Relating glass forming ability and mechanical behavior to the structure of metallic liquids and glasses

非技术摘要金属玻璃是一类独特的金属，其结构与更常见的晶体材料非常不同。大多数金属中的原子以规则、有序的方式排列，而金属玻璃则具有类似于液态的无序原子结构。这种原子水平上的差异往往导致大规模性能的显著不同，包括更高的强度、更好的耐腐蚀性，以及令人惊讶的能力，可以轻松而廉价地模塑成复杂的形状。然而，并不是所有的金属都能形成玻璃。最近的研究表明，形成玻璃的能力可能与高温液体的原子结构有关。定量描述无序结构以预测金属合金的玻璃形成能力是具有挑战性的，因为原子位置没有长程模式。以前识别原子结构中常见多面体形状的尝试很有启发意义，但对这些形状之间的相似程度以及结构有序性的总体程度提供的洞察力有限，特别是对于由不同原子大小的多个元素组成的合金。该项目转而使用基于人工智能的算法，在计算机模拟的液体中自动识别结构相似的原子团簇。这种结构相似程度较低的液体组合物被认为是良好的玻璃成形剂。模拟的结果将在广泛的成分上进行实验验证，使用紧凑的材料库，并使用先进的基于3D打印的方法快速合成。这项工作将加速发现用于各种应用的新的高性能金属材料，并有助于培养一支受过计算和高效制造方法培训的科学和工程劳动力队伍。最后，它将通过针对不同学区的初中和高中教师的研讨会来影响K-12科学和工程教育。技术总结金属玻璃作为潜在的结构材料具有巨大的前景，因为它们具有优异的机械性能和独特的热塑性成形复杂形状的能力。尽管具有潜在的变革性，但关于玻璃结构和性能之间的关系，仍有几个基本问题。特别是，液体的结构与其淬火形成玻璃的难易程度之间的关系仍然知之甚少。该项目的动机是最近的研究表明，合金系统中成分的相对玻璃形成能力可以通过一个简单的参数来预测，该参数表征了液体中远高于玻璃化转变温度的最近邻原子团簇的布居分布。这项工作结合了分子动力学模拟、模式识别和机器学习聚类算法，以及最先进的高通量合成和表征方法，研究了模拟液体和玻璃中原子团簇的几何和布居与实验观察到的性质之间的相关性，涉及广泛的合金成分和家族。一种基于激光沉积的合成技术被用于快速构建用于评估的合金库。通过纳米压痕将这些库中玻璃形成区的机械性能映射为组成的函数。玻璃形成能力和机械性能的组成趋势将与模拟液体和玻璃中原子团簇的布居分布趋势进行比较。对于每个合金家族，将使用点模式匹配和机器学习聚类算法来识别一组独特和健壮的原子基元，这些基元组成了模拟液体和玻璃的结构。通过将模拟结构与实验性能测量相结合，这项工作将极大地加强对金属玻璃结构-性能关系的理解，并使具有理想性能组合的新合金的合理设计成为可能。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Non-Technical SummaryMetallic glasses are a unique class of metals with structures very different from the more common crystalline materials. Whereas atoms in most metals are arranged in a regular, ordered pattern, metallic glasses have a disordered atomic structure, similar to the liquid state. This difference at the atomic-level often results in dramatically different properties at the large scale, including higher strength, better corrosion resistance, and a surprising ability to be easily and inexpensively molded into complex shapes. Not all metals are able to form glasses, however. Recent studies suggest that the ability to form a glass may be related to the atomic structure of the high-temperature liquid. Quantitatively describing the disordered structure in order to predict the glass-forming ability of a metal alloy is challenging, since there is no long-range pattern to the atomic positions. Previous attempts to identify common polyhedral shapes in the atomic structure are instructive, but offer limited insight into the degree of similarity among those shapes and therefore the overall degree of structural order, particularly for alloys consisting of multiple elements of different atomic sizes. This project instead uses artificial-intelligence-based algorithms to automatically identify structurally ‘similar’ atomic clusters in computer-simulated liquids. Liquid compositions with a low degree of such structural similarity are postulated to be good glass formers. The results of the simulations will be experimentally verified over a wide range of compositions using compact material libraries, rapidly synthesized with an advanced 3D-printing-based method. This work will accelerate the discovery of new, high-performance metallic materials for a variety of applications, as well as contribute to the development of a science and engineering workforce trained in computational and efficient manufacturing methods. Finally, it will impact K-12 science and engineering education through workshops targeting middle- and high-school teachers from diverse school districts.Technical SummaryMetallic glasses hold tremendous promise as potential structural materials, due to their unique combination of excellent mechanical properties and unusual ability to be thermoplastically formed into complex shapes. While potentially transformative, several fundamental questions remain about the relationship between glass structure and properties. In particular, the relationship between the structure of the liquid and the ease with which it can be quenched to form a glass remains poorly understood. This project is motivated by recent studies suggesting that the relative glass-forming ability of compositions within an alloy system may be predicted by a simple parameter characterizing the population distribution of nearest-neighbor atomic clusters in the liquid, well above the glass transition temperature. The work integrates molecular dynamics simulations, pattern-recognition and machine-learning clustering algorithms, and state-of-the-art high-throughput synthesis and characterization methods to investigate correlations between the geometry and population of atomic clusters in simulated liquids and glasses and experimentally observed properties, over a wide range of alloy compositions and families. A laser-deposition-based synthesis technique is used to rapidly construct alloy libraries for evaluation. Mechanical properties of the glass-forming regions in these libraries are mapped as a function of composition via nanoindentation. Compositional trends in glass-forming ability and mechanical properties will be compared with trends in the population distribution of atomic clusters in simulated liquids and glasses. For each alloy family, point-pattern matching and machine-learning clustering algorithms will be used to identify a set of unique and robust atomic motifs that comprise the structure of the simulated liquids and glasses. By integrating the simulated structures with experimental property measurements, this work will dramatically strengthen the understanding of structure-property relationships in metallic glasses, and enable the rational design of new alloys with desirable combinations of properties.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.