The Method And Application Of SDG Fault Diagnosis Based On Principal Component Analysis

Signed Directed Graph (SDG) model as a fault diagnosis methods which is not depended on accurate mathematical can be used to express complicated cause-effect relationship, and has the capacity of containing large-scale potential information. When used for fault diagnosis, the SDG not only has good completeness, excellent robustness and strong adaptability, but also can provide fault propagation paths and fault explanation. In recent years, SDG fault diagnosis method develops rapidly, and gets the extensive attention of scholars, and has achieved good application effect in petroleum and chemical industry. But the fault diagnodis method based on SDG model inevitably has the shortcomings of multi-meanings reasoning in qualitative method, which can result in low resolution, and there is other problems to be solved:determing the threshold limit too hard, relying on the model too much, slow diagnosis rate of large systems, bad real-time and so on. So it is necessary to integrate other fault diagnosis methods with SDG, The combine of the advantages of the two methods can reduce the diagnosis time and increase the automation degree of the diagnostic process.Principal Component Analysis (PCA) is a multivariate statistical method. It is suitable for handling the high dimensional, noisy and highly relevant data because of the superiority of its application of data compression, and also the method does not need to master the precise mathematical model of the process. The PCA model can be established according to the process historical data under normal operating conditions of system, and can be used to process monitoring and fault diagnosis of system variables for. However, traditional diagnostic methods based on PCA is difficult to find the real fault sources from a number of abnormal variables, especially when multiple fault roots exist. That is because the PCA method does not have the ability of derive the relationship between the abnormal variables and reason fault evolution path.In order to solve the problems exsit in SDG fault diagnosis, the fault monitoring and contribution methods of the PCA are introduced into SDG fault diagnosis in this paper, and the following work is done:(1) In order to improve the precision of the SDG model, avoid the slow diagnosis speed of large-scale system and improve the capabilities of fault diagnostic, a series of measures is used to simplify the SDG system model. First, according to the steady-state analysis, we can identify the main branch which have a greater impact on the proliferation of system fault, and then remove the secondary branch with less effect which greatly simplifies the SDG model. Then, speed format outputed by controller can effectively deal with variation issue of node symbol in the non-unidirectional transition phase.(2) When multiple faults occur simultaneously in the industrial process, the common DCS system can only give a simple alarm, and the operator is difficult to determine which one is the real root cause immediately, which would greatly reduce the the opportunity of correct operation and increase the risk of system operation.For this the paper propose a SDG multi-source fault diagnosis algorithm,which is used to find fault roots by the reasoning of the abnormal variable in the SDG model in the case of multiple fault roots.(3) A fault diagnosis method which combined PCA and SDG is proposed. Take the TEP system for example, frist the SDG model and the PCA model is established, then the PCA method is used to monitor all the process variables. When the fault occurs, the variables state which is obtained according to the contribution chart is marked on the SDG model, and finds possible fault sources by backward reasoning in SDG model. The simulation results show that the methods improve the diagnosis rate and resolution under the premise of retained the many advantages of SDG fault diagnosis.