Accelerating Optimization Design of Bio-inspired Interlocking Structures with Machine Learning

Abstract

Structural connections between components are often weak areas in engineering applications. In nature, many biological materials with remarkable mechanical performance possess flexible and creative sutures. In this work, we propose a novel bio-inspired interlocking tab considering both the geometry of the tab head and neck, and demonstrate a new approach to optimize the bio-inspired interlocking structures based on machine learning. Artificial neural networks for different optimization objectives are developed and trained using a database of thousands of interlocking structures generated through finite element analysis. Results show that the proposed method is able to achieve accurate prediction of the mechanical response of any given interlocking tab. The optimized designs with different optimization objectives, such as strength, stiffness, and toughness, are obtained efficiently and precisely. The optimum design predicted by machine learning is approximately 7.98 times stronger and 2.98 times tougher than the best design in the training set, which are validated through additive manufacturing and experimental testing. The machine learning-based optimization approach developed here can aid in the exploration of the intricate mechanism behind biological materials and the discovery of new material designs boasting orders of magnitude increase in computational efficacy over conventional methods.