Multi-faults Classification of Gear System Based on Combining Wavelet Packet Transform with Support Vector Machine

By friendly combining the wavelet packet transform (WPT) with support vector machine (SVM), a new multi-damages detection and multi-faults classification method for gear system is successfully proposed. Through testing and analyzing different faults of gear system under different running conditions, collect the vibration testing signal, using wavelet packet decomposition, the signals processed by wavelet threshold denoising are decomposed into several levels signal, then reconstruct the decomposed signal, and calculate the energy of each level signal, which is regarded as characteristic of fault. On this basis, the support vector machine is introduced to damage detection and diagnosis of gear system, establish the classifiers for binary classification and multi-classification, and investigate the solution of binary classification and multi-classification SVM. And then consider characteristics of each level signal as input, use support vector machine to classify them, and compare their results with the results of neural network classification. The study has shown that characteristics of denoised signal are more distinct than the original signal's, and the method of combining wavelet packet transform with support vector machine shows more superiority in classification than simply using SVM. In 300r/min, 900r/min, 1200r/min and 1500r/min running conditions, single fault and multi-faults, such as no fault, tooth root short crack, tooth root long crack, pitch circle short crack, pitch circle long crack, tooth wear, tooth root short crack & tooth wear and multi-faults, can be effectively distinguished and diagnosed, the rates of classification and diagnosis are more than 92%, especially, there is a high recognition accuracy for multi-faults.