Recognition of penetration state in CMT based on Swin Transformer using erratic arc acoustic

Automatic recognition of weld penetration is one of the approaches to evaluate the performance of cold metal transfer (CMT) welded joints, and the recognition of penetration based on arc acoustic signal has long been a focal point and a major challenge in this field. To achieve automatic recognition of weld penetration, this paper proposes a Swin Transformer based on irregular arc acoustic signals. The weights of this model are trained using this method, and its hyperparameters are optimized during the training process. The results show that the optimized model achieves an accuracy of up to 98.33% on the validation set, significantly outperforming other advanced identification models (CNN-LSTM, VGG16, ResNet50, Vit-16). Finally, the training process of the model is visualized to further verify the reliability and effectiveness of the proposed method. This study expands the application of Swin Transformer in the welding field and achieves remarkable results. Furthermore, the inference speed of the model can reach 10.82 ms, enabling intelligent monitoring of the weld penetration state.