This paper presents a machine learning (ML)-based approach for designing active composite (AC) plates with 3D shape changes. A residual network ML model predicts the actuated shape from the material distribution, while a global-subdomain design strategy using ML-integrated gradient descent (GD) and evolutionary algorithm (EA) optimizes the material distribution for desired shape changes. The method demonstrates high efficiency for various target shapes, achieving optimized designs for multiple irregular targets using ML-EA and a normal distance-based loss function. This approach provides an efficient tool for designing 4D-printed active composites.

Xiaohao Sun, Liang Yue, Luxia Yu, Connor T. Forte, Connor D. Armstrong, Kun Zhou, Frédéric Demoly, Ruike Renee Zhao, H. Jerry Qi