Mining Education Australia Research Projects Review
Transcription
Mining Education Australia Research Projects Review
Mining Education Australia Research Projects Review 2012 Editor: Paul Hagan Published by: The Australasian Institute of Mining and Metallurgy Ground Floor, 204 Lygon Street, Carlton Victoria 3053, Australia i © Mining Education Australia 2013 All papers published in this volume were independently peer reviewed prior to publication. Mining Education Australia is not responsible as a body for the facts and opinions advanced in any of its publications. ISSN: 2201 - 716X Desktop published by: Kylie McShane and Claire Lockyer The Australasian Institute of Mining and Metallurgy Printed by: BPA Digital 11 Evans Street, Burwood Victoria 3125, Australia ii PReFACe The 2012 MeA Research Projects Review includes a selection of papers summarising the work undertaken by final year students in the mining engineering program across the four Mining education Australia (MeA) universities. The papers were co-authored by students who completed their mining research project in 2012 and their academic supervisor. All the papers in this volume were peer-reviewed by academics from the MeA member universities. Preparation of a conference paper is one of the assessment requirements in the Mining Research Project (or thesis). The project is undertaken over two consecutive semesters in the final year of the MeA undergraduate degree program. while the learning outcomes are primarily aimed at applying the principles of the scientific process to reallife issues facing the industry, it is also concerned with students applying the principles of project management and risk management; preparing a proposal with clearly defined objectives and benefits; planning and scheduling tasks and activities; liaising with people from a range of different backgrounds; and developing both verbal and written communication skills. The MeA program is an education initiative between the four major universities in Australia that offer a degree program in Mining engineering; Curtin University through the western Australian School of Mines (wASM), University of Adelaide (UoA), University of New South wales (UNSw) and University of Queensland (UQ). The program, which was launched in 2007, was collaboratively developed by academic staff from each of the partner institutions with advice from education consultants to ensure the latest in teaching and learning practices were embedded into it. The program is supported by the Minerals Council of Australia (MCA) who, through its various industry partners, has helped ensure MeA Graduates will be able to meet current and future challenges of the global minerals industry. over 190 students across the four MeA universities completed the research project course in 2012 and consequently many will have completed the degree program. As graduate mining engineers they are now set to begin their professional careers in the mining industry. As in previous years, the research projects span a broad spectrum of topic areas, reflecting the range of issues being faced by the mining industry. The distribution of project categories across MeA in 2012 is shown in Figure 1 while Figure 2 shows the topic distribution at each of the four universities reflecting to some degree the differences in expertise and focus at each university. Fig 1 - Distribution of research topic areas across MEA in 2012. iii Fig 2 - Distribution of research topic areas by university. The fourth MeA Student Conference was held in october 2012 at the wASM campus of Curtin University in Kalgoorlie; a copy of the program is shown in Appendix 1. This year the event was enlarged to a half-day conference with up to three students selected from each university to present a paper on their research project. The list of presenters is shown in Appendix 2. The three 2012 prize winners at the conference were: • • • Casey Costello (UQ) Brenton Goves (UQ) Tim Graham (UNSW). Members of the MeA course team who coordinated the research project are acknowledged for their efforts. This team included Dr Basil Beamish (UQ), Dr Paul hagan (UNSw), Prof Roger Thompson (wASM) and Associate Professor Chaoshui Xu (UoA). It is also important to acknowledge the important role of the academic staff who supervised the students and the professional staff who supported the students in their project. Special acknowledgments are made to Dr Beamish who will be standing down from his full-time academic role at UQ in early 2013, and Dr Andrew jarosz who is retiring from wASM also in early 2013. Dr Beamish was a founding member of the course team when it was established in 2006. he has had a great passion for the course and as a result of his drive and enthusiasm has contributed to the continual development and improvements in the course that have translated into a better learning experience for students. Dr jarosz similarly has been a long-term supporter of the course, understanding the importance it plays in contributing to the education of mining engineers. Finally, the contribution made by the mining industry and especially those people who assisted the students with their research projects must be acknowledged. Many of the projects were initiated while students were on vacation employment which again emphasises the important contributions work experience makes to student education. It is important that industry recognises the role it plays in contributing to student education and ensuring the continual supply of high quality graduate engineers. Paul Hagan Mining research Project course convenor, and chairman MEa Program leaders committee iv MEA Student Conference Organising Committee Dr Andrew Jarosz (WASM) – Conference Convenor Associate Professor Emmanuel Chanda (UoA) Dr Paul Hagan (UNSW) Associate Professor Mehmet Kizil (UQ) Dr Rudra Mitra (UNSW) Review Panel Mr Duncan Chalmers (UNSW) Associate Professor Emmanuel Chanda (UoA) Dr Chris Daly (UNSW) Dr Nazife Erarslan (UQ) Dr Paul Hagan (UNSW) Dr Adrian Halim (WASM) Dr Andrew Jarosz (WASM) Dr Murat Karakus (UoA) Associate Professor Mehmet Kizil (UQ) Dr Rudra Mitra (UNSW) Dr Simit Raval (UNSW) Associate Professor Serkan Saydam (UNSW) Professor Warren Seib (UQ) Associate Professor Chaoshui Xu (UoA) v viii CONTENTS An Investigation into Semi-Intact Rock Mass Representation for Physical Modelling Block Caving Mechanics Zones P Carmichael and B Hebblewhite 1 Maximising Production at Brockman 4 by Minimising Truck Delays at the Crusher V Collins and M S Kizil 7 Grizzly Modifications at Ridgeway Deeps Block Cave Mine C Costello and P Knights 13 Multi-Objective Optimisation of MiningMetallurgical Systems V Golding, D Jafari, B Rajopadhyaya, D Sundquist and E Chanda 19 Evaluating the Performance and Productivity of Continuous Surface Miners in Iron Ore at Christmas Creek Mine B Goves, M S Kizil and W Seib 25 Strategic Project Risk Management for an Emerging Miner T K C Graham and S Saydam 31 Metaheuristic Analysis of Current Open Cut Blasting Fragmentation Prediction Models A Holland, M Iles and M Karakus 37 Ventilation Requirement for Electric Vehicles in Underground Hard Rock Mines – A Conceptual Study M Kerai and A Halim 45 Prediction and Modelling of Blast Vibration and its Effects at Glendell Colliery M K McKenzie and D Chalmers 51 Modelling Depletion of Queensland Coal Resources using Historical Production Trends A Meikle and W Seib 57 Management of Trailing Cables on Electrically Powered Load-Haul-Dump Units W Paterson and P Knights 67 The Effect of Joint Properties on a Discontinuous Rock Mass M Sherpa and P Hagan 75 83 Author Index ix x an Investigation into Semi-Intact rock Mass representation for physical Modelling Block caving Mechanics zones P Carmichael1 and B hebblewhite2 aBStract With the continual advancement of block caving operations comes a need for greater understanding of the governing physics of the propagation process. Hence, it is desirable that cavability be more reliably predicted in the future to ensure safety and efficiency is achieved throughout the mining process. This paper presents the results of an experimental program that examined the potential use of semi-intact rock masses in the physical modelling of block caves, to analyse cave propagation and the associated mechanics zones. The findings of the investigation showed that it is possible to model caving mechanics zones with a material consisting of gravel, sand and a gypsum plaster such as Plaster of Paris. The paper also identified the two major limitations of the current material: similitude and texture of the material. Regarding similitude, both geometrical and dynamic similitudes were not met in all areas, whilst the texture of the material makes delineation of the caving mechanics zones impossible at this time. The results of the investigation are presented in this paper such that they can be used for further refinement of the modelling material and for immediate preliminary cave propagation modelling. IntroductIon Block caving is fast becoming the method of choice for extracting large, low-grade deposits at depth. It has been identified by Laubscher (1994) and later by Chitombo (2010) that block caving is entering a new era of ‘super caves’. As such it has been recommended by Chitombo (2010) that extensive research be undertaken in all aspects of block caving with particular focus on propagation, fragmentation and particle migration to ensure block caving remains safe and efficient in the future. Cave propagation and fragmentation is governed by five caving mechanics zones, first analysed by Duplancic and Brady (1999) by way of a seismic and long hole drilling survey. These various caving zones are indicated in Figure 1. Extensive time and resources are required to analyse these zones in the full scale and as such are unavailable to most operations. Therefore, numerical modelling is currently the method of choice to analyse cave propagation due to the current accessibility of computational power; however, the results obtained are only as accurate as the inputs provided. As such, there is a requirement for greater understanding the caving process to be able to provide accurate inputs for the calibration of numerical models. It is believed that with continual research and analysis, the governing physics of the caving process can be fully Fig 1 - The five regions associated with caving, determined by Duplancic and Brady (1999). understood to not only aid in numerical modelling investigations but also ensure accurate and reliable cave prediction at existing and new operations in the future. The primary objective of this research project was to provide a method by which the industry can study and analyse the caving process in a controlled environment. 1. The University of New South Wales, Sydney NSW. 2. MAusIMM, The University of New South Wales, Sydney NSW. Email: [email protected] MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 1 p carMIchael and B heBBlewhIte By using readily available resources, low-cost quick-setting modelling materials were investigated for use in a twodimensional (2D) block cave modelling frame to allow for prompt analysis of caving mechanics. It must be noted that the research at this stage was largely qualitative and subjective due to the preliminary nature of the investigation. apparatuS The main apparatus used in the experimentation was a purpose built 2D modelling frame, which is shown in Figure 3 The critical dimensions of the frame are outlined in Tables 1 and 2. Methodology preliminary testing A preliminary analysis was first undertaken to determine a range of cemented rock masses suitable for large-scale testing. This preliminary testing was largely qualitative as the investigators were seeking an indication of material behaviour rather than accurate material properties. The materials chosen for the testing were sand, 5 mm gravel, 15 mm gravel, Boral casting plaster and water due to their availability and low cost. The preliminary testing involved slump testing approximately 40 mixtures ranging from zero per cent to six per cent plaster and zero per cent to 30 per cent sand, using the moulds shown in Figure 2. Fig 3 - Two-dimensional physical modelling frame in the testing orientation. TABLE 1 Modelling frame internal dimensions. internal model dimensions (mm) Height 1250 Width 1000 Depth 188 TABLE 2 Modelling frame critical dimensions. Fig 2 - Modelling moulds used in preliminary analysis. Laboratory tests were also undertaken to determine the strength of the material. By using a 15 cm core mould (10 × the largest particle) and plaster capping, the material was tested for its uniaxial compressive strength (UCS). large-scale testing Overview Once the preliminary testing was completed the successful materials were applied to a newly acquired 2D physical modelling frame. The purpose of this testing was to determine the compatibility of the cemented rock mass to model cave propagation with varying stress and fault regimes in place. The procedure involved mixing up approximately 380 kg of a material for loading into the frame. Once loaded it was allowed to set for 24 hours. When set an undercut was created and stresses were applied to induce caving. The resulting cave propagation was documented and analysed. With the frame being new and little work having been done in this area, the process was largely ‘trial and error’ and as such, the procedure changed for each test undertaken, based on recommendations from previous tests. 2 Modelling frame critical measurements (mm) Max drawpoint area Stress-inducing plate area 56 × 300 1250 × 188 The frame has the ability to rotate in two dimensions to allow for the placement of faults in the rock mass. It also has a system by which horizontal stresses can be applied during the caving process. The stress application system is shown in Figure 4. It consists of two stress screws on each side, which act upon a plate and then onto the mixture. The stresses were measured by cantilever load cells, which were connected to a National Instruments card for recording and analysis during testing. This system is shown in Figure 5. reSultS and analySIS preliminary testing From the preliminary tests, the following key findings were made: • sand content had the largest effect on material strength and behaviour in the first 48 hours due to its water retention abilities MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 an InveStIgatIon Into SeMI-Intact rock MaSS repreSentatIon for phySIcal ModellIng Block cavIng MechanIcS zoneS TABLE 3 Large-scale tests undertaken. Plaster (per cent) Fig 4 - Load cell positioning on force plates and amplification system (Kell, 2012). Fault (degrees) Stress regime - Free flow gravel - Wooden block test 1 - Low 2 - High 1.5 - High 1.2 - High 1 - Medium 1 45 Medium 1 - Very high 1 45 Medium 1 - Normal (prestressed) Fig 6 - Comparison of cross-sections of test one and numerical modelling investigation. Fig 5 - Load cell monitoring equipment for use with the model. • • • • a material of approximately one to two per cent plaster and zero per cent sand was suitable for large-scale testing the material forms its own inherent weakness planes, along which failure occurs the material does not instantly break into individual gravel pieces, but rather clumps, allowing for secondary fragmentation a material consisting of one per cent plaster and zero per cent sand had a UCS of approximately 11 kPa. fracturing increased in size, to be able to be detected visually. Examples of this fracturing are shown in Figures 7 and 8. Both faulted tests resulted in similar findings, with the cave changing direction slightly while passing through the faulted zone, as shown in Figure 9. This process showed that the cemented rock mass is suitable for testing caving through fault zones. large-scale testing Eleven tests were undertaken using the developed material and modelling frame. These tests are outlined in Table 3. A number of key findings were made from these large-scale tests. Firstly, it was identified that, as with full-scale operations, air gap management was key to the success of the test. In one test the air gap was not managed, consequently the cave instantly broke through to the surface. Despite the failure to manage caving, the cross-section of the propagation struck similarities with a numerical modelling investigation undertaken by Vyazmensky (2012), as shown in Figure 6. Another finding was the identification of fracturing above the cave back, starting in the seismogenic zone and further expanding in the zone of discontinuous deformation. Fracturing could not be visually identified in the seismogenic zone; however, fracturing was audible. It was not until the rock mass entered the zone of discontinuous deformation that the MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Fig 7 - Fracturing occurring above the cave back. In the majority of the large-scale tests, an air gap was formed before stresses were applied to allow caving to take place, allowing the material to manage its stresses. It was decided 3 p carMIchael and B heBBlewhIte on material strength and swell factor, which was identified to be lower than the 20 per cent to 40 per cent experienced in full-scale operations. The result of this test was that the material was stronger during testing and withstood stresses of up to 0.04 MPa as opposed to the previous rock mass that failed with 0.035 MPa. It was also identified that the rock mass had a higher swell factor as the material had fewer voids due to setting under compression. Fig 8 - Fracture opening up during the caving process. Two major issues were also identified during the large-scale testing and subsequent analysis. The first major issue was the porosity and texture of the material not allowing for simple detection of fracturing above the cave back, hence making it difficult to delineate the five caving mechanics zones. The second major issue was that similitude was largely not met during the investigation. Recommendations have been made to rectify both problems, which are outlined in the recommendations section. concluSIonS The results of this investigation show that it is possible to utilise a semi-intact rock mass to model cave propagation and the associated mechanics zones when used in conjunction with a block cave modelling frame. However, at this stage the delineation of the caving mechanics zones is impossible due to the inability to visually detect fracturing above the cave back. This was a result of material texture. Fig 9 - Cave propagation through faulted zone. in one test to place the material under high stress without allowing caving, to determine whether the cave would stall, form wider, form narrower or undergo any other changes during the caving process. The stresses of the high-loading test compared to a normal test are shown in Figures 10 and 11 respectively. It can be seen that the stresses applied were higher than those normally applied and they were continuously reapplied throughout the test. The result of this test was that when the drawpoint was opened the cemented rock mass flowed freely from the frame, similar to the free-flow gravel test. It was hypothesised that this was due to the high stress breaking bonds in the material as it could not de-stress through caving. It is believed that this is largely due to the material not being set under stress; therefore, there are significant voids for the material to move into under stress. As a recommendation from this test a cemented rock mass was allowed to set under stress to determine the effect this had From the preliminary investigations, a cemented rock mass consisting of one per cent Plaster of Paris, three per cent water and 96 per cent gravel (5 mm to 15 mm) was deemed suitable for testing in the 2D modelling frame. The material showed changes in cave propagation when subjected to differing fault and stress regimes; however, the material had two significant flaws, relating to the texture and similitude. recoMMendatIonS Although caving was achieved, it was identified that there are still a number of flaws in the cemented rock mass. The two major flaws are outlined below. The first issue associated with the material is the texture, more specifically the porosity. Due to the porosity and dark colour of the material, it is difficult to identify any small fracturing above the cave back, making it impossible to delineate the mechanics zones for analysis. As such, it is recommended that an infill material be investigation to Fig 10 - Stresses applied during high stress test. 4 MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 an InveStIgatIon Into SeMI-Intact rock MaSS repreSentatIon for phySIcal ModellIng Block cavIng MechanIcS zoneS Fig 11 - Stresses applied during normal test. decrease this porosity. A key property of this material is that it must not increase water retention in the model, as did sand. This is to ensure the material can set rapidly for timely testing. The second major issue with the material was a lack of similitude. It was determined through the literature that complete dynamic and geometric similitude is not achievable; however, it should be attempted to replicate similitude in the areas that most affect the phenomenon being tested. As such, it is recommended that further work be undertaken into the area of dynamic similitude concerning the scaling of material UCS, tensile strength and other properties that have a significant effect on the caving process. acknowledgeMentS The authors acknowledge the support of James Tibbett and thank the contributions made by Jake Kell, Alison Tibbett and Kanchana Gamage. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 referenceS Chitombo, G P, 2010. Cave Mining – 16 years after Laubscher’s 1994 paper ‘Cave mining – state of the art’, in Proceedings 2010 Second International Symposium on Block and Sublevel Caving (ed: Y Potvin), pp 45-61 (Australian Centre for Geomechanics: Perth). Duplancic, P and Brady, B H, 1999. Characterisation of caving mechanisms by analysis of seismicity and rock stress, in Proceedings 9th International Congress on Rock Mechanics 1999, pp 1049-1053 (Balkema: Rotterdam). Kell, J, 2012. Personal communication (undergraduate student): University of New South Wales, 6 September. Laubscher, D, 1994. Cave mining - state-of-the-art, in Proceedings 6th Underground Operators Conference 1995 (ed: T S Golosinski), pp 165-175 (The Australasian Institute of Mining and Metallurgy: Melbourne). Vyazmensky, A, 2012. Personal communication Geotechnical Engineer) Kazakhmys PLC, 14 June. (Principal 5 6 Mining Education Australia – Research Projects Review 2012 Maximising Production at Brockman 4 by Minimising Truck Delays at the Crusher V Collins1 and M S Kizil2 ABSTRACT The truck and shovel mining method is the most commonly used method in surface mining. Trucks are loaded by dig units such as shovels, excavators and front-end loaders. Haul trucks then transport the mined material from the pit to the crusher, where it is broken down and sent to the plant for further processing. Brockman 4, a surface iron ore operation in Western Australia, utilises the truck and shovel method of mining. Observations show that the truck cycle is not operating at its full capacity. This paper investigates ways in which truck delays at the crusher can be minimised in order to effectively maximise production. Possible solutions identified in this paper can be used as an interactive tool to further understand the nature of the mine-plant interface at Brockman 4 and how wait-for-truck and wait-for-crusher delays can be effectively reduced. This could translate to an annual revenue increase of A$130 M should wait-for-truck delays be eliminated entirely. This paper also provides a number of suggestions to reduce delays for both the crusher and the trucks. Introduction Due to the high operational cost related to the truck and shovel method, it is vital that the efficiency of the truck cycle is maximised. The smallest improvement to the efficiency of any truck and shovel operation presents the opportunity for an increase in revenue of millions of dollars. At Brockman 4, a Rio Tinto Iron Ore mine site in Western Australia’s Pilbara, observations show that often, numerous trucks are found queuing at the crusher or there are no trucks available to supply feed. This paper investigates possible solutions and ways to minimise wait-for-crusher and wait-fortruck delays, and the effects that implementing such solutions can have on production and revenue in the long term. Modular DISPATCH® is a dispatching system used on all mobile equipment on site to record various parameters such as the time taken to load, haul, queue and dump. This data, collected over a six month sample period, was used to generate a series of inputs into a computer software package known as TALPAC. TALPAC is a haulage fleet evaluation system that is used to determine the productivity and related economics of any truck and shovel haulage system (Runge, 1993). This software package was used to create a series of haulage simulations that replicate those at Brockman 4, and whose effect on crusher delays could be further analysed. Furthermore, a computer system known as Citect is installed throughout all plant processes on site and effectively determines when trucks are able to dump in the crusher by changing a light signal alongside. The possibility of integrating the Modular and Citect systems at Brockman 4 has been investigated, as the exchange of information across the mine-plant interface is expected to increase production. The prime focus of this paper is to minimise wait-forcrusher and wait-for-truck delays by investigating both the fleet composition and external solutions in order to achieve this. Truck and Shovel Operations Truck and shovel method The truck and shovel mining method is considered to be the most predominant and flexible system used in open pit mining (Holke, 2004; Oberisser, 2010). Despite this, the costs associated with the truck and shovel mining method account for more than 50 per cent of the total cost involved with surface mining (Kizil, Knights and Nel, 2011). It is vital then, that this component of the mining process is made as efficient as possible. An example of a truck and shovel production cycle can be seen in Figure 1. Truck-shovel matching In a truck and shovel operation, it is imperative that the capacities of the dig units are compatible with the capacities of the truck fleet. Fleet efficiency can be investigated by first categorising the truck and shovel fleet as being truck-shovel matched, over-trucked or under-trucked. The term ‘truckshovel matched’ refers to situations where the ideal capacity and number of haul trucks is available for any given dig unit. Operations that are ‘over-trucked’ involve a larger number of trucks than the optimal number, yielding unnecessary additional costs with no improvement to production. 1. SAusIMM, Graduate Mining Engineer, School of Mechanical and Mining Engineering, The University of Queensland, St Lucia Qld 4072. Email: [email protected] 2. MAusIMM, Associate Professor, School of Mechanical and Mining Engineering, The University of Queensland, St Lucia Qld 4072. Email: [email protected] Mining Education Australia – Research Projects Review 2012 7 v collInS and M S kIzIl in open pit mines. Modular Mining Systems, established in the United States of America in 1979, created a computerbased DISPATCH® system designed to be used in mining operations worldwide. The company developed the system to be used primarily as a large scale mine management/ dispatch system to control the operation of haul trucks in open pit mines. ‘The primary objectives of DISPATCH® are to maximise production, minimise rehandle, supply the plant and meet blending objectives (Olsen and White, 1992)’. To achieve this, algorithms must be developed that use raw data to intelligently dispatch the trucks in all situations. Fig 1 - The truck/shovel production cycle (modified from: Kizil, Knights and Nel, 2011). Operations that are ‘under-trucked’ involve a smaller number of trucks than the optimal number, which adversely affects production, yielding substantial revenue losses. Runge (1993), as cited in Kizil, Knights and Nel (2011), observed that when an incorrect match between trucks and shovels or the fleet is not being efficiently assigned, the following operational characteristics resulted: • • • • excessive truck queuing times excessive wait on truck delays abnormal queuing times at the dump truck bunching (typically observed with mixed fleet haulage). types of dispatch systems There are three general types of dispatching methods: 1. manual 2. semi-automatic 3. automatic. Manual dispatching systems involve decision-making, where trucks are assigned to shovels by pit supervisors. The pit supervisor uses their informed opinion on what assignment is best to be conducted at that point during the shift (Bonates and Lizotte, 1988; Georgieff, 2006). Manual dispatching systems are usually only used in small operations. Semi-automatic dispatching systems incorporate real-time data into the decision-making process. Data is obtained directly from mining equipment via wireless communication systems and the computer systems installed in all the pit equipment (Bonates and Lizotte, 1988). Data is then transmitted from the trucks and shovels to the pit controller who can effectively make the decision (hence the term ‘semi-automatic’) as to where and how trucks are assigned (Georgieff, 2006). Automatic dispatch systems involve truck assignments that are dictated entirely by the computer system. The computer system analyses its own data in order to compute the best assignment to be conducted in any circumstance. Once an assignment has been determined, a system alert is displayed on the computer screen within the truck and the operator can see what their next assignment is. This system relies less on the competency of the pit controller (Georgieff, 2006). MonItorIng SySteMS In Surface MInIng Modular dISpatch® production monitoring It is widely accepted in modern mining operations that truckshovel dispatching by computer-based systems is invaluable 8 Since the start of production in 2010, Brockman 4 has incorporated the use of Modular DISPATCH® to control the operation of haul trucks in the pit, and assist the grade controller in directing or appointing shovels where to dig. Each haul truck is equipped with a live computer system that emits wireless signals back to the control centre. The accuracy of the data obtained from Modular depends on the competency of the operator and the accuracy that they provide. Such data includes (Banfield, 2012): • • • • • • • material grade, truck ID and number of tonnes per load block number (source feed) and dump location (dump feed) time truck arrives at the excavator time excavator or shovel starts and stops loading truck time the truck arrives at the dump time the trucks starts and stops dumping into the crusher time spent queuing at the dig unit or dump location. citect production monitoring Citect is a software development company who specialises in the automation and control industry (Schneider Electric, 2012). Citect has released numerous different software packages, one of which is known as Citect Ampla. For a particular application, Citect Ampla can graphically replicate on the control monitor a representation of the facility showing ‘animation points’ (Schneider Electric, 2012). The facility operator can then supervise the control of the facility via the screen. The Citect screen at Brockman 4 allows the crusher circuit to be displayed on the screen. The dynamic representation displays the crusher bin level at all times, along with the ‘dump light status’ at the crusher. This dump light status refers to the signal that trucks can see, and corresponds to the crusher bin level. Modular dISpatch® and citect integration In 2011, an interface between Modular DISPATCH® and Citect was implemented at Paraburdoo Mine Site, a different Rio Tinto Iron Ore operation (Rio Tinto Iron Ore, 2011b). The integration of the two systems between mine and plant allowed the transfer of information between the two sectors to aid in real-time amendments or changes to the operation to make it more efficient. The objective of this entire process was to minimise wait-for-crusher and wait-for truck (Rio Tinto Iron Ore, 2011b). Citect transfers the following information to DISPATCH® every five seconds (Rio Tinto Iron Ore, 2011a): the crusher bin level (per cent) the time until the apron feeder stops running, ie runs out of feed (seconds) • the truck dump light status (green/red/flashing red). Modular DISPATCH® then combines this information with estimated truck arrival times to make real-time decisions about (Rio Tinto Iron Ore, 2011a): • • MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 MaXIMISIng productIon at BrockMan 4 By MInIMISIng truck delayS at the cruSher when to reassign trucks to the run-of-mine (ROM) stockpile, without compromising feed and hence risking a wait-for-truck (or wait-for-product delay) on the plant • when to instruct the ROM loader to feed the crusher so that plant feed is maintained when no trucks are in close proximity, but stop the feed in time so that the bin is not full when the next truck arrives and unable to dump, inducing a wait-for-crusher delay • how the queue at the crusher should be managed. After the Modular-Citect integration was implemented at Paraburdoo, a number of improvements were observed. These were: • a reduction in truck queue and dump times at the crusher (166 to 134 seconds per load) • a reduction in wait-for-truck/wait-for-product delays (163 to 113 seconds per load) • an increase in throughput tonnes delivered to crusher (median shift from 16 861 tonnes to 18 161 tonnes). The increase in throughput tonnes delivered to crusher resulted in a reduction in truck queue and dump times and a reduction in wait-for-truck delays as was expected. It was also found that the average throughput increased by approximately 1300 tonnes per shift (Rio Tinto Iron Ore, 2011a). Using a conservative iron ore price of A$100 per tonne, this corresponded to an increase of approximately A$94.9 million dollars over the course of the year. • data collectIon Data were collected from Brockman 4 mine site in Western Australia over a six month sample period. Various data collected included the following: modular data (cycle times) haul routes/strings from Vulcan daily wait-for-crusher and wait-for-truck delays maintenance/downtime data daily production reports hourly crusher throughput. The sample period began on 1 January, 2012, at 6.00 am and ended on the 1 July 2012, at 6.00 am. The sample period consisted of 180 operating days and 360 shifts. • • • • • • Fig 2 - Crusher delay breakdown categories. duration and the portion of which is attributed to wait-fortruck delays which is hightlighted in the grey secind bar of each month. Figure 2 also shows the total monthly duration of waitfor-crusher delays gradually decreased over the six month period. In January, Brockman 4 utilised six Komatsu 730Es amongst their fleet of Komatsu 830Es and only two Komatsu 730Es by the end of June, 2012. This decreasing trend in waitfor-crusher delays is most likely attributed to less use of mixed fleet haulage over the sample period. This supports the literature findings by Kizil, Knights and Nel (2011). operating standby delay analysis – plant: wait-for-truck The operating standby delays for the crusher were further broken down into their subcategories. The data used was obtained between 1 January and 1 July, 2012. The pie chart in Figure 3 shows the respective proportions each subcategory contributes to the overall operating standby delay duration over this six month period. From the graph, it is apparent that wait-for-truck delays constitute the largest proportion of operating standby delays, a total 62 per cent of the total duration. There are three main pits at Brockman 4: Pit 2 (P02), Pit 3 Central (P03C) and Pit 3 East (P03E), which are located approximately 7.7, 3.5 and 2.7 km from the crusher respectively. Brockman 4 operates using 12 hour shifts with the day shift beginning at 6.00 am and the night shift beginning at 6.00 pm. Four crews (Team 1, 2, 3 and 4) are employed on an eight days on and six days off roster and alternate between dayshift and night shift each swing. The mine operates 24 hours per day, every day of the year. reSultS characterisation of delays – primary crusher Crusher delays over the six month period were categorised according to four different delay types: operating delays, operating standbys, scheduled loss and unscheduled loss. The total delay duration for each month was then broken down into the various delay categories, and presented in the stacked bar graph as seen in Figure 2, which demonstrates their relative proportions. Wait-for-truck delays exist as a subcategory of operating standby delays. The graph in Figure 2 shows a separate series of stacked bars alongside the total delay durations for each month. These bars represent the total operating delay MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Fig 3 - Operating standby crusher delays. operating standby delay analysis – mine: waitfor-crusher When analysing the operating standby delay breakdown from the production side of the mine-plant interface, it was expected that wait-for-crusher delays would constitute the largest portion of delays. The results of the operating standby 9 v collInS and M S kIzIl delays for the Brockman 4 production truck fleet are shown in Figure 4. As expected, the wait-for-crusher delays constituted the largest portion of operating standby delays, representing 44 per cent of the total delay duration. TABLE 1 Economical analysis. Month Potential utilisation increase (%) Potential production increase (t) Potential revenue increase @$100/t January 5 143 389 14 338 900 February 6 143 603 14 360 300 March 4 105 034 10 503 400 April 3 73 680 7 368 000 May 4 99 158 9 915 800 June 4 86 016 8 601 600 TOTAL Fig 4 - Operating standby truck delays. key delay periods Figure 5 shows the average hourly production throughput for the crusher over the course of the day over the six month period. From the graph, it was evident that there were significant production throughput loses at 6 am, 12 pm, 6 pm and 12 am. These significant production drops were attributed to shift change (6 am and 6 pm) and production lunch break (12 pm and 12 am) for both day and night shift. Truck driver change-outs during shift change led to a recurring window where throughput ceased. Lunch breaks also contributed to the significant slumps in production. 65 088 000 1. as the potential revenue increase that could have been achieved at Brockman 4 in the first six months of 2012 2. the revenue lost at Brockman 4 due to wait-for-truck delays in the first six months of 2012. From the analysis conducted, it was found that approximately A$65 M was lost due to wait-for-truck delays over the six month period. This potential gain easily justified the cost involved in implementing various methods and solutions to achieve this. talpac analysis – input variables To run the TALPAC simulations, a series of input data was required. This data input included: equipment availability equipment machine models and rated capacity haul segment lengths, gradients and curve angles rolling resistances speed limits per segment. Haul road routes, digitised in Vulcan were imported to TALPAC for haulage simulations using a rolling resistance of three and the following speed restrictions: • • • • • 60 km/h empty 40 km/h loaded on flats 20 km/h downhill/uphill whilst loaded 10 km/h around corners. For the purpose of this exercise, two Komatsu 730E were used as the fixed variable for the mixed-fleet haulage scenarios as this was the number of Komatsu 730Es on site as of June, the month from which the haul roads were modelled. The number of Komatsu 830Es was then allowed to run on ‘auto’ in order to determine the ideal number required for each digger. • • • • Fig 5 - Average hourly crusher throughput. economical analysis – impact of eliminating wait-for-truck delays Using the data obtained between 1 January and 1 July, 2012, an economic analysis was conducted on the impact of eliminating wait-on-truck delays. By hypothetically eliminating the waitfor-truck delays from the data obtained, the monthly increase in utilisation of the crusher was determined. Using a crusher rating of 3600 tonnes per hour, the potential increase in throughput each month was then quantified. Using an iron ore price of A$100 per tonne, potential revenue increases were estimated for each month as shown in Table 1. These revenue values can be seen from two different perspectives: 10 truck-shovel matching outcomes In order to determine the ideal number of trucks for each of the dig units, parameters for the longest haul route from each of the three pits was used. From Pit 2, Pit 3 Central and Pit 3 East (P02I, P03C and P03E respectively), the longest haul routes determined were the 530, 540 and 590 RL (relative level) respectively. The series of simulations run using TALPAC involved a mixed truck fleet consisting of both Komatsu 830Es and Komatsu 730Es. Each of the different dig units were placed in each pit one at a time and the ideal number of trucks for the operation to be considered truck-shovel matched was conducted. The truck numbers corresponding to non-mixed fleet haulage can be found in Table 2. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 MaXIMISIng productIon at BrockMan 4 By MInIMISIng truck delayS at the cruSher TABLE 2 Mixed fleet truck numbers (Komatsu 830Es when two Komatsu 730Es were fixed). Pit Loader Letourneau L-1850 Shovel Hitachi 5500 Excavator Hitachi 3600 P02I 11 15 11 P03E 6 6 6 P03C 5 6 5 alleviating wait-for-crusher delays. This material then provides the loader with feed that it can transfer between the ROM pad and the crusher when there is another gap in the truck cycle. An area alongside the crusher approximating 60 000 m3 is already available at Brockman 4. A photo of this ROM pad can be seen in Figure 6. Brockman 4 has an annual production rate of 22 million tonnes ore with 44 million tonnes of waste. When simulations were run placing each type of dig unit in each of the three pits and analysing the production rates that could be achieved, the results shown in Table 3 were achieved. TABLE 3 Mixed fleet production capabilities. Mixed Loader Letourneau L-1850 Shovel Hitachi 5500 Excavator Hitachi 3600 P02I 18 742 632 26 668 760 19 795 772 P03E 18 814 835 26 978 261 19 924 772 P03C 19 147 615 27 174 059 20 101 023 As shown in Table 3, the Hitachi 5500 is capable of moving the largest number of production tonnes. At Brockman 4, each shovel and every truck mines high-grade, low-grade and waste. Trucks and shovels are not assigned to one particular type of material all year round, and thus, it is necessary to analyse how much material can be moved using the current fleet. The required truck numbers given in Table 2 and the production rate calculated in Table 3 were used in order to determine the minimum number of trucks required to move 66 million tonnes of material per annum. The smallest number of trucks required can be achieved if two Hitachi 5500 and one Hitachi 3600 excavators were used, which will require 21 Komatsu 830E and 2 Komatsu 730E trucks. Brockman 4 currently has 26 Komatsu 830E and 2 Komatsu 730E trucks. According to the TALPAC analysis, the Brockman 4 operation appears to be over-trucked. Over-trucking can therefore be considered to be one of the factors contributing to truck delays at the crusher and consequently the constantly out-of-sync truck cycle at Brockman 4. It should be noted that the implications of having different capacity trucks will inevitably result in longer queuing times and a more frequent chance of truck cycle disruptions. delay reductIon StrategIeS utilising a loader on run-of-mine One possible way to alleviate crusher delays is to utilise a loader on the ROM. If the loader were to remain on the ROM at all times, it could supply feed to the crusher when no trucks are in the vicinity of the crusher, thereby minimising wait-fortruck delays. Conversely, if there is already a truck waiting at the crusher, the incoming truck can dump on the ROM pad, thereby MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Fig 6 - Run-of-mine pad. The graph in Figure 5 illustrates the key times that the ROM loader can be further utilised to counteract production slumps. Therefore, not only can it be used to minimise wait-for-truck and wait-for-crusher delays, but also to increase significant production slumps during crew shift change or lunch breaks. Additionally, having a permanent ROM loader would help the crusher throughput increase from approximately 3000 tonnes per hour toward the crusher rating of 3600 tonnes per hour. The approximate cost of incorporating a permanent ROM loader at Brockman 4 operations can be found in Table 4, with input from R2Mining (2011) and an estimated average loader availability of 85 per cent. TABLE 4 Initial cost associated with ROM loader. Component Cost (A$) Capital cost $550 000 $550 000 $250/h × 0.85 × 365 × 24 $1 861 500/year $120 000/year $240 000/year Operating costs per year Personnel to operate equipment Cost in first year acquired Total $2 651 500 When the potential increase in revenue that can be gained by eliminating wait-for-truck delays (as determined in Table 1) is compared to the cost to acquire, operate and man a loader, it is evident that the acquisition of a permanent ROM loader is easily justified. Approximately A$65 M can be saved over a six month period, and thus a potential of A$130 M each year if wait-for-truck delays are essentially eliminated. The A$2 651 500 that it would cost to attain and operate a loader in the first year (and less in subsequent years) is insignificant and would be well worth the investment. hopper near crusher Another option that may help to reduce wait-for-truck and wait-for-crusher delays is to install a hopper system alongside the crusher. This system is similar to dumping on the ROM, except that the ore from the hopper can be transferred to the crusher instantaneously and does not need to be rehandled. 11 V Collins and M S KiziL When trucks are unavailable to tip at the crusher, the hopper system can be activated and material can immediately be fed at a constant rate until the next truck arrives. If the crusher is at full capacity, trucks then have the choice to dump on the ROM or in the hopper and immediately return to the truck cycle. The cost of implementing a hopper system would be far outweighed by the potential increase in revenue that could be achieved using this system. Table 5 outlines the approximate cost involved with implementing a hopper system with input from R2Mining (2011). TABLE 5 Hopper system cost analysis. Component 250 tonne hopper Operating costs per year Operators Total ($) 450 000 450 000 15/h × 0.90 × 365 × 24 118 260/year 120 000/year 240 000/year 808 260 Similarly, when the potential increase in revenue that can be gained by eliminating wait-for-truck delays is compared to the cost to acquire, install and operate a hopper system, the costs involved are easily justified. These costs were determined with input from R2Mining (2011) and would be considered insignificant compared to the A$130 million that could be saved by implementing it. Integrating Modular and Citect After the implementation of the Modular-Citect integration at Paraburdoo, numerous improvements to the mine-to-plant process were achieved. The integrated system would be able to accurately determine when to cue the loader to feed the crusher and when to cue the truck to dump at an alternative location. This is the image of production at Brockman 4 in the near future. If the implementation of Modular and Citect proves to be feasible after further data analysis, production increases of up to 1300 tonnes per shift may result. Using a conservative iron ore price of $100/tonne, this could see revenue increases of up to A$94.9 M per year. CONCLUSIONS By categorising delays from either side of the mine-plant interface, it was possible to quantify the significance of waitfor-crusher and wait-for-truck delays at Brockman 4 (44 and 62 per cent respectively). The data analysis showed a decrease in wait-for-truck delays from January to June, 2012 as a result of gradually eliminating the use of Komatsu 730E trucks. When comparing these observations to previous research (Kizil, Knights and Nel, 2011), it could be concluded that the reduction in wait-for-truck times over the sample period is attributed to the gradual reduction in mixed fleet haulage over this time. By analysing Modular data outputs over a six month period, it was possible to determine where the production cycle at 12 The economical analysis also determined that if the waitfor-truck delays were removed from the production cycle, there would be a potential increase in revenue of A$130 M/a for the company. Acknowledgements Cost (A$) Cost in first year acquired: Brockman 4 could be improved. By reassessing the truck cycle at Brockman 4 as a preliminary step towards minimising delays at the crusher and increasing throughput, it is to be expected that Brockman 4 will see significant improvements to their operations. The interdependent nature of mining operations, including the truck and shovel method, highlights the importance of maintaining each link in the mine-to-plant production chain. The authors acknowledge the support and guidance provided by Dr Basil Beamish and industry professional Gareth Banfield. References Banfield, G, 2012. Personal communication, 11 May. Bonates, E and Lizotte, Y, 1988. A computer simulation model to evaluate the effect of dispatching, International Journal of Surface Mining, 2:99-104. Georgieff, D, 2006. Improving truck utilization during shovel delays, undergraduate thesis (unpublished), University of Queensland, Brisbane. Holke, B, 2004. Planning and application of hydraulic shovels in open pit mines: shovels/truck operation and shovel/crusher operation, Journal of Mines, Metals and Fuels (0022-2755), 34(4). Kizil, M, Knights, P and Nel, S, 2011. Improving truck-shovel matching, in Proceedings 35th APCOM Symposium, pp 381-391 (The Australasian Institute of Mining and Metallurgy: Melbourne). Oberisser, H, 2010. Fully mobile IPCC solutions [online]. Available from: <http://www.mining magazine.com/equipment/fullymobile-ipcc-solutions> [Accessed: 26 April, 2012]. Olsen, J P and White, J, 1992. On improving truck/shovel productivity in open pit mines, in Proceedings 23rd International Symposium 1992 – Application of Computers and Operations Research (SME: Arizona). R2Mining, 2011. Australia mine and mill equipment costs [online]. Available from: <http://calc2011.au.costs.infomine.com/ Projects/projectlist.aspx> [Accessed 30 September, 2012]. Rio Tinto Iron Ore, 2011a. ER crusher Dispatch/Citect integration project valuation (unpublished), Paraburdoo Mine Site, Paraburdoo. Rio Tinto Iron Ore, 2011b. Paraburdoo Modular-Citect interface: automated truck reassignment and loader reassignment (unpublished), Paraburdoo Mine Site, Paraburdoo. Runge, 1993. Section 6: Truck and loader productivity estimation, section 7: factors affecting truck and loader productivity, section 10: sample TALPAC output, Planning and Operation of Truck and loader Mining Systems – Technical Training Course Notes, Course Ref: 1593, pp 59-81, 83-100, 125-141 (Runge Mining (Australia) Pty Ltd: Mackay). Schneider Electric, 2012. Products and Services [online]. Available from: <http://www. schneider -electric.com/site/home/index. cfm/ww/> [Accessed 11 May, 2012]. Mining Education Australia – Research Projects Review 2012 Grizzly Modifications at Ridgeway Deeps Block Cave Mine C Costello1 and P Knights2 ABSTRACT Grizzlies are used in underground mining to act as size separators at the crusher tipping stations. Acceptable size ore will fall through the apertures and oversize ore will remain on the screen for secondary breakage. At Ridgeway Deeps block cave mine the rock breaker operating on the oversize at the grizzlies is causing approximately two hours per shift downtime. This is vastly affecting production rates. This paper presents the results of an investigation to modify the static grizzly aperture design to reduce downtime caused by the rock breaker attending to oversize rocks. This paper reviews existing alternatives to underground grizzlies and proposes a new design. The new design has taken into consideration the conclusions found through research and data analysis conducted on the tipple stations at Ridgeway Deeps gold mine. A cost benefit analysis of installing new grizzlies was conducted considering the remaining life-of-mine of four years. Introduction Grizzlies are used in underground mining to act as size separators at the crusher tipping stations. Acceptable size ore will fall through the apertures of a selected size, and anything too big (oversize) will remain on the screen for secondary breakage. Due to the arrangement of the tipple stations at Ridgeway Deeps (RWD) the rock breaker cannot act independently of the LHD’s tipping, so therefore the LHDs must stop tipping each time the grizzly has a blockage or a build-up of oversize ore. RWD is suffering from an average of two hours downtime per shift due to the rock breaker operation. An investigation was undertaken to determine alternatives to modifying the current grizzly and how other block cave mines have dealt with this problem. As RWD is an ‘experimental’ mine for Newcrest it is important that the larger neighbouring Cadia panel cave mine base their designs upon lessons learnt from Ridgeway. This study developed an alternative design based upon the findings of investigations into: •• •• •• •• •• RWD feasibility study crushers different shapes of oversize occurring on the grizzlies frequency of the oversize shapes occurring time required to break the different shaped rocks. Material handling system Overview The Ridgeway gold-copper mine, owned by Newcrest Mining Limited, is a block cave (BC) mine which was developed under a pre-existing sublevel cave (SLC) mine (Brunton, Sharrock and Lett, 2011). The material handling system (MHS) is integrated with the existing truck and portal conveying system to deliver ore to the surface stockpile (Maunsell, 2007). The primary crushing system consists of two single toggle jaw crushers (nominal size 2000 × 1500 mm). The significant changes from SLC (which has a single gyratory crusher) to BC mining impacted the crusher selection as follows: •• BC fragmentation – the models predicted significantly coarser material compared to SLC •• increased footprint – simulation in the prefeasibility indicated that two tipping stations approximately 200 m apart in bypass drives emanating from the southern perimeter, with run-of-mine (ROM) ore transition, drawpoints to crusher, performed by four to five LHDs is required to achieve production •• geotechnical constraints limit the size of the crusher chambers and their location relative to the orebody (Maunsell, 2007). The feasibility study outlined that from these conditions a two-stage crushing philosophy was implemented adopting two jaw crushers fed with preconditioned ore via static 1200 mm square grizzlies with fixed hydraulic rock breakers at the tipping stations. Alternate crusher types including gyratory crushers, jaw-gyratory crushers and rotary breaker type were also considered for primary crusher duties. The rotary breaker type was deemed unproven for primary crushing duty for abrasive hard rock application, and discounted from any comparisons due to lack of relevant benchmarking applications. Gyratory and jay-gyratory crushers were found to return low business reliability ratings (Allman, Dunstan and Syme, 2007). Fragmentation The predicted cave fragmentation as per the feasibility study for RWD BC reveals that the initial fragmentation 1. SAusIMM, School of Mechanical and Mining Engineering, The University of Queensland, Brisbane Qld 4072. Email: [email protected] 2. MAusIMM, Head of Division of Mining Engineering, School of Mechanical and Mining Engineering, The University of Queensland, Brisbane Qld 4072. Email: [email protected] Mining Education Australia – Research Projects Review 2012 13 c coStello and p knIghtS was expected to be coarse, becoming finer as the cave propagates. The crusher size selection was based upon the coarse prediction to minimise the requirement for secondary breakage at the drawpoints. The BC fragmentation geometry was predicted to be 1:0.75:0.5 (x, y, z). Secondary breakage would be required at the drawpoints for rocks > 3 m3 (y > 1500 mm) (Maunsell, 2007). The cave fragmentation sizing indicated at the time of the feasibility report that a crusher feed F80 of approximately 1480 mm. The crusher nominal size is 1200 mm square, therefore, the grizzlies were implemented to reduce the chance of bridging at the crusher. grIzzlIeS overview A grizzly is a screen that acts as a scalping device for oversize material (Darling, 2011). Very coarse material is usually screened on a grizzly, which in its simplest form consists of heavy parallel bars set in a frame. The most common use of grizzlies is the feed to a primary crusher. Clogging is a major problem, especially on the cross-members that hold the bars together (Darling, 2011). Square mesh grizzlies are used to get a more accurate maximum size of material, whereas rectangular and non-gridded grizzlies can allow material of larger size to pass through. Heavy duty grizzly bars are cast from manganese steel and have double tapers. grizzly design When designing a grizzly for a specific purpose, the openings between the grizzly bars should be commensurate with the size of the receiving hopper where the product has to be discharged (Gupta and Yan, 2006). Stationary screens are operated either at horizontal or inclined planes. A relatively steep installation is preferred for higher throughputs but the quality of separation is likely to be affected as the effective aperture and open area are decreased. During the process of screening, particles either fall through the apertures or get held back. Obviously larger particles than the aperture opening cannot pass through (Gupta and Yan, 2006). Size and shape of particles A fraction of particles, although smaller than the aperture also do not pass through the first time they encounter an aperture. Particles that are elongated, but with cross-section less than the aperture, will pass through provided they approach the aperture at an appropriate angle. Figure 1 shows the effect of shape and size of particles through screening. Both A and Fig 1 - Particle size and shape at screen surface (Gupta and Yan, 2006). 14 C particles are prevented from passing through, A being larger in size and C elongated. Particle C will, however, pass through if rotated to look like particle D. Particle B will always pass through. Thus both shape and size are of importance in a screening operation. Particle sizes that are near to the aperture size are the most difficult to screen. This is defined as particles having a size 0.75 to 1.5 times the aperture are the most difficult to screen (Gupta and Yan, 2006). The passage of each particle will also depend on the angle at which it reaches the aperture. If the screen were sufficiently long it could eventually approach the aperture at the appropriate angle and pass through. If the length was insufficient then in spite of the particle being smaller than the aperture it may report as oversize (Gupta and Yan, 2006). Blinding or clogging of screens during operation is one of the most contentious and difficult factors that a screen designer has to face. The most effective way to reduce clogging is to impart vibratory or circular motion to the screen. To impart the motion, the screen surfaces are rigidly fixed onto a frame, the frame in turn is fixed to moving devices that are either mechanically or electrically driven (Gupta and Yan, 2006). alternatIveS to grIzzlIeS underground A primary concern in underground mining is where and how to handle ‘moveable’ oversize. The oversize can be: handled at the extraction points moved to a special gallery for blasting moved to an orepass equipped with a grizzly and handled there • directly dumped into an orepass for later handling. All variations are used, and each company has its own philosophy in this regard (Bullock and Hustrulid, 2001). Palabora BC mine in South Africa and Northparkes BC mine in New South Wales both have the same crushing level design as RWD BC mine in that the ore is delivered straight to the crusher; however, they have no grizzlies. • • • northparkes block cave mine As part of the E26 Lift 2 mine, a single jaw-gyratory crusher was installed. The primary advantage of this size of crusher over conventional crushers is its ability to accept a maximum rock size of 3 m3 whilst achieving consistent P80 of 120 mm. As a result, a grizzly screen is not required on the crusher feed bin, although a rock breaker is fitted to the crusher to address periodic blockages in the crusher and on the plate feeder, which can be seen in Figure 2 (Butcher et al, 2011). Fig 2 - Northparkes E48 crusher chamber with rock breaker. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 grIzzly ModIfIcatIonS at rIdgeway deepS Block cave MIne Rio Tinto’s philosophy was to minimise the stages in ore block size reduction, size the crusher to handle the largest rock size that the LHD could carry, and install a continuous system ‘rock factory’ concept. It was the first time the jawgyratory crusher was installed in an underground mine. It was the preferred option as it accepts a larger feed size than Northparkes’ current crushers and produces a finer product (Duffield, 2001). palabora block cave mine The feasibility report for Palabora depicts the LHDs will be tipping onto grizzlies at the bottom of an 8 m orepass. The grizzlies at the tipples have been designed such that any oversize material would slide to the central breaking area between the grizzlies. The tips would be monitored by CCTV and thus the control centre would dispatch one of the mobile rock breakers to the tip if this should occur. The rock breaker would then proceed to break the boulder. The broken material would then fall into the grizzlies on either side of the breaking area. No allowance was made for delays due to oversize rocks being removed from the grizzly as it was assumed that rocks would migrate down the grizzly to the breaking area where they would be broken by the mobile rock breakers (Palabora Mining Company Limited, 1996). The ore is then crushed by a Krupp 1700 mm × 2300 mm double toggle jaw crusher (Maiwasha, Ngidi and Pretorius, 2010). The grizzlies were reportedly removed by 2007 due to the 8 m drop height causing excessive damage and production downtime at the crusher. Oversize is now handled in the levels (Capes, 2007). has stopped, hence the downtime. The following occurrences can activate a red light at the tipple stations: ROM bin level high (bin greater than 80 per cent full) grizzly airlock door alarms (door to the bird cage is opened) • rock breaker in operation. Another cause of downtime recorded in Citect is when maintenance shutdowns occur. At RWD there are two shutdowns per month. When a shut occurs Citect sometimes records the rock breaker as operating (out of park position); however, it is in fact only out of park position for maintenance rather than actually in operation. • • Whilst carrying out data analysis the following steps were taken to ensure accuracy of results: Each time a maintenance shut occurred the downtime data was discounted as it would incorrectly bias the average. • If both the high bins and rock breaker in-use alarms were activated at the same time only the minutes from the high bins would count. This is because if the bins are full LHDs aren’t allowed to tip anyway so if the rock breaker is not the cause of the downtime. This will allow for accurate quantification of when the rock breaker is causing downtime. Through analysing data from 13 December 2011 to 13 February 2012 it was concluded that both the east and west tipples clocked just over four hours of downtime per shift on average. Figure 3 displays the different triggers for red lights and the average downtime they caused per shift for this time period. • removal of grizzlies The removal of the grizzlies at Northparkes and Palabora has proved successful for both mines (Butcher et al, 2011). Northparkes made the initial decision to not include grizzlies at the feasibility stage as the crusher could handle the largest size rock a LHD could transport. Palabora, similar to RWD suffered from downtime due to the rock breaker. Palabora educated their workforce on identifying oversize, relying upon the capability of their crushers, and the effect of the impact resulting from an 8 m drop to the ore bin to assist breakage (Capes, 2007). It can be concluded that the requirement of a grizzly rests upon the crusher requirements and specifications. Typically the maximum rock size the crusher will accept and crush without bridging is approximately 90 per cent of the crusher gape for a single toggle machine. The largest jaw crusher gape is around the 1500 mm mark, which results in the largest practical rock size that could be accepted by the crusher of approximately 1350 mm in size. The cave fragmentation predicts an F80 of 1480 mm, which is greater than the nominal size (Maunsell, 2007). Therefore, a grizzly scalping screen is required to guard against damage to the crusher. reSultS and dIScuSSIon estimating downtime The initial data was collected using Citect to calculate the total downtime at the tipping stations and the cause of the downtime. Downtime at the tipple stations has been defined as any time the red traffic lights are activated. A set of traffic lights are installed at each access point to the tipples so the LHDs know when to tip (green light) or wait (red light). Anytime a LHD is waiting to tip due to a red light, production MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Fig 3 - Average downtime caused by three triggers of the red lights over two months. It can be seen that the biggest contributor to downtime on average, is the rock breaker at both the east and west tipples. On average the rock breaker causes two hours of downtime per shift. It can be concluded from these findings that in order to increase productivity the root cause of why the rock breaker is operating excessively will need to be identified. It is to be noted that the cause for high bins could be due to downtime further along the MHS, such as the conveyor or crusher; however, this is outside the scope of the investigation. The focus is on what occurs above the grizzly panels. rock behaviour analysis In order to design a grizzly that would reduce downtime, the problem with the current grizzly design first needs to be identified. To do this data was collected using the CCTV footage of the tipples. The footage was initially reviewed in segments of 24 hours for 60 days. Figure 4 displays a snapshot of the footage that was reviewed. Different types of analysis carried out using the footage has determined: 15 c coStello and p knIghtS across the oversize rock shapes, and is reduced in the jammed category as the rock breaker merely needs to move the rocks rather than break any. It can be concluded from these results that the most problematic rock shape is round and nesting-round. As these rock shapes also take longer than one minute and 26 seconds to break, these are the rock shapes that will be targeted in the modified grizzly design. As the nesting due to round rocks only occur due to the initial plug round rock, if the round rock shape problem can be alleviated there will only be a small fraction of oversize occurring on the grizzly panels. tipping style An investigation regarding if the style of tipping affects oversize was also conducted. Three tipping styles were identified: • Fig 4 - Snapshot of the east tipple station. the shapes of the rocks that occur on the panels the frequency of the shapes of rocks the location on the grizzly panel each rock shape occurs the average time required by the rock breaker to break each shape of rock • the effect of different tipping styles from the LHD buckets. The CCTV footage has been examined using different categories of oversize rocks. The different shapes of rocks being examined are: • • • • round: oversize rock (>1200 mm) that is round in shape nesting – round: nesting or clogging, caused by a plug round rock • nesting – jammed: nesting or clogging, caused by rocks of acceptable size jamming together upon impact resulting in a blockage • slab – oversize (>1200 mm)rock that is of rectangular shape • nesting – slab: nesting or clogging, caused by a slab rock. By understanding the most problematic rock shapes in the highest frequented area will allow for the most effective grizzly design. The results from the analysis can be seen in Figure 5. • • Drop – this is when the LHD stops at the grizzly, raises the bucket and dumps the load in one position with most of the material landing in the one area (ie zones 2 and 7 if the LHD came from the left, refer to Figure 6 for grizzly zones). Fig 6 - Grizzly zones. Spill – similar to drop except the bucket is dumped slower allowing for the material to move around and spread out as it is being dumped. • Launch – occurs when the LHD doesn’t fully stop at the tipple. The LHD drives in and starts emptying the bucket, then begins reversing. This causes the material to ‘launch’ over the grizzlies allowing the material to scatter and distribute more between the holes. Figure 7 outlines the results indicating the type of tipping style and the grizzly zone in which the oversize occurred. • Fig 7 - Location of oversize relative to the tipping style of the LHD. Fig 5 - Frequency of rock shapes versus the average time to break them. It can be seen that nesting due to round rocks and round oversize rocks are on average the highest frequented rock shapes. The time to break each rock shape is fairly similar 16 It can be seen that the spilling method was prevalent and the middle four holes are the most frequented. The data is, however, inconclusive. There were too many variables to be considered for this study, including: different operators, which side of the tipple the LHD approached from, and whether to only record plug rocks or the loads that contributed to nesting as well. Data was only recorded for six shifts. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 grIzzly ModIfIcatIonS at rIdgeway deepS Block cave MIne trial – minor modification An initial trial of a modified grizzly was carried out. Two old grizzly panels were installed on the west tipples. The old panels consisted of three holes approximately 900 mm × 1000 mm in size. The bars separating the holes were removed in order to increase passing size to approximately 3000 mm × 900 mm. The west tipple with the modified grizzly panels can be seen in Figure 8. any component further along the MHS by allowing a greater rock size through. fIndIngS froM analySIS The following conclusions have been made from the initial data analysis and preliminary trials: • • • • • Fig 8 - Minor modification to grizzly panels. The purpose of this trial was to: evaluate that if by increasing the passing size less oversize would accumulate on the grizzly thereby increasing productivity • monitor the effects of potential rock sizes of 2 m2 damaging or bridging on the crusher gape. Data for this modification was accumulated for 28 shifts (14 days). It was found that downtime was reduced marginally by six per cent. Keeping in mind that only half the grizzly panels were modified, this was an encouraging outcome. The results can be seen in Figure 9. • • • • The rock breaker operation was causing approximately two hours of downtime per shift. The cause of the excessive rock breaker operation was due to oversize on the grizzlies. The most problematic rock shapes were round oversize rocks which also cause nesting. These rock shapes also require more time to break, making them a target in the modified design. If the round rocks were no longer a problem it would free 79 per cent of the rock breaker time, allowing production time to increase by approximately 1.5 hours. The oversize predominantly occurs in the middle four holes of the grizzly panels, and this is where the most wear and tear occurs. The most common tipping style is spilling; however, due to data collection methods the tipping style analysis is inconclusive toward the objectives of this research project. The initial modified trial found that increasing passing size helped to reduce downtime, however only by six per cent. Concerns regarding large rocks passing through and damaging or bridging on the crusher gape were disregarded. The option to remove the grizzlies altogether was disregarded due to the crusher requirements. fInal ModIfIed deSIgn The final modified design takes into account findings from the literature review, initial data analysis and the preliminary trials. Figure 10 displays the modified design. Fig 10 - Final modified grizzly design (1 and 2 represent transverse and cross bar designs). The modified grizzly design encompasses the same area as the old design so no new excavations are required for installation. It is a static design with eight apertures similar to the old design. However, the apertures are square, unlike the previous holes, which were filleted and can be seen for comparison in Figure 11. Fig 9 - Downtime comparison between the new and old grizzly. It was concluded from this investigation that by increasing the passing size, less oversize accumulated on the grizzlies, thereby reducing downtime by not requiring the rock breaker. It was also found that no damage was done to the crusher or MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Removing the fillets allows for greater throughput and reduced flat surface area in order to alleviate room for nesting to occur. All the bars are now angled; this has been done for two reasons, firstly to alleviate nesting by removing flat surfaces where material can build up and secondly, having an angled impact zone for the material allows for the material to potentially be broken up more and moved around before passing through the apertures. The aperture size has been 17 c coStello and p knIghtS Investigation into the grizzly zones showed oversize occurred and the effect of LHD tipping styles was found irrelevant and had no impact on the final design. • An initial trial of a slightly modified grizzly found that by increasing passing size downtime was reduced marginally. The modified grizzly design was based on the identified problems and initial results. The final design has targeted round rock shapes and nesting through a larger passing size and eliminating flat surfaces where nesting could occur. A cost benefit analysis was carried out and it has been found beneficial to manufacture and trial one modified grizzly design. • Fig 11 - Original grizzly schematics. increased to 1200 × 1330 mm, this is a 16 per cent increase in passing size. The maximum rock size passable is 1.27 m3, which is still under the maximum rock size allowable into the crusher. The total weight of the grizzly is approximately 22.5 tons. A quote was sourced from an Australian company for manufacturing the grizzly at a cost of $278 750. cost benefit analysis As the remaining life-of-mine is only four years, it is important to calculate if the cost of implementing a new grizzly will be beneficial. As the apertures have been made larger a greater amount of material can pass through. It has been estimated that there would be an increase of 7.7 per cent passing feed for the new grizzly. This increase in throughput has been used to calculate the net present value (NPV) base on the revenue the new grizzly could make. Over the remaining four year mine life, the net present value of replacing the east and west grizzlies with the new design was estimated at $4.8 M. concluSIonS An investigation into the grizzlies at Ridgeway Deeps mine was undertaken in order to design a modified grizzly design that will reduce downtime. The chosen research topic has proved difficult to investigate as the design requirements are site specific. The dimensions and/or requirement of a grizzly are based upon the crusher feed requirements. Preliminary research found that the crushers at RWD can accept a maximum rock size (90 per cent of the crusher gape) of 1350 mm. The option of removing the grizzlies was investigated as it has proven successful for other block cave mines. It was found, however, that due to the predicted cave fragmentation and the crusher capacity grizzlies are required. The original grizzlies consisted of eight round apertures 1200 × 1200 mm in size. The data gathered produced the following conclusions: • • 18 The rock breaker in operation is the biggest contributor to downtime (approximately two hours per shift). Analysis of the oversize rocks indicated that round rock shapes are the most problematic rock as they are the most frequented type of oversize. Round rocks were also found to cause the most nesting. Newcrest Mining Limited has manufactured one modified design and it will be implemented on the west tipple for trial. Newcrest will monitor the downtime at the tipple and also the production tonnes. Further research is recommended to monitor any change in crusher downtime resulting from the new design. referenceS Allman, A, Dunstan, G and Syme, T, 2007. Ridgeway Deeps feasibility study, Newcrest Mining Limited. Brunton, I, Sharrock, G and Lett, J, 2011. Full scale near field flow behaviour at the Ridgeway Deeps block cave mine, in Proceedings MassMin 2012 (Canadian Institute of Mining, Metallurgy and Petroleum: Montreal). Bullock, R L and Hustrulid, W A, 2001. Underground Mining Methods: Engineering Fundamentals and International Case Studies, 718 p (Society for Mining, Metallurgy, and Exploration: Littleton). Butcher, A, Cunningham, R, Edwards, K, Lye, A, Simmons, J, Stegman, C and Wyllie, A, 2011. Northparkes Mine [online], The Australasian Institute of Mining and Metallurgy. Available from: <http://www.ausimm.com.au/content/docs/northparkes_ example_paper.pdf> [Accessed: 16 March 2012]. Capes, G, 2007. Palabora 2010 visit: Production and secondary break benchmark, Newcrest Mining Limited. Darling, P, 2011. SME Mining Engineering Handbook, third edition, 1835 p (Society for Mining, Metallurgy, and Exploration: Littleton). Duffield, S, 2001. Design of the second block cave at Northparkes E26 Mine, in Proceedings MassMin 2000, pp 35-46 (The Australasian Institute of Mining and Metallurgy: Melbourne). Gupta, A and Yan, D, 2006. Mineral Processing Design and Operation – An Introduction, pp 293–353 (Elsevier BV: Amsterdam). Maiwasha, C, Ngidi, N and Pretorius, D, 2010. Competent persons report on the mineral assets of Palabora Mining Company Limited, South Africa [online]. South Africa, Rio Tinto. Available from: <http://www.jse.co.za/Libraries/JSE_Documents_and_ Statistics_-_CPR_-_Rio_Tinto/Palabora_Mining_Company.sflb. ashx> [Accessed: 19 April 2012]. Maunsell, 2007. Ridgeway Deeps Mine – MHS and fixed infrastructurefeasibility study report, Maunsell Pty Ltd, Brisbane. Palabora Mining Company Limited, 1996. Palabora feasibility report, Palabora Mining Company Limited. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Multi-Objective Optimisation of Mining-Metallurgical Systems V Golding1, D Jafari2, B Rajopadhyaya3, D Sundquist4 and E Chanda5 Abstract Optimisation of integrated mining and metallurgical systems poses some significant challenges due to the conflicting nature of business objectives. Traditionally, mine planners would seek to optimise the net present value (NPV) of a project, but the issue of sustainability in mining requires consideration be given to a wider range of goals. Standard linear programming can optimise a system, based on a single objective such as maximising the NPV. However, the single objective function approach is not adequate to accommodate multiple objectives that are often conflicting in nature. A mine optimisation model should be formulated in such a way as to include multiple objectives and applicable constraints. A model that can simultaneously optimise a number of conflicting objectives is required to handle such complexity in mine planning. Multi-objective optimisation is a linear programming method, which can handle mine optimisation problems with multiple goals. In this paper, a weighted goal programming (WGP) approach was used, whereby each objective function had priority over the other(s), depending on corporate decisions. The aim was to develop a mine optimisation model that includes energy, water consumption and equivalent carbon emissions as objective functions. The model is based on WGP and is used to find the optimum material flow rates in a network model of a large-scale mining and metallurgical operation. The network system and relevant data is based on a large-scale open pit mining operation to add realistic values to the generic system. The LINGO software package was used for the analysis. Preliminary results show that the model can successfully consider the different objective function goals, weights and constraints to optimise the given network by changing resource utilisation. The program is quick to solve and useful for undertaking what-if analyses. INTRODUCTION An essential part of the mine planning process is optimising the net present value (NPV) for the operation (Rafiee and Asgahri, 2008; Topal and Ramazan, 2012). The NPV essentially determines the value of the operation and helps to predict the life of the mining operation. Whilst this is a valuable tool in the planning process, the optimisation is simplistic in that it only considers a single objective. This single objective optimisation (SOO) provides a single, unique solution for the production schedule of the operation, but does not consider some of the operational constraints that have the potential to significantly alter the capacity and practicality of the mining operation. Such operational constraints could be in the form of energy consumption, carbon emission and water use. SOO does not have the capacity to include these factors into the optimisation strategy. (Rangaiah, 2008). This optimisation strategy has the capacity to optimise a number of objective functions, based on specified variables and constraints, and output a range of optimum conditions. Whilst the SOO will output a unique optimum solution in the form of local and global optima, the MOO will have many optimal solutions; the exception is when the objective functions are non-conflicting. In this special case only one unique solution is expected. Real life problems involve several objectives to be considered simultaneously. Often these objectives are conflicting, which means that optimum solution to one objective function might compromise one or more other objective functions. In this respect, MOO is an ideal tool for representing a realistic mining operation as it is able to consider a number of variables within the mining environment. Multi-objective optimisation (MOO), also known as multicriterion optimisation, involves finding the optimum values of decision variables corresponding to all objective functions A mathematical model will be created and applied to a largescale mining-metallurgical operation in South Australia. The model will ensure that the cost associated with carbon, energy 1. SAusIMM, Student, School of Civil Environmental and Mining Engineering, University of Adelaide, SA 5005. Email: [email protected] 2. SAusIMM, Student, School of Civil Environmental and Mining Engineering, University of Adelaide, SA 5005. Email: [email protected] 3. SAusIMM, Student, School of Civil Environmental and Mining Engineering, University of Adelaide, SA 5005. Email: [email protected] 4. SAusIMM, Student, School of Civil Environmental and Mining Engineering, University of Adelaide, SA 5005. Email: [email protected] 5. MAusIMM, Director of Teaching, School of Civil, Environmental and Mining Engineering, University of Adelaide SA 5005. Email: [email protected] Mining Education Australia – Research Projects Review 2012 19 v goldIng et al and water consumptions are minimised while maximising the economic value (NPV). MultI-oBjectIve optIMISatIon MethodS There have been many achievements in optimising single objective functions in the last century. However, MOO has only been thoroughly studied in the past few decades. Many complex organisational problems involve multiple decision functions, which are difficult to model due to a lack of information. These types of planning problems are modelled using multi-objective linear programming (MOLP). There are two major methods for solving MOO problems: generating and preference-based method. The preferencebased method requires inputs and suggestions from the decision makers (DMs). The decision maker is a person or group of people assigned with the task to accept the best optimal solution, based on their experience and knowledge. The generating method involves finding the solution’s space to the MOO hoping that one will be the optimal solution, without any inputs from the DM. These two methods are further divided into subgroups, illustrated in Figure 1 (Rangaiah, 2008). Therefore, it is important to have an algorithm to solve these multi-criterion fractional problems. Kornbluth and Steuer (1981) have used goal programming and multi-objective scalarisation programming to solve multi-criterion fractional objective functions. There are several papers on solving multi-objective fractional functions using either GP or MOLP; however, very few have utilised both methods collectively. Methodology A network was adapted from a typical large-scale mining operation that encompasses a comprehensive miningmetallurgical system. That is, from the extraction of a raw material through to a smelted and refined saleable product. network model The modelling of the mining-metallurgical system involved identifying nodes where a decision must be made about ore utilisation. Nodes are also required where a constraint must be applied, such as the limitations of a smelter or refinery. Processes that do not require a decision to be made, or a constraint implemented, can be simplified by combining associated costs. The revised network for the hypothetical mine is presented in Figure 2. The operation has been modelled for a one-year period. This allows a yearly production schedule to be generated that will satisfy the annual targets for carbon production, water consumption and energy use. Stockpiles have not been considered as an ore source although they could be easily added to the network. The primary purposes of stockpiles are short-term grade control and processing beyond the life of mining, which is not applicable to the time frame being modelled. Model formulation The network represents the flow of ore from one node to another. The ore is extracted, crushed, processed and then concentrated or smelted into a refined product. Fig 1 - Multi-objective optimisation methods (Rangaiah, 2008). Multi-objective goal programming Jones et al (2010) has described multi-objective goal programming (GP) as a model that is based on achieving certain goals assigned by the decision maker. The decision maker will put forward preferences, which are used as tools to solve the decision functions. GP is classified into two major subsets, known as weighted goal programming (WGP) and lexicographic goal programming (LGP) (Tamiz, Jones and Romero, 1998). The WGP assigns weights to the unwanted positive and negative deviations of the decision maker, based on their importance. LGP, also known as pre-emptive goal programming, relies on the existence of the priority levels. The priority levels are the objective functions to be optimised and each priority level will have negative and positive deviation from the goal target. Each objective function will have priority on one or more of the other objective functions. Weighted goal programming has been chosen to formulate the optimisation model. Multi-objective linear fractional goal programming Some real life modelling problems are best dealt with in terms of fractions such as investment, costs, profit per share, etc. 20 If we consider the amount of copper within each arc of the network, we can estimate an equivalent quantity of ore by using the average grade of the deposit. By doing this, the material flowing along the arcs between nodes can be represented as a tonnage as shown in Figure 3. In Figure 3 ‘X’ represents the equivalent amount of ore in tonnes and ‘C’ represents the coefficient associated with costs of carbon, energy, water or the economic value of the arc. The source ‘i’ could represent an ore source (a mine) and the sink ‘j’ a concentrator (Chanda, 2004). decision variables The variable Xij as described above represents the tonnes of equivalent ore from source ‘i’ to sink ‘j.’ These are the decision variables used in the optimisation model. The number of decision variables corresponds to the number of arcs within the described network. For the chosen network, each of the decision variables is described below. objective functions Using the described decision variables, a number of objective functions can be defined. This will allow the MOO analysis to be formulated. For the purposes of this optimisation, the objective functions will focus on carbon emissions, energy use, water use and economic value. X0,1: Tonnage of ore extracted from the pit MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Multi-Objective Optimisation of Mining-Metallurgical Systems Fig 2 - Revised mining network model. X0,2: Tonnage of ore hoisted from underground XR,3: Combined tonnage of pit and underground ore (ie from ROM pad) going to concentrator X3,4: Equivalent tonnage of ore in tails transported to Counter Current separator (hydrometallurgy) X4,5: Equivalent tonnage of ore in underflow from hydrometallurgical plant transported to tailings storage facility X4,6: Equivalent tonnage of ore in overflow from hydrometallurgical plant sent for solvent extraction X6,7: Equivalent tonnage of ore contained within U3O8 solvent X7,8: Equivalent tonnage of ore contained within U3O8 concentrate transported to market X6,9: Equivalent tonnage of ore contained in copper solvent transported to refinery X9,10:: Equivalent tonnage of ore in refined copper transported to market X2,3: Tonnage of ore extracted underground and transported to concentrator X3,11: Equivalent tonnage of ore contained in concentrate leach X11,12: Equivalent tonnage of ore from concentrate leach transported to smelter Mining Education Australia – Research Projects Review 2012 X11,14: Equivalent tonnage of ore in copper concentrate transported to market X12,13: Equivalent tonnage of ore in smelted material transported to refinery X13,10: Equivalent tonnage of ore from refined copper cathode transported to market X13,15: Equivalent tonnage of ore extracted into Anode slimes X15,16: Equivalent tonnage of ore from gold and silver bullion transported to market � Source X ij C ij � Sink Fig 3 - Network representation of material flow. Minimisation of carbon cost The first objective function to be minimised is the amount of carbon produced in the entire system. The amount of ore passing through the process is optimised such that carbon emission through whole system is minimised: MinZ1^Xh = / Cij # Xij i,j (1) where: Cij represents the cost of equivalent carbon emissions in the form of a dollar value per equivalent tonne of ore. Xij represents the material flowing from the respective source ‘i’ to sink ‘j’ in terms of equivalent tonnes of ore. Minimisation of energy usage The second objective function is the amount of energy consumed in the mine. Again, the amount of energy used per tonne of ore through different arcs is known. 21 V Golding et al MinZ 2 ^Xh = / Eij # Xij i,j (2) where: Eij represents the cost of energy use (per kwh) in the form of a dollar value per equivalent tonne of ore. Xij represents the material flowing from the respective Minimisation of water usage The third objective function is the minimisation of water through the system. The water consumption per tonne of ore passing through the different process in the system is known. The optimisation will produce the tonnages of ore passing through the various nodes, which will aim to optimise the water usage in the system. (3) Wij represents the cost of water use (per litre) in the form of a dollar value per equivalent tonne of ore Xij represents the material flowing from the respective source ‘i’ to sink ‘j’ in terms of equivalent tonnes of ore Maximisation of economic value The final objective function is the maximisation of the economic value of the ore through the system. This function will counter the three previous cost functions creating a more realistic scenario. The model will thus be analysing trade-offs such as sacrificing profit to decrease costs. This will optimise the value per tonne of ore through the system. (4) where: Vij represents the present value ($) in the form of a dollar value per equivalent tonne of ore Xij represents the material flowing from the respective source ‘i’ to sink ‘j’ in terms of equivalent tonnes of ore Constraints In order for the model to reflect a real situation it must produce results within a realistic range. That is, it should have appropriate flows of ore through the network, and it has to simulate the environment that a normal mine can operate under. In order to ensure that this is achieved a number of constraints are implemented. These include non-negativity constraints, production constraints and flow constraints. The use of these constraints will be explained further. Non-negativity constraints Material flowing through nodes in the network represents ore as it travels through the stages of mineral processing in the mine. It does not make sense for ore to travel backwards across the arcs. That is, the amount of material moving through each path must be positive. This non-negativity constraint applies to all of the decision variables as described above. Production constraints The amount of material passing through the network cannot exceed the capabilities of the mining operation or processing facilities. Constraints were placed on the relevant nodes in 22 Further constraints are necessary to ensure that the model produces a solution that is realistic and satisfies the conditions of the optimisation process. One example is the definition of flow constraints, which need to be set so that mass is balanced within the network. This ensures that the amount of material entering the system, ie extracted through mining, must equal the amount of material exiting the system through various paths. These include not only the relevant markets, but also outlets such as the tailings storage facility (TSF) or similar waste products. Modelling where: MaxZ 4 ^Xh = / Vij # Xij i,j Further constraints were determined for the capability of the processing systems in the mine. An upper limit was estimated for facilities such as the concentrator, smelter and refinery. Additional constraints source ‘i’ to sink ‘j’ in terms of equivalent tonnes of ore MinZ 3 ^Xh = / Wij # Xij i,j order to ensure that the optimised conditions are realistic in terms of the mining operation. The underground and open pit capacities were estimated at 10 Mt/a and 50 Mt/a, respectively. The multi-objective linear optimisation is analysed and evaluated using the linear programming software called LINGO. This software package is a comprehensive tool designed to make building and solving linear, nonlinear and MOO faster, easier and more efficient. Weighted goal programming (WGP) was determined to be the most suitable method for optimising this system. It provides flexibility in that it allows a decision-maker, ie a mining emgineer to assign preferential weights to the goal functions that are of higher importance, hence, minimising the cost (Tamiz, 1998). Goal functions The summation objective functions are converted to equations by adding the deviation variables (ni, pi) and the goal. The goals are based on hypothetical data for a typical operation. The target values are $90 M, $150 M and $3 M for carbon, energy and water cost objective functions respectively. They represent costs that the company does not want to exceed in a year of operation. The goal for economic value was assumed to be A$2.43 B. Equation 5 shows the four goal functions, ie carbon, energy, water and economic value. / Cij # Xij + n1 - p1 = 90 000 000 i,j / Eij # Xij + n2 - p2 = 150 000 000 i,j / Wij # Xij + n3 - p3 = 3 000 000 i,j / Vij # Xij + n4 - p4 = 2 430 000 000 i,j (5) Achievement function It is required to assign weights to the different deviation variables, based on the importance of their goal functions. For instance, the energy goal function has greater importance over the other costs due to its higher overall cost, therefore, p2 will have a higher weight assigned to it. Similarly, p1 will have lower weight and p3 will have the lowest weight. The deviations are scaled by the goal assigned to their respective goal functions. The achievement function is the sum of all the unwanted deviations, which is minimised (see Equation 6). Min f^ xh = w1 # ` 1 # p1 + w2 90 000 000 j 1 # p2 + w3 150 000 000 j 1 #` # p3 + w4 3 000 000 j 1 #` # n4 2 430 000 000 j #` (6) Mining Education Australia – Research Projects Review 2012 Multi-Objective Optimisation of Mining-Metallurgical Systems where: a value of zero, which means that the target goal has been perfectly reached. w1, w2, w3 and w4 are the weights assigned to the The water objective function was underachieved with an n3 value of $682 421. In other words, money has been saved from water use. The goal could thus be reduced even further to $3 000 000 – $682 421 = $2 317 579. The value of ore has been overachieved as the p4 value is $1 412 376 000. This means that we have increased the value of the system in comparison to our goal value. It must be noted that this economic value is determined from addition across the network, so it does not correspond to a real profit. corresponding goal functions p1, p2 and p3 are negative deviations of the carbon, energy and water goal functions n4 is the positive deviation of the economic value, which will act to maximise this objective Results The LINGO output represents the result of the optimisation process. The achieved objective value is 0.1021345, which represents the sum of the deviations from the set goals. This objective value is very small, which means that the goals assigned to the carbon, energy and water objective functions are almost achieved. In other words, the system is working and the mining operation can make profits as well as achieve these targets. Data validation The outputs from LINGO were compared to a hypothetical mining operation as shown in Table 2. The estimated tonnages for each of the relevant nodes were converted to equivalent tonnages of ore, so that it can be compared directly to the outputs from LINGO. The amount of infeasibilities is zero in this case, which means that no constraints or variables are violated. LINGO has run four cycles to achieve this result, which corresponds to the four criteria in the optimisation. The total number of variables is equal to 21 and all the constraints and equations are linear in nature. The total number of constraints used in optimisation is 43. Discussion The mining, metallurgical and smelting/refining components of a mining operation were represented in a network flow diagram as shown in Figure 2. The network displays a number of different streams that produce different saleable products in accordance with the described mining operation. From the data validation, values appear to be within a reasonable range of a typical large-scale open pit operation; however, a number of shortcomings or biases are observable. The most significant of these appears to be a bias towards minimisation of cost, which may have been brought about by an insufficient weighting of the revenue function. In this respect, the object functions focused mainly on the minimisation of cost, with respect to carbon emissions and water and energy usage. The revenue generated was not given appropriate weighting in this optimisation process. This is mostly highlighted in the fact that a lot of material was sent to the TSF instead of being passed on to flow through the The values of the decision variables are also produced, which represents the amount of ore passing through each arc. All the variables and their resulting values are provided in Table 1. The value column shows the deviation from the target value in dollar terms. The p1 value of $9 192 105 means that the goal for the carbon function is overachieved and the system is unable to reach this target. The optimisation will achieve a carbon function goal of $90 000 000 + $9 192 105 = $99 192 105. The p2 and p3 deviations are zero, which means that the goals have been underachieved. The energy objective function has TABLE 1 LINGO modelling results. Variable Ore (mt) Reduced cost Variable Value ($) Reduced cost Variable Weight Reduced cost X0,1 50.0 0 P1 9 192105 0 W1 1.00 0 X0,2 10.0 0 P2 - 0 W2 1.00 0 XR,3 53.3 0 P3 - 0 W3 1.00 0 X3,4 6.5 0 P4 1 412 376 000 0 W4 1.23 0 X3,11 46.8 0 N1 - 0 X4,5 1.0 0 N2 - 0 X4,6 5.5 0 N3 682 421 0 X6,7 2.5 0 N4 1 0 X6,9 3.0 0 X7,8 2.5 0 X9,10 3.0 0 X11,12 10.0 0 X11,14 36.8 0 X12,13 10.0 0 X13,10 6.0 0 X13,15 4.0 0 X15,16 4.0 0 Mining Education Australia – Research Projects Review 2012 23 V Golding et al TABLE 2 Comparison to a hypothetical mining operation. Node Min (t) Mid (t) Max (t) Converted (t) Open pit 50 000 000 50 000 000 50 000 000 50 000 000 Underground 10 000 000 10 000 000 10 000 000 10 000 000 Concentrator 61 000 000 71 434 800 81 869 000 60 000 000 Smelter and acid plant 14 000 000 16 155 970 18 311 940 25 000 000 Refinery 12 500 000 16 520 250 20 540 500 29 000 000 Tailings storage facility 45 000 000 45 375 000 45 750 000 4 000 000 Hydrometallurgy 6 500 000 9 864 280 13 228 560 8 000 000 Cu Conc 35 000 000 35 000 000 35 000 000 33 000 000 Cu Cathode 6 000 000 8 155 973 10 311 945 29 000 000 U3O8 concentrate 2 500 000 2 500 000 2 500 000 53 800 000 Au and Ag bullion 4 000 000 5 875 000 7 750 000 45 300 000 refinery and smelter. These nodes have high values associated with carbon and energy use, but also contribute significantly to the amount of valuable material extracted. This is especially true for the production of gold and silver bullion. Conclusion and recommendations A MOO strategy was successfully implemented to describe a comprehensive mining-metallurgical system. The optimisation process focused on the minimisation of the costs associated with carbon emission and water and energy consumption, whilst also incorporating the revenue generated by the end saleable products. LINGO was found to be an extremely efficient program, capable of handling the model and the associated number of constraints. It is believed that this method of optimisation is ideal for a mining operation to model a system in terms of flow of material through relevant nodes or processes. Furthermore, the model is capable of addressing a number of objective functions, or optimising criteria, in order to realistically model the expected cash flow of the mine. 24 The LINGO model output results that are within an acceptable range of an existing mining operation (see Table 2). The objective value was found to be 0.1 dollars, which suggests that most of the targets set by the decision-maker have been achieved. In terms of the individual objective functions, the carbon goal was overachieved by $9.1 M. The energy goal was satisfied within a reasonable amount. The water goal was under achieved by $0.7 M and the economic value goal was over achieved by $1.4 B. More detailed information from a mining operation will assist with data validation. The actual application of the model will need to be tailored so as to be site-specific for the mining operation in question. Weighting can be altered, with respect to the objective function, to represent the different operating conditions of the mine in question and the goals required. A limitation of this method is that the model is affected significantly by the goals set by the decision-makers. For this reason it is important for these goals to be set with a high level of confidence by a competent person(s). REFERENCES Chanda, E K, 2004. Network linear programming optimisation of an integrated mining and metallurgical complex, in Orebody Modelling and Strategic Mine Planning, pp 13-19 (The Australasian Institute of Mining and Metallurgy: Melbourne). Jones, D and Tamiz, M, 2010. Practical Goal Programming, International Series in Operations 11, Research and Management Science 141, DOI 10.1007/978-1-4419-5771-9_2, Springer Science Business Media, LLC. Kornbluth, J S H and Steuer, E, 1981. Multi objective linear fractional programming, Management Science, 27:1024-1039. Rafiee, V and Asghari, O, 2008. A heuristic traditional MIP solving approach for long term production scheduling in open pit mine, Journal of Applied Sciences, 8(24):4512-4522. Rangaiah, G P, 2008. Process optimisation, multi objective optimisation, techniques and application in chemical engineering, Advances in Process Systems, 1:1-12. Tamiz, M, Jones, D and Romero, C, 1998. Goal programming for decision making: An overview of the current state-of-the-art, European Journal of Operational Research, 111(3):569-581. Topal, E and Ramazan, S, 2012. Strategic mine planning model using network flow model and real case application, International Journal of Mining, Reclamation and Environment, 26(1):29-37. Mining Education Australia – Research Projects Review 2012 Evaluating the Performance and Productivity of Continuous Surface Miners in Iron Ore at Christmas Creek Mine B Goves1, M S Kizil2 and W Seib3 ABSTRACT Continuous surface mining involves milling, crushing and loading iron ore in one systematic process. Whilst the method is relatively new in the iron ore industry, it is also beneficial in selective mining of orebodies with challenging geologies. This paper details an evaluation of the performance and productivities of continuous surface miners in iron ore mining at Christmas Creek Mine in Western Australia. The study involved ten months of data collection and analysis to successfully complete the project. In total, 31 surface mining machines were examined involving five different machine models. The five machine models investigated were the Wirtgen SM2200, Wirtgen SM2500, Wirtgen SM4200, Vermeer T1255 and Vermeer T1655. The data is comprised of two different types of analysis concerning operational and instantaneous production reports. The operational production data objective was based on five key criteria including productivity, performance, influence of material type, availability, utilisation and overall unit cost. The objective of the instantaneous data was to correlate instantaneous productivities with material type. This resulted in the construction of plots using mapping software packages. Introduction Christmas Creek Iron Ore Mine is located approximately 110 km north east of Newman in the Pilbara region, Western Australia. The deposit is hosted within the Nammuldi Member at the base of the Marra Mamba Iron Formation (Clout and Rowley, 2009). Selective mining is used to extract the iron ore at Christmas Creek, utilising the continuous surface miners that are innovative to the region and the iron ore industry. With advancements in technology, continuous surface miners offer an alternative to traditional orebody extraction techniques. The load and haul process is the method primarily used in open cut mining applications. Whether surface miners are used in conjunction with load and haul practices, or as a completely different mining method, the potential advantages may directly affect the economic viability of a deposit. Selective mining poses significant difficulties concerning the profitability of a mine, especially, in the current market where large-scale deposits with small stripping ratios are preferred. With the utilisation of surface mining methods, thin seam mining of iron ore bands, not only increases the economic viability of a project, but assists with mine planning, drill and blast design, transportation, processing and flexibility. Surface miners allow high quality products to be excavated with economic efficiency even from technically challenging mineral deposits (Wirtgen GmbH, 2010). Additionally the method can dramatically improve run-of-mine (ROM) quality, reduce dilution and increase mining reserve. The material is excavated and crushed in one operation before it is loaded onto trucks, saving further handling, transportation and process costs (Williams, Mendelawitz and Castle, 2007). With this in mind, the current surface mining equipment at Christmas Creek mine is not being actively tracked with regard to production rates, locations and the type of geological material. An analysis and evaluation will allow a baseline of performance for any future improvement projects, and ensure effective tracking of the surface miner productivities. The study was limited to comparing five machine types at Christmas Creek Iron Ore Mine. The Wirtgen surface miners evaluated included the SM2200, SM2500 and SM4200. The Vermeer surface mining machines evaluated included the T1255 and T1655. Mining Method and Equipment The surface mining method at Christmas Creek involves continuous surface miners cutting the material to ground rather than utilise the direct loading conveyor system implemented at Cloudbreak mine. Whilst trials have been conducted on the conveyor system, as well as a harvesting method, this paper is only concerned with the conventional method. 1. The University of Queensland, St Lucia Qld 4072. Email: [email protected] 2. MAusIMM, Associate Professor, School of Mechanical and Mining Engineering, The University of Queensland, St Lucia Qld 4072. Email: [email protected] 3. FAusIMM(CP), Adjunct Professor, School of Mechanical and Mining Engineering, The University of Queensland, St Lucia Qld 4072. Email: [email protected] Mining Education Australia – Research Projects Review 2012 25 B Goves, M S Kizil and W Seib Fig 1 - Cross-section of mining process (Walker, 2011). The operation is separated into an overburden crew and surface mining crew. The two crews are unlikely to work in the same pit at any given time due to increased traffic and safety issues. The top of surface miner (TOSM) is the key factor in the mining profile. Material above this line is drilled and blasted before overburden removal takes place through hydraulic excavators and haul trucks. Figure 1 illustrates the mining process and cross-section profile. The orebody is located below the TOSM and is milled in strips using the surface miners. Hump cuts are initiated to provide a smooth flat running surface for the surface miners to run on. The surface miners mill the ore whilst separating any waste layers. The material is mined in situ alleviating the need for drill and blast. Once the stripping process has been completed, the broken material is then ‘walked’ by geologists to confirm and mark up blocks of ore and waste. The designs are uploaded on to dozers’ computer to ‘push up’ the material into suitable piles for loading out. Front-end loaders are used to load haul trucks, which then dump at a designated ROM pad or waste dump. In order to effectively utilise the surface miners, a minimum of three strips are required in a rotational system. This includes a milling strip, a strip being loaded out and a strip that has finished being loaded and is cleared. Wirtgen and Vermeer surface miners were analysed throughout the project. Both manufacturers utilise a mechanical cutting drum to fragment the rock. The key difference between the two is the opposite cutting directions. The Wirtgen machines cut upwards compared with the Vermeers, which cut top-down. A Wirtgen SM2500 machine is illustrated in Figure 2 equipped with a conveyor system. Depending on the machine configuration, surface miners can cut from 2.20 to 4.57 m wide and at a depth of 0.10 to 0.71 m. The differences in size, weight, horsepower and cutting dimension deliver advantages and disadvantages for each machine model. Methodology and Data Analysis Operational production data The operational production data involved a large amount of data collection and analysis. The data was collected for 10 months on a daily basis from July 2011 to April 2012. It was recorded and captured by the surface miner operators, before compilation into Microsoft Excel spreadsheets. Each individual spreadsheet included: •• day shift and night shift for a particular day •• every surface mining machine that was operated during that day 26 Fig 2 - Wirtgen SM2500 at Christmas Creek mine (Truong and Nguyen, 2012). •• SMU hours, working hours, standby time, operating delays, weather delays, unscheduled maintenance and scheduled maintenance •• the amount of picks, tool holders and fuel used •• the tonnages milled •• the area of mining (pit/strip/flitch) •• the ground conditions that were encountered and any additional comments. This information was obtained via email and then compiled into monthly summary spreadsheets before analysis could continue. Due to the large amount of data, a significant period of time was spent assembling the information into readable and comprehensive data sets. It is estimated that approximately 600 to 800 spreadsheets were used in the analysis. Instantaneous Production Data The instantaneous production data was collected for a number of different shifts for the Wirtgen surface miners. The instantaneous productivity of a surface miner is defined as the production rate at a certain point in time. This rate also details the immediate GPS position of the machine along with an estimate of the type of material the machine is milling. The data was recorded and captured through access to the Trimble GPS on-board the surface miners. On average three to four points are recorded every two seconds when operating. The production reports contain information relating to: •• easting, northing, Z RL (x, y, z) position of the surface miner •• GPS points for both ends of the cutting drum •• GPS tolerances •• the speed the machine is operating at (km/h) •• the design and location the machine is operating in (pit/ strip/flitch) •• which gear the machine is operating in and the direction of travel. The instantaneous productivity is dependent on several key constraints and variables. The width and depth of the cut must be fixed for a given strip. The density must also be fixed to accurately calculate tonnage per hour. The key variable is the speed the machine is milling at in a given strip. The instantaneous productivity is calculated by: P = S × CW × CD × ρ (1) Mining Education Australia – Research Projects Review 2012 evaluatIng the perforMance and productIvIty of contInuouS Surface MInerS In Iron ore where: P is the instantaneous productivity S is the speed (metres/hour) CW is the cutting width (metres) CD is the cutting depth (metres) ρ is the density (kg/m3) reSultS operational production results Productivity of surface mining fleet Calculating the productivity of the surface miners was one of the key pieces of analysis throughout the duration of the project. Figure 3 portrays a comparison of the five machine models and their average production rates. The targeted production rates also highlight whether a machine made target or not. was lower than expected, there was an upward trend and improvement from an initial reading of 444 t/h before finishing with 528 t/h. The Vermeer T1655 machine produced results on par with its target of 1700 t/h, with an average of 1699 t/h. The machine met its target in the months of September, October and November 2011, before a downward trend developed at the beginning of 2012. The highest monthly average was 2137 t/h in September 2011 and the lowest being 635 t/h in March 2012. The unusual result in March dropped the project average significantly, triggering an investigation into the machine’s availability, utilisation, operational delays and maintenance. When comparing all of the machines, the SM2500s produced the most impressive results in terms of improvement, where a sharp increase was observed to almost triple the production rate by project end. The SM4200s achieved the highest production rates above target and consistently achieved positive results. The SM2200 machines operated at approximately a quarter the productivity of a SM4200. Finally, the T1255s consistently struggled to meet the targeted rates and operated at approximately a quarter the production rate of a T1655. In combination with the productivity, the performance of all the surface mining machines was calculated. Figure 4 details the performance comparison. The SM2500 and SM4200 produced positive performances of approximately 13 per cent and 27 per cent respectively. The SM2200 and T1255 machines produced results below the expected level at -8 per cent and -17 per cent respectively. The Vermeer T1655 produced a result that was on par with its target. All of the machine models except for the T1655 observed an upward trend in performance from the beginning of the study to the end. The improvement trend was due to a combination of factors concerning: • Fig 3 - Productivity comparison of surface mining fleet. The Wirtgen SM2200 machines averaged a production rate of 579 t/h, which was slightly lower than the targeted 630 t/h. Over the five-month period from July to November 2011, the machines met target only twice and almost reached target in October. A trend analysis, revealed a gradual improvement in productivity. A recording of 386 t/h in July before finishing with three months averaging over 600 t/h highlighted the improvement. It was concluded that had there been a larger period of data collected, there may have been more consistent results observed. Due to the decommissioning of the SM2200s in December, data would need to have been sourced earlier than April 2011. • • • a better understanding of the limitations and capabilities of the machines improvement of selected preconditioning of the orebody enhanced knowledge and understanding of geological model development of operator skills and efficiency. Influence of material type This analysis detailed the average production rates when a machine was operating in a certain material type or rock hardness. As expected, there was a correlation between material type and productivity, as highlighted by the three linear trend lines in Figure 5. Based on the average production rate of 579 t/h, it is estimated that the SM2200 machines were operating on average in medium material. The SM2500s were The Wirtgen SM4200 machines produced the highest productivities of all the machines analysed. The average production rate was 2161 t/h, which was significantly above the targeted value of 1700 t/h. The machines were implemented in December after the decommissioning of the SM2200s and studied through to April. A sharp improvement was observed from averaging 1138 t/h in December to finishing with an average of 2198 t/h in April. The highest monthly average was 2451 t/h in January. The Vermeer T1255 machines recorded the lowest productivity rates and struggled to meet targets. An average of 458 t/h was recorded with a target rate of 550 t/h. The machines were analysed for a period of six months meeting target only in February 2012. Although the productivity MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Fig 4 - Performance comparison of surface mining fleet. 27 B goveS, M S kIzIl and w SeIB Fig 5 - Comparison of material type productivity for Wirtgen surface miners. The total rostered time for each machine model was broken down into a comparison of working, standby, operational delay, unscheduled maintenance and scheduled maintenance. This comparison is shown in Figure 6 where T1255 machines achieved the highest amount of working time. Considering they had the worst performance at -16.7 per cent, this suggests that the target rates were too high at 550 t/h. The SM2200 machines had the lowest working time at 24 per cent, resulting in large periods of operational delay and unscheduled maintenance at 36 per cent and 27 per cent respectively. This excessive amount of downtime was the cause for not meeting targets and its -8.1 per cent performance. The most productive machine the SM4200, had a working time of 37 per cent. By reducing its excessive operational delay a further increase in performance would prevail. Overall, across all 31 surface miners, the working time was 34 per cent, standby eight per cent, operational delay 18 per cent, unscheduled maintenance 21 per cent and scheduled maintenance 18 per cent. estimated to have milled medium to medium hard material. The SM4200 machines operated in medium soft material, based on an average production rate of 2161 t/h. The smaller SM2200 and SM2500 machines almost doubled their production rates when moving from hard to soft material. This was displayed in Figure 5 by the 343 t/h and 543 t/h in hard material compared with 622 t/h and 1043 t/h respectively in soft material. Based on the result of 2021 t/h in hard material for the SM4200s, a production rate of approximately 4000 t/h is expected in soft material. This correlation between material type and productivity, suggests the smaller Wirtgen machines are more sensitive to fluctuations in rock hardness when looking at production rates. The bigger SM4200 machines are less susceptible to variation in geology, suggesting a more robust machine configuration. The result suggests the SM4200 machines can obtain much higher and consistent production rates when milling different types of rock without having an adverse or detrimental effect on the productivity. Depending on the type of material the SM4200 can operate between two to four times greater than the SM2500. The SM2200 operates at approximately two-thirds the production of a SM2500. Further analysis of the results exposed unusual values in medium hard material for SM4200 and in soft material for SM2200 machines. The values do not follow the linear relationship and are significantly lower than the linear trend line. If the correlation were correctly modelled a value of around 2040 t/h would be expected in medium hard material for the SM4200. A value of above 800 t/h would be expected for the SM2200 in soft material. The differences in the linear trend are attributed to the smaller periods of data analysed. The SM4200 and SM2200 were both analysed for five months due to decommissioning and arrival of new machines. The SM2500 machines had comprehensive data from the full 10 month study and exhibit an almost perfect linear line. Human errors would also be evident due to the manual process involved in recording and capturing the data. This estimation process provides an idea of the material type operated in, as majority of the iron ore specified by the geological model, is in the medium to hard range. Availability and utilisation The availability is defined as the percentage of time a machine is mechanically available for work. The utilisation represented the percentage of scheduled machine hours that were productive. Finally the effective utilisation is the percentage working time per total calendar time. The effective utilisation is also described as the product of availability and utilisation. 28 Fig 6 - Comparison of rostered time for surface mining fleet. When comparing all the machine models on a month-bymonth basis, a number of important trends were observed, as illustrated in Figure 7. One key area to improving productivity is by limiting the amount of unscheduled maintenance. From July to April a distinct reduction from 40 per cent down to 17 per cent in unscheduled maintenance was achieved. By more than halving this downtime, the machines saw a rise in productivity and performance. The months of November through to February experienced increased operational delay. In particular, the month of January with almost 50 per cent of total rostered time attributed to operational delay. This was due to the summer months where increased cyclone activity and dust storms were prevalent. Scheduled maintenance was fairly constant at 17 per cent and represented proactive measures taken on the machines. This resulted in limiting the unscheduled maintenance. The average availability, utilisation and effective utilisation were calculated with a number of key observations found. Table 1 shows that the T1255 had the highest utilisation and effective utilisation, however produced the lowest results in terms of performance, further suggesting that the target rate of 550 t/h was potentially too high. A cause for concern is the underutilisation of the SM4200. Whilst the machine had the highest availability at 76 per cent, it was only utilised 49 per cent of the time. This resulted in 37 per cent effective utilisation, which requires improvement considering the SM4200 machines achieved the highest production rates. Overall, when considering all machine types in the study, MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 evaluatIng the perforMance and productIvIty of contInuouS Surface MInerS In Iron ore Fig 7 - Comparison of rostered time on a monthly basis for surface mining fleet from July 2011 to Apr 2012. TABLE 1 Average availability, utilisation and effective utilisation for the surface mining fleet. Machine Availability (%) Utilisation (%) Effective utilisation (%) SM2200 64 45 24 SM2500 55 59 32 SM4200 76 49 37 T1255 68 64 43 T1655 66 48 32 Average 61 59 34 Fig 8 - Monthly trends in availability, utilisation and effective utilisation for surface mining fleet. a cost per tonne basis, the results indicate that the two largest machines are preferred over the smaller machine models. The overall unit cost was also calculated regarding material type. This provided a cost of operating in different strengths of rock hardness for the Wirtgen surface miners. In Figure 10, different line types correspond to unit cost and production rates for the same machines. The increase in rock hardness produces an increase in overall unit cost. When perfectly modelled, a linear relationship should be observed as per the red trend line for the SM2500 machines. the availability was 61 per cent, utilisation 59 per cent and effective utilisation 34 per cent. The availability, utilisation and effective utilisation were then examined on a month-by-month basis for all the machines. A number of key trends were found as illustrated by the coloured trend lines in Figure 8. The availability experienced a positive trend from just above 40 per cent to in excess of 60 per cent by the end of the study. The utilisation on the other hand, experienced a downward trend. Utilisation of 70 per cent was observed at the beginning of the study before falling to around 50 per cent by the project end. Finally the effective utilisation was fairly constant and was maintained at around 35 per cent for the ten month study. The analysis of overall unit cost is aimed at quantifying which models are preferred on a cost per tonne basis. The study required the total operating and maintenance (O&M) cost to be calculated in order to determine the cost per milling hour. The cost was inclusive of overheads, supervision, depreciation, interest, fitter labour, components, oil and grease, tool holders, fuel and the operator’s salary. However, it was not inclusive of picks. The most expensive piece of machinery based on total O&M cost was the SM4200 surface. The least expensive machine was the T1255 surface. Using the average production rates and the total O&M cost, the overall unit cost was determined for each machine model. This is shown in Figure 9. The study found that the T1255 machine was the most expensive, whilst the largest machines the T1655 and the SM4200 were cheaper by 63 per cent. Figure 9 details the most expensive to least expensive machines on the x-axis. The analysis suggests a relationship between production rate and cost per tonne. An increase in production rate results in a decrease in overall unit cost. On MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Fig 9 - Comparison of overall unit cost for surface mining fleet. When milling in hard material the SM4200 machines were over three times cheaper than the SM2500 and 4.5 times cheaper than the SM2200. In soft material, the SM2200 was 1.5 times more expensive than the SM2500 machine. The SM4200 was a further three times cheaper in soft material compared with the SM2200 machines. It is conclusive that the SM4200 machines were significantly cheaper to operate than the other two Wirtgen machines in a wide range of materials. The costing analysis suggests that when milling below the targeted rates, the cost per tonne will be higher. This is associated with operating in harder rock strengths, which leads to increased wear and tear on the machine. If the cost of picks were included in this study, it is predicted that the overall unit cost would see a further rise. Instantaneous production results Instantaneous parameters The instantaneous productivity calculation involved using a number of parameters. The analysis uses the forward speed of the machine as the key variable in combination with the 29 B goveS, M S kIzIl and w SeIB TABLE 3 Instantaneous operating ranges for SM2200. Material Speed range (m/min) Very hard 0 - 1.7 Hard 1.7 - 5 Medium 5 - 6.7 Soft 6.7 - 10 TABLE 4 Instantaneous operating ranges for SM2500. Fig 10 - Comparison of material type overall unit cost for surface mining fleet. cutting dimensions and density. The density used was 2.8 t/m3 based on the rock properties at the Fortescue deposits (Hutchins, 2012). For the calculation to work correctly, the cutting width and depth of a machine is fixed for a given strip. Table 2 details the cutting dimensions used for each of the Wirtgen surface miners in the calculation. As the forward speed of the machine is the principal TABLE 2 Surface miner cutting dimensions for instantaneous productivity calculations. Machine Cutting width (mm) Cutting depth (mm) Wirtgen SM2200 2200 300 Wirtgen SM2500 2500 500 Wirtgen SM4200 4200 650 determining factor, it was necessary to establish the penetration rate or the amount of metres per minute a machine could operate at. Using industry research and trials conducted at the Fortescue deposits (Hutchins, 2012), operating ranges were calculated for this particular shift. Tables 3 to 5 show the operating ranges for the SM2200, SM2500 and SM4200 surface miners. concluSIonS The paper investigated two types of data with regard to several different surface mining machines. The SM4200 and T1255 were the largest machines and produced the highest production rates. The best performance was recorded by the SM4200 and the SM2500. However, the worst performed machines were the T1255 and SM2200 over the project duration. There was a distinct linear correlation between material type and productivity. The SM2200 and SM2500 followed the linear trend, whilst the larger SM4200 machine experienced smaller fluctuations in productivity as the material type varied. In terms of efficiency, a decrease in utilisation was an alarming trend, as the availability steadily increased over project duration. An assessment of the total rostered time exposed a target rate too high for the T1255 machines. However, an improvement in unscheduled maintenance downtime by more than half was promising. Regarding the overall unit cost, the larger T1655 and SM420 were the cheapest machine to operate, whilst the smaller T1255 was the most expensive on a dollar per tonne basis. The instantaneous productivity calculations resulted in the construction of plots for three shifts. Areas of high and low productivity were identified with corresponding zones 30 Material Speed range (m/min) Very hard 0 - 1.7 Hard 1.7 - 5 Medium 5 - 8.3 Soft 8.3 - 11.7 TABLE 5 Instantaneous operating ranges for SM4200. Material Speed range (m/min) Very hard 0 - 3.3 Hard 3.3 - 6.7 Medium 6.7 - 15 Soft 15 - 25 of material hardness. Sections of tramming and erroneous points were observed and were excluded from the analysis. An important correlation was found between instantaneous productivity, speed of the surface miner and material type. acknowledgeMentS The authors would like to thank the continued support of Downer EDi Mining’s technical services department in Brisbane and at the Christmas Creek Mine site. Additionally, the author would like to thank Fortescue Metals Group, Western Australian Surface Mining and UEA for providing the data. The author gratefully acknowledges the advice and information provided by Jim Hutchins. referenceS Clout, J M F and Rowley, W G, 2009. The Fortescue Metals Group story – From exploration to the third largest iron ore producer in Australia, in Proceedings Iron Ore 2009, pp 63-71 (The Australasian Institute of Mining and Metallurgy: Melbourne). Hutchins, J, 2012. Personal communication, 9 August. Truong, B and Nguyen, A, 2012. Xmas Creek Operation, workshop delivered at the 2012 Engineers and Surveyors Conference, Mackay, March 2012. Walker, B, 2011. Surface mining – An overview, presentation, CC Operations. Williams, J, Mendelawitz, D and Castle, M, 2007. Applicability of using Wirtgen surface miners as a mining tool for iron ore, in Proceedings Iron Ore 2007, pp 427-432 (The Australasian Institute of Mining and Metallurgy: Melbourne). Wirtgen GmbH, 2010. Job report surface mining: iron ore mining in Australia: surface miners are main mining equipment [online], Germany, Wirtgen Group GmbH. Available from:<http://www. wirtgengroup.de/en/aktuelles_und_presse/presseberichte/ uebersicht/Presseartikel_detail_4703.html> [Accessed: 10th May 2012]. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Strategic Project Risk Management for an Emerging Miner T K C Graham1 and S Saydam2 Abstract Development of a new minerals project represents a time of high strategic risk for a mining company, particularly in the case of an emerging mining company; one in the process of making the transition from predominantly an exploration organisation to a full-scale mining organisation. A number of methods have been identified as having the potential to assist these companies in understanding the risk of their projects. A case study based the Cockatoo Coal’s Baralaba Coal Expansion Project (BCEP), was used as the basis to evaluate three quantitative scenario analysis methods; these being: sensitivity analysis, program evaluation and reviewing technique (PERT) and Monte Carlo simulation. The study found that the Monte Carlo simulation provided the most comprehensive results combined with the flexibility to be selective in terms of output data. By contrast, the PERT approach appeared to be a simplified modelling method with potentially unreliable and biased results; whereas the sensitivity analysis provided limited results for the purpose of analysing the prospects of and risks in the project. Furthermore, a model for risk management in a minerals development project has been suggested. This highlights the unique stages through the life of the project, their respective milestones and the deliverables required. The Monte Carlo simulation has subsequently been applied to this model and recommendations provided for which deliverables should be used for risk analysis for the project. Introduction Emerging Miners are mining companies that are at the development stage; these companies have proved up mineral reserves and are undergoing the process of advancing into production. This transformation from explorer to producer is characteristically a high risk phase of a company’s life cycle. There is generally quite a high level of investor, management and corporate pressure to succeed in bringing the projects from scoping to construction and operations; as well as a distinct lack of operating experience hampering the ability to achieve these goals. Additionally, it is quite common for these companies to run, what could be classified as, medium to high levels of risk. Such companies are not faced with the capital allocation issues of the larger, more established companies; and having a mine may be more desirable for shareholders than having no mine at all (Stirzaker, 1997). Emerging miner – Cockatoo Coal Limited In 2005 Cockatoo Coal Ltd, an Australian coal exploration company, was formed and listed on the Australian Stock Exchange. The company began with only one exploration lease, however it has since grown a substantial portfolio containing five Coal exploration leases throughout Queensland and New South Wales as well as one small-scale operating mine (Cockatoo, 2012). Cockatoo is a prime example of the modern ‘Emerging Miner’, with multiple projects undergoing development and a goal to grow these into operational mines within the next five years. It is a company with big aspirations, and equally large risks. In early 2012, Cockatoo identified the unique nature of its position as an emerging company. The company also realised that if it was to succeed in making an effective transformation to full-scale mining operations then it would need to have a thorough understanding about how to manage its projects successfully, and how to understand and manage its various risks in the process. These business risks became more apparent in 2012 when the Australian mining industry appeared to contract slightly leading to a slump in share prices across the market and as a consequence a 70 per cent reduction in the Cockatoo share price between February 2012 and September 2012. As such Cockatoo recognised that in order to ensure the success of its projects, and with them, the company, it must implement new initiatives to identify where risks lie and how best to manage and mitigate them. The topic of this paper was thus inspired by this company with the intention to assist with their objectives. Research objectives With reference to the issue of risk in an emerging miner’s project development system, the objective of this study was to analyse potential techniques for incorporating risk effectively 1. GAusIMM, The University of New South Wales, Sydney NSW. Email: [email protected] 2. MAusIMM, The University of New South Wales, Sydney NSW. Email: [email protected] Mining Education Australia – Research Projects Review 2012 31 t k c grahaM and S SaydaM into a project, and to provide recommendations on how they could be integrated into a project management structure. project rISk ManageMent theory As defined by Dobie (2007), the general concept of project management is the disciplined process in the planning, organising, monitoring and controlling of all aspects of a project. This concept can be expanded upon further by using nine functional elements to serve as the basis of the project management framework. These elements and their relationships are shown in Figure 1. Importantly, in Figure 1, it must be noted that the relationship of most relevance to this study is that between Risk and the four key elements, and how a variation in one element can alter the inherent risk in the project. Fig 2 - Mine project stages (Chinbat, 2011). Fig 1 - Elements in project management (Dobie, 2007). the mine development project For the purpose of this study, a mine development project has been defined as a number of stages separated by milestones. This concept is illustrated by Chinbat (2011) in Figure 2 and emphasises that, before each stage is passed, a review must be undertaken to conclude whether the stage met its objectives, if there is adequate prospects to continue to the next stage, and what the best approach to the subsequent stage is. Scenario analysis of mine project options It is at the stage gates concluding the stages shown in Figure 2 that a decision has to be made about what the future direction for the project will be. Importantly, there are a number of options available to management for the progress of the project at these points. These range from delay, terminate and abandon through to expand, shrink, revise and proceed (Eschenbach et al, 2007). In order to summarise how these options fit into the mine project structure, a decision tree has been constructed as shown in Figure 3. This tree runs through the feasibility and prefeasibility stages of a mine development project and demonstrates what options are available to management at the stage gate completing each stage. It is at these decision points that a new problem is raised, ‘Which of the above options is the best for our project at this stage?’ In order to assist with this problem, this paper proposes that scenario analysis could be employed. The intention of scenario analysis is to thoroughly analyse each of the options at the stage gate (from a time, cost, quality and scope perspective) so that they can be accurately understood and compared objectively, so that the ‘best’ option can be selected. The purpose of this process is to introduce reliability into 32 Fig 3 - Options in a mine development project. the information used by decision makers. By having more reliable and accurate data, it is proposed that there will be a reduced chance of uncertainty and thus risk of the project going ‘wrong’ in the future. It is at this point that attention now turns to how this scenario analysis should be undertaken. In order to answer this, five scenario analysis methods have been identified for the potential to assist in this application. These are listed in Table 1 and all take a quantitative approach to model the outcome, and additionally, uncertainty in the outcome of a project. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 StrategIc project rISk ManageMent for an eMergIng MIner TABLE 1 Scenario analysis methods. Event tree analysis (probabilities) Sensitivity analysis Monte Carlo simulation Risk adjusted discount rate Program evaluation and reviewing technique (PERT) coMparISon of ScenarIo analySIS MethodS In order to compare the methods found in Table 1 they have each been applied to the Cockatoo Coal Baralaba Coal Expansion Project. This is a project currently at the completion of the Bankable Feasibility Study (BFS) with the intention to expand the small Baralaba Coal Mine from 750 kt/yr to 3.5 Mt/yr through the addition of two new pits and an associated wash plant. As such, to evaluate and compare the above methods, an analysis has been undertaken at the BFS stage gate analysing the option to proceed through construction stage. This has been undertaken on both the schedule (time) and budget (cost) for this option. Furthermore, of the above five methods, three have been selected for the comparison; sensitivity analysis, PERT and Monte Carlo simulation. The other two have been deemed either too limited in their scope to be applicable across all stages of a minerals project (risk adjusted discount rate) or un-achievable with the limited level of information available (event tree analysis). Sensitivity analysis The sensitivity analysis has been used to model a variation in each of the input variables in both the schedule (duration) and the budget to see what kind of a variation will be produced in the overall outcome (total). This analysis produced some results of minor relevance to the research’s objectives. It has identified which jobs in the schedule the overall duration is most sensitive to (shown through Figure 4). However, these results are limited as they have not given any indication about what the actual expected outcome is likely to be. This is further exaggerated in the case of the budget analysis (not shown) which only indicated the obvious conclusion that the total budget sum is most sensitive to a variation in the larger budget items – an outcome which provides little assistance to managers attempting to make a decision about which option to pursue. program evaluation and reviewing technique The program evaluation and reviewing technique (PERT) simulation makes use of three input values for each of the variables – a best case, worst case and most likely outcome. These are then substituted into the PERT formula to produce a weighted average estimate of the outcome. Subsequently this method has been applied to both the Baralaba Coal Expansion Project construction schedule and budget with a weighted average outcome being produced for each of these factors, and the overall results (showing the variation) found in Table 2. TABLE 2 Results of program evaluation and review technique analysis. Schedule duration Budget total Original PERT 1491 days 1515 days (101.6%) 100% 100.2% This method has, in some regard showed an improvement over the sensitivity analysis; this is as an actual estimate for the outcome that has been produced, a result which could be used by management in better understanding the project and applying appropriate contingency in planning. However, these results produced do have some limitations. Firstly, they do not have a high statistical reliability – only one iteration has been undertaken and as such they are reliant wholly on the assumptions made through the PERT formula to produce a result. Secondly, the PERT method has been shown to be susceptible to bias through the use of subjective estimates for the three input values; this means that it is possible for management to be grossly misled by the outputs of this technique (Klingel, 1966). Monte carlo simulation Monte Carlo simulation uses a process whereby each of the independent variables (job durations in the schedule/items in the budget) is allocated a distribution function. A simulation is then undertaken with the input values generated randomly from within these probability distributions for each of the inputs. This method has subsequently been applied (through the use of Palisade’s @Risk MC Simulation Program) to both the construction schedule and budget and a 1000 to 5000 iteration simulation been carried out. Resulting from this technique, a density histogram has been produced. Shown in Figure 5 is the density histogram for the construction budget. It can be seen this has effectively formed a normal distribution of results which can resultantly be used to produce a mean, medium and mode result; as well as various other distribution data such as confidence intervals and percentiles – all of which can be applied in the calculation of project contingency and budgeting. Similar results were produced for the construction schedule (not shown). Furthermore, as shown in Figure 6, this technique (through the software package used) has the ability to produce its own sensitivity analysis – in this case for the sensitivities in the construction schedule – effectively making the previous stand-alone sensitivity analysis void. Fig 4 - Major sensitivities in Baralaba Coal Expansion Project schedule. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 As such the Monte Carlo simulation has shown a greater flexibility and reliability in the results produced, which could be used by management in the decision making process, than both the Sensitivity Analysis and PERT simulation. 33 t k c grahaM and S SaydaM of ‘deliverables’ (Project Development Institute, 2012). A deliverable for the purpose of this model comprises of all the elements/factors produced in each of the respective stages and are used as inputs to the stage gate review. For the purpose of a mine development project, it is additionally proposed that these deliverables both address the outcomes of the previous stage, as well as an analysis of the options for future progression into further stages. Fig 5 - Monte Carlo simulation – budget results. The emphasis of the guide is to roughly highlight the areas of uncertainty at each stage gate in a minerals development project and to provide guidance on the most appropriate method to be undertaken in order to understand the impact these uncertainties could have on the outcome. As such, a summary of the mine project process and associated deliverables is found in Figure 7. Fig 6 - Monte Carlo sensitivities – schedule results. risk scenario analysis methods – discussion of results and findings Resulting from this comparison, the Monte Carlo simulation method has been deemed the most valuable method for use in evaluating the options in a mine development project. It provides a greater level of detail, accuracy and reliability in the information in its results, which could thus be applied by management in making the correct or most appropriate decisions. However it must be noted, that there is an inherent risk associated with its use. If the analysis is not carried out correctly and improper distributions are used, it is highly likely that inaccurate results are produced. This can lead to misleading information or poor decisions being made. As Mooney (1997) notes, one of the key problems in conducting Monte Carlo Simulation is determining which distribution function a random variable in the population should assume. Additionally as McCabe (2003) explains, experts are very comfortable estimating the most likely values of activity duration, but are not as experienced at estimating lower and upper limits. This inaccuracy could lead to errors in the results and thus the integrity of the simulation. None the less, this method, utilised properly, has the ability to best assist managers (compared to the other methods) in making decisions regarding the options in a project. project rISk ManageMent guIde The final outcome of this study was to construct a system to provide guidance for Emerging Miners on how Monte Carlo simulation could be employed to decrease risk in the decision making process of a mine development project. In order to do this, the stage gate model discussed previously has been expanded to include the third category 34 Fig 7 - Proposed flow diagram summarising stages and deliverables for analysis in a mine project. For example, using Figure 7 at the prefeasibility stage gate it is recommended through this system that the resource should be mostly defined and the mineable properties understood. Therefore the major variables which are a cause for uncertainty are based primarily on financial, infrastructure, logistical and approval variables. Risk modelling using Monte Carlo simulation can now be undertaken at this stage looking at these variables in order to make the best recommendations on the financial and business position associated with the project. Further risks should also be identified regarding the schedule of on-going activities. This should include an analysis of identified hazards which could impact on the progression of the project as well as a scenario analysis of the Feasibility study schedule and budget, thus assisting in the accurate planning and allocation of resources for on-going studies. concluSIonS The presence of uncertainty has been recognised as a primary factor leading to the existence of risk in an Emerging Miner’s MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Strategic Project Risk Management for an Emerging Miner minerals development project. This uncertainty results from a number of limiting characteristics unique to this class of miner, and can potentially have a detrimental impact on the ability of the company to identify and pursue the best options in a project. Subsequently this paper has been focused on assisting emerging miners (characterised through the Cockatoo Coal case study) understand this risk, reduce uncertainty within a project and consequently lead to better decisions being made. The following conclusions have been produced and, furthermore, are recommended for implementation into the project management system for both Cockatoo Coal and other existing or future emerging mining companies: •• the use of the mine project stage and stage-gate theory is recommended as a tool for managing the general project progress •• a real options modelling system is recommended for use in identifying all the options available to decision makers at the corresponding stage-gates •• the Monte Carlo simulation method has been identified as the top technique for analysing these options in order to make a decision on the best path to pursue •• finally, a guide has been produced to provide recommendations on which deliverables at each of the stage gates should be analysed through the use of Monte Carlo simulation. It is hoped that by applying this system, companies in this phase of their life will be able to better appreciate the risk in their projects and thus account for and mange it better in their business decisions. Acknowledgements The authors wish to acknowledge the assistance provided by the Cockatoo Coal senior management team for their support and provision of resources and data vital for this study; Andrew Lawson, Luca Rocchi, Steve Jukes, and particularly Dane Stewart for his assistance and supervision in this process. References Chinbat, U, 2011. Risk analysis in the mining industry, in Risk Management in Environment, Production and Economy (ed: M M Savino), pp 103-122 (InTech: Croatia). Cockatoo, 2012. Cockatoo Coal Limited [online], Available from: <http://www.cockatoocoal.com.au> [Accessed: 12 March 2012]. Dobie, C, 2007. A Handbook of Project Management – A Complete Guide for Beginners to Professionals (Allen and Unwin: NSW). Eschenbach, T, Lewis, N, Henrie, M, Baker, I V E and Hariman, J C, 2007. Real options and real engineering projects, Engineering Management Journal, 19(4):11-19. Klingel, A R, 1966. Bias in PERT project completion time calculation for a real network, Management Science, 13(4):194-201. McCabe, B, 2003. Monte Carlo simulation for schedule risks, in Proceedings 2003 Simulation Conference: Driving Innovation, Winter Simulation Conference, New Orleans. Mooney, C Z, 1997. Monte Carlo Simulation (Sage Publications: USA). Project Management Institute, 2012. Stage-Gate – your roadmap for new product development [online], Available from: <http:// www.prod-dev.com> [Accessed: 15 October 2012]. Stirzaker, M, 1997. Feasibility studies - Risks and sensitivities, in Proceedings MINDEV 97 – The Third International Conference of Mine Development (The Australasian Institute of Mining and Metallurgy : Melbourne). Mining Education Australia – Research Projects Review 2012 35 36 Mining Education Australia – Research Projects Review 2012 Metaheuristic Analysis of Current Open Cut Blasting Fragmentation Prediction Models A Holland1, M Iles2 and M Karakus3 ABSTRACT Open cut mining operations require blasting optimisation through appropriate rock fragmentation to minimise cost and maximise efficiency of the entire mining process. A good blast fragmentation prediction model is critical for blasting optimisation. This paper addresses shortcomings of the current models through a critical comparison of their accuracy. It then proposes a new model based on genetic programming (GP), which takes into consideration a more comprehensive set of parameters related to blast design, explosive properties and rock mass description. Data used to derive the model was collected from blast trials conducted in conjunction with Orica Mining Services. The fragmentation of the material after blasting was estimated using Orica’s image analysis software, Powersieve 3®. The current prediction models compared included the Kuz-Ram, corrected Kuz-Ram and the Kuznetsov-Cunningham-Ouchterlony (KCO) model using the three-parameter Swebrec function. It has been found that inclusion of the Kuz-Ram correction factor offers a reasonable improvement in the accuracy of prediction for both the Kuz-Ram and KCO models. Orica’s blasting prediction software, Blast Design Assistant®, was used as an additional tool to evaluate both the accuracy of the Swebrec function to predict fragmentation, and its capacity as the underlying model of a software program. The GP approach predict 80 per cent passing size of the muck pile with a high level of accuracy at CITIC Pacific’s Sino Iron magnetite operation in Western Australia. Further validation of the proposed equation is required to determine its effectiveness at other mine sites and as a general blasting fragmentation prediction model. INTRODUCTION Blasting in mining has been used for centuries to effectively and efficiently break rock for excavation. The fragment size after blasting has a major impact on the performance of all aspects of the mining operation and on the economics of the mine and mill (Marton and Crookes, 2000). The term ‘rock fragmentation’ is used to describe the range of sizes, including the distribution, of rock portions after blasting. Fragmentation size affects all downstream processes, from load and haul (including immediate digging), crushing and grinding, through to processing. Improved fragmentation through blasting can increase diggability, thereby increasing productivity of loaders and excavators, reduce the need for secondary breakage of oversize material, increase mill throughput and decrease energy consumption in crushing (Hudaverdi, Kulatilake and Kuzu, 2010). Therefore, the ‘mine to mill’ approach has gained popularity, optimising blast design for the profitability of the overall operation, rather than individual processes (Hudaverdi, Kulatilake and Kuzu, 2010). Each operation has an optimum fragment size, which when achieved by blasting, will result in the greatest efficiency of the operation. Tools and technologies that can predict the performance of individual blasts make it possible to ensure that the optimum fragment size will be achieved every time (Marton and Crookes, 2000). Prediction tools make it possible to adopt real-time adjustments to the blast design as rock properties vary between blast areas. Factors affecting the outcome of a blast are grouped into three main categories: blast design variables, rock properties and explosive properties. Rock properties can vary significantly throughout the orebody and overburden material, but are not variables that can be controlled by the blast design engineer. The blast design, and to a lesser degree the explosive properties, are variables, which can and should be changed to achieve the desired fragmentation of the resulting muck pile. Fragmentation Prediction Models Blast fragmentation models were reviewed in chronological order to observe developments in parameter inclusion, complexity and practicality. Kuznetsov (1973) developed the fundamental equation relating mean particle fragment size and applied blast energy per unit volume of rock. Later, a standardised comparison of ammonium nitrate/fuel oil (ANFO) to trinitrotoluene (TNT) was incorporated, yielding the following equation 1. SAusIMM, Graduate Mining Engineer, School of Civil, Environmental and Mining Engineering, University of Adelaide, SA 5005. Email: [email protected] 2. SAusIMM, Graduate Mining Engineer, School of Civil, Environmental and Mining Engineering, University of Adelaide, SA 5005. Email: [email protected] 3. MAusIMM, Lecturer, School of Civil, Environmental and Mining Engineering, University of Adelaide, SA 5005. Email: [email protected] Mining Education Australia – Research Projects Review 2012 37 A Holland, M Iles and M Karakus for the prediction of the mean fragment diameter, XM (cm) (Aghababaei et al, 2009; Kuznetsov, 1973): XM = BI * K-0.8 Qe 1/6 `S 115 ANFO 19/30 j (1) where: A = 0.06 # ^ RMD + JF + RDI + HF h BI: Lilly’s Blastability Index (7 for medium hard rocks; 10 for hard, but highly fissured, rocks; and 13 for very hard, weakly fissured rocks) K: powder factor, kg/m3 Qe: mass of TNT equivalent in energy to the explosive charge in one blast hole, kg SANFO: relative weight strength of the explosive to ANFO Rosin and Rammler’s (1933) historic empirical distribution was obtained from crushed coal data. The Rosin-Rammler formula was used to predict the fragment size distribution of blasted rock in the form: X n RM = 1 - e- c Xc m (2) RM: proportion of material passing the screen X: screen size, cm XC: characteristic size through which 63.2 per cent of particles pass, cm n: index of uniformity In 1983, Cunningham published a historic paper outlining his empirical fragmentation model, combining the Kuznetsov and Rosin-Rammler formulae with a unique algorithm deriving the coefficient of uniformity, n (Ficarazzo and Morin, 2006). First rearranging the Rosin-Rammler Equation (2) gave: (3) Cunningham (1983) identified that substituting the mean passing size XM from Kuznetsov’s equation, with R=0.5 (assuming 50 per cent of material passes through the mean size), into Equation 3 yielded: XM XC = n 0.693 (4) In addition to this, Cunningham (1983) suggested an equation estimating the blasting parameter, n: 0.5 n = ` 2.2 - 14 B j` 1 + S j ` 1 - W j` L j D 2 2B B H where: B: burden, m S: spacing, m D: borehole diameter, mm W: standard deviation of drilling accuracy, m L: total charge length, m H: bench height, m TABLE 1 Rock factor parameters (Aghababaei et al, 2009). RMD (rock mass description) Powdery/friable 10 Vertically jointed JF* Massive 50 JPS (vertical joint spacing) <0.1 m 10 0.1 m to MS 20 MS* to DP* 50 Dip out of face 20 Strike perpendicular to face 30 Dip into face 40 RDI (rock density influence) RD, rock density (t/m3) RDI = 25RD* -50 A fragmentation curve can be plotted with values for the characteristic size XC and uniformity index n (Aghababaei et al, 2009). X XC = n - ln ^ 1 - R h M (6) JPA (joint plane angle) where: 38 Another important recommendation by Cunningham (1983) was the adaptation of Lilly’s Blastability Index, BI, shown in Table 1, to better describe the rock factors in Kuznetsov’s model (Aghababaei et al, 2009). Attempting to account for density, mechanical strength, elastic properties and structure, the equation for rock factor, A, is: (5) HF (hardness factor (GPa)) HF = Y/3 If Y<50 HF = UCS*/5 If Y>50 * Meaning Unit MS Oversize m DP Drilling pattern size m Y Young’s modulus GPa Uniaxial compressive strength MPa UCS JF=JPS+JPA Whilst the Kuz-Ram fragmentation model has been fundamental in the development of fragmentation theory, there are some important drawbacks. A major issue with the Kuz-Ram model is inadequate parameter inclusion. Due to the simplified nature of the Kuz-Ram model, some important blasting aspects are not explicitly accounted for, including geomechanical properties and blast initiation timing. The Kuz-Ram model generally gives poor prediction for the fines fraction. In addition, there is no upper limit to fragmentation size, which is highly unrealistic. In 2004, Spathis identified a technical error in the original Kuz-Ram model. The error arose from the assumption that the mean passing size from Kuznetsov’s equation could be equated with 50 per cent passing size in the Rosin-Rammler Equations 3 and 4. The correct expression for mean fragment size XC was in fact presented in Kuznetsov’s original paper as: XC = XM C` 1 + 1 j n (7) Mining Education Australia – Research Projects Review 2012 MetaheurIStIc analySIS of current open cut BlaStIng fragMentatIon predIctIon ModelS The upper limit of the fragmentation size (XMAX) is defined as the minimum of the in situ block size, the blasthole spacing or the blast hole burden. where: is the gamma function Γ(n) = (n - 1)! Γ This corrected expression was shown to reduce errors by 179 - 105 per cent for uniformity index n in the range 0.8 2.2. The error partially explained the underestimation of fines from the original Kuz-Ram model (Spathis, 2004). To apply to the Kuz-Ram model, a multiplying factor g(n) was created as a ratio of the two XC functions, to be applied in the calculation of XM : ln ^ 2h1/n g (n) = C^ 1 + 1/nh (8) In 1999, two models were published by the Julius Kruttschnitt Mineral Research Centre (JKMRC) to address the issue of poor fit of the Kuz-Ram model to actual sieved blast data: the crush zone model (CZM) and the two-component model (TCM) (Ouchterlony, 2005). These models used multiple applications of the Kuz-Ram model to describe the fragmentation distribution. These two models were considered more realistic, as they contain more parameters than a single application of the Kuz-Ram model; however they are considerably more complex, and may require data input from sample blasts (Ouchterlony, 2005). The CZM model has also attracted criticism for the incorrect assumption that almost all fines are generated in the blast crush-zone. The Swebrec function was developed to address the residual flaws of the Kuz-Ram and JKMRC models. Ouchterlony (2005) fitted the Swebrec function to hundreds of sets of blasted fragmentation data, with excellent results. The Swebrec function, depicted in Equation 9, generates a fragmentation curve of mass percentage passing a given sieve size, shown in Figure 1. In addition, a third parameter was introduced in the Swebrec function: the curve-undulation parameter, b,given in Equation 12. The curve-undulation parameter results in an inflection point on the fragmentation curve (in log-log space), allowing for very accurate non-linear fines prediction. X b = = 2 ln 2× ln e MAX oG n X50 n is Cunningham’s uniformity index. The improvement in fragmentation prediction by the Swebrec function led to the development of a new model: the Kuznetsov-Cunningham-Ouchterlony (KCO) model. The KCO model was created by combining the Kuznetsov equation for X50 with the Swebrec function describing particle distribution. The KCO model offers distinct improvements over the Kuz-Ram model, with better prediction capacity in the fines and a finite upper limit to block size. Blast design variables are included via the curve undulation parameter, b, which uses n in its expression. The use of the Kuznetsov equations ensures rock mass properties and powder factor are also accounted for. The g(n) factor, which is applied to the Kuz-Ram model to offer the corrected Kuz-Ram, could also be incorporated in the KCO model. The factor is applied to both models in Kuznetsov’s formula, given in Equation 13. The Swebrec function already offers a vast improvement for prediction in the fines region; therefore, it may not be necessary to include this factor in the KCO model. (9) where: XMAX is the upper limit of the fragmentation size X50 is the fragmentation size of 50 per cent passing b is the curve undulation parameter j (13) 19/30 j In the current fragmentation prediction models, many physically significant parameters are not accounted for. Inclusion of more parameters was expected to increase the accuracy of fragmentation prediction, specifically considering the Kuz-Ram and corrected Kuz-Ram, KCO and Swebrec models; stemming, subdrill and explosive distribution, density of explosives, tensile strength, Posission’s ratio of rocks, joint aperture and joint plane area are not included in these models. The inclusion of the g(n) factor into the KCO model is a relatively new advancement and, as such, its use in prediction models requires further investigation. The 50 per cent passing size is defined as in Kuz-Ram: ANFO ANFO 19/30 reSearch gapS Z R Vb_-1 X ] S ln c MAX m W b ] Wb X P^ x h = [ 1 + S S XMAX W ` ] S ln e Wb ] S X50 o W b \ T Xa `S `S 115 Although the Swebrec function has improved the prediction model for blast fragmentation, it has not incorporated any further parameters than those that have been included since the development of the Kuz-Ram model in 1983. The inclusion of further parameters may be necessary to offer a more realistic and reliable prediction model. Incorporating more parameters also offers a more tailored prediction model, specific to each mine site. Fig 1 - Fragmentation curve generated by Swebrec function (Ouchterlony, 2005). 115 (12) where: X50 = g^nh AK-0.8 Qe 1/6 X50 = BI # K-0.8Qe 1/6 (11) XMAX = min (in situ block size, S, B) (10) MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Another identified ‘gap’ in the inputs to these models is the description of maximum fragment size. The expression 39 a holland, M IleS and M karakuS for XMAX is a simple selection of the minimum value of in situ block size, burden or spacing. In reality, the maximum fragment size will be much more variable, especially in the case of highly fractured rock masses. A comprehensive understanding of rock mass structure, and all in situ rock mass characteristics is, therefore, pivotal for blast fragmentation modelling. All rock properties used as inputs have potential to be improved with better understanding of the rock mass. Identifying a new expression for XMAX, or suggesting methods for collection of more detailed geological data, is outside the scope of this research paper. Methodology A series of trial blasts were undertaken to provide a data set, which was used for the three main areas of research: 1. Blast Design Assistant® (BDA) analysis 2. development of a new prediction equation using genetic programming 3. comparison of current models, BDA and the newly proposed equation and investigation into the benefit of the g(n) factor. The blasts were conducted at CITIC Pacific’s Sino Iron Project, located at Cape Preston, 100 km south west of Karratha, in the Pilbara region of Western Australia. Results were collected from nine specifically designed trial shots, with assistance from Orica Mining Services. The actual fragmentation of the material after blasting was estimated using Orica’s image analysis software Powersieve 3®. Blast design assistant analysis This phase of the research involved the use of Orica’s blasting prediction software, BDA. The software utilises the threeparameter Swebrec function to predict the fragmentation distribution of a blast, based on various blast design and geological inputs. The program requires the same input parameters required by the Swebrec function. BDA also uses a ‘base case’ to improve predictions. Information required for the base case includes all blast design, explosive and rock property inputs for a single blast area, as well as the ‘result’ or actual fragmentation distribution of that blast, which is imported from Powersieve 3®. Direct output from the program includes the 20 per cent, 50 per cent, and 80 per cent passing sizes, a graphical fragmentation distribution, illustrated in Figure 2 and the Swebrec parameters: b, XMAX and X50. It is these output variables that were used to compare the accuracy of BDA’s predictions against the actual values obtained from Powersieve 3®. The BDA software served as a tool to evaluate both the accuracy of the Swebrec function to predict fragmentation, and its capacity as the underlying theoretical model of practical, user-friendly software. genetic programming analysis The research addressed blast design, explosive and rock property parameters that are absent in current prediction models. These variables have significant physical impact on the fragmentation outcome of a blast and, therefore, their inclusion into prediction models is recommended. Using all current design parameters, as well as additional parameters, an equation to predict the 80 per cent passing size of the muck pile after blasting was developed using genetic programming. Model comparison Finally, a model comparison was carried out by comparing each model to the actual fragmentation results from site and calculating the prediction error. The models compared in this research were the Kuz-Ram, corrected Kuz-Ram and KCO (using the three-parameter Swebrec function with and without the inclusion of the g(n) correction factor), the BDA software and the newly proposed equation. The models were compared in their ability to predict the 80 per cent passing size. Fig 2 - Blast Design Assistant® output, the base case and predicted fragmentation curves. 40 MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Metaheuristic Analysis of Current Open Cut Blasting Fragmentation Prediction models The trials also assisted to clarify the effect of the inclusion of the g(n) factor in the KCO model. A conclusion as to whether the g(n) factor improves the accuracy of the KCO model was not expected as the number and variety of trials was limited. However, this trial laid the foundation for this investigation. Results Blast design assistant analysis For each of the nine trial shots, the following blast design and explosive properties were collected from site and entered into BDA: •• •• •• •• •• •• burden (m) and spacing (m) number of holes and their diameters (mm) bench height (m) and subdrill (m) stemming (m) total mass of explosive (kg) explosive density (g/cm3). The spacing, dip and dip direction of joint sets and bedding planes were also required by the software. The geological properties were entered for each of the ten ore zones present across the trial area: •• •• •• •• •• •• uniaxial compressive strength (UCS) (MPa) tensile strength, σt (MPa) Young’s modulus, Y (GPa) Poisson’s ratio density (t/m3) P and S-wave velocities (m/s). Each of the nine blast areas were used as the base case scenario to predict the fragmentation distribution of all eight remaining blast areas, resulting in 72 different predictions. The Swebrec parameters (b, X50 and XMAX) were predicted by BDA and then substituted into the Swebrec function (Equation 9) to generate the predicted fragmentation distribution for each shot. The actual fragmentation distribution for each shot was obtained from Powersieve 3® and the error could then be obtained at each increment of sieve size and a total average error calculated. A summary of the errors of the BDA prediction against the actual fragmentation distribution is displayed in Table 2. The accuracy of the fragmentation distribution predicted by the BDA program was also assessed using the 20 per cent, 50 per cent and 80 per cent passing sizes. These values were predicted and output directly from the program and, hence, could be compared to the actual percentage passing sizes obtained from Powersieve 3®; the error calculated is given Table 3. Genetic programming analysis This section of analysis was conducted using GPLAB, a genetic programming toolbox for MATLAB, developed by Silva and Almeida (Karakus, 2011). Genetic Programming (GP) automatically creates several computer programs (CP), which are composed of functions and terminals to solve a particular problem. The functions in this case were standard arithmetic operators: f(x) = {×, ÷, −, +, √, sin, cos, log, power, exp}. The terminals are the independent input values that the CPs use, along with the functions (arithmetic operators), to arrive at the solution or ‘result’, in this case the 80 per cent passing size. The terminal set included all parameters used in the current prediction models, with the addition of stemming, subdrill, explosive density, tensile strength and Poisson’s ratio. Mining Education Australia – Research Projects Review 2012 TABLE 2 Summary of average prediction errors for Swebrec fit. BDA fit base case 1 Blast design assistant prediction Average error (%) 2 3 4 5 6 7 43.1 26.3 41.3 18.9 60.4 45.7 80.9 22.2 1 3 4 5 6 7 4.9 10.0 3.9 3.9 1.9 11.0 8.8 16.5 1 2 4 5 6 7 0.5 12.2 1.8 1.5 3.0 7.9 6.6 13.1 4.7 1 2 3 5 6 7 9 6.2 2 3 4 0.8 8 8 8 8 9 9 9 1.6 1.5 2.4 7.6 6.2 12.6 4.4 1 2 3 4 6 7 9 0.4 10.4 5.1 3.8 4.5 9.2 8.0 13.9 1 2 3 4 5 7 8 9 0.8 14.0 4.5 2.7 5.4 4.5 1.3 2.7 7.9 1 2 3 4 5 6 8 9 6 7 0.2 8 1.6 9 5.2 3.7 6.1 5.3 2.0 1.7 8.7 1 2 3 4 5 6 7 9 14.4 5.1 3.2 6.1 4.9 1.4 1.2 8.4 7 8 9.8 6.4 6.1 7.3 4.1 14.8 1 0.0 8 7.3 2.2 12.0 5 42.4 2 3 4 5 6 17.0 13.1 15.9 8.9 23.3 20.5 29.5 5.4 6.0 5.6 17.2 Initially there were 18 parameters all of which were required to be included at least once in the equation that is generated by the program. Dimensionless variables were made by dividing parameters of the same units of measurement with each other. The result parameter (80 per cent passing) was also made dimensionless by dividing it by joint plane spacing, both in meters. Additionally, charge length was removed as a parameter as this was accounted for by stemming, subdrill and bench height. Similarly, charge per blast hole was removed as this was used in the calculation of powder factor. Some variables were simply added or subtracted to one another, which further reduced the number of input variables and improved the chance of the program using all variables in the generated equation. The aforementioned changes reduced the total number of input variables from 18 to 7, which represented 16 individual parameters. An iterative approach was taken: for each attempt the terminals were manipulated, input into GPLAB and an equation was generated using the data, to predict the result parameter (P80/JPS). This equation was examined for parameter inclusion, overall simplicity and function fitness. The seventh attempt proved to be successful as all variables were employed in the equation offered by GPLAB, given in Equation 14, and the fitness (sum of absolute difference, SAD) of 0.008919 demonstrates an excellent fit between the predicted P80 from the generated equation and the actual P80. As the result used in GPLAB was P80/JPS, the generated equation is simply multiplied by joint plane spacing to obtain the equation, which predicts the 80 per cent passing size. 41 A Holland, M Iles and M Karakus TABLE 3 Summary of average prediction errors based on percentage passing. Average error (%) per shot 1 2 3 4 5 6 7 8 9 Total average error (%) P20 53.7 32.8 27.3 26.9 32.4 62.7 41.1 94.7 27.8 44.4 P50 51.3 24.2 21.1 21.6 21.1 33.8 33.8 36.2 27.6 30.1 P80 45.9 20.2 18.6 17.9 19.9 21.0 27.3 21.0 41.4 25.9 Blast Z ]] x5 [ f log x1× log > ln x3 x7.e Hp o P80 = ln ]] e 2 10 e x4 \ ×JPS _ x2 . log x6b 10 b x4 ` b b a(14) Using logarithmic rules, Equation 14 can be simplified to: x5 x3 Go P80 = ln )eeln x1×e ×x7×0.627× ln =ln x4 x2 × ln x6 JPS ×0.434 × 1 x4 (15) where: |1 = Burden Spacing |2 = Stemming + cos (Joint Plane Angle) Hole Length | 3 = Drillhole diameter Hole length |4 = Powder Factor + Rock Density Explosive Density Powder Factor |5 = |6 = Re lative Weight Strength 100 Young' s Modulus + Tensile Strength Uniaxial Compressive Stength |7 = Poisson' s Ratio The new equation was intended to provide a simple tool for a quick analysis of the predicted P80 of a blast. Engineers can simply input the design and rock property variables to determine the P80 that will be achieved for each blast. Model comparison The final step in the data processing was to compare the accuracy of the Kuz-Ram, corrected Kuz-Ram, KCO model and the developed model, using the actual field data. Each of these models required specific inputs, some of which called for manipulation of gathered field data, or assumptions. The BDA predictions were included in this comparison as an additional test of the Swebrec function. The P80 as a direct output from BDA was averaged, and the P80 was also calculated from the averaged output Swebrec parameters. The error in prediction was calculated for each blast area and the average absolute error for each prediction model was used to rate the models by accuracy (see Table 4). The calculated P80 values are tabulated in Table 5. 42 TABLE 4 Average absolute error for each prediction method. Model Absolute error (%) Swebrec 100 Kuz-Ram 93 Kuznetsov-Cunningham-Ouchterlony: Swebrec with g(n) factor 57 Corrected Kuz-Ram 51 Blast Design Assistant (parameters averaged and P80 calculated) 50 Proposed model 9 Blast Design Assistant (P80 averaged) 5 ANALYSIS and INTERPRETATIONS Blast design assistant analysis The accuracy of BDA as a fragmentation prediction tool was measured in two ways, through the overall average error of the fragmentation curve and through the percentage passing parameters. Overall prediction error The overall prediction error was found to be quite low (5.37 per cent - 7.34 per cent) for most cases, and significantly higher when predicting with a base case of blast area 29 (17.23 per cent error) and blast area 60 (42.35 per cent error). The consistently high errors when predicting from an area could be an indication that the actual fragmentation data input collected from the CITIC Pacific Sino (PS) Iron project site in June 2012 may not be accurate. Percentage passing errors The percentage passing errors were averaged over each blast area and decreased from P20 to P50 and P80, with respective errors 44.4 per cent, 30.1 per cent and 25.9 per cent. These errors are quite high, and could stem from a number of factors. The actual percentage passing values were read directly from the Powersieve 3® graph, a process fraught with human error. The precision of the readings is likely to be low, and random in the magnitude of the error. The Powersieve 3® program is also assumed to contain some inherent error in estimating percent passing size. This error can be minimised by thorough preprocessing of images, and a good understanding by the user of the programs functions. It should be noted that the P80 errors were consistently the lowest, indicating P80 is the most accurate parameter to predict. This reinforces the use of P80 as the prediction variable in the generated model. Mining Education Australia – Research Projects Review 2012 Metaheuristic Analysis of Current Open Cut Blasting Fragmentation Prediction models TABLE 5 Actual and predicted P80 values. Blast area 80% Passing Size, P80 (m) Actual Kuz-Ram Corrected Kuz-Ram Swebrec KCO BDA (parameters averaged) BDA (P80 averaged) Proposed model 1 0.310 0.449 0.348 0.464 0.361 0.262 0.216 0.250 2 0.230 0.448 0.350 0.464 0.364 0.355 0.230 0.229 3 0.190 0.350 0.272 0.364 0.284 0.303 0.190 0.171 4 0.240 0.399 0.312 0.414 0.325 0.362 0.240 0.237 5 0.225 0.439 0.342 0.454 0.356 0.279 0.253 0.230 6 0.215 0.462 0.361 0.478 0.375 0.372 0.215 0.226 7 0.200 0.495 0.384 0.511 0.399 0.375 0.200 0.224 8 0.200 0.414 0.323 0.429 0.336 0.363 0.199 0.152 9 0.250 0.462 0.361 0.478 0.375 0.239 0.250 0.242 Proposed model The new equation, developed via genetic programming in GPLAB to predict the 80 per cent passing size of a muck pile, has incorporated additional parameters compared to current models and hence offers a robust description of the factors that affect blast outcomes. In theory, this equation should, therefore, offer highly accurate predictions in various rock types, with a variety of design parameters. Presently, the equation offers highly accurate predictions of the 80 per cent passing size for the geological conditions specific to CITIC Pacific’s magnetite site. This is to be expected as the field data that was input and utilised by GPLAB to develop the equation came from this site. Therefore, validation of the newly proposed model against a wide variety of field data will be beneficial to evaluate the equation’s merit. The equation’s use of 18 different parameters, which describe a myriad of factors affecting fragmentation from blasting, is vigorous and should be encouraged. This is a development on the present position of fragmentation prediction modelling. The equation, although initially appearing complex, is a straightforward substitution of variables (χ1 – χ7) and their corresponding design, explosive or rock property parameters, all of which are readily available at a mine site. Often, this information is required for other purposes and is, therefore, being recorded at site anyway. Therefore, collection of the input information required by the P80 equation will not only be readily available, but should not interfere with current operations nor create more work. When used with everyday computer applications such as Microsoft Excel, the formula for the equation can be set up so that the design parameters are simply entered into corresponding cells, offering a quick and easy prediction. The equation can be utilised to check the proposed design of a shot to determine the resulting P80 of the muck pile. If the prediction does not satisfy the target P80 required by downstream processes, changes to the blast design can be made immediately until the target P80 is reached. All members of the drill and blast team can simply and quickly enter the parameters of the proposed design into the spreadsheet containing the prediction equation. The equation is, therefore, highly practical and efficient. Model comparison The prediction error for the 80 per cent passing size was compared for each model. The Kuz-Ram, corrected Kuz-Ram, Swebrec and KCO models consistently over-estimated the P80 Mining Education Australia – Research Projects Review 2012 size, while the two BDA predictions and the generated model both over- and under-estimated the P80 size. It was expected that the Swebrec function would provide a highly accurate prediction; however, the results suggest otherwise, showing the Swebrec function to have the worst error (100 per cent) of all the models. Therefore, the inputs were investigated and the XMAX, XM and b values were examined to try to identify where the high error stemmed from. Both the b and XM values were calculated from the same data as used in the Kuz-Ram model. The XMAX value, however, was taken as the minimum value of burden and spacing, as the in situ block size was not known. The XMAX values for the Swebrec and KCO models are much higher than the actual XMAX, taken from CITIC Pacific Sino Iron site. The XMAX values predicted from each base case in BDA were averaged, and were also quite high. As the XMAX value is a direct input into the Swebrec equation, this large over-estimation would account for some of the overall prediction error. It is pleasing to note that the corrected Kuz-Ram and KCO model have increased accuracy due to the inclusion of the g(n) factor. In both cases (corrected Kuz-Ram versus KuzRam and Swebrec versus KCO) the error was almost halved using the g(n) factor. This is a promising result, and should be developed on with further research to clarify the effect of the g(n) factor on prediction models. The results showed that the direct P80 size prediction from BDA was the most accurate of all the models. An average error of only five per cent was found over nine sets of blast data. This showed that BDA is a good prediction tool, and makes good use of the Swebrec function. The program also includes most of the available parameters. The program itself proved to be interactive and user-friendly. Also worth note is that BDA has two other modelling functions: HelFire for damage, and DMC for the cast and heave of a muck pile. The model generated in this research showed high accuracy in predicting the P80 size of the trial data, but this is due to the model being constructed from this data. Unfortunately, more data was not able to be collected to validate the model with a new data set. Given a longer time frame, more field data could be collected as it became available, and used to validate the results found. CONCLUSIONS A new model predicting the 80 per cent passing size was developed using genetic programming and field data. 43 A Holland, M Iles and M Karakus It incorporates 18 independent parameters to represent the variables that affect blast fragmentation: blast design, explosive and rock properties. Five new parameters, which are not used by earlier models: stemming, subdrill, explosive density, Poisson’s ratio and tensile strength were introduced in this model. All input parameters for the newly developed model can be easily obtained from site, adding to the ease of practical application of the equations. The equation offers a convenient tool, which allows for immediate adjustments to proposed blast designs to ensure the target P80 is achieved. A comparison study was carried out on the prediction of 80 per cent passing size, using the Kuz-Ram, corrected KuzRam, Swebrec and KCO models, the BDA program and the new prediction equation developed in this research. The data obtained from site was used as inputs in the study. Results from the BDA were the most accurate, followed by those from the newly developed equation. Of the remaining models, the results from the Swebrec function had the lowest accuracy of prediction. This could be due to the difficulty experienced in obtaining an accurate value for the significant parameter XMAX. The inclusion of the g(n) factor into the Kuz-Ram and KCO models proved to be worthwhile as the accuracy of prediction improved significantly. REFERENCES Aghababaei, H, Gheibie, S, Hoseinie, S H and Pourrahimian, Y, 2009. Modified Kuz-Ram fragmentation model and its use at the Sungun Copper Mine, International Journal of Rock Mechanics and Mining Sciences, 46:967-973. 44 Cunningham, C, 1983. The Kuz-Ram model for prediction of fragmentation from blasting, in Proceedings First International Congress in Rock Fragmentation by Blasting, Lulea, 2:439-453. Ficarazzo, F and Morin, M A, 2006. Monte Carlo simulation as a tool to predict blasting fragmentation based on the Kuz-Ram model, Computers & Geosciences, 32:352-9. Hudaverdi, T, Kulatilake, P H S W and Kuzu, C, 2010. Prediction of blast fragmentation using multivariate analysis procedures, International Journal for Numerical and Analytical Methods in Geomechanics, 35:1318-1333. Karakus, M, 2011. Function identification for intrinsic strength and elastic properties of granitic rocks via Genetic Programming (GP), Computers and Geosciences, 37:1318-1323. Kuznetsov, V M, 1973. The mean diameter of fragments formed by blasting rock, Sov Min Sci, 9:144-8. Marton, A and Crookes, R, 2000. A case study in optimizing fragmentation, in Proceedings of The Australasian Institute of Mining and Metallurgy, 1:35-43. Ouchterlony, F, 2005. The Swebrec© function: Linking fragmentation by blasting and crushing, Trans Inst Min Metall, Mining Technology, 114:A29-A44. Rosin, P and Rammler, J, 1933. The laws governing the fineness of powdered coal, Journal of the Institute of Fuel, 7:29. Spathis, A T, 2004. A correction relating to the analysis of the original Kuz-Ram model, International Journal for Blasting and Fragmentation, 8(4):201-5. Mining Education Australia – Research Projects Review 2012 Ventilation Requirement for Electric Vehicles in Underground Hard Rock Mines – A Conceptual Study M Kerai1 and A Halim2 ABSTRACT In the past five years, electric power price in Australia has increased significantly and is likely to continue to increase in the foreseeable future. This can make a mine uneconomic to operate. One option to reduce underground mine power consumption is to reduce ventilation power consumption. The electrical power required for the ventilation system is one of the major components of the total mine electrical power consumption, which typically represents more than one-third (de la Vergne, 2003). To reduce ventilation power consumption, ventilation requirement must be reduced. One option to do this is to replace diesel vehicles with electric ones. An electric motor produces zero emissions and only emits one-third the heat of an equivalent diesel engine. Airflow specification can therefore be less (Marks, 2012). Ventilation requirement for an underground hard rock mine is determined based on the engine power of diesel vehicles used in the mine. The current Australian regulatory requirement for a diesel vehicle ranges from 0.05 to 0.06 m3/s per kW engine power. However, no such requirement stated for an electric vehicle. This paper evaluates the ventilation requirement of an electric vehicle operating in an underground hard rock mine. Introduction In the past five years, electric power price in Australia has increased significantly and is likely to continue to increase in the foreseeable future. This can make a mine uneconomic to operate. The electrical power required for the ventilation system for a mine is one of the major components of the total electrical power consumption, which typically represents more than one-third (de la Vergne, 2003). To reduce ventilation power consumption, ventilation requirement must be reduced. One option to achieve this is to replace diesel vehicles with electric ones. An electric motor produces zero emissions and only emits one-third the heat of an equivalent diesel engine. Airflow specification can therefore be less (Marks, 2012). Currently ventilation requirement in Australian underground hard rock mines is determined by multiplying the rated diesel engine power of all vehicles with the regulatory airflow requirement. For example, 0.05 m3/s per kW rated engine power is the requirement in Western Australia Mines Safety and Inspection Regulation (WAMSIR) 1995, regulation 10.52 (6) (WA Government, 1995). Currently, there is no such requirement for electric vehicles in Australia. The only regulation concerning ventilation for electric vehicles is WAMSIR 1995 regulation 9.34 which states that a minimum air velocity of 0.25 m/s is maintained in all underground areas in the mine where vehicles or locomotives powered by electricity is used. However, this means that the airflow quantity will be different depending on the dimension of the area where they work. For example, a 14 tonne LHD unit with motor power of 178 kW will have airflow quantity of 4 m3/s in a 4 m × 4 m heading and 6.25 m3/s in a 5 m × 5 m heading. This means that the smaller heading will have higher temperature than the larger one. This is not the right approach as the amount of heat produced by an electric vehicle depends on the size of the motor, not on the dimension of the heading. The airflow quantity should therefore be governed by the size of the motor. Methodology The first step of this research was to quantify the contaminant produced by an electric heavy vehicle, which is heat. The measurement to quantify heat produced by an electric heavy vehicle was carried out in Rio Tinto’s Northparkes Copper-Gold mine in New South Wales (NSW). This mine was selected as it exclusively uses electric vehicles, which are electric load haul dump (LHD) units, as production equipment. The LHD unit is Sandvik LH514E. Data collected were Dry bulb (DB) temperature, wet bulb (WB) temperature, Barometric pressure, airflow quantity (air velocity and airway dimension). By using psychometric equations, the amount of heat produced by an electric LHD can be calculated. In addition to this, power consumed by the unit was also determined by measuring current, voltage and power factor. 1. Western Australian School of Mines (WASM), Curtin University, Kalgoorlie WA. 2. MAusIMM, Western Australian School of Mines (WASM), Curtin University, Kalgoorlie WA. Email: [email protected] Mining Education Australia – Research Projects Review 2012 45 M Kerai and A Halim This is to validate a popular assumption that heat produced by an electric vehicle is equal to the power consumed by it. The literature review found that this assumption had never been validated prior to this research. Ventsim Visual simulations were conducted as the next step. The aim of these simulations was to determine the total airflow quantity that is required in order to have Wet Bulb (WB) temperature at the deepest part of the mine less than or equal to 30°C. Although the common design limit used is 28°C, this limit is considered to be out of date since it was determined based on the working condition in the old days where the application of air conditioned cabin vehicles was not extensive (Van den Berg, 2012). Today, air conditioned cabin vehicles are common in mines in developed countries. These vehicles assist heat stress management by providing microclimate cooling, ie the vehicles’ operator spend most of his/her time in cool environment. Due to this, Rio Tinto’s Resolution Copper project in USA uses design limit of 30°C (Van den Berg, 2012). Ventsim Visual models of two currently operating hard rock underground mines in Kalgoorlie were used with their known fleet of machinery. The first is a deep mine which extends to over a kilometre deep and the second is a shallow mine which extends to 600 m deep. The models were generated using two distinct geothermal parameters, namely Kalgoorlie (cool strata) and Mount Isa (hot strata). These parameters can be altered in Ventsim Visual software. One input of a thermodynamic simulation using Ventsim Visual is the heat emitted by an electric vehicle. For this purpose, it was essential to determine the ratio between the heat emitted and total electric motor output of the LH514E. The heat emitted was measured as 145 kW and the total motor output of this LHD is 178.7 kW. This results in a ratio of 81 per cent. However, since this ratio varies in every electrical equipment, a conservative approach was taken in the simulation. Therefore, it was assumed that the amount of heat emitted from every conceptual electric vehicle is equal to its motor output. heat. However, Ventsim Visual is unable to calculate sensible and latent heat from a fuel cell. Hence, the assumption above was used. QUANTIFYING HEAT EMITTED BY A MINING ELECTRIC HEAVY VEHICLE As described before, the first step of this research was to quantify the contaminant produced by an electric heavy vehicle, which is heat. The measurement to quantify heat produced by an electric heavy vehicle was carried out in Rio Tinto’s Northparkes Copper-Gold mine in New South Wales (NSW). This mine was selected as it exclusively uses electric vehicles as production equipment. Northparkes’ heavy electric machinery fleet is mainly comprised of electric Load Haul Dump (LHD) units, specifically the Sandvik LH514E. The machine is operating on a delivered voltage of about 1,000 V. It is equipped with three electric motors, one drive motor (132 kW), one pump motor (45 kW), and one fan motor (1.5 kW). The LH514E carries an arrow head design bucket with a capacity of 7 m3. To determine heat emitted by an LH514E, the following parameters were measured upstream and downstream of the LHD operating in a production drive, refer to Figure 1: •• dry bulb (DB) temperature •• wet bulb (WB) temperature •• barometric pressure. It was then necessary to determine the motor output of every conceptual electric vehicle. The only way to do this is to compare the power output of a diesel vehicle and an electric vehicle that have same workload. The only large vehicle that is manufactured as a diesel and as an electric is Sandvik’s LH514 LHD unit. The motor output of the LH514 electric is 70 per cent of its diesel equivalent (Sandvik Mining, 2012). Therefore, it was assumed that every conceptual electric vehicle in the fleet has motor power of 70 per cent of its diesel equivalent. With this assumption, the existing mining diesel fleet was converted into a conceptual electric fleet by multiplying their engine power with 70 per cent. The final step was running Ventsim Visual simulations. After thermal parameters and heat emitted from conceptual electric vehicles were set, the primary fan was adjusted with various values of fixed quantity that is sufficient enough to have the WB temperature less than or equal to 30°C at the deepest part of the mine. This fixed quantity was then subtracted with leakage and fixed facilities quantities and then divided by the total fleet output power to obtain unit requirement as m3/s per kW output. Refrigeration plant was simulated if required. It was assumed that these vehicles are battery powered, which do not produce moisture. Hence all the machinery will be inputted as a point source of sensible heat. Although an electric vehicle can have fuel cell as its power source, fuel cell produces moisture and therefore emits sensible and latent 46 Fig 1 - Locations of LH514E heat quantification measurements. In addition to these, airflow quantity that flows in the production drive was measured by measuring air velocity and drive cross sectional area. Using these measurement results and psychometric equations, the amount of heat that was emitted by the unit could be calculated. The measurements were taken while the LHD unit had come to a halt. Unlike diesel vehicles, there is no difference between idling and full throttle as the electric motor always runs at constant speed. This is because when accelerator pedal is pressed, the motor is engaged to the drive wheel and when the pedal is released, the motor is disengaged from the drive wheel. Mining Education Australia – Research Projects Review 2012 Ventilation Requirement for Electric Vehicles in Underground Hard Rock Mines – A Conceptual Study Various instruments were required in order to complete the measurements. They were supplied by the Western Australian School of Mines (WASM). The Kestrel4000 pocket weather meter was used to measure the Dry Bulb (DB) and Wet Bulb (WB) temperatures, the Druck DPI 740 digital barometer was used to measure the barometric pressures, Alnor RVA501 digital anemometer was used to measure air velocity, and the Leica Disto was used to measure dimension of the airway. The main aim of the measurements is to have sufficient and reliable data regarding the heat emitted by the LHD unit. Therefore, the measurements were undertaken as close as possible to the LHD unit in order to prevent the data being interfered from other heat sources present in an underground hard rock mine such as ground, the concrete pavement, the caved ore in draw points, and water in the production drive. However, it was also understood that there is a high safety risk for personnel to get close to a running electric LHD unit. Therefore, a Risk Assessment (RA) and a Job Safety Analysis (JSA) were conducted before the measurements were taken. There is a popular assumption that heat emitted by an electric vehicle is equal to the power consumed by the vehicle. In order to validate this assumption, the current, voltage and power factor were also measured. Power consumption by the unit was calculated using the following equations: E = (√3 × V × I x cosθ)/1000 where E = power consumption (kW) V = voltage (V), I = current (Amp), and cosθ = power factor Once these calculations were done, comparison between power consumed by the LHD unit and heat emitted by it could be done and the assumption could be validated. Measurements were carried out across five points at the front and rear of the LHD unit to minimise the errors. Results of these measurements are shown in Table 1. As shown in Table 1, the variations between the values of each point do not deviate by a huge amount. The air quantity in the drive was measured as 17.2 m3/s Using these results, heat emitted by the electric LHD was determined using psychometric equations included in an Excel spreadsheet. Table 2 shows the heat emitted by the LHD for different trials. The average generated from Table 2 was then compared with the power consumed by the LHD. Power consumption was calculated as 147 kW. The voltage, current and power factor for the LHD was measured as 1000 Volts, 100 Amps, and 0.85 respectively. The heat generated (145 kW) and power consumption (147 kW) is very similar and the variation between these Table 2 Heat emitted by LH514E. Trial Heat (kW) 1 161.8 2 156.6 3 138.8 4 128.3 5 139.2 Average 145.0 values is due to the measurement errors. This finding validates assumption that heat generated by an electric vehicle equals to its power consumption. VENTSIM VISUAL SIMULATIONS Simulation of the deep mine The mine fleet (existing diesel and conceptual electric) of the deep mine is listed in Table 3. Not all vehicles listed in Table 3 are present inside the mine at the same time. The simulation was run with the vehicles listed in Table 4 operating in the mine at the same time. This is the typical operational situation in this mine. Jumbos and longhole drills were not included as they run with their small electric motor while drilling. Their motor output is so small and therefore can be negligible. As described before, two distinct geothermal parameters were used in this simulation, namely Kalgoorlie (cool strata) and Mount Isa (hot strata). This is to analyse the impact of two distinct geothermal conditions that are present in Australian mining areas upon the ventilation requirement of electric vehicles. Table 5 and 6 list the geothermal parameters for cool and hot strata respectively. In order to address the shortage of airflow in its deepest block, this mine utilises a booster fan. Table 7 shows the conceptual electric vehicle fleet that is used in the deepest block of the mine. This fleet was based on the actual diesel fleet used in this mining block. The electric motor output was estimated based on the 70 per cent diesel power assumption described before. The fixed quantity of the booster fan was adjusted along with that of the primary fan in order to find total quantity that will cause WB temperature less than or equal to 30°C in the deepest part of the mine. As with the fixed quantity in the primary fan, this fixed quantity was then adjusted to exclude leakage quantity and then divided by the total motor output shown in Table 7 to Table 1 Data collected at Northparkes Mine. Trial Front of load haul dump End of load haul dump DB (°C) WB (°C) Barometric pressure (Pa) DB (°C) WB (°C) Barometric pressure (Pa) 1 21.00 16.20 106 197.00 18.90 13.40 106,205.00 2 21.50 16.30 106 189.00 18.90 13.60 106,201.00 3 21.30 16.00 106 117.00 18.90 13.60 106,202.00 4 21.20 16.20 106 181.00 19.00 14.00 106,211.00 5 21.50 16.20 106 209.00 18.90 13.80 106,217.00 Average 21.30 16.18 106 178.60 18.92 13.68 106 207.20 Mining Education Australia – Research Projects Review 2012 47 M Kerai and A Halim Table 3 Deep mine fleet. Unit Diesel engine power (kW) Electric motor output (kW) Fleet size Total electric output (kW) Light vehicle 98 68.6 12 823.2 Light truck (stores) – long 221 154.7 1 154.7 Light truck (stores) – short 176 123.2 1 123.2 Charge up rig 104 72.8 1 72.8 IT 152 106.4 2 212.8 Jumbo 110 77 2 154 Production drill 104 72.8 2 145.6 Medium load haul dump 231 161.7 2 323.4 Large load haul dump 321 224.7 5 1123.5 Small truck 485 339.5 6 2037 Large truck 548 383.6 5 1918 Grader 152 106.4 2 212.8 Water cart 152 106.4 2 212.8 Shotcrete rig 82 57.4 1 57.4 Concrete agitator 170 119 Table 4 Vehicles operating in the deep mine at the same time. Equipment Fleet size 2 238 TOTAL 7809.2 Table 6 Mount Isa (hot strata) geothermal parameters. Parameters Values IT 1 Rock thermal conductivity Large truck 4 Rock thermal diffusivity Grader 1 Rock temperature at surface/portal Water cart 1 Geothermal gradient 19.92 °C per km vertical meters Large load-haul-dump 1 Airway wetness factor 10% Small truck 2 Average surface barometric pressure Light vehicle 5 Average airways age 1.23 W/m°C 0.55 × 10-6 m2/s Parameters Rock thermal conductivity Rock thermal diffusivity 25°C WB, 35°C DB Table 7 Electric vehicle fleet in the deepest block of the deep mine. Values 1.75 W/m°C 98.5 kPa 9 years Average surface temperatures Table 5 Kalgoorlie (cool strata) geothermal parameters. 28°C Unit 0.75 × 10-6 m2/s Motor output (kW) Fleet size Total output (kW) 23°C Light vehicle 68.6 5 343 Geothermal gradient 8.5°C per km vertical meters Charge up rig 72.8 1 72.8 Airway wetness factor 10% IT Rock temperature at surface/portal 106.4 1 106.4 77 1 77 Average surface barometric pressure 98 kPa Jumbo Average airways age 9 years Production drill 72.8 2 145.6 Large load-haul-dump 224.7 2 449.4 Average surface temperatures 23°C WB, 35°C DB Small truck 339.5 1 339.5 obtain unit ventilation requirement in m3/s per kW. No fixed facilities quantity was subtracted from this fixed quantity as the fixed facilities are located above this block. This value was then compared with that obtained from the fixed quantity in the primary fan. Both fixed quantities were then varied in the model until WB temperature of 30°C was reached in the deepest part of the mine and both produced similar unit ventilation requirement. Large truck 383.6 5 1918 Grader 106.4 1 106.4 Water cart 106.4 1 106.4 Shorcrete rig 57.4 1 57.4 Concrete agitator 119 1 119 TOTAL 3840.9 48 Mining Education Australia – Research Projects Review 2012 Ventilation Requirement for Electric Vehicles in Underground Hard Rock Mines – A Conceptual Study It was found that the mine located in Kalgoorlie (cool strata) requires a total quantity of 400 m3/s in its primary fan and 172.5 m3/s in its booster fan. A spot cooler of 500 kW(R) was required in the deepest block. The fixed quantity in the primary fans takes into account the 20 m3/s requirement for fixed facilities (fuel bay and magazine) and the leakage factor of 25 per cent. Hence the quantity required for the fleet had to be adjusted accordingly as shown below: Total quantity = 400 m /s 3 Quantity without leakage = 400/1.25 = 320 m3/s Quantity for vehicle fleet (less quantity for fuel bay and magazine) = 320 - 20 = 300 m3/s The total output power of the conceptual electric vehicles is noted to be 7809.2 kW. Hence, by simply dividing the required quantity of 300 m3/s by the total power output of 7809.2 kW, the unit requirement is 0.0384 m3/s per kW electric motor output. For the fleet used in the deepest block, the quantity that was allocated for the fleet was calculated by excluding 15 per cent leakage factor from the booster fan fixed quantity. It was found that the fleet quantity is 172.5/1.15 = 150 m3/s. The unit requirement was calculated by dividing this quantity by the total output power of the conceptual electric vehicles used in this block (3840.9 kW). The requirement is 0.0391 m3/s per kW, which is very similar with that calculated from the fixed quantity in the primary fan. It was therefore concluded that a requirement of 0.04 m3/s per kW applies to every electric vehicle used in this mine. This is 20 per cent less than the requirement for the diesel fleet which is 0.05 m3/s per rated kW diesel engine power. For the mine that is located in Mount Isa (hot strata), it was found that the requirement is similar with that in the mine located in Kalgoorlie (0.04 m3/s per kW). However, a 4.3 MW(R) surface refrigeration plant has to be utilised. This simulation shows that refrigeration plant is required in a deep mine located in hot strata. Without this plant, ventilation quantity will have to be increased significantly. However, this will result air velocity in the decline beyond manageable limit. Therefore, it is inevitable that refrigeration plant has to be utilised in a deep mine located in hot strata. Simulation of the shallow mine The mine fleet at this mine is listed below in Table 8 with the vehicles operating in a mine at the same time listed in Table 9. The shallow mine is a series ventilation circuit in which each level (ore drive) is ventilated by auxiliary fan and layflat duct. In addition to heat from the vehicles listed on Table 9, eight 90 kW auxiliary fans and one 180 kW decline development auxiliary fan were turned on in the model to simulate heat emitted by them. This reflects the typical condition in this mine where eight ore drives and decline development face are active at a time. The average airway age in this mine was three years. The shallow mine located in Kalgoorlie (cool strata) requires a total quantity of 70 m3/s at its primary fan without any refrigeration plant. This quantity includes 20 per cent leakage factor. Hence the ventilation requirement was obtained by adjusted this quantity to exclude leakage and then dividing Mining Education Australia – Research Projects Review 2012 Table 8 Shallow mine fleet. Unit Diesel engine power (kW) Electric motor output (kW) Fleet size Total electric output (kW) Light vehicle 98 68.6 5 343 Charge up rig 66 46.2 2 92.4 Small LHD 123 86.1 2 172.2 Medium LHD 142 99.4 2 198.8 Large LHD 298 208.6 2 417.2 Large truck 600 420 1 420 Medium truck 392 274.4 1 274.4 IT 82 57.4 2 114.8 Shotcrete rig 82 57.4 1 57.4 Concrete agitator 100 70 1 70 TOTAL 2160.2 Table 9 Vehicles operating in the shallow mine at the same time. Unit Fleet size Large truck 2 IT 1 Light vehicle 3 Small load haul dump 2 the adjusted quantity by the total output power of 2160.2 kW. This resulted in required quantity of 0.027 m3/s per kW electric motor output. It was therefore concluded that requirement of 0.03 m3/s per rated kW electric motor output applies to this mine, which is 40 per cent less than the current requirement of 0.05 m3/s per rated kW diesel engine power. The shallow mine located in Mt Isa (hot strata) required a total quantity of 120 m3/s at its primary fan without any refrigeration plant. The higher requirement reflects the additional heat load from the strata. Therefore, the requirement was calculated as 0.046 m3/s per kW electric motor output, which is eight per cent less than the current requirement of 0.05 m3/s per rated kW diesel engine power. The requirement would be less if a refrigeration plant was utilised. However, given that the mine is only 600 m deep, it was assumed that utilising refrigeration plant in this mine is not financially feasible. CONCLUSIONS This study has found an indication that utilising electric vehicles will require less ventilation and therefore will save some primary fan power cost. The reduction of ventilation requirement varies between mines located in cool and hot strata, and between deep and shallow mines. This is because mines located in hot strata have more heat load than that located in cool strata due to the larger heat emitted by the strata. Deep mines have more heat load than shallow mines as heat emitted by the strata increases along with depth. A refrigeration plant must be used in deep mines located in hot strata. Utilising refrigeration plant reduces the quantity requirement; however, it comes with increasing cost of operating and maintaining the plant. 49 M Kerai and A Halim The popular assumption that heat emitted by electric vehicle equals to its power consumption was validated in this study RECOMMENDATION FOR FUTURE WORK The simulation done in this study was only based on two mines. As each mine is unique and has its specific ventilation circuit, vehicle fleet, and geothermal parameters, a similar simulation of other mines might produce different results. It is recommended that similar simulation is done in different mines in order to support the indication found in this study. This study was done with assumption that all electric vehicles are powered by battery since Ventsim Visual is unable to calculate latent heat produced by fuel cell. It is recommended that the software is upgraded to include capability to calculate latent heat produced by fuel cell, and therefore is able to simulate heat emitted by fuel cell powered vehicles. ACKNOWLEDGEMENTS This study required visiting Rio Tinto’s Northparkes mines, which was made possible through the funding from Mining Education Australia (MEA). The authors would like to thank Prof Peter Knight, Executive Director of MEA and Mrs Paulette 50 Schmidt, Finance Officer and Executive Assistant of MEA for arranging the visit to Northparkes mine, Mr Eddy Samosir, Ms Claudia Vejrazka, and Mr Mat Allan from Northparkes mine for their assistance during the visit. REFERENCES De la Vergne, J, 2003. Hard Rock Miner’s Handbook [online]. Available from: <http://www.altomines.com/pdfs/ HardRockMinersHandbook.pdf> [Accessed: 30 March 2012]. Marks, J R, 2012. Airflow specification for metal and non-metal mines, in Proceedings of 14th US/North American Mine Ventilation Symposium (eds: F Calizaya and M Nelson), pp 191-195, (University of Utah: Salt Lake City). Sandvik Mining, 2012. Specification of LH514 and LH514E [online]. Available from: <http://www.mining.sandvik.com> [Accessed: 15 August 2012]. Van den Berg, L, 2012. Personal communication regarding refrigeration plant design limit, Senior Consultant and Manager – Ventilation, Snowden, 25 September. Western Australian (WA) Government, 1995. Mines Safety and Inspection Regulation 1995 [online]. Available from: <http://www.mirmgate.com/docs/compliancegate/wa/ MineSftyAndInspectionRegs1995_05-c0-01.pdf> [Accessed: 30 March 2012]. Mining Education Australia – Research Projects Review 2012 Prediction and Modelling of Blast Vibration and its Effects at Glendell Colliery M K McKenzie1 and D Chalmers2 Vibration management is an integral part of maintaining best practice standards at Glendell Coal’s truck and shovel operation in the Hunter Valley of New South Wales. This mining operation is adjacent to several infrastructure items. A site vibration model was developed, to explore economic, engineering and community impacts. Due to the proximity of public infrastructure to the mine it is unlikely that blasting would be able to continue in the future without modifying its blast designs. A blast vibration model was required to manage the effect of blasting operations on stakeholders. Charge weight scaling laws provide an ideal tool for this application, but it needed to be carefully tailored to local site conditions to achieve best results. By analysing vibration measurements from previous blasts at the mine in conjunction with other data such as geological conditions and blast design features, it was possible to optimise the charge weight scaling laws. These predictions where used to evaluate vibration management for the mine, in terms of blast design and financial consequences. Consequently, a site optimised vibration model was created for the mine site with derived values for site constants that account for various blasting scenarios. The model provides Glendell Colliery with a predictive tool for blast design and was used as the basis for a number of recommendations to improve future blast designs. The model also indicated that the final pit shell may need to be modified with a resultant loss of coal reserves, an important finding as any modification to the final pit shell changes the economics of the operation and will also require modification to the mine operations plan. Introduction Blasting is an integral aspect of almost all modern surface mining operations. However, despite its importance, blasting has the potential to create a large source of dust, noise and, importantly, vibration. If not well designed, blast vibration can be a source of concern for nearby stakeholders in the vicinity of mining operations due to the potential effects on dwellings and infrastructure, as well as causing general annoyance in communities. The Hunter Valley region of New South Wales (NSW) is the most productive thermal coal producing region in Australia. The valley is also home to horse studs and wineries that support a thriving tourism industry. This mixture of land uses causes local mining operations to be held to some of the highest operational standards in the world (Richards and Moore, 1995). This paper provides technical justification for establishing a vibration model to accurately predict vibration levels from blasting at Glendell Colliery, to enable a more thorough consideration of the resulting effects and implications. Glendell Coal is an open cut mining operation in the Hunter Valley of NSW. It is part of the Mt. Owen Complex and is owned and operated by Xstrata Coal NSW (XCN). The deposit is mined using conventional truck and shovel methods, featuring large scale production blasts. The Great Northern Rail Line bounds the operation on the southern side of the final pit shell, presenting special complications. At this stage there is no blasting deed in place between Australian Rail Track Corporation (ARTC) and Glendell Coal. This means that full scale production blasting can only be conducted at distances greater than approximately 500 M from the rail line to ensure compliance with a 25 mm/s vibration limit (Davidson, 2012). In consideration of these concerns, this paper will provide an assessment of the current state of knowledge in the field of blast vibration prediction whilst exploring the potential effects on the Glendell mining operation. Vibration Modelling Prediction of vibration has traditionally been performed using charge weight scaling laws or some derivative of such. These laws make their base predictions on the assumption that the peak particle velocity (PPV) caused by a blast will be the same as that generated if a single blast hole containing a charge equal to that of the maximum instantaneous charge (MIC) was detonated at the centroid of the area (Blair, 2004). Recent times have seen an increase in use of wave form models, which take each individual blast hole and model the sum of the orthogonal wave components. These individual wave forms can then be superimposed after applying the appropriate delays to form a complete model of the blast. 1. SAusIMM, Graduand Mining Engineer, The University of New South Wales, Sydney. 2. Senior Lecturer, The University of New South Wales, Sydney. Email: [email protected] Mining Education Australia – Research Projects Review 2012 51 M k MckenzIe and d chalMerS One drawback of superposition models is that they assume the ideal case in which vibration radiates uniformly in all directions. This is almost certainly not the case as any geological structures present changes in stratum and the screening effect of shot ground will affect the propagation of blast waves. While the limitations of scaling laws are well recognised, they are still widely used throughout the mining industry. The use of the charge weight laws is particularly prevalent amongst site-based personnel as, once established, they are simple to use and the required data is easily sourced. These laws expressed in their typical form (in Equation 1 contain two site-specific values: a site constant (K) and exponent (β). PPV = K e where: D MIC -b o (1) mm/s D: distance from the charge, metres MIC: Max Instantaneous Charge, kg In the preliminary stages of a vibration prediction, the formula can be applied using values typical for the relevant type of mining. However, using these so-called ‘standard values’ is fraught with danger as their upper and lower limits, shown in Table 1 can produce wildly different outcomes when they are varied by a small amount. TABLE 1 Typical values for site laws (constants) in coal mines (source: Richards and Moore (1999)). Site Value K β Upper limit 500 -3.0 Lower limit 10000 -1.4 Typical value in coal mines 1600 -1.6 SIte vIBratIon Model The investigation of vibration behaviour at Glendell used a large amount of data from past blasts. Before this data was used, identifying possible outliers and then sorting data into different categories for assessment was required. Groupings trialled were based on factors such as lithology, monitoring location and blast horizon. For each of the considered data configurations, statistical analysis of key parameters was required to ensure the integrity of the data being used, as well as to maintain the representative nature of the model being created. For quantitative comparison, regression analysis was performed, which was then used to derive values to be used in the charge weight scaling formulae. After deciding on the structure of the vibration models and deriving the constants, the results were plotted. This step was important, not only because visual representations are a useful communication aid, however as it also provided a useful quality assurance mechanism. Using this vibration model, the consequences for Glendell can be assessed to determine the impact from an operational, blast design, community and environmental perspective. Having taken into account the results of the regression analysis and investigated the types of blasting that typically occur at the mine site, the structure of the vibration model was decided. The choice of constant values was based on both accuracy, and operational considerations. The constant values are generally skewed towards the mining horizons that contribute most significantly towards annual waste movements because some horizons account for a much greater portion of waste movements than others, as illustrated in Figure 1 where waste movements are stated in units of millions of bench cubic metres (Mbcm). Given the dependence of scaling formulae on appropriate values of the site constant and site exponent being used, the derivation of these values should be a priority. To derive values for these constants data from past blasts is required. The information gathered should include the maximum charge per delay, coordinates of both the blast and monitoring points and the PPV at these monitoring locations. The PPV can then be plotted versus the scaled distance for each blast and a curve fitted. The decision of what values to use for the constants will vary upon the criteria set out for the prediction. Some scenarios will call for a best-fit case, calculated using a least-squares method or similar. Given the potential consequences often associated with exceedance, Humphreys (2012) suggests a conservative approach will usually be taken, such as 95 per cent less than or in especially sensitive cases 100 per cent less than. An important aspect of the conservative approach is deciding what data to exclude as outliers, as these methods take no account of directional influences on vibration. Thus, it is dangerous to exclude unusually high results, as they can be the most indicative of what conditions will produce high vibration levels. A useful technique to improve the robustness of the site exponent estimate is to place a series of vibration monitors in a concentric line away from a blast. This method of gathering data, suggested by Gilbert (2012), is an excellent way of estimating the attenuation characteristics of the rock mass as it provides multiple data points all while keeping the directional influence component constant. 52 Fig 1 - Waste movements during study period. The rationale being that compromise is a necessary part of such a model, but the negative effects of the compromises made can be mitigated by ensuring it comes at the expense of the horizons that are less important to the overall mine schedule. The values for the constants used in the model varied in the range from 1587 to 3668 for K, and -1.34 to -1.72 for β. A lithology-based approach was found to give the best result in most cases. The following formulae are recommended for use in the described situations at Glendell. power lines For the monitoring sites ‘Power line 1’ and ‘Power line 2’: PPV = 3668 e D MIC -1.72 o mm/s (2) MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 predIctIon and ModellIng of BlaSt vIBratIon and ItS effectS at glendell collIery Sandstones and conglomerates D PPV = 1652 e MIC -1.34 o mm/s (3) operations claystone, siltstone and mudstone PPV = 1587 e D MIC -1.44 o mm/s (4) conSequenceS The vibration model developed for Glendell was used to consider if changes to current blast practices are likely to be needed. In cases where changes are anticipated to be required, this model will be used to explore the flow-on effects for the main stakeholders involved. Due to the nature of blasting, measures to minimise vibration levels seldom improve blasting activities from the perspective of performance, efficiency or cost. Whilst refinement of the initiation timing can reduce vibration levels considerably, these improvements soon plateau, making further design changes necessary. The likely effect on production activities of the proposed changes was also considered in order to gauge the implications of these changes by reducing productivity and increasing costs. Blast design When using charge weight scaling laws to predict and subsequently manage vibration levels, the only variable a blast engineer can directly influence is what the MIC for the designed blast will be. As MIC is also a main variable in the commonly used charge weight scaling laws, it can be used as the main driver for vibration constrained blast design. By rearranging the charge weight scaling formula to the form below and substituting the site constants determined earlier, the highest MIC able to be used while staying within vibration limits can be determined. MIC = e K.D 2 1 o PPV b kg Currently, Glendell uses 229 mm holes almost exclusively which, when using ANFO, will not become an issue until blasting comes within 350 m of infrastructure as blast holes are seldom much more than 20 m. (4) When the limits on charge weights are calculated and compared to current practices it becomes apparent that staying under these limits will require changes to blast designs in the future. In Figure 2, the maximum column length (charge weight) for a range of blasthole diameters and a 25 mm/s vibration limit is shown in relation to distance from infrastructure. The calculations in this figure were made for the case of 100 per cent ANFO. Fig 2 - Maximum column length for three blasthole diameters. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 All of the proposed blast changes to meet vibration limits are likely to impact productivity at Glendell. In terms of excavation units, there is often a reduction in dig rates associated with decked blasts. This reduction is attributable to the uneven distribution of explosive energy in the blast area, as well as the lessened muck-pile heave and the associated ‘looseness’. As the size of blasts is reduced, more blasts will be required to maintain the same level of production. When blasts are fired, equipment often has to be relocated so that it is outside the blast exclusion zone. As tramming speeds are low, these relocations can be time consuming, causing a loss of productive time and an increase in downtime due to the extra wear on undercarriage components. Additionally, smaller blast sizes will result in more machine relocations and the described loss to production. Usually, dozer and grader work is required to make an area suitable for drilling equipment to access and work on, including the erection of windrows. With smaller blasts, these tasks must be completed more frequently, increasing demand on support equipment. This draws them away from other tasks that more directly add value to the business, such as haul road maintenance and coal preparation. As blast sizes are reduced to deal with vibration, so too will bench heights. A reduction in bench heights has scheduling consequences, requiring a larger operational footprint with more active benches. Previous long term planning is unlikely to have taken these things into account and may need to be modified to reflect the changed operating conditions. economic Implementing the blast designs required to manage vibration levels at Glendell in the future, will come at an economic cost to the operation. These costs will be direct, such as an increase in initiation products, and indirect, such as lost production, increased down time due to maintenance and greater compliance costs. As blast hole diameters are reduced, less volume is blasted per metre of drilling, costing more in terms of consumables, equipment and fixed costs such as labour. In terms of blasting activities, the use of decked blasts and smaller diameter holes will increase shot-firing labour requirements and decrease the productivity of the existing shot crew. Additionally, when decked blasts are used, there is an increase in the amount of initiation equipment required, as each deck is primed separately. This cost is exacerbated when electronic detonators are used, which at $55.03 each, plus $10.69 for a booster, increases the cost per hole by $65.72. Further, to add a second deck, the initiating equipment expense increased to $20 767 for an average blast before any other costs are factored in. In bulk mining operations such as Glendell, equipment productivity is an important issue. The method by which ground is blasted directly influences the excavation rate achieved and is sensitive to the degree of fragmentation and muck-pile heave. As Glendell operates three Hitachi EX5500 that are responsible for the majority of the waste moved on site, the effect of reduced production rates was examined for these machines. The consequence of a reduction in the productivity of EX5500 is significant, as illustrated in Figure 3. 53 M k MckenzIe and d chalMerS capable of being used in wet ground. However, even with this design, the blast size needs to be restricted to approximately 40 000 bcm, otherwise reinforcement becomes an issue. other The site vibration model highlights the narrowing margins for error that exist as mining approaches the infrastructure items, presenting a challenge to maintaining a healthy social licence. As such, ongoing attention should be directed towards cultivating a positive relationship between Glendell and its stakeholders. Fig 3 - Hitachi EX5500 productivity losses. These losses were calculated by dividing the foregone waste movement by the product stripping ratio to determine the lost coal production and then applying the current spot price for thermal coal ex Newcastle. From the above, it is obvious that a small drop in the productivity of an EX5500 will cost the operation large sums of money, especially if the other two excavators experience a drop in production rates. While a drop in annual coal production due to lower machine productivity is undesirable, the coal is not necessarily sterilised, rather its value is simply diminished by being mined at a later date. Conversely, if blasting is unable to occur, the underlying coal is effectively lost. Due to the proximity of the Great Northern Rail line to the ultimate pit limit, it is unlikely that blasting could be used without exceeding the current 25 mm/s limit at the pits margins. Based on an average cumulative coal thickness of 18.5 m and pit width of 3.5 km, each metre that the final pit wall is moved away from the rail line will sterilise 90 kt of coal, constituting a $5.9 million loss at current spot prices. Blasts with an average MIC cannot be used once within 400 m to 500 m of the rail line. Figure 4 shows the standoff distance required to stay under the 25 mm/s threshold, calculated using the constant values that were determined using regression analysis. Also seen in Figure 4 is the lowest standoff distance that can be achieved using viable blasting practices for Glendell, though other operations such as Liddell have managed to mine very close to the line. The blast design used for the limiting case consists of a 10 m bench loaded in two decks, separated by a gravel deck, with a bulk product Fig 4 - Limiting distances for 25 mm/s. 54 Challenges to stakeholder relations are unlikely to be confined to external parties such as ARTC and Camberwell Village, the rationale is stakeholders such as crew members and the CFMEU have a vested interest in maintaining employment levels at the operation. These factors may lead to conflict if the production rate is altered or mine life is reduced due to the challenges presented by managing vibration. The possibility of a loss of reserves also raises the issue of resource utilisation, and reduced revenue to the state government due to the foregone royalties that would have been payable on the coal sterilised. concluSIonS The research project conducted highlighted that rule of thumb values are not appropriate for vibration prediction at Glendell Coal’s mining operation. Whilst rule of thumb values can be used currently without ill-effect, future blasts will be within the margin for error in managing vibration as the site matures. While preparing and analysing the data, as well as in a consideration of the results of the project, it became apparent that the current system of record keeping for blasting activities at Glendell is comprehensive as it captures useful information in a manner that is conducive to subsequent study and examination. The high standard of information retained about blasting practices at the mine was a significant aid to the completion of the research project and lends confidence to derived results. By analysing data from past blasts using methods such as regression analysis, it was possible to exclude abnormal blasts, leaving only those that are representative of typical blasts at Glendell. This data was then grouped in several ways to determine what the most effective and efficient way of predicting vibration at this site would be. Based on the analysis performed, grouping data based on the lithology of the area being blasted produces the most satisfactory result in the sense of statistically fit, broad classifications where used. It is likely that this method could be improved by more rigorously exploring the potential correlations between rock-mass attributes and vibration behaviour. Having developed a vibration model, it was applied to determine what changes to drill and blast practices are required and when they will need to be implemented. Based on order of magnitude style calculations, it was immediately apparent that some degree of change is required. With the aid of a vibration model tailored to the site conditions, accurate predictions of the adaptations required could be made. Key findings included the stand-off distances at which significant changes are likely to be required, but perhaps the most important finding is that the current blasting practices are not suitable at distances closer than 500 m to most infrastructure items. This is important considering the percentage of the mining area within this zone. The changes are unlikely to MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Prediction and Modelling of Blast Vibration and its Effects at Glendell Colliery be significant until blasting reaches within 250 - 350 m of infrastructure, at which point it is unlikely that 229 mm holes can continue to be used. Using highly constrained blasting practices, including 10 m benches, decking, small diameter blast holes and reduced overall blast size, blasting can come within 150 - 200 m of infrastructure for most blast horizons. As changes are made to blast designs and practice, mining costs will rise. The increases will come from drill and blast, where the cost of drilling, blasting products and labour will all increase, and from the opportunity cost associated with the probable reduction in equipment productivities, brought about by smaller blasts and having to dig through decked blasts. With a 150 - 200 m limit on the viability of blasting, it is quite likely that areas of the current final pit shell will fall inside this zone. The consequence of this is that if vibration limits remain unchanged, it is quite possible that reserves will be sterilised directly due to an inability to blast, or indirectly as the result of diminished returns due to high costs. Acknowledgements The author acknowledges the support of Glendell Colliery, specifically Mr Shaun Leary and Mr Anthony Davidson for supporting the project and providing the data needed to perform the investigation. References Blair, D P, 2004. Analysis and modelling of airblast and ground Vibration, in Proceedings EXPLO, pp 1-10 (The Australasian Institute of Mining and Metallurgy: Melbourne). Davidson, A, 2012. Personal communication. Drill and Blast Engineer, Glendell Coal, 14 February. Gilbert, S, 2012. Personal communication, Drill and Blast Engineer, Ravensworth Surface Operations. 18 January. Humphreys, M, 2012. Personal communication, Consulting Mining Engineer, MiPlan, 9 February. Richards, A B and Moore, A J, 1995. Blast vibration control by wavefront reinforcement techniques, in Proceedings EXPLO ’95, pp 323-328 (The Australasian Institute of Mining and Metallurgy: Melbourne) Richards, A B and Moore, A J, 1999. Predictive assessment of surface blast vibration, in Proceedings EXPLO ‘99, pp 91-97 (The Australasian Institute of Mining and Metallurgy: Melbourne). Mining Education Australia – Research Projects Review 2012 55 56 Mining Education Australia – Research Projects Review 2012 Modelling Depletion of Queensland Coal Resources using Historical Production Trends A Meikle1 and W Seib2 ABSTRACT Coal is a finite energy resource. Therefore, by definition, that resource will inevitably be depleted. The reporting of coal resources by the Queensland Government includes a projected resource life value based on current production levels. This paper reviews the accuracy of government reporting of resource life and explores more accurate means of modelling depletion profiles. The modelling approaches considered historical production trends in order to forecast future coal production. Future coal production would then dictate the depletion of the Queensland coal Ultimate Recoverable Resource (URR), which was determined to consist of base and growth resource groups. Using this process, low and high URR depletion profiles are provided in the paper. In addition, using the same modelling process and past URR estimates, resource life trends were investigated. However, as stated, modelling in this paper considers only historical production trends. Therefore, a number of factors were determined to exist outside the scope of this paper with the potential to impact on the depletion profile that will eventuate. Subsequently, they were assessed in order to gauge the potential degree of influence and thus, the accuracy of the modelling forecasts. Introduction Due to the finite nature of coal as an energy resource, it will inevitably be depleted. Therefore, in the interest of the Queensland coal industry the state and federal governments provide annual coal resource life estimates. Such publications include Australia’s Indicated Mineral Resources (Geoscience Australia, 2012) and Queensland Exploration Scorecard (Queensland Exploration Council, 2011). However, the method in which the estimates are calculated in publications such as these can be inaccurate and misleading. Therefore, this paper reviews past approaches to depletion modelling of coal and other finite energy resources, such as those by Hubbert (1956) and Mohr et al (2011), with the aim of producing a new method that will offer increased accuracy. All past studies indicate that depletion will follow a bell-curve shape. Therefore, the desired model will mirror this trend, using annual coal production and coal resource data, inclusive of all coal types. Using the proposed model, forecasts will be made regarding peak coal and resource depletion. However, in order for the forecasts to remain accurate a number of factors that influence coal resource depletion will have to follow historical rates of change. If this occurs, the accuracy of historical production data will hold, as will depletion forecasts made using the proposed model. Queensland Coal Resources Queensland has an extensive coal resource base (The Queensland Coal Board), accounting for approximately 60 per cent of Australia’s coal resources, which is the fifth largest in the world (Geoscience Australia, 2012). A wide variety of coal types are present in Queensland and they are exported to a variety of international markets, which include highgrade metallurgical coking coal, blending coals, pulverised coal injection coals, high and lower volatile thermal coals for power generation and high and low volatile thermal coals for industrial market (DME, 2007). However, all of Queensland’s metallurgical coal is found in the Bowen Basin, whereas, thermal coals are found throughout the state. The distribution of Queensland’s coal-bearing basins is shown in Figure 1, and the major basins include the: •• •• •• •• •• Bowen Surat Galilee Clarence-Moreton Eromanga/Cooper. In addition to the large resource size of both Queensland and Australia, the region is positioned well as a coal exporter and to handle increasing global demand for coal because of (Productivity Commission, 1998): •• •• •• •• •• •• abundant supplies of good quality coal readily accessible coal deposits coal deposits relatively close to shipping ports established rail and port facilities close proximity to major Asian coal markets a proven reputation as a reliable supplier. 1. Mining Engineer Undergraduate, The University of Queensland, 25 Oakmont Street, Rothwell Qld 4022. Email: [email protected] 2. FAusIMM(CP), Casual Lecturer, School of Mechanical and Mining, The University of Queensland, 14 Crestview Street, Kenmore Qld 4069. Email: [email protected] Mining Education Australia – Research Projects Review 2012 57 a MeIkle and w SeIB logistic curves The concept of modelling depletion rates and future production of finite resources was first analysed by Hubbert (1956). He theorised that production levels of a given finite resource begin at zero, prior to production commencing, rise to one or more maxima before returning to zero, when resource exhaustion occurs. In following with this assumption, a bellshaped curve was suggested as an accurate model for the production curve. Hubbert (1959) later adopted a generalised logistic curve as the representation of production trends (Mohr et al, 2011). The logistic curve has since been widely used in subsequent studies when modelling fossil fuel depletion such as coal (Höök et al, 2010; Patzek and Croft, 2010). The logistic equation is defined as (Mohr et al, 2011): QL ^ t h = QT c 1 1 m e`-rL `t - t pjj + 1 where: QL(t) is the cumulative production for the logistic curve QT is the Ultimately Recoverable Resources (URR) tp is the peak year of production rL is the rate constant. ModellIng approacheS A recent study by Mohr et al (2011), compared four modelling approaches, including logistics curves, to project Australian coal production. The study considered a standard or conservative URR estimate and a high or optimistic estimate of URR when conducting modelling. The results indicated depletion of Queensland coal resources occurring in approximately 2135 and 2150, for the standard and high URR scenarios, respectively. government resource life gompertz curves Fig 1 - The distribution of Queensland’s coal basins (Mutton, 2003). The reporting of national or state coal resource endowment by the government often includes a resource life estimate. The resource life is calculated by dividing the reported resource endowment by the current annual production level at the time of the report. While this figure provides an interesting point of reference, it is highly inaccurate and often misleading (Bartlett, 2006). The main reason for the inaccuracy of the figures is the assumption of constant production and zero resource growth. In reality this would mean constant levels of production until the year of resource depletion, at which time production would abruptly cease. This represents a highly unlikely scenario for the coal industry or the depletion of any resource in general. The government method of calculating resource life is shown in Figure 2. The Gompertz curve has also been used in past studies to model fossil fuel depletion (Fitzpatrick, Hitchon and McGregor, 1973; Moore, 1966). The general shape of the Gompertz curve is similar to the logistic curve, in that they are both bell-shaped. However, the Gompertz curve is not constrained to being symmetrical. The Gompertz equation is defined as (Mohr et al, 2011): r t - t pj c ` G` m QG ^ t h = QT e - e (2) where: QG(t) is the cumulative production for the Gompertz Curve QT is the URR; tp is the peak year of production rG is the rate constant The depletion rate of the remaining recoverable resources is defined as (Mohr et al, 2011): ddG ^ t h = P^ t h QT - Q^ t h (3) where: Fig 2 - Government resource life representation. 58 dδG(t) is the depletion rate for the Gompertz curve at time t QT is the URR; Q(t) is the URR at time t P(t) is the annual production at time t In a study by Mohr et al (2011), Gompertz curves were used to project Australian coal production. The study considered a standard or conservative URR estimate and a high or optimistic estimate of URR when conducting modelling. The MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Modelling Depletion of Queensland Coal Resources using Historical Production Trends results indicated depletion of Queensland coal resources beyond 2200 in both scenarios, standard and high URR. Supply-demand models The supply-demand model investigated was designed by Mohr and Evans (2009). The model consisted of a defined number of idealised mines, each consisting of a mine life, L, and production rate, P. Depending on the production requirements at the time, mines would either begin or cease production for the given time period. The time of first production was defined as T0, the production rate constant, r, dictated the rate at which new mines began production. As with previous modelling methods, a standard and high URR estimate was used, and it influenced the number of mines allocated to a given region. Using this approach, two constraints were used: 1. the static case – no interaction between supply and demand would occur 2. the dynamic case – both supply and demand influenced each other. Modelling of the supply-demand scenario hypothetically began in 1880, with an initial demand of 1 Mt that would increase at 4.8 per cent per annum. The production rate constants for each state and coal type were obtained from historical production data (Mohr et al, 2011). Depletion of Queensland coal resources was found to occur approximately in 2125, for both URR scenarios using the dynamic scenario. This would represent a resource life of approximately 113 years. Peak coal Peak coal, or the point where coal production reaches a maximum, is key when predicting resource depletion. Bardi and Lavacchi (2009) theorised that in the initial stages of extraction of an abundant and cheap resource, economic growth and increasing investments into extraction/ exploration will result. As the more easily accessed and cheaply extracted resources become depleted, extraction costs begin to climb due to increased overburden depths and reduced resource quality. In time, investments are outweighed by rising extraction costs; growth slows and eventual production enters a depletion-driven decline. The three modelling approaches used in Mohr et al (2011) all indicate similar years when peak coal will be reached: 2048 - 2052, for the lower URR and 2056 - 2057 for the higher URR. Excluding the Gompertz curve model, the production rate experienced at peak coal is also relatively similar 453 489 Mt/a for the lower URR and 564 - 644 Mt/a. Modelling Conclusions From the three modelling approaches investigated, it is clear that the proposed model should produce a depletion profile that is a bell-shaped curve. Furthermore, the results of the study conducted by Mohr et al (2011) can be used as a rough guide to the desired outcomes of the proposed modelling scenario. Therefore, peak coal should occur approximately in the period 2048 - 2057, with a magnitude of 564 - 644 Mt/a. Additionally, resource depletion should approximately occur between 2125 - 2150, depending on the determination of URR used in the proposed scenario. Modelling Methodology Resource and production data In order to produce an accurate model of Queensland coal resource deletion, accurate historical data must be sourced. Mining Education Australia – Research Projects Review 2012 In addition, in order to achieve maximum accuracy, where possible, a single data source was used to collect data, so that no disparity in data occurred due to differing reporting standards. Raw coal production data for Queensland was sourced from the Australian Bureau of Agricultural and Resource Economics and Sciences (ABARES) for 1950 - 2010. The 60-year period was deemed sufficient to indicate any production trends for coal production in Queensland and was, therefore, suitable for the purpose of the project. For the purposes of the proposed model, the coal resources were separated into two categories: base and growth resources. The base resource group consisted of Measured and Indicated JORC Resources. Whereas, the growth category of resources comprised Inferred JORC Resources and coal exploration targets. However, due to the reduction in resource size that occurs when a resource is upgraded to a JORC category of higher geologic certainty, it is assumed that only 20 - 40 per cent of the resource will be maintained through the upgrading process. The 20 to 40 per cent range was decided upon following review of resource statements for companies such as New Hope Coal and BHP Billiton. Whilst the resource statements did not indicate a definitive figure, the range was considered a best judgment of the trends present. Queensland coal resources for the base resource group could not be obtained from a single data source. Therefore, the Queensland Coal Board Industry Review was used for the 1980 - 2003 period. Whereas the Geoscience Australia publication, Australia’s Identified Mineral Resources, was used for the later 2004 - 2011 period. Whilst the two data sources used different reporting standards, they were found to correlate strongly with one another, and were, therefore, used interchangeably. Resources in the growth resource category were only collected for 2011, and were sourced from the most recent resource statements of approximately 40 coal mining and exploration companies. Symmetric Gaussian modelling From the review of past depletion modelling techniques it was determined that coal resource depletion would follow a general bell-curve shape. Gaussian curves were deemed to produce a suitable depletion profile. Furthermore, Gaussian curves could be simply manipulated, which would aid the modelling process. A modelling approach of this nature will aim to accurately project annual production until which point that the cumulative production is equal to the proposed URR. At this point in time, the Queensland coal resource will be declared as depleted. The depletion profile will, therefore, be directly related to annual production figures, thus making the assumption that coal resources will be mined in full. The symmetric form of the Gaussian function is its most basic form; it is symmetrical about its central peak and is defined by the equation (Guo, 2011): y = a.e - ^ x - bh 2c 2 2 (4) where the following variables define: a: curve amplitude b: location of the curve peak on the x axis c: curve width. For the purpose of this report and the modelling process, the height of the curve will indicate the magnitude of peak 59 a MeIkle and w SeIB coal, ie the variable a. Variable b will dictate when peak coal occurs and variable c will relate to the annual growth in coal production. Using the historical coal production data for Queensland, the Gaussian curve can be manipulated in order to best match the historical trend. Due to the symmetric nature of the curve, the growth profile of the curve, dictated by historical data, will be mirrored to indicate the projected decline in production post-peak coal. The area under the projected curve, or sum of annual coal production, will correspond to the URR for Queensland coal. Therefore, changes in peak coal are expected to be the most evident result for different URR scenarios. The different URR scenarios will be produced by modelling each combination of the full base resource with either 20 per cent or 40 per cent of each of the Inferred JORC Resources and exploration targets, resulting in seven hypothetical scenarios. From the seven modelled scenarios, the scenario using 20 per cent of both Inferred JORC Resources and exploration targets will represent the theorised low URR depletion scenario. Whereas the scenario using 40 per cent of both growth resource categories will represent the theorised high URR depletion scenario. The low and high URR scenarios will aim to encompass the true Queensland coal depletion scenario within their bounds. asymmetric gaussian modelling Whilst symmetric modelling techniques have been used in the past to make depletion projections for finite resources, a distinct problem exists within the modelling approach. The problem stems from the ability of the modelling technique to adapt to a larger and expanding URR scenario. A larger URR simply results in a peak coal of greater magnitude using the symmetric modelling approach and minimal increases in resource life; this result is expected to be a poor representation of reality. Therefore, an asymmetric modelling approach was also tested. An asymmetric Gaussian curve will rectify this problem. The asymmetric Gaussian curve is defined by the equation (Kato, Omachi and Aso, 2011): 1 c 2 x-b 1 1 z y = y0 + a ec 2 ` d j - d mc 2 + 2 erf c 2 mm d (5) historical resource life comparison This historical resource life comparison component of this project aimed to review resource life trends present in government resource life calculations. Those trends were then compared with those made by modelling past Queensland coal resource inventories. This was done by using the asymmetric Gaussian model to model the base resources, ie Measured and Indicated JORC Resources, present in 1980, 1995 and 2010. Ideally, growth resources would have been accounted for in each of these years and used when modelling. However, Inferred JORC Resources were rarely reported quantitatively prior to 2003 and were instead only given a qualitative measure. In addition, whilst exploration targets were reported for all time periods, locating mining and exploration companies operating at that time and their current exploration targets was deemed to be too difficult for the time frame allowed for this project. Therefore, using only base resources was deemed to be the most viable option as it would still indicate any trends that exist in resource life projections using the asymmetric Gaussian model. Furthermore, 1980 and 1995 have been chosen as the periods to be tested for the comparison process. This will indicate any trends on a 15-year basis. In addition to using the asymmetric Gaussian model for comparison, two different techniques were used: 1. Technique 1: use the full range of historical production data (1950 - 2010) combined with the individual URR for each period of time to be tested. 2. Technique 2: use only the historical production data to date for each period of time to be tested, combined with the individual URR for each period of time to be tested. The individual results for each technique were then compared to see if any differences eventuated, and the more accurate technique was selected. depletIon ModellIng reSultS where: z = x-b - c c d lower range will be defined by the peak coal projection made by the symmetric Gaussian curve for the base growth scenario. Three peak coal values will then be selected for testing within the indicated range. Three URR scenarios will also be tested, corresponding to the base, low and high URR scenarios tested using the symmetric Gaussian modelling approach. (6) and the following variables define: a: curve amplitude b: location of the curve peak on the x-axis c: curve width resource and production data Using the sources identified earlier in this paper, the annual coal production and resource data shown in Figure 3 was produced. From Figure 3, annual coal production can be seen to steadily increase from 1950 to 2010, as expected. Resource d: modification factor or skewness. The key difference between the two Gaussian curves is the addition of the modification factor. The modification factor decreases the rate of reduction in annual production, therefore, extending the resource life. Therefore, when using the asymmetric Gaussian curve, the curve is fit to historical production data in the same manner as before, except a larger URR can be accounted for by increasing the modification factor, thus extending the projected resource life. However, for the asymmetric Gaussian curve the magnitude of peak coal must be defined. This value will be determined by reviewing the symmetric Gaussian curve data to discern an expected time frame for the occurrence of peak coal. A simple polynomial fit of historical production data will then be used to project an upper range for the magnitude of peak coal. A 60 Fig 3 - Queensland coal resources and production. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 ModellIng depletIon of queenSland coal reSourceS uSIng hIStorIcal productIon trendS data appears to indicate that peak levels have been reached. However, this is most likely in response to changes in the JORC Code, resulting in the downgrade of resources due to the requirement of increased geological certainty. Therefore, the increase in resource size from 2008 onwards can be expected to continue in coming years. Futhermore, Table 1 shows the Resource and production data for 2011, which was used for the various modelling approaches and scenarios tested. TABLE 1 Coal Resource and production data for 2011. Cumulative production (Mt) 4742 Resource (Measured and Indicated) (Mt) 37 156 Resource Inferred (Mt) 60 187 Exploration targets (Mt) 62 417 Fig 4 - Symmetric modelling of base, low and high ultimate recoverable resource scenarios. TABLE 3 Symmetric modelling projected depletion bounds. Symmetric modelling The symmetric Gaussian modelling approach was successful in modelling each of the seven URR scenarios proposed for testing. Table 2 shows the individual modelling results that were produced. From Table 2 the impact of increasing URR size is evident. By almost doubling the URR from scenario one to seven the projected peak coal level increased from 558 Mt/a to 892 Mt/a. Whereas the resource life only increases by 32 years from scenario one to seven. Therefore, the initial assumption that the symmetric model would have difficulty accurately reflecting an increase in URR, was proved correct. Figure 4 graphically depicts the depletion profiles of scenarios 1, 3 and 7, corresponding to the base, low and high URR scenarios. For all scenarios a high degree of accuracy was present in representing the historical production data, reflected by an average R2 value of 0.99 for all tested scenarios. Ultimate recoverable resource (Mt) 56 430 - 75 966 Resource life (y) 136 - 153 Peak coal (Mt/y) 709 - 892 Peak coal (y) 2054 - 2063 as the inaccuracy of the symmetric model would be more prevalent for the increased URR used in each case. Initially, literature was consulted, and the Department of Infrastructure and Planning (2010) indicated that annual coal production in Queensland would double by 2030. This would mean annual production levels of roughly 500 Mt for the state by this time. Modelling conducted using the symmetric Gaussian approach also indicated that peak coal would occur Therefore, from the symmetric modelling results, the bounds of the low and high URR scenarios indicate that the true depletion scenario for Queensland will reside between the bounds shown in Table 3. These bounds are also represented in Figure 4, between the low and high URR depletion profiles. asymmetric modelling Prior to conducting any asymmetric Gaussian modelling, suitable peak coal levels had to be determined. In the case of the base URR scenario, peak coal levels of 500, 550 and 600 Mt were selected due to their proximity to the initial peak coal prediction made by the symmetric. However, for the low and high URR scenarios, a different approach had to be taken Fig 5 - Peak coal projection for asymmetric low and high ultimate recoverable resource scenarios. TABLE 2 Symmetric modelling results. inferred resources (%) Explore target (%) Ultimate recoverable resource (Mt) Res life (y) Peak coal (Mt) Peak coal (y) Case 1 0 0 41 902 121 558 2047 Case 2 20 0 53 925 131 700 2052 Case 3 20 20 56 430 136 709 2054 Case 4 20 40 58 930 139 732 2056 Case 5 40 0 65 969 145 798 2059 Case 6 40 20 70 960 149 846 2061 Case 7 40 40 75 966 153 892 2063 MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 61 a MeIkle and w SeIB approximately between 2050 and 2060. Using this information historical data was then plotted, as shown in Figure 5, with high and low level projections. The high-level production projection consisted of a constant growth scenario and the low scenario was produced using the same bell-curve modelling approach used to produce the symmetric modelling scenarios. The peak coal range was determined to exist within the bounds of the low and high production projections. Therefore, a conservative range of 650 - 750 Mt was selected to be tested using the asymmetric Gaussian modelling approach. This would mean calculating projected scenarios for peak coal levels of 650, 700 and 750 Mt of annual coal production. Using the projected peak coal estimates for each of the three URR scenarios, projections were made. Once again, high R2 values were produced. However, using the asymmetric modelling approach, the resource life of each scenario was increased, compared to the symmetric approach, especially in the case of lower projected peak coal levels. Therefore, as expected, lowering the peak coal level for a given scenario increases the projected resource life. The modelling results at this stage are shown in Table 4. Increased resource life for similar URR scenarios with lower peak coal levels is a trend the symmetric coal model was unable to produce. Also, the asymmetric model, as with the symmetric approach, produces a consistent estimate of peak coal. This result is supported by the qualitative reasoning that annual coal production will not be dictated by the size of the URR, but rather factors such as coal price and global supply/ demand. In order to produce a final upper and lower projection of the expected depletion profile for Queensland coal resources, an appropriate peak coal level had to be selected from within the tested 650 - 750 Mt range. Furthermore, to choose a level for peak coal, the symmetric modelling results were reviewed. Due to the method in which the symmetric model is fit to a URR scenario by adjusting the magnitude of peak coal, the suitable peak coal level should reside below that calculated by the symmetric model for a similar URR. As a result, the low URR scenario was used to make the comparison. The symmetric model indicated a projected peak coal magnitude of 709 Mt of coal; therefore, the 650 Mt peak coal level should be used for the asymmetric model as the scenario for comparison. Using the 650 Mt peak coal level for both the low and high URR scenarios, the projections in Table 5, were produced. From Figure 6, the low and high URR scenarios indicate the theorised bounds for which the actual depletion profile of Queensland coal resources will occur. Therefore, a resource life of between 186 and 314 years is projected with peak coal occurring in 2050 - 2053, with a magnitude of 650 Mt. The TABLE 5 Asymmetric modelling low and high ultimate recoverable resource scenarios, with specified peak coal level. Unit Low ultimate recoverable resource High ultimate recoverable resource Resource Mt 56 447 75 975 Resource depletion y 2198 2326 Resource life y 186 314 Peak coal Mt 650 650 Peak coal y 2050 2053 Fig 6 - Asymmetric modelling low and high ultimate recoverable resource scenario. proximity of the true scenario to either of the bounds should be related to the actual URR of Queensland coal that will be dictated by the proportion of growth resources, ie Inferred JORC Resources and exploration targets, which are upgraded. Modelling technique comparison For the purpose of this project, two modelling techniques were used to forecast the depletion of Queensland coal resources: symmetric and asymmetric Gaussian curves. As a point of comparison, the low and high URR scenarios were used for each of the modelling techniques. The projected data for the two comparison scenarios is shown in Figure 7. The most substantial difference in the way the two modelling methods are calculated is the way a larger URR is accounted for by each model. The symmetric modelling technique increases the peak coal level in order to account for the larger URR; whereas the peak coal level is fixed for the asymmetric modelling approach and instead a modification factor extends the resource life by decreasing the rate of production decline post peak coal. This trend is evident in Figure 7. In fact, a 35 per cent increase in URR from the low to high scenario results in a 69 per cent increase in resource life for the asymmetric model, whereas only a 13 per cent increase TABLE 4 Asymmetric modelling low and high ultimate recoverable resource scenarios. Ultimate recoverable resource Low ultimate recoverable resource High ultimate recoverable resource 56 447 75 975 Mt 0.99 0.99 0.99 0.99 0.99 0.99 Peak coal Mt 650 700 750 650 700 750 Peak coal y 2050 2051 2052 2053 2054 2057 Resource depletion y 2198 2170 2143 2326 2291 2257 Resource life y 186 158 131 314 279 245 R 2 62 MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 ModellIng depletIon of queenSland coal reSourceS uSIng hIStorIcal productIon trendS at about 40 - 55 for the past four decades. Continuing the trend of 40 - 55 operating coal mines in Queensland, in order to reach the projected 650 Mt of annual production at peak coal for the asymmetric scenario, annual production of 11.8 - 16.3 Mt/ mine is projected. This projection is supported by current coal projects, with expected annual coal production in excess of 30 Mt/a such as Carmichael, Alpha, Galilee and Kevin’s Corner (Department of Natural Resources and Mines, 2012). historical resource life comparison Using the annual production and coal resource data collected as part of this paper, resource life estimates can be made in the same method as shown in government reports. These estimates are shown graphically in Figure 9. Fig 7 - Comparison of symmetric and asymmetric modelling results. occurs for the symmetric modelling technique. However, the symmetric model figure is affected by the increase in peak coal that occurs from the low to high URR scenario. Due to the greater ability of the asymmetric model to adapt to a larger URR by extending the resource life of the scenario, it is considered the more accurate of the two models. Therefore, all final projection figures will be based on the depletion bounds produced by the low and high URR scenarios using the asymmetric Gaussian model. This translates to a resource life of 186 - 314 years depending on the URR that eventuates for Queensland coal. Peak coal is projected to occur between 2050 and 2053, with a magnitude of 650 Mt. projected annual production per mine As mentioned in the previous section, modelling indicates a resource life of 186 - 314 years, with a peak coal of 650 Mt occurring in 2050 - 2053. However, in order to reach the projected peak coal level for 2050 - 2053, the current average annual production levels per operating coal mine will have to increase dramatically. Geoscience Australia (2012) indicates that 57 coal mines were operating in Queensland in order to produce 257.44 Mt of coal for that year (Australian Bureau of Agricultural and Resource Economics, 2010). This equates to an average annual production per mine of approximately 4.5 Mt/mine. From data collected from Geoscience Australia (2004 - 2011) and the Queensland Coal Board (1950 - 2003), trends in the number of operating coal mines and annual production per mine has been shown graphically in Figure 8. Fig 9 - Government resource life projections. The resource life trend indicated from the government resource life figures indicates that annual production is increasing at a greater rate than resource expansion and, therefore, resource life has steadily declined. However, in recent years, the decline has plateaued and resource life for Queensland coal resources has remained between 130 and 160 years. This trend can be linked to record coal prices for the 2008 - 2010 period, resulting in increased exploration expenditure. Using the first comparison technique, the trends shown in Figure 10 indicate that resource life and both the magnitude and expected date of peak coal has increased from the 1980 to 2010 scenario. This is the opposite of the trend indicated by the government resource life calculation method and further emphasises its misleading nature. From Figure 8, a clear trend of increasing annual production per mine can be observed. This trend is expected to continue, at least until the time of projected peak coal. The number of operating coal mines in Queensland appears to have plateaued Fig 10 - Historical resource life comparison - Technique 1. Fig 8 - Annual production per mine site. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Using the second comparison technique, the indicated trends shown in Figure 11 are quite different, in comparison to the first. The resource life for the three periods of comparison is relatively constant (146 - 151 years), with only five years variation. Although peak coal for each time period varies in a similar way to the previous technique, the reason for the relatively constant resource life is production trends indicated at each period of time. When the production data is stacked 63 a MeIkle and w SeIB sufficient demand will exist and that the global coal market does not become over-supplied. one consistent trend is shown, which is represented by the 1950 - 2010 production data. However, when the production data time frame is shortened for the other periods, the projected rate of increase in annual production decreases. This effectively lengthens the resource life for the 1980 and 1995 scenarios to mirror that of the 2010 scenario, by decreasing the rate in which the URR is depleted. Of the two historical resource life comparison methods, the second method is best representative of depletion projections that would have been made at each period, ie 1980 or 1995. This conclusion stems from the fact that production data for the full 1950 - 2010 period would not have been known in either 1980 or 1995. ModellIng lIMItatIonS Introduction With any modelling process, there will invariably be limitations involved. In the case of the modelling approach taken in this project, the limitations stem from assumptions made, and factors of influence not included, within the project scope. Two main assumptions were made in order to produce the modelling results of this project. The assumptions were: 1. historical production will indicate future production trends 2. Queensland coal resources and annual production data are directly related, ie coal resources are mined in full. Whilst past studies indicate that historical data can be a good indication of future trends when modelling, future changes in the coal industry have the ability to dramatically alter the true depletion profile of Queensland coal resources. Furthermore, the scope of this project considers solely historical production data when projecting depletion profiles for Queensland coal resources. Therefore, all other factors that influence coal production and coal resource size have the ability to reduce the accuracy of the projections made. Some of these factors include: • • • • • coal supply coal demand coal price government policy coal-related infrastructure. coal supply and demand For the purpose of this project, only coal produced in Queensland was included within the scope. However, the greater coal supply market includes other coal-producing states in Australia as well as other nations. By only including coal supplied by Queensland, the assumption is made that 64 350 Annual Coal Exports (Mt) Fig 11 - Historical resource life comparison - Technique 2. Future coal exports are the factor that will affect most on the forecasts made as part of this project. Therefore, export trends for the past decade were assessed in Figure 12, for the countries of high export quantities or areas of projected coal export growth, as nominated by the Queensland Department of Infrastructure and Planning (DIP, 2010). The DIP indicated that Indonesia, Kazakhstan and Mongolia would all experience considerable export growth in the coming decade as coal export infrastructure is developed and expanded (DIP, 2010). Figure 12 shows that substantial export growth has occurred in both Australia and Indonesia during the past decade. 300 250 200 150 100 50 0 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Year Australia USA Mongolia Indonesia South Africa Russia Kazakhstan Fig 12 - Global coal export trends, 2000 - 2010. Using Indonesia as an export growth example, it appears approximately a decade is required for a nation to transition from small-cap coal exporter (<50 Mt/a) to one of the global leading coal exports, exporting in excess of 200 Mt/a. If this assumption proves to be true, Kazakh and Mongolia exports, combined with continued growth in Indonesian exports, have the potential to over-supply the global coal market. In this event, the forecasts made as part of this project will most likely prove to be inaccurate, with the peak coal event occurring prior to the 2050 - 2053 forecast. However, the likelihood of over-supply occurring in the global coal market can be best assessed by also investigating global coal demand. As with coal supply, coal demand was outside of the scope of modelling conducted as part of this project. However, global coal demand will ultimately play a large role in dictating annual production in Queensland. If the global demand for Queensland coal does not exist, then the resource will simply not be extracted. Therefore, expected global coal demand should be assessed in order to determine possible impacts on depletion forecasts made. Figure 13 indicates total coal imports for China and India to reach approximately 280 Mt by 2025. If the Chinese and Indian import levels continue to 2050 - 2053, as with Japanese demand, approximately 70 per cent of annual coal production will be exported to these nations during that period. It would then be reasonable to assume that domestic coal demand, combined with the imports of other nations, will account for the remaining annual coal production that occurs at the peak coal level devoted to coal export. Therefore, provided there are no drastic global events affecting international coal supply and demand, the projections should be relatively accurate regarding resource depletion and annual production. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Modelling Depletion of Queensland Coal Resources using Historical Production Trends Coal price and exploration spending The price of coal influences a variety of factors affecting the depletion forecasts made in this project. Coal price influences profitability margins for extraction and overall company revenue. Historical coal price data is shown in Figure 14; exploration spending is also shown. 200 Coal Imports (Mt) 180 160 140 120 Coal-related infrastructure The final projected depletion scenario indicates a peak coal of 650 Mt to occur between 2050 and 2053. While the operating coal mines during the 2050 - 2053 period may have the capability to extract the required 650 Mt of coal to reach the peak coal projection, there will be no economic motivation to do so if the required infrastructure to export and sell the coal does not exist. In order to meet the projected peak coal level, sufficient coal systems will have sufficient capacity. A coal system is comprised of (DIP, 2010): 100 80 60 40 20 0 2009 2011 2013 2015 2017 2019 2021 2023 2025 2027 2029 Year China India Japan Fig 13 - International coal imports. 700 350 600 300 500 250 400 200 300 150 200 100 100 50 0 •• mines and associated coal preparation plants, coal stock piles and load-out facilities •• regional road, power and water infrastructure •• a rail network and rail-haul providers that deliver product coal to domestic users and export ports •• one or more coal export terminals •• a deep-water export coal port, and supporting infrastructure and services •• bulk carriers to transport the coal overseas. The Queensland coal systems are divided into four areas, and are shown in Figure 15. The four subsystems are the: 1. Newlands system 2. Goonyella system 3. Blackwater-Moura system 4. Western system. In the case of Queensland, the main limiting factors of the coal system are rail or port facilities. Therefore, by assessing current and planned rail and port facilities, the potential Coal Price ($/t) Figure 14 indicates a strong correlation between coal price and exploration spending. Exploration spending leads to exploration drilling, which invariably results in increases in overall resource size for Queensland coal. Therefore, the conclusion can be made that coal price will play a large part in the determination of the final URR scenario that eventuates. Coal price is expected to increase with demand and as the resource becomes depleted. Whilst it is difficult to predict the coal price required to produce the projected URR and annual production scenarios, trends indicate that coal price will continue to increase for the near future. Therefore, the accuracy of the projections made should be maintained. Exploration Spending (million $) government policy, namely royalties and taxes, will not overly affect the accuracy of the depletion projections made as part of this project. 0 1980 1985 1990 1995 2000 2005 2010 Year Exploration Spending Coking Coal Price Thermal Coal Price Fig 14 - Historical coal price and exploration spending. Government policy The state and federal government has the ability to impact coal industry revenue in the form of royalties and taxes. The size of the government royalties and taxes impacts on the overall economic viability of coal extraction and, thus, resource size and annual production levels are impacted. However, whilst both governments increase royalties and taxes with time it is in their best interest to keep the tax at a level that does not impact annual coal production or exploration spending in an overly negative manner. If it does, the revenue generated for either government will be diminished, which would be an undesirable scenario. Therefore, it can be concluded that Mining Education Australia – Research Projects Review 2012 Fig 15 - Queensland coal systems (DIP, 2010). 65 ability of Queensland coal infrastructure to export 550 Mt of coal during the peak coal scenario can be assessed (assuming an 85 per cent export rate). Current port expansion plans and current projects under construction indicate that Queensland will have sufficient port facilities to meet the projected export requirements throughout the depletion forecast. However, rail limitations could prove to be an issue. The main capacity limitation will exist where the proposed Surat Basin Railway links with the Blackwater-Moura rail system. However, the Surat Basin Railway will be a key component of any future coal projects in the Surat Basin (DIP, 2010). Additionally, increased capacity will be required of all aspects of the Queensland coal system when coal mining begins in Galilee Basin and at a later point, the Eromanga Basin. Currently, there are no operating mines in these areas, and this is largely due to the costs involved in establishing sufficient coal transport infrastructure for the two basins. In order to handle the additional coal supplied annually by mines produced in the Galilee and Eromanga Basins, a standalone rail link will have to be created to one of the existing port facilities. Provided planned expansion projects progress as planned and a stand-alone rail link is established for the Galilee and Eromanga Basins, coal production can feasibly progress as forecasted. Conclusions Having conducted two separate modelling techniques, using symmetric and asymmetric Gaussian curves, the final and most accurate forecast was determined to be that using the asymmetric curve for the low and high URR scenarios. Using this modelling approach it was found that the true Queensland coal depletion scenario would consist of the following characteristics: •• •• •• •• •• URR – 56 447 - 75 975 Mt Resource depletion – year 2198 - 2326 Resource life – 186 - 314 years Peak coal – 650 Mt/a Peak coal – year 2050 - 2053. Using the 650 Mt peak coal projection and historical data on the number of operating mines, it was concluded that 40 - 55 mines would be operating at the time of peak coal, with average annual production between 11.8 - 16.3 Mt/mine. Additionally, historical resource life estimates were assessed. It was found that using the government method of calculating coal resource life, 130 - 160 years has been projected for the past five years. Prior to this period, a steady trend of decreasing resource life was evident. In comparison, resource life projections were made using the asymmetric Gaussian curve modelling technique. Projections were made for 1980, 1995 and 2010. The modelling indicated that resource life varied between 146 - 151 years for the three time periods tested. However, the productions trend for each period indicated that peak coal would increase by 100 Mt for each period following the 350 Mt/a projection for 1980. The modelling techniques used in this project indicated that projections correlated strongly with historical production data. It was determined that a number of factors existed outside the scope of the project with the ability to influence the true depletion scenario. These factors included: •• coal supply and demand •• coal price 66 •• government policy •• coal infrastructure systems. Review of the identified factors indicated that whilst they could affect the true depletion scenario, accuracy of the projections should be maintained. References Australian Bureau of Agricultural and Resource Economics, 2010. Australian commodity statistics 2010 [online]. Available from: <http://www.adl.brs.gov.au> [Accessed: 7 May 2012] Bardi, V and Lavacchi, A, 2009. A simple interpretation of Hubbert’s model of resource exploitation, Energies 2009, 2:646-661. Bartlett, A, 2006. A depletion protocol for non-renewable natural resources: Australia as an example, Natural Resources Research, 15: 151-164. Department of Natural Resources and Mines, 2012. Queensland’s coal – Advanced mines and projects [online]. Available from: <http://mines.industry.qld.gov.au/assets/coal-pdf/new_coal_ min_adv_proj.pdf> [Accessed: 20 September 2012]. DIP, 2010. Coal Plan 2030 – Laying the foundations of a future [online], Queensland Government. Available from: <www.deedi. qld.gov.au> [Accessed: 7 May 2012]. DME, 2007. Department of Mines and Energy. Queensland’s worldclass coals [online]. Available from: < http://mines.industry.qld. gov.au/ assets/coal-pdf/wcc_nov_07_1.pdf> [Accessed: 10 May 2012]. Fitzpatrick, A, Hitchon, B and McGregor, J R, 1973. Long-term growth of the oil industry in the United States, Mathematical Geology, 5:237-267. Geoscience Australia, 2012. All publications from 2004 to 2011. Australia’s identified mineral resources [online]. Available from: <http://www.ga.gov.au/cedda/publications/1201> [Accessed: 20 September 2012]. Guo, H, 2011. A simple algorithm for fitting a Gaussian function, IEEE Signal Processing Magazine, 89:134-137. Höök, M, Zittel, W, Schindler, J and Aleklett, K, 2010. Global coal production outlooks based on a logistic model, Fuel, 89:3546-3558. Hubbert, M, 1956. Nuclear energy and the fossil fuels, Drilling and Production Practice, 205: 7-25. Kato, T, Omachi, S and Aso, H, 2011. Asymmetric Gaussian and its applications to pattern recognition, PhD thesis (unpublished), Graduate School of Engineering, Tohoku University, Sendai-shi. Mohr, S, Höök, M, Mudd, G and Evans, G, 2011. Projection of longterm paths for Australian coal production – Comparisons of four models, International Journal of Coal Geology, 86(4):329. Mohr, S and Evans, G, 2009. Forecasting coal production until 2100, Fuel, 88:2059-2067. Moore, C L, 1966. Projections of US petroleum supply to 1980, Office of Oil and Gas, Washington DC, p 47. Mutton, A J, 2003. Queensland Coals 14th Edition: Physical and Chemical Properties, Colliery and Company Information, Queensland Department of Natural Resources and Mines. Patzek, T and Croft, G, 2010. A global coal production forecast with multi-Hubbert cycle analysis, Energy, 35:3109-3122. Productivity Commission, 1998. The Australian Black Coal Industry [online]. Available from: <http://www.pc.gov.au> [Accessed: 7 May 2012]. Queensland Exploration Council, 2011. Queensland Exploration Scorecard, Queensland Resources Council. The Queensland Coal Board, 1980 - 1997. The Queensland Coal Board annual reports/reviews from 1980 to 1997 [online]. Available from: <http://www.nrm.qld.gov.au/mines> [Accessed: 4 May 2012]. Mining Education Australia – Research Projects Review 2012 Management of Trailing Cables on Electrically Powered Load-Haul-Dump Units W Paterson1 and P Knights2 ABSTRACT Electric load-haul-dump units (eLHDs) using trailing cables are more increasingly being used in underground hard rock mining operations in place of conventional diesel load-haul-dumps (LHDs). This is due to electric LHD’s desirable operational characteristics and lower operating costs. Unlike diesel LHDs, electric LHDs do not release diesel particulates, have lower heat and noise emissions and have cheaper energy requirements. This in turn reduces requirements on the installed capacity of the mines’ ventilation system. A number of mine design restrictions that impact on the operational application of electric LHDs were identified as part the research project. Electric LHDs were found to be most suited to block and panel caving operations, due to their centralised long life extraction levels. The extraction level layouts identified as most suited for the implementation of electric LHDs were the offset herringbone and herringbone layouts. Other mine design considerations identified as part of this study were: trailing cable length, automation barrier systems, gate end bays, electrical requirements, type of trailing cable and roadway design. Data pertaining to electric LHD downtime was sourced from Northparkes Mines. The average availability, utilisation and effective utilisation of the electric LHDs was significantly different from previously projected values. The mean time between failure and the mean time to repair were also determined as part of the analysis. The key causes of maintenance downtime were identified as part of the study. Scheduled maintenance was the largest contributor, followed by drive and gearbox faults and then trailing cable and electric faults. INTRODUCTION LOAD-HAUL-DUMPS Electric LHDs using trailing cables are becoming more common-place in the mining industry and have many benefits over traditional diesel LHDs. However there is a lack of information relating to the practical application of eLHDs. This project summarises information relating to the safe operation and reliability of trailing cables. Data provided by Northparkes Mine (NPM) is used to assess the impact maintenance of eLHDs using trailing cables has on their availability and performance. The data provided allows for the determination of the primary causes of unplanned maintenance. Significant advancements were made in underground metalliferous mining in the 1960s. One of the more significant advancements made was the introduction of new underground mining machinery such as LHDs (Matikainen, 1991). LHDs are a mid-sized mining vehicle that are used in underground metalliferous operations to load, haul and dump material (Ward, 2009). The introduction of LHDs in the 1960s resulted in substantial changes to mine planning and production rates. A number of problems were encountered with the LHDs after they were initially introduced; these related to their lower than expected availability and significant amount of maintenance required. Most of these problems were overcome with better understanding of the machines and how to effectively and properly manage them. Since the 1960s there have been a number of advancements relating to LHDs with a focus on developing electrically powered and automated LHDs (Matikainen, 1991). When assessing the application of eLHDs in a mining environment, consideration must be given to the impacts their use will have on the mines’ design. One of the key design issues associated with the application of eLHDs is their limited movement due to the length of the trailing cable. This movement restriction will in turn affect the layout of the extraction level and the manner in which the ore is extracted. This information is summarised allowing NPM and industry to easily assess what design changes would need to be implemented or improved for the use of eLHDs using trailing cables. This project is part of Mining Education Australia’s (MEAs) electric mine initiative and can be used by the industry to help assess the benefits of using eLHDs. Most LHDs are powered by either diesel or electricity through the use of trailing cables. Each of these methods has their own specific advantages and disadvantages. Diesel powered LHDs have many desirable operating characteristics. These include high productivity, high equipment utilisation, high mobility and operational flexibility (Weiss et al, 1981; Ross, 2008). Problems associated with using diesel LHDs include: 1. SAusIMM, School of Mechanical and Mining Engineering, The University of Queensland, Brisbane Qld. Email: [email protected] 2. MAusIMM, Head of Division of Mining Engineering, School of Mechanical and Mining Engineering, The University of Queensland, Brisbane Qld. Email: [email protected] Mining Education Australia – Research Projects Review 2012 67 W Paterson and P Knights higher noise levels, increasing fuel price, exhaust emissions and increased heat emissions. One of the consequences associated with the use of diesel LHDs, is the need to increase and improve ventilation due to the presence of diesel exhaust emissions. Additional ventilation requirements would result in increased ventilation costs (Grigg, 1985; Brannon et al, 2008). repositioned to the gate end bay at the opposite end of the extraction drive in order to bog the remaining drawpoints. LHDs would also be unable to reverse into the opposite drive without running over the trailing cable and risking damage (deWolfe, 2009). Electrically powered LHDs have a number of associated advantages and disadvantages. The advantages include: minimal exhaust emissions, reduced heat emissions, cheaper energy requirements and reduced noise levels. Due to the reduced heat and exhaust emissions, a reduced ventilation capacity can be installed, thus reducing associated ventilation costs (Abrahamsson and Johansson, 2008; Grigg, 1985). Disadvantages associated with the application of eLHDs include: reduced flexibility, limited haul range, limited mobility, additional development and their limited application to specific mining methods and extraction level layouts (Grigg, 1985; Weiss et al, 1981). The continuous trough layout is characterised by its extensive longitudinal undercut trough drift which is developed and blasted prior to the production drifts and drawpoints. The design of the layout results in no minor apices, meaning that there is no lateral restraint to the major apex. The continuous trough layout has been used and considered by a number of mines, however, no recorded cases could be found of eLHDs being utilised for this layout. The layout can be modified to allow for the use of eLHDs, by orientating the drawpoints towards the incoming LHD. Other infrastructure would have to be installed to cater for the use of eLHDs (Brown, 2003). METHODS SUITABLE FOR THE APPLICATION of eLOAD-HAUL-DUMPS The El Teniente layout was developed and named after the El Teniente copper mine in Chile. Drawpoint cross-cuts are developed in straight lines at a 60˚angle to the major apices and production drifts. The minor apex of the El Teniente layout is shorter and more stable than other designs; this is in part due to the rectangular shape of the columns (deWolfe, 2009; Esterhuizen and Laubscher, 1992). Diesel LHDs are widely used in metalliferous mining operations and can be used in a variety of circumstances and mining methods. This contrasts with eLHDs, which are limited to the few mining methods that have long-life centralised extraction levels. The limited application of eLHDs is associated with the necessary infrastructure and the limited range of operations caused by the length of the trailing cables. These limitations mean that eLHDs are most suited to block caving operations and mining methods based on similar principles. Mining methods that could utilise eLHDs are: panel caving, inclined drawpoint caving and front caving (Brown, 2003). Block caving and panel caving are the most common of these methods and are discussed in more detail. EXTRACTION LEVEL LAYOUT Introduction The design and performance of the extraction level in block or panel caving operations have significant impacts on the operational efficiency and cost effectiveness of the operation as a whole (Brown, 2003). Several horizontal LHD layouts are utilised in block caving operations; however of these only four are significantly different from each other. The four extraction layouts to be considered are: Henderson, Continuous Trough, El Teniente and Offset Herringbone. The application and feasibility of using eLHDs varies for each of the layouts (Brown, 2003). Henderson (Z design) The Henderson layout was developed at Henderson mine, USA. The layout has since been phased out of Henderson’s operations and there are no other known cases of this layout being used. The Henderson layout consists of opposite drawpoint drifts that are in line with each other but inclined to the production drift. Drawbells and drawpoints at right angles to the production drifts (Brown, 2003). The application of eLHDs has not been trialled for the Henderson layout. A number of modifications would need to be made to allow for their use. The modified layout would require power sources to be located at opposite ends of the extraction drives, as only half of the drawpoints would be orientated towards an incoming LHD. Once half of the drawpoints had been bogged the LHD would need to be 68 Continuous trough (trench layout) El Teniente (parallelogram) Electric LHDs using trailing cables were trialled at El Teniente in the 1980s, this resulted in unsatisfactory outcomes. However electric LHD technology has changed significantly since the 1980s and more is understood about the mine design requirements for their application (Anon, 2010). To allow for the use of electric LHDs a power feed would be required at each end of the panel. The LHD would be required to reposition to the different feeds more frequently, which would cause operational delays. Similar to the Henderson design, the LHDs would not be able to reverse into the opposite drawpoint without running over the trailing cable and potentially damaging it (deWolfe, 2009). Offset herringbone The offset herringbone layout is a modification of the herringbone layout. Due to the improved geotechnical and operational aspects associated with the offset herringbone layout, the original herringbone layout has all but been phased out. The offset herringbone layout has become the most commonly used layout for new block and panel cave operations. The layout is characterised by drawpoints which are offset and at right angles to the major apex (Esterhuizen and Laubscher, 1992). The offset herringbone layout was used as the extraction level layout for the block caving operations at NPM (Brown, 2003). Electric LHDs using trailing cables are most suited in their application to the offset herringbone layout. An advantage of the offset herringbone layout is that the LHD will always face the same direction. This is particularly important when the LHD trams to a permanent crusher or ore pass. As a result a power feed only needs to be installed at one end of the panel; this means that there will not be operational delays from repositioning the LHD. The Northparkes E26 Lift 2 offset herringbone extraction level layout can be seen in Figure 1. This layout was designed with the intention of using eLHDs with trailing cables, and a number of modifications have been made to accommodate them including permanent gate end bays (deWolfe, 2009). Mining Education Australia – Research Projects Review 2012 ManageMent of traIlIng caBleS on electrIcally powered load-haul-duMp unItS first extraction level design for Lift 2 E26 did not include any dedicated GEBs. Instead, the LHDs were to be parked in the extraction drives and plug in points were to be installed on the sidewall opposite the drive entrance. This design would have prevented any vehicular movement along the perimeter drive whilst LHDs were operating. The design was later changed to one which had permanent GEBs allowing for improved operations (Rio Tinto, 2002; Ross, 2008). electrical The electrical requirements of the mine should be calculated and the appropriate infrastructure should be installed to allow for the operation of eLHDs. The designed electrical system must meet certain predetermined criteria. Hartman (1992) defines the ten key criteria for a designed electrical system: Fig 1 - Northparkes E26 Lift 2 (deWolfe, 2009). deSIgn conSIderatIonS for electrIc load-haul-duMpS trailing cable length One of the key factors that must be considered when designing an extraction level layout for eLHDs is the required length of the trailing cable. If the cable is too short the LHD will not be able to function for its designed task, however if the cable is too long the LHD will become less efficient as it is carrying unnecessary weight and any future cable replacements will be more costly. NPM encountered this problem when transitioning from Lift 1 to Lift 2 of the E26 block cave. The eLHDs used for Lift 1 had a total trailing cable length of 280 m, this fell short of the 330 m length required to operate in Lift 2. A program was put in place to refurbish the cable reels and associated motors allowing them to operate in Lift 2 (Rio Tinto, 2006; Ross, 2008). 1. safety to personnel and property 2. reliability of operation 3. simplicity 4. maintainability 5. adequate interrupting ability 6. current-limiting capacity 7. selective-system operation 8. voltage regulation 9. potential for expansion 10. initial capital cost. Sandvik (2011) outlines a number of considerations that need to be taken into account when designing an electrical system for eLHDs: • • • • Barrier systems NPM is in the process of trialling the automation of its eLHD fleet. As part of this process a barrier system using a light curtain is installed. If one of the light curtains is breached, the automated eLHDs shut down and cannot be reactivated until the cause or the breach is investigated and they are manually reset. Barrier systems were originally installed close to the floor, however this resulted in unforseen operational delays. During everyday operation of eLHDs, trailing cables are occasionally subject to recoiling. The recoil from the trailing cables was enough to set off the safety barriers, shutting down all automated production for a period of time. Modifications were made to the height of the barrier systems from the floor reducing the downtime associated with trailing cable recoil (Rio Tinto, 2011). However this problem is ongoing and still results in LHD downtime. To avoid downtime from trailing cable recoil further research should be conducted into improving the design of the automation barrier system however, this exceeds the scope of this project. gate end bays Gate end bays (GEBs) are designed to allow for the parking of eLHDs while providing space for the Gate End Panels (GEPs) or ‘plug in’ points. GEBs are not required for the operation of eLHDs using trailing cables, but allow for improved operations that reduce the potential for damage, to trailing cables and their associated infrastructure. NPMs MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 • • • electric LHDs typically demand 1.5 to 1.8 times the nominal power while mucking the power demand on the supplying network will be moderately high the transformer size should be dimensioned to the expanded power requirements ±10 per cent tap changer should be equipped to the supplying transformer to regulate source voltage and to compensate for voltage and torque drops it is not advisable to connect any other loads other than LHDs to the same substation advisable not to supply all of the LHDs from the same substation, and recommended limit of four LHDs per substation. trailing cable Modern electrical cables used in underground mining operations are constructed using a complex layering system. Cables may comprise of a number of layers including a conductor, conductor shield, electrical insulation (dielectric), insulation shield, auxiliary shield, primary shield, dielectric field, insulation and an outer cover (Thue, 1999). Many manufactures recommend specific trailing cables for their electric LHDs. Sandvik (2011) recommends: • • • • a four core rubber trailing cable pilot wireless trailing cables pilot wireless earth continuity monitoring vector-plus connectors and receptacles. roadway design The typical operation of eLHDs poses many hazards that can result in damage to the trailing cable. One of the hazards which trailing cables are most commonly exposed is the risk of the trailing cable becoming stuck or damaged by debris in the extraction drive. Other hazards include the cable being 69 W Paterson and P Knights damaged or suffering abrasion from a rough roadway surface. These hazards were taken into consideration during the design of extraction levels of the block caves at Northparkes. Once the extraction level had been developed, a concrete roadway was constructed for the extraction level. The concreting of the roadways resulted in less damage and abrasion to the cables, reduced machine maintenance, improved water drainage and an increased tyre life (Rio Tinto, 2006). The block cave concrete roadway designs were modified to improve upon the previous design. The E48 Feasibility Study Report (2006) outlined a number of learnings from the previous roadway designs, as follows: •• the use of 80 MPa concrete is important for abrasion resistance •• the practice of installing dowel bars for join bonding between the road panels successfully inhibits vertical movement •• the use of steel reinforcement in the extraction drives and no steel reinforcement in the draw point slabs is suitable •• the blow-off of the floor to solid rock is time and material consuming •• the road panel thickness previously used is sufficient •• efforts need to be made to further inhibit cracking •• the design with the sharp corner on the drawpoint panel is insufficient •• the implementation of a QA-system as used in E26 Lift two effectively ensures the production of high quality roadways •• kerb construction lowers maintenance costs significantly •• water runoff, coming mainly from the dust suppression, is very limited making drains redundant. The extraction level of lift one E48 required 4862 m of concrete roadway to be laid at a rate of 20 m a day (Rio Tinto, 2006): •• The concrete laid in the drawpoints is subject to significant wear due to the mining operations. This was taken into consideration as part of NPM’s roadway design of the drawpoints which only used in situ cast concrete slabs without reinforcement; this differed from the typical roadway design which used reinforcement. The primary reason for this design alteration was that the drawpoint roadway could be easily repaired and did not contain any steel reinforcement which could pose a safety hazard to the operation of the eLHDs. OPERATIONAL CHARACTERISTICS of LOADHAUL-DUMPS Overview of data set Data was provided by NPM pertaining to breakdowns, scheduled maintenance and machine downtime for their LHD fleet. The data set contains information from the 1st August 2011 to the 17th September 2012, a period of 414 days. The data provided contained information relating to the operating characteristics of 12 eLHDs and two diesel LHDs. Six of the eLHDs had insufficient data or were not scheduled for production over the period that the data was recorded and were excluded from the analysis. The data recorded for the remaining six eLHDs was extensive with an average of 1660 separate incidents recorded per LHD over the period of operation. Table 1 summarises information relating to the provided data set while Table 2 contains information relating to the data sets of the LHDs analysed. 70 TABLE 1 Summary of the data set provided by Northparkes Mine. Parameter Value Shift period 12 hours Period over which data recorded (dates) 1/08/11 to 17/09/12 Period over which data recorded (days) 414 Total number of recorded incidents 14 218 Number of eload-haul-dump units 12 Number of eLHDs with adequate data sets 6 Average number of recorded incidents (eLHD) 1660 Number of diesel load-haul-dump units 2 Number of diesel LHDs with adequate data sets 2 Average number of recorded incidents (diesel LHD) 1415 TABLE 2 Summary of the data from the load-haul-dump units analysed. Load-hauldump type/no LHD No Period (days) No of incidents Electric LHD8 414 1746 Electric LHD9 414 1740 Electric LHD10 414 1712 Electric LHD11 414 1701 Electric LHD12 414 1739 Electric LHD14 414 1331 Diesel LHD7 414 1446 Diesel LHD13 414 1384 Previous studies Before the commencement of Lift 1 E48 block cave the availability and utilisation of the eLHDs had to be determined. NPM operates a fleet of six Tamrock 1400E Electric LHDs and one Tamrock Diesel 1400D LHD in their current production operations. Rio Tinto’s feasibility study of Lift 1 E48 (2006) determined the availability and utilisation the eLHDs and the diesel LHD to be 82 per cent and 92 per cent respectively. Table 3 provides a breakdown of the various losses associated with the availability and utilisation. Service time was determined to be the largest loss associated with eLHD availability; this was closely followed by contingency for major component failures and break-downs, including cable and tyre damage. Shift change was found to be the only loss associated with the utilisation, most of the time lost is associated with travel to and from the shift changeover. The E48 feasibility report (2006) made a number of recommendations to improve the utilisation of the LHDs, these include: •• hot seat changeovers •• utilising the automation of the eLHDs to remotely control them from the surface, meaning that the operator does not need to waste time traveling down to the level. Electric load-haul-dump-units The operating characteristics of the eLHDs were calculated using information provided by NPM. Table 4 summarises Mining Education Australia – Research Projects Review 2012 Management of Trailing Cables on Electrically Powered Load-Haul-Dump Units TABLE 3 Availability and utilisation values for load-haul-dump units (Modified from Rio Tinto, 2006). Availability Value Losses 82% Service time (6%) Breakdowns, including cable and tyre damage (4%) Contingency for major component failures (5%) Utilisation Effective utilisation 92% Shift change (8%) 75.44% the values calculated for all of the eLHDs. The calculated availability of the eLHDs was similar for all machines. The availability of the eLHDs ranges between 85.91 per cent for LHD11 and 92.93 per cent for LHD14. The average availability of the eLHDs was found to be 88.29 per cent, this value is higher than the 82 per cent availability predicted in the E48 feasibility study. The utilisation was also determined for the eLHDs to range between 75.13 per cent and 79.07 per cent, with an average value of 77.18 per cent. The values calculated for the utilisation of the eLHDs were found to be similar to each other with no outlying values. The effective utilisation was found to range between 65.80 per cent for LHD8 and 72 per cent for LHD14. The average effective utilisation for the eLHDs was calculated to be 68.13 per cent this value is significantly lower than the 75.44 per cent predicted in the E48 feasibility study. The discrepancy in these two values is due to an unrealistically high utilisation being used in the feasibility study which only takes into account shift change and does not consider either other operating delays or no scheduled production. As part of the their feasibility study for the E48 block cave Rio Tinto completed a productivity simulation as part of this simulation they used a MTBF of 120 operational hours and a MTTR of 251 minutes. The MTBF for the eLHDs ranged between 18.26 hours for LHD8 and 31.42 hours for LHD11. The average MTBF was found to be 21.95 hours, this value is significantly lower than the one used in the productivity simulation. The variation between the two MTBF values can be partially attributed to the difficulties of operating an automated eLHD fleet which at the time of compiling the E48 feasibility report had not been trialed at other sites. The MTTR was found to range between 136 minutes for LHD11 and 188 minutes for LHD12. The average MTTR for the eLHDs was found to be 167 minutes this value is significantly smaller than the 251 minute value used in the E48 feasibility study. The lower MTTR value means repairs to the eLHDs are taking a significantly shorter period of time than previously predicted, this could be due to a number of things, including: •• improved competency of workforce (both operators and service personal) •• improved maintenance procedures •• increased frequency of minor repairs •• reduced frequency of major repairs. Diesel load-haul-dump-units The two diesel LHDs in operation at NPM were analysed as a comparison point for the eLHDs. These LHDs are used in both production and development operations and are exposed to different operating conditions than their electric counter parts. The smaller sample size of diesel LHDs means that the results from the analysis are not as representative as that of the eLHD analysis. Table 5 shows the calculated operating characteristics of the diesel LHDs. The availability of the diesel LHDs was found to range between 89.30 per cent for LHD7 and 94.57 per cent for LHD13, with an average availability of 91.94 per cent. The average availability of the diesel LHDs was higher than the previously predicted value of 82 per cent in the E48 feasibility study. The increased availability of the LHDs may be attributed to any number of operational or environmental factors. The utilisation was determined to range between 65.75 per cent and 68.01 per cent, with an average value of 66.88 per cent. The effective utilisation of the diesel LHDs were found to range between 60.73 per cent for LHD7 and 62.18 per cent for LHD13, with an average effective utilisation of 61.46 per cent. The values calculated for the effective utilisation of the diesel LHDs were found to be significantly lower than the previously indicated in the E48 feasibility study. The main reason for the large discrepancy between that of the feasibility study and the calculated utilisation can be attributed to unrealistic estimates being used for the feasibility study. Both the MTBF and MTTR were calculated for the diesel LHDs. The MTBF for the diesel LHDs ranged between 27.50 hours and 38.02 hours with an average MTBF of 32.07 hours. The MTBF values for both diesel LHDs differ significantly and a representative MTBF value cannot be determined due to the small sample size. The calculated MTBF values are significantly smaller than the ones used in the E48 feasibility studies productivity simulation. Possible reasons for the large discrepancy between the calculated and the predicted MTBF values include: TABLE 4 Calculated operating characteristics of Northparkes Mine’s eload-haul-dump units. Load-haul-dump unit no Availability (%) Utilisation (%) Effective utilisation (%) MTBF (hours) MTTR (minutes) LHD8 87.52 75.18 65.80 18.26 169 LHD9 89.29 75.13 67.08 19.24 180 LHD10 86.01 78.28 67.32 20.78 164 LHD11 85.91 79.07 67.93 31.42 136 LHD12 88.07 77.95 68.65 20.15 188 LHD14 92.93 77.47 72.00 21.83 166 Average 88.29 77.18 68.13 21.95 167 Mining Education Australia – Research Projects Review 2012 71 W Paterson and P Knights TABLE 5 Operating characteristics of diesel load-haul-dump units at Northparkes Mine. Load-haul-dump units no Availability (%) Utilisation (%) Effective utilisation (%) MTBF (hours) MTTR (minutes) LHD7 89.30 65.75 60.73 27.50 226 LHD13 94.57 68.01 62.18 38.02 142 Average 91.94 66.88 61.46 32.76 184 •• unrealistic MTBF values used for the productivity simulation •• changes to the environmental or operational conditions •• untrained or unskilled workforce (both operational and service personal) •• ageing LHD fleet. The calculated MTTR values range between 142 minutes for LHD13 and 226 minutes for LHD7, with an average MTTR of 184 minutes. There is a large variation between the two LHDs MTTR values; this means making a meaningful representative value impossible. Both the calculated MTTR values are smaller than the ones predicted in the E48 feasibility study. Electric and diesel load-haul-dump-unit comparison A comparison, of the electric and diesel LHDs operating at NPM, was made to ascertain which of these LHDs has more desirable operating characteristics. Due to the smaller sample size of diesel LHDs their operating characteristics are not as conclusive as the eLHDs. Both electric and diesel LHDs were found to have similar availabilities, however the diesel LHDs were found to have an availability that was 3.65 per cent higher than the eLHDs. There was a greater variance between the two types of LHD for the utilisation. The eLHDs were found to have an average utilisation that is 10.3 per cent higher than that of the diesel LHDs. The larger variation in utilisation can be in part attributed to the diesel LHDs operating under more varied and challenging conditions than the eLHDs. The variation TABLE 6 Classification system used by Northparkes Mine to categorise various types of maintenance. Category Description Planned maintenance Scheduled maintenance Communications fault Power failure and not defined Drive/gearbox fault Foreign object/contamination, incorrect installation, mechanical overload, temperature and not defined Leak Foreign object/contamination, incorrect installation and not defined Low flow Hydraulic oil flour Low pressure Tyre failure Roller fault Foreign object/contamination, fines/carry back and not defined Sensor/proxy fault Human error, incorrect installation, not defined, power failure and sensor error Structural fault Bent, collapsed, cracked and not defined Trailing cable/electrical fault Electrical fault, electrical overload, foreign object/ contamination, human error, incorrect installation, not defined, power failure, temperature, PLC fault and trailing cable fault 72 between the two types of LHD for the utilisation may also be attributed to a preference to use eLHDs over diesel LHDs due to their lower operating costs. The average MTBF variation between the types of LHD was 10.81 hours, with diesel LHDs having an average MTBF value of 32.76 hours and eLHDs having an average MTBF value of 21.95 hours. This is a considerable variation and is indicative of diesel LHDs having a lower failure frequency than eLHDs. There is also considerable variation between the MTTR values for both types of LHD, with electric and diesel LHDs having a MTTR value of 167 and 184 minutes respectively. This means that it takes on average 17 minutes less to repair an eLHD than it does a diesel LHD. Although this variation may be due to a number of things, there are two likely causes: 1. failures that occur in diesel LHDs are more labour intensive 2. eLHDs have an increased number of minor failures when compared to diesel LHDs. DISTRIBUTION OF MAINTENANCE DOWNTIME Introduction It is important for a mining operation to understand the main causes for machine maintenance so appropriate plans and procedures can be introduced so as to improve machine availability. NPM classifies maintenance preformed on LHDs into ten distinct categories, which are outlined in Table 6. Electric load-haul-dump-units Time lost due to maintenance for the eLHDs was quantified into the ten distinct categories used by NPM. Planned maintenance was found to be the largest contributor to downtime for all of the eLHDs analysed. Planned maintenance ranged between 36.96 per cent and 49.49 per cent. Faults associated with the drive/gearbox were the second largest contributor of maintenance time for all of the eLHDs with the exception of LHD14. These ranged between 14.15 per cent and 27.20 per cent. LHD14 had trailing cable and electrical faults as its second largest contributor to maintenance downtime. The availability and utilisation of LHD14 was also higher than all of the other eLHDs assessed. There may be a number of reasons for this, however the mostly likely reasons are that LHD14 is newer than the other eLHDs or it has been recently refurbished and therefore does not have as many problems with its drive/gearbox. The third largest contributor to downtime for all of the eLHDs, excluding LHD14 was trailing cable/electrical faults. The trailing cable/electrical faults had a range between 13.04 per cent and 21.06 per cent. Structural faults were the fourth most time consuming repairs, this was followed by communication faults, leaks and sensor/proxy faults which all accounted for approximately the same amount of downtime. The downtime associated with low flow and roller faults was insignificant. Low pressure only accounted for a small amount of downtime on LHDs 8 and 9. Mining Education Australia – Research Projects Review 2012 ManageMent of traIlIng caBleS on electrIcally powered load-haul-duMp unItS LHDs accounts for 47 per cent of all maintenance downtime whereas it only accounts for 42 per cent of downtime in eLHDs. The variation of planned maintenance is most likely due diesel LHDs requiring scheduled maintenance every 125 hours compared to eLHDs which only require it every 500 hours (Rio Tinto, 2006). Dive/gearbox faults accounted for 31 per cent of all of the maintenance downtime for diesel LHDs this value is significantly larger than the 22 per cent for eLHDs. Maintenance downtime from trailing cables/electrical faults accounted for six per cent of total maintenance time for diesel LHDs whereas it accounted for 15 per cent for eLHDs. This large variation between the two types of LHD, for both drive/ gearbox faults and trailing cable/electrical faults is probably due to diesel LHDs not using trailing cables resulting in a redistribution of the maintenance downtime. Fig 2 - Graph of the average classified eLHD maintenance downtime. Communications and sensor/proxy faults contributed marginally less to downtime for diesel LHDs than for the eLHDs. The variation between these values may be due to the smaller sample size of the diesel LHDs. Structural faults, leaks, low pressure, low flow and roller faults account for a similar proportion of maintenance downtime for both electric and diesel LHDs. concluSIonS Fig 3 - Graph of the average classified diesel LHD maintenance downtime. diesel load-haul-dump units The percentage of total repair time required for each of the ten distinct maintenance categories used at NPM for diesel LHDs was calculated. The largest contributor to downtime for diesel LHDs was planned maintenance, which ranged between 38.85 per cent and 63.34 per cent of total downtime. The main cause of downtime for LHD7 was drive/gearbox faults which accounted for more downtime than planned maintenance however the difference between the two accounted for less than one precent of lost time. The second largest contributor to downtime was the faults associated with the dive/gearbox. Trailing cable/electrical faults and structural faults were the next largest contributors to downtime, followed by leaks. Minimal amounts of downtime were associated with low flow, low pressure, roller faults and sensor/proxy faults. comparison of electric and diesel load-hauldump units A comparison of diesel and eLHD maintenance downtime was completed using the average downtime associated for each classification of maintenance. Figure 2 and Figure 3 show the distribution of maintenance downtime for electric and diesel LHDs respectively. The maintenance downtime distribution of the two types of LHD varies significantly. Planned maintenance for the diesel MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 This project determined that eLHDs are suited to block caving operations and other mining methods which are based on the same fundamental principles. The offset herringbone and herringbone extraction level layouts were found to be most accommodating to the application of eLHDs. A number of mine design restrictions that impact on the operational application of eLHDs were identified. These included; trailing cable length, automation barrier systems, gate end bays, electrical requirements, type of trailing cable and roadway design. The project identified the need for further research, into ways in which the design of the automation barrier systems, could be modified so as to avoid interference from trailing cable recoil. The average availability of eLHDs and diesel LHDs operating at NPM was determined to be 88.29 per cent and 91.94 per cent respectively. The utilisation of eLHDs was found to be 77.18 per cent. This value was larger than the 66.88 per cent determined for diesel LHDs. This variation in utilisation may be attributed to the diesel LHDs operating under more varied conditions. Electric LHDs were found to have a significantly lower MTBF value than diesel LHDs. The MTTR for eLHDs was found to be 17 minutes less than diesel LHDs. This difference was attributed to eLHDs requiring less complex maintenance than diesel LHDs. Planned maintenance was found to be main cause of electric and diesel LHD downtime. Drive and gearbox faults were determined to be the second largest contributor to LHD downtime. Electrical and trailing cable faults accounted for 15 per cent of total eLHD maintenance downtime. This value varied greatly from the 6 per cent recorded for diesel LHDs. A larger sample size of diesel LHDs is required if a comprehensive analysis between diesel and eLHDs is to occur. acknowledgeMentS This research project would not have been possible without the support and assistance of several people and parties. The 73 W Paterson and P Knights authors would like to thank Rio Tinto Northparkes Mine for providing data for this project, in particular Brendon Burrows and Trent Burke for their time and assistance with the collection of data. The authors would also like to thank Markus Chang and Matthew Elston for their assistance in proofreading and the discussion of technical concepts. Matikainen, R, 1991. How to increase LHD productivity in hard rock mining, in Proceedings International Conference: Reliability, Production and Control in Coal Mines, pp 85-88 (The Australasian Institute of Mining and Metallurgy: Wollongong). REFERENCES Rio Tinto Pty Ltd, 2002. Northparkes mines Lift 2 north extension feasibility study, Rio Tinto report E26L2NPFS. Abrahamsson, L and Johansson, J, 2008. Work culture and gender issues in a changing technical context – examples from LKAB iron ore mine in Kiruna, in Proceedings 5th International Conference and Exhibition on Mass Mining, pp 1129-1138 (Lulea University of Technology, Division of Mining and Geotechnical Engineering: Lulea). Anon, 2010. El Teniente goes deeper, International Mining, October, 38:10-15. Brannon, C, Casten, T, Hewitt, S and Kurniawan, C, 2008. Design and development update of the Grasberg block cave mine, in Proceedings 5th International Conference and Exhibition on Mass Mining, pp 433-442, Lulea University of Technology, Sweden. Hartman, H, 1992. SME Mining Engineering Handbook, 2161 p (Society for Mining, Metallurgy, and Exploration: Littleton). Rio Tinto Pty Ltd, 2006. E48 project feasibility study October 2006, Rio Tinto report E48PFS. Rio Tinto Pty Ltd, 2011. Automation at Northparkes [online]. Available from: <http://www.dpi.nsw.gov.au/__data/assets/ pdf_file/0004/353479/Automation-@-Northparkes-Rio-Tinto. pdf> [Accessed: 1 May 2012]. Ross, I, 2008. Northparkes E26 Lift 2 block cave – A case study, in Proceedings 5th International Conference and Exhibition on Mass Mining, pp 25-34 (Lulea University of Technology, Division of Mining and Geotechnical Engineering: Lulea). Sandvik Mining and Construction, 2011. Sandvik electric LHDs, Northparkes. Brown, E, 2003. Block Caving Geomechanics, 511 p (Julius Kruttschnitt Mineral Research Centre: Brisbane). Thue, W, 1999. Electrical Power Cable Engineering, 417 p (Marcel Dekker: New York). deWolfe, C, 2009. Drawpoint layouts for block caves, Rio Tinto report 12733-2. Ward, B, 2009. Load haul dump (LHD) automation [online], Commonwealth Scientific and Industrial Research Organisation. Available from: <http://research.ict.csiro.au/research/labs/ autonomous-systems/field-robotics/mining-robotics/loadhaul-dump-lhd-automation> [Accessed: 1 May 2012]. Esterhuizen, G and Laubscher, D, 1992. A comparative evaluation of production-level layouts for block caving, in Proceedings 2nd International Conference and Exhibition on Mass Mining, pp 63-70 (University of Pretoria, Department of Mining Engineering: Pretoria). Grigg, F, 1985. Future developments in underground mobile equipment, in Proceedings First International Federation of Automatic Control Symposium, pp 165-174 (University of Queensland, Department of Mining Engineering: Brisbane). 74 Weiss, P, Fettweis, G, Moschitz, I, Olsacher, A and Riedler, H, 1981. Relevant factors for development and draw control of block caving, in Proceedings Design and Operation of Caving and Sublevel Stoping Mines (American Institute of Mining, Metallurgical, and Petroleum Engineers: Denver). Mining Education Australia – Research Projects Review 2012 The Effect of Joint Properties on a Discontinuous Rock Mass M Sherpa1 and P Hagan2 Abstract When considering the strength of a rock mass having joint sets, the dominant parameters are joint orientation, joint frequency and joint strength. Under experimental testing conditions, these parameters strongly influence rock mass strength, elastic properties and the mode of failure. A review of the literature identified a lack of information concerning the influence of joint frequency on the properties of a rock mass. This study aimed to assess the effects of joint frequency on rock mass properties, particularly in a confined stress state. A total of 24 uniaxial and 87 tri-axial compressive strength tests were undertaken using intact and discontinuous, jointed sandstone core specimens, the latter having sawn-cut joints orientated across the longitudinal axis of the core. Based on the results, several conclusions have been made concerning the Anisotropic Effect factor, the Joint Factor and the mode of rock failure. First, the Aniostropic Effect factor can be adjusted for joint frequency. Second, the joint strength parameter of the Joint Factor can also be modified and finally, a single shear failure plain was observed for intact and jointed rock specimens under biaxial confinement. These conclusions provide an insight into the effects of joint frequency in a discontinuous rock mass. Introduction Rock masses often contain discontinuities having varying length, frequency and orientation. Discontinuities are considered as planes of weakness, that can control and influence the strength, deformation and failure behaviour of a rock mass (Brady and Brown, 2004). In the laboratory, a discontinuous rock mass can be modelled through induced joints in intact rock samples. Laboratoryscale testing can provide an insight into the influence of joint properties on a discontinuous rock mass and hence in understanding rock mass behaviour. Background The Joint Factor method as proposed by Ramamurthy and Arora (1994) provides a method of estimating the unconfined compressive strength ratio (σcr) of a discontinuous rock mass. This was based on a series of confined and unconfined strength tests using plaster and sandstone cores that contained induced joints of different configurations. The Joint Factor (Jf) is defined in terms of joint frequency (Jn) expressed in units of fractures per metre (f/m), joint orientation (n), and joint strength (r) as shown in Equation 1 (Ramamurthy and Arora, 1994). Jf = Jn n×r (1) The Joint Factor represents a reduction in rock mass quality. A modification was proposed to the Joint Factor by Jade and Sitharam (2003) who found it could be used to estimate the compressive strength of a rock mass as shown in Equation 2. J f m c vcr = 0.039 + 0.893e - 160.99 (2) The Joint Strength parameter (r), is defined as: r= x vn (3) where τ and σn are the shear and normal stresses acting along a joint plane. In their original proposal, Ramamurthy and Arora (1994) recommended direct shear testing to obtain joint strength parameter values. If direct shear testing is not possible then they provided suggested values for the joint strength parameter which for clean joints varied with the compressive strength of the intact rock as shown in Table 1. An earlier study by Byerlee (1978) found that for joints containing no infill material, when a high normal stress is applied, the joint strength parameter is independent of rock type and UCS. Byerlee (1978) suggested that the joint shear stress, t, is dependent on the applied joint normal stress, σn, as shown in Equation 4. x = 0.85 × v n (4) Equation 4, known as Byerlee’s Law, is valid over the normal stress range of 5 to 200 MPa for clean joints only (Paterson and Wong, 2005). 1. GAusIMM, The University of New South Wales, Sydney. Email: [email protected] 2. FAusIMM, The University of New South Wales, Sydney. Email: [email protected] Mining Education Australia – Research Projects Review 2012 75 M Sherpa and p hagan TABLE 1 Suggested values for the Joint Strength parameter, for various strength rock samples (after Ramamurthy and Arora, 1994). Uniaxial compressive strength of intact rock (MPa) Joint strength parameter, r 2.5 0.30 5 0.45 15 0.60 25 0.70 45 0.80 65 0.90 100 1.0 Remarks Fine-grained micaceous to coarse-grained Singh, Rao and Ramamurthy (2002) stated that the joint strength parameter can be estimated at ‘sufficiently low normal stress.’ When subjected to compressive strength testing, a discontinuous rock mass and its joints are quite often subjected to high normal stresses as is the case when joints are orientated at or near to perpendicular to the axial loading direction. As joint orientation rotates from the perpendicular to parallel to the axis of loading, the rock mass properties are significantly affected. In an unconfined stress state, the rock mass strength can be predicted by a single plane of weakness theory, which was first proposed by Jaeger (1960). The effect of joint orientation on a discontinuous rock mass is further complicated by the addition of confining pressure. Einstein and Hirschfeld (1974) and Jade and Sitharam (2003) suggested that, at high confining pressures, strength is independent of joints and failure often occurs across joints, through the intact rock. Fig 1 - Effect of confinement on a discontinuous rock mass containing a single induced joint at varying orientation angles, β (Ramamurthy and Arora, 1994). The addition of confining pressure can suppress the influence of joints in the rock mass. This can be seen in Figure 1 where Ramamurthy and Arora (1994) showed how the effect of joint orientation on the ratio of maximum to minimum principle stresses decreases as confining pressure is increased. To quantify the effect of joint orientation on a discontinuous rock mass under a confining pressure, Ghazvinian, Hadei and Madani (2008) proposed the Anisotropic Effect factor (Ae) such that: Ae = via - v ja via (5) Fig 2 - Effect of joint orientation on the principle stress ratio (Ghazvinian, Hadei and Madani, 2008). where σia and σja are the intact rock and jointed rock mass compressive strengths at some confining pressure a. The Anisotropic Effect factor can be used in conjunction with the Joint Factor method to predict confined compressive strength ratios. They based their findings on tests using plaster core specimens having an intact strength of 8.7 MPa and two orthogonal joints at varying joint orientations. Their results are similar to those of Ramamurthy and Arora (1994) in Figure 1, although the results of Ghazvinian, Hadei and Madani (2008) in Figure 2 are symmetrical at about 45°, as a result of the two orthogonal joints. The Anisotropic Effect factor is greatest for joint orientations which produce minimum strength. As the confining pressure is increased, the Anisotropic Effect factor is reduced for all joint orientations as can be seen in Figure 3. To date, the Anisotropic Effect factor has only accounted for joint orientation. It could be expected that joint frequency would similarly influence rock mass strength. 76 Fig 3 - Variation in the Anisotropic Effect factor (Ae) with joint orientation (after Ghazvinian, Hadei and Madani, 2008). MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 the effect of joInt propertIeS on a dIScontInuouS rock MaSS eXperIMental work research aims The objective of the study was to assess whether the spacing of discontinuities in a rock or the joint frequency alters the Anisotropic Effect factor in a jointed rock mass. To achieve this, an experimental testing regime was undertaken, consisting of unconfined and confined compressive strength tests using intact rock and discontinuous rock mass samples. Tests were undertaken using a Hawkesbury Sandstone sample. Joints were sawn across the diameter of the diamond cored rock specimens having a diameter of 41 mm with length of 102 mm. These cuts were orientated perpendicular to the axial loading direction as shown in Figure 4. The joints were free of any infill. Due to the fine grained nature of the sandstone, the diameter was considered sufficient for compressive strength testing. Fig 5 - Effect of loading platens that are stiffer than the test material (Tang and Hudson, 2010). Compressive strength testing The testing apparatus recorded axial force and axial displacement, allowing for calculations of axial stress-strain values. Estimations of peak strength, linear elastic modulus and normal stiffness were made from the stress-strain curves. The joint frequencies tested were 10, 20, 40 and 60 f/m. The discontinuous rock mass samples contained one, two, four and six sawn cut joints respectively for these joint frequencies. Fig 4 - Examples of sandstone specimens containing induced joints or discontinuities. Methodology The confining pressures (σ3) selected for testing are presented in Table 2. Originally, the confining pressures selected were much higher, in the range of 5 to 50 per cent of the intact rock strength, σci, as used by Ghazvinian, Hadei and Madani (2008) in their testing. Preliminary testing at the five per cent level resulted in a three-fold increase in rock strength. It was assumed that the effects of confining pressure on joint frequency would be observed at lower strengths, thus the confining pressures were lowered to the values presented in Table 2. TABLE 2 Range of confining pressures (σ3) used in the test program. Experimental considerations The rock strength tests were undertaken using an MTS815 model stiff testing machine. A tri-axial cell provided confinement to the rock specimens during testing. A constant deformation rate of 0.003 mm/s was used in all tests. This rate achieved peak strengths within five to ten minutes of loading. Differences in elastic properties between rock and the steel loading platens used in testing were expected to influence results. Brady and Brown (2004) stated that the applied axial stress can become non-principle when the test material and loading platens are of different stiffnesses. This issue and its effects are described by Tang and Hudson (2010) who stated that pre-peak and peak stress-strain behaviour is independent of loading platen properties and are only found to influence post-peak stress-strain and failure behaviour. Being made of steel, the platens were much stiffer than the sandstone. The effect that this difference in stiffness would have on the stress distribution can be seen in Figure 5. With the sample/platen interface at the top and bottom of the rock specimen, compressive failure cones are observed. A tensile failure plane is also evident in the middle of the sample (Tang and Hudson, 2010). Tests were replicated three times to account for variability in the rock sample and test method. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Levels of confining pressure σ3 (as per cent of σci) 0.5 1.5 2.5 4 6 σ3 (MPa) 0.1 0.25 0.5 0.8 1.2 reSultS A summary of the results of the uniaxial and triaxial tests at increasing levels of joint frequency and of confining pressure is shown in Table 3. In the unconfined state, it was found that the peak stress reduced with joint frequency as shown in Figure 6. This is consistent with the results of Hagan et al (2012). Similarly, elastic modulus also decreased with joint frequency. Considering the combined effects of the discontinuities and confinement, it was found that the presence of discontinuities in the rock reduced the peak strength; it was more sensitive at the highest joint frequency of 60 f/m as shown in Figure 8. Whereas at the lower discontinuity frequencies of 10 and 20 f/m, there was little difference between the peak strength values at the various levels of confinement attained compared to the intact specimens. A similar variation was observed with Elastic Modulus as shown in Figure 9. 77 M Sherpa and p hagan TABLE 3 Test results. Joint frequency (f/m) 0 f/m 10 f/m 20 f/m 40 f/m 60 f/m Confining pressure (MPa) Peak stress (MPa) Elastic modulus (gPa) Normal stiffness (MPa) 0.00 20.6 3.8 931 0.10 25.8 4.6 683 0.25 38.3 4.7 504 0.50 50.6 6.4 563 0.80 65.5 7.3 518 0.00 21.3 5.0 986 0.10 24.1 4.9 464 0.25 37.5 6.4 404 0.50 56.9 7.7 419 0.80 69.6 8.5 562 0.00 18.8 4.0 549 0.10 25.6 4.8 581 0.25 37.7 5.8 452 0.50 54.3 6.6 457 0.80 71.5 7.6 363 0.00 15.7 2.4 410 0.25 31.2 3.3 328 0.50 47.2 4.3 404 0.80 59.8 4.8 412 0.00 13.5 2.0 312 0.10 21.0 2.5 308 0.25 28.9 2.9 346 0.50 43.1 3.9 333 0.80 53.7 4.2 295 Fig 7 - Variation in the peak strength of intact specimens with confinement pressure. Fig 8 - Variation in peak strength with confinement pressure for varying levels of discontinuity frequency. pronounced change in the elastic modulus, however, which decreased with confinement pressure and joint frequency. Elastic properties For the intact samples, the load fell sharply after achieving the peak load to the residual strength level. By contrast, in those test specimens containing discontinuities there was a more gradual reduction in load as can be seen in Figure 10. The discontinuous samples underwent significantly more deformation which, as suggested by Einstein et al (1970), is due to closing of the induced joints. Failure observations Fig 6 - Variation in peak stress and elastic modulus joint frequency in unconfined tests. When considering the effects of confinement on the intact test specimens, strength was found to increase with confinement pressure, σ3, as shown in Figure 7. Initial observations showed no significant difference in the levels of peak strength between the intact and discontinuous samples at joint frequencies of 10 and 20 f/m over the range of confining pressures tested. There was, though, a more 78 When testing specimens in an unconfined state, a buckling tensile failure at the midpoint of the specimen was often observed as seen in Figure 11 as well as vertical splitting. This is consistent with the type of failure predicted by Tang and Hudson (2010) due to differences in the elastic properties of the steel loading platens and the rock material, resulting in creating a tensile plane, as seen in Figure 5. Furthermore, compressive cones were observed at the top and bottom of the test specimens, also described by Tang and Hudson (2010), resulting in minimal fracturing within these areas with all tested specimens. An example of the effect of the compressive cone can be seen in Figure 12. In this example, the failure zone is circumferential around the top and bottom edges of the test specimen. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 the effect of joInt propertIeS on a dIScontInuouS rock MaSS were subjected to confinement, the middle section underwent significantly more radial deformation compared with the intact and the more widely spaced, lower joint frequency specimens. The cause of the increase in radial deformation is a result of top and bottom sample segments acting as intermediate loading platens and concentrating the applied stress within the middle segment. Fig 9 - Variation in the Elastic Modulus with confinement pressure for varying levels of discontinuity frequency. Fig 12 - Failure around sample end. Shear failure proposal For a sample to result in a single shear failure, confinement was required. As a result of an increase in the joint frequency, the confining pressure required for a single shear also increased. A linear relationship has been identified between confining pressure and joint frequency for a single shear failure, and is presented in Figure 13. The relationship is presented in Equation 6. Fig 10 - Stress-strain curves of intact and discontinuous (40 f/m) samples. This relationship provides an indication of when the failure of a discontinuous rock mass under confinement stops being influenced by joint frequency. Equation 6 has been developed from discontinuous rock mass samples containing induced joints perpendicular to the major loading direction. Considering the major effect joint orientation has on rock mass strength and failure mode, it would be expected that this relationship would not hold for joints sets orientated at other angles. v 3 = 0.0006 # J f + 0.0043 (6) where σ3 is expressed as a percentage of the intact rock compressive strength (per cent σci). dIScuSSIon anisotropic effect factor Fig 11 - Example of buckling failure observed in unconfined tests. In the confined state, however, and particularly at the higher levels of confinement, the dominant failure mode resulted in a single shear failure plain. For tests with a joint frequency greater than 20 f/m, it was observed that failure would always occur through the middle section of the test specimen. When the specimens MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 Based on the earlier observations by Hagan et al (2012), it was expected that the presence of joints in the discontinuous samples would result in a reduction in peak strength when compared to the intact rock. However, this was not observed in testing specimens having widely spaced discontinuities of 10 and 20 f/m as can be seen in Figure 14, where there was little significant difference in the peak strength between the intact and jointed rock specimens over the range of confining pressures studied. The findings of peak strength behaviour between the intact and discontinuous samples have a significant effect when comparing the experimental results with the Anisotropic Effect factor, as presented in Figure 15. 79 M Sherpa and p hagan Fig 13 - Required confining pressure for shear failure with an increase in joint frequency. Fig 14 - Comparison of the peak strength values between the intact and low frequency jointed (10 and 20 f/m) specimens. For the 10 and 20 f/m specimens there was a poor correlation between joint frequency and the Anisotropic Effect factor. At higher joint frequencies of 40 and 60 f/m, a reduction in the Anisotropic Effect factor was observed as confining pressure was increased. This behaviour is similar to the observations for joint orientation by Ghazvinian, Hadei and Madani (2008). It can be concluded that at low joint frequencies when the rock mass quality is still high, there is little difference in peak strength between an intact rock and discontinuous rock mass. However, as the rock mass quality decreased through an increase in joint frequency, the Anisotropic Effect tends to become more dominant. Fig 15 - Experimental results analysing the Anisotropic Effect for joint frequency. Fig 16 - Experimental results and Joint Factor method predictions. For joints orientated perpendicular or near perpendicular to σ1, Byerlee’s Law should be used in the Joint Factor method. concluSIonS Compressive strength testing of intact rock and discontinuous rock mass sandstone core samples was undertaken. The discontinuous samples contained induced joints, sawn cut at varying spacing to replicate the effects of joint frequency. joint strength parameter modification A shear failure relationship was proposed. This relationship, a function of the applied confining pressure and joint frequency, can be used to identify the influence of joint properties on the failure mode of a rock mass sample. The Joint Factor method has been compared with the experimental results. Initial Joint Factor predictions were made using the values recommended by Ramamurth and Arora (1994), that is, 0.98 for joint orientation and 0.65 for the joint strength parameter. These resulted in a significant underestimation compared with the experimental results, as seen in Figure 14. It was found that, at the low joint frequencies tested of 10 and 20 f/m, there was little significant differences in strength between the intact and jointed rock specimens. Consequently at these joint frequencies, there was no correlation with the Anisotropic Effect factor. As joint spaced decreased and joint frequency increased, the impact of the Anisotropic Effect factor was similar to that previously seen for joint orientation. By substituting the joint strength parameter value of 0.65 with Byerlee’s (1978) recommendation of 0.85, this reduced the underestimation also shown in Figure 16. If direct shear testing is unavailable to estimate a joint strength parameter, care should be taken when using recommended values provided by Ramamurthy and Arora (1994). 80 A modification of the Joint Factor method has been proposed. It was found that for clean joints orientated perpendicular to the major loading direction, the joint strength parameter is dependent upon the applied axial stress and not on the intact rock UCS. Modifying the joint strength parameter reduced the error in the empirical predictions of the Joint Factor method compared with the experimental results. MINING eDUCATIoN AUSTRALIA – ReSeARCh PRojeCTS ReVIew 2012 The Effect of Joint Properties on a Discontinuous Rock Mass References Brady, B H G and Brown, E T, 2004. Rock Mechanics for Underground Mining, third edition (Kluwer Academic Publishers: Netherlands). Byerlee, J, 1978. Friction of rocks, Pure Applied Geophysics, 116:615-626. Einstein, H H and Hirschfeld, R C, 1974. Model studies on mechanics of jointed rock, Journal of the Soil Mechanics and Foundations Division, 99(3):229-248. Einstein, H H, Nelson, R A, Bruhn, R W and Hirschfeld, R C, 1970. Model studies of jointed rock behaviour, in Proceedings 11th US Rock Mechanics Symposium, California (American Rock Mechanics Association). Ghazvinian, A, Hadei, M R and Madani, B, 2008. Behaviour of mechanical anisotropic specimens under triaxial testing, in Proceedings 42nd US Rock Mechanics Symposium, San Francisco (American Rock Mechanics Association). Hagan, P, Zhang, L, Mitra, R and Kodama, J, 2012. The effect of discontinuities in uniaxial compressive strength testing of rock, in Proceedings 7th Asian Rock Mechanics Symposium, Seoul, Korea. Mining Education Australia – Research Projects Review 2012 Jade, S and Sitharam T G, 2003. Characterisation of strength and deformation of jointed rock mass based on statistical analysis, International Journal of Geomechanics, 3:43-54. Jaeger, J C, 1960. Shear failure of anisotropic rocks, Geological Magazine, 97:65-78. Paterson, M S and Wong, T, 2005. Experimental Rock Deformation: The Brittle Field, second edition (Springer: Netherlands). Ramamurthy, T and Arora, V K, 1994. Strength predictions for jointed rocks in confined and unconfined states, International Journal of Rock Mechanics and Mining Sciences and Geomechanics Abstracts, 31(1):9-22. Singh, M, Rao, K S and Ramamurthy, T, 2002. Strength and deformation behaviour of a jointed rock mass, Rock Mechanics and Rock Engineering, 35(1):45-64. Tang, C and Hudson, J A, 2010. Rock Failure Mechanisms: Explained and Illustrated (CRC Press/Balkema: Netherlands). 81 82 Mining Education Australia – Research Projects Review 2012 APPeNDIX 1 – MeA STUDeNT CoNFeReNCe PRoGRAMMe oCToBeR 2012 2012 MEA Student Research Conference Curtin University (WASM), 12 Oct 2012, Kalgoorlie WA Programme (12 Oct 2012, WASM, Hobson Lecture Theatre): 08:45 – 09:00 09:00 – 10:20 09:00 – 09:20 09:20 – 09:40 09:40 – 10:00 10:00 – 10:20 10:20 – 10:40 10:40 – 12:00 10:40 – 11:00 11:00 – 11:20 11:20 – 11:40 11:40 – 12:40 12:40 – 14:00 12:40 – 13:00 13:00 – 13:20 13:20 – 13:40 13:40 – 13:50 13:50 – 14:00 Welcome & Opening of the Symposium (A. Jarosz & P. Hagan) Session 1 Vanessa Collins (UQ): Maximising production rates at Brockman 4 by Minimising Truck Delays at the Crusher Tim Graham (UNSW): Strategic Project Risk Management for an Emerging Miner Dermott Sundquist, Vaughn Golding, Davood Jafari, Brajesh Rajopadhyaya (UA): Multi-‐Objective Optimisation of Mining-‐Metallurgical Systems Weilin Wang (WASM): ANSYS for Stress Analysis of Underground Structures Morning Tea in Rm.703-‐211 Session 2 Seung Hyeon Lee (WASM): Comparing Neural Networks with JKSimblast Prediction Model in Purpose of Optimal Ground Vibration Induced by Blasting McLeod Mackenzie (UNSW): Prediction and Modelling of Blact Vibration and its Effects at Glendell Colliery Casey Costello (UQ): Grizzly Modifications at Ridgeway Deeps Block Cave Gold Mine Lunch in Rm.703-‐211 Session 3 Brenton Goves (UQ): Continuous Surface Miner Operations at Fortescue Paul Carmichael (UNSW): An Investigation into Semi-‐ intact Rock Mass Representation for Physical Modelling of Block Caving Mechanics Zones Vadim Strukov (WASM): A Comparative Study of Truck Cycle Time Prediction Methods Afternoon Tee in Rm.703-‐211 & Meeting of Prize Committee Announcements of Prizes & Closing of the Conference The order of presentations was determined by a ballot drawn on 10/10/2012 at WASM (Present: A.Jarosz, A.Halim, B.Suwardi, J. Hackney) vi 2 Appendix 2 – Presenters in MEA Student Research Conferences Year/Venue University Student presenter(s) Paper title UoA Dermott Sundquist and Davood UNSW Paul Carmichael An investigation into semi-intact rock mass representation for physical modelling of block caving mechanics zones McLeod McKenzie Prediction and modelling of blast vibration and its effects at Glendell Colliery Multi-objective optimisation of mining-metallurgical systems Jafari Tim Graham (third prize) 2012 WASM UQ Vanessa Collins Maximising production rates at Brockman 4 by minimising truck delays at the crusher Casey Costello (first prize) Brenton Goves (second prize) WASM (Curtin) 2011 UQ 2010 UoA Strategic project risk management for an emerging miner Seung Hyeon Lee Grizzly modifications at Ridgeway Deeps block cave gold mine Continuous surface miner operations at Fortescue Comparing neural networks with JKSimblast prediction model in purpose of optimal ground vibration induced by blasting Vadim Strukov A comparative study of truck cycle time prediction methods Weilin Wang ANSYS for stress analysis of underground structures UoA James Boffo Acoustic emission monitoring of impregnated diamond drilling for deep exploration UNSW Owen Riddy (first prize) UQ Brendan Murphy WASM (Curtin) Colin Thomson UoA Adam Schwartzkopff and Developing a truck allocation model for Bengalla Mine Development of an underground coal scheduling and simulation program Stoping analysis of Golden Grove under high stress conditions Comparisons between three-dimensional yield criteria for fractured and intact rock Daniel Hardea 2009 UNSW UNSW James Tibbett Failure criteria of three major rock types affecting Ridgeway Deeps UQ William Hartley (first prize) WASM (Curtin) Nathan Colli Analysis of the principles for sound waste dump design and placement UNSW Kali Dempster Quantifying time-dependent rock behaviour at Ridgeways Deeps block cave operation UQ Michael Baque Optimisation of coal scheduling at Dawson Mine WASM (Curtin) Kieran Rich (first prize) vii An investigation into final landform criteria required for a safe, stable, sustainable and non-polluting landform in the Bowen Basin Paste fill for stope stability at the Challenger Gold Mine Author Index C Carmichael, P K 1 Karakus, M 37 45 Chalmers, D 51 Kerai, M Chanda, E 19 Kizil, M S 7, 25 Collins, V 7 Knights, P 13, 67 Costello, C 13 G M McKenzie, M K 51 Golding, V 19 Chalmers, D 51 Goves, B 25 Meikle, A 57 Graham, T K C 31 Paterson, W H Hagan, P 75 Halim, A 45 Hebblewhite, B Holland, A 37 Mining Education Australia – Research Projects Review 2012 R Rajopadhyaya, B 19 S Saydam, S Seib, W 37 J Jafari, D 67 1 I Iles, M P 31 25, 57 Sherpa, M 75 Sundquist, D 19 19 83 84 Mining Education Australia – Research Projects Review 2012