Hidden semi-Markov model based monitoring algorithm for multimode processes

Abstract

Several studies have adopted hidden Markov model (HMM) to monitor multimode processes. The drawback of HMM is that its inherent duration probability density is exponential and hence it is inappropriate for the modeling of multimode processes. To address this problem, hidden semi-Markov model (HSMM), which introduces the mode duration probability into HMM, is combined with principal component analysis (PCA) in this paper, named as HSMM-PCA. With the restriction of mode duration probability, HSMM-PCA can successfully identify the operation mode affiliation and build the precise PCA model for each mode. As a result, HSMM-PCA is more sensitive to abnormal conditions and has better fault detection ability for multimode processes.