Fault Diagnosis and Analysis of Rotating Machinery Bearings Based on JDA-SVM

Bearing faults are primary factors affecting rotating machinery safety. Under the condition of limited data samples of laboratory rotating machinery, traditional fault diagnosis methods have problems of low accuracy and insufficient persuasiveness. To expand the experimental samples and improve the diagnostic accuracy, this paper proposes the JDA-SVM fault diagnosis method integrating Joint Distribution Adaptation (JDA) and Support Vector Machine (SVM). Firstly, the vibration signal feature vectors of bearings are extracted through time-frequency domain analysis; then, the feature vectors extracted from the source domain (the public dataset of the Case Western University.) and the target domain (rotating machinery) are adapted by JDA. The adapted feature vectors are input into the SVM pattern recognition model to construct an intelligent classification model for bearing faults. Experiments show that, when the training data is insufficient, the mis-detection rate of the improved SVM (JDASVM) based on joint distribution adaptation significantly decreases. JDA-SVM provides a stable solution for cross-domain joint diagnosis problems. The feature transfer and classifier joint optimization algorithm proposed herein enhances visual fault diagnosis accuracy for rotating machinery operational status, establishes a novel theoretical framework for machinery diagnosis and transfer learning, and provides practical solutions for subsequent industrial equipment maintenance.