Abstract
There exist many studies on optimization of shell and spatial structures including shape optimization of shells and topology optimization of trusses. Recently, with the rapid development of machine learning and generative AI, these technologies have increasingly been applied to optimization of trusses and shell structures.
In this presentation, several applications of machine learning to optimization of shells, spatial frames, trusses, and building frames are introduced. Machine learning is a fundamental technology underlying AI, and the purpose of its applications to structural optimization include accelerating the optimization process through function approximation and the identification of feasible solutions, one-shot generation of optimal solutions from design conditions and initial solutions, reducing the number of design variables, post-processing of optimal solutions, etc.
Machine learning is generally classified into three categories: supervised learning, unsupervised learning, and reinforcement learning. Although supervised learning is the most widely used approach, it requires many optimal solutions or numerical analysis results to be prepared as training data before starting the learning process. By contrast, reinforcement learning learns the optimization process itself and can generate training data during the learning process.
It is shown in the numerical examples that various types of reinforcement learning can be successfully applied to optimization of latticed shells and trusses. In addition to application to shells and spatial frames, optimization results are shown for building frame structures using Transformer architectures utilizing a similar architecture of learning process employed in large language models (LLMs). Finally, the current state of generative AI applications to structural optimization and their future prospects are discussed.
Biography
Prof. Makoto Ohsaki received Ph.D. from Kyoto University, and is currently an Emeritus Professor, Kyoto University. He continues research and education as Visiting Professor at Ritsumeikan University, Japan. He has contributed to academic development of analysis and design of shells and spatial structures for more than 40 years mainly through his research on structural optimization, which has evolved into machine learning for optimization and geometric design. He contributes to IASS as the Treasurer and a member of Executive Council. He published more than 300 reviewed journal papers, and co-authored several monographs including six English books. He also contributed to academic fields as a Vice-President of Architectural Institute of Japan, and the President of Asian Society of Structural and Multidisciplinary Optimization.
Keywords
- optimization
- machine learning
- reinforcement learning
- spatial frame
- truss