Machine Learning-Enabled Prediction of Spatiotemporal Strain Evolution in Stereolithography-Fabricated Architected Hydrogels

This study developed a machine learning framework to predict the spatiotemporal evolution of strain fields in stereolithography-fabricated architected hydrogels. The materials consisted of stiff re-entrant honeycomb cellular domains embedded within softer hydrogel matrices, creating heterogeneous structures with nonlinear deformation behaviour. Full-field digital image correlation measurements were collected during uniaxial tensile deformation and used to train a sequence-to-sequence deep learning model combining convolutional encoders with ConvLSTM layers. The model learned spatial strain-localization patterns and temporal deformation evolution from DIC strain maps and accurately predicted subsequent strain fields. Experimental results showed that deformation initially appeared relatively uniform, then transitioned into geometry-controlled strain localization along pathways governed by the programmed hydrogel architecture. The model reproduced both global strain distributions and localized deformation features while reducing high-frequency DIC noise. Error analysis showed that prediction discrepancies were concentrated in regions with high strain gradients, indicating that prediction difficulty depended more on deformation-field spatial complexity than on strain magnitude. Overall, the work links experimentally programmed structural heterogeneity to deformation behaviour and provides a data-driven pathway for predicting and designing architected soft material systems.