Spiral Concentrator Interface Monitoring Through Image Processing: A Statistical Learning Approach

Spiral concentrators are robust gravity separation devices that allow for concentration of slurry streams. Optimal splitter position (which determines recovery and grade) is dependent on the interface positions of the concentrate, middlings and/or tailings in the trough. Various image processing techniques have been proposed to automatically detect interface positions, which could be useful for spiral concentrator monitoring and control. Two methods are compared in this work: the first is the previously-described genetic algorithm optimization of the parameters of a traditional edge detection algorithm. The second uses logistic regression, a well-known statistical classifier. The performance of the two methods was compared on two data sets, for ilmenite and chromite concentration. The logistic regression method was shown to outperform the genetic algorithm approach, in terms of computational cost of training and successful interface detections on test data, for both the relatively easy ilmenite concentrate interface, and the more challenging chromite concentrate interface.