Collaborative Research: CDS&E: Charge-density based ML framework for efficient exploration and property predictions in the large phase space of concentrated materials

合作研究：CDS

• 批准号: 2302764
• 负责人: Pejman Tahmasebi
• 金额: $ 30.38万
• 依托单位: Colorado School of Mines
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-09-01 至 2026-08-31
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2302764&HistoricalAwards=false
• 关键词: Collaborative; Research; CDS& Charge; density

Non-technical summaryFuture engineering applications require complex materials to withstand extreme environments. Electronic structure calculations have played an integral role in developing fundamental understanding of atomic and electronic level properties of materials. To tackle the challenges of growing materials complexities, in this project, the investigator at Clemson University will collaborate with investigators at Colorado School of Mines to integrate data-science based image-recognition techniques with electronic structure calculations to predict materials properties. Image recognition is widely used for face recognition, lane-assisted driving, food-contaminant detection, cancer-cell detection, etc. In this project, the charge density of materials will be used in the form of images to learn the electronic structure of materials to enable property predictions in complex materials. The project will contribute to technical, educational and workforce development. A fundamental understanding of lattice distortion in complex alloys will be delivered, namely in high entropy alloys that consist of multiple principal elements in large concentrations. The project will develop an open-source machine learning framework with a curated database of charge densities and alloys’ properties. It will also train undergraduate and graduate students for future digital economy at the intersection of materials physics and data science via a new ‘data science in materials science’ course, and summer workshops.Technical summaryThe chemical randomness in high entropy alloys engenders unique nearest neighbor environments causing lattice and electronic distortions that result in large uncertainties in properties both qualitatively and quantitively. The uncertainties scale with compositional (atomic fraction) and chemical (different elements) diversities resulting in an extremely stiff challenge for density functional theory (DFT) to explore the phase space. This technical challenge runs parallel to the scientific challenge of mechanistic reasons of composition-property correlations. Since, charge density is the fundamental quantity from which the physics and property correlations can be extracted, the investigators will develop a charge-density based machine learning framework that will elucidate the role of disruptive energy landscape on the emerging properties, and simultaneously remove the bottleneck to trace the large phase. The machine learning models will learn the charge density distributions and properties from simpler alloys and predict them in complex alloys while bypassing expensive DFT calculations altogether. The investigators will work under the hypothesis that larger asymmetry in charge density leads to larger uncertainty in properties. The students will learn to perform electronic structure calculations, data generation and interpretation, and application of machine learning models to predict materials properties. The students will also learn image-recognition techniques applied to materials science problems. Summer workshops will be organized by the investigators to engage girls in STEM with an interactive, engaging and hands-on approach. The investigators will also organize a virtual workshop with a specific focus on feature recognition techniques for materials predictions.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.

非技术摘要未来的工程应用需要复杂的材料来承受极端环境。电子结构计算在发展对材料的原子和电子级特性的基本理解方面发挥了不可或缺的作用。为了应对材料日益复杂的挑战，在该项目中，克莱姆森大学的研究人员将与科罗拉多矿业学院的研究人员合作，将基于数据科学的图像识别技术与电子结构计算相结合，以预测材料特性。图像识别广泛应用于人脸识别、车道辅助驾驶、食品污染物检测、癌细胞检测等。在该项目中，将以图像的形式利用材料的电荷密度来学习材料的电子结构，从而实现复杂材料的特性预测。该项目将有助于技术、教育和劳动力发展。将提供对复杂合金中晶格畸变的基本理解，即由高浓度的多种主要元素组成的高熵合金。该项目将开发一个开源机器学习框架，其中包含电荷密度和合金特性的精选数据库。它还将通过新的“材料科学中的数据科学”课程和暑期研讨会，在材料物理和数据科学的交叉点上为未来的数字经济培养本科生和研究生。技术摘要高熵合金中的化学随机性产生了独特的最近邻环境，导致晶格和电子扭曲，从而导致定性和定量性质的巨大不确定性。不确定性随着成分（原子分数）和化学（不同元素）的多样性而变化，这给密度泛函理论（DFT）探索相空间带来了极其严峻的挑战。这一技术挑战与成分-性质相关性的机械原因的科学挑战并行。由于电荷密度是提取物理和性质相关性的基本量，研究人员将开发一种基于电荷密度的机器学习框架，该框架将阐明破坏性能量景观对新兴性质的作用，同时消除追踪大相的瓶颈。机器学习模型将从更简单的合金中学习电荷密度分布和特性，并在复杂合金中预测它们，同时完全绕过昂贵的 DFT 计算。研究人员将在以下假设下开展工作：电荷密度的不对称性越大，性质的不确定性越大。学生将学习执行电子结构计算、数据生成和解释，以及应用机器学习模型来预测材料特性。学生还将学习应用于材料科学问题的图像识别技术。调查人员将组织夏季研讨会，以互动、参与和实践的方式让女孩参与 STEM。研究人员还将组织一个虚拟研讨会，重点关注材料预测的特征识别技术。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。