We propose a neural-operator-driven design framework that enables accurate and diverse shape-morphing programming for soft materials with complex geometries. The mapping from the material property field to the deformation field is formulated as an operator learning problem. The proposed Spectral and Spatial Neural Operator (S2NO) can enable high-fidelity shape-morphing prediction for soft materials with complex geometries. The efficient forward prediction of S2NO empowers evolutionary algorithms to thoroughly explore the design space, thereby achieving accurate and diverse morphing designs.
To test the performance of our S2NO-driven design framework, we design a series of shape-morphing soft materials featuring various complex geometries, including irregular-boundary shapes (dart, human, stingray, and blade shapes), porous structures (dome and butterfly structures), and thin-walled structures. The codes for model training and optimisation design are available here. The trained S2NO model is in the "Modelling\logs" folder. The optimisation results are in the "Optimisation\final_results" folder. Moreover, all test data in this study have been deposited at (https://drive.google.com/drive/folders/1OUKJacsNhWbBd7ByzSLCTpTYepj2V0ju?usp=drive_link).