AWCDN: an interpretable transferable Siamese network model for few-shot fault diagnosis

Abstract In engineering, fault diagnosis faces numerous challenges, including scarcity of fault samples, noise interference, and variations in operating conditions, which significantly degrade the performance of traditional deep learning methods. To address these issues, this paper proposes an Adaptive wavelet contrastive diagnosis network. Firstly, a transferable adaptive laplace wavelet convolution layer is designed, which effectively enhances the model’s generalization ability and transfer learning capability under different operating conditions by dynamically adjusting the kernel function parameters. Secondly, a discrete wavelet transform layer is introduced to convert time-domain signals into the wavelet domain, capturing more refined fault features. Finally, convolutional layers are utilized to extract features from the wavelet-domain signals, and fault classification is achieved through similarity comparison. Experimental results on two bearing datasets demonstrate that the proposed method outperforms seven other advanced deep learning models, exhibiting excellent noise robustness and cross-domain transfer learning capability.