Answers are grounded in this paper and cite their sources.
Air-gap eccentricity in permanent magnet synchronous generators produces non-stationary vibration responses with strong cross-band coupling, making fault-sensitive features difficult to extract. This study proposes a mechanism-guided diagnosis framework combining VMD-STFT-RGB representation with a multi-scale attention-enhanced ConvNeXt network. The electromechanical relationship among air-gap permeance distortion, unbalanced magnetic pull and structural vibration is first analyzed. VMD is then used to separate bearing-housing vibration signals into low-, middle- and high-frequency components, which are transformed by STFT and fused into RGB time-frequency images. The proposed MSA-ConvNeXt integrates multi-scale depthwise convolution, squeeze-and-excitation attention and cross-stage residual fusion. Under repeated five-fold acquisition-batch-wise cross-validation, the method achieved an accuracy of 97.86 ± 0.36%, a macro-F1 score of 0.9781 ± 0.0037 and a macro-AUC of 0.9885 ± 0.0032. The accuracy remained 93.18 ± 0.58% under 5 dB noise and reached 91.63 ± 1.13% on unseen compound eccentricity configurations. These results demonstrate the effectiveness, stability and robustness of the proposed method.
No takes yet. Share an insight, caveat, or question.
Xu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: