Fault Diagnosis Based Minimum Risk Bayesian Decision Approach In Industrial Process

It is of great importance to quickly detect and identify an abnormal event occurring in industrial manufacturing processes, which will improve the safety of the processes and avoid the production loss. The rapid progress in computer control technology makes the data collection and storage industrial processes possible. Making full use of the data information for process monitoring and fault diagnosis has received a lot attention. And multivariate statistical process monitoring is one of current hot research.Fault detection and fault diagnosis are two crucial parts during process monitoring. In this paper, the principal component analysis method was adopted for fault detection, which is a popular method based on multivariate statistical methods for fault detection. After a fault was detected, the following step was to diagnose its root causes. The common fault diagnosis methods are contribution plots and reconstruction method, while both have the smearing effect. The smearing effect is that the contributions of non-faulty variables were affected by the faulty variables due to the variables correlation, which may cause the wrong fault diagnosis. In addition, the problem of fault propagation during the variables due to the actions of the process controllers and variables correlation was not solved when a process fault occurred. This paper focused on the two problems and developed a new method combining reconstruction-based contribution(RBC) with minimum risk Bayesian rule. Normalized relative RBC was used to represent the characteristic of the observation of the samples, and beta distribution was adopted to approximate the probability density distribution of the variable being faulty or normal. On the basis of the scaling factor, the conditional risk was obtained. This fault diagnosis method can reduce the influence of smearing effect, improve the diagnosis rate, handle faults with a small magnitude and identify multiple process faults.Through the numerical simulation, the Tennessee Eastman process and penicillin fermentation process, the proposed method was verified. The experimental results showed the effectiveness and superiority of the approach, which can enhance the safety and reliability in industrial processes. Comparing the proposed approach with two existing methods, the proposed method can faster and more accurate to identify the faulty.