Modeling shape transformations in liquid crystal elastomers: a machine learning approach to inverse design

Liquid Crystal Elastomers (LCE) are stimuli-responsive, programmable actuators that undergo shape-morphing in response to a change of temperature, illumination, or other environmental cues. The resulting actuation trajectory is programmed by patterning the nematic director field, for instance by forming the material between two glass substrates with prescribed surface anchoring patterns which may be identical or entirely different. Removing one of the substrates leaves behind an LCE coating, while removing both produces a freestanding thin LCE film. Using a GPU-based finite element simulation developed in-house, we explore mechanisms by which arrays of topological defects in the microstructure of LCE thin coatings give rise to complex transformations in surface topography. A key challenge in LCE device engineering is the inverse design problem: find the nematic director field that morphs a sample to a desired target shape e.g. when heated uniformly. Inspired by recent collaboration with the Lavrentovich group [1], we investigate the inverse design problem for LCE surface coatings that morph to form complex topographies. We present a machine learning approach to address this inverse design challenge. First, we solve 1500 “forward” problems using our fast finite element approach [2] for different director configurations to form a training dataset. Next, we train a stacked ensemble regression model using the Autogluon framework [3]. Here 80% of the dataset was used to train the models including tree-based and deep learning algorithms. The prediction of the parameters defining the director field on the remaining test dataset was evaluated. The ensemble model outperformed any individual model and could also predict configurations that departed from the initial geometry by adding noise. As an example we model the design of morphing LCE suction cups to mimic the gripping capabilities of octopus suckers. We discuss plans to extend this approach to a broader class of LCE device geometries.

[1] G. Babakhanova et al, Journal of Applied Physics 128, 184702 (2020)

[2] A. H. Gelebart et al, Nature v. 546, 632 (2017)

[3] auto.gluon.ai