Abstract: Because of the multiplicity of operation phases in batch process， faults may have obvious features in one or more specific operation phases， but do not show any features in other operation phases. That is， faults have corresponding feature phases. A fault diagnosis method based on identification of fault feature phases was proposed. Identification of fault feature phases was realized by comparing the differences between the centroids of historical faulty data set and the normal data set. Different fault diagnosis models were respectively developed based on multiway Fisher discriminant analysis (MFDA) to reduce the search space to specific fault feature phases and improve the diagnosis performance of the models. Simulation experimental results showed the feasibility and validity of the proposed method.

Key words: batch process, multiphase, fault feature phase, multiway Fisher discriminant analysis (MFDA), fault diagnosis