Recent developments in Deep Learning (DL) suggest a vast potential for Topology Optimization (TO). However, while there are some promising attempts, the subfield still lacks a firm footing regarding basic methods and datasets. We aim to address both points. First, we explore physics-based preprocessing and equivariant networks to create sample-efficient components for TO DL pipelines. We evaluate them in a large-scale ablation study using end-to-end supervised training. The results demonstrate a drastic improvement in sample efficiency and the predictions' physical correctness. Second, to improve comparability and future progress, we publish the two first TO datasets containing problems and corresponding ground truth solutions.
@article{arxiv.2209.05098,
title = {SELTO: Sample-Efficient Learned Topology Optimization},
author = {Sören Dittmer and David Erzmann and Henrik Harms and Peter Maass},
journal= {arXiv preprint arXiv:2209.05098},
year = {2023}
}
Comments
25 pages, 10 figures, submitted to the International Journal for Numerical Methods in Engineering