The Study On Monitoring And Fault Diagnosis Of Large-scale Chemical Process

Online process monitoring and fault diagnosis are important issues for large-scale chemical plants. A new method is proposed for online process monitoring and fault diagnosis based on Kernel Principal Component Analysis, Probabilistic Neural Network and Fault Tree Analysis.First of all, the processing method based on wavelet denoising and KPCA is establised. Based on the data collected from the operation of the system, the wavelet denoising is carried out to improve the quality of the process data. Then the KPCA model is established by using the data of noise reduction. The SPE and control limits can be obtained through calculating to monitor the process condition online. According to the four different kinds of fault monitoring from Tennessee Eastman process, the results shows that the combination of WD and KPCA can be used more effective in the on-line monitoring and find the faults in time.Then, the fault diagnosis method based on probabilistic neural network and fuzzy c-means algorithm is put forward, because the WDKPCA cannot recognize the fault. In the case of few types of fault, the correct rate of diagnosis is very high. However, the diagnosis accuracy will decrease when the large training date are input into the neural network. In order to improve the quality of training samples, Fuzzy C-Means algorithm is introduced to cluster the training sample data. The method can remove the redundant samples and improve the training efficiency of neural network. The results show that the training time of neural network is reduced and the accuracy of fault diagnosis is improved obviously when the sample data is redundant.Finally, the fault trees of important component in system is established. The managers could find out the weak link and fault module of the system and optimize the production process to reduce the probability of failure. When the PNN cannot figure out the faults, the managers can analysis the fault tree and find out the root cause of the failure to prevent the occurrence of faults.