Research on automatic fault diagnosis based on time -frequency characteristics and PCA-SVM

Abstract

According to the operation of the automaton transient impact, nonlinear, non-stationary signal, a method which is based on the time-frequency characteristics and PCA-SVM automaton fault diagnosis is proposed. Firstly, this paper uses statistical analysis and overall empirical mode decomposition method to construct high dimensional mixed domain initial feature vector from the characteristics of different indicators, and characterized by different description of the automaton fault analysis of different angles, then through principal component analysis, dimensionality reduction and feature extraction, the high dimensional variables changes into less dimensional independent eigenvector. At last, the principal component vectors using PCA to support vector machine, realizes the fault recognition. The experimental results show that the hybrid domain initial feature vector can accurately describe the fault characteristics, and the main feature vector extracted by PCA can discard the redundant information and simplify the classifier, and the SVM network diagnosis can improve the classification accuracy.