MACHINE LEARNING–BASED PREDICTION AND OPTIMIZATION OF RADIATOR COOLING EFFICIENCY USING AL₂O₃–WATER NANOFLUID

Abstract

This study investigates and predicts the enhancement of thermal performance in an automotive radiator using aluminum oxide (Al₂O₃)–water nanofluid compared to conventional water coolant. Experiments were conducted with nanoparticle concentrations ranging from 0.001 to 0.01 vol% and coolant flow rates between 7 and 10 L/min using a closed-loop test rig at a constant inlet temperature of 50 °C. Key thermal–hydraulic parameters—including heat transfer rate, convective heat transfer coefficient, Reynolds number, and pressure drop—were experimentally measured and then used to train machine-learning models for predictive analysis. The nanofluid was prepared using a two-step mechanical stirring method to ensure uniform dispersion and stability. Among the models tested, the Random Forest Regressor achieved the highest accuracy (R² = 0.97) in predicting the convective heat transfer coefficient and pressure drop. Model-based optimization identified 0.007 vol% concentration and a flow rate of 10 L/min as the optimal conditions, yielding a heat transfer coefficient of 1437 W/m²·K—over 25 % higher than that of water. The hybrid experimental–machine-learning approach not only validated the superior thermal performance of Al₂O₃ nanofluids but also demonstrated the potential of data-driven optimization for designing efficient, next-generation automotive cooling systems.