Hidden Markov Model Based Adaptive Independent Component Analysis Approach for Complex Chemical Process Monitoring and Fault Detection

For complex chemical processes with multiple operating conditions and inherent system uncertainty, conventional multivariate process monitoring techniques such as principal component analysis (PCA) and independent component analysis (ICA) are ill-suited because they are unable to characterize shifting modes and process uncertainty. In this article, a novel hidden Markov model (HMM) based ICA approach is proposed for process monitoring and fault detection. First the hidden Markov model is built from measurement data to estimate dynamic mode sequence. Further the localized ICA models are developed to characterize various operating modes adaptively. HMM based state estimation is then used to classify the monitored samples into the corresponding modes, and the HMM based I-2 and SPE statistics are established for fault detection. The effectiveness of the proposed monitoring approach is demonstrated through the Tennessee Eastman Chemical process. The comparison of monitoring results shows that the proposed HMM-ICA approach is superior to the conventional ICA method and can achieve accurate detection of various types of process faults with minimized false alarms.