Machine Learning‐Assisted Inverse Design of 3D Shape Morphing in Liquid Crystal Elastomer Composite Strips

Abstract The design of active composites with programmable 3D shape‐morphing capabilities is essential for functional applications. However, the inverse design problem, determining optimal material profiles to realize target shapes, remains challenging for structures with vast design spaces and highly nonlinear mechanics. Conventional approaches, often based on intuition or finite element (FE) analysis, are computationally prohibitive. Here, a machine learning (ML)‐assisted framework is presented for rapid inverse design of 3D morphing in 4D‐printed voxelized liquid crystal elastomer (LCE) composite strips. The approach integrates both forward and inverse processes. The forward model employs recurrent neural networks (RNNs) trained on FE simulation data to deliver accurate and efficient 3D shape predictions. Leveraging this forward model, a Sequential Genetic Algorithm (SGA) is developed to effectively explore the large design space and identify the optimal material profiles. The combined ML‐SGA framework demonstrates both high accuracy (fitness > 0.995) and efficiency. Designs obtained through this approach are validated by comparison between simulations and experiments, showing excellent agreement with target 2D and 3D morphologies. This work provides a new scheme for the 3D morphing inverse design of beam‐like 4D‐printed composite structures and highlights the promise of data‐driven approaches for the on‐demand design of complex functional structures.