EI、Scopus 收录
中文核心期刊
胡雅楠, 余欢, 吴圣川, 奥妮, 阚前华, 吴正凯, 康国政. 基于机器学习的增材制造合金材料力学性能预测研究进展与挑战. 力学学报, 2024, 56(6): 1-24. DOI: 10.6052/0459-1879-23-542
引用本文: 胡雅楠, 余欢, 吴圣川, 奥妮, 阚前华, 吴正凯, 康国政. 基于机器学习的增材制造合金材料力学性能预测研究进展与挑战. 力学学报, 2024, 56(6): 1-24. DOI: 10.6052/0459-1879-23-542
Hu Yanan, Yu Huan, Wu Shengchuan, Ao Ni, Kan Qianhua, Wu Zhengkai, Kang Guozheng. Machine learned mechanical properties prediction of additively manufactured metallic alloys: progress and challenges. Chinese Journal of Theoretical and Applied Mechanics, 2024, 56(6): 1-24. DOI: 10.6052/0459-1879-23-542
Citation: Hu Yanan, Yu Huan, Wu Shengchuan, Ao Ni, Kan Qianhua, Wu Zhengkai, Kang Guozheng. Machine learned mechanical properties prediction of additively manufactured metallic alloys: progress and challenges. Chinese Journal of Theoretical and Applied Mechanics, 2024, 56(6): 1-24. DOI: 10.6052/0459-1879-23-542

基于机器学习的增材制造合金材料力学性能预测研究进展与挑战

MACHINE LEARNED MECHANICAL PROPERTIES PREDICTION OF ADDITIVELY MANUFACTURED METALLIC ALLOYS: PROGRESS AND CHALLENGES

  • 摘要: 增材制造是现代高端装备制造领域的革命性突破技术之一. 其中, 增材构件的大批量生产和高可靠应用, 关键在于制造可重复性、质量可靠性与性能可预测性. 而在各向异性组织、广域分布缺陷、深部残余应力和复杂表面粗糙度等诸多因素的共同影响下, 基于传统经验模型和有限数据的增材制造金属力学性能预测效率与准确性面临着严峻挑战. 近年来, 作为大数据与人工智能发展到一定阶段的必然产物, 机器学习(machine learning, ML)方法为有效处理高维物理量之间的复杂非线性关系提供了契机, 在增材制造合金材料力学性能预测领域得到持续关注. 文章综述了机器学习在增材制造材料及构件力学性能预测中的国内外研究进展. 首先简述了常见的机器学习算法和通用的机器学习流程, 重点分析了融合物理信息的机器学习(physics-informed machine learning, PIML)方法的特点与构造方式; 然后概述了增材制造合金材料力学性能4大影响因素的形成原因及机器学习在这些影响因素预测中的应用现状; 重点介绍了ML和PIML在拉伸性能和疲劳断裂性能预测中的代表性研究成果; 最后指出当前机器学习在力学性能预测中的局限性, 并探讨了发展趋势和技术前景.

     

    Abstract: Additive manufacturing (AM) is a revolutionary breakthrough in the manufacturing of modern high-end equipment. In order to promote the mass production and reliable applications of AM-processed components, the major determining factors include manufacturing repeatability, quality reliability, and performance predictability. However, the combined effects of anisotropic microstructure, randomly distributed defects, internal residual stresses, and surface roughness pose a challenge for the prediction accuracy and efficiency of mechanical properties through traditional empirical models and limited testing data. Recently, as an inevitable product of the development of big data and artificial intelligence to a certain stage, machine learning (ML) has demonstrated a great potential for modelling the complex nonlinear relationships among high-dimensional physical quantities, which has received continuous attention in the field of predicting the mechanical properties of AM-processed materials. This paper offers a comprehensive review of the research progress in predicting the mechanical properties of AM-processed metals and components using ML methods. First, the common ML algorithms (parametric and non-parametric models) and general ML procedures (data preparation, model establishment, and model application and evaluation) are briefly introduced. Special attention is devoted to exploring the characteristics and construction methods of the advanced physics-informed machine learning (PIML), with specific discussions on the physics-informed model input, construction and output. Furthermore, the reasons for the formation of the four major influencing factors on the mechanical properties of AM-processed materials (anisotropic microstructure, manufacturing defects, residual stresses, and surface roughness), and the current application status of ML in predicting these influencing factors are summarized. This paper focuses on the representative research results of ML and PIML in predicting the tensile and fatigue fracture properties of AM-processed metals. Finally, the limitations of ML in predicting the mechanical properties of AM-processed metals, as well as the hot topics and technological prospects, are pointed out.

     

/

返回文章
返回