Monitoring of back bead penetration based on temperature sensing and deep learning

Abstract

The monitoring of back bead penetration has always been an important topic in welding field, the vision inspection and arc sound recognition are main methods to monitor welding penetration state. In this paper, an innovative method based on temperature sensing and deep learning is proposed to monitor weld bead penetration. Firstly, it is verified that the distribution of welding temperature field is separable under various welding penetration states. Secondly, a region of interest (ROI) in the welding heat affected zone near molten pool is identified for temperature field detection. Taking the temperature field image of ROI as the input, the prediction model for weld bead penetration is built on the basis of deep residual network. Finally, the generalization performance of the proposed method is verified. The experimental results indicate that detection precision of the model for weld bead penetration is higher than 99%, and the generalization performance is strong.