| It is acknowledged that an effective method for industrial process fault diagnosis can not only identify faults to avoid possible danger in the first time, but also diagnose the types and causes of the faults to prevent their further going on.However, due to the complexity of actual industrial processes, it is difficult to conduct fault diagnosis by means of process mathematical models. Based on the data analysis metrics, data driven approaches are able to mine a lot of useful information recorded during process running,obtaining required knowledge of processes, which is of more practical significance. In this thesis, we use unsupervised multivariate statistical analysis method and clustering algorithm, simply and efficiently handling the fault detection and diagnosis problems for industrial processes. The main contributions are presented as follows.Firstly, the kernel principal component analysis (KPCA) and fuzzy C-means clustering (FCM) algorithms are investigated in details.Further, the KPCA and FCM are combined to formulate the novel KPCA based FCM clustering method.Secondly, the proposed method is used for industrial fault detection and diagnosis. The kernel principal component analysis (KPCA)algorithm is employed to map the input data to a high dimensional feature space, which helps extract the features from the original data,effectively reducing the data dimension and enhancing the processing efficiency. SPE and T2 statistics are utilized to determine whether a fault has occurred so as to identify the corresponding process variable.Subsequently, the FCM is employed to cluster the fault features,implementing the fault diagnosis.Finally, the TE chemical process is used as the test-bed for the proposed method. Taking advantage of KPCA-FCM, the fault features are classified before the online diagnosis of process faults is realized. It is shown that this data-driven algorithm can enjoy effective fault detection and diagnosis of industrial processes. |
PDF Full Text Request