Abstract: Resistance spot welding (RSW) of heterogeneous materials is a complex nonlinear physical process,and it is difficult to establish a mathematical mapping model between welding process parameters and nugget diameter.The adaptive evolution algorithm was used to predict nugget diameter by combining with cascade forest.Considering the characteristics of the experimental data of Resistance Spot Welding (RSW) of two-layer unequal thickness heterogeneous plates,the material information was characterized by physical property and chemical composition of the material.The process and material were selected as the quality influencing factors to be the model's input features.Random forest,XGBoost and Light GBM were cascaded into the traditional cascade forest as meta-learners respectively,and the adaptive differential evolution algorithm was applied to the cascade forest's optimal parameters selection.The SHAP theory was introduced to analyze the impact of input features on the nugget diameter.The experimental results showed that the R2、MSE of the cascade forest model optimized by hyperparameters on the test set were 0.9480 and 0.0315 respectively,which were better than the baseline model.The important influencing factors of the nugget diameter were determined,which provided a basis for the design of process parameters for welding of new materials.

Key words: resistance spot welding, cascade forest, SHAP value, differential evolution

摘要： 异质材料电阻点焊是非线性的复杂物理过程,焊接工艺参数与熔核直径的数学映射模型难以建立。鉴于此,提出结合自适应差分进化算法与级联森林进行电阻点焊熔核直径预测。考虑两层不等厚异质板材电阻点焊实验数据的特征,以材料的物理特性和化学成分表征材料信息,并将工艺和材料视为质量影响因素作为模型的输入特征;分别选择随机森林、XGBoost和LightGBM作为元学习器级联到传统级联森林中,利用自适应差分进化算法搜寻最优超参数确定级联结构,引入SHAP理论分析影响熔核直径的因素及其作用。实验结果表明,经超参数优化的级联森林模型在测试集上的R2、MSE分别为0.9480、0.0315,优于基线模型,同时确定了熔核直径的重要影响因素,为新材料焊接的工艺参数设计提供依据。

关键词: 电阻点焊, 级联森林, SHAP值, 差分进化