Medium Range Ordering in liquid Al-based alloys: towards a machine learning approach of solidification

Abstract Ab initio molecular dynamic simulations of liquid Al93Cr7 and Al83Zn10Cr7 alloy have evidenced the presence an icosahedral short range order (iSRO) which develops into medium (iMRO) as melt is undercooled. This atomic arrangement accounts for Dynamic Heterogeneities characterized by Al fast-dynamics regions Cr-rich slow-dynamics regions. Characterisation was carried out a direct connectivity approach. However, given small size simulation (256 atoms), such characterisation remains partial. In to better describe both iMRO formation more dilute alloys closer industrial compositions, new modelling strategy has been initiated allow in long term large-scale atomic-level simulations. Molecular Dynamics (MD) million billion atoms may indeed lead meaningful results. Exploitation large amounts MD-generated big data can be means Machine Learning (ML) tools provide relevant powerful analysis methods. An unsupervised ML approach based on topological descriptors using persistent homology concepts proposed reveal structural features arrangements without priori knowledge studied system. applied so far pure melts. Both translational orientational orderings are thus together with nucleation pathways, whose revealed beyond hypotheses Classical Nucleation Theory.