LUMEN : Liquid structure Understanding for Microstructure Engineering
with Novel upscaling assisted by artificial intelligence.
Coordinator: Julien ZOLLINGER
Université de Lorraine
Keywords: Microstructure, processes, atomic scale, artificial intelligence, high-speed metallurgy, metal alloys, multi-scale modeling, high-performance computing, solidification
The LUMEN project aims to develop new aluminum alloys based on knowledge of the structure of the liquid. The objective is to explore the addition of transition elements and possible additives in aluminum alloys to optimize their microstructure in the context of increased recycling, which inevitably leads to an increase in impurities that alter the mechanical properties of the alloys. Ab initio molecular dynamics calculations show that certain more favorable microstructures can be obtained in connection with a liquid structure dominated by 5-fold symmetry. Certain impurities could therefore become beneficial elements: the aim is to understand under what conditions.
We also know that in additive manufacturing, the addition of slowly diffusing elements improves properties by supersaturating the matrix during solidification and slowing down solid-state precipitation.
The project uses the DIAMS high-throughput metallurgy platform to screen alloys and the DIAMOND platform to develop a multiscale modeling approach. Atomic modeling using molecular dynamics employs interatomic potentials derived from machine learning on ab initio modeling results to capture the influence of alloying elements on the liquid structure and solid-liquid interface behavior. Phase field microstructure modeling predicts the dynamics of solidification microstructures, taking into account the anisotropy of interfacial energy and atomic attachment kinetics. Process-scale modeling using the grain envelope model (GEM) simulates the solidification macrostructure and studies intergranular growth competition. The link between the different scales is provided by new tools for transferring the constitutive laws from the atomic scale to the phase field and then to the GEM. The experimental results will serve as a basis for calibrating and validating the models, with a particular focus on Al-Cr-Fe-Ti systems.
The project comprises five work packages (WP). WP1 focuses on atomic-scale simulations of liquids and interfaces, performing large-scale molecular dynamics simulations to obtain an accurate representation of solid-liquid interfaces and atomic diffusivity in the liquid. WP2 focuses on phase field modeling, developing codes to quantitatively predict the dynamics of solidification microstructures. WP3, dedicated to grain envelope modeling, aims to extend GEM to multi-component alloys and simulate grain growth competition. WP4 uses machine learning to link atomic and microstructural descriptions and develop a graph-based generative model for coarse-graining. Finally, WP5 is dedicated to the generation and characterization of microstructures using high-throughput manufacturing approaches, creating databases on the effect of alloying elements and their combinations on solidification microstructures.
Innovations include the deployment of combinatorial metallurgy on multi-component systems, multi-scale modeling incorporating new physical models, and the use of artificial intelligence tools for upscaling between models. The identification of new products tolerant to recycling impurities will enable the aluminum industry to become more resilient in the face of increased recycling. In conclusion, LUMEN aims to overcome scientific barriers through innovative approaches ranging from the atomic scale to the process scale, with the goal of developing aluminum alloys adapted to the challenges of recycling and additive manufacturing.