Monitoring Multiphase Batch Operations using Hidden Semi-Markov Statistical Model

J. Chen and Y.-C. Jiang (Taiwan, China)


Hidden segmental semi-Markov model, multiway principal component analysis and process monitoring


This paper addresses the automatic on-line monitoring batch operation. A novel and flexible approach is proposed based on hidden segmental semi-Markov models (HSMM) and multiway principal component analysis (MPCA) for different phases in the dynamic batch operation but ambiguity between phases exists. MPCA is used for the elimination of the cross-correlation among process variables; HSMM, for construction of the temporal behavior among process variables at different phases. Although prior process knowledge may not be available in many batch processes, the probability distribution of the transition time among different phases can be trained from the batch operation data. Subsequently, recognition of the pattern in batch runs with the normal operation is achieved by a recursive Viterbi algorithm, which can find the transition state sequence from the series observation data. The control limit for the normal operation condition is presented to compare process features of on-line batch processes. The proposed method is successfully demonstrated in a simulated fed-batch penicillin cultivation process. The comparisons of the existing stage based PCA and HMM methods are also shown, explaining the power and the advantages of the proposed method.

Important Links:

Go Back