Design of Novel Auxetic Bi-Materials Using Convolutional Neural Networks.

Abstract

A convolutional neural network (CNN) was developed to predict the Poisson's ratio of representative volume elements (RVEs) composed of a bi-material system with soft and hard phases. The CNN was trained on a dataset of binary microstructure configurations, learning to approximate the effective Poisson's ratio based on spatial material distribution. Once trained, the network was integrated into a greedy optimization algorithm to identify microstructures with auxetic behavior. The algorithm iteratively modified material arrangements, leveraging the CNN's rapid inference to explore and refine configurations efficiently. The results demonstrate the feasibility of using deep learning for microstructure evaluation and optimization, offering a computationally efficient alternative to traditional finite element simulations. This approach provides a promising tool for the design of advanced metamaterials with tailored mechanical properties.