Improving Process Fault Detection and Diagnosis Using Robust PCA and Robust FDA

The performance of PCA and FDA based fault detection and diagnosis procedures could deteriorate with the violation of the normality assumptions made during conventional approaches. The consequence is a reduction in accuracy of the models and efficiency of the methods, which results in an increase of misdetection and misclassification rate. A robust method is proposed to deal with the normality violation, especially the multivariate outliers existing in the data. This method, using a winsorization procedure with an M-estimator based on the generalized t distribution, possesses both robustness and effectiveness, and results in better PCA and FDA models when the assumption is violated in practical cases. Comparisons between the proposed and the conventional PCA and FDA modeling techniques and their applications to process fault detection and diagnosis are illustrated through a multipurpose chemical engineering pilot-facility.