Liquid crystal polymers (LCPs) are a versatile class of smart materials that serve orientational order while pose rubber elasticity. They are programmable and stimuli-responsive, undergoing reversible shape transformations under external stimuli. [1] For instance, the surface of an LCP coating can transition from flat to complex, non-planar topographies, governed by the alignment of mesogenic units during photopolymerization. [2] Although diverse surface morphologies have been demonstrated through the introduction of topological singularities in molecular orientation [3], the quantitative link between these orientations and the resulting shapes remains poorly understood. To address this gap, we formulate an inverse problem: given a target shape, determine the molecular alignment that produces it. We trained a series of finite-element simulations using AutoGluon [4], an automated machine-learning (AutoML) framework, to solve this tabular regression task. Remarkably, the model predicts alignment fields with high accuracy, even without explicit prior knowledge of liquid-crystalline or polymer physics. Our findings provide a critical first step toward replacing traditional finite-element analyses with differentiable neural-network surrogates, paving the way for rapid design and optimization of LCP-based systems.

Acknowledgements

This research forms part of the research program financed by the Dutch Research Council NWO InfoSkin 19440 and NWO TacPlay 9966.

References

[1] D.J. Broer, and T.J. White, Nature materials, 2015, 14(11), 1087-1098.

[2] M.O. Astam, Y. Zhan, T.K. Slot, and D. Liu, ACS Applied Materials & Interfaces, 2022, 14(20), 22697-22705.

[3] Y. You, Y.M. Golestani, D. J. Broer, T. Yang, G. Zhou, R.L.B. Selinger, D. Yuan, and D. Liu, Materials Horizons, 2024, 11(13), 3178-3186.

[4] N. Erickson, J. Mueller, A. Shirkov, H. Zhang, P. Larroy, M. Li, and A. Smola, arXiv, 2020, 2003.06505.