Online quality inspection of resistance spot welding for automotive production lines

Abstract

Reliable quality control of resistance spot welding (RSW) is a long-standing challenge, due to random disturbance on automotive production lines. In this paper, a quality evaluation framework is proposed based on dynamic resistance (DR) signals, aiming to accurately predict welding quality. The proposed framework integrates welding process stability with deep learning models. Given the uniform variation pattern of each weld with the same schedule, process stability can be determined based on the reference curve constructed by the low-rank and sparse decomposition method. Subsequently, a one-dimensional convolutional neural network (1DCNN) with channel attention mechanism is developed to further predict welding quality, which can perform channel-wise feature recalibration to enhance the classification performance. Extensive experiments substantiate that the proposed network yields a remarkable classification performance compared with typical algorithms on several RSW datasets collected on an actual production line. This study provides a valuable reference to achieve an intelligent online quality inspection system in the automotive manufacturing industry.