A Data-Driven Few-Shot Fault Diagnosis Method for Rolling Bearings

Introduction: To address the severe sample imbalance problem in rolling bearing fault diagnosis, where normal samples are abundant while fault samples are scarce, this study proposes an intelligent diagnostic method based on simulation–experimental data fusion to enhance diagnostic accuracy under small-sample conditions. Materials and Methods: A multi-source bearing dataset is constructed by fusing high-fidelity simulated vibration signals from a dynamic fault model with experimental data. To enhance feature extraction, a Particle Swarm Optimization-based adaptive Variational Mode Decomposition (PSO-VMD) method is developed to automatically optimize key parameters. To reduce distribution discrepancies between simulated and experimental data, a sample entropy-based cross-domain alignment strategy is introduced. The method is validated on the Case Western Reserve University bearing dataset. Results: Experimental results indicate that the proposed method achieves a diagnostic accuracy of 98.58%, representing a 7.53% improvement over using experimental data alone, along with superior numerical stability. Discussion: The method demonstrates superior robustness and numerical stability across different fault categories and small-sample scenarios. Although the quantitative results are obtained from a specific benchmark dataset, the proposed framework is not limited to a particular bearing type. By adjusting simulation parameters such as bearing geometry, rotational speed, and load conditions, the method can be extended to other bearing systems without altering the overall diagnostic framework. Conclusion: The results confirm that simulation–experimental data fusion combined with PSOVMD feature extraction and entropy-based cross-domain alignment provides an effective solution to small-sample and data imbalance challenges in rolling bearing fault diagnosis. The proposed method offers high diagnostic accuracy and strong industrial applicability, and has entered the stage of formal patent protection.