Chris Aldrich

The interpretation of flotation froth surfaces by using digital image analysis and neural networks

Abstract The rapid developments in computer vision, computational resources and artificial intelligence, and the integration of these technologies are creating new possibilities in the design and implementation of commercial machine vision systems. In chemical and minerals engineering, numerous opportunities for the application of these systems exist, of which the characterization of flotation froth structures is a good example of the utilization of visual data as a supplement to conventional plant data. In this paper images from pyrite batch flotation tests conducted after a factorial design as well as images from a copper flotation plant were used to understand the relationship between froth characteristics and flotation performance better. The results show that a significant amount of data can be extracted from flotation surface froths. Techniques have been developed to characterize chromatic information, average bubble size, froth texture, froth stability and mobility of surface froths. It has been shown that most of the froth characteristics of this study can be explained in terms of the concentration of solids in the froth and the factors that affect the solids concentration. The techniques developed proved to be useful in investigating the effect of a mixed collector and the addition of copper sulphate. The depressing effect of the copper sulphate and the higher grades and recoveries made possible by the mixed collector under these conditions were explained by analysis of the froth features. Excellent results were obtained in modelling the relation between froth characteristics or froth grade and recovery by using a backpropagation neural network. A sensitivity analysis showed that the most important froth features for the experimental conditions of this study are the froth stability, mobility and average bubble size. This computer vision system constitutes a powerful research tool for the investigation and interpretation of the effect of various flotation parameters. This paper also shows how the rapid development in computer technology and related disciplines can be used to transform recently developed concepts and available technology into a new generation of intelligent automation systems.

ANN-DT: an algorithm for extraction of decision trees from artificial neural networks

Although artificial neural networks can represent a variety of complex systems with a high degree of accuracy, these connectionist models are difficult to interpret. This significantly limits the applicability of neural networks in practice, especially where a premium is placed on the comprehensibility or reliability of systems. A novel artificial neural-network decision tree algorithm (ANN-DT) is therefore proposed, which extracts binary decision trees from a trained neural network. The ANN-DT algorithm uses the neural network to generate outputs for samples interpolated from the training data set. In contrast to existing techniques, ANN-DT can extract rules from feedforward neural networks with continuous outputs. These rules are extracted from the neural network without making assumptions about the internal structure of the neural network or the features of the data. A novel attribute selection criterion based on a significance analysis of the variables on the neural-network output is examined. It is shown to have significant benefits in certain cases when compared with the standard criteria of minimum weighted variance over the branches. In three case studies the ANN-DT algorithm compared favorably with CART, a standard decision tree algorithm.

Digital image processing as a tool for on-line monitoring of froth in flotation plants

Abstract As the most important separation technique in mineral processing, flotation has been the subject of intensive investigation over many years, but despite these efforts it remains a poorly understood process that defies generally useful mathematical modelling. As a result the control of industrial flotation plants is often based on the visual appearance of the froth phase, and depends to a large extent on the experience and ability of a human operator. These types of processes are consequently often controlled suboptimally owing to high personnel turnover, lack of fundamental understanding of plant dynamics, inaccuracy or unreliability of manual control systems, etc. By using techniques based on image colour analysis and Fast Fourier Transforms to process videographic data of the froth phase in a copper flotation plant, it is shown that an image processing system can distinguish between different copper levels in the froth down a rougher bank and extract global features from the visual characteristics of the surface froth. In this way it is possible to quantify the mineral content of the froth (based on colour), the average bubble size distribution, the direction of flow and the shape of the bubbles or the mobility of the froth. The overall image of the froth is analysed instead of attempting to identify the boundaries between bubbles.

The significance of flotation froth appearance for machine vision control

Abstract The development of robust automatic control systems has proved difficult because of the complexity of the problem. Flotation is notorious for its susceptibility to process upsets and consequently its poor performance, making successful flotation control systems an elusive goal. Machine vision systems provide a novel solution to several of the problems encountered in conventional flotation systems for monitoring and control. In previous work powerful techniques have been developed for the extraction of flotation froth appearance features such as average bubble size, froth mobility and stability, chromatic information and textural properties of surface froth. A methodology has been developed for the classification of froths, based on appearance and metallurgical significance. The objective of this paper is to provide a clear framework and motivation for the development of a machine vision system for flotation control. A systematic discussion of the diffuse literature descriptions about the relation between froth appearance and fundamental flotation principles is presented. A preliminary classification strategy for flotation froths is proposed and an example of how process deviations can be related to froth appearance is provided. Design constraints and principles imposed on a vision system by flotation are also discussed.

Effect of particle size on flotation performance of complex sulphide ores

Flotation processes occurring in the bulk and froth phases have a characteristic influence on the structural features and dynamics of the flotation froth. It is recognized that the structure and texture of a mineral froth is a good indicator of flotation separation performance. The surface froth feature and dynamics are presented by three features extracted from the digitized images of the froths, i.e. SNE, a rough indication of the average bubble size of the froth, froth stability and the average grey level of the froth, an indication of mineral loading. Particle size is an important parameter in flotation operation. Nowadays, particle size is often measured and controlled in flotation concentrators. In this study the dependence of the froth structures on the particle size variation was investigated on the batch flotation of a sulfide ore from the Merensky reef in South Africa, and the size by size recovery curves were studied as well. In general medium particles produced bubbles smaller than those observed in the presence of fine and coarse particles, and the recovery rates were larger. Entrainment was a contributory mechanism to the recovery of fine particles. The fluctuation of flotation indices on the particle size change can be diagnosed and predicted by the froth structures change with a high degree of accuracy.

The classification of froth structures in a copper flotation plant by means of a neural net

Abstract By making use of grey level dependence matrix methods, digitized images of the froth phases in a copper flotation plant were reduced to feature vectors without losing essential information of the characteristics of the froth. Classification of features extracted by means of both spatial grey level dependence matrix (SGLDM) methods, as well as neighbouring grey level dependence matrix (NGLDM) methods was investigated. By using a learning vector quantization (LVQ) neural net it was shown that froth structures could be classified satisfactorily when either NGLDM or SGLDM methods were used. When these feature sets were combined, however, the success rate of classification improved to almost 90%. This is sufficiently accurate to enable incorporation of the neural net classifier into on-line plant control systems.

The interrelationship between surface froth characteristics and industrial flotation performance

Abstract This paper discusses the rapid development in computer technology and neural networks that are used to transform recently developed concepts and available technology into a new generation of intelligent automation systems. In this study features extracted from images of froths by an on-line machine vision system in an industrial precious metal flotation plant were used to relate froth characteristics with the performance of the plant by using self-organising and Sammon maps. This intelligent vision system constitutes a powerful tool for the investigation and interpretation of the effect of various flotation parameters. Previous work is extended by relating surface froth characteristics with industrial flotation control and performance variables. This method of system identification represents a significant development towards an automatic control system.

The cross-entropy method in multi-objective optimisation: An assessment

Solving multi-objective problems requires the evaluation of two or more conflicting objective functions, which often demands a high amount of computational power. This demand increases rapidly when estimating values for objective functions of dynamic, stochastic problems, since a number of observations are needed for each evaluation set, of which there could be many. Computer simulation applications of real-world optimisations often suffer due to this phenomenon. Evolutionary algorithms are often applied to multi-objective problems. In this article, the cross-entropy method is proposed as an alternative, since it has been proven to converge quickly in the case of single-objective optimisation problems. We adapted the basic cross-entropy method for multi-objective optimisation and applied the proposed algorithm to known test problems. This was followed by an application to a dynamic, stochastic problem where a computer simulation model provides the objective function set. The results show that acceptable results can be obtained while doing relatively few evaluations.

Neurocontrol of a ball mill grinding circuit using evolutionary reinforcement learning

Abstract A ball mill grinding circuit is a nonlinear system characterised by significant controller interaction between the manipulated variables. A rigorous ball mill grinding circuit is simulated and used in its entirety for the development of a neurocontroller through the use of evolutionary reinforcement learning. Reinforcement learning entails learning to achieve a desired control objective from direct cause—effect interactions with a simulated process plant. The SANE (symbiotic adaptive neuro-evolution) algorithm is able to learn implicitly to eliminate controller interactions in the grinding circuit by taking a plant wide approach to controller design. The ability of the neurocontroller to maintain high performance in the presence of large disturbances in feed particle size distribution and ore hardness variations is demonstrated. The generalisation afforded by the SANE algorithm in dealing with considerable uncertainty in its operating environment attests to a large degree of controller autonomy.

The application of neural nets in the metallurgical industry

Abstract Although the potential of new techniques for the construction of accurate plant models, such as those based on connectionist methods, is generally acknowledged, little on their practical application can be found in the chemical and metallurgical engineering literature. In this paper the use of neural nets to model gold losses on a reduction plant and the consumption of an additive on a leach plant, as well as the pyrometallurgical processing of zinc and aluminium is discussed. The gold and leach plant models performed better than the multilinear regression models used on the plants, even where relatively few data were available. The neural networks used to model the recovery of lead and zinc from industrial flue dusts, process synthesis of zinc recovery plants and the processing of secondary aluminium in a rotary salt flux furnace produced realistic results that could be used by plant personnel to optimize their operations.