This study leverages deep feed-forward neural networks (DNNs) to develop a predictive model for estimating the compressive behavior of porous metallic bio-metamaterials based on their geometric and material characteristics. A DNN architecture comprising two hidden layers was trained on an extensive dataset of 3D-printed porous metamaterials with various relative densities and mechanical properties. The model's performance using Mean Absolute Error, Mean Squared Error, and R2 demonstrated high accuracy. Sensitivity analysis identified relative density and applied strain as the most influential parameters. The results underscore the potential of machine learning in rapid design of porous bio-metamaterials, reducing reliance on costly experimental procedures.
Khalvandi et al. (Mon,) studied this question.