Search Documents

Large deposits present computational challenges to todayÆs commercial orebody modelling systems that are based on conventional block models. We describe a system that is specifically engineered to model and visualise large deposits. Central to the system is linear octree encoding of conventional block models and real-time conversion of three-dimensional (3D) meshes to linear octrees. We provide both methodological treatment and practical experience in implementing such a system. A commercial implementation is also presented to demonstrate the validity of the approach.

Sep 26, 2011

By K Muller, T Oosthuizen

Blast exclusion zones are instrumental in protecting equipment and personnel. When there is no constraint in footprint, the exclusion zone can be enormous, practically eliminating the hazard of flyrock. Unfortunately, this may not always be the case and in many circumstances the blast design needs to be altered to suite such scenarios. When a conservative design approach is taken, it could completely eliminate any form of ejection, whether it is from the free face or the bench itself. The degree to which typical flyrock models describe fragment motion through air with ballistic flight equations are not justified when such a high level of control is applied. This paper examines various methods of risk analysis and suggests an approach to be used for controlled blasting.The outcome of principles discussed in this paper can be seen in video form (Mavmedia, 2014).CITATION:Muller, K and Oosthuizen, T, 2015. Should the blast exclusion zone be based on a qualitative or quantitative analysis? A practical implementation conundrum, in Proceedings 11th International Symposium on Rock Fragmentation by Blasting, pp 425–428 (The Australasian Institute of Mining and Metallurgy: Melbourne).

Aug 24, 2015

By G Lane, M Bueno

In a recent publication (Bueno and Lane, 2011) the authors presented the analysis of a consolidated dataset composed by 137 autogenous/semi-autogenous grinding (AG/SAG) mill pilot trials. By comparing the measured data against theoretical specific energy consumption, the analysis indicated which circuit configurations were more or less energy efficient. However, the analysis has also indicated that the energy efficiency of each of the types of circuits vary significantly.There are numerous factors that contribute to the variations of energy efficiency of AG/SAG mill-based circuits, such as feed size distribution, recycle ratios and ore characteristics. In view of this, the present work accesses these factors individually and discusses what may lead a particular circuit configuration to be more or less efficient in terms of energy and costs.CITATION:Bueno, M and Lane, G, 2012. Key drivers of energy and cost efficiency in autogenous/semi-autogenous grinding circuits, in Proceedings 11th AusIMM Mill OperatorsÆ Conference, pp 113-118 (The Australasian Institute of Mining and Metallurgy: Melbourne).

Oct 29, 2012

By M Brockman, M Cowan, A Robeson

"Sunrise Dam Gold Mine (SDGM) is located 820 km from Perth, 220 km from Kalgoorlie—Boulder and 55 km south of Laverton in Western Australia. The SDGM main pit had successfully operated since 1997 and was due to be completed during 2012. An opportunity existed to extract the high-grade crown pillar – 600 kt @ 7 g/t for 135 koz by either open pit or underground methods.The crown pillar was the 60 m (vertically) thick remaining part of the high-grade Cosmo orebody between the top of the paste filled underground mine stopes and the planned base of the final open pit shell.There were two major difficulties being considered when choosing the underground or open pit options:a major wall failure in the upper section of the main pit wall directly above the crown pillar restricting access from the base of the open pit to the top of the crown pillarimpact of a high rainfall event while the crown pillar was being mined from either surface or underground resulting in flooding of the underground mine workings.This paper will discuss the following issues:the process to identify and evaluate the risks in both options using knowledge based modelling software; this technique allowed the evaluation of a complex set of assumptions to arrive at the preferred optionremediation of the Watu failurethe selection of the preferred option and the challenges encountered before, during and after the extraction of the crown pillarthe planning and geotechnical work to ensure the safe extraction of the crown pillarsuccessful outcomes transitioning from open pit operation to a solely underground operation.Citation:Robeson, A, Cowan, M and Brockman, M, 2016. A Case Study on the Assessment, Planning and Extraction of the Crown Pillar at Sunrise Dam Gold Mine, in Proceedings Ninth AusIMM Open Pit Operators' Conference 2016, pp 115–136 (The Australasian Institute of Mining and Metallurgy: Melbourne)."

Nov 15, 2016

By A Tawadrous, D S. Preece

Cave mining is of interest to the mining community as it offers a chance to extract large, deep orebodies with low operating costs and high production rates. It allows for extracting large low-grade orebodies, which may not have previously been considered for mining. Cave mining is done either top down (ie sublevel caving) or bottom up (ie block/panel caving). Both approaches rely on gravity, stresses and weakness of the rock to fracture and achieve a controlled collapse.Block caving starts at the base of the orebody. It progresses upward and across through continuous collapse of the roof. To achieve this, a critical hydraulic radius must be exceeded. This is accomplished by blasting drawbells, through which the caved ore passes to the production level. An entire level above the drawbells (undercut level) is excavated to help initiate the cave. The undercut level ensures the connection between drawbells and achievement of the required hydraulic radius.Conventionally, drawbells are generated by drilling and blasting several times from the production level as well as the undercut level. Mucking between each shot is required to generate enough void space for the rock to swell. The need for several blasts necessitates some downtime between each blast. Also, the cost of developing and maintaining the undercut level can constitute up to 35 per cent of the cost of mining the panel.A new blasting technique was developed to create the drawbells in a single blast from the production level. The technique ensures the connection of the drawbell to the neighbouring ones, thus, the critical hydraulic radius is reached and propagation of the cave achieved without the need for the undercut level. The technique has the potential to reduce the capital cost of a block caving operation, increase its feasibility and reduce time to reach the ore and consequently the payback period. This paper presents the modelling of a traditional drawbell excavation process, the philosophy of the developed technique, the modelling of that new design and its implementation.CITATION:Tawadrous, A and Preece, D S, 2015. A novel blasting technique to create drawbells and eliminate the undercut level in block cave mining, in Proceedings 11th International Symposium on Rock Fragmentation by Blasting, pp 617–624 (The Australasian Institute of Mining and Metallurgy: Melbourne).

Aug 24, 2015

By B Patterson, Y Lanwin

Over the last decade the adoption of autonomous technology has proven its potential to fundamentally alter the mining industry, offering significant improvements in safety and productivity. Automation levels are now increasing exponentially across all parts of the supply chain from drilling, haulage, processing, rail and shipping to the maintenance of the above. However, the management of human error risk is also becoming increasingly complex as the disconnect between the people who operate and interact with mining equipment and those that design the autonomous systems grows in terms of skill and expertise. Furthermore, as comfort levels with automation increases amongst operators, complacency is increasing as the initial functional safety benefits have been realised. This paper introduces an innovative concept that unites two distinct fields, functional safety and behavioural economics, to address the complex issue of human error risk in autonomous mining operations. Functional safety, a well-established field in engineering involves identifying potential hazards, assessing risks and designing systems in a way that minimises the chances of accidents and limits their consequences. Meanwhile, behavioural economics takes a data centric approach to analysing the psychological aspects of human decision-making, acknowledging and seeking to understand the errors that can originate from cognitive biases and heuristics. This paper seeks to bridge these domains by emphasising that human error in autonomous mining is not solely a technical issue but is deeply influenced by human behaviour and decision-making processes. By incorporating insights from behavioural economics, we can better predict, prevent and manage human errors in the mining sector. By applying principles from both disciplines to a recent incident example we will explore how cognitive biases can impact peoples’ interactions with autonomous mining equipment. This interdisciplinary approach will not only enhance the design and implementation of functional safety measures in mining but also lead to the development of people centric solutions that consider the inherent fallibility of people. We will explore how human centred design can help bridge the gap between these two domains and apply systems thinking for a universal approach. Through an indepth case study analysis and empirical data, our research aims to provide actionable insights and practical recommendations for mining industry professionals, regulators and stakeholders involved in the safe integration of autonomous systems.

Apr 16, 2024

By R Goddard

An alternative stoping method has been adopted for narrow vein stoping at the Tasmania Mine in Northern Tasmania. Colloquially described as ‘Radial-in-Reef’ (RiR) stoping, the method employs the drilling of radial rings in the plane of the Reef from a drill stub located at the geographic centre of the proposed stope panel.Production from the Western Zone of the Tasmania Reef has been undertaken remotely from the Reef since 2007, after a detailed geotechnical engineering analysis indicated that the integrity of the sill drives could not guarantee an acceptable level of confidence should an anticipated seismic event occur within a radius of 10 m. The key tenet of the remote mining approach was the exclusion of personnel from sill drives once stoping had commenced. The remote mining method had to be ‘retrofitted’ to four existing stoping blocks totalling 11 sill drives which totalled 160 m vertically. The remote mining method proved to be complex, cumbersome and very expensive. However, the ore was extracted successfully.By early 2010, previously developed ore had largely been exhausted and the future development with an orebody open at depth presented the opportunity to review the current extraction method. However, the new extraction method needed to maintain the tenet of excluding personnel from sill drives of active stoping blocks, but seek opportunities to reduce the difficult and costly aspects of the previous method. The new ‘Radial-in-Reef’ stoping reduced the development requirement from six to two drives per ore block. Cost implications aside, the significant reduction in prestoping extraction ratio was expected to also reduce the likelihood of seismicity. Transverse production rings were designed to be drilled radially from a drill stub mined from a mid-stope footwall drive. Stopes totalling up to 7000 tonnes were planned be fired in a single electronic blast. Extraction was designed to take place from a basal sill drive under teleremote operation. At the time of writing, the method was under trial.

Mar 21, 2011

By R J. Goldie

The Monty Copper Gold Deposit is an ultra-high grade mine in Sandfire Resource’s Doolgunna region which was completely drilled out and developed within two years of commencing mine development. During this period, a total of 986 grade control diamond drill holes were completed, assayed, and validated, 15 levels developed, and 3136 faces mapped. The aggressive schedule meant a high flow of geological data was produced, validated, and incorporated into regular geological model updates. To process this data at a rate which allowed for timely use, more robust data processing methodologies needed to be generated. A new validation system was implemented which required two levels of validation being applied to any diamond drilling update. Step one was the validation of initial physical data including survey pick-ups and logging. This was done prior to new drill data being added to the geological model. Step two was the validation of assay data which was required prior to resource estimation. The implementation of Seequent’s Leapfrog Geo modelling software also allowed for a more seamless and efficient modelling workflow. The use of the Leapfrog Geo vein systems allowed for more rapid generation of geological models which then afforded more time for stringent assessment of the mineralisation system. An example of this was the introduction of cross-shift peer reviews in which opposing rosters of geologists would rigorously evaluate the previous rosters work prior to reporting. This additional level of validation resulted in more robust geological models and the identification of new ore trends. This presentation will detail the evolution of the Monty Geological model from a simple tabular indicated resource into a complex highly refined measured resource. Themes will include how the model was produced using workflows developed and optimised to meet key deliverables.

Mar 22, 2022

Australia had an active uranium mining industry in the 1950s with uranium oxide being supplied under contract to the atomic energy authorities of the U.K. and U.S.A. In the 1960s, when these con- tracts concluded, domestic production ceased, apart from Rum Jungle, whose production until 1971 was stockpiled. By the 1970s, the demand for uranium from the electrical utility industry had risen sharply and the effects on Australia were notable. The Mary Kath- leen mine re-opened in 1976 but, more significantly, exploration activity suddenly expanded with the re- sult that several world-standard deposits were locat- ed. Western world requirements for uranium based on increasing energy consumption and a changing energy mix, will warrant the development of Aust- ralia's resources. By 1985 our mines could be pro- ducing 9,500 tonnes of uranium oxide yearly and by 1995 the export value from uranium could reach that from wool. In terms of benefit to the community the economic rewards are considerable but, in terms of providing energy to the world, Australia's uranium is vital.

Jan 1, 1978

By S Lehane, S Pearce

The characterisation and assessment of waste rock has been the subject of considerable research, and many guidance documents have been published both internationally and within Australia. While these documents provide detailed information on how to characterise waste rock, there is not a great amount of detail on how to determine the optimal waste management options from an operational viewpoint. For example, typical questions that a mine operator would seek answers to include: What placement method should be used (eg end tipping versus paddock dumping)? What is an optimal tip head height for waste placement in a storage facility? How should sulfide grade control be carried out?Defining the contributing factors to waste management and waste material placement and assessing the contribution of each variable directs these assessments towards a semi-qualitative and quantitative methodology. Quantitative methodologies show an increase in the confidence of these assessments by providing logical, measureable assessments.As part of a strategic assessment for waste management at an active mine site, a detailed risk-based investigation of the influence of variables common to waste placement was completed. These variables included placement method, placement height, sulfur grade and material conditioning (eg compaction) on the associated risks of waste material storage. The assessed risks include temperature assessments (potential for thermal take-off as a result of spontaneous combustion), toxic gas production and flow and acid production potential. Applying a numerical modelling approach to the assessment of the different waste material placement strategies allows for the variation in risk to be graphically presented and communicated through risk matrices and histograms.CITATION:Pearce, S and Lehane, S, 2015. Quantitative risk assessment tools to assist with waste management and placement guidelines, in Proceedings Tailings and Mine Waste Management for the 21st Century, pp 353–364 (The Australasian Institute of Mining and Metallurgy: Melbourne).

Jul 27, 2015

By J Ireland, S Rose

The main outcomes of mining grade control include defining the boundary between ore and waste and then categorising the ore according to which stockpile it should be sent. A typical workflow includes using general mine planning software (GMP) – such as SurpacTM, VulcanTM or CAE StudioTM – to create three-dimensional wireframes of mineralisation domains, which are used to flag the relevant samples of interest and constrain estimation during block modelling. Creating the wireframes is a time-consuming process, requiring reasonable software skills, patience and an eye for detail. The interpretations can be subjective; indeed, even the same geologist will give different shapes if presented with the same data at different times. The process can become particularly tedious when trying to carry out updates due to new data. It can also be difficult to test the effect of alternative interpretations or grade thresholds because of the time constraints in production.There are now several implicit modelling tools available, either as standalone programs or as add-on modules to existing GMP. In this paper, the authors develop a workflow using implicit modelling methods to show how grades and geology can be rapidly interpreted to provide effective domains for grade estimation. The paper includes case studies from open pit gold mines. The implicit modelling approach can save several days of interpretation work for each grade control drilling campaign.CITATION:Rose, S and Ireland, J, 2014. Implicit modelling applied to grade control data, in Proceedings Ninth International Mining Geology Conference 2014 , pp 115–122 (The Australasian Institute of Mining and Metallurgy: Melbourne).

Aug 18, 2014

Conditional simulation can be used to determine the probabilities of block iron-grades being above selected cut-offs. However, there are much faster alternatives that can produce acceptable results. In this paper blasthole data from a selected area of Joffre Member hosted ore of the Brockman Iron Formation at the Mt Whaleback orebody is used as the basis of a case study to assess alternatives to simulation. Iron grade, in per cent, is interpolated from a blasthole dataset into a block model using ordinary kriging. Samples are then removed from this blasthole dataset to produce data subsets. A complete distribution of iron grades for points or blocks is obtained utilising the data subsets and the following techniques: conditional simulation, median indicator kriging, the probability from conditional expectation and confidence intervals attached to the estimation of blocks in the framework of the discrete Gaussian model. The complete blasthole dataset is considered to represent the true iron grades at sample point support while the ordinary kriging estimate derived from these samples is considered to represent the true block grades. The distributions obtained from point estimates, block estimates and simulations are compared to each other and to the true estimates of iron grades using graphs and statistics. The conditional expectation technique reproduced the results obtained from conditional simulation extremely well and it is a viable alternative to simulation at point support. At block support the confidence intervals technique reproduced the results of the conditional simulation more closely than the median indicator kriging and in the data subsets tested the confidence intervals technique actually reproduced the true grade distribution more closely than the conditional simulation.

Jan 1, 2005

By E Philippa, B Miller

Material characterisation, ore sequencing and process capability are commonly regarded as the key parameters to consider in designing and operating an effective pit to plant supply chain. Therefore, an understanding of rock mass properties is the starting point for controlling the presentation of problematic material to the ore handling plant and optimising process utilisation. There are different methodologies for characterising problematic material and ore handleability which exist across a spectrum of sophistication. At one end of this spectrum are simple indices built using first principles geology and at the other are complex algorithms developed using advanced analytical methods, machine learning and ore traceability. Regardless of the method, without an integrated process for careful management of problematic ore, material handling delays (ie blocked, bridged, blinded and buried events) will interrupt the stable flow of production. The ability of the ore control geologist to mitigate these delays is central to the ore handleability challenge. This paper describes a quick and easy solution that links the presentation of ore to the operating window at the crusher. This operating window is based on material characterisation, using the physical, textural and mechanical properties of the rock mass, and crusher performance. It also describes how a basic problematic ore index (POI) can be developed using ore tracking to the production face, petrographic studies of ore texture and mineralogy, and existing variables in the geological model. Finally, this paper will demonstrate that by coding the POI into the geological model for mine planning, together with performance monitoring through compliance to plan routines, material handling delays at the crusher can be reduced, unlocking significant resource value.

Mar 22, 2022

The N3500 orebody is the northern continuation of the 3500 orebody at Xstrata Copper's Enterprise Mine and occurs at a depth of 1.3 to 1.8 km. At the nearby S3500 orebody, extreme deformation was observed in access drives where there was a high number of bedding planes per metre combined with either graphite or talc infill. The observed failure mechanism was largely controlled by the geometry of the inclined bedding submitted to high horizontal stresses. Due to the anisotropic strength and deformation behaviour of the shale rock mass, traditional isotropic methods of geomechanical property estimation based upon GSI are not appropriate. A numerical modelling approach, based upon the distinct element codes UDEC (2D) and 3DEC (3D), has been developed to accurately simulate the failure mechanisms and deformation observed in the initial infrastructure surrounding the N3500 orebody.

Nov 25, 2010

By G M. Jeffress, D C. Arne

There are two discrete aspects to collecting portable X-ray fluorescence (pXRF) data to obtain reliable results and to comply with reporting requirements under the 2012 edition of the JORC Code (JORC, 2012) – sampling method and quality assurance. A robust sampling program is required to obtain representative material for analysis. The appropriate method should be established for a particular project using a small orientation study. A variety of approaches to sampling soils, drill cuttings, rock faces and diamond drill core are discussed. These include using multiple spot analyses, scanning using a single operating beam and the collection of cutting saw pulps. The operator must also be aware of the attenuating effects of analysing through sample bags of different material types for various elements as well as the limitations of current pXRF instruments in reporting reliable data for certain elements. The analysis of the samples must also be accompanied by a quality assurance program in order to generate data that are ‘fit for purpose’. This includes recording X-ray tube resolution on a routine basis, using matrix-appropriate standards for which total elemental concentrations are available, using silica blanks, capturing sufficient duplicate analyses to constrain data precision and obtaining check assays to calibrate pXRF data against laboratory data. Examples of quality control approaches and data are presented. Table 1 of the 2012 JORC Code also recommends that information on instrument make and model, reading times and calibration be provided in any reporting of pXRF data. To this should be added information on operating mode and temperature, software version and moisture content of the samples. A robust sampling methodology with a suitable quality assurance/quality control program should produce pXRF data of sufficient quality for public reporting purposes, provided that the data are presented using appropriate cautionary language and adequate supporting information.CITATION:Arne, D C and Jeffress, G M, 2014. Sampling and analysis for public reporting of portable x-ray fluorescence data under the 2012 edition of the JORC Code, in Proceedings Sampling 2014 , pp 57–64 (The Australasian Institute of Mining and Metallurgy: Melbourne).

Jul 29, 2014

By I Gipps

Each of us has a vision of what the future mine will be and, generally, these visions have many similarities and some differences. One of the major discrepancies in visions of the future mine is not the ultimate goal but the path to achieve it. Many of the future concepts require a radical re-thinking of how things are and will be done yet. So how do we reach this visionary future mine? We need a path that satisfies both our future aspirations of a high tech gee-whiz operated from our iPads and the unfortunately mundane approach that says ‘we must know the method works, ie nothing new’.This paper explores how, within a research based organisation, we firstly set our ultimate goal and then divide the path into realistic technology development chunks that would appeal to mining companies to progress within their organisational constraints. Through examples, such as, automated LHDs, communications systems, surveying and localisation, rock cutting and geological sensing, we illustrate how this path will get us from the here and now to the out there future.This paper also demonstrates CSIRO’s short, medium and long-term research in many of the areas needed to create the Future Mine: mining methods, automation and remote control, mine communications, mapping and mission planning, geological sensing and, importantly, human-system integration, which examines how to effectively present data as information to operators and supervisors in an intelligent and useful manner.

Nov 22, 2011

By P Marianelli, J Theiler, B Beamish, A Spillman

The presence of reactive pyrite in waste rock contributes to the hazard of spontaneous combustion in emplacement facilities as it acts as an initiator of self-heating at low ambient temperatures. Organic carbon also present in the waste rocks acts as a fuel at elevated temperatures above 120°C and can lead to thermal runaway resulting in eventual ignition. A new laboratory test method (adiabatic incubation testing) is now available that can identify the presence of reactive pyrite and quantify the self-heating behaviour of the waste rock. McArthur River Mine has a waste rock profile that contains pyritic shale units, which have created spontaneous combustion events in the past. A systematic sampling and testing program is underway to characterise the spontaneous combustion hazard likelihood of the McArthur River Mine waste rocks to provide a sound database for future mine planning of the waste removal and emplacement. This paper presents some initial outcomes of this program and shows the leading practice in spontaneous combustion hazard assessment and management that has been adopted at McArthur River.CITATION:Beamish, B, Marianelli, P, Theiler, J and Spillman, A, 2018. Spontaneous combustion hazard assessment of McArthur River Mine waste rock, in Proceedings Mine Waste and Tailings Stewardship Conference 2018, pp 422–426 (The Australasian Institute of Mining and Metallurgy: Melbourne).

Jul 23, 2018

By A L. de Ambrosis, S J. Clarke

In the non-codified realm of tunnel design there are issues which are sure-fire instigators of controversy. This paper presents a discussion regarding one particular issue which in the authors’ experience often results in comment and debate. This issue only forms part of the rock load assessment process, but it is specifically addressed with the intention of presenting a rational basis for design, which may result in broader acceptance within the tunnel design community.The issue considered relates to the characterisation of in situ rock stresses for the design of tunnel support in the Sydney Basin. A summary is given of the literature on the topic, with emphasis on the main sources of uncertainty and contention. A design approach is proposed, with particular focus given to the numerical modelling of in situ ground stresses and the approach used to ensure that the adopted stress regime is consistent with the assumed ground model. The resulting variability in the initial stress field is discussed and comparisons made to the test database.CITATION:de Ambrosis, A L and Clarke, S J, 2014. Characterisation of in situ stresses for tunnel lining design in the Sydney?Basin, in Proceedings 15th Australasian Tunnelling Conference 2014 , pp 247–252 (The Australasian Institute of Mining and Metallurgy: Melbourne).

Sep 17, 2014

By D Olwagen, C Stanton, J Gerber

From December 2013 to February 2014 the Southern Middleback Ranges magnetite concentrator went through a successful optimisation process that included the installation of extra grinding power to treat a wider range of ore types. After the initial improvements, sampling surveys were conducted to identify the mineral deportment through the concentrator. Samples collected during the survey were analysed using X-ray fluorescence, quantitative X-ray diffraction and QEMSCAN™ to cross-validate the analysis. Certain streams were also put through a laboratory-scale Davis Tube Recovery washing process to produce clean concentrates for analysis. The QEMSCAN™ analysis identified various metallurgical characteristics (mineral association, particle sizes, grain sizes, particle densities, etc) and these were evaluated for each separation and comminution stage to both understand the processes involved and identify opportunities for improvement. Key findings from the investigation indicated the importance of understanding the elemental deportment of target elements throughout the various minerals. The standard practice of tracking iron deportment through assay provided misleading results on losses to tail through the plant. The investigation provided insight into both wanted and unwanted minerals and the efficiency of each separation stage. The efficiencies of separation are critical due to their impact on downstream capacity, and the investigation provided insight into where opportunities exist to extract more benefit. The findings of the geometallurgical analysis have been implemented on the concentrator, and improvements in the rejection of non-magnetic minerals have been seen.CITATION:Gerber, J, Stanton, C and Olwagen, D, 2015. Using geometallurgy during process optimisation activities at the Southern Middleback Ranges magnetite concentrator, in Proceedings Iron Ore 2015, pp 561–566 (The Australasian Institute of Mining and Metallurgy: Melbourne).

Jul 13, 2015

By E J. Y Koh, G J. McLachlan, E Amini, N Beaton

Preconcentration greatly reduces the ore mass input to the processing plant while upgrading the grade. One of the commonly used methods of preconcentration is optical sorting. These methods rely on optical sensors which classify the material stream into “accept-reject” streams. Currently, it is common to use near infrared (NIR) or colour cameras to classify streams based on reflectance or colour thresholds set by experts. The imaging is done through line-scan sensors where the material is scanned on a belt or a chute. Then, a separation apparatus like a diverter gate or air jet is controlled to separate the streams according to the classification made by the algorithm. Separation efficiency is dependent on the classification algorithm. Existing classification algorithms rely on low-level feature discrimination like observing individual pixel values for reflectance or colour hues. This limits discrimination between minerals with low contrast in colour. Human experts perform mineral segmentation using texture, colour distribution and shape, but it is difficult to translate these visual rules to mathematical thresholds. Deep learning methods like Convolutional Neural Networks (CNNs) provide high-level feature discrimination similar to a human expert. Instead of relying on pre-defined features, the CNNs learn complex features from the dataset. CNNs can construct a hierarchy of features based on pixel clusters to learn texture, shapes, and colour spectrums of the mineral grain surface to provide better separation. In this study, a state-of-the-art deep learning instance segmentation method is used to outline boundaries and classify grains from the background (mineral segmentation). The algorithm is trained on images collected from a flotation feed of a gold deposit, where the gold is mechanically locked inside pyrite. This method can process on-line video inputs at 5 frames per second to provide a live grade. The proposed method is inexpensive as it only uses a camera and a desktop computer.

Nov 10, 2020