A Variable Selection Method for Fault Isolation through Bayesian Information Criterion

In industrial process, fault isolation technology can identify the major variables leading to faults. Contribution plots and reconstruction-based methods are common tools, but both of them have the smearing effect. To solve this problem, Bayesian theory based method have been developed. Unfortunately, they usually have a high misdiagnosis rate when handling multiple faults with small magnitudes. In this paper, a new fault isolation method based on Bayesian information criterion is proposed. Firstly, the fault isolation problem is transformed into a mixed integer nonlinear programming problem. Then, to reduce the difficulty of calculation, the original problem is simplified into a series of nested mixed integer quadratic programming problems by using forward selection strategy. Finally, these problems can be solved by branch and bound algorithm. The effectiveness of the proposed method is verified by Monte Carlo simulation.