[Paper Review] 3D Topology Optimization using Convolutional Neural Networks

Saurabh Banga, Harsh Gehani+3 more|arXiv (Cornell University)|Aug 22, 2018

Topology Optimization in Engineering15 references53 citations

TL;DR

The paper proposes a 3D encoder-decoder CNN approach to accelerate topology optimization by predicting final optimized structures from early iteration data, achieving about 40% time reduction with ~96% structural accuracy.

ABSTRACT

Topology optimization is computationally demanding that requires the assembly and solution to a finite element problem for each material distribution hypothesis. As a complementary alternative to the traditional physics-based topology optimization, we explore a data-driven approach that can quickly generate accurate solutions. To this end, we propose a deep learning approach based on a 3D encoder-decoder Convolutional Neural Network architecture for accelerating 3D topology optimization and to determine the optimal computational strategy for its deployment. Analysis of iteration-wise progress of the Solid Isotropic Material with Penalization process is used as a guideline to study how the earlier steps of the conventional topology optimization can be used as input for our approach to predict the final optimized output structure directly from this input. We conduct a comparative study between multiple strategies for training the neural network and assess the effect of using various input combinations for the CNN to finalize the strategy with the highest accuracy in predictions for practical deployment. For the best performing network, we achieved about 40% reduction in overall computation time while also attaining structural accuracies in the order of 96%.

Motivation & Objective

• Motivate a data-driven alternative to computationally intensive physics-based 3D topology optimization.
• Develop a 3D encoder-decoder CNN to accelerate prediction of optimized material distributions.
• Explore how early iteration data from Solid Isotropic Material with Penalization (SIMP) can serve as input.
• Compare training strategies and input configurations to maximize prediction accuracy for deployment.

Proposed method

• Use a 3D encoder-decoder CNN architecture to map early SIMP iteration data to the final optimized structure.
• Investigate how different input combinations affect CNN performance.
• Analyze iteration-wise progress of the SIMP process to guide input selection for the network.
• Train and evaluate multiple training strategies to identify the best performing configuration.
• Quantify computation time reduction and structural accuracy of CNN predictions.

Experimental results

Research questions

• RQ1 Can a CNN predict the final topology optimization result directly from early SIMP iterations?
• RQ2 What input configurations and training strategies yield the highest accuracy and fastest inference for 3D topology optimization?
• RQ3 How does the CNN-based approach compare to traditional physics-based topology optimization in terms of time and accuracy?

Key findings

• Best-performing network achieved about 40% reduction in overall computation time.
• Structural accuracies of the CNN predictions are on the order of 96%.