Diffusion, dynamics and solidification processes in Fe-Al-Si melts

本方案旨在研究控制Fe-Al-Si熔体扩散动力学的微观机制，以及这些系统在过冷状态下的早期凝固阶段。德法联合体建立了一项联合实验和理论研究，以了解Fe-Al-Si混合物中的这些现象，这具有基本的和技术上的兴趣。对金属液体进行最先进的无接触测量，应用静电和电磁悬浮技术，并使用大型欧洲设备在凝固前测量液体和过冷合金的高质量结构和动态性能。所获得的实验数据将提供高温Fe-Al-Si熔体的广泛的热物理性质，即密度、热膨胀、剪切粘度、静态结构因子、自扩散系数和互扩散系数以及Soret系数。这些数据与量子计算的数据库相关联，使我们能够使用基于人工智能概念的机器学习方法来构建原子间力场。有了这些力场，在大规模分子动力学模拟方面的真实建模将成为可能。从这些模拟中，我们的目标是对动力学和凝固过程的微观洞察。此外，以Fe-Al-Si系统为例，应发展从机器学习技术获得三元金属系统相互作用模型的方法。这些相互作用模型应对熔体的材料性质以及凝固过程和晶相的性质提供真实的描述。

The present proposal aims at investigating microscopic mechanisms that govern the diffusion dynamics in Fe-Al-Si melts as well as the early stages of solidification of these systems in their undercooled state. The German-French consortium set up a joint experimental and theoretical study for understanding these phenomena in Fe-Al-Si mixtures which is of fundamental and technological interest. State-of-the-art contact-free measurements on metallic liquids are performed, applying both electrostatic and electromagnetic levitation techniques, and using large-scale European facilities to measure high quality structural and dynamic properties in the liquid and undercooled alloys prior to solidification. The obtained experimental data will provide a wide spectrum of thermophysical properties of high-temperature Fe-Al-Si melts, namely density, thermal expansion, shear viscosity, static structure factors, self- and interdiffusion coefficients, and Soret coefficients. These data are associated to a data base from quantum calculations and enable us to construct interatomic force fields using machine learning methodologies based on Artificial Intelligence concepts. With these force fields, a realistic modelling in terms of large-scale molecular dynamics simulations shall be possible. From these simulations, we are aiming at a microscopic insight into kinetic as well as solidification processes. Moreover, for the example of the system Fe-Al-Si, the methodology shall be developed to obtain interaction models for ternary metallic systems from machine learning techniques. These interaction models shall provide a realistic description of materials properties of melts as well as of solidification processes and the properties of crystalline phases.