基于组合式机器学习的Cr系合金结构钢的TTT图  

Research on TTT diagram of Cr series alloy structural steel based on combined machine learning

作　　者：任鑫[1] 王浩鑫 樊献金 孙涛 王港 REN Xin;WANG Hao-xin;FAN Xian-jin;SUN Tao;WANG Gang(School of Materials Science and Engineering,Liaoning Technical University,Fuxin 123000,China)

机构地区：[1]辽宁工程技术大学材料科学与工程学院,辽宁阜新123000

出　　处：《材料热处理学报》2023年第5期201-209,共9页Transactions of Materials and Heat Treatment

基　　金：辽宁省教育厅科学研究经费项目(LJ2020JCL027)。

摘　　要：时间-温度-转变(TTT)图是制定钢的优化热处理工艺的重要工具,但是实验测定相当耗时且昂贵,并且计算方法也较为单一。因此,有必要开发一种快速准确预测TTT图的替代方法。本文将Artificial Neural Network(ANN)、K-Nearest Neighbor(IBK)、Bootstrap Aggregating(Bagging)、Random Committee(RC)、Gaussian Processes(GP)、Self-Organizing Maps(SOM)和Random Forest(RF)算法结合起来,形成组合式机器学习(CML)算法,用于预测Cr系合金结构钢的TTT图,其参数包括了合金化元素、奥氏体化温度和时间。使用相关系数(CC)、误差分析(RMSE、MAE)和性能拟合进行验证和筛选模型。本项工作应用CML模型来预测4种Cr系合金钢的TTT图,以评估模型的预测能力。最后,还将CML模型的预测结果与JMatPro软件的预测结果进行了比较。结果表明:与JMatPro相比,CML模型误差值小,预测结果与实际值较为接近。The time-temperature-transition(TTT)diagram is an important tool for formulating the optimal heat treatment process of steel,but the experimental measurement is time-consuming and expensive,and the calculation method is relatively simple.Therefore,it is necessary to develop an alternative method for fast and accurate prediction of TTT diagram.In this paper,the Artifical Neural Network(ANN),K-Nearest Neighbor(IBK),Bootstrap Aggregating(Bagging),Random Committee(RC),Gaussian Processes(GP),Self Organizing Maps(SOM)and Random Forest(RF)algorithms were combined to form a combined machine learning(CML)algorithm to predict the TTT diagram of Cr series alloy structural steel.Its parameters include alloying elements,austenitizing temperature and time.Correlation coefficient(CC),error analysis(RMSE,MAE)and performance fitting were used to validate and screen the model.In this work,CML model was used to predict the TTT diagram of four kinds of Cr series alloy steel,so as to evaluate the prediction ability of the model.Finally,the predicted results of CML model were compared with those of JMatPro software.The results show that,compared with JMatPro,the error value of CML model is small,and the predicted results are close to the actual values.

关 键 词：机器学习 合金结构钢 时间-温度-转变图 算法 

分 类 号：TG142.33[一般工业技术—材料科学与工程]