Physics-Informed Neural Network Prediction of Nanofluid Thermal Transport in TPMS Gyroid Heat Exchangers

Triply periodic minimal surface (TPMS) heat exchangers offer high surface-area-to-volume ratios and interconnected flow pathways, making them attractive for compact thermal management. However, accurately predicting nanofluid heat transfer over a wide range of nanoparticle concentrations and operating conditions in complex TPMS geometries remains computationally challenging because of the coupled effects of porous architecture, flow dynamics, and concentration-dependent thermophysical properties. In this study, a hybrid physics-informed neural network (PINN) framework was developed to reconstruct concentration-dependent Al2O3−water nanofluid temperature fields in TPMS gyroid heat exchangers. The originality of the proposed approach lies in integrating sparse thermocouple measurements, a steady-state convection–diffusion equation, boundary condition residuals, concentration-dependent nanofluid property models, and a physics-based concentration scaling procedure within a unified framework. The proposed framework was applied to aluminum and silver TPMS heat exchangers over a wide range of nanofluid volume fractions and flow conditions. The trained PINN accurately reconstructed the experimentally measured temperature field, demonstrating excellent agreement with the reference experimental data. Predictions at concentrations beyond the experimentally measured reference condition were obtained using the physics-based concentration scaling model. The effective heat transfer coefficient and Nusselt number were subsequently evaluated from the predicted mean TPMS temperature through an energy balance formulation. Increasing nanoparticle concentration reduced the predicted TPMS temperatures by approximately 17.5–18.5%, while the combined increase in concentration and flow rate produced an overall temperature reduction of about 33.5%. Relative to the selected baseline condition, the combined variation in concentration and flow rate was associated with calculated increases of 62.08% in  heff , 58.33% in Nu, and 59.32% in Re. These results demonstrate the potential of the proposed hybrid PINN framework as a computationally efficient surrogate for evaluating nanofluid-enhanced TPMS heat exchangers, while acknowledging that predictions away from the training concentration depend on the validity of the concentration scaling model.