Fault Diagnosis of Motor Roller Bearing Based on Relative Wavelet Energy and Decision Tree

Wavelet transform is often used in fault diagnosis of rolling bearings. However, in practical applications, the number of layers decomposed by wavelet transform needs to be manually specified. If there are too many layers, the corresponding features will also increase, which increases the difficulty of feature selection. To solve this problem, this paper proposes a fault diagnosis method based on relative wavelet energy and decision tree algorithm. This method has strong interpretability, which is mainly reflected in the following two aspects: (1) Establishing classification rules that is easy to be understood with clear tree structures. (2) The feature importance index can be used to quantify the importance of each feature and screen important features. The results show that: when the number of layers decomposed by wavelet increases to more than 6 layers, the accuracy rate will increase to more than 95%. In addition, due to the pre-pruning strategy in the decision tree, the decision tree actually uses fewer features, and the features with higher feature importance are selected when building the decision tree, which avoids the complicated tree structure and increases its interpretability.