Accurately predicting resistance spot welding (RSW) quality is essential for the manufacturing process. In this study, the RSW process signals of 2219/5A06 aluminum alloy under two assembly conditions (including gap and spacing) were analyzed, and then artificial intelligence modeling was carried out. To improve the performance and efficiency of RSW quality evaluation, this study proposed a multi-signal fusion method that was performed by combining principal component analysis and a correlation analysis. A backpropagation neural network (BPNN) model was optimized using the sine-chaotic-map-improved sparrow search algorithm (SSA), and the input and output of the model were the variables after multi-signal fusion and the button diameter, respectively. Compared with the standard BPNN model, the Sine-SSA-BP model reduced the MAE by 42.33%, MSE by 51.84%, and RMSE by 31.45%. Its R2 coefficient reached 0.6482, which is much higher than that of BP (0.2464). According to various indicators (MAE, MSE, RMSE, and R2), the evaluation performance of the Sine-SSA-BP model was better than that of the standard BPNN model. Compared with other models (BP, GA-BP, PSO-BP, SSA-BP, and Sine-PSO-BP), the evaluation performance of the Sine-SSA-BP model was best, which can successfully predict abnormal spot welds.
Keywords: backpropagation neural network; quality prediction; resistance spot welding; sine chaotic mapping; sparrow search algorithm.