基于NSGA-Ⅲ-EBM模型的电阻点焊接头性能可解释预测  

Interpretable prediction of joint performance inresistance spot welding based on NSGA-Ⅲ-EBM model

作　　者：王恒 杨凯[1] 何奕程 黄海松[1] 陈家兑[1] 高鑫[1,3] WANG Heng;YANG Kai;HE Yicheng;HUANG Haisong;CHEN Jiadui;GAO Xin(Key Laboratory of Advanced Manufacturing Technology of the Ministry of Education,Guizhou University,Guiyang 550025,China;Wenzhou Polytechnic,Wenzhou 325000,China;School of Mechanical Engineering,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学现代制造技术教育部重点实验室,贵州贵阳550025 [2]温州职业技术学院,浙江温州325000 [3]贵州大学机械工程学院,贵州贵阳550025

出　　处：《计算机集成制造系统》2025年第4期1215-1227,共13页Computer Integrated Manufacturing Systems

基　　金：国家自然科学基金资助项目(52265062);贵州省科技计划资助项目(黔科合基础-ZD[2025]099,黔科合基础-MS[2025]620,黔科合支撑[2023]一般302);大学生创新创业训练计划资助项目(gzugc2023032,gzugc2023033)。

摘　　要：基于焊接过程信息和机器学习模型的质量预测方法,是实现动力锂电池组焊接接头性能可靠评估的主要途径。为解决传统机器学习模型存在的超参数选择不合理和预测结果可解释性差等问题,建立了锂电池电阻点焊过程信息数据集,构建了接头性能预测机器学习模型,对比分析了不同机器学习模型对电阻点焊小样本数据集的预测性能;基于第三代非支配排序遗传算法(NSGA-Ⅲ)提出了NSGA-Ⅲ-EBM模型,研究了NSGA-Ⅲ-EBM模型对不同特征数据的泛化性,并对输入特征进行了全局解释和局部解释分析。结果表明,针对焊接接头的熔核直径以及拉伸剪切载荷的预测,EBM模型相较于MLP、MLS-SVR和XGBoost模型具有更好的预测性能,在测试集上的平均RMSE、R^(2)分别为2.4127、0.8466;采用NSGA-Ⅲ进行超参数优化后的NSGA-Ⅲ-EBM模相较于未优化的EBM模型,在测试集上的平均RMSE和R^(2)分别提升了17.2%、2.1%;此外,还确定了影响接头性能的重要特征,为焊接工艺参数的动态调整提供了依据。Based on the welding process information,the machine learning model quality prediction method is the main way to realize the reliable assessment of the performance of welded joints in power lithium battery packs.To address the issues of unreasonable hyperparameter selection and poor interpretability in traditional machine learning models,a process information dataset for lithium battery resistance spot welding was established.A machine learning model for joint performance prediction was constructed,and the prediction performances of different machine learning models on a small dataset of resistance spot welding were compared and analyzed.The NSGA-Ⅲ-EBM model was proposed based on the third-generation Non-dominated Sorted Genetic Algorithm(NSGA-Ⅲ).The generalizability of the NSGA-Ⅲ-EBM model to different feature data was investigated,and the input features were analyzed both globally and locally.The results indicated that,for predicting the welded joint’s kernel diameter and tensile shear load,the EBM model outperformed the MLP,MLS-SVR,and XGBoost models,achieving an average RMSE and R^(2)of 2.4127 and 0.8466,respectively on the test set.After hyperparameter optimization using NSGA-Ⅲ,the NSGA-Ⅲ-EBM model improved the average RMSE and R^(2)on the test set by 17.2%and 2.1%,respectively,compared to the unoptimized EBM model.Additionally,important features that affected joint performance were identified,providing a basis for the dynamic adjustment of welding process parameters.

关 键 词：电阻点焊 接头性能 可解释模型 超参数优化 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TM912[自动化与计算机技术—控制科学与工程]