DMREF: Physics-Informed Meta-Learning for Design of Complex Materials

一类高性能材料，包括固-固复合材料、多孔固体、泡沫、生物材料和增材制造材料，具有复杂的微观结构，这在决定其性能和性能方面起着主导作用。这个多学科项目将利用人工智能（AI）的最新创新，为复杂材料建立一个新的设计和发现周期，这将极大地加速材料创新。该项目将创造新的方法，通过这些方法，人类材料科学家和人工智能将合作发现这种复杂材料的最佳微观结构设计，以实现目标特性和性能。通过性能驱动设计和微结构优化来创新下一代材料有巨大的机会和需求。在这个设计材料以革新和工程我们的未来（DMREF）项目中，人工智能驱动的设计框架将通过人工智能在复杂物理过程的设计和机器学习方面的根本性突破，开创这些机会，并将对材料研究界产生重大影响。该项目的成功将导致人工智能驱动的材料微观结构设计框架，从而大大加快复杂材料的发现过程，并通过节省不必要的“反复尝试”实验来降低材料创新所需的成本和劳动力。人工智能驱动的设计框架将易于扩展并适用于广泛的复杂材料，这将有利于功能材料，聚合物，复合材料，生物材料等的设计和制造。通过加快发现周期和降低成本，该设计框架将使美国工业受益，从而有助于安全、国家安全和社会的技术进步。因此，该项目将显著加速和推进具有理想性能和功能的材料的发现和开发，这与DMREF项目的愿景是一致的。该项目将受益于与空军研究实验室（AFRL）就含能材料的制造过程以及根据实验结果对人工智能框架的设计输出进行测试和验证的合作。通过AFRL提供的机会，学生培训和劳动力发展将得到加强。作为对性能具有强烈微观结构影响的复杂材料的原型，该项目将重点关注高能材料（EM），其中包括推进剂，烟火和爆炸物的广泛范围-各种推进和弹药系统中的关键部件，对美国军方至关重要，以及民用应用（建筑，运输，采矿等）。该项目将建立新的方法和工具，通过先进的机器认知和博弈论决策，为人工智能驱动的机械制造设计提供闭环，指导表征、设计/优化、制造、实验和验证的整个过程。为了实现这一目标，研究人员将首先构建广泛的CHNO EMs空间并组装机器学习数据集。然后，研究人员将为em等复杂材料开发一种新的物理信息元学习（PIML）框架，然后用实验数据和实际用例进行验证。在实现这些目标的同时，该项目将在与小数据学习、弱监督学习和可解释人工智能等主题相关的人工智能领域取得根本性的、受使用启发的突破，为材料和人工智能社区服务。该项目将受益于与空军研究实验室（AFRL）就含能材料的制造过程以及根据实验结果对人工智能框架的设计输出进行测试和验证的合作。通过AFRL提供的机会，学生培训和劳动力发展将得到加强。该项目由NSF的数学和物理科学部（MPS）材料研究部（DMR）设计材料以革新和工程我们的未来（DMREF）计划，刺激竞争研究的既定计划（EPSCoR），土木，机械和制造创新（CMMI）的工程（ENG）部门以及信息和智能系统（IIS）的计算机和信息科学与工程（CISE）部门共同资助。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

A wide class of high-performance materials, including solid-solid composites, porous solids, foams, biological materials, and additively manufactured materials, have complex microstructures, which play a dominant role in determining their properties and performance. This multidisciplinary project will harness recent innovations in artificial intelligence (AI) to establish a novel design and discovery cycle for complex materials that will dramatically accelerate material innovations. This project will create new methodologies through which human materials scientists and AI will collaborate to discover optimal microstructural designs of such complex materials for targeted properties and performance. There are enormous opportunities and needs for innovating next-generation materials through performance-driven design and optimization of microstructures. The AI-driven design framework in this Designing Materials to Revolutionize and Engineer our Future (DMREF) project will pioneer these opportunities through fundamental breakthroughs in AI for the design and machine learning of complex physical processes and will have a high impact on the materials research community. The success of this project will lead to an AI-driven material microstructure design framework, resulting in significant speedup in the discovery process of complex materials, as well as reducing the cost and labor required for material innovation by saving unnecessary “cut-and-try” experiments. The AI-driven design framework will be easily scalable and applicable to a broad range of complex materials, which will benefit the design and manufacturing of functional materials, polymers, composites, biomaterials, etc. By providing an accelerated discovery cycle and reduced costs, the design framework will benefit the US industry and, thereby, contribute to the safety, national security, and technological advancement of society. As such, this project will significantly accelerate and advance the discovery and development of materials with desirable properties and functionality, which aligns with the vision of the DMREF program. This project will benefit from collaboration with the Air Force Research Laboratory (AFRL) with respect to the manufacturing process of energetic materials and the testing and validation of design outputs of the AI framework against experimental results. Student training and workforce development will be enhanced through opportunities provided through AFRL.As the archetype of a complex material with strong microstructural influence on performance, this project will focus on energetic materials (EM), which cover the wide swathe of propellants, pyrotechnics, and explosives—key components in a variety of propulsion and munition systems critical to the US Military, as well as to civilian applications (construction, transportation, mining, etc.). This project will build new methods and tools to close the loop for AI-driven design of EMs, guiding the overall process of characterization, design/optimization, fabrication, experimentation, and validation through advanced machine cognition and game-theoretical decision making. To accomplish this goal, the investigators will first construct the space of a wide range of CHNO EMs and assemble machine learning datasets. The investigators will then develop a novel physics-informed meta-learning (PIML) framework for complex materials such as EMs, which will then be validated with experimental data and real uses cases. While achieving these, this project will make fundamental, use-inspired breakthroughs in AI-related to topics such as small data learning, weakly-supervised learning, and explainable AI, serving both the materials and AI communities. This project will benefit from collaboration with the Air Force Research Laboratory (AFRL) with respect to the manufacturing process of energetic materials and the testing and validation of design outputs of the AI framework against experimental results. Student training and workforce development will be enhanced through opportunities provided through AFRL.This project is jointly funded by NSF’s the Mathematical and Physical Sciences (MPS) Division of Materials Research (DMR) Designing Materials to Revolutionize and Engineer our Future (DMREF) program, the Established Program to Stimulate Competitive Research (EPSCoR), the Engineering (ENG) division of Civil, Mechanical, and Manufacturing Innovation (CMMI), and the Computer and Information Science and Engineering (CISE) division of Information and Intelligent Systems (IIS).This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.