Classification of the Resistance Spot Weld Failure Mode Using Convolutional Neural Network

Abstract

The resistance spot welding (RSW) is a comprehensive used process in the automotive industry. However, in the production line the quality of the weld is normally roughly assessed from its appearance by human. In this work, we propose the prediction of the quality of the resistance spot weld by using the convolutional neural network (CNN). The architecture of CNN applied in this work was MobileNetV3. The quality of the weld in this work was the failure mode of RSW determined by strength of the weld by tensile shear test. The external apparent image of welds was used as information for predicting the quality of the welds. For building the data set, the RSW was conducted with specific welding conditions and the heat trace image of welds was captured. The parameters of the CNN were trained from the apparent image of the welds and their failure mode. As a result, the CNN model was able to predict the class of the failure mode of the RSW from the welds image with satisfactory F1-score of 94.32% for unseen validation data set.