GO-GAN: Geometry Optimization Generative Adversarial Network for Achieving Optimized Structures with Targeted Physical Properties

TL;DR

GO-GAN introduces a novel GAN architecture that efficiently generates optimized geometries based on user-defined physical properties, combining a Pix2Pix framework with a dynamic training loop leveraging dataset symmetries.

Contribution

This work presents GO-GAN, integrating a Pix2Pix GAN with a new input mechanism and symmetry-based training for geometry optimization based on physical properties.

Findings

Successfully generates optimized geometries from scalar property inputs.

Demonstrates rapid and acceptable design variations.

Ensures designs adhere to specifications while allowing creative exploration.

Abstract

This paper presents GO-GAN, a novel Generative Adversarial Network (GAN) architecture for geometry optimization (GO), specifically to generate structures based on user-specified input parameters. The architecture for GO-GAN proposed here combines a \texttt{Pix2Pix} GAN with a new input mechanism, involving a dynamic batch gradient descent-based training loop that leverages dataset symmetries. The model, implemented here using \texttt{TensorFlow} and \texttt{Keras}, is trained using input images representing scalar physical properties generated by a custom MatLab code. After training, GO-GAN rapidly generates optimized geometries from input images representing scalar inputs of the physical properties. Results demonstrate GO-GAN's ability to produce acceptable designs with desirable variations. These variations are followed by the influence of discriminators during training and are of practical significance in ensuring adherence to specifications while enabling creative exploration of the design space.