• Becker, Sébastien

Abstract

Metals and their alloys represent a large part of the materials used in industry. Their design is often made from the liquid phase by solidification and such since the Iron Age. With the energy transition we are facing, a better control of manufacturing processes and optimization of these materials are important issues and require a fine understanding of the structural mechanisms at the smallest scales of matter. The crystal nucleation represents the phenomena acting at the first stages of the transition from a supercooled liquid to a crystalline solid, and it can be understood at the atomic scale. However, such study remains very difficult to access experimentally, and large-scale numerical simulations, beyond the million atoms, offer a way to model these phenomena thanks to the improvement of the computing performances. On the other hand, the very large amount of data makes the classical approaches less relevant to fully exploit and analyze the underlying information. This work is devoted to the study of the homogeneous crystal nucleation using large-scale molecular dynamics simulation of four pure metals, namely zirconium, tantalum, aluminium, and magnesium, and two compositions of the binary alloy aluminium-nickel (AlNi), chosen for the diversity of their crystalline phase and their industrial interest. The originality here consists in combining persistent homology, a topological data analysis tool, and an unsupervised machine learning algorithm, based on Gaussian mixture models, for the autonomous identification of local atomic structures from the simulations. The results obtained for pure metals show a heterogeneous scenario of homogeneous nucleation in which the appearance of nuclei is driven by low density areas of structures with five-fold symmetry in the liquid. The growth of the nuclei shows a concurrent emergence of the translational and orientational orderings as well as a diffuse interface between the nuclei and the liquid. In the case of the magnesium, a two-step nucleation process is highlighted, where the nuclei go through a body-centered cubic transient phase that precedes the hexagonal close packed stable phase. In the case of alloys, different nucleation pathways were also observed depending on the composition. For Al50Ni50, the nucleation follows a one-step process towards the body-centered cubic phase, with a symmetrical role of Ni and Al atoms. For Al25Ni75, a polymorphism of phases, body-centered cubic, hexagonal close packed, and face-centered cubic is observed, with the emergence in a first step of nuclei with a body-centered cubic phase in a similar way that the nucleation of pure Ni. For these two alloys, the chemical short-range order of the underlying stable crystalline phase precedes the orientational ordering. This methodology opens the way more generally to further study of structural mechanisms at the atomic scale in relation to the macroscopic properties of materials.

Topics

• density
• impedance spectroscopy
• nickel
• simulation
• Magnesium
• Magnesium
• crystalline phase
• aluminium
• molecular dynamics
• laser emission spectroscopy
• zirconium
• iron
• liquid phase
• solidification
• tantalum
• machine learning