Predicting mechanical properties of elastomer honeycombs using fast image-based measurements

• Fast image-based metrology data and mechanical properties are measured for 65 parts, resulting in length and thickness information for every lattice wall. • Precise measurements of lattice wall dimensions reveal part distortions that affect mechanical properties. • Feature measurements train machine learning models to predict mechanical properties. • A graph neural network demonstrates the best performance to predict and extrapolate mechanical properties. Elastomer lattice structures provide tunable mechanical properties and full elastic recovery over large strains. However, the design of elastomer lattice materials can be challenging when the manufacturing process induces geometric deformations that can significantly affect the material properties. This paper presents a method for fast, image-based measurements of geometric deformations in additively manufactured elastomeric honeycombs and their impact on mechanical properties. A total of 55 parts were designed, fabricated, and tested; these parts have linear elastic stiffness 15.4–1422.8 kPa and plateau stress 1.4–164.3 kPa. Imaging measurements extract the length and thickness dimensions of every wall in every part, revealing small but random distortions that arise from the manufacturing process. An additional 10 parts were modified to have larger deformations for the purpose of the study. Machine learning models trained using the measured wall dimensions accurately predict the mechanical properties. Models that use measured wall dimensions outperform models based on design values only. A Graph Neural Network (GNN) predicts the linear elastic stiffness, plateau stress, and densification onset with 18.2 %, 16.0 %, and 3.2 % error. This method could aid designers in understanding the properties of elastomer lattice structures and the impact of manufacturing-induced deformations.