Small sample fault diagnosis method based on dual convolutional kernel feature fusion and channel attention weighted temporal convolutional network (DCK-CAM-TCN)

In actual industrial environments, equipment failures often occur sporadically during operation, resulting in insufficient labeled data for training. To address the issues of difficult feature extraction and poor generalization caused by insufficient data in small-sample fault diagnosis, a small sample fault diagnosis method based on dual convolutional kernel feature fusion and channel attention weighted temporal convolutional network (DCK-CAM-TCN) is proposed. Firstly, dual convolution kernels are employed to extract signal features, with the large kernel capturing low-frequency components and the small kernel extracting additional features to enhance the network's expressiveness. Secondly, the channel attention mechanism adaptively adjusts the feature responses of each channel, enabling the network to focus on the most informative and relevant features while suppressing unimportant ones. Finally, the Temporal Convolutional Network (TCN) is utilized to capture dependency features within long time series, further improving the model's ability to process sequential data. Experimental results demonstrate that the DCK-CAM-TCN model significantly outperforms traditional Convolutional Neural Networks (CNNs) and other comparison models in small-sample scenarios. The results indicate the significant advantages of the DCK-CAM-TCN model in small-sample fault diagnosis.