PCA Based Characteristic Parameter Extraction and Failure Recognition Using LS-SVM

An intelligent fault diagnosis method based on principal component analysis (PCA) and least squares support vector machines (LS-SVM) is proposed. The characteristic parameter set is obtained by wavelet packet transform (WPT). And PCA is used to extract the principal features associated with the diagnosing object. Then, the training data set which is reduced from the original parameters are used as inputs to a LS-SVM for founding the classifier. In the paper, the PCA and LS-SVM method successfully realizes the multi-class failure recognition on the centrifugal pump circulation system. The experimental results demonstrate that WPT based characteristic parameters construction method and PCA based feature extraction technology are effectively, and the LS-SVM algorithm using the RBF kernel function had good multi-classification properties.