Rolling Bearing Fault Diagnosis under Variable Working Conditions Based on Joint Distribution Adaptation and SVM

Abstract

The traditional fault diagnosis methods for rolling bearing usually require the test data and training data to follow the same distribution, which cannot be always meet in real-world scenarios, since the working condition of rolling bearing is often variable. Hence, to overcome the low performance of fault diagnosis traditional methods for different data distributions, a fault diagnosis approach based on transfer learning is proposed in this paper. And the main idea of our approach is to combine joint distribution adaptation and support vector machine to diagnose bearing faults under variable working conditions. In this research, kernel-JDA is used to reduce the difference between distributions of datasets taking both the marginal and conditional distributions into consideration, while the parameters of kernel-JDA are optimized to improve the performance. Besides, multi-features including time domain features and the relative wavelet packet energy are constructed at first to prepare for fault diagnosis. After mapping the multi-features through kernel-JDA, SVM is utilized to diagnose faults of rolling bearing under different working conditions. In addition, comparison experiments on vibration signal datasets of rolling bearings are carried out to verify the effectiveness and applicability of this approach for both the normal and small sizes of the sample sets.