Study on Nugget Diameter Prediction of Resistance Spot Welding Using an Artificial Neural Network

Abstract

Resistance spot welding, which has the advantages of low cost and high productivity, is the most common method used in the automobile industry for joining steel sheets. However, in practice, resistance spot welds are typically tested for welding quality using destructive rather than non-destructive inspection methods because of their lower cost. However, in destructive inspection, quality defects can be found only after the completion of the process. Accordingly, several studies are currently being conducted to predict the quality of welding in real time. Welding quality is determined by the diameter of the nugget, and its size depends on several independent variables. In this study, a linear regression model and artificial neural network model were constructed to predict the nugget diameter. An electric power pattern was obtained from the results of a welding experiment, and nine types of electric power characteristic values were extracted from the obtained electric power pattern as independent variables. From the nine electric power characteristic values, six having the highest correlation with the nugget diameter were determined as final independent variables through correlation analysis. The linear regression model was constructed using multiple linear regression analysis, and the artificial neural network model was built using a deep neural network model with two hidden layers and nodes of 64 and 16. In this study, the error between the actual measured and predicted nugget diameters was taken as 0.2 mm or less as a good predictive value. When the linear regression model was used to predict the nugget diameter, only approximately 36% were predicted well. By contrast, when the artificial neural network was used, approximately 86% were predicted well. Thus, the artificial neural network model yielded better results. It was determined that with more welding data and information on steel types, the proposed welding quality prediction system could be improved.