LEAPS-MPS: Machine Learning-guided Identification of Mechanically Stabilizing Solid-state Electrolytes

锂离子电池在我们的日常生活中扮演着越来越重要的角色，为智能手机、平板电脑和电动汽车等设备供电。然而，目前的电池有很大的局限性，比如安全问题和需要经常充电。为了满足日益增长的能量存储需求，需要能够存储更多能量的更持久的电池。一个有希望的解决方案是用锂金属取代电池负极中使用的石墨，锂金属有可能储存大约10倍的能量。然而，锂金属的一个主要挑战是树突的形成，树突是一种小的、树枝状的结构，随着时间的推移会生长，并可能导致短路，导致电池故障。本项目旨在通过研究材料的力学性能，识别具有优异力学特性的电解质，了解如何防止枝晶的形成。这项研究是在休斯顿大学进行的，这是一个主要的西班牙裔服务机构，为扩大代表性不足的群体的参与提供了肥沃的土壤。本项目将招收研究生和本科生，在与国民经济和科学发展高度相关的大数据、人工智能、计算材料科学等新兴交叉领域进行专业培训。该项目旨在发现具有定制机械性能的固体材料，用于锂金属阳极的全固态电池的电解质。用固态电解质取代商用锂离子电池中的液体电解质被认为是最有前途的抑制枝晶的方法，因为固体材料具有优越的机械性能。然而，尽管进行了大量的研究，还没有发现能够完全抑制枝晶的固体材料。目前对枝晶抑制的理解存在一些空白，包括(i)对机械性能标准的理解有限，（ii）缺乏准确探测固体材料全部机械行为的工具，以及（iii）缺乏系统的方法来识别新的固体材料作为候选电解质。基于最近的理论和实验工作，PI假设力学各向异性，以弹性性能的方向依赖性为特征，在Li金属阳极和固体材料之间的固-固界面上起着重要作用，因此可以利用该各向异性来设计抑制枝晶形成的固态电解质。该项目遵循材料基因组计划（MGI）的原则，并建立了一种独特的数据驱动方法，用于材料的各向异性弹性性能的生产和分析，以及它们对锂枝晶成核和生长的影响的研究。具体来说，该项目将通过(i)开发不确定性量化的机器学习模型来预测全弹性张量，从而预测材料的各向异性行为，以及（ii）进行高通量筛选以识别机械稳定的固态电解质来解决这些空白。这种机器学习引导的含锂材料的计算筛选是一种高效和有效的方法，可以识别有前途的候选材料，以进行进一步的实验验证。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。

NON-TECHNICAL SUMMARY Lithium-ion batteries are playing an increasingly important role in our daily lives, powering devices like smartphones, tablets, and electric vehicles. Current batteries, however, have major limitations such as safety issues and the need for frequent recharging. To meet the growing demand for energy storage, longer-lasting batteries that can store more energy are needed. A promising solution is to replace the graphite used in the negative electrode of batteries with lithium metal, which has the potential to store about 10 times more energy. However, a major challenge with lithium metal is the formation of dendrites—small, branch-like structures that grow over time and can cause short circuits, leading to battery failure. This project aims to understand how to prevent dendrite formation by studying the mechanical properties of materials and identifying electrolytes with superior mechanical characteristics. The research is conducted at the University of Houston, a major Hispanic-Serving Institution, which provides a fertile ground for broadening participation from underrepresented groups. Graduate and undergraduate students will be recruited for this project and professionally trained in the new cross-disciplinary area of big data, artificial intelligence, and computational materials science, which is highly relevant to national economic and scientific advancement.TECHNICAL SUMMARYThis project aims to discover solid materials with tailored mechanical properties to be used as electrolytes in all-solid-state batteries with lithium metal anode. Replacing the liquid electrolyte in commercial Li-ion batteries with solid-state electrolytes is considered the most promising approach to suppress dendrites due to the superior mechanical properties of solid materials. However, despite extensive research efforts, no solid material that can completely suppress dendrites has been successfully identified. There are several gaps in the current understanding of dendrite suppression, including (i) limited understanding of the criteria on mechanical properties, (ii) lack of tools to accurately probe the full mechanical behaviors of solid materials, and (iii) lack of a systematic approach to identifying new solid materials as candidate electrolytes. Based on recent theoretical and experimental work, the PI hypothesizes that mechanical anisotropy, characterized by the directional dependence of elastic properties, plays a significant role at the solid-solid interface between Li metal anode and a solid material, and can thus be leveraged to design solid-state electrolytes that suppress the formation of dendrites. The project embraces the principles of the Materials Genome Initiative (MGI) and establishes a unique data-driven approach for the production and analysis of anisotropic elastic properties of materials and the investigation of their effects on Li dendrite nucleation and growth. Specifically, the project will address the gaps by (i) developing uncertainty-quantified machine learning models to predict the full elastic tensors and thus anisotropic behaviors of materials and (ii) conducting high-throughput screening to identify mechanically stabilizing solid-state electrolytes. This machine learning-guided computational screening of Li-containing materials is an efficient and effective approach to identifying promising candidates for further experimental verification.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.