Optimization of BP neural network based on Harris Hawk algorithm to predict resistance spot welding quality

Resistance spot welding (RSW), which is still extensively used in the automobile, aerospace, and other equipment production industries, depends critically on the quality of the weld. The size of the welded joint nucleus diameter was used as the output of the spot welding quality monitoring model, with welding current, welding voltage, welding air pressure, and welding time selected as input parameters. This study developed a neural network-based online monitoring model for resistance spot welding. The Harris Hawk Herd Algorithm Optimized BP Neural Network (HHOBP) neural network model, which is the basis for the spot welding quality prediction model, has an R² coefficient value of up to 0.9981 and may somewhat predict the quality of welded connections. Compared with PSO-BP, GWO-BP, CS-BP, GA-BP, traditional BPNN and support vector machine models, HHOBP has achieved 93.06% ~ 96.15% (MAE), 99.43% ~ 99.81% (MSE), 92.18% ~ 95.46% (RMSE) and 87.54% ~ 96.83% (MAPE) in different error indicators, reflecting excellent modeling capabilities.