This paper is about multivariate process monitoring (detection and diagnosis) with Bayesian networks. It allows to unify in a same tool (a Bayesian network) some monitoring dedicated methods like multivariate control charts or discriminant analysis. After the context introduction, we present Bayesian networks (with discrete and Gaussian nodes) in section 2. In sections 3 and 4 we respectively propose the modeling of the two tasks (detection and diagnosis) in the Bayesian network framework, unifying the two steps of the process monitoring in a sole tool. An application is given in section 5 in order to demonstrate the effectiveness of the proposed approach. This application is a benchmark problem in process monitoring: the Tennessee Eastman Process. Efficiency of the network is evaluated for detection and for diagnosis. Finally, we give conclusions on the proposed approach and outlooks concerning the use of Bayesian network for process monitoring.
Sylvain Verron, Teodor Tiplica, Abdessamad Kobi