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

Architected materials composed of stiff cellular frameworks embedded within soft matrices exhibit highly heterogeneous and nonlinear deformation behaviors governed by structural heterogeneity. However, understanding and predicting the spatiotemporal evolution of strain fields in such systems remain challenging due to their complex, multiscale mechanical responses. Here, we present a machine learning-enabled framework to predict the spatiotemporal evolution of strain fields in stereolithography-fabricated architected hydrogels using full-field digital image correlation (DIC) measurements. A sequence-to-sequence neural network integrating convolutional encoders with convolutional long short-term memory (ConvLSTM) layers is trained on time-resolved strain maps to capture strain localization dynamics and forecast subsequent deformation states. The model accurately reproduces both global strain distributions and localized deformation pathways, demonstrating strong agreement with experiments while providing denoised and physically consistent predictions. Error analysis reveals that prediction discrepancies are primarily concentrated in regions with high strain gradients, indicating that model performance is governed by the spatial complexity of localized deformation rather than strain magnitude. By linking experimentally programmed structural heterogeneity to deformation behavior through data-driven modeling, this work establishes a predictive framework for understanding structure–property relationships in architected soft materials and provides a pathway toward machine learning-assisted design of advanced material systems.