A particle filter driven dynamic Gaussian mixture model approach for complex process monitoring and fault diagnosis

Abstract

Complex non-Gaussian processes may have dynamic operation scenario shifts so that the conventional monitoring methods become ill-suited. In this article, a new particle filter based dynamic Gaussian mixture model (DGMM) is developed by adopting particle filter re-sampling method to update the mixture model parameters in a dynamic fashion. Then the particle filtered Bayesian inference probability index is established for process fault detection. Furthermore, the particle filtered Bayesian inference contributions are decomposed among different process variables for fault diagnosis. The proposed DGMM monitoring approach is applied to the Tennessee Eastman Chemical process with dynamic mode changes and the results show its superiority to the dynamic principal component analysis (DPCA) and regular Gaussian mixture model (GMM) in terms of fault detection and diagnosis accuracy.