Applying Machine Learning to local structure in metallic systems for structural applications

机器学习正迅速成为计算科学中最有用的工具和最受欢迎的技能之一。在这个项目中，学生将学习如何使用机器学习算法，并将其应用于材料科学中一个关键的现实问题。理解结构-性能关系对新型材料设计至关重要。近年来，高熵合金作为冶金领域中最活跃的领域之一，已经证明了局部结构（即原子尺度上的相互作用）对合金物理性能的重要性。同样，已经在航空航天工业中服役的合金，已经证明了局部结构影响其强度的现象。如果要在工业上成功地利用这些特性，就需要充分了解局部结构和特性之间的这种联系。这需要计算驱动局部结构形成的潜在能量学。计算机模拟通常用于根据输入电势预测系统内发生的顺序，这使得生成任意大的局部结构数据相当容易。然而，由于多壳相互作用的复杂性和由此产生的数学概率，这个问题很难反转。这使得这个问题非常适合应用机器学习技术，可以学习能量学和局部结构之间的映射。这个项目将使用机器学习模型来计算能量参数。这将使我们能够估计驱动在金属系统中观察到的短程有序的能量学。结构合金的x射线和中子散射数据将在大型设施（钻石光源，ISIS中子源）上获得，并提取局部结构，并用于验证机器学习模型。力学测试也将在与所创建的模型相关的属性上进行。

Machine learning is rapidly becoming one of the most useful tools and highly sought-after skills in computational science. In this project, the student will learn how to use machine learning algorithms and work to apply them to a key real-world problem in materials science. Understanding structure-property relationships is fundamentally important for novel materials design. Recently, high-entropy alloys, one of the most active fields in metallurgy, have demonstrated the importance of local structure (i.e. interactions on the atomic scale) to alloy physical properties. Similarly, alloys already in service in the aerospace industry, have demonstrated phenomena suggesting local structural effects on their strength. If these properties are to be successfully industrially exploited, this link between local-structure and properties needs to be fully understood. This requires the calculation of the underlying energetics driving the formation of local structure.Computer simulations are commonly used to predict order occurring within a system from an input potential, making it fairly easy to generate arbitrary large data of local structure. However, this problem is difficult to invert due to the complexity of multi-shell interactions and resulting mathematical probabilities. This makes the problem ideal for the application of machine learning techniques that can learn the mapping between energetics and local structure.This project will use machine learning models for the calculation of energetic parameters. This will enable us to estimate the energetics driving the short-range ordering observed in metallic systems. X-ray and Neutron scattering data from structural alloys will be obtained at large scale facilities (Diamond Light Source, ISIS Neutron Source), and the local structure extracted, and used to validate the machine learning models. Mechanical testing will also be carried out at the properties correlated with the models created.