Machine-learning based design of active composite structures for 4D printing

Active composites are a class of materials that have environmentally responsive components within them. One key advantage of active composites is that through mechanics design, a variety of actuation can be achieved. The development of active composites has been significantly enhanced in recent years by multimaterial 3D printing where different materials can be precisely placed in 3D space, enabling the achievement of shape-shifting of 3D printed parts, or 4D printing. In practical applications, it is highly desirable that the part shape can change in a pre-described manner, which requires the careful design of where to place different materials. However, designing an active composite structure to achieve a target shape change is challenging because it requires solving an inverse problem with spatially heterogenous, highly nonlinear (active) material behavior within a potentially complex boundary value problem. In this paper we present a machine learning approach to the design of active composite structures that can achieve target shape shifting responses. Our strategy is to combine the finite element method with an evolutionary algorithm. In order to achieve a target shape, we compose the structures of equally sized voxel units that are made of either a passive or an active material and optimize the distribution of these two material phases. The optimization method is tested against several illustrative examples in active composite design to show the agreement between the target shape and the best machine learning solution obtained.