Fault Diagnosis On Tennessee-Eastman Process

With the expending of the industrial process scale, the complexity of the industrial system is increasing so that ensuring the efficient and safe of industrial system has became the key of improving product quality and economic efficiency. To ensure the efficient and safe of industrial system, we must establish an effective process monitoring and fault diagnosis. In the process monitoring, we can capture and store large amounts of measurement datas. Through the process of those datas, we can get the state of industrial system and timely process of industrial system fault. In this context, fault diagnosis is gradually developing.TCP contains large number of variables, and some of these variables is related. So that efficiency is low if we use all of the TEP measurement datas. We use Principle Component Analysis(PCA) to detect the fault. Through the result of PCA fault diagnosis, we see that traditional PCA can not identify system's fault accurately. To get better result, we use the Kernel Principle Component Analysis(KPCA) and we compare the different results of PCA and KPCA.For fault identification and fault classification, we use Support Vector Machine(MVC) method. Basing on VC dimension of statistical learning theory and structural risk minimization, SVM can find best compromise between complexity of model and learning ability on limited sample information, so that get the best ability to promote. On fault identification and fault classification's application, SVM has achieved good results.In the simulation, false alarm rates and missing alarm rates of PCA is very high in fault diagnosis. So PCA can not correctly detect whether TEP occurs fault. But KPCA's result has been greatly improved and KPCA can apply to TEP fault diagnosis. SVM recognize the variables of causing faults and class of faults correctly. So SVM achieve good results.