SELTO: Sample-Efficient Learned Topology Optimization

TL;DR

This paper introduces SELTO, a deep learning approach for topology optimization that enhances sample efficiency through physics-based preprocessing and equivariant networks, supported by new datasets for benchmarking.

Contribution

It presents a novel, sample-efficient DL pipeline for topology optimization and provides the first datasets with ground truth solutions for the field.

Findings

Significant improvement in sample efficiency.

Enhanced physical correctness of predictions.

Availability of new datasets for benchmarking.

Abstract

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.