Optimizing base fluid composition for PEMFC cooling: A machine learning approach to balance thermal and rheological performance.

The Proton Exchange Membrane Fuel Cell (PEMFC) is a highly efficient and eco-friendly technology, making it a pivotal solution for sustainable energy systems. Effective thermal management of PEMFCs is essential, and nanofluids have emerged as superior coolants compared to conventional fluids. Less exploration in PEMFC cooling, particularly using reduced graphene oxide (rGO) suspended hybrid nanofluids, supports the present work on the thermal and rheological properties of rGO-based hybrid nanofluids. The experimental exploration involves five different mixtures of base fluid composition comprising ethylene glycol (EG) and water (W). The hybridization of Al₂O₃ and rGO nanoparticles was performed by dispersing both at four different concentrations in the 50:50 base fluid mixture. The experimental procedure involves evaluation of dispersion stability, viscosity, and thermal conductivity of hybrid nanofluids. The results showed that increasing the EG proportion reduced thermal conductivity while increasing viscosity. The maximum thermal conductivity ratio of 1.23 occurred at 80:20 W: EG for 1 vol% concentration at 60 °C, while the highest viscosity ratio of 1.48 was observed at 20:80 W: EG at 30 °C. The developed correlation for viscosity shows an 11.2% reduction in the coefficient of determination obtained for the thermal conductivity model. This study explores the application of Linear Regression (LR), Decision Tree (DT), and eXtreme Gradient Boosting (XGBoost) models for predicting thermal conductivity and viscosity using experimental datasets. The thermal conductivity model showed that XGBoost has the best predictive power, with Test R² = 0.9941, Test mean square error (MSE) = 0.0000, and Test KGE = 0.9613. XGBoost again beat other models in predicting viscosity, with Test R² = 0.9944, Test MSE = 0.0269, and Test KGE = 0.9903. SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) graphs showed that the model outputs were greatly affected by the base fluid ratio (BFR), temperature, and concentration. This made the model outputs easy to understand both globally and locally. These findings provide valuable insights for designing efficient cooling solutions for PEMFCs, supporting their broader adoption in energy applications.