New configuration of truss structures through new numerical modeling and optimization method

Developing methods to obtain optimal cross-section, topology, and geometry of trusses is an established field of research. However, due to nonlinearity and multi-objectivity of structural requirements, it is still difficult to obtain desirable solutions. This presentation comprises two parts with respect to core methods handling the difficulty. In the first part, based on force density method, a new method of simultaneous optimization of geometry and topology is presented for trusses. The difficulties due to existence of melting nodes are successfully avoided by considering force density, which is the ratio of axial force to the member length, as design variable. This method does not need constraints on nodal locations to avoid coalescent nodes, and enables to generate optimal solutions with a variety in topology and geometry. In the second part, several machine learning methods, such as clustering analysis and q-learning, are integrated into the optimization process. Since machine learning is able to extract specific feature of the data structure, it has potential to enhance the efficacy of optimization, despite the nonlinearity of the optimization problem.

Kazuki Hayashi is currently a doctoral student at Kyoto University. He obtained his Masters of Engineering at Kyoto University in 2018. During his master program, he studied in the Building Technology discipline group at the Massachusetts Institute of Technology (MIT) and worked in the research center of MakMax group, a company specializing in membrane structures. His current research agenda is to develop a novel design process that allows close collaboration between architects and structural engineers, by means of machine learning and structural optimization. For more details about his projects, visit: hayashikazuki.net