Monitoring of a Platinum Group Metal Flotation Plant with an Isolation Forest

Froth flotation is one of the most important techniques in mineral processing to beneficiate valuable minerals from ore. As a consequence, advanced control of industrial flotation plants can significantly affect the efficiency of mineral beneficiation. Videographic monitoring of flotation cells is a vital element of advanced process control systems, as the froth phase richly reflects the operational state of the process. However, traditional multivariate process monitoring systems based on principal component analysis (PCA) may not necessarily capture the nonlinear behaviour reflected features extracted from the froth phase. Therefore, in this paper a novel unsupervised monitoring scheme is proposed which is based on the isolation forest (IF) model and its statistical metric anomaly score. The results show that this novel monitoring scheme can be a promising alternative or complementary to traditional monitoring schemes based on the PCA model.