机器学习技术在材料科学领域中的应用进展

Research Progress of Machine Learning in Material Science

材料是国民经济的基础,新材料的发现是推动现代科学发展与技术革新的源动力之一,传统的实验“试错型”研究方法具有成本高、周期长和存在偶然性等特点,难以满足现代材料的研究需求。近些年,随着人工智能和数据驱动技术的飞速发展,机器学习作为其主要分支和重要工具,受到的关注日益增加,并在各学科领域展现出巨大的应用潜力。将机器学习技术与材料科学研究相结合,从大量实验与计算模拟产生的数据中挖掘信息,具有精度高、效率高等优势,给新材料的研发和材料基础理论的研究提供了新的契机。机器学习技术结合了计算机科学、概率论、统计学、数据库理论以及工程学等知识,计算速度快、泛化能力强,能有效地处理一些难以运用传统实验及模拟计算方法解决的体系和问题。近10年,机器学习在材料科学研究中的应用呈现出爆炸式的增长,尤其在新材料的合成设计、性能预测、材料微观结构深入表征以及改进材料计算模拟方法几个方面,均有着出色的表现。当然,作为一项数据驱动技术,如何获取大量实验数据并将其构建为行之有效的数据集仍是现阶段机器学习技术在材料科学领域应用的热点和难点。本文概述了机器学习技术的基本原理、主要工作流程和常用算法,简述了机器学习技术在材料科学领域中的研究重心及应用进展,分析了机器学习在材料学研究中尚存在的问题,并对未来此领域的发展热点进行了展望。

Materials are the foundation of the national economy,the discovery of new materials gives impetus to the development of modern science and technological innovation.The traditional“trial and error”experimental methods are no longer applicable to the research of modern materials owing to the disadvantages of high cost,long period and great contingency.In recent years,with the rapid development of artificial intelligence and data-driven approach,as a main branch and an important tool of them,machine learning is receiving increasing attention and showing tremendous potential.The integration of machine learning into material science research can greatly improve the precision and efficiency,and provide new opportunities for the research and development of new materials and the study of the basic theory.Machine learning technology combines knowledge of computer science,probability theory,statistics,database theory and engineering.It shows a faster computing speed and good generalization ability,and can effectively deal with some systems and problems difficult to tackle by traditional experiments and numerical simulation.In the past decade,the applications of machine learning in material science research have shown explosive growth,especially in the synthesis and design of new materials,the property prediction,characterization of the microstructure,and the improvement of material calculation and simulation methods.Machine learning will be indispensable in the development of material science and engineering in the future.At present,how to obtain a large number of experimental data and build effective data set is still a hot spot and difficulty in the application of machine learning in the field of material science.This paper outlines the basic principles,workflows and common algorithms of machine learning,briefly describes the research focus and application progress of machine learning technology in the field of materials science,and analyzes the existing problems of machine learning in mate-rials science research.Meanwhile,some hot spots of the material field in the future are pointed out.