Search Documents

By S J. Tear, L D. Queen

"An EXTENDED ABSTRACT is available for download. A full-length paper was not prepared for this presentation. The Frieda River Project is located in the north-west of Papua New Guinea. The target commodities for a bulk mining operation are copper and gold associated with a large felsic porphyry system. The project comprises a series of deposits, with the Horse-Ivaal-Trukai (HIT) being the largest. It also includes the adjacent smaller Koki and Ekwai deposits.Previous resource estimates for the HIT were made using 14 separate mineralised domains based on structure, copper grades, lithology and alteration. H&S Consultants (H&SC) has revisited the deposit and concluded that a simpler methodology could be used for resource estimation without using wireframes as hard domain boundaries. This conclusion is based on reviewing all the available geological data combined with geological and resource experience of porphyry deposits.It is generally acknowledged that porphyry deposits form by diffusion of mineralising fluids through the host lithologies, meaning that sharp boundaries are unlikely unless these fluids encounter specific physical or chemical boundaries. It is important to prepare a comprehensive geological model prior to attempting resource estimation as it forms the framework for the estimates; however, not all geological features may be relevant to the distribution of mineralisation and there can be a tendency to overcomplicate the estimation process by using too many domains. Unless a hard boundary can be identified consistently and unequivocally, it is generally better not to use a boundary in that location.H&SC completed unconstrained models with no top-cutting using both a simple flat search ellipse and a steep search ellipse to generate global resource estimates reported within a predetermined pit shell at a 0.2 per cent copper cut-off. Both outcomes are within 10 per cent for the tonnes, contained copper tonnes and gold ounces compared to the constrained resource estimate. The slight differences are most likely due to the different search strategies, but also indicate the robustness of the unconstrained modelling approach.CITATION:Queen, L D and Tear, S J, 2015. The Frieda kiss – keeping it simple , in Proceedings PACRIM 2015 Congress, pp 361–368 (The Australasian Institute of Mining and Metallurgy: Melbourne)."

Mar 18, 2015

By J Kuva, M Parian, B I. Pålsson

One of the main factors affecting the suitability of an ore for autogenous grinding is the ore’s competency, ie providing enough critical stones for the grinding process. Several experimental test routines exist and are used for assessing the viability of the ore for autogenous grinding. However, very little attention is given to the generation of fatigue in large stones experiencing repeated shocks in the mill. To investigate this, large pebbles sampled from an industrial autogenous grinding mill running hard and soft ores were categorised based on the grinding energy. From the pebbles, small drill core samples were prepared and went through a series of fatigue cycle tests. Both hard and soft ores showed similar average resistance to failure in compression tests, but the hard ore had a consistent resistance with lower variations. The cores were scanned by XCT before and in-between fatigue tests at the highest reachable resolution, 1.5 µm voxel size. The outcome showed that a higher number of micro-cracks were visible in the soft ore compared to the hard ore. The frequency of micro-crack development in the soft ore may be the reason for its lower specific grinding energy compared to the hard ore. It is obvious that for realistic conditions in an autogenous grinding mill, stones are pre-weakened by fatigue before they fracture. Therefore, shortcut methods focusing only on running tests on fresh and small samples may operate in unrealistic conditions by ignoring the fatigue phenomenon. This is even more important for ores that are on the borderline to be considered as competent for autogenous milling.

Aug 24, 2022

By Bure R

Approximately 2.4 million ounces of gold and 11.3 million ounces of silver have been produced from the Umuna Lode by Misima Mines Pty Limited between June 1989 and the end of Decemeber 1996. Reserves remaining at the end of December were 32.1 million tonnes at 1.0 gram per tonne gold and 9.7 grams per tonne silver. To attain precious metal production of this volume, an estimated K60,000,000 (1996 Kina) was spent on exploration alone. from the 1940s to the present day. Exploration expenditure for the 1996 calendar year was approximately K4,400,000. With such a large investment in exploration, large quantities of data were generated and stored in various forms. To make best use of this data, it was necessary to organise and systematically store this data in a form compatible with a modern Geographical Information System (GIS). The Umuna Lode on Misima Island is a low sulphidation gold- silver deposit dated at 3.5 ¦ 0.2 Ma. The deposit is controlled by the 50 to 100 metre wide Umuna Fault- a moderate dipping, brecciated clay- chlorite- carbonate altered, normal fault zone. There is approximately 230 metres of mostly dip slip offset on this fault, which is measurable by the offset of the Cretaceous schist, marble and greenstone host sequence. Late Miocene microdiorite spatially related to the Umuna Lode does not appear to be genetically related to the gold- silver mineralisation. GIS was introduced to Misima at the end of 1995 and until recently has been mostly used for data storage and retrieval, the over- laying of various data sets to assist in targeting prospective areas, and as a drafting tool. The use of data-driven, predictive data analysis is the most recent effort in maximising the use of the extensive Misima database. By characterising the key controls and properties of the Umuna- type mineralisation, a Weights of Evidence, data-driven analysis produced a prospectivity map of Misima Island. The purpose of this map was to identify prospective areas yet to be explored and to assess the effectiveness of exploration to date.

Jan 1, 1997

By D Will

RedrossThe Redross orebody averages 1 m wide, has a strike of 50 - 350 m, dips at 55° and has a basalt footwall and an ultramafic hanging wall. The most successful and innovative aspects of mining Redross were:•ore development innovations, 90 per cent of development was split fired with a shanty back profile to enable larger and more productive equipment to be used and resulted in higher development grades, improved project payback time and improved ore recovery;•hand held stoping using a multiple level, bottom up ‘V’ shaped hand held stoping method that allowed mining on multiple levels with hand held productivities between 50 and 90 tonnes per shift;•the use of 1 m cement pillars and 7 m high crown pillars were used to create multiple mining areas that enabled high overall production rates from hand held mining this orebody; and•the use of vertical cement pillars to enable hand held mining of adjacent stopes without leaving ore pillars.MarinersThe Mariners orebody averages 3 m wide, has a strike length of 50 - 250 m and dips at 50° with a basalt footwall and poor ultramafic hanging wall. The most successful innovative ideas of re-opening Mariners were:•Dewatering the existing mine workings utilising ‘oilfield’ technology by installing a long submersible pump in on existing pump rising main. This enabled the mine to be dewatered at rates up to 100l/ sec and helped overcome the 45 to 55 L/s inflows.•Overcoming the technical problems associated with Mining N07 ore blocks which had severely deteriorated from being underwater for five years by using reverse firing techniques to improve extraction rates and reduce dilution.•Mass firing a large mining block in the N08 with minimal void space because of deteriorating ground conditions.•Use of a longhole version of the Redross hand held mining method to achieve 100 per cent extraction of the N09 higher grade lenses.

Mar 21, 2011

By M J. Nimmo

extremely powerful tools for building predictive and generative models. ML can be used for building highly accurate regression and classification models. But without careful data science and statistical analysis, a regression or classification model could be highly biased and completely wrong – and we may not even realise. An example rock density estimation task will be used to illustrate the potential gains and possible pitfalls in using ML and AI in Mineral Resource estimation. What could possibly go wrong? Nothing! For most simple regression or classification problems, very little. For density estimation, using the global mean may be all that is possible, but the resource classification would need to reflect the uncertainty in global estimation of tonnage. Everything! From data collection and measurement, from selecting ML and AI techniques and designing the analysis workflow to our own cognitive bias. The things that could go wrong are vast and may include but not limited to asking the wrong question, not asking questions, data bias, measurement errors, insufficient data coverage, not collecting the right variables, not collecting enough observations, collecting clustered samples, unbalanced data, filtering out data without reason (outliers), clipping data (outliers), relying on automatic feature selection, including irrelevant variables, excluding relevant variables, geological interpretation, changing context, blindly following best practice, splitting small data sets, subsetting large data sets, using the wrong tool for the task, insufficient budget to complete the analysis, not allowing enough time for testing and experimentation (the science), focusing too much on building the model (the engineering), not learning from the data, our assumptions, our preconceptions, and our skill. This paper will explore the question of how training regression models to predict rock density are affected by gaps in the data (missing observations and missing variables). What could possibly go wrong?

May 24, 2023

By Q Zhuo, P C. Pistorius, M N. Al-Harbi

A likely increasingly important pathway for future low-carbon ironmaking and steelmaking is the combination of gas-based direct reduction with electric furnace steelmaking (DRI-EAF route), or with electric smelting (DRI-ESF route). In the DRI-EAF route, the gangue that is present in the iron ore is fluxed and removed as slag in the melting (steelmaking) step. This is in contrast with integrated steelmaking, in which the gangue is removed as part of blast furnace ironmaking. For DRI-EAF steelmaking, the amount and composition of slag depend on the iron ore composition, flux additions and iron yield. Slag volume and composition affect phosphorus removal, foaming behaviour and the process energy requirements. In this work, heat data from a year of DRI-EAF production was analysed to test whether the expected relationship between slag volume and electricity consumption for EAF steelmaking was observed. The data include slag analyses (one sample per shift), and heatlevel information on all inputs, final steel temperature and dissolved oxygen concentration, electricity consumption, tapped steel mass and tap-to-tap time. In analysing the data, the slag volume was estimated from the assayed slag CaO concentration and CaO inputs to the EAF. The expected effect of slag volume on steelmaking energy consumption was calculated with a mass and energy balance, considering the heat of mixing of the slag. The calculated energy requirement of slag formation is in line with previous estimates reported in the literature.

Aug 21, 2024

By D Wyborn, P A. Dowd

The characterisation of rock fracture networks is an important component of rock engineering applications involving stability assessment or fluid flow analysis. However, the derivation of a reliable rock fracture model remains a very challenging problem in practice. This paper describes a Bayesian framework, in the form of Markov Chain Monte Carlo (MCMC) simulation, for the construction of such a model. Model conditioning using different data sources is discussed including seismic events recorded during hydraulic fracture stimulation, rock face fracture mapping data and downhole geophysical survey data. The freeware FracSim3D is used for the simulations.

Sep 26, 2011

Mt Muro was a gold and silver mine in Central Kalimantan on the Island of Borneo, Indonesia. Aurora Gold through its Indonesian subsidiary Indo Muro Kencana owned the project. Processing commenced in November 1994, and ceased in June 2002. Mining, including prestripping, commenced in March 1994, and ceased in May 2002. Nominal throughput was 1.5 Mtpa, at average grades of 3.80 g/t Au and 99 g/t Ag. From the start of prestripping until the end of 1996, Drill Blast sampling was the main grade control method. After several studies into the viability of Reverse Circulation, including an assessment of costs and benefits, the switch from Drill Blast to Reverse Circulation was made in late-1996. From 1997 until the end of the project Reverse Circulation was the main grade control method. The mine-to-mill call factors û the reconciliation between predicted and actual tonnes and grades delivered to the mill û for the period from commencement of processing to December 1996 were 1.069 for tonnes, 0.949 for gold, and 0.969 for silver. The mine-to-mill call factors between January 1997 and June 2002 (excluding a period of significant illegal mining) are 1.086 for tonnes, 0.977 for gold and 1.025 for silver. At face value this indicates a 2.8 per cent absolute improvement for gold, and 5.6 per cent for silver. Using average grades for the project, an additional 5112 oz of gold, and 267 173 oz of silver were delivered annually to the mill using Reverse Circulation sampling instead of Drill Blast sampling. The additional recovered metal value is estimated at $US 2.87 M annually, significantly exceeding the cost of Reverse Circulation drilling. This paper discusses a number of issues that may affect the estimated benefits.

Jan 1, 2003

By S Kost, R Nievergelt, F Cortes

In recent years, the use of steel fibre-reinforced concrete (SFRC) for segmental tunnel linings has become increasingly popular due to numerous advantages including durability, resistance to spalling, reduced cracking at corners and edges and significant cost-saving potential. One of the key parameters for measuring the performance of SFRC is the residual (post crack) flexural strength. A problem often faced in design and quality control of SFRC is the high variability of results obtained by standard beam test methods, with reported coefficients of variability (CoV) in the order of 25 per cent. The high test variability results in a relatively low permissible design strength relative to the average flexural strength of the material and commonly leads to non-conformances with respect to residual flexural strength during production testing.This paper presents the implementation of an alternative testing method using large round determinate panels (RDPs) for quality control testing of the SFRC used in the segmental lining of the City East Cable Tunnel (CECT) in Sydney. The alternative RDP test method was found to correlate well with standard beam tests and significantly reduce the variability of test results, minimising the risk of outlying residual flexural strength results, with zero non-compliant test results reported during the production period.This successful implementation of an alternative test for post-crack flexural strength with lower variability in results constitutes a relevant precedent to demonstrate the feasibility of this approach which contributes to the development of more cost-effective concrete mixes and minimises the risk for non-compliant results and the contractual disputes that commonly arise from these.CITATION:Cortes, F, Nievergelt, R and Kost, S, 2014. Development of a segmental lining specification for steel fibre reinforced concrete flexural strength testing using round panel tests, in Proceedings 15th Australasian Tunnelling Conference 2014 , pp 461–468 (The Australasian Institute of Mining and Metallurgy: Melbourne).

Sep 17, 2014

By C Xuan, J H. Park, Q Wang, C Shen, D Kumar, W Mu

Dissolution of non-metallic inclusions in the metallurgical slag is of vital importance for cleanliness control of the steel manufacturing. With the development of high temperature confocal laser scanning microscope (HT-CLSM), new insights have been obtained due to its in situ observation characteristics, higher resolution and precise control. However, HT-CLSM measurement has the limitation, eg the slag composition cannot include high amount of transition metal oxides. In addition, it is time consuming for the experimental procedure and not so simple to succeed for every measurement. It is known that digitalisation has made a significant progress in recent years. Machine learning (ML), a sub-domain of artificial intelligence (AI), is the key enabling technology for the digitalisation of the material science and industry. The database for ML model is collected using almost all available HT-CLSM experimental data and subsequently the established database is trained by different ML methods. Unseen data is used as the benchmark of the ML model. Al2O3 dissolution is the main process to be predicted in the current study. A good agreement between the HT-CLSM data and the ML model prediction results show the possibility to apply ML in process metallurgy.

Jun 19, 2024

There are many compelling reasons why correct sampling schemes should be standard practice in the mining industry, yet there are numerous examples where poor sampling practices are known to exist in the business yet permitted to continue. It is not just technical issues that can be ascribed to the reasons for the continued use of incorrect methods. This paper proposes non-technical reasons why incorrect sampling practices continue, and identifies potential ways to overcome them.

Mar 1, 2010

By J Wen, P Hou, W Yu, M Hu, Z Jiang, C Bai

Vanadium titano-magnetite (VTM) is one of the important iron ore resources in the world. High viscosity and low fluidity are the basic characteristics of hot metal containing vanadium and titanium during the ironmaking process. Clarifying the influencing mechanism of the melt viscosity would be a significant issue for the high efficiency smelting of VTM. In this paper, a self-developed melt viscometer based on the principle of torsional vibration theory was employed to measure the viscosity of hot metal accurately. Simultaneously, the thermodynamic calculations and some modern detection methods were also used to investigate the precipitation behaviour of high-melting-point phases in hot metal containing vanadium and titanium. The measuring results show that the viscosity of molten iron gradually increases with the increase of titanium content when the mass fraction of titanium is between 0.09 wt per cent – 0.50 wt per cent. Meanwhile, the viscosity of the hot metal was affected significantly by the vanadium content. When the vanadium mass fraction was higher than 0.30 wt per cent, the viscosity of hot metal increased sharply, which will have a serious impact on the fluidity of the hot metal. Furthermore, it was revealed by thermodynamic calculations that an increase in titanium and vanadium content would result in the precipitation of high melting point phases such as TiC, TiN, VC, and VN in the molten iron, ultimately resulting in a decrease in its fluidity. Meanwhile, it was confirmed by Scanning Electron Microscope (SEM) observation that the precipitates in the hot metal were compounds consisting of V, Ti, C, and N elements, which agrees well with the thermodynamic analysis results.

Aug 21, 2024

By K M. P Singh, K M. K Sinha, G S. Jha, T Gouricharan, K K. Sharma

The majority of the powerplants in India are fed with non-coking coals, with the coals dispatched from their respective mines via wagons loaded by a rapid loading system or pay loader. The usual top size of coal is 200 mm and sampling is carried out manually from wagons. The manual method of wagon top sampling of large size raw coal is not only difficult, but also violates some of the fundamental principles of sampling. As per sampling requirements, samples are to be drawn from the full depth of the wagons, which is impossible to be done manually. Since the ash distribution in the different size fractions is not homogeneous, the results from the samples, which do not reflect the true size distribution of the lot, are likely to be biased. More importantly, sample collection by shovel from the top is a function of human discretion and not governed by the equiprobability rule. In another case, the sample has been collected by a cross belt-type auto mechanical sampler for 200 mm size coal at one of the super thermal power stations. Test analysis shows that coal collected using a cross belt-type mechanical sampler also has significant deviations from stopped belt coal samples of 200 mm size due to size segregation. Studies carried out on wagon top sampling through specially designed experiments reveal some interesting features. The presence of bias in wagon top sampling is apparent when compared with the ‘stopped belt’ sample. In this paper, a comparative study between manual wagon top sampling/auto mechanical sampling with respect to stopped belt sampling has been conducted for coal of size 200 mm. In addition, precision studies have been performed for wagon top sampling/mechanical sampling for Indian thermal powerplants.CITATION:Sinha, K M K, Sharma, K K, Jha, G S, Singh, K M P and Gouricharan, T, 2014. A comparative study of manual wagon top sampling and auto mechanical sampling of 200 mm size coal with respect to stopped belt sampling of thermal coal at Indian super thermal powerplants, in Proceedings Sampling 2014 , pp 43–48 (The Australasian Institute of Mining and Metallurgy: Melbourne).

Jul 29, 2014

By M Alston

Many organisations struggle to yield meaningful outcomes from incident investigations. This paper will demonstrate that we can discover systemic issues by rethinking the ‘Why’ and ‘How’ of investigations, fostering genuine risk reduction. Key points: 1. Shift from ‘prevent recurrence’ to ‘learn and reduce risk’: conventional investigations focus narrowly on recurrence prevention, resulting in narrowly focused investigations with limited ability to identify and reduce organisational risk. Redirecting our objective to learning and improvement opens avenues for broader risk mitigation. 2. Understanding daily work for risk insights: recognise that work is not always completed in full compliance with procedures. Workers will adapt how they complete tasks influenced by the conditions and management systems that exist every day. Understanding the conditions and systems affecting everyday work is vital for uncovering systemic risks. 3. Unearth systemic issues through in-depth exploration: many investigations address surface concerns such as lack of training, incorrect procedures, equipment not being maintained, and missing overarching organisational risks. Systemic, organisation-wide risks are not identified. To discover these systemic issues and create shared learnings, investigators must question management systems, including resourcing, planning and risk management. 4. Design actions for safe failure: common actions like discipline and training fall short and do not reduce risk. Resilience emerges from well-designed controls. Incorporate planning, better resources, interlocks, and guarding to bridge the gap between task execution and risk. 5. Investigation Reports as a Business Case: Investigation reports should not simply state historical information regarding an incident. They are to facilitate decision-making and be presented as a business case to drive change. Organisations can move beyond localised risk and ineffective administrative actions by changing investigation objectives and methods. They can facilitate learning, understand daily work intricacies, identify systemic issues, and enhance the capacity for safety, driving lasting risk reduction.

Apr 16, 2024

By A Bennett, S Arndt

Minerals and Metals Group Pty Ltd (Golden Grove Mine) commissioned a comprehensive study of the life of mine using advanced 3D stress modelling. Essential input into the modelling was the proposed mining sequence, itself the culmination of years of planning and geotechnical experience gained in comparing actual performance to predictions. The modelling approach addressed the whole of mine at a scale of 5 km. There were however several magnitudes of scale to address, from the scale of laboratory testing (0.05 m), to drive size (5 m), extraction size (50 m) and stope size (500 m). Scaling was therefore an integral part of the procedure. The constitutive rock properties in terms of peak and post-peak behaviour varied with the scale. At the laboratory scale, the constitutive relationships for all the main rock units were determined in 3D as well as recovering the in situ stress using the rock memory technique, deformation rate analysis (DRA), whereas at the larger scale the properties were varied to match observations.The modelling, using non-linear analysis to account for rock mass yield and peak-residual strength, demonstrated that if the planned extraction sequence and stope geometry had been pursued there was a high probability of large volumes of excess shear stress being released. However, it was not possible to identify whether the excess shear stress (ESS) will release as one large event or many smaller events. The mining approach was to modify the extraction sequence/stope geometry to manage the hazard. This is where the economic advantage can be realised.Alternatively or in combination with further modelling, the volume identified with the predicted high ESS or energy release could be intensively explored for structures having adverse orientation to the stress change and with sufficient areal extent to release the ESS. The microseismic data can be interrogated for the likelihood of a large event to support the hypothesis of a singular release versus multiple releases of smaller events. If the initial planned extraction sequence had to proceed as designed, changes to dimensioned stopes would be recommended for risk mitigation. Mine planning could as an option, modify the sequence and the extraction could then be remodelled.CITATION:Bennett, A, Arndt, S and Dight, P M, 2014. The economic advantages of detailed geotechnical assessments for extraction sequencing, in Proceedings 12th AusIMM Underground Operators’ Conference 2014 , pp 243–250 (The Australasian Institute of Mining and Metallurgy: Melbourne).

Mar 24, 2014

By R Castedo, P Segarra, I del Castillo, L M. López, J A. Sanchidrián

This work investigates some suggested methods for blast vibration monitoring on a hard surface based on 53 tests on a vibration exciter. Errors in the measurement attributable to coupling have been described as function of frequency by the rock-to-mount (or coupling) transmissibility, ie ratio of the velocity of the vibration sensor mount to the velocity of the rock motion. Transmissibility has been measured from 2 to 200 Hz, at two vibration levels (5 and 20 mm/s), in four seismographs. Attachment methods with an anchor, thermal adhesive and gypsum plaster allow stiff rock-to-mount coupling. They ensure accurate measurements (errors below 0.3 dB) for frequencies commonly found in blasting (below 100 Hz) irrespective of the mount characteristics and the vibration level. Free laid mounts and sandbagged vibration sensors lead generally to a poor, weak coupling. Errors are non-linear errors with the input velocity, ie the operating frequency window narrows as the velocity of the ground increases. Free mounts follow ground motion acceptably well (error up to 0.4 dB) at both vibration levels and frequencies below 15 Hz. At higher frequencies, free mounts lead to the largest distortions with expected errors ranging from 0.8 to 6.6 dB. Sandbagging limits the expected errors below 1.3 dB, but it does not prevent large errors that can be up to 4.5 dB. Both frequency and velocity should be considered together to assess the measurement errors instead of the expected accelerations, as it is currently made.CITATION:Segarra, P, Sanchidrián, J A, Castedo, R, López, L M and del Castillo, I, 2015. Effect of vibration sensor mounting method on the quality of vibration measurements, in Proceedings 11th International Symposium on Rock Fragmentation by Blasting, pp 545–552 (The Australasian Institute of Mining and Metallurgy: Melbourne).

Aug 24, 2015

By V Yadi

Geostatistics has been widely used for more than four decades as a powerful tool in ore resource/reserve estimates and grade control. This method has gradually replaced the conventional method such as polygon and cross-section for grade estimation by more mathematical techniques such as kriging. The advances in geostatistical understanding and the availability of appropriate software packages have given much more contribution in speed and high accuracy during production. However, in many cases, these techniques are sometime applied without support with a good thorough understanding of the geological controls on metal distribution. To produce the result close to actual, the geological control in interpretation and appropriate geostatistical task is a must. Geological logging was described in the core must be consistent with the level plan produced by commercial modelling software used as a guide for grade control during production. Experience gained through evaluation of the Central I deposit resulted that the most effective systems continue to be based on geostatistical methods of resource estimation. Computer applications modelling software packages are necessarily as the core of an efficient in undertaking these assignments. They contribute to make a valued to data management and are becoming increasingly effective during processing. The latter role is particularly important to check the actual versus estimated values where will significantly to improve personnel to have a confident with their knowledge combined with experience and complete their jobs satisfy.

Aug 22, 2011

By Marshall GA, Dickerson AW

The efficient operation of large haul trucks in large open pit mines requires an economic trade off between the haul route quality, the haul truck payloads and the level of haul truck reliability. This paper describes an investigation into the relation- ship between haul route surface quality and the level of structural reliability of a haul truck main frame. An instrumentation system for unattended recording of data on a haul truck operating normally in a mine and com- puter based analytical procedures were developed and used to identify which road segments contribute significantly to fatigue cracking at various sections of a haul truck main frame. INTRODUCTION In many open pit mines there is exten- sive use of large haul trucks to move over- burden and minerals. In most open pit hard rock mines, both the overburden and mineral ore are moved by large rear dump type haul trucks. Similar haul trucks are also widely used for coal overburden and mine site coal haulage operations in New South Wales, South Australia and Western Australia. Large bottom

Jan 1, 1982

By I Canbulat, J Emery, L Yao and C Zhang

This paper documents a process of data analytics and machine learning applied to coalmine roof convergence data. It follows a publication documenting the formation of the database. For each set of analyses Python scripts were utilised to extract data from an SQL database, with subsequent analyses also executed in Python, utilising open-source tools such as scikit-learn. Preliminary analyses with classic machine learning algorithms included multiple logistic regression, K-nearest neighbours, decision tree, random forest, support vector machine, kernel support vector machine, artificial neural network (ANN), naïve Bayes and multiple linear regression. Results indicated that relationships were complex and non-linear, with the artificial neural network most suitable. An ANN model was then refined through a standard process of hyperparameter optimisation and data augmentation to arrive at a final model. Due to the black box nature of the ANN further insight into how the parameters interacted was sought and found with Shapley additive explanations (SHAP values). SHAP values utilise Game Theory, where each input parameter or feature is a ‘player’ while the data set is the ‘team’. The SHAP value is the impact of each player on the target value, essentially establishing the contribution of each input parameter. The results from all analyses are delivered in the paper, together with insights, applications, strengths, and limitations of the methodology.

Nov 29, 2022

By P Luo, J E. Danoczi, P Nakutnyy

Sensor-based sorting methods to remove waste or for upgrading ore prior to downstream metallurgical processing is becoming more prevalent for mining operations. X-ray transmission (XRT) sorters are a popular choice for mineral processing plants since they can sort a wide variety of minerals. XRT sorters pass high intensity X-rays through particles to generate images of varying greyscale, based on their effective atomic density. The ore sorter then mechanically separates the particles using algorithms based on the images. A popular test currently used to evaluate the amenability of the ore to XRT sorting is to pass the particles through the sorter and obtain the greyscale images showing mineralogical differences. This test provides information on the mineralogical heterogeneity between ore particles for their separation. However, it does not provide an understanding of the actual mineralogical composition of the ore or information pertaining to which other minerals may respond in a similar manner, nor does it provide liberation information. This paper describes a test that combines high resolution X-ray micro Computer Tomography (CT) combined with QEMScan to obtain information about amenability of ore to XRT sorting and to obtain valuable information that can streamline metallurgical testwork programs. This test also provides information relating to particle thickness and associated X-ray attenuation coefficients, desired mineral presence and mineralogical information relating to associated minerals and clays to assist with developing sorting algorithms. Combining the CT and QEMScan can address many limitations of the current method to determine whether XRT sorting technology is appropriate for particular ores. Mineralogy information from QEMScan can identify incidences of; mineral encapsulation, problematic clays, detrimental minerals as well as calibrate the greyscale values of CT data. Additional 3D CT imaging carried out on cores provides spatial information of the minerals within the ore and mineral grades.

Nov 10, 2020