机器学习加速相场计算的发展及应用  

Development of Phase-Field Computations Accelerated by Machine-Learning and Their Applications

作　　者：苏阳 吴文华 杨志刚 陈浩 SU Yang;WU Wenhua;YANG Zhigang;CHEN Hao(Key Laboratory of Advanced Materials of Ministry of Education,School of Materials Science and Engineering,Tsinghua University,Beijing 100084,China)

机构地区：[1]清华大学材料学院先进材料教育部重点实验室,北京100084

出　　处：《鞍钢技术》2024年第6期50-60,115,共12页Angang Technology

基　　金：国家重点研发计划(2022YFE0110800);国家自然科学基金(52201011,51922054)。

摘　　要：在钢铁材料设计等领域中,相场计算作为一项基础而关键的技术,面临的主要挑战是计算精度与效率的平衡。随着材料科学计算领域的迅速发展,机器学习为提高相场计算的效率和精度开辟了新途径。通过综合评述机器学习在加速相场计算方面的应用,总结了利用机器学习技术求解相场及类似的偏微分方程问题的不同策略和实现方式,并对这些方法的计算结果进行了评估。通过比较分析各种机器学习方法在加速相场计算方面的优势、局限和适用场景,讨论了目前挑战和未来的发展方向,为机器学习加速相场计算研究提供了方向性指导。The computing technique for computations of phase fields,as a fundamental and critical technology in such fields as designing of steel materials,faces the main challenge of balancing calculation accuracy and efficiency.With the rapid development of the computation field for materials science,machine learning has opened up a new way for improving the efficiency and accuracy of phase field computations.By comprehensively reviewing the applications of machine learning in accelerating phase field calculations,different strategies and implementation methods for solving phase fields and similar partial differential equations by using machine learning technology were summarized,and the calculation results by these methods were evaluated.By comparing and analyzing the advantages,limitations and applicable scenarios of various machine learning methods in accelerating phase field computations,the current challenges and future development directions were discussed,which provided directive guidance for carrying out the studies of accelerating phase field calculations by machine learning.

关 键 词：相场计算 机器学习 材料设计 偏微分方程 

分 类 号：TG14[一般工业技术—材料科学与工程] TP3[金属学及工艺—金属材料]