Data-driven inverse design of a dual-stiffness TPMS fusion cage with experimental validation

Data-driven design of architected biomedical structures requires controllable geometry-property relationships and experimental validation of manufacturable designs. This study proposed an inverse design framework for a dual-stiffness triply periodic minimal surface (TPMS) fusion cage by integrating parametric finite element analysis (FEA), deep learning, laser powder bed fusion (LPBF), and mechanical testing. A TPMS-Gyroid cage was parameterized by wall thickness, internal gap, pillar dimension, and unit-cell size to regulate contact-induced stiffness transition. FEA identified parameter effects on staged stiffness, transition load, and porosity. Based on 800 simulated designs, a multilayer perceptron (MLP) was trained to establish the nonlinear geometry-property mapping, achieving high predictive accuracy ( ) for force, elastic modulus, cage height, and porosity, and enabling inverse generation of geometric parameters from target mechanical performance. Optimized Ti6Al4 V cages were fabricated by LPBF and evaluated under uniaxial compression, three-point bending, and eccentric loading, representing flexion, extension, and bending, respectively. The experimental results agreed well with the numerical predictions and confirmed a robust bilinear, contact-induced dual-stiffness response. The cages showed a compliant first stage, second stage modulus of to , and porosity of to . This work provides a complete, transferable data-driven design route for additively manufactured TPMS fusion cages.