Optimal design of nature-inspired structures with machine learning and theoretical analysis

Abstract

For the last decade, biomimicry has been a good reference to design a system such as high toughness-high strength materials and reversible-versatile adhesive surface. However, the nature-inspired systems have definite limitation because the structures in nature are mimicked based on intuition and domain expertise within the limitation of fabrication, and the structures are designed considering various factors such as survivability in nature. Thus, the structures differ from the optimal structures that are required in practical applications, and there is mismatch between the industrial requirement and the nature-inspired design. Accordingly, in this dissertation, we analyzed the limitation of nature-inspired system such as nacre-inspired composite, gecko-inspired adhesive pillar, and present the optimal design beyond the existing nature-inspired system through machine learning and simulation-based theoretical analysis. Furthermore, for the designs having superior properties, we present machine-learning based design framework, and validate it through grid composite optimization problem.