Application Of Dynamic Uncertain Causality Graph To Dynamic Fault Diagnosis And Stats Prediction In Chemical Processes

Chemical processes are massive, complex and high-risky. It is important to diagnose the fault online in order to avoid losses of economy and lives. This article applies a model named Dynamic Uncertain Causality Graph (DUCG) to solve this problem. As a new model, DUCG graphically represents the uncertain causalities of the process system compactly, and provides probabilistic diagnosis results accurately. DUCG can not only reduce the scale and complexity by simplifying the DUCG graph online according to the observed evidence, but also deal with incomplete knowledge and uncertain evidence.The contents of this research include:①The fault diagnosis of chemical processes using DUCG is achieved. Some special technical treatment is developed to deal with the vibrational signals that makes the DUCG system have a widely scope of application.② The well-known Tennessee Eastman (TE) simulator is taken as the experimental platform to test the functions of DUCG methodology and software.54 variables and 114 causalities are included in the constructed DUCG knowledge base. The diagnosis of all the 20 failures simulated by TE are performed. The correct diagnosis rate is 100%. In comparison, the mean correct diagnosis rate with well-known Bayesian Network (BN) is reported as 79.71%.③Based on the fault diagnosis result, the states which will change to the abnormal in the future are predicted.Though the predictive result is not exactly same as the real result, most of the abnormal states which real occurred are predicted successfully, the prediction accuracy rate is about 70%.