Bearing fault diagnosis method for variable operating conditions based on improved joint distribution adaptation with small sample learning

Purpose To enhance the accuracy and robustness of bearing fault diagnosis, this study proposes an improved Joint Distribution Adaptation (JDA) method that incorporates joint adaptation of marginal and conditional distributions. The method aims to address the issue of domain discrepancy between source and target domains in bearing fault diagnosis. Design/methodology/approach First, time-domain and frequency-domain features of the bearing vibration signals are extracted. Principal Component Analysis is then applied to reduce the dimensionality of high-dimensional features, constructing diagnostic feature vectors. Second, to overcome the limitation of traditional JDA methods, which use fixed weights for marginal and conditional distribution adaptation, a dynamic weight adjustment strategy is introduced. During the adaptation process, the strategy prioritizes the adaptation of marginal distributions in the early stages to alleviate the imbalance in target domain samples and subsequently focuses on conditional distribution adaptation to improve the classifier’s ability to distinguish fault categories in the target domain. Findings The results demonstrate that, under conditions of limited target domain samples, the proposed method achieves average diagnostic accuracies of 94.5 and 97.5% on two datasets, significantly outperforming traditional JDA methods and conventional classification approaches. Originality/value The improved JDA (IJDA) method is combined with a Relevance Vector Machine (RVM) for classification, and its effectiveness is validated using bearing experimental datasets. The results demonstrate that, under conditions of limited target domain samples, the proposed method achieves average diagnostic accuracies of 94.5 and 97.5% on two datasets, significantly outperforming traditional JDA methods and conventional classification approaches.