Minimum risk Bayesian decision based fault diagnosis of industrial chemical process

Abstract

Fault identification is a critical step of the fault diagnosis of an industrial process. The faults in chemical processes rarely show a random behavior. Generally, they will be propagated to different variables because of the influence of the process controllers and the correlations between variables. Thus, it is helpful to take the pervious fault diagnosis results into consideration during the current determination of faulty variables. In the presented work, an unsupervised data-driven fault diagnosis method is developed based on the minimum risk Bayesian decision theory. This approach combines reconstruction-based contribution and the minimum risk Bayesian inference method. The loss function is introduced into the method. The benchmark Tennessee Eastman (TE) process is used to verify the effectiveness and applicability of the proposed method.