Hidden Markov Model-Based Statistics Pattern Analysis for Multimode Process Monitoring: An Index-Switching Scheme

Multiple operating modes pose a challenge for process monitoring in industry. Although many monitoring approaches have achieved quite success, most of them neglected the dependency of sampled data and only dealt with samples in a separate fashion. This paper proposes a sequential framework for multimode process monitoring with hidden Markov model-based statistics pattern analysis (HMM-SPA). To begin with, a hidden Markov model is trained on the basis of the historical data. Statistics pattern analysis mixture models (SPAMM) are constructed to characterize the distinctive statistical pattern of each operating mode. Then, during online monitoring period, the mode vector is obtained using the Viterbi algorithm, and the differential mode vector is calculated. At last, the proposed method switches to an appropriate monitoring index automatically, according to the norm of the differential mode vector. The effectiveness of the proposed method is demonstrated by a numerical simulation, a continuous stirred tank heater (CSTH) process, and the Tennessee Eastman process.