Improved Support Vector Machine And Its Application In Fault Diagnosis

Fault diagnosis technique is an effective and important method to improve the safety and reliability of industrial process. This paper introduces some fault diagnosis method and a typical chemical process:Tennessee Eastman Process. Then PCA method and its application in fault diagnosis of TE process is presented. The result is not satisfactory because TE process is a complex process and PCA is a linear method. While SVM method in fault diagnosis of TE process performs better for it is a intrinsically non-linear method. But the performance of this method can still be improved.The focus of this paper is to make the improved SVM method performs better in fault diagnosis. Though it is a popular method of machine learning, there is no unified standard in selection of kernel function and parameters of SVM method. In order to reach both strong learning ability and generalization ability, a hybrid kernel function is constructed and Particle Swarm Optimization (PSO) algorithm is applied to optimize the parameters of SVM. Results show that the improved SVM method performs much better than PCA method in fault diagnosis of TE process and better than the basic SVM method too.