基于BP神经网络18%~25%Cr奥氏体耐热钢持久性能预测  

Prediction of creep-rupture property of 18%-25%Cr austenitic heat-resistant steel based on BP neural network

作　　者：王双莲 冷心悦 赵名扬 朱昌浩 程聪 刘省 WANG Shuanglian;LENG Xinyue;ZHAO Mingyang;ZHU Changhao;CHENG Cong;LIU Sheng(School of Materials and Chemical Engineering,Hubei University of Technology,Wuhan 430068,Hubei,China;Hubei Provincial Key Laboratory of Green Light Industry Materials,Wuhan 430068,Hubei,China;Hubei Engineering Laboratory of Light-weight Materials and Processing for Automobiles,Wuhan 430068,Hubei,China)

机构地区：[1]湖北工业大学材料与化学工程学院,湖北武汉430068 [2]绿色轻工材料湖北省重点实验室,湖北武汉430068 [3]车用轻质材料与加工湖北省工程实验室,湖北武汉430068

出　　处：《钢铁研究学报》2025年第1期104-116,共13页Journal of Iron and Steel Research

基　　金：湖北省教育厅科学技术研究计划中青年人才资助项目(Q20311412);大学生创新创业训练计划资助项目(S202210500066);湖北工业大学博士科研启动基金资助项目(XJ2021009402)。

摘　　要：为建立不同温度和应力服役条件下18%~25%(质量分数,余同)Cr奥氏体耐热钢的成分-热处理工艺-持久性能的关系,以9种奥氏体耐热钢的成分、热处理工艺、持久试验断裂时间和持久试验温度作为输入参数,持久强度作为输出参数,并利用SHAP值和最大边缘相关性筛选特征,建立基于BP神经网络的18%~25%Cr奥氏体耐热钢持久性能预测模型。结果表明,相比于传统方法,该模型可构建18%~25%Cr奥氏体耐热钢体系在复杂服役条件下的成分-热处理工艺-持久性能的映射关系,实现不同18%~25%Cr成分的钢种的持久性能精准预测。针对具体钢种,结合BP神经网络模型和热力学计算方法可从组织-性能角度进一步筛选和优化其成分组合。To establish the relationship between composition,heat treatment process,and creep performance of austenitic heat-resistant steels with 18%-25%Cr under various temperature and stress service conditions,the chemical composition,heat treatment process,creep test rupture time,and creep test temperature of nine types of austenitic heat-resistant steels were used as input parameters,with creep strength as the output parameter.Using SHAP values and maximum marginal correlation to select features,a prediction model for the creep performance of 18%-25%Cr austenitic heat-resistant steel based on a BP neural network was established.The results indicate that when compared to traditional methods,this model can construct the relationship between the composition,heat treatment process,and creep rupture properties of 18%-25%Cr austenitic heat-resistant steels under complex service conditions,achieving accurate prediction of creep rupture property for 18%-25%Cr steels.For specific steels,the combination of the BP neural network model and thermodynamic calculation methods can further screen and optimize the composition combination of the steel from the perspective of structure-property.

关 键 词：奥氏体耐热钢 BP神经网络 化学成分 热处理工艺 持久性能 

分 类 号：TG142.33[一般工业技术—材料科学与工程]