A tutorial introduction to nonlinear process monitoring for practicing engineers

This is a tutorial paper on industrially successful approaches for nonlinear process monitoring. Online process monitoring is essential to continuous operation of plants at high efficiency. The problem of process monitoring is one where an impending but undesirable process condition has to be identified and used to alert process engineers. This problem has traditionally been solved by practicing engineers using statistical techniques that detect a change in the underlying correlation structure of process variables. For linear correlation structures, algorithms such as principal components analysis, independent component analysis and partial least squares have been proposed. However, many industrial processes exhibit nonlinear correlation structure between the process variables. In this tutorial we describe various nonlinear monitoring techniques and elaborate on two recently developed methods, namely Support Vector Machine and Local Linear Embedding. The two methods are evaluated by application on two experimental case studies, a SAG mill and a polymer extruder.