机器学习在高强汽车钢板工艺中的优化研究  

Optimization Research of Machine Learning in High Strength Automotive Steel Plate Process

作　　者：王俊雄 徐鑫 刘仁东 时晓光 陈家浩 倪常皓 何燕霖[3] 李麟[3] WANG Junxiong;XU Xin;LIU Rendong;SHI Xiaoguang;CHEN Jiahao;NI Changhao;HE Yanlin;LI Lin(State Key Laboratory of Metal Material for Marine Equipment and Application,Anshan 114009,Liaoning,China;Ansteel Iron&Steel Research Institutes,Anshan 114009,Liaoning,China;School of Materials Science and Engineering,Shanghai University,Shanghai 200444,China)

机构地区：[1]海洋装备用金属材料及其应用国家重点实验室,辽宁鞍山114009 [2]鞍钢集团钢铁研究院,辽宁鞍山114009 [3]上海大学材料科学与工程学院,上海200444

出　　处：《鞍钢技术》2025年第2期37-44,54,共9页Angang Technology

摘　　要：为了满足高强汽车钢板定制化研发的需求,采用随机森林、极限梯度提升树、支持向量机、自适应提升树和高斯过程回归五种机器学习方法,对不同产线的590 MPa级汽车钢板的化学成分、工艺参数、力学性能指标等数据进行了研究,以构建抗拉强度和伸长率的预测模型。结果表明,特征参数的数量以及质量对模型预测精度的提高至关重要,对数据量为514条,特征参数为27个的M钢而言,采用自适应提升树方法训练后的模型预测精度R2值可达到0.9以上,对于数据量过少或者关键特征缺失的产线钢种,其预测效果并不理想;基于该钢种组织中残余奥氏体发挥相变诱发塑性的强塑化机理,引入热力学参数特征可有效简化模型。另外,用平行坐标图以及SHAP值评估,探讨了不同工艺参数在模型预测中的重要性,并分析了这些参数对抗拉强度和伸长率的影响程度,为机器学习方法在高强汽车钢板产线工艺优化中的应用提供了理论参考。In order to meet the demand of customized research and development of high-strength automobile steel plates,five machine learning methods,including random forest regressor,extreme gradient boosting,support vector machine,adaptive boosting and Gaussian process regressor,were used to study the data of chemical composition,process parameters and mechanical properties of 590 MPa automobile steel plates from different production lines,so as to build a prediction model of tensile strength and elongation.The results showed that the quantity and quality of characteristic parameters were very important to improve the prediction accuracy of the model.For M steel with 514 data and 27 characteristic parameters,the prediction accuracy R2 value of the model trained by adaptive boosting method could reach above 0.9,and the prediction effect was not ideal for production line steel with insufficient data or a lack of key features.Based on the strengthening and plasticization mechanism of retained austenite in the steel structure,the model could be effectively simplified by introducing thermodynamic parameter characteristics.In addition,the importance of different process parameters in model prediction was discussed by using parallel coordinate diagram and SHAP value evaluation,and the influence degree of these parameters on tensile strength and elongation was analyzed,which provided theoretical reference for the application of machine learning method in process optimization of high-strength automobile steel plate production line.

关 键 词：机器学习 汽车钢板 工艺优化 力学性能 

分 类 号：TG1[金属学及工艺—金属学] TP3[自动化与计算机技术—计算机科学与技术]