[Paper Review] A Novel Topology Optimization Approach using Conditional Deep Learning

Sharad Rawat, M.-H. Herman Shen|arXiv (Cornell University)|Jan 14, 2019

Topology Optimization in Engineering17 references50 citations

TL;DR

The paper presents a conditional Wasserstein GAN (CWGAN) framework to mimic conventional topology optimization for planar structures, enabling ultra-low-cost generation of optimized designs with settable design conditions.

ABSTRACT

In this study, a novel topology optimization approach based on conditional Wasserstein generative adversarial networks (CWGAN) is developed to replicate the conventional topology optimization algorithms in an extremely computationally inexpensive way. CWGAN consists of a generator and a discriminator, both of which are deep convolutional neural networks (CNN). The limited samples of data, quasi-optimal planar structures, needed for training purposes are generated using the conventional topology optimization algorithms. With CWGANs, the topology optimization conditions can be set to a required value before generating samples. CWGAN truncates the global design space by introducing an equality constraint by the designer. The results are validated by generating an optimized planar structure using the conventional algorithms with the same settings. A proof of concept is presented which is known to be the first such illustration of fusion of CWGANs and topology optimization.

Motivation & Objective

• Motivate reducing computational cost in topology optimization by learning from conventional algorithms.
• Develop a CWGAN-based model to reproduce quasi-optimal topologies with conditioned design parameters.
• Demonstrate that conditioning truncates the design space and enables pre-specified optimization criteria.

Proposed method

• Use a conditional Wasserstein GAN with deep CNN generators and discriminators.
• Train on quasi-optimal planar structures produced by conventional topology optimization algorithms.
• Incorporate an equality constraint to condition the CWGAN on designer-specified requirements.
• Validate by generating an optimized planar structure using the same settings as the conventional method.

Experimental results

Research questions

• RQ1 Can CWGANs reproduce quasi-optimal topology optimization results for planar structures?
• RQ2 Does conditioning CWGANs on design requirements effectively constrain the design space?
• RQ3 How does the CWGAN-generated design compare to conventional topology optimization outputs under identical settings?

Key findings

• Demonstrates a proof of concept for fusing CWGANs with topology optimization.
• CWGANs can generate topology samples conditioned to predefined design values.
• The approach aims to replicate conventional optimization results in a computationally inexpensive manner.