307 / 2025-06-22 15:32:37

Cascaded wavelet-driven global interpretable model for bearing fault diagnosis

Interpretable, Morlet wavelet filter, Attention mechanism, Bearing fault diagnosis

Deep learning has been widely applied to bearing fault diagnosis due to its outstanding performance in feature extraction and pattern recognition. However, the lack of interpretability in deep models significantly limits the trustworthiness of their diagnostic results. To address this issue, this paper proposes a cascaded wavelet-driven global interpretability model for bearing fault diagnosis. Specifically, a multi-scale cascaded Morlet wavelet filter bank is designed in the frequency domain to effectively extract key frequency features from bearing vibration signals based on physical priors. A channel attention mechanism is further introduced to adaptively weight and fuse multi-channel signals, dynamically emphasizing the relative importance of each channel and enhancing interpretability during signal processing. Moreover, the model employs statistical features with clear physical significance for quantitative analysis, ensuring the physical traceability of the extracted features. Extensive experiments conducted on both public and private datasets demonstrate the superior interpretability, reliability, and diagnostic performance of the proposed method.