Cluster analysis for granular mechanics simulations using Machine Learning Algorithms

RIM, DANIELA; MILLÁN, EMMANUEL NICOLÁS; PLANES, MARÍA BELÉN; BRINGA, EDUARDO M; MOYANO, LUIS G.

Mendoza

Congreso; Congreso Internacional de Ciencias de la Computación y Sistemas de Información; 2018

Universidad Champagnat y Universidad Nacional de San Juan

Abstract. Molecular Dynamics (MD) simulations on grain collisions allow to incorporate complex properties of dust interactions. We performed simulations of collisions of porous grains, each with many particles, using the MD software LAMMPS. The simulations consisted of a projectile grain striking a larger immobile target grain, with different impact velocities. The disadvantage of this method is the large computational cost due to a large number of particles being modeled. Machine Learning (ML) has the power to manipulate large data and build predictive models which could reduce MD simulation times. Using ML algorithms (Support Vector Machine and Random Forest) we are able to predict the outcome of MD simulations regarding fragment formation, after a number of steps smaller than in usual MD simulations. We achieved a time reduction of at least 46%, for 90% accuracy. These results show that SVM and RF can be powerful yet simple tools to reduce computational cost in collision fragmentation simulations.