Deep learning assisted PM metamaterial design

Publication Date


Document Type

Conference Proceeding

Publication Title

Proceedings of the International Congress on Acoustics


In this work, we present deep learning-assisted models for design of tetrahedral PM unit cells to construct a 3D lattice structure that is highly anisotropic and mimics the acoustic properties of water. We optimize unit cell design via forward techniques by altering the design parameter and building datasets to train deep learning models. We implement conditional Wasserstein generative adversarial networks (cWGAN) and convolutional neural networks (CNN) models utilizing TensorFlow Python libraries with Keras API. The purpose of establishing the deep learning framework is to create more data on the parameters of unit cells such that we can select the cells whose mechanical properties fit our design goal.


Convolutional Neural Networks, Deep Learning, Generative Modeling, Pentamode Metamaterials, Wasserstein Generative Adversarial Networks


Biomedical Engineering; Mechanical Engineering