Efficient numerical and ANN models for optimization of filler gradation of particulate-filled composites

Abstract

Based on hard-sphere-like truncated Lennard-Jones interactions, a box compression simulation method, which belongs to molecular statics, was proposed to calculate the random close packing density of spherical particles. This method accurately provides the gradation corresponding to the maximum random close packing density with higher efficiency than the discrete element method (DEM). We trained an artificial neural network (ANN) model based on the data generated using this method to predict the packing densities of ternary and quaternary packing systems and discussed the applicability of this model in design of particulate-filled composites for electronic packaging. We found that the negative correlation between viscosity and random close packing density only applies to fillers of sizes greater than several microns. When the average filler size is reduced to submicron level, the specific surface area becomes the dominant factor in determining the upper limit of filler loading and viscosity of composites.