The Feature Selection in Rolling Bearing Fault Diagnosing Based on Parts-Principle Component Analysis

In this study, PCA (principal component analysis) was used to select features and eliminate the redundancy features in process of rolling bearing fault monitoring. And then a new method was mentioned out to optimize the feature space with P-PCA (parts principal component analysis), which needs to deal with the data of each fault categories with PCA firstly, and then reconstructed the feature space with parts principal components that were got previously. Then recognize the rolling bearing fault patterns based on artificial neural network. The result of experiment indicate that compared with the PCA, the P-PCA avoid the interfering between features which belong to different fault patterns. Which make new feature space contains more useful information, decline the training error rate of artificial neural network, and raise the speed and accuracy of fault pattern recognizing.