塑性变形理论建模新范式:人工智能赋能和数据科学驱动  

New paradigm of plastic deformation theory modelling:Artificial intelligence empowered and data science-driven

作　　者：何霁 江晟达[1,2] 刘霄 郭聪 钱昌明[1,2] 李淑慧 HE Ji;JIANG Sheng-da;LIU Xiao;GUO Cong;QIAN Chang-ming;LI Shu-hui(State Key Laboratory of Mechanical System and Vibration,Shanghai Jiao Tong University,Shanghai 200240,China;Shanghai Key Laboratory of Digital Manufacture for Thin-walled Structures,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学机械系统与振动国家重点实验室,上海200240 [2]上海交通大学上海市复杂薄板结构数字化制造重点实验室,上海200240

出　　处：《塑性工程学报》2024年第1期2-14,共13页Journal of Plasticity Engineering

基　　金：国家自然科学基金面上资助项目(51975364);国家自然科学基金重大项目(51790170);机械系统与振动国家重点实验室自主课题资助项目(MSVZD202207)。

摘　　要：传统塑性变形理论基于唯象方法,主要依赖于开发者的经验并通过拟合已得到的实验数据来获取,具有强烈的局限性。随着材料微观结构和力学性能的复杂程度不断提高,开发更为复杂的唯象本构模型已经极其困难。材料模型不通用、难以联系材料性能和制造过程成为制约塑性成形加工的主要问题之一,也为塑性理论的发展提出严峻挑战。人工智能与数据科学的发展为材料和机械科学带来了新的机遇,随着材料设计和应用的快速发展,不需要传统的显式本构模型,反映材料微观结构和宏观性能之间关系的塑性变形理论建模方法应运而生。其中,以物理规律增强神经网络建模、高效多尺度聚类分析、数据驱动免本构建模为代表的理论建模新范式具有其鲜明的特色和显著优势,是极具潜力的模拟预测方法。通过对这3个方面的总结梳理,探讨了人工智能赋能和数据科学驱动的塑性理论建模及多尺度模拟技术开发的发展现状与未来趋势。Traditional plastic deformation theories are based on the phenomenological approach which mainly rely on the experience of the developers and are obtained by fitting the experimental data,which have severe limitations.With the increasing complexity of microstructures and mechanical properties of materials,it has become extremely difficult to develop more complex phenomenological constitutive models.Non-universal material models,which are difficult to relate to material properties and manufacturing processes,have become one of the main problems limiting plastic forming and processing,and also pose a serious challenge to the development of plasticity theory.The development of artificial intelligence and data science opened up new possibilities in materials and mechanical sciences.With the rapid development of material design and application,the traditional explicit constitutive model is not needed,the theoretical modelling approaches to plastic deformation that reflect the relationship between the microstructures and macroscopic properties of materials come into being.Among them,the new paradigm modelling theories represented by mechanistically informed neural networks,efficient multi-scale clustering analysis and model-free data-driven computational mechanics have their distinctive features and significant advantages,which are the most promising simulation methods.Through the summary of these three aspects,the development status and future trend of plastic theoretical modeling and multi-scale simulation technology development driven by artificial intelligence and data science were discussed.

关 键 词：数据驱动 多尺度 计算力学 机器学习 无本构模拟 

分 类 号：TG156[金属学及工艺—热处理]