Adaptive control in mineral processing

"The raw ores which are processed in mineral processing plants exhibit very strong variations of their physical and chemical properties. As a consequence, the performances of the comminution and separation processes vary greatly from time to time. Model-based control strategies improve the performance of the plant; however, they require some form of adaptation to cope with the variations of the are properties.This paper deals with the use of adaptive techniques for the control of mineral processes. It begins with a survey on adaptive control algorithms including parameters estimation aspects. Then several examples dealing with crushing, rod, ball, autogenous and semi-autogenous grinding, flotation , drying and calcination processes are reviewed.IntroductionThe operation of the various processes of mineral processing plants (crushing, grinding, flotation, dewatering, etc.) is generally controlled using empirically tuned regulators of constant structure, such as PID controllers. To improve the operation performance, mineral processors try to implement specifically designed controllers, i.e. controllers based on plant dynamic models. Most of the time, these techniques improve the control of the mineral processing installations. However, modelling is a complex procedure because of the problems of measurement and data management, specific to this industry, and because of the variations of the model parameters. In this context model-based control strategies do not always bring the expected improvements or are sometimes discarded and replaced by more conventional techniques .The sources of variation of the dynamic behaviour of mineral processing systems can be classified into three categories: (I) the inherent changes of the raw ore properties (composition, grindability, mineralogical texture, etc.); (2) the process tuning modifications (grinding media loads, number of hydrocyclones, slurry level in flotation cells, agitation speeds, etc.); (3) the feedrate variation of the ore, water, chemical reagents, etc. All these factors change the fundamental properties of the process dynamics such as the grinding or flotation rates and the residence time in the equipment. The simplified linear models which are used to describe the process dynamics are not able to track the complex behaviour of the mineral processing systems when the amplitude of the sources of variation are too large. The mismatch between the model and the plant behaviour is then frequently a cause of model-based controller failures."