Spanning the Scales: Insights into Dislocation Mobility Provided by Machine Learning and Coarse-Grained Models

金属是如何断裂的？我们如何才能让他们变得更强大？缺陷和杂质的作用是什么？材料的强度最终由原子水平上的微观相互作用决定，这种相互作用可以精确地建模。然而，挑战在于，从计算上讲，不可能一步到位地将信息从纳米级传播到毫米级。在这个项目中，你将使用量子力学-分子力学和高斯近似势(一种机器学习方法)相结合的方法来建立位错的粗晶模型，并对金属和合金中的塑性变形进行定量预测。在金属抗塑性变形中产生的应力由位错的流动性决定，位错在晶体中移动的容易程度。位错的运动受缺陷对扭结和钉扎的形成和迁移过程的限制。在原子尺度上对快速运动位错在大应力下的这些过程进行了模拟，对应于冲击波上升的条件，但在低应变率下，时间尺度太长，无法使用原子方法。粗粒化方法绕过了显式建模原子的需要，使用统计力学根据扭结对激活热等量来确定位错的迁移率。需要一个多尺度的方法来揭示位错的结构和能量学的细节，包括位错扭结结构和对非滑动应力分量的依赖关系，并为改进的粗晶模型提供输入。该项目将与AWE合作，通过开发体心立方金属中螺位错运动的粗晶模型，探索位错能量学和位错运动之间的联系。将使用基于机器学习的原子间势和QM/MM方法来跨越从头计算建模能力与所需时间和长度尺度之间的差距。除了研究纯金属外，还将研究杂质的影响。将考虑较小的间隙杂质和较大的置换合金化。在先前工作的基础上，W将被初步考虑。还有空间考虑其他系统，如铁和钢，Ta，Taw，或V，将通过与位错运动的直接分子动力学模拟相比较，寻求在强驱动区对粗晶位错模型的验证。将继续调查粗粒度模型将如何在弱驱动的制度下得到验证。

How do metals break? How can we make them stronger? What are the roles of defects and impurities? The strength of materials are ultimately determined by the microscopic interactions on the atomic level, which can be modelled accurately. However, the challenge is that computationally it is not possible to propagate information in one step from the nanometer to the millimeter scale. In this project, you will use combined Quantum Mechanics-Molecular Mechanics and Gaussian Approximation Potentials, a machine learning approach, to develop coarse-grain models of dislocations and to make quantitative predictions of plastic deformations in metals and alloys.The stress generated in a metal resisting plastic deformation is governed by the dislocation mobility, the ease by which dislocations move through the crystal. Dislocation motion is limited by the processes of formation and migration of kinks and pinning by defects. Modelling of these processes on the atomistic scale has been carried out for fast-moving dislocations under large stresses, corresponding to conditions in the rise of a shock wave, but at lower strain-rates, timescales are too long to access using atomistic methods.Coarse grained methods bypass the need to model the atoms explicitly, determining the dislocation mobility from quantities such as the kink-pair activation enthalpy using statistical mechanics. A multiscale approach is needed to reveal the details of the structure and energetics of dislocations, including the dislocation kink structure and dependencies on non-glide stress components, as well as providing inputs for improved coarse-grained models.In collaboration with AWE, this project will explore the link between dislocation energetics and dislocation mobility, through the development of coarse-grained models for screw dislocation mobility in bcc metals, including the effect of non-glide stresses. Machine-learning based interatomic potentials and QM/MM approaches will be used to span the gap between the capabilities of ab initio modelling and the required time and length scales. In addition to studying the pure metal, the effect of impurities will be investigated. Small interstitial impurities and larger substitutional alloying will be considered. Building on prior work, W will be considered initially. There is scope for considering other systems such as Fe and steel, Ta, TaW, or V. Validation of the coarse-grained dislocation model in the strongly driven regime will be sought by comparing with direct molecular dynamics simulations of dislocation mobility. Investigation of how the coarse-grained model will be validated in the weakly driven regime will be pursued.