Molecular Dynamics

Molecular dynamics simulation of alloying characteristics of Al–Mg nanoparticles under different process heating conditions 

ABSTRACT

The effect of thermal process parameters on the alloying of Al–10Mg (wt%) system has been studied in molecular dynamics approach. Five distinct values of heating rate have been considered for Al and Mg nanoparticles (NPs) to melt, coalesce followed by identical cooling. Detailed investigation of Al-Mg NPs alloying has been done in terms of alloying temperature, atomic migration, coalescence kinetics and mechanical characteristics for various heating rates were carried out. Prior to the alloying simulation, component melting simulations of Al and Mg single particles (SPs) were conducted to determine the melting temperature of NPs and to inspect their individual thermo-stability. From the change in potential energy, its evident that the melting temperature of the component NPs significantly depends on heating rates. Following the component melting, alloying simulations were conducted in three distinct phases: (i) heating, (ii) relaxation, and (iii) cooling and solidification. Following the solidification, uniaxial tensile test simulation was done on a block cut to characterise the mechanical behaviour in context to dislocation analysis, HCP evolution, stacking fault analysis and corresponding stress–strain relationship. Obtained results showed a substantial affiliation of heating rates and coalescence kinetics but showed minimal effect on mechanical properties.

Graph neural network-based predictions of potential energy landscape in hybrid monte-carlo molecular dynamics simulations of a medium-entropy alloy (In Progress)

ABSTRACT

Hybird Monte-Carlo molecular dynamics (MCMD) simulations are employed to achieve a thermally stable structure across various temperatures in a medium-entropy alloy (MEA) using LAMMPS. The simulation generated dump files and potential energy labels, which were used to construct graph reperesentation of the atomic configurations. Edges were created between each atom and its 12 nearest neighbors without incorporating explicit edge features. These graphs then served as input for for a Graph Convolutional Neural Network (GCNN)- based machine learning (ML) model to predict the system’s potential energy. The GCNN architecture effectively captured the atomic interactions, local environment and chemical ordering within the MEA structure. The GCNN-based ML model demonstraed strong performance in predicting potential energy at different steps, showing satisfactory results on both the training data and unseen configurations. Our approach highlights the potential of integrating MCMD simulations with GNN based-ML models to efficiently predict the and analyze the properties of complex alloys without the need of interatomic distance data.

Keywords: Graph neural network, Hybrid Monte-Carlo molecular dynamics, Medium-entropy alloy, Machine learning, Graph

representation