In complex industrial conditions, fault diagnosis is severely affected by unreliable sensor signals. To address this issue, this paper proposes a multi‑scale wavelet time‑frequency enhancement network for uncertainty‑aware weighted fusion fault diagnosis. Firstly, a multi‑scale wavelet enhancement layer is adopted to extract differentiated fault features from multi‑sensor signals using multiple wavelet bases. Secondly, a KAN‑Fourier attention module is designed to jointly enhance time‑domain transient impulse features and frequency‑domain periodic fault features. Next, the Dirichlet distribution is employed to quantify sensor uncertainty. Combined with the feature distance among sensors, fusion weights are learned adaptively to complete evidence‑level weighted fusion for multi‑sensor evidence. Experimental results on the SDUST bearing dataset show that the proposed method achieves high diagnostic accuracy under different loads. Analysis results verify that the method can dynamically assign weights according to sensor reliability and make full use of complementary information among multiple sensors.

rotating machinery,fault diagnosis,uncertainty,Kolmogorov-Arnold Networks,Fourier transform