Using a generative adversarial network for the inverse design of soft morphing composite beams

Abstract

The inverse design of structures having tailored properties is challenging mainly due to the multiple design solutions that can satisfy the prescribed conditions. For example, in the inverse design of morphing composite beams, different fabrication solutions exist because the material, geometry and actuation can be varied. On the other hand, the problem can be highly nonlinear due to the large deformations present in such problems. For this reason, we present a generative adversarial network-based inverse design method for constructing soft composite beams that morph into target shapes and can carry out complex prescribed motions. Our approach makes use of composites with passive and active layers that deform into prescribed shapes due to the strain mismatch induced by the non-homogeneous geometric and material properties as well as temperature actuation. To test the proposed method and explore the parametric space much faster than with heating and cooling, we established a mechanical analog (a toy model) that exploits the mechanical stretching of highly elastic, active layers. Experiments and numerical examples demonstrate the effectiveness of our simple toy model, for which the generator network takes the target shapes as inputs and generates the corresponding design parameters for the fabrication of composite beams that self-deploy into prescribed shapes when released. We extended our method for generating the design parameters for forming soft, morphing composite beams that exhibit complex targeted motions when actuated by temperature. Our data-driven method is simple, yet robust enough to provide solutions to complex problems and aid in the future design of soft robots and smart-deployable structures.

Topics & Concepts