Quality evaluation by classification of electrode force patterns in the resistance spot welding process using neural networks

Since resistance spot welding (RSW) has become one of the safest and most reliable processes for fabricating sheet metals, many quality estimation methods have been developed to ensure the welding qualities. In this paper, two kinds of quality evaluation method by classification of electrode force patterns using neural networks are proposed in a servo-controlled RSW system. Firstly, experiments were conducted under different welding conditions with various process parameters such as welding currents and electrode forces in order to determine the relations between force patterns and qualities. Secondly, experiments were conducted in order to generate basic data to train the proposed neural networks and finally to evaluate welding qualities through the classification into standard patterns. The proposed learning vector quantization (LVQ) net indicates the fast classification, showing a total success rate of 90 per cent for test data with five standard patterns. The proposed back-propagation (BP) net shows the precise classification with a total success rate of 95 per cent, considering a slightly longer time for classification due to the additional data process time. The results evaluated with the standard welding quality classes show the practical feasibility of the proposed classification methods.