WADE-MoE: A Knowledge-Guided Wavelet-Based Multi-Expert Framework for Bearing Fault Diagnosis under High-Noise Conditions

Bearing fault diagnosis faces significant challenges under high-noise conditions, particularly when fault signatures are submerged in intense noise. Existing methods struggle with insufficient coupling between wavelet transforms and deep models, and lack adaptive parameter selection when dealing with such scenarios. This paper presents WADE-MoE (Knowledge-Guided Wavelet-Adaptive Multi-Expert System), a novel framework that achieves accurate bearing fault diagnosis under high-noise conditions by integrating traditional signal processing knowledge with deep learning techniques. The framework introduces three key innovations: 1) A Knowledge-guided Wavelet Denoising Module (KWDM) that incorporates wavelet transform priors into deep learning architectures, enabling end-to-end parameter optimization; 2) A Multi-scale Hierarchical Feature Transformation Network (MHFT-Net) that effectively captures multi-scale time-frequency characteristics of signals; 3) A multi-expert collaborative mechanism that enhances system adaptability to various noise types by integrating multiple expert modules with distinct wavelet properties. Experiments on the Case Western Reserve University bearing fault dataset demonstrate exceptional diagnostic performance under high-noise conditions (-10dB to 10dB SNR), maintaining a 90.51% F1-score even under extreme noise (-10dB), significantly outperforming existing methods. Theoretical analysis and experimental results validate the effectiveness and robustness of our proposed method in high-noise environments.