Chemical Engineering Science, Vol.201, 82-96, 2019
Dynamic process fault detection and diagnosis based on a combined approach of hidden Markov and Bayesian network model
The present study introduces a novel methodology for fault detection and diagnosis (FDD), based on a combined approach of data and process knowledge driven techniques. The Hidden Markov Model (HMM) detects the abnormalities based on process history while the Bayesian Network (BN) diagnoses the root causes of faults. An HMM is trained with standard operating condition data while the structure of BN is developed based on process knowledge. The log-likelihoods (LL) of process historical data strings are used to establish the conditional probability tables of the BN. HMM identifies the abnormal behaviour of the system. The time of detection of abnormality, respective log-likelihood value, and the probabilities of being in the process condition at the time of detection are used as evidence to BN. The updated BN is then used to diagnose the root cause by considering the respective changes in the probabilities. Performance of the new technique is tested and validated with published data of Tennessee Eastman Process. Eight of the ten selected faults were successfully detected and diagnosed. (C) 2019 Elsevier Ltd. All rights reserved.