Part Mutual Information Based Quality-related Component Analysis for Fault Detection

Abstract

In this paper, a novel part mutual information based quality-related component analysis (PMIQCA) method is presented to detect quality-related faults and reduce the interference alarms. The low-dimensional subspace of process variables can be found, which reflects real-time changes in quality. The detection rates of quality-unrelated faults can be reduced while the detection rates of faults that are related to quality are increased. The basic idea is to select the most relevant process variables and principal components (PCs) with the maximal part mutual information (PMI) for each iteration, so as to build a more accurate supervisory relations between process variables and quality. Afterwards, two appropriate statistics are established for quality-related fault detection. Finally, the Tennessee Eastman Process (TEP) is carried out to demonstrate the effectiveness of PMIQCA.