A cross-domain fault diagnosis method for vehicle motors under variable operating conditions based on attention-guided graph convolutional networks

As a critical component of the automotive powertrain system, the performance status of electric motors directly impacts vehicle operational safety. During vehicle operation, sensor signals are frequently subject to noise interference, while motors typically operate under varying conditions, posing significant challenges for fault diagnosis. To address these issues, this paper proposes a cross-domain fault diagnosis method based on attention-guided graph convolutional networks, effectively countering interference from variable operating conditions. First, to tackle noise in extracted signal data, time-frequency domain feature fusion is employed to capture signal characteristics from both temporal and spectral perspectives, comprehensively extracting useful information. Subsequently, a wavelet kernel convolution layer is introduced, leveraging the multi-resolution properties of wavelet transforms to enhance feature extraction capabilities. Second, a graph-generation structure based on attention mechanisms is employed. This structure combines autoregressive moving average filter graph convolutions with multi-order graph convolutions based on Chebyshev polynomials to achieve further feature extraction and capture multi-scale information. Finally, the advancement of this method has been verified through a series of experimental cases, which demonstrates superior performance compared to other models across various cross-domain tasks.