Algorithms to control coal segregation under non-stationary conditions: Part I: Moving window and SPC-based updating methods

With highly heterogeneous run of mine coal streams, physical segregation, using on-line quality measurements, can serve as an economically important first stage of the mineral beneficiation process. Segregation enables high quality fractions of the stream to bypass cleaning operations, thereby saving cleaning costs and avoiding the yield losses inherent in any downstream separation process. This paper develops control strategies that address the objectives of quality targeting of the no-wash coal to meet desired quality specifications while maximizing yield of the segregation process to the no-wash stock. It extends earlier work by the authors to deal with the case where the stochastic nature of the quality levels is non-stationary over time, a situation any practical segregation system should accommodate. Two techniques, involving the use of moving windows and auto-correlated statistical process control techniques, are employed to define intervals of time where the process might be considered as “locally” stationary. Subsequently, two different methods, involving the use of empirical and Gaussian distributions, are used to characterize the distribution of quality levels over these intervals. Given this characterization, a special decision criterion is applied to make segregation decisions that enables one to meet the control objectives stated above. Performance of different variants of these control strategies are compared with each other and to an existing industrial segregation algorithm using data obtained at a representative mine. The methods show dramatically improved capability to control quality targets and increase yields. Moreover, different variants of the algorithms are more effective in different application circumstances.