Improved data-driven performance of Charpy impact toughness via literature-assisted production data in pipeline steel

作     者：SHANG ChunLei WANG ChuanJun WU HongHui LIU WenYue CHEN YiMian PAN GuangFei WANG ShuiZe WU GuiLin GAO JunHeng ZHAO HaiTao ZHANG ChaoLei MAO XinPing SHANG ChunLei;WANG ChuanJun;WU HongHui;LIU WenYue;CHEN YiMian;PAN GuangFei;WANG ShuiZe;WU GuiLin;GAO JunHeng;ZHAO HaiTao;ZHANG ChaoLei;MAO XinPing

作者机构：Beijing Advanced Innovation Center for Materials Genome EngineeringInnovation Research Institute for Carbon NeutralityUniversity of Science and Technology BeijingBeijing 100083China State Key Laboratory of Metal Materials for Marine Equipment and ApplicationAnshan 114009China Ansteel Beijing Research Institute Co.Ltd.Beijing 102200China 

出 版 物：《Science China(Technological Sciences)》 (中国科学（技术科学英文版）)

年 卷 期：2023年第66卷第7期

页      面：2069-2079页

核心收录：

学科分类：080706[工学-化工过程机械] 08[工学] 081104[工学-模式识别与智能系统] 0817[工学-化学工程与技术] 0807[工学-动力工程及工程热物理] 0903[农学-农业资源与环境] 0901[农学-作物学] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：supported by the National Natural Science Foundation of China(Grant Nos.52122408,51901013,52071023) financial support from the Fundamental Research Funds for the Central Universities(University of Science and Technology Beijing)(Grant Nos.FRF-TP-2021-04C1,and 06500135) supported by USTB MatCom of Beijing Advanced Innovation Center for Materials Genome Engineering。 

主　　题：data-driven design pipeline steel Charpy impact toughness machine learning 

摘      要：Pipeline transportation is one of the most economical ways to transport crude oil and natural gas over long distances.High toughness is one of the important qualities of pipeline steel to ensure safe transportation,wherein a key factor characterizing toughness is Charpy impact toughness(CIT).In this work,according to the production line data provided by a steel mill and the experimental data collected in literature,two machine learning model construction strategies were proposed.One was based solely on the production line dataset,and the other was based on the production line dataset together with the literature dataset.In these two strategies,the random forest model displayed the best prediction results,the accuracy of strategy I was 0.58,and the accuracy of strategy II was 0.90,wherein literature data effectively improved the CIT prediction accuracy.Finally,an optimized CIT model based on machine learning algorithms was established.The proposed strategy of literature data-assisted production line data provides a new perspective for optimizing and predicting the performance of traditional structural materials.