Bayesian Fault Diagnosis Using Process Knowledge of Response Information

Process fault diagnosis is a topic of significant practical interest. Bayesian fault diagnosis methods have been developed to identify the problem source from all monitors of the process. However in a large scale industrial process, taking all the monitors into account not only increases computation burdens but also leads to spurious diagnosis. This paper proposes a new approach to obtain a more reliable diagnosis under Bayesian frame. It explicitly takes the process knowledge expressed as response matrix into consideration to estimate the likelihood in Bayesian inference. The simulation demonstrates that the proposed approach is able to improve the diagnosis even when some abnormal mode data is sparse or not available in the historical dataset.