Intelligent Fault Identification Based On Wavelet Packet Energy Analysis and SVM

Abstract

In the machinery fault diagnosis field, many new and powerful methods play the important role in improving the veracity and the reliability. Fault feature extract is the premise for the fault diagnosis. Wavelet packet transform (WPT) is a mathematical tool that has a special advantage over the traditional Fourier transform in analyzing non-stationary signals. It adopts redundant basis functions and hence can provide an arbitrary time-frequency resolution. The signals are decomposed into different frequency bands with the WPT, then, the energy percents of every frequency band components are calculated as the fault detection index. In this paper, the fault signals are sampled from one gear with pitting fault. According to the characteristic of the gear pitting signals, decompose the signals with the WPT and the changes of the energy percents in the frequency bands including the gear natural frequency will be used as the fault index. Fault style identification is the other vital issue of the fault diagnosis process. Support vector machines (SVM) is a new general machine-learning tool for classification, forecasting and estimation in small-sample cases. The principle and the process of gear pitting identification using SVM is presented. This paper only shows the availability for pitting fault diagnosis with the integration of the WPT and SVM, but the conclusion is also flexible for other machinery fault style classification.