Intelligent fault diagnosis is a set of methods for the health monitoring of mechanical bearings. To address the problem of insufficient applicability of diagnostic models due to differences in the domain distribution between laboratory data and actual working conditions, this study constructs a complete transfer learning diagnostic system. Firstly, the Hilbert transform technique was introduced to extract time-domain and frequency-domain features, as well as periodic correlations and other indicators; then, three models, i.e., transfer learning (TL), gradient boosting machine (GBM), and random forest (RF), were used to classify the data and compare their accuracy. It was found that TL had the highest accuracy in testing, with an F1 score of 0.9631. In the transfer task of the target domain samples, compared with the direct application of the source domain model with a classification accuracy of 70.3%, the transfer learning method achieved a classification accuracy of 97.6%, and the transfer gain increased by 27.3 percentage points, proving the superiority of the model constructed in this paper. Finally, SHapley Additive exPlanations (SHAP) was used to provide a detailed explanation of the transfer learning model, and the basis for model decision making was revealed through feature importance analysis.
Gao et al. (Sun,) studied this question.