DMREF: AI-Guided Accelerated Discovery of Multi-Principal Element Multi-Functional Alloys

形状记忆合金（SMA）是一类在施加应力、磁场和/或温度变化时经历可逆且可重复的马氏体转变（MT）的金属合金。这些转变可以实现广泛的技术，包括紧凑型固态致动器，固态制冷机，热存储和管理系统以及对广泛温度变化稳定的结构。不幸的是，已经发现当前的合金配方（具有相对简单的化学性质）在其性能方面具有显著的限制，这阻止了其在变革性技术中的广泛部署。这推动了该领域朝着探索具有越来越复杂的化学性质并且具有以显著量存在的多于三种或四种成分的合金的方向发展[即，多主元素多功能合金（MPEMFAs）。在这个巨大的化学空间中航行极具挑战性。为了应对这一挑战，该项目将开发一种新型的闭环材料设计框架，该框架可以集成实验，计算材料科学模型和机器学习（ML）/人工智能（AI）方法，并通过定制接口连接实验，模型，现有数据，更重要的是，跨学科的研究人员。这个设计材料来革命和工程我们的未来（DMREF）项目旨在提高对一类重要材料的理解，以实现广泛的技术。该项目旨在发现具有极端性能组合的MPEMFAs，例如具有低滞后的超高温马氏体转变（MT），在应力下稳定可逆的形状变化，在温度显著超过最先进水平的超弹性;极端的性能，如高达800°C的因瓦和埃林瓦效应;或独特的定制性能，如SMA作为相变材料（PCM），具有高热导率和相变焓，但MT温度差异很大。为了导航这个巨大的化学空间，将开发一个新的框架：（i）采用新的物理信息机器学习来有效地识别适合优化的可行区域;（ii）融合模拟和实验以获得有效的ML模型;（iii）开发新的批处理方法。（并行）贝叶斯优化（BO）策略，以进行全局最优的迭代实验设计;以及（iv）能够同时考虑多个目标和约束。其目的是超越加速发现，寻求解决有关负责MPEMFA中多功能行为的潜在因素的问题。生成的元数据，加上计算和ML模型，开放访问代码，端到端工作流以及高质量的数据库，将为在数据稀缺的情况下学习复杂系统时开发和验证ML/AI框架提供测试平台，特别是在ML/AI驱动的材料发现中。该项目将利用最近建立的材料科学，信息学和设计跨学科研究生证书，能源材料的数据驱动的发现和设计（D3 EM），以培训受此努力支持的博士生，该奖项反映了NSF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。

Shape Memory Alloys (SMAs) are a class of metallic alloys that undergo reversible and repeatable martensitic transformations (MT) upon applying stress, magnetic fields, and/or temperature changes. These transformations can enable a wide range of technologies, including compact solid-state actuators, solid-state refrigerators, thermal storage and management systems, and structures that are stable against wide temperature changes. Unfortunately, current alloy formulations (with relatively simple chemistries) have been found to have significant limitations in their performance that prevent their widespread deployment in transformative technologies. This has pushed the field towards exploring alloys with increasingly complex chemistries and with more than three or four constituents being present in significant amounts [i.e., multi-principal element multi-functional alloys (MPEMFAs)]. Navigating this vast chemical space is extremely challenging. To address this challenge, this project will develop a novel closed-loop materials design framework, which can integrate experiments, computational materials science models, and machine learning (ML) / artificial intelligence (AI) approaches, with customized interfaces connecting experiments, models, existing data, and more critically, researchers across disciplines. This Designing Materials to Revolutionize and Engineer our Future (DMREF) project aims to result in an enhanced understanding of an important class of materials to enable a wide range of technologies. Participating students will be trained in interdisciplinary approaches to materials discovery in the spirit of the Materials Genome Initiative (MGI).This project aims to discover MPEMFAs with extreme property combinations, such as ultra-high temperature martensitic transformations (MTs) with low hysteresis, stable reversible shape change under stress, superelasticity at temperatures significantly beyond state-of-the-art; extreme properties, such as Invar and Elinvar effects up to 800°C; or uniquely tailored properties, such as SMAs-as-phase-change-materials (PCMs) with high thermal conductivity and transformation enthalpy but also with widely different MT temperatures. To navigate this vast chemical space a new framework will be developed that: (i) employs novel physics-informed machine learning to efficiently identify the feasible regions amenable to optimization; (ii) fuses simulations and experiments to obtain efficient ML models; (iii) develops new Batch (parallel) Bayesian Optimization (BO) strategies to make globally optimal iterative experimental design; and (iv) is capable of simultaneously considering multiple objectives and constraints. The aim is to go beyond accelerated discovery, seeking to address questions about the underlying factors responsible for the multi-functional behavior in MPEMFAs. The generated metadata, together with the computation and ML models, open-access code, end-to-end workflows, as well as high quality databases, will provide a testbed for developing and validating ML/AI frameworks when learning complex systems under data scarcity, particularly in ML/AI-drive materials discovery. The project will leverage the recently established interdisciplinary graduate certificate on materials science, informatics and design, Data-Enabled Discovery and Design of Energy Materials (D3EM), to train the PhD students supported by this effort, contributing to the workforce development goals of the Materials Genome Initiative.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.