Search Documents

By I. Soto, E. Córdova, A. Anani

The new caving megaprojects that are planned to go into production in the next decade have scheduled horizontal developments at very high rates, which are difficult to achieve. Research has been conducted that seeks to model the construction times for underground developments to determine the feasibility of achieving the planned rates. However, these previous attempts do not consider all the operational aspects that make up the mine development cycle. Here, we present a discrete event simulation (DES) model that individualizes each unit operation of the development cycle to determine the construction rates that new projects will be able to attain. The developed model was applied to the New Mine Level (NML) of the El Teniente Division (DET). The results indicate that it would be impractical to achieve the advance rates scheduled in the original plans if the unit operations of the mine continue to be executed using the status quo. The metres developed after six years in the sections analysed can be significantly improved by changing the maintenance strategy for the equipment (457 metres increase), orepass availability (194 metres increase), and the initial planning sequence (1 598 metres increase). It is recommended that the mine implements a flexible maintenance strategy, defines a strategy to ensure and increase orepass availability to contractors, and makes use of current simulation tools for a more robust development sequence planning.

Oct 1, 2022

By N. Theron, R. F. Meeser

The global emphasis on conserving energy resources has led to the adoption of hybrid power systems in vehicles. Optimally applied, hybrid systems can save up to 40% on fuel costs. To optimally manage a hybrid vehicle's energy flow it is necessary to know, in real time, all the energy requirements to complete a given route. The energy consumption depends mainly on the vehicle mass, speed profile, and route topography. Of these, the topographic profile is the one factor that is only route-dependent and not influenced by the vehicle or driving styles. The heading vs. distance profile is also an example of a route characteristic not influenced by the vehicle or driving style. In this study the topographic and heading profiles are used to identify routes, and are easily measured by means of digital barometric pressure and compass sensors. Correlations between the current route and previously travelled/stored routes are performed based on their topographic and heading profiles in a point-by-point manner. Above-ground tests were first performed using a road vehicle and six routes to evaluate the system. The system consistently and correctly identified a 20 km route within the first 4 km. It also proved to function correctly in underground tests through the implementation of an instrumented hand-held surveyor's wheel. This system finds direct practical application in optimizing the energy management of an underground hybrid locomotive used by the mining industry in South Africa, but can also be of benefit in applications where route identification is required and using GPS is not feasible.

Jan 26, 2023

By N. Tatalias, P. Ilunga, M. Lampi, J. Edwards, K. Valkama, M. Nicolle, D. Samal

"Developing a greenfield mining project is a complex exercise: doing so in an increasingly volatile, uncertain, and ambiguous global economy adds significantly to this complexity. In the context of an elevated cobalt price, this paper outlines the metallurgical and process development of the Mutoshi copper–cobalt project for copper cathode and cobalt hydroxide production in the Democratic Republic of Congo. Chemaf has a long production history on the Copperbelt, a region widely perceived to be tough to operate in. This history stretches back to 2001, during which time the company has developed two copper–cobalt processing facilities. The experience gained in the development and operation of these two facilities has been used extensively in the Mutoshi copper–cobalt project. The metallurgical test work, process design, and project delivery approach are presented. Additionally, the decisions taken to minimize risk and maximize upside to the project are discussed. IntroductionChemaf (Chemicals of Africa), a mining company located in Katanga province of the Democratic Republic of Congo (DRC), has been operating since 2001 with two major copper– cobalt processing facilities: the Usoke and Etoile plants. In 2016, Chemaf acquired the Mutoshi mine, which belonged to Gecamines and had been explored by Anvil Mining since 2004 under a joint venture with Gecamines. Chemaf appointed African Mining Consultants Ltd (AMC) to undertake a preliminary economical assessment, then develop a JORCcompliant Mineral Resource estimate and finally a full life-of mine feasibility study. Both a Mineral Reserve and Mineral Resource were developed and the final life-of-mine feasibility study is to be published shortly. A metallurgical test work programme was initiated and carried out by Chemaf’s research and development laboratory with assistance from the Chemaf process team. After completion of the initial test work, Chemaf developed a process design criteria document for a 20 kt/a copper cathode and 8 kt/a cobalt in hydroxide production plant, followed by a flow sheet and supporting METSIM mass balance. As the drilling progressed, the plant design capacity was revised to 16 kt/a cobalt production. The life of the mine at this production rate is over 40 years."

Nov 1, 2018

By M. M. Ali, C. Musingwini, R. A. Phillips

The long-term open-pit mine production scheduling problem (OPMPSP) is typically solved in two broad stages. The first stage entails optimisation of the ultimate pit limit, which delimits the economical final pit and maximises the undiscounted value of the deposit. The second stage entails optimisation of the long-term open-pit mine production scheduling, which maximises the discounted value of the deposit. The long-term OPMPSP is a type of an NP-hard problem, such as the travelling salesman problem (TSP), because it is intrinsically complex to solve since it incorporates both hard and soft constraints. Therefore, metaheuristic approaches have been used to find global optimum solutions to the long-term OPMPSP. This paper presents a metaheuristic approach utilising a dual interchange algorithm (DIA) to solve the long-term OPMPSP. The DIA is a parent algorithm which supervises two subservient algorithms, namely, simulated annealing (SA) and Tabu search (TS). The DIA selects the better solution generated in each run from the two algorithms, and simultaneously penalises the algorithm that produced an inferior solution, while rewarding the one that produced a better solution. The DIA was tested on the three-dimensional Arizona Copper Deposit (KD) dataset, which is publicly available for research purposes from the online MineLib repository. It was set to generate net present values (NPVs) in five-minute intervals and the solution improved until it plateaued at about 30 minutes of runtime. The NPVs compared favourably to those obtained on the same KD dataset but obtained over 24-hour runtimes, when solved using simulated annealing and Tabu search algorithms, and the NPV generated from the work of Muñoz (2012) using the modified TopoSort as found on the MineLib repository at the time of this study. Therefore, this paper demonstrates that DIAs can be developed and used to solve the three-dimensional long-term OPMPSP.

Sep 1, 2021

Exxaro installed their first ultra fines treatment plant at Leeuwpan Coal mine. The plant is now fully operational and results already show that the plant has great potential. The plant was designed and installed by DRA. Several challenges were faced during the installation due to the space constraints in the Leeuwpan plant. Efficiency tests were conducted on both modules after commissioning and showed very promising results. A lower ash product with a higher yield can be produced with the fine coal dense medium cyclones compared to the spirals previously employed.

Jan 1, 2010

By G. L. Dyer, R. D. Alorro, R. E. Browner

A potential new rare earth extraction process has been suggested (Lazo, et al., 2017) (Lazo, et al., 2018) based on converting the rare earth elements to solid phase rare earth oxalates in an oxalic acid leach. Oxalic acid consumption is a significant issue in the process, therefore to improve the viability of this proposed new extraction method, a process to recover oxalate from the leach solution and recycle oxalic acid is required. A review of the literature and confirmation through thermodynamic analysis of the potential options identified precipitation of calcium or ferrous oxalate as the best potential options. The use of metallic iron to both provide ferrous ions as a precipitant and reduce ferric ions in solution would provide the most economical option, however slow kinetics and unreacted iron created significant issues with this process. The direct use of ferrous ions, while faster and more efficient, produced a precipitate with relatively poor solid/liquid handling characteristics and would require an intermediate stage utilising calcium in regenerating the oxalate. Utilising calcium chloride and oxide created very similar precipitates with better physical properties than the ferrous oxalate and that a reasonable proportion of the oxalate could be recovered before neutralisation of the solution induced precipitation of other phases such as iron oxy-hydroxides and calcium phosphate. The produced calcium oxalate was easily converted to oxalic acid and gypsum via immersion in sulphuric acid. It was also found that there is a significant loss of oxalic acid due to factors other than rare earth oxalate precipitation during leaching. This leads to a maximum oxalate recovery in the range of 50 %.

Jan 1, 2020

By A. Drumond, J. F. C. L. Costa, M. G. Lemos, A. L. Rodrigues, F. G. Niquini

The past decade has seen a renewed possibility of using machine learning algorithms to solve a large collection of problems in several fields. Data acquisition for mining operations has increased with the growth in sensor-based technologies, and therefore the amount of information available for mining applications has dramatically increased. Ore sorting equipment is available for separating ore from waste based on differences in physical properties detected by a real-time analyser. The separation efficiency depends on the contrast in these properties. In this study we investigate the application of machine learning models trained using data from the output of a dual-energy X-ray ore sorting apparatus at a gold mine. The particles were first hand-sorted into ore and gangue classes based on their mineralogical composition. Classification models were then used to help decide the balance between the number of true and false positives for ore in the concentrate, with a view to economic parameters, using their receiver operator characteristic (ROC) curves. The results showed AUC (area under the ROC curve) scores of up to 0.85 for the classification models and a maximum reward condition Fpr/Tpr around 0.5/0.9 for a simplified economic model.

Aug 8, 2024

By F. Hörtkorn, Y. Narainsamy

Rather quietly, a new generation of European norms dealing with earthworks has been developed over the past few years, in particular a part of EN 16907 pertaining to extractive waste. Whereas the majority of the parts deal with routine geotechnical aspects, one part deals specifically with tailings dams. The standard is currently in draft form but if this becomes a European Standard, members of the European Union are bound to comply with the regulations. As such, the standard has the capacity to influence mining in Europe and potentially in South Africa. In this paper we review the standard and compare it to the South African standard on mine residues and the recently published Global Industry Standard on Tailings Management (GISTM). We conclude found that although the European norm makes great strides towards increasing tailings dam safety, there is room for improvement. The current South African standard has served well since its inception, but can be improved by using concepts from the European and global tailings standards.

Jun 2, 2022

By B. Amankwaa-Kyeremeh, W. Skinner, R. K. Asamoah, C. Greet

With the continuous increase in ore mineralogical complexity coupled with decreasing head grades, there is a need for mining industries to process large tonnages of material to meet global metal demands. Smarter, cheaper techniques are required to better map ore mineralogical variations proceeding to pre-concentration (flotation) stages for higher performance and metal recoveries. This paper is aimed at briefly reviewing previous studies on different selected ore mineralogy, with the focus on their pulp chemistry, to identify relationships which aid quicker flotation feed variation monitoring. Very little published work exists linking pulp chemistry parameters to flotation feed mineralogical variation. A successful application implies maximising grades at higher metal recoveries and reducing the cost of reagent for a better performing flotation circuit. Keywords: Mineralogical variation, flotation, pulp chemistry

Jan 1, 2020

By F. Yang, S. S. Long

Mixed fatty acids, such as oxidised paraffin soap, are frequently used as a substitute for oleic acid in flotation of scheelite in China. Not only because it is cheaper than oleic acid, but also because it is superior in flotation performance to oleic acid. This superiority is generally attributed to the mixed fatty acids' synergistic effect. In this paper, the synergistic effect of oxidised paraffin soap was investigated. We found it mainly consists of a series of saturated fatty acid salts, with six to nineteen carbon atoms. All the contained fatty acids were classified into four groups based on their chain length to make the research work practical. Also, the groups were labelled as 1, 2, 3, and 4 in order. The research was simplified into studying the synergistic effect of four equivalent collectors, instead of fourteen fatty acids. It is justified by the fact that neighbouring fatty acids are expected to have a similar flotation performance. The synergistic effect of the four groups of fatty acids was studied by single-mineral flotation with single groups, binary mixtures, ternary mixtures, and the full component, as collector respectively. The flotation performance index and synergy value were introduced to evaluate flotation performance and synergistic effect of groups, respectively. The results show that the mixture 1+2+3 has the strongest synergy of all followed by the mixtures 1+3 and 2+3. The mixture 3+4 and 1+2, on the other hand, have negative synergy. The synergistic effect of mixtures of groups indicate that being consecutive in length and a broad range of lengths are vital for fatty acids to enhance synergy. Group 3 is fundamental for the collecting action of 733 oxidised paraffin soap in the flotation of scheelite. The fatty acid soap in group 4 is not suitable for flotation due to its low solubility. Keywords: Mixed collectors, synergistic effect, scheelite

Jan 1, 2020

By S. Dey, C. Bhar, S. K. Mandal

Loading and hauling of ore and waste are the key operations of an opencast coal mine and entail a high operational cost. The productivity of a mine can be increased by reducing the cycle time of loading equipment as well as utilizing dumpers optimally. In this paper we discuss the impact of rock type, bucket fill factor, rock fragmentation, the height of the cut, and angle of swing of the bucket on shovel performance. A time study is conducted on shovels in an opencast coal mine with experimental blasts of rocks to assess the impact of different factors on the performance of the shovel. Based on the data, the authors applied multiple regression (MR) and artificial neural network (ANN) techniques to develop different models for the prediction of the shovel cycle time. Developed models are validated by comparing the predicted data with actual field data. With the help of the best model, the plausible fleet size is determined in order to utilize the shovel and dumper optimally and to improve the performance of shovel–dumper operation.

Oct 1, 2022

By S. Sinha, S. Ganguly

"In an opencast mine explosives are used for fragmentation of rock. Inefficient use of explosive energy in an opencast operation produces excessive ground vibration, which is measured by peak particle velocity (PPV). To mitigate ground vibration, it is essential to develop a model to predict PPV. At present empirical models are used. These models are based on only a few input variables, hence they fail to take into account the effects of the myriad factors that cause ground vibration. Due to lack of explicit knowledge about the complex mine blasting system the scope of application of mathematical and statistical modeling techniques is limited. The artificial neural network (ANN) technique is a learning algorithm that can remove some of these limitations and can be applied to predict PPV. In this paper an ANN model is developed for prediction of blast vibration using 248 data records collected from three coal mines with diverse geomining conditions. The correlation coefficient between measured PPV and model output was found to be 0.96 and the average error percentage 11.85. The ANN model output was compared with the output of three empirical models that are widely used for prediction of PPV. The correlation coefficient between the PPV predicted by an empirical model and measured PPV data was 0.63 and the relative error percentage 38.47. This result demonstrates the superiority of the ANN model compared to empirical blast models. By using site-specific structural discontinuities as input the model performance can be further improved. Sensitivity analysis and 3D plotting were used to gain further knowledge about blast-induced ground vibration. IntroductionIn an opencast coal mine explosives are used for fragmentation of coal and overburden. If the explosive energy is not fully utilized it causes blast-induced ground vibration, which may damage nearby structures. Ground vibration is expressed as peak particle velocity (PPV). During different stages of mine planning and operation, it is necessary to use a ground vibration prediction model for blasthole design. Selection of the modelling technique is crucial. Mathematical and statistical modelling techniques have limited application because of the lack of explicit knowledge about the complex mine blasting system. Vogiatzi (2002) highlighted the problem of multicollinearity in case of statistical modeling techniques. Mutalib et al. (2013) stated that mathematical models are unable to capture the nonlinear relationship between several blasting-related parameters due to the complexity of the model input data. However, the difficulty involved in modelling complex blast vibration problems can be removed by adopting an alternative soft computing modelling approach. One of the soft computing techniques is the artificial neural network (ANN). Ragam and Nimaje (2018) developed an ANN model for predicting PPV using six input variables. Kosti et al. (2013) stated that the conventional predictors fail to provide acceptable prediction accuracy. They showed that a neural network model with four mine blast parameters as input could make significantly more accurate on-site predictions. Sayadi et al., (2013), using a database from Teheran Cement Company limestone mines, found that a neural network resulted in maximum accuracy and minimum error. Khandelwal and Singh (2009) developed an ANN model using 150 data records from an Indian coal mine with site-specific rock characteristics and geomining setting. Khandewal and Singh (2007) built a ground vibration prediction model for a magnesite mine using four prediction variables with 20 data records. Kamali and Ataei (2010) predicted PPV in the structure of the Karoun III power plant and dam using an ANN. El Hafiz et al. (2010) evaluated ground vibration predictors using data from a single-station seismograph at a limestone quarry in Egypt."

Feb 1, 2019

By M. Khodadazadeh, K. G. van den Boogaart, T. Leißner, L. Tusa, J. Gutzmer, R. Gloaguen, M. Kern

Ore sorting is a technology that is increasingly used to process primary raw materials. Timeconsuming and expensive empirical state-of-the-art test work is carried out to assess whether the use of ore sorting for the enrichment of a particular ore makes technical and economic sense. With the innovative simulation-based approach presented here, it is possible to direct the selection of a suitable sensor based on quantitative mineralogical and textural data, thus avoiding much of the empirical studies. Required data can be collected quickly and cost-effectively using available methods of automated mineralogy. The obtained parameters such as mineral grain size distribution, modal mineralogy, mineral area and mineral density distribution have been utilised in this study to simulate the prospects of success of ore sorting, applying different types of sensors. Empirical tests with commercially available sensor systems have been conducted to experimentally validate the predictions of the simulations. The estimation of the target mineral grade can be further optimised by the use of machine learning algorithms for the integration of automated mineralogy data and sensor data. The approach can easily be adapted to other types of raw materials and thus has great potential to become a key technology for the optimisation of processing experiments. Keywords: Automated mineralogy, hyperspectral imaging, machine learning, ore sorting

Jan 1, 2020

By M. M. Bwalya, F. K. Mulenga

"SynopsisThe application of the attainable region (AR) technique to the analysis of ball milling is currently limited to batch data. This paper introduces the use of the technique to continuous milling.To this end, an industrial open milling circuit processing a platinum ore was surveyed. Samples were collected and later characterized by means of laboratory batch testing. On site, several milling parameters were varied systematically so as to collect data for modelling purposes. These paramters included ball filling, slurry concentration, and feed flow rate. After data analysis, a simulation model of the open milling circuit was developed under MODSIM®, a modular simulator for mineral processing operations. The mill was then simulated and the data generated was analysed within the AR framework.Initial findings reveal an opportunity to gain valuable insight by studying milling using the AR technique. From an exploratory perspective and inasmuch as this study is concerned, feed flow rate, ball size, and ball filling were identified as being pivotal for the optimization of open ballmilling circuits. Mill speed, on the other hand, had only a limited effect on the production of particles in the size range -75 +10 µm.IntroductionGlasser and Hildebrandt (1997) propounded the attainable region (AR) as a technique for the analysis of chemical engineering reactor systems. Results have since been produced and tested on both the laboratory and pilot scales. The method results in a graphical description of chemical reactions by considering the fundamental processes taking place in the system, rather than the equipment. From the plotted graphs, the process and the reactors can be synthesized optimally into a flow sheet.The use of the AR method still has a long way to go as far as mineral processing is concerned. For instance, several articles have reported the application of the AR technique to ball milling: Khumalo et al. (2006, 2007, 2008); Khumalo (2007); Metzger et al. (2009, 2012); Metzger (2011); Katubilwa et al. (2011); Hlabangana et al. (2012). Their main shortcoming has been the exclusive use of laboratory batch grinding data.To address this deficiency, Mulenga and Chimwani (2013) proposed a way by which the technique could be extended to continuous milling. In effect, the batch milling characteristics of a platinum-bearing ore (Chimwani et al., 2013) were used and scaled up to an open milling circuit. Then, with simplifying assumptions, an attempt was made to optimize the residence time of particles inside the mill. Later, Chimwani et al. (2014a) presented some optimization examples involving various milling parameters. The sequence of published articles then paved the way for the study of industrial milling systems with the AR methodology."

Jan 1, 2015

By Z. Levinson, R. Dimitrakopoulos

A mining complex is an engineering system where raw materials are extracted from several mines, processed, transported, and sold to the market. The uncertainty and local variability of material grades and their locations within the related mineral deposits are critical for determining the optimal life-of-mine extraction sequence, destination policy, operating modes, and capital investment decisions, which define the production schedule of a mining complex. Infill drilling is used to further characterise a mineral deposit and reveals information that changes the representation of the resource model and potentially the resulting optimal long-term production schedule. A self-learning reinforcement learning approach is proposed to determine the optimal drilling locations that maximise the net-present value of the long-term production schedule and the associated financial forecasts by adapting the schedule to additional information. A case study at a copper mining complex demonstrates a $70 M increase in project value by collecting drilling information with a $1 M budget.

Sep 1, 2021

By A. Navarra, A. Quelopana

The mine production scheduling problem (MPSP) has been studied since the 1960s, and remains an active area of computational research. In extending the concepts of the MPSP, the automated mine may now be regarded as a hierarchical intelligent agent in which the bottom layer consists of distributed robotic equipment, while strategic functionality occupies the higher layers. Here we present a disambiguation of artificial intelligence, machine learning, computational optimization, and automation within the mining context. Specifically, the Q-learning algorithm has been adapted to generate the initial solutions for a high-performing strategic mine planning algorithm, originally developed by Lamghari, Dimitrakopoulos and Ferland, based on the variable neighbourhood descent (VND) metaheuristic. The hierarchical intelligent agent is presented as an integrative conceptual platform, defining the interaction between our new Q-learning adaptation and Lamghari’s VND, and potentially other hierarchically controlled components of an artificially intelligent mine, having various degrees of automation. Sample computations involving Q-learning and VND are presented.

Dec 1, 2021

By Jean Michel Marino, Jean-Pierre Slama

"The computer methods used by the French Commissariat a l'Energie Atomique, which are applicable to any type of ore deposits, are:(i) Ore reserve estimation. The data are either the core analyses or gamma leggings. To determine the spatial distribution of the ore, three different methods of interpolation are used. Two of these are based on the least-squares theory and the third one is an application of the regionalized variable theory. The interpretations of the orebody by the geologists are introduced in the computer program as often as possible.(ii) Optimal open pit design. From the spatial distribution of ore, the three-dimensional outlines of the optimal pit are determined by a geometrical algorithm which allows different slopes varying with the azimuth.Finally, the optimization of annual successive plans arc discussed."

Jan 1, 2014

By S. T. L. Harrison, M. A. Fagan-Endres, M. D. Odidi

This paper investigated the ability of eight common hydrodynamic models – tanks in series (TIS); three compartmental model configurations (Comp-1, Comp-2 and Comp-3); advection dispersion (AD); piston exchange (PE); piston exchange-diffusion variant (PE-D) and piston dispersion exchange (PDE) – to simulate residence time distribution (RTD) data generated from tracer tests conducted on a packed bed composed of different size fractions of malachite ore. The chosen materials were packed into 0.1 m internal diameter columns to a depth of 0.155 m and drip irrigated at 5.11 L/ m2.h. The four size fractions tested were: 1 – 2 mm, 2 – 2.8 mm, 4.75 – 5.6 mm and 13.6 – 16 mm. Step-up tracer tests were conducted using potassium chloride. The results showed that an increase in the particle size of the malachite ore had little effect on overall bed porosity but resulted in significantly lower values of liquid holdup. Two of the monoporosity models tested, TIS and Comp-2, were the most affected by the changes in bed size fraction due to their inherent inability to cope with high degrees of non-ideal flow present in the larger particle systems. The dual porosity models performed relatively better with the dominant trend being a reduction in dynamic to total liquid holdup as particle size increased. Of all the models tested, the PDE and TIS models were the best and worst performers respectively, with R2 values averaging at 0.997 and 0.679 across all experiments respectively. Though the Comp-1, Comp-3 and PE-D models also performed well, the addition of the dispersion coefficient to take the axial spread of fluid molecules into account added a layer of complexity that allowed the PDE model to perform relatively better than the other dual porosity models. Keywords: Heap leaching, packing material, hydrodynamics, malachite, PSD, modelling

Jan 1, 2020

By V. W. Lumadi, S. Nyasha

The mining industry has yet to fully accept and embrace the strategic role of technology and innovation in successful business planning and execution. In this study, we empirically determine the factors that influence acceptance of technology solutions in the mining industry in South Africa, assess critical success factors influencing adoption of advanced technology in the South African mining industry, and explore challenges faced by the mining industry in South Africa in adopting advanced technology in its operations. Thematic content analysis is employed in the data analysis. Results show that critical success factors to the adoption of mining technology include learning culture, knowledge sharing, high labour costs, and behaviour of other firms in the industry; commonly identified challenges include inadequate engagement with external stakeholders, uncertainties, the cyclical nature of the sector, and high risk related to the adoption of unproven technology and performance systems focused on volumetric production. The study encourages mining companies to implement usage of technology. This can be properly done through regular engagement with the Minerals Council South Africa.

Apr 14, 2024

By Y. Fu, C. Aldrich

This study investigates the use of dense convolutional neural networks to estimate bubble size distributions in flotation froths. 20 videographically captured images of froths from an industrial platinum flotation plant in South Africa were labelled semantically with an image annotation tool, LabelMe (https://github.com/wkentaro/labelme). As a result, pixels in the images could be categorized into three classes, namely ‘bubbles’ (class 0), ‘bubble boundaries’ (class 1) and ‘background’ (class 2). A U-net architecture with a ResNet34 backbone was trained to segment the images via classification of the pixels in the images. 5 of the 20 images were used to validate the performance of the neural network. Explicit inclusion of the bubble boundaries as a class proved to be significantly more effective than training the network with two classes only (i.e. bubbles and background). However, the thickness of the boundary layer regions were critical. If too thin, it was difficult to distinguish between some contiguous bubbles and if too thick, some fine bubbles may not be identified. The results were promising, although there is still significant scope for improvement, among other by making use of more images for training and validation. Keywords: Flotation Froths, Convolutional Neural Networks, Deep Learning, Bubble size Analysis

Jan 1, 2020