Identifying steady-state operation in Semi-autogenous Grinding (SAG) mills is crucial in process control for optimising grinding circuit performance and reducing operational disturbances. It also enables evaluation of control system performance and identifying periods of process stability suitable for performing plant surveys and applying steady state process models. Current practice relies on visual trends to assess steady states in SAG mills, with no formal multivariate method for automated detection.,This study introduces a steady state identification technique based on multivariate Dynamic Time Warping (DTW) in a sliding window framework. The methodology identifies periods of steady state operation in SAG mills across various scenarios, and is validated using data from two processing plants and sampling resolutions. The method uses findings from previous studies to set parameters such as window size based on mill residence time. The main advantage of the proposed methodology is that it does not require extensive pre-processing, such as noise filtering, missing value imputation, or removing stoppage conditions. Results show that the approach accurately detects steady-state periods, classifying approximately 63% of operations as steady across both sites. This generalisable methodology offers a practical foundation for real-time monitoring and future autonomous plant control.

This study explores the application of machine learning techniques for predicting generic mill liner wear in semi-autogenous grinding (SAG) mills used in mineral processing. Various models were developed and compared using data from 143 liner measurements across 36 liner cycles from ten different SAG mills. The research initially focused on individual mill modeling, employing simple linear regression, first-order kinetic approach, Multiple Linear Regression (MLR), tree-based methods (Decision Trees, Random Forests, XGBoost), and Multilayer Perceptron (MLP). Results showed that simple linear regression provided sufficient accuracy, with other methods only slightly improving performance. This study then developed a combined model using data from multiple mills. MLR and advanced machine learning techniques were applied for this generic model, with XGBoost emerging as the most successful. In the interpolation scenario involving a mill similar to those in the training data, the XGBoost model achieved a mean absolute percentage error (MAPE) of 5.27%. For the extrapolation scenario, with a mill larger than those in the training set, the MAPE increased slightly to 6.12%. These results demonstrate the potential of machine learning approaches in creating effective generic models for mill liner wear prediction. However, this study also highlights the potential for improving predictive models by incorporating additional key parameters such as liner and ball material properties.

This paper describes a 3D cellular automaton (CA) for dynamically modelling ore piles with continuous feeding and discharging and incorporating two separate size segregation mechanisms. Stockpiles are an integral part of materials handling and storage, and their operation plays an important role in the overall performance of the mineral processing plant. Size segregation can create fluctuations and cause processing units to receive uneven feeds. However, this issue has not received enough attention. Therefore, a three-dimensional dynamic stockpile model has been developed at the Julius Kruttschnitt Mineral Research Centre (JKMRC), which can model the dynamic response of a wide-size-range stockpile. A Continuous Cellular Automata was developed, dividing the volume of the stockpile into a three-dimensional grid of cells with each cell containing an independent set of properties that are tracked throughout the simulation. Modelling size segregation during material flow was presented in Part 1. The variation in the stockpile surface profile and size distribution of the stockpile can be predicted. In this paper, the stockpile model with size segregation extends to the discharging and migration for real-time simulation. The model structure and industrial validation are discussed in this paper, which will show the application of the dynamic model, including the particle size distribution and the height variation of the stockpile. The three-dimensional dynamic stockpile model can contribute to the control of the comminution circuit because different operational strategies can be predicted using the model.

Size segregation is an unavoidable problem in materials handling and storage. The size segregation in a stockpile will send different particle sizes to different locations in the stockpile. This phenomenon will result in the comminution process receiving varying feed sizes depending on the feeders' layout. The fluctuations in feed size will then impact the performance of comminution and downstream processes. Therefore, qualifying size segregation is an important topic for any research related to size segregation.Whilst size segregation is a recognised factor in operations, there is no mature index to quantify the degree of size segregation for multi-sized materials. In this research, after reviewing the existing size segregation indices and identifying their strengths and limitations, two novel-sized segregation indices are proposed for quantifying size segregation in stockpiles and bins. The proposed indices are based on the variation in the particle size distribution of different locations of a stockpile. The first index is an improvement to an existing index (Li et al., 2017). The authors propose the second index and it is based on the differences in the particle size distribution curves at each section of the stockpile or bin. A number of small-scale experiments have been conducted in the laboratory to investigate the suitability of the proposed indices and their advantages over the existing indices. These experimental results validate that the proposed indices are suitable for quantifying size segregation, and they perform better than the existing indices. As a result, the new size segregation indices can be utilised to quantify the segregation in stockpiles, which can then be used to model and predictor inform the downstream process for reacting and optimising the process.

This paper describes a 3D cellular automaton (CA) for modelling ore pile formation that incorporates size segregation due to surface stratification. Ore stockpiles and bins are essential in mining operations as a buffer between the mine and the mineral processing plant, and their operation plays an important role in the overall performance of the downstream equipment. Size segregation can occur if the feed size to the pile varies over time or due to a variety of segregation mechanisms occurring in the pile itself. In particular,it can occur as a result of percolation stratification within surface flows. The structure of the newly developed CA model is described, and the simulated surface profile and size segregation response is validated through a series of laboratory-scale piles for characterizing the segregation potential of the feed particles. It is found that the model adequately describes the segregation behaviour at a range of model scales where coarse particles roll to the outside of the pile and fine particles are concentrated around the centre of the pile. Moreover, it is found that the model is sufficiently fast to use in real-time applications such as dynamic process control and digital twins.

Most mine sites use large coarse ore stockpiles as a buffer between the mine and the processing plant. Stable operation of the downstream processes demands a uniform output from the stockpile with respect to particle size, however, particle size distributions of the product drawn from stockpiles can vary significantly over time due to various size segregation phenomena. Although quantifying size segregation occurring in industrial stockpiles would provide valuable information for operators, it is not practical due to the size of these piles which can exceed 100 kt. This research is focused on developing a laboratory test, which aims to quantify the propensity of an ore pile to segregate and to correlate these results to industrial stockpiles. The data generated by the laboratory-scale experiment could enable the modelling of size segregation for industrial-scale stockpiles. The results of the comprehensive laboratory tests indicate that it is possible to quantify size segregation in the laboratory and scale up the results.

In mineral processing, accurate characterisation of the mechanical properties of particles of various mineral composite, texture and scale is critical for the development of improved breakage modelling. Additionally, these models should emphasise the primary breakage properties of the particles on the models rather than breakage in test devices that introduce substantial secondary fragmentation events. This work explores the use of the Short Impact Load Cell (SILC) test to investigate the mechanical properties of fabricated 3D-printed (3DP) specimens of single and binary mineral composition, i.e., iron oxide, silica, and layered specimens with both minerals. The fabricated quasi-identical specimens are useful to explore the repeatability and contrast of controlled additively manufactured 3DP specimens and the SILC testing performance. The fracture characteristics of printed quasiidentical specimens were observed using an ultra-high-speed digital camera. The study showed that the specimen properties, such as tensile strength and fracture energy are strongly influenced by the number of beds, bed thickness, and mineral composition. The force-deformation as well as force-time profiles and specimen fragmentation are studied to understand better the variability of the results and the specimen's physical response to a single impact. The contour of force-time profiles allows for interpretations of how the striker contacts the specimen and to infer the crack initiation and fracture propagation through beds of brittle or ductile material. The use of hierarchical clusters facilitates the analysis as they enhance contrasts and give more insight into breakage and fragmentation, which is worthwhile investigating for natural rocks in the future.

The fraction of mill volume occupied by rocks, grinding media and slurry in grinding mills are dominant factors influencing AG and SAG mill power draw and grinding rate. Hence accurate mill filling measurements are needed for modelling mills. To that end, it is usually necessary to enter the confined space of a grinding mill, a task that involves a degree of safety risk and a potentially lengthy plant stoppage. This paper discusses a range of available methods for measuring and calculating mill filling and introduces a new technique for accurately measuring filling that eliminates the need to enter the mill in most cases. The new method offers engineers and researchers a safer and faster method for quantifying the filling level in industrial SAG and ball mills, and allows the surface profile to be surveyed, without requiring the contents to be level. The paper also details formulas for calculating the filling level and load volume, which unlike previously published equations, also consider the volume in the conical ends and the volume occupied by the mill shell lifters.

A mechanistic model for tumbling mills was developed based on breakage characteristics and tumbling mill operational features. The concept was presented at the IMPC (International Mineral Processing Congress) 2014, followed by progress in a sub-process of the model presented at the IMPC 2016. Additionally, a number of papers on the sub-models and breakage function have been published. This paper provides a consolidated summary of the outcomes and status of the model. The overall model structure is presented along with the sub-models such as appearance functions, breakage rate functions, energy distribution, transport, and dual component grinding interaction model. The strengths and capabilities of the model structure as achieved to date are presented. The approach developed can be used as a platform for building multicomponent models. The modelling work can be done quicker by using an existing structure such as the one presented in this paper. It is recommended that researchers assess compatibility prior to embarking on model development work if the intention is to use this model structure. (c) 2021 Elsevier B.V. All rights reserved.

Automation is a critical element for the sustainability of mining operations at the process level. However, robust and reliable advanced process control remains the main objective to achieve for many mineral processing plants. This major limitation is despite advances in digital technologies and the industry's gradual uptake of the Internet of Things (IoT) which made it possible to measure and monitor a wide range of parameters in operations. The uncertainty on measurement makes it challenging to develop robust and reliable advanced control systems. Thousands of sensors and instruments are installed on mining and mineral processing equipment to measure critical parameters for monitoring and control. Although there is a wide range of operating parameters measured directly using sensors and instruments, there are parameters that cannot be measured practically due to the measurement's complexity, lack of appropriate access or harsh environment for measurement. Furthermore, the reliability of measurements provided by some sensors is low. Soft sensors are software that use the measurements of parameters from existing instrumentation to calculate a parameter that is not practical to measure physically. With advances in mathematical modelling and computation power, particularly edge computing, and the need for redundancy to enhance measurement reliability, accurate measurement of parameters that are complicated to measure physically but are essential for advanced control, the development and use of soft sensors are growing. This paper focuses on the role of soft sensors to enhance reliability and accuracy of measurement and enable advanced process control for mineral processing plants. The development and implementation of the Julius Kruttschnitt Minerals Research Centre (JKMRC) Mill Filling Inference Tool (Mill FIT) is presented in this paper. These examples demonstrate the value of integrating expert knowledge as mathematical models with the data that operating plans generate daily to measure key operating parameters accurately.

Stockpiles are an integral part of mineral processing plants, and their operation plays a significant role in plant performance. However, it usually does not get enough attention from plant engineers. Size segregation is a common issue in most storage and materials handling facilities, and it can have a severe impact on the performance of downstream processes. Size segregation will create fluctuations and cause the downstream equipment to receive uneven feeds. A dynamic stockpile model incorporating size segregation has been developed at the Julius Kruttschnitt Research Centre (JKMRC), which can model the dynamic response of stockpiles, including the stockpile height variation and size distribution of feeders while dynamically varying feed and discharge rates. This model is important for enabling operators to enhance the operation and control of stockpiles and minimize the impact of size segregation on the performance of downstream units. A typical stockpile design with an asymmetric layout is studied and is shown to inherently result in a segregated discharge. Hence three industrial strategies have been simulated to investigate ways of reducing segregation. The simulation results provide ideas for future stockpile designs and operations. The application of this model also contributes to the process control of the comminution circuit. This paper will increase the understanding of how to run a stockpile in the industry and will show the importance of modelling materials handling for process control and optimization.

In fine grinding applications, gravity induced stirred mills have a demonstrated higher energy efficiency compared to conventional tumbling ball mills, leading to their increasing utilisation in unlocking value from fine grained and low-grade ore deposits. The efficiency of the current strategy for compensating for wear of the agitator screw, by maintaining constant power draw through increasing mill media filling during the liner lifecycle, has not been evaluated. This paper provides a laboratory evaluation of the alternative strategy of maintaining constant mill filling in batch wet grinding. The comparison is based on the size specific energy (SSE) approach. It was established that the grinding efficiency improves when maintaining a constant filling, whereas it decreases when running at a fixed mill power. These results encourage continuing work with continuous and full-scale mills, and potentially utilising increasing mill speed to compensate for the loss of power draw, and thus throughput, with screw wear.

Many of the existing liberation models combine a model of texture that describes the intact ore structure with a model of particle production. The Geometric Texture Model (GTM) proposed by the authors calculates particle compositions for describing multi-mineral liberation distributions, but these can also be used to generate particles with representative composition distributions to feed particle-based models of various unit processes. In this paper, a model of a meso-texture from the George Fisher deposit is used in combination with a particle-based model that responds to surface composition of particles to predict the overall flotation response of the ore.

Large coarse ore stockpiles are very common in mine sites. The uniform output from the stockpile is essential for stable operation in the downstream equipment and the whole comminution circuit. However, due to size segregation in stockpiles which is inevitable, the particle size distribution of stockpile product can vary significantly. Therefore, it is important to understand size segregation in materials handling and storage units. However, it is very difficult to quantify size segregation of an industrial stockpile due to the size of the stockpile. Therefore, this research has focused on developing a laboratory experiment which allows quantifying the size segregation of stockpiles and extends the results to industrial scale. The result of preliminary lab tests indicated that it is possible to quantify size segregation in a laboratory and to scale up the results.

In expanding the mine to process considerably more competent ore sources, this semi-autogenous-ball mill-crusher (SABC) circuit with a single ball mill is not just throughput constrained but will shift to being permanently ball mill limited. The application of a fully integrated processing objective that relies on close cooperation between mine, dispatch, and mill is required to address this challenge. Moving beyond the general perception of mine–to-mill, a deeper processing knowledge is applied along the mining chain, considering blasting as the first stage of comminution and recovery. Grade deportment and dilution are considered at the mining stage and modelled with the new Sustainable Minerals Institute (SMI) blast movement simulator, linking with the block model data. Based on field trials, blast design and blending strategies are developed to couple with new operating strategies at the mill.It has been found that accounting for blast movement for the higher intensity blasts could generate additional value of over $1 million per high-intensity blast. Strategies to shift the workload and debottleneck the milling circuit were proposed and proven during the milling trials, demonstrating an increase in throughput of 16% is achievable. A number of process improvement opportunities, including changing the semi-autogenous (SAG) mill control strategy, have been identified to enhance current productivity and ensure long-term capability to process the considerably more competent future ores. In a departure from traditional once-off applications of mine-to-mill changes, on-site technology transfer is being embedded in online tools to sustain advanced mine-to-mill capability in the daily planning and operation.

Comminution is the most energy intensive process in mining industry, due to its low efficiency. Comminution efficiency decreases with particle size, providing a great incentive for the optimisation of fine grinding applications. Related to this issue, the Size Specific Energy (SSE) approach is a technique that provides a better evaluation of comminution efficiency, once it takes in account the generation of fines and it is less dependent on particle shape. Based on this, the SSE can be used as a benchmarking for fine grinding evaluation. The aim of this paper is to apply the SSE approach to the evaluation of grinding efficiency of gravity induced stirred mills. Batch grinding experiments were conducted in a laboratory gravity induced stirred mill using an iron ore sample with a top size of 0,180mm.

The new Aitik autogenous grinding (AG) milling circuit was based on the successful original milling plant at Aitik, with a focus on low life-of-mine operating costs. At the heart of this is eliminating steel grinding media, maximizing the use of gravity flow, and using spiral classifiersinstead of hydrocyclones. The two 22.5 MW AG mills, 11.6 metres (m) (38 feet [ft]) diameter by 13.7 m (45 ft) long, are the largest mills by volume operating in the world. Each AG mill feeds a 10 MW pebble mill via a unique coupling of flow and pebble feed with the pebble mill product recycling to the spiral classifier that is closed with the AG mill. The massive size of the AF circuit and unique layout make this an interesting case study in alternative plant layout and operation compared to the strong trend towards finer feed, high ball load and multiple ball mills per semi-autogenous grinding (SAG) mill. Operating capability, data on specific energy, throughput, and plant stability are presented as a comparative case study.

In the mining industry, stockpiles and bins are very common, and their design can play a significant role in circuit performance. A typical dry comminution circuit includes crushing, screening, stockpiles, bins, conveyors and transfer chutes. Design of stockpiles and bins plays an important role in the performance of the comminution process. However, in process modelling and optimisation, materials handling and storage does not receive as much attention as comminution and classification units. Size segregation, which means particles with different sizes seating apart in a bin or stockpile, is an important phenomenon in the materials handling process. In mineral processing, the size segregation occurs throughout the whole process from the load and haul at the mine to waste disposal. However, this issue in bins and stockpiles, in particular, during the filling and discharging affects the performance of downstream processes. Therefore, modelling the size segregation is important for addressing operational challenges. This paper presents the structure of a 3-D bin and stockpile model, which is developed to model the size segregation. The model is validated using data from a laboratory scale test. The structure and details of materials storage models with size segregation and the application of these model are presented in this paper. The simulation of the size segregation phenomena could improve the understanding of materials flow and help designers to optimise the design of equipment and circuits.

Semi-autogenous mills in general are fed from a coarse ore stockpile. Providing a uniform feed from the stockpile is essential to maintain a stable performance of the SAG mill and consequently downstream equipment. However, due to size segregation in stockpiles, the particle size from stockpile draw points can vary. The size distribution of feed to the SAG mill could change over the time depending on which feeders are running, the location of feeders, and the level of material at different section of the stockpile. Therefore, simulation of stockpile and SAG mills without incorporating size segregation will not accurately represent the operational performance of grinding circuits because in reality particle size distribution varies over time. A model of a stockpile that incorporates size segregation is presented. This model can be used to simulate different feeding conditions from the stockpile to the SAG mill and provide an opportunity for optimization. The simulation result!s are compared to experimental data from a small-scale stockpile tests in different situations. The results of the stockpile simulations have been used to evaluate the performance of the SAG mill using JKSimMet. Comparison with steady-state data shows the impact of size segregation in stockpiles on the performance of the SAG mill. The results illustrate that the size segregation of stockpile will increase the throughput variance by 404% over 100h of operation. In this example, simulation results indicated that implementing a control strategy to maintain the size distribution of the feed to the SAG mill will increase the throughput by 922 t/h. This paper demonstrates the need for dynamic models of stockpile and materials handling with size segregation to quantify the impact of size segregation on performance of grinding circuits.

The paper describes a methodology to simulate product particle size distribution of an industrial scale VertiMill (R) engaged in regrinding duty. Using survey data, the time-based population balance model has been utilized to simulate mill product particle size distribution. A sub-model relating mill power to particle residence time is developed and applied to the population balance model to predict mill response in different operating conditions. The result showed that the new time-based population balance model applied to the VertiMill (R) is capable of predicting product particle size distribution with a change in mill power, feed size distribution, mill feed rate, and slurry solids concentration.