Research on bearing fault feature transfer diagnosis based on balanced distribution adaptation under feature fusion

Abstract

In practical industrial applications, the operating conditions of bearings frequently change, posing significant challenges for reliable fault diagnosis. Traditional machine learning methods, which rely on the assumption of independent and identically distributed samples, often experience a significant decline in diagnostic accuracy under such variable conditions. To address this issue, this paper proposes a bearing fault transfer diagnosis method that combines the Balanced Distribution Adaptation (BDA) algorithm with a Back Propagation neural network (BPNN) classification algorithm. Firstly, time-domain features of the bearing signals are extracted to comprehensively reflect the operational state of the bearings. Principal Component Analysis (PCA) is then utilized to reduce the dimensionality of the high-dimensional features, preserving the main information while reducing computational complexity. Subsequently, the BDA algorithm is employed to align the features of the source and target domains, balancing distribution differences and achieving effective feature space transfer. Finally, the BP neural network classification algorithm is used to classify the transferred features, thereby diagnosing bearing faults. Experimental results demonstrate that, compared to traditional fault diagnosis methods, the proposed approach achieves higher diagnostic accuracy and robustness under different working conditions. This method not only addresses the challenges posed by changing operating conditions but also holds significant practical value, providing a robust and efficient solution for real-world industrial applications such as predictive maintenance and condition monitoring in critical engineering systems.