Quality Estimation of Resistance Spot Welding by Using Pattern Recognition With Neural Networks

Abstract

A quality estimation system of resistance spot welding has been developed using a dynamic resistance pattern. Dynamic resistance is monitored in the primary circuit of the welding machine and is mapped into a bipolarized vector for pattern recognition. The Hopfield neural network classifies the pattern vectors and utilizes them to estimate weld quality. Weld strength measurements have been made to examine the performance of the estimation system. Good agreement is obtained between the classified results and tensile-shear strengths. For a better understanding of the estimation process of the network, an example in which the dynamic resistance is classified into the stored pattern is also illustrated.