An Efficient and High-Resolution Topology Optimization Method Based on Convolutional Neural Networks

Abstract

Topology optimization is a pioneering design method that can provide various candidates with high mechanical properties. However, the high-resolution for the optimum structures is highly desired, normally in turn leading to computationally intractable puzzle, especially for the famous Solid Isotropic Material with Penalization (SIMP) method. In this paper, we introduce the Super-Resolution Convolutional Neural Network (SRCNN) technique into topology optimization framework to improve the resolution of topology solutions with extremely high computational efficiency. Additionally, a pooling strategy is established to balance the number of finite element analysis (FEA) and the output mesh in optimization process. Considering the high training cost of 3D neural networks, several 2D neural networks are combined to deal with 3D topology optimization design problems. The combined treatment method used in 3D topology optimization design eliminates the expense of retraining 3D convolutional neural network and guarantees the quality of 3D design. Some typical examples justify that the high-resolution topology optimization method adopting SRCNN has excellent applicability and high efficiency.

Topologically optimal design and failure prediction using conditional generative adversarial networks

“…45 Among many methods in finding the topologically optimum structures, deep learning-based techniques recently became the tools of choice. 11,13,14,[22][23][24][25][26][48][49][50] Specifically, cGANs attracted many research communities in the past few years due to their advanced synthesizing ability. 11,14,24,51 TopologyGAN is one of the most recent studies which uses cGANs to predict the optimal topology.…”

“…In the literature, interesting research on topology optimization has been done using deep learning and image processing techniques. 6,11,13,14,[22][23][24][25][26] A detailed discussion on some of these important and related works is presented in Section 2. Though the aforementioned papers discuss various ways to produce (or predict) topologically optimal designs, they have not considered stress predictions on the optimal designs.…”

“…In the literature, interesting research on topology optimization has been done using deep learning and image processing techniques 6,11,13,14,22‐26 . A detailed discussion on some of these important and related works is presented in Section 2.…”

A Systematic Study of Deep Generative Models for Rapid Topology Optimization

“…Researchers have explored the application of deep learning model architectures based on Convolutional Neural Network (CNN) and generative models to reduce the computational cost of optimization algorithms [4]. Liu [5] proposed a hybrid technique which uses a CNN to enhance the resolution of a lowfidelity topology optimization algorithm. On similar lines, Banga [6] used an intermediate result of a conventional topology optimization to accelerate the prediction of the optimal design and reducing the time cost of training data and potential inaccuracy rate using a 3D CNN which is trained on a fixed discretized domain and a single material type while having the variation of location, number and magnitude of the external forces.…”