Welding quality analysis and prediction based on deep learning

Abstract

With the development of intelligent manufacturing, a certain automobile production line has accumulated a large number of automatic collection of welding process data. But the traditional automobile spot welding process lacks of welding data analysis. Now through artificial intelligence means to analyze data, so as to achieve the prevention of quality defects. Considering the difficulty of welding quality prediction due to the high nonlinearity of welding process and the complex interaction of various factors, a welding quality analysis and prediction model based on deep learning was proposed. Based on the analysis of weld spatter factors, a welding quality prediction model was established by Residual Neural Network (ResNet), taking the characteristic point at the end of metal bonding process, the maximum voltage and the wear state of electrode cap as input parameters. The results show that the accuracy of the model is 88.9%, which provides a basis for automatic control of welding quality.