Abstract: In automotive body welding, the quality of resistance spot welding is primarily determined by the shear strength that the weld nugget can withstand. However, measuring this shear strength requires destructive testing, which is not only inefficient and costly but also impractical for inspecting welds on actual production parts. Therefore, quality inspection is limited to sample testing using coupons or manual ultrasonic spot checks. To enable real-time online monitoring of weld quality, this paper designs multiple prediction models for spot welding strength. The predictive performance of these models is compared based on metrics such as the adjusted coefficient of determination (R 2 ) and mean absolute percentage error (MAPE). The genetic algorithm-optimized backpropagation neural network model is ultimately selected. This model achieves an adjusted R 2 of 0.992 8 and a MAPE of 2.292%, demonstrating excellent performance and contributing to the overall improvement of automotive body quality

Key words: resistance spot welding; genetic algorithm-optimized neural network; intelligent optimization algorithm; weld shear strength prediction; prediction model

摘要： 在汽车车身焊接过程中，电阻点焊的质量主要通过控制焊点能承受的拉剪力控制。但 是检测焊点能承受的拉剪力需要进行破坏性实验，不仅效率低、成本高，且产品件上的焊点 是无法进行此类破坏性的检验，只能用试片进行样本抽检或者通过超声波进行人工抽检。为 能实现实时在线监测焊点质量，文章设计了多种点焊强度的预测模型，对比各模型在调整拟 合优度（R 2）、平均绝对百分比误差（MAPE）等性能指标上的预测效果，最终选择了遗传算 法优化反向传播神经网络模型。该模型的 R 2、MAPE 分别达到了 0.992 8、2.292%，效果良好， 有助于汽车车身质量的整体提升。

关键词: 电阻点焊；遗传算法优化神经网络；智能优化算法；焊点拉剪力预测；预测模型