Answers are grounded in this paper and cite their sources.
Hard-magnetic soft materials (HMSMs) exhibit programmable large deformations under external magnetic fields, which allows for a wide range of applications in soft robotics and actuators. Conventional computational methods for the design and optimization of HMSM-based structures require iterative adjustments of configurations and material parameters, thereby demanding numerous repeated simulations and leading to high computational cost. Here, we develop a mesh-free computational framework based on the deep energy method (DEM) by integrating the potential energy of HMSMs into neural networks, enabling efficient prediction of large deformation. The accuracy of the proposed DEM is validated against experimental results on beam bending and twisting deformations, while its generality is further demonstrated by extending to polydomain structures. Furthermore, a transfer learning strategy is adopted, where pre-trained neural networks enable the learned physical principles to be applied to new cases. By varying geometry, applied magnetic fields, and remanence distributions, we demonstrate that this strategy achieves a reduction in training epochs and computational time compared to training from scratch, while maintaining identical prediction accuracy. Based on this method, we construct magnetically driven biomimetic plants, achieving programmable tuning of the morphology for magnetically actuated thin-walled structures. The proposed DEM-based transfer learning framework provides an efficient and stable computational tool to predict large deformations of HMSMs, offering valuable guidance for rational designs of soft robots and functional surface devices.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: