Multimode Process Monitoring with Mode Transition Constraints

Abstract

Multimode process monitoring has gained widespread attention both in industry and academia recently, and the hidden Markov model (HMM) has been introduced to handle the multimodality of process data. However, most of HMM-based approaches cannot effectively detect mode disorder faults, if multimode processes operate under mode transition constraints. In this article, a new HMM-based method is developed to address this problem. The moving window Viterbi (MW-Viterbi) algorithm is proposed to identify operating modes, where reconstructed samples are utilized for mode identification if a fault occurs. Then, the Mahalanobis distance is adopted for fault detection, and a reconstruction-based method is derived for fault identification. The numerical experiment illustrates the superiority of the developed method in multimode process monitoring with mode transition constraints.