Bridging fidelity gaps in the design of TPMS-based metamaterials using neural networks

Accurate prediction of wave transmission in architected metamaterials remains challenging due to discrepancies between low-fidelity simulations and high-fidelity experimental measurements. Low-fidelity simulations rely on assumptions about material properties and are further influenced by manufacturing uncertainties and fabrication errors, leading to deviations from measured behaviors. To address this fidelity gap, we propose a novel multi-fidelity learning framework that integrates simulation and experimental data for precise prediction of transmission spectra based on design geometry parameters. Triply Periodic Minimal Surface (TPMS) lattices are selected as a proof of concept due to their mathematically defined geometries, which allow precise control of structural properties and their influence on vibration transmission. The proposed framework captures the intricate relationship between lattice type, relative density, and transmission response by fusing low-fidelity simulations with high-fidelity experimental data from vibration testing of 3-D-printed samples. Results demonstrate the framework’s ability to predict experimental transmission curves with minimal error across a frequency range of 500 Hz to 12 kHz. This work represents a significant advancement in the predictive modeling of TPMS lattices and establishes a scalable, data-driven methodology for optimizing architected materials in vibration control and beyond.