Machine learning framework to predict the mechanical properties of photopolymer gyroid lattices at various strain rates

Photopolymer gyroid lattices are advanced materials owing to their exceptional mechanical performance and impact energy absorption capability. However, determining their mechanical properties is challenging because of the complex interactions between their intricate shapes, material properties, and manufacturing defects. In this work, a machine learning (ML) framework was developed to forecast the mechanical properties, including Young's modulus, yield strength, fracture strength, fracture strain, and toughness, of photopolymer gyroid lattices at low computational costs. This framework employs a recursive feature elimination technique to reduce less important features, followed by a grid search method to optimize the hyperparameters of the model. Also, we applied random forest, extra tree, artificial neural network, and long short-term memory techniques and evaluated their performance using coefficients of determination (R²), mean absolute error (MAE), and root mean square error (RMSE). The extra tree and random forest algorithms produced the best results. In addition, we examined the effects of strain rate and porosity on the mechanical properties. The SHapley Additive exPlanations (SHAP) procedure was applied to show the effects of the individual attributes on the modeling results. This framework provides a cost-effective and time-efficient alternative to experimental procedures, enabling precise prediction of mechanical properties of photopolymer gyroid lattices.