Research On The Fault Diagnosis Method Based On MBKPCA And PSDG

Abstract:Fault diagnosis is an important problem in the industrial production especially for the process industry, there has been much research devoted to this subject in the past decades. The fault diagnosis method based on SDG is capable of expressing the complex causal relationship and containing the large-scale potential information, then the method is of categoricalness and adaptability, meanwhile can provide the fault propagation path and the explanation of the fault evolution, therefore it obtains the widespread attention of scholars. However, as a pure qualitative method the fault diagnosis method based on SDG still has some disadvantage such as low resolution of the diagnosis result, low accuracy and reliability of the model. In order to solve those problems we combine the transfer entropy, MBKPCA and the probability theory with SDG to do the fault diagnosis method research, it is very significance for science and application. The main research work and innovation achievements of the dissertation are as follows:(1) Aiming at the process knowledge is not complete, false and the lack of the precise mathematical model in the traditional modeling method, we introduce the transfer entropy into the SDG modelling. Considering the transfer entropy that can quantify the dependency between variables and detect the direction between them, by making full use of the process historical data the variable node chain model can be constructed based the transfer entropy, then the SDG model can be obtained. The accuracy and reliability of the SDG model can be improved by using the proposed method.(2) Aiming at the characteristic that the relation between variables is complex and the fault diagnosis is difficult in the large-scale industrial process, a fault diagnosis method based on MBKPCA and SDG is proposed. First, by proposing a partition strategy based on SDG and priority, the process is divided into multiple blocks according to the strong connected component as the highest priority, the multiple input or output degree node group as the second priority and the node chain as the lowest priority; on that basis, MBKPCA is used for the process monitoring. If the fault is detected, MBKPCA will determine which block the fault occurs in, then SDG is triggered to complete the fault location in the fault block. The proposed method could improve the accuracy and rapid of the fault diagnosis for the complex industrial process by overcoming the disadvantage of the incomplete fault isolation of MBKPCA and combination explosion in SDG reasoning process.(3) For the existence of a positive feedback loop, it’s unable to use the above method to isolate the fault completely, an improved method based on PSDG is further proposed that uses a new measure to open the cyclic structure and determine the branch probability, according to the posterior probability calculated the possible root cause of the fault can be sorted, then the fault isolation is made according to the probability sort order, the completeness of the fault diagnosis process based on SDG is ensured.The simulation research on Tennessee Eastman process is performed to indicate the effectiveness of the proposed method. Figures(37), tables(7), references(80).