Research on recognition methods of spot-welding surface appearances based on transfer learning and a lightweight high-precision convolutional neural network

摘要

Resistance spot welding is the most commonly used welding method in the welding process of automotive body-in-white manufacturing, but the appearance quality of the welding spot still relies on manual inspection, which is inefficient and error-prone. To this end, two methods based on deep learning are proposed to recognize welding spot appearances in this paper. In the first method, a practical convolutional neural network (CNN) model is quickly obtained by fine-tuning the VGG net. In the second method, the Release-Compression (RC) block is designed to fully utilize the power of convolution operation and greatly reduce the parameter number, and the information retention strategies are proposed to optimize the bottom and top of the network, so an ad-hoc CNN model named RswNet is obtained by combining RC block and information retention strategies. Experiment results show that the accuracies of the proposed two models are both higher than existing models, and RswNet has the higher accuracy and its parameters are reduced by more than 56 compared with existing models.