Comprehensive Prediction of the Relative Modulus of Strut-Based Gyroid Lattice Structures Employing an ML-Based Surrogate Model

Due to its porous structure, tunable properties, and nearly isotropic characteristics, the Gyroid Lattice Structure (GLS) is widely utilized in orthopedic implant applications. To effectively reduce stress shielding and enhance implant longevity, it is essential to accurately predict the GLS's elastic moduli across a broad range of relative densities for various material selections. This study conducted a comprehensive finite element analysis of strut-based GLS models, considering unit cell arrangements, a wider range of relative densities, and variations in lattice orientations to predict its relative elastic moduli. GLS models with relative densities of 10%, 30%, 50%, and 75% were experimentally tested and numerically analyzed to capture properties across a broader density range. The well-known Gibson-Ashby model and a machine learning (ML)-based surrogate model employing Gaussian Process Regression were developed to extend predictions across the full-density spectrum. The results showed that different relative densities required varying numbers of unit cells to achieve elastic modulus convergence. The improved Gibson-Ashby model provided closer predictions to experiments over a wider density range but struggled to fully capture behavior at high relative densities near bulk material properties. In contrast, the ML-based surrogate model accurately predicts elastic moduli across the entire relative density range. Compared to experimental results, this approach demonstrates greater accuracy in predicting the elastic modulus of GLS, with reduced error compared to other methods. These findings are particularly valuable for optimizing implant and scaffold designs, as accurate modulus predictions contribute to improved performance and longevity, helping to mitigate the stress-shielding effect.