly considered together with the requirements for a useful simulation environment in terms of the simulator architecture. Finally, a case study is presented on the application of the discussed methodologies to an unmanned ground vehicle (UGV) for earth-moving tasks, which is being developed in an accelerated time frame with a transient development team.

INTRODUCTION Mining automation is a difficult domain for robotics development, with significant challenges in development and operation due to the unstructured environments (both geometrically and in material properties), common use of multiple sensing modes and the high degrees of freedom inherent in most experimental mining robotics3 platforms. Additionally, there are pragmatic difficulties associated with the development of mining robotics platforms stemming from long set-up and shut-down times (pre- and post-start checks) and access to suitable test sites that do not occur with most indoor and small platform robotics. These issues are further compounded by those applications wherein dependability of the robot is essential due to any number of commercial, remote deployment and/or risk constraints. This paper argues that a successful development paradigm for mining robotics includes significant use of simulation environments and that certain aspects of the simulation engine architecture allow for more rapid and flexible development of the simulation itself. For clarity, the discussion here is typically constrained to the simulation of robotics in a full three-dimensional (3D) environment. Furthermore, the authors consider that robotics is a singular field in the engineering disciplines in that testing on the developed product is currently by far the predominant means of development. Most (if not all) other engineering fields involve significantly more extensive use of simulation or numerical modelling tools before prototype and production testing. There are many reasons for the slow uptake of such tools in robotics, not the least being that the popular approach of replaying log files is sufficient for the development and testing of many robotics algorithms and that the appropriate simulation tools are only in their infancy. The work in Weatherly et al (2006) and Grabowski et al (2006) details the software architecture and core development processes used in the development of a 2005 DARPA Grand Challenge entrant (the MITRE Meteor). The 2005 Grand Challenge was a DARPA (United States of America’s Defence Advanced Research Projects Agency) sponsored event requiring autonomous off-road vehicles to 1. Research Scientist, CSIRO, PO Box 883, Kenmore Qld 4069. Email: [email protected] 2. Post-Doctoral Researcher, CSIRO, PO Box 883, Kenmore Qld 4069. Email: [email protected] 3. In this work, the authors make a distinction between mining automation and mining robotics, in that robotics includes some element of mobile actuation (such as a robotic arm,

or an automated vehicle), whilst automation covers a wider field, including robotics but also process automation and the like. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

257

N HILLIER AND J RYDE

travel a 132-mile desert route in under ten hours without any external assistance. These works highlight a number of advantages and also a significant caveat to the use of simulation heavy field robotics development and so a summary is presented here to introduce some key concepts. The fundamental restrictions based upon this team were: x accelerated development (concept to deployment was approximately 11 months), x no previous Grand Challenge experience, x no immediately local off-road test site (the authors claim that in ten months, only ten days of testing was conducted outside of a parking lot), and x a relatively small development team (five core members). The team chose to use a simulation heavy development cycle including full ten hour race simulations to test software robustness as well as data-replay from limited off-road real-world testing. Both the simulated and replayed data interfaces were transparent to the code operating the real robot, a concept that can readily be achieved through the use of modern robotics middleware (eg the Robotic Operating System (ROS) (Quigley et al, 2009), Dynamic Data Exchange (DDX) (Corke et al, 2004) and Player/Stage (Gerky, Vaughan and Howard, 2003)). The software was also transparent to realtime and simulation-time concepts to overcome flexible computation availability during simulation for faster and slower than real-time execution without timing problems. Ultimately, the Mitre Meteor DARPA Grand Challenge entry passed all the qualification rounds to start the final robot race (only 23 of the original 195 entrants started the desert race); however, it ultimately failed due to misinterpretation of dust as an obstacle. This highlights a fundamental issue for simulation driven development: the question of the appropriate level of simulation fidelity and the simulation versus field testing trade-off. The MITRE team described the increase in model fidelity due to the incorporation of field testing data as a ‘lasting repository of project experience’, and the take away message is that simulation development and testing alone is insufficient for practical applications; however, its judicious use can significantly speed up the development cycle for field-based robotics.

BRIEF SUMMARY OF THE STATE-OF-THE-ART IN ROBOTICS SIMULATION Simulators The robotics community has recently seen rapid development of a number of capable 3D, physics enabled simulation platforms. Beyond the ever-growing list of custom simulation solutions that institutions are developing for their own one-off systems (for example the abundance of simulators for the RoboCup four legged league), those that the authors’ consider to be the most useful or seem to be currently the most popular for robotics research are very briefly described in Table 1. Amongst the numerous offerings available, the OpenRAVE platform is currently the tool of choice for the authors. Features that are critical to the authors’ work, provided by this solution, which stand out from the others include: x Simple, flexible interfaces in a variety of languages (C++, Octave/Matlab and Python). x A dynamically loadable plug-in type architecture that allows for significant customisation of very low-level interfaces (such as the employed physics engine and collision checking) and ease of integration into middleware (comes with interfaces to ROS, and the authors trivially added support to DDX as required). x A modular design providing an opportunity for a light-weight customised version to be run live on platforms with less processing power via replacement of almost any component. x Platform independence (support for Windows, MacOS and Linux) with the transparency of open source code. x The ability to dynamically create, destroy and change the geometric properties of items in the simulation environment. This is a particularly powerful option for operations that manipulate the environment. Furthermore, the OpenRAVE environment allows for environment cloning, a powerful feature that could be exploited for more advanced simulation tasks such as parallel execution for run-time planning tasks, although the authors are yet to exploit this feature. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

258

SIMULATION FOR DEPENDABLE MINING AUTOMATION

TABLE 1 Summary of key (comparable) simulator features of some more popular robotics simulation packages. Simulator

Employed Physics

Source

Platform

Gazebo (Koenig and Howard, 2004)

ODE

Open

Linux

MS VSE (http://msdn.microsoft.com/en-us/ robotics/default)

PhysX

Closed

Windows

Unity (http://unity3d.com)

PhysX

Closed

Windows and MacOS

Unreal Engine

Sim open, Engine closed

Windows, MacOS and Linux

Any (ODE by default)

Open

Windows, MacOS and Linux

Various custom algorithms

Open

Windows and Linux

ODE (customisable in ‘Pro’ version)

Closed

Windows, MacOS and Linux

OpenSimulator (http://www.opensimulator.org)

ODE (+ others)

Open

Windows, MacOS and Linux

Matlab and similar tools (http://mathworks.com)

Custom algorithms

Closed

Windows, MacOS and Linux

USARSim (Carpin et al, 2007; Zaratti, Fratarcangeli and Locchi, 2007) OpenRAVE (Diankov and Kuffner, 2008) OpenHRP (Kanehiro, Hirukawa and Kajita, 2004) Webots (Michel, 2004)

Applications in automation The focus of this paper is on the use of simulators for algorithm and system development, by emulating a robotic platform and its environment in a virtual world. Although this is probably the most well known application of simulation tools in robotics, there are some other notable applications that may not be as widely known. The competitive robotics industry has adopted simulation use wholeheartedly, with a wide range of robotics competitions sporting either a simulation component, alternative stream (eg RoboCup’s simulation leagues) or having no real-world component at all (eg Rat’s Life is a pure simulation competition, although the equivalent hardware can be purchased). This is extremely beneficial to the robotics community as simulation competitions significantly reduce the financial barriers to entry into the industry by removing the associated costs and skills required to purchase and maintain robotic hardware and supporting infrastructure. The advent and rise of augmented reality interfaces (Milgram and Kishino, 1994) is boosting work in the area that some refer to as augmented virtuality, a field which is particularly prevalent in teleoperation applications with large latencies. In these applications, consistency of the user experience is provided through a virtual interface consisting of a GUI and a simulation back-end that synchronises with the real-world when possible. The simulation engine is then used to extrapolate the system state and present the user with a real-time interface. The use of simulation in robotics development is not restricted to software-only manifestations. Hardware-in-the-loop (HIL) simulation has been used extensively in many industries as a developmental and debugging tool. In robotics, it is often used as part of a staged development cycle, whereby sensor performance, timing issues or control demands require critical testing before deployment. HIL simulation often acts as a substitute for sensory information, particularly where an appropriate real-world test environment is difficult to achieve (eg off-world Rupp et al, 2009) or is deemed high-risk in the case of potential failure (eg aerial systems). Simulators are now also being increasingly used in a hybrid manner to artificially increase robotic populations. A typical example of this is in the field of self-reconfigurable (or modular) robotics, whereby one may wish to evaluate system performance of many tens, hundreds or thousands of components, but have hardware for only a few tens of modules (see for example, the HIL simulation work of Lal and Fitch, 2009). There is also scope for simulation tools to be used within an automated machine’s programming paradigm in an online manner, particularly with respect to planning tasks. This is often referred to as decision support. Here, the robot holds some model of the world and is able to run a simulation of the robot’s execution of the plan in a faster than real-time manner to evaluate various decision metrics. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

259

N HILLIER AND J RYDE

The application of the virtual world provided by simulation to robotics is almost endless. Some of the other uses of which the authors are aware include visualisation, training (of both operators and for evolutionary type systems) and as a tool for the design of user interfaces. Each of the above presented applications for simulation have different performance criteria and ideally a simulation engine should be sufficiently flexible to allow reuse in all of the above scenarios with minimal customisation.

TEST-BASED DEVELOPMENT System dependability (reliability, trustworthiness) is a key criteria for the development of robotics within many groups due to any number of commercial, remote deployment and/or risk constraints. In practice, there appear to be three streams of development methodology in use: x restriction of development tools, algorithms and platforms to those that are time-proven and tested; x verification via theorem proof (see the methodologies of Bensalem, Ingrand and Sifakis, 2008); and/or x thorough, test-based development. Given the current pace of worldwide robotics research and the rapid development of new algorithms that can be considered crucial to the performance of advanced mining automation systems (eg simultaneous localisation and mapping – SLAM, computer vision), the authors consider that the achievement of the dependability criteria is best developed through test-based methodologies. Such methodologies also present a measurable metric upon which we can assess the system’s dependability. Furthermore, test-based development allows for aggressive re-factoring and experimentation without fear of a regression in functionality. Finally, the test-based development methodology has a demonstrated track record in other disciplines.

Practicalities for automating machinery Test-based development has been a very successful paradigm in software engineering with various derivatives such as test driven development seeing success (see for example Nagappan et al, 2008). It has the potential to be beneficial for robotics development; however, the translation of this paradigm to the robotics domain can be problematic. While it may be argued that test-based development is possible with indoor robots, it is often impractical for application to the mining domain for a number of reasons: x Tests on real robots have to run in real time by definition, this is time consuming for the development cycle of test-based development, as such cycles consist of multiple iterations of writing test code, writing program (deployment) code and running tests. x The repeatability of the test suite is important to the methodology’s success. Ensuring that the test harness and robotic system are in the same initial state before testing can be difficult in real mining robotics deployments, but is readily achievable for a software simulator. x While regression testing in software engineering has negligible cost, the regular execution of regression tests on autonomous systems can be costly in terms of fuel consumption and mechanical wear. Performing such tests for automated machines can be segmented into two distinct categories: 1. Those tests that exercise passive algorithms, and can be run on recorded data, such as simultaneous localisation and mapping (SLAM), where the system is an observer of the data. 2. Those tests that exercise active algorithms, whereby the sensory data is altered by the action of the algorithm (the system is a participant in the creation of the data – eg path planning and control execution). Such tests are required to be run on the robot (live) or through some element of simulation. We argue that simulation is typically the only practical means by which to exercise active tests, and typically also allows for the exercising and evaluation of passive algorithms. Additionally, a simulator allows for fully automated testing at any time, which is vital for regression testing and rapid development. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

260

SIMULATION FOR DEPENDABLE MINING AUTOMATION

Ideally, tests should be deterministic and idempotent, namely, that $f(x) = f(f(x))$. It is important to appreciate that success in simulation is most often a necessary but not sufficient condition for success in the field. For this reason, simulation tests should be run following all code changes and before every operation in the field. Furthermore, there is an array of tests that are substantially easier to write in simulation due to the ready availability of expected results in the simulator internals (eg those that require or generate pose estimates). Some may argue that such tests are not productive, because they do not easily translate to implementation on deployed robots (the truth is not readily available); however, the mere existence of the test and the ease with which it can be used for underlying algorithm development purposes can greatly accelerate the development time frame and again, provides some increase in the measure of system dependability.

SIMULATION FIDELITY The final aim of the use of the simulator in robotics development is to maximise the intersection of code that can be run in the real-world and simulator. If such a system can pass all required tests in the simulation environment, this infers that a high level of trust can be placed in the deployed code. The issue of simulation fidelity is then raised. We refer to simulation fidelity as the level of detail and accuracy to which the simulation reflects the real system; for example, what sensor models are used or the accuracy of the physics (or indeed, if a kinematic only simulation is sufficient). As touched upon in the introductory discussion with reference to the MITRE Meteor DARPA Grand Challenge entrant, the fidelity with which the simulation reflects the real world system(s) can both be crucial to the success of the robot and a valuable log of the development process. Whilst it is easy to argue that one should aim for a simulation environment with the lowest possible complexity (and hence minimal fidelity) required for the task at hand (this reduces the possible number of failure points in the simulation, easing debugging and ensures that time is not spent developing redundant functionality in the simulation environment). The ready availability of physics simulation engines gives a base level of fidelity that can aid in identifying unforeseen, but common execution failures. An example of this is the classic planar world assumptions, which the authors found to be immediately impractical on the UGV discussed earlier, due to the pitch and roll effects of the platform in acceleration manoeuvres and the associated errors in scanning laser returns when mapped to a two-dimensional (2D) environment. The issue of required fidelity for the task at hand has implications on the choice of simulation package. Where by default the simulator is lacking the minimal required fidelity, the simulation engine must be sufficiently flexible to allow the researchers to increase the fidelity of the simulation environment as required.

TIMING Beyond the fidelity issues of non-simulated phenomena that may be critical to the robots success (or failure), the authors argue that the other most critical element to ensuring the successful transfer of simulator developed code to real-world robotics is that of timing. Simulators offer the ability to control the passage of time; however, this introduces a wide array of possible failure mechanisms to the system that may not be apparent in systems developed without the use of a simulation environment.

Controlling time The control of the passage of time is a powerful tool only available in simulation that, amongst other advantages, enables real-time debugging. Conventional debugging pauses the execution to allow inspection and possible modification of the program state. For systems operating in the real world such debugging practice is typically impossible, as whilst the program execution is paused, time in the real world continues (pausing the system in itself typically alters the system state, and in some instances may induce catastrophic failure – eg debugging a helicopter control system during flight). Such debugging tools, however, are easily applied in simulation by pausing both the simulator and the robot program when a break condition is reached. Altering simulator time can also aid with visualising actions that would otherwise occur too fast in reality. Without simulators these sorts of debug actions are typically done by real-time configurable logging and post mission inspection of log files. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

261

N HILLIER AND J RYDE

Accelerating simulation time also allows the execution of experiments that would otherwise be difficult to run. An example is that of training a neural network or an evolutionary system. Once the algorithms have been trained in simulation they require less verification and intervention before deployment on the real robot.

Simulation time-induced failures The alteration of time in the simulation environment, whether intentional as outlined previously, as a side-effect of under-specified hardware or a by-product of the discrete time-stepping employed can reveal a number of failure modes. When properly addressed, a significantly more robust robotic solution is developed. These variations in time affect the independence of processes and computation nodes and reveal any dependence of algorithms on clock timing. Care must be taken that the time reference provided by the simulation and that used in the operational code is consistent. Furthermore, if computing resources are scarce on the deployed platform, there arise issues surrounding slower and faster than real-time simulation, whereby required computations may use all available resources and unpredictable behaviour may result.

CASE STUDY – SKID STEER LOADER UNMANNED GROUND VEHICLE This section discusses a few uses of simulation tools in the development of an autonomous skid steer loader (modified Bobcat S185) displayed in Figure 1. This is a versatile and ubiquitous vehicle typically deployed on construction sites. Its main defining characteristics are its compact size, ability to turn in place and the positioning of the operator between the boom arms. The platform has significant operational safety risks induced by high turn rates, large actuation forces and high moving mass.

(A)

(B)

FIG 1 - (A) Implementation of the bobcat simulator in OpenRave; (B) the real bobcat.

Sensor noise failure During early platform development, a control system failure occurred and was found to be due to harmonic noise on a velocity sensor. This manifested in the UGV drifting in position when commanded to hold a static position. An increase in simulator fidelity was made by implementing a capped random walk as noise to the velocity sensor. The addition of this type of noise to the simulated sensor resulted in the simulated UGV exhibiting similar behaviour to the real platform. A simple station hold test was written and the controller adjusted to account for the sensor noise. The existence of this sensor noise and associated test has resulted in a system that allows for the identification of future regression in this control functionality and allows for increased confidence during control system code re-factoring. We have since extended this type of testing to include adjusting and increasing sensor noise levels to identify when algorithms fail, and what level of sensing accuracy is required. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

262

SIMULATION FOR DEPENDABLE MINING AUTOMATION

Sensor timing analysis As part of the simulator development, a number of issues related to timing became apparent (similar to those discussed in section timing). Figure 2 shows a comparison of the differences between various sensor timestamps as reported by the system middleware. There is a significant variation with some interesting multi-modal distributions in the real UGVs reported timestamps. Those reported by the simulator are much tighter in tolerance and closer to what we would expect (see the bi-modal and tri-modal distributions of the scanning laser sensors (lmsLeft, lmsRight, Spin0) and the GPS sensor (novatelGPSPos, novatelGPSVel)). We know that such sensors report very regular information, yet this is lost between the sensor and the provision of a timestamp in our middleware. The source of the error is unclear as other sensor data (eg the inertial solution (novatelINS) and control demands (lllDemand)) is reported within a tight distribution. These timing issues may become critical in the future and efforts are underway to understand such errors.

(A)

(B)

FIG 2 - (A) Comparison of time deltas between updates for simulation; (B) the real unmanned ground vehicle.

Trajectory comparison As part of the development of way-point following code, a reduced set of the UMBmark odometry benchmark test (Borenstein and Feng, 1995) was performed via way-point execution (a bi-directional square path). The results of this test are presented in Figure 3. The simulation environment was used to develop the way-point code almost exclusively with only minor alterations following field



FIG 3 - Comparison of the RTK-GPS antenna trajectories measured experimentally and in simulation for a simple bi-directional square path, starting at the origin and moving north first. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

263

N HILLIER AND J RYDE

testing. It can be seen that the four corner way-points are consistently reached both in simulation and in the experimental data set. However the simulation lacks an appropriate level of fidelity in the modelled terrain – there are significant altitude changes (0.5 m over the 6 m sides of the square) and much higher roughness in the test area as opposed to the perfectly flat and uniform surface of the simulation. These non-uniformities in the experimental test site’s surface perturb the controller and result in the poor tracking performance. Efforts are under way to import point-cloud terrain information into the simulation space. Nonetheless, the simulation provides easy access to the true trajectory in all circumstances. Typically, for field deployment, RTK GPS information is the closest (position error around 0.05 m) to the true trajectory available and in some test sites, even this measure is unreliable due to multi-path effects. This testing has also prompted the development of a trajectory following controller rather than the simple way-point code currently employed, wherein access to the true trajectory is highly beneficial to the controller’s development.

Earth moving demonstration The first significant project milestone for this UGV is the demonstration of a material moving task. This task involves autonomously moving material from one location to another and is executed in both reality and the simulation environment. The simulation of earthen material is typically not an implemented feature of robotics simulators, and whilst work on accurately modelling earthen material in simulation for robotics has been conducted (Halbach, 2007; Cleary, 1998), we chose to use a number of 0.1 m sided cubes as an approximate representation. The authors believe that the use of the simulator significantly reduced algorithm test and development time as well as the required skill level to get a high-risk robotic platform conducting useful tasks. We estimate development time to take a platform with open-loop, normalised control inputs to demonstrable performance of an earth-moving task at under 900 hours. Development time spent customising the simulation environment for this task (as opposed to learning and developing the simulator itself) is estimated at a further 100 hours.

CONCLUSION It has been argued that test-based development for mining robotics provides benefits such as: x tests allow for aggressive refactoring and experimental algorithms development without fear of regression in functionality, and x a higher confidence in software system dependability is achievable. Simulation is an enabling technology for test-based development. The key issues in using simulation for algorithm development and testing for mining robotics applications are: x Fidelity: does the simulation provide sufficient fidelity to be a useful tool (especially for testing)? Is the architecture of the simulation engine appropriately designed for the developers to alter the simulation fidelity as required, not only to change the representation of the physical world, but also to implement a common low-level interface for code and maximise the intersection of code that is run on the simulator and the real robot. This flexibility allows the users to better examine edge-case scenarios and the robustness of developed code. x Timing: does the code framework, including the simulator, middleware and extensible code elements have sufficient capability to handle non-real-time execution (if desired, or otherwise is there sufficient resourcing available to guarantee real-time execution)? Running faster or slower than real-time and the ability to pause both program execution and the simulation environment concurrently provides a powerful debugging tool and a means to artificially increase a learning or evolutionary robot’s exposure to experience. We have presented some results from a basic exemplar of the test-based development methodology’s use in the mining robotics domain. Further to the results presented, the authors believe that the approach leads to accelerated progress and that the test structure gives a more robust development process.

ACKNOWLEDGEMENTS The authors would like to thank Leon Stepan and Hendrik Erckens, who provided much of the control system development and field trials for the UGV work presented; the Engineering Support MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

264

SIMULATION FOR DEPENDABLE MINING AUTOMATION

team from the CSIRO Autonomous Systems Lab, who provided the automation interfaces to the bobcat platform and the CSIRO’s Minerals Down Under Flagship, who provided funding for the work presented.

REFERENCES Bensalem, S, Ingrand, F and Sifakis, J, 2008. Autonomous robot software design challenge, in Proceedings Sixth Joint Workshop on Technical Challenge for Dependable Robots in Human Environments (IARP/ IEEE-RAS/EURON: Pasadena). Borenstein, J and Feng, L, 1995. UMBmark: A benchmark test for measuring odometry errors in mobile robots, in Proceedings SPIE Conference on Mobile Robots, pp 113-124 (SPIE: Philadelphia). Carpin, S, Lewis, M, Wang, J, Balakirsky, S and Scrapper, C, 2007. USARSim: A robot simulator for research and education, in Proceedings International Conference on Robotics and Automation, pp 1400-1405 (IEEE: Rome). Cleary, P W, 1998. The filling of dragline buckets, Mathematical Engineering in Industry, 7:1-24. Corke, P, Sikka, P, Roberts, J and Duff, E, 2004. DDX: A distributed software architecture for robotic systems, in Proceedings Australasian Conference on Robotics and Automation (ARAA: Canberra). Diankov, R and Kuffner, J, 2008. OpenRAVE: A planning architecture for autonomous robotics, Robotics Institute, technical report, CMU-RI-TR-08-34 [online]. Available from: [Accessed: 10 September 2010]. Gerky, B P, Vaughan, R T and Howard, A, 2003. The player/stage project: Tools for multi-robot and distributed sensor systems, in Proceedings International Conference on Advanced Robotics, pp 317-323, Coimbra, Portugal. Grabowski, R J, Weatherly, R M, Bolling, R H, Seidel, D, Shadid, M and Jones, A, 2006. MITRE meteor: An off-road autonomous vehicle for DARPAs grand challenge, Journal of Field Robotics, 23(9):811-835. Halbach, E, 2007. Development of a simulator for modeling robotic earth-moving tasks, Master’s thesis, Luleå University of Technology, Scandinavia. Kanehiro, F, Hirukawa, H and Kajita, S, 2004. OpenHRP: Open architecture humanoid robotics platform, International Journal of Robotics Research, 23(2):155-165. Koenig, N and Howard, A, 2004. Design and use paradigms for gazebo: An open-source multi-robot simulator, in Proceedings International Conference on Intelligent Robots and Systems, pp 2149-2154 (IEEE/RSJ: Sendai). Lal, R and Fitch, R, 2009. A hardware-in-the-loop simulator for distributed robotics, in Proceedings Australasian Conference on Robotics and Automation (ARAA: Sydney). Michel, O, 2004. Cyberbotics Ltd – WebotsTM: Professional mobile robot simulation, International Journal of Advanced Robotic Systems, 1(1):40-43. Milgram, P and Kishino, A F, 1994. Taxonomy of mixed reality visual displays, IEICE Transactions on Information and Systems, E77-D(12):1321-1329. Nagappan, N, Maximilien, E M, Bhat, T and Williams, L, 2008. Realizing quality improvement through test driven development: Results and experiences of four industrial teams, Empirical Software Engineering, 13(3):289-302. Quigley, M, Conley, K, Gerkey, B, Faust, J, Foote, T B, Leibs, J, Wheeler, R and Ng, A Y, 2009. ROS: An opensource robot operating system, in Proceedings International Conference on Robotics and Automation, Open-Source Software workshop (IEEE). Rupp, T, Boge, T, Kiehling, R and Sellmaier, F, 2009. Flight dynamics challenges of the German on-orbit servicing mission DEOS, in Proceedings 21st International Symposium on Space Flight Dynamics, Toulouse, France. Weatherly, R M, Kuhl, F S, Bolling, R H and Grabowski, R J, 2006. The Mitre Meteor robot control software: Simulate as you operate, in Proceedings Winter Simulation Conference 2006, pp 1294-1298 (WSC Foundation). Zaratti, M, Fratarcangeli, M and Locchi, L, 2007. A 3D simulator of multiple legged robots based on USARSim, in Proceedings Tenth International Robocup 2006 Symposium, pp 13-24 (Springer).

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

265

Coal Bed Boundary Detection using Infrared Technology for Longwall Shearer Automation K Nienhaus1, F Mavroudis2 and M Warcholik3 ABSTRACT Automation in underground mining enhances safety and leads to economic efficiency. In the past many efforts have been made on the automation of longwall shearer operation. With respect to the ultimate goal of automation, ie self controlled operation, coal bed boundary detection is one of the key problems. For longwall human labor is necessary in order to detect the exact position of the coal seam to be excavated. The authors present a sensor concept based on infrared technology and pattern recognition which will solve this problem. The very harsh underground conditions set high demands on the implemented sensors. The combination of infrared sensors and powerful processing algorithms is part of the innovative concept for automation of the new Eickhoff Bergbautechnik GmbH SL 750 shearer loader. The multi-sensor automation system EiControlPlus which is designed for the complete detection of the machine environment fills in the gap in the coal-excavation automation by building a recognition system that turns the shearer loader into an ‘intelligent machine’ which is capable of detecting the correct position of the seam to be excavated and thus enables human-free coal mining.

INTRODUCTION Automation in underground mining is promoted for two basic reasons: improved safety and economic efficiency. The prospect of improved safety is invaluable, where the ultimate goal is to utilise as little human labour as possible to avoid injuries and exposing miners to hazardous, noisy and dusty environment. Besides evaluation and exploration of the deposit, extraction is an essential key to the economic success of the mining operations. Improved economic efficiency has also come to the fore, as automation streamlines and speeds up the whole excavation process. The automation of individual processes in recent decades led to a considerable increase in productivity and has improved the utilisation of the capacity of the individual components. Intelligent assistant systems and multisensor concepts relieve and assist machine operators and thus contribute to greater safety in the mining industry. The difficult ambient conditions in underground mining impose high requirements on the sensors used. Rough environmental conditions in combination with very high mechanical demands restrict the use of the sensors underground and may lead to complete failures. Regarding the ultimate goal of automation, ie self controlled operation, coal bed boundary detection is one of the key problems. The EiControlPlus shearer loader automation system equipped with Department for Mining and Metallurgical Machinery’s (IMR’s) infrared technology as the key component could be the solution to safe and highly efficient underground coal mining.

COAL INTERFACE DETECTION Currently the control systems of shearer loaders working underground are conducted manually. The operator controls the shearer and all its functions and navigates the mining machine threedimensionally through the deposit to be extracted. The operator must detect the geological and geometrical environment with his eyes and ears (cutting sound). Dust, rock fall, humidity, noise and constraints to the operator’s vision are all reasons for miners demands for far-reaching automation of the mining process. The accuracy of deposit boundary detection relies to a great 1. Professor, Head of the Department, Department for Mining and Metallurgical Machinery (IMR), RWTH-Aachen University, Wuellnerstrasse 2, Aachen 52056, Germany.

Email: [email protected] 2. Research Engineer, Department for Mining and Metallurgical Machinery (IMR), RWTH-Aachen University, Wuellnerstrasse 2, Aachen 52056, Germany. Email: [email protected] 3. Research Engineer, Department for Mining and Metallurgical Machinery (IMR), RWTH-Aachen University, Wuellnerstrasse 2, Aachen 52056, Germany. Email: [email protected]

MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

267

K NIENHAUS, F MAVROUDIS AND M WARCHOLIK

extend on the ability of the operator. Corresponding imprecision often leads to low recovery rates. In addition, the extraction of waste rock leads to more wear of cutting tools and increased operating and processing costs. In the past many efforts have been made to develop sensors for coal-roof/coal-floor interface detection. In principle two approaches were used, namely surface recognition and depth thickness recognition. Whereas the former requires periodical cutting in the host rock to follow the coal/ rock boundary, the latter is used to predict remaining coal thickness. Consequently, for extracting a complete coal seam the surface recognition approach is suited whereas for leaving a defined thickness of coal at the roof or floor usually requires a depth thickness approach. The following figure gives an overview of the coal interface detection (CID) sensors that were found to be more promising as well as an assignment to the CID approach (Vinzelberg and Nienhaus, 2006). By using characteristic layers (such as clay layers) in the seam the first approach can be used without cutting in the roof/floor repeatedly. Characteristic rock-layers can be traced inside the seam along the longwall and can be used as guidance for the shearer’s path. Material characterisation and detection of interlayering rock layers and geological strictures can be achieved with the aid of infrared and vibration measurement. The infrared sensors can detect textures of the mineral to be mined and utilise these textures as a control aid for the shearer loader.

BASICS OF INFRARED TECHNOLOGY AND SENSOR SYSTEMS There are certain environmental challenges that the infrared sensors that capture the images of the coal face have to endure, the most important being the necessity to withstand enormous vibrations and capture clear images through dust. Infrared radiation is part of the electromagnetic spectrum, which includes various types of radiation The infrared range comprises the range of wavelengths from 0.78 mm to 1 mm. A distinction is made between the near, middle and far range. Figure 1 shows the electromagnetic spectrum and the subdivision of the infrared range into ranges of different wavelengths (Figure 2).

FIG 1 - Overview of coal-interface detection systems.

In infrared measurement the radiation of the measured object is measured without contact via a sensor which is sensitive to infrared radiation. The distance between the object to be measured and the sensor is of vital importance. The radiation must penetrate the atmosphere before it impinges on the sensor and can be absorbed and scattered. The atmospheric effects may falsify the measurements by attenuation of the transmitted radiation and reduction of the contrast by the natural radiation of the atmosphere. MINE PLANNING AND EQUIPMENT SELECTION (MPES) CONFERENCE / FREMANTLE, WA, 1 - 3 DECEMBER 2010

268

COAL BED BOUNDARY DETECTION USING INFRARED TECHNOLOGY FOR LONGWALL SHEARER AUTOMATION

FIG 2 - Classification of the infrared ranges in the electromagnetic spectrum.

According to Wien’s displacement law the maximum radiation from matter with ambient temperature is in the wavelength range from 8.0 to 12.0 mm. To counteract the atmospheric effects infrared sensors are used in the wavelength range from 7.5 to 14 mm. The high transmission degree of the atmosphere in the long wave range and the resulting large distance from the visible light also produce the clear advantages of infrared sensors in comparison to the sensors working in the visible area (Figure 3).



FIG 3 - Emissions of the sun and earth in the electromagnetic spectrum.

The infrared sensors used offer high resistance to rough ambient conditions such as dust, haze, rain, fumes and air humidity. In addition images can be taken in poor weather and visibility conditions. Images can also be taken at night because the infrared image is generated by the reception of thermal radiation. The development of infrared sensors in the last years has led to the production of optical devises without mechanical scanners and complex cooling systems and with extremely compact dimensions. This enables the use of sensors in areas with high oscillations and guaranties that the sensors are robust even in conditions where vibrations and shocks occur. The modern sensors can endure shocks of up to 70 g and vibration of up to 4.3 g and also have small dimensions (35 × 37 × 49 mm) and low weight of