[論文レビュー] Shape-morphing programming of soft materials on complex geometries via neural operator

Lu Chen, Gengxiang Chen+5 more|arXiv (Cornell University)|Jan 16, 2026

Advanced Materials and Mechanics被引用数 0

ひとこと要約

The paper introduces S2NO, a Spectral and Spatial Neural Operator, to predict and inverse-design shape-morphing of soft materials on complex geometries, enabling high-fidelity morphing and super-resolution material distributions via discretisation-invariant operator learning and evolutionary optimization.

ABSTRACT

Shape-morphing soft materials can enable diverse target morphologies through voxel-level material distribution design, offering significant potential for various applications. Despite progress in basic shape-morphing design with simple geometries, achieving advanced applications such as conformal implant deployment or aerodynamic morphing requires accurate and diverse morphing designs on complex geometries, which remains challenging. Here, we present a Spectral and Spatial Neural Operator (S2NO), which enables high-fidelity morphing prediction on complex geometries. S2NO effectively captures global and local morphing behaviours on irregular computational domains by integrating Laplacian eigenfunction encoding and spatial convolutions. Combining S2NO with evolutionary algorithms enables voxel-level optimisation of material distributions for shape morphing programming on various complex geometries, including irregular-boundary shapes, porous structures, and thin-walled structures. Furthermore, the neural operator's discretisation-invariant property enables super-resolution material distribution design, further expanding the diversity and complexity of morphing design. These advancements significantly improve the efficiency and capability of programming complex shape morphing.

研究の動機と目的

• Motivate shape-morphing design of soft materials on complex geometries for applications like biomedical devices and aerospace systems.
• Develop a neural-operator model that captures global and local morphing on irregular domains.
• Enable voxel-level material distribution optimization via integration with evolutionary algorithms.
• Demonstrate discretisation-invariant capabilities for super-resolution morphing designs.
• Show generalisation across multiple geometries through multi-geometry modelling.

提案手法

• Propose S2NO, combining Laplacian-eigenfunction spectral encoding with spatial graph-convolutional processing to learn operators from material distributions to deformation fields.
• Encode geometry with Laplacian eigenfunctions to perform spectral convolution; couple with graph-based spatial convolution and input/output gates.
• Train with FE-generated data (55,000 random distributions per case; 7 cases) using a lifting layer, 8 S2NO layers, and a projection layer; use MSE-based loss.
• Leverage discretisation-invariance to enable super-resolution material distribution design without retraining.
• Integrate S2NO with evolutionary optimization to perform exhaustive design space exploration for inverse morphing tasks.
• Demonstrate fine-tuning of low-to-high resolution voxel models to realise high-resolution morphing with limited data.

実験結果

リサーチクエスチョン

• RQ1 Can S2NO accurately predict shape morphing for soft materials on irregular and complex geometries?
• RQ2 Does discretisation-invariance enable reliable super-resolution material distributions without extensive retraining?
• RQ3 Can the framework perform robust inverse design to achieve diverse target morphologies on complex geometries?
• RQ4 How well does multi-geometry modelling enable knowledge sharing across different geometries?
• RQ5 What is the comparative performance of S2NO versus existing neural-operator models on these tasks?

主な発見

• S2NO delivers state-of-the-art predictive performance across seven geometry tasks compared with PODNN, DeepONet, POD-DeepONet, NORM, and Transolver.
• Across seven cases, S2NO achieves lower L2, MAE, and M-Max errors than baselines, with relative improvements up to 54.75% in L2, 55.56% in MAE, and 55.00% in M-Max for certain cases.
• S2NO enables accurate inverse design of target morphologies for irregular-boundary, porous, and thin-walled structures, with maximum errors as low as 1.90–2.85% relative to height or length in reported cases.
• The discretisation-invariant property supports super-resolution design, demonstrated by zero-shot generalisation from low to high resolution voxel data and effective fine-tuning with limited high-res data.
• Multi-geometry modelling improves prediction accuracy across five petals, outperforming single-geometry training, enabling shared parameters across geometries.
• Modular assembly demonstrations show the framework can design complex surfaces (hemispherical, morning-glory, butterfly wing-like morphologies) from simple units.