Class-incremental fisher discriminant analysis with principal component analysis for process monitoring

A new approach using class-incremental Fisher discriminant analysis (FDA) with principal component analysis (PCA) is proposed for process monitoring. FDA seeks directions that are efficient for discrimination and shows smaller error rate for detecting and diagnosing the known faults. However, it may not detect and identify those new faults that aren't included in the model. As its complementarities, PCA and PCA-based joint angle plot are used for detect and identify the new fault, respectively. After the new fault is detected and added to the known fault library, class-incremental FDA updates FDA model so as to improve the monitoring performance for the new fault. The approach is applied to TEP for process monitoring. The results show the effectiveness of the method.