Identification and Online Classification of Ore Crushing Circuit Process Modes

Ore crushing circuits are often characterized by distinct operating modes, knowledge of which contributes towards improved control and operation. Process modes were extracted in an unsupervised manner from an expanded, feature-rich (i.e., including auxiliary measurements) historical dataset by nonlinear dimensionality reduction and clustering. Subsequently, a kernel support vector machine was trained to classify operating conditions according to previously identified modes using routine process measurements. Applying the classifier to unseen data indicated that the identified process modes were distinct and meaningful. Finally, Shapley additive explanations (SHAP) analysis was shown to be a valuable tool for understanding the classifier’s outputs.