The rotating rectifier is the key part of multiphase annular brushless excitation systems. Nevertheless, the rectifiers often experience faults caused by diode failures, which brings security risks in practice. Accurately diagnosing faults in the rotating rectifier is pivotal for ensuring the safe operation of multiphase annular brushless excitation systems. However, the types of rotating rectifier faults are diverse, and the characteristics of different faults are inherently weak. Traditional mechanism-driven diagnostic schemes offer interpretability but often struggle with precise fault diagnosis. New data-driven diagnostic schemes exhibit speed and accuracy but encounter challenges in training and debugging in practical applications. This paper proposes a hybrid mechanism-data-driven diagnostic scheme for rotating rectifier faults.

Based on the fault mechanism, the frequency domain characteristics of the excitation current after the fault are derived, and the fault characteristic patterns are summarized. Then, thresholds of the mechanism diagnosis model are calculated using finite element simulation data. Extracting the frequency domain characteristics of the excitation current allows the fault mechanism to be clearly described, thus providing a solid foundation for subsequent fault diagnosis. The current waveform under normal operation and different fault conditions can be simulated by adjusting the models, which allows for determining thresholds for various operating conditions.

Then, the fast dynamic time warping (Fast-DTW) algorithm is introduced to calculate the similarity of excitation current time-domain waveforms, subsequently forming a data-driven model combined with the k-nearest neighbors (kNN) classifier. The fast-DTW algorithm can align waveforms of different time lengths and start points to capture subtle differences between waveforms. By combining the fast-DTW algorithm with the kNN classifier, the data-driven model can realize the diagnosis of rotating rectifier faults.

Mechanism-driven and data-driven diagnostic schemes are integrated based on ensemble learning principles. Ensemble learning significantly enhances the overall performance of the model by combining the results of multiple learners. Five mechanism-driven and five data-driven models are established to obtain a final diagnostic result based on the absolute majority voting method. The hybrid diagnostic scheme exhibits the advantages of mechanism-driven and data-driven models, effectively overcoming the limitations of a single-driven model.

Finally, the verification of prototype experiments indicates that the hybrid scheme’s diagnostic accuracy reaches 100%, significantly surpassing single-driven models. Establishing diagnostic models requires offline simulation data, reducing training difficulty and improving practicality on-site. The hybrid scheme maintains a reasonable diagnostic speed while ensuring high accuracy.

In conclusion, the proposed hybrid mechanism-data-driven fault diagnosis scheme combines mechanism analysis and data-driven methods to enhance the accuracy and robustness of fault diagnosis, demonstrating excellent test performance in prototype experiments. The diagnostic approach based on the time-frequency characteristics of the excitation current demonstrates excellent interpretability, achieving accurate fault diagnosis solely through training with simulation data.