Modified DWT for Feature Extraction of Bear Failure Vibration Signal

Abstract

In this paper, bear fault is automatically diagnosed by using pattern recognition. To improve the resolution of lower frequency part, we introduce scale factors to discrete wavelet composition (DWT). The modified DWT combined with high order cumulates are used for vibration signal feature extraction. Besides we use principle component analysis to reduce dimension of the feature data. This feature extraction method has a lower dimension and a higher resolution for lower frequency parts. Therefore it can not only reveal the characteristics of non-linear relationship between amounts of features, but also help to improve the speed and accuracy of classification. Finally neural network algorithm is used for fault classification. Result shows that our method can accurately and efficiently identify the type of bearing failures.