Machine learning-based design and optimization of curved beams for multistable structures and metamaterials

Abstract

Curved beams have been wildly used in MEMS (Micro-electromechanical systems) devices and energy absorption materials owing to its bistability. Almost all curved beams in previous studies have a constant thickness. Although better performance can be achieved by changing the thickness distribution, such as beams of uniform strength, lack of design and optimization tool limits the development and application of curved beams with varying thickness. In this paper, we demonstrate a new approach to design and optimize curved beams based on machine learning, which has been successful in many fields owing to its ability to process big data that can also be used in structural design and optimization. This machine learning-based model is able to achieve accurate predictions of nonlinear structure–property relationships. The optimized designs with different optimization objectives, such as stiffness, forward snapping force, and backward snapping force, are obtained efficiently and precisely. Experimental testing is conducted on specimens with optimized profiles, which are fabricated using a high-resolution multi-material 3D printer. The computational results are validated by the experimental results. The machine learning-based optimization approach developed here can provide a promising tool for the design and optimization of beam-based structures and mechanical metamaterials.