Modelling effective electrical resistance in particle reinforced composites using Generative Adversarial Network

TL;DR

This paper introduces a cGAN-based model trained on finite element data to accurately predict electrical resistance in particle-reinforced composites, surpassing traditional resistor network methods.

Contribution

The study presents a novel cGAN approach for modeling electrical resistance in composites, capturing complex particle interactions more accurately than existing methods.

Findings

cGAN outperforms resistor network models in accuracy

Model effectively captures particle connectivity effects

Finite element data enhances model training

Abstract

Polymer matrix composites embedded with conductive particles are widely utilized for applications that demand stringent control of the effective electrical resistance (or conductivity) of the material. This property is highly sensitive to the particle shape and size distribution within the composite and their percolation threshold. One of the most widely utilized numerical strategies to model this property is the Resistor Network method. However, it is based on many assumptions of the particle shape and inter-particle contact which limits its practical applications. In this work, we have proposed a conditional Generative Adversarial Network (cGAN) based modelling strategy that can accurately capture the flow of electrical current through particles which are connected to multiple other particles in the matrix. The cGAN is trained on data generated by finite element simulations that can model the physics of the problem accurately. It is shown that the GAN based model predicts the electrical flow within the composite and hence the effective electrical resistance much more accurately than the Resistor Network model.