Overload Detection in Semi-Autogenous Grinding: A Nonlinear Process Monitoring Approach

Abstract

Detecting the onset of overloading in a semi-autogenous grinding (SAG) mill is a challenging task for operators to perform due to the complex and nonlinear nature of an overload. To detect an overload, operators must simultaneously monitor the correlations between several measurements of the SAG process. However, overloading often goes unnoticed at its early stages because the subtle changes in the correlations between measurements are difficult for an operator to observe. In addition, linear process monitoring techniques such as principal component analysis (PCA) provide inconsistent results with overload detection because of the process nonlinearity. Recently, locally linear embedding (LLE) with a linear classifier has been proposed to detect the early onset of an overload in a SAG mill. In this paper, we compare the suitability of LLE to detect the early onset of an overload against kernel PCA and support vector machines.