Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

A. Padmaprabhan, Shriram Hari, Nived Philip Thomas, Khaish Singh Chadha, Sai Sidhardh, Viswanath Chinthapenta, Prabhat Kumar

TL;DR

GO-GAN introduces a geometry-optimization GAN based on Pix2Pix for producing optimized structures conditioned on scalar inputs. It integrates a U-Net generator with a PatchGAN discriminator and a novel input mechanism that encodes scalar properties as images, guided by a dynamic, symmetry-exploiting training loop. The method is validated on cantilever beam topology optimization and RC49 chair design, achieving fast generation with low volume and compliance errors and controlled design variations. The work highlights potential for rapid, constraint-aware design iterations while identifying opportunities to incorporate multi-physics constraints and hybrid generative models to enhance robustness and diversity.

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.

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

TL;DR

GO-GAN introduces a geometry-optimization GAN based on Pix2Pix for producing optimized structures conditioned on scalar inputs. It integrates a U-Net generator with a PatchGAN discriminator and a novel input mechanism that encodes scalar properties as images, guided by a dynamic, symmetry-exploiting training loop. The method is validated on cantilever beam topology optimization and RC49 chair design, achieving fast generation with low volume and compliance errors and controlled design variations. The work highlights potential for rapid, constraint-aware design iterations while identifying opportunities to incorporate multi-physics constraints and hybrid generative models to enhance robustness and diversity.

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.

Paper Structure

This paper contains 17 sections, 6 equations, 3 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methodology
  3. Problem Description
  4. Topology Optimization of Cantilever Beam
  5. Chair Design Prediction
  6. Neural Network Architecture
  7. Generator Network
  8. Encoder:
  9. Adaptive Dense Layer:
  10. Decoder:
  11. Discriminator Network:
  12. Input Mechanism
  13. Loss Function and Optimization
  14. Results and Discussion
  15. Topology optimization of a cantilever beam
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Schematic diagrams for the different architectures used in the proposed neural network model
  • Figure 2: Cantilever beam and corresponding optimized design. (a) Cantilever beam design domain. $F$ and $L$ denote the applied load and dimension of the domain. (b) Optimized design obtained using a MATLAB code andreassen2011efficient
  • Figure 3: Different parts of the proposed architectures