[Paper Review] A deep Convolutional Neural Network for topology optimization with strong generalization ability

Yi‐Quan Zhang, Bo Peng+3 more|arXiv (Cornell University)|Jan 23, 2019

Topology Optimization in Engineering13 references64 citations

TL;DR

The paper presents a deep CNN with U‑Net like architecture for topology optimization, trained on SIMP-generated data, achieving faster designs with comparable optimality and good generalization to unseen boundary conditions.

ABSTRACT

This paper proposes a deep Convolutional Neural Network(CNN) with strong generalization ability for structural topology optimization. The architecture of the neural network is made up of encoding and decoding parts, which provide down- and up-sampling operations. In addition, a popular technique, namely U-Net, was adopted to improve the performance of the proposed neural network. The input of the neural network is a well-designed tensor with each channel includes different information for the problem, and the output is the layout of the optimal structure. To train the neural network, a large dataset is generated by a conventional topology optimization approach, i.e. SIMP. The performance of the proposed method was evaluated by comparing its efficiency and accuracy with SIMP on a series of typical optimization problems. Results show that a significant reduction in computation cost was achieved with little sacrifice on the optimality of design solutions. Furthermore, the proposed method can intelligently solve problems under boundary conditions not being included in the training dataset.

Motivation & Objective

• Develop a neural network framework for structural topology optimization.
• Leverage an encoder–decoder (U‑Net) architecture to perform downsampling and upsampling.
• Use a well-designed multi-channel input tensor to encode problem information.
• Output the layout of the optimal structure from learned representations.
• Evaluate performance against conventional SIMP optimization in terms of speed and accuracy.

Proposed method

• Adopt an encoder–decoder CNN architecture with downsampling and upsampling components.
• Incorporate a U‑Net style design to improve performance on topology optimization tasks.
• Design input as a multi-channel tensor where each channel encodes different problem information.
• Train the network on a large dataset generated by traditional SIMP topology optimization.
• Output the predicted optimal structure layout as the network’s final prediction.
• Compare efficiency and accuracy against SIMP across typical optimization problems.

Experimental results

Research questions

• RQ1 Can a deep CNN generalize topology optimization solutions to boundary conditions not seen during training?
• RQ2 What are the trade-offs between computational cost and design optimality when replacing SIMP with a learned model?
• RQ3 How does U‑Net based architecture affect performance on topology optimization tasks?

Key findings

• The proposed CNN achieves significant reductions in computation cost with only minor sacrifices in design optimality.
• The network demonstrates intelligent generalization to boundary conditions outside the training set.
• Training relies on a dataset produced by conventional SIMP optimization, enabling the model to learn canonical solutions.