Universal Design Methodology for Printable Microstructural Materials via a New Deep Generative Learning Model: Application to a Piezocomposite

TL;DR

This paper introduces a universal microstructural design methodology using a novel deep generative model, enabling efficient creation of multifunctional piezoelectric composites for sensing and energy applications.

Contribution

It presents a new physics-aware generative neural network combining VAE and vision transformer for designing complex microstructures with desired properties.

Findings

Developed an FFT-based method for property prediction of microstructures.

Created HetMiGen, a generative model for 3D microstructure design.

Demonstrated the model's ability to generate functional piezoelectric microstructures.

Abstract

We devised a general heterogeneous microstructural design methodology applied to a specific material system, elasto-electro-active piezoelectric ceramic embedded plastics, which has great potential in sensing, 5G communication, and energy harvesting. Due to the multiphysics interactions of the studied material system, we have developed an accurate and efficient FFT-based numerical method to find the multifunctional properties of diverse cellular microstructures generated by our HetMiGen code. To mine this big dataset, we used our customized physics-aware generative neural network in the format of a VAE with convolutional neural layers augmented by a vision transformer to learn long-distance features which may affect the properties of the 3D voxelized microstructures. In training, the decoder learns how to map the property distribution to the appropriate high-dimensional distribution of 3D microstructures. Therefore, it can be considered an online material designer within the explored design space during its inference phase.