Machine Learning-assisted Computational Structural Design
The project "Machine Learning-Assisted Computational Structural Design" aims to integrate Machine Learing (ML) techniques into the structural design process, facilitating the exploration of novel structural shapes and enhancing the efficiency of form-finding and structural optimization processes. The research focuses on graph-based representation of structures, Automatic Differentiation (AD) and Graph Neural Networks (GNNs) as core methodologies for addressing challenges in structural design.
The project builds upon the Combinatorial Equilibrium Modeling (CEM) framework, a form-finding method based in graphic statics and graph theory. A key development within this research is the introduction of Logic-Informed Graph Neural Networks (LIGNNs), which incorporate logical constraints as semantic loss functions to ensure that automatically generated topolgies adhere to specified validity conditions. This approach facilitates the automated generation of valid topology diagrams, thereby reducing the manual effort required in structural form-finding.
Collaborators:
Lazlo Bleker, Pierluigi D'Acunto, Professorship of Structural Design, School of Engineering & Design, Technical University Munich
Prof. Dr. Kam-Ming Mark Tam, Department of Architecture, University of Hong Kong
Funding programme:
International Graduate School of Engineering and Design (IGSSE)
Duration:
2021-2025
Related publications:
Bleker, Lazlo; Tam, Kam-Ming Mark; D’Acunto, Pierluigi: A Graph-Based Grammar for Structural Design using Deep Reinforcement Learning. Proceedings of the IASS 2024 Symposium, 2024