Monitoring weld penetration of laser-arc hybrid welding joints without full-penetration requirement based on deep learning

Abstract

In-situ monitoring weld penetration of laser-arc hybrid welding (LAHW) joints without full-penetration requirement is challenging since no back reinforcement occurs. The novelty of this study is to develop convolutional neural network (CNN) models for real-time monitoring of weld penetration in LAHW based on front weld pool images. A dataset, including front weld pool images and corresponding weld penetration labels determined by weld cross-sections, is established by conducting weld tests under different laser powers. CNN classification models, i.e., AlexNet, ResNet18, and ResNet50, are trained and validated by the dataset. The ResNet18 model can classify four weld penetration states with the highest accuracy, achieving 99%. Then, a regression model is constructed to predict the weld penetration depth via the ResNet18 model parameters. The comparison between predicted and actual penetration depths indicates that 98.8% of the errors are less than 0.5 mm, and 61.43% are less than 0.1 mm.