Novel symbolic regression for turbulence flows of nano-coolant and air in louvered-fin flat-tube automotive radiators: A combination of experimental properties and device real dimensional

This study explores the thermal performance improvement of louvered-fin flat-tube car radiators using a hybrid nanofluid of Al 2 O 3 and SiO 2 nanoparticles in a water-ethylene glycol (50:50) mixture. Through detailed CFD simulations combined with symbolic regression, the research analyzes the impact of various air velocities and nanocoolant flow rates on heat transfer efficiency, pressure drop, and friction factor in turbulence regime. The model enjoys experimental values for each thermo-physical property of this nanocoolant to boost the precision of results. The hybrid nanofluid showed significant improvements in heat transfer performance, with Nusselt numbers increasing by 2 % as nanoparticle concentration rose from 0.05 % to 0.2 %. The highest air velocity of 39 m/s yielded a maximum heat flux of 160 kW/m 2 , with a corresponding augmentation in the Nusselt number. The heat flux distribution along the fins indicated a linear rise with increasing wall temperatures (up to 170 kW/m 2 ) and an oscillating trend due to flow separation and vortex formation between the louvers. Increasing the nanocoolant flow rate from 7 to 10 L/min resulted in a 40 % enhancement in the system's thermal performance, with the Nusselt number peaking at around 90 in the center of the tube. Higher flow rates amplified heat transfer, while the friction factor showed an 11 % increase due to the additional pressure drop. Despite this, the overall performance enhancement coefficient (PEC) remained stable, fluctuating only slightly (around 1.002), indicating that the trade-off between enhanced heat transfer and increased flow resistance was well-managed. Symbolic regression provided high-accuracy mathematical correlations for the heat flux as a function of temperature and position along the fin, reducing computational time while maintaining R 2 values exceeding 0.996. These correlations enabled a more efficient and cost-effective approach to modeling radiator performance, significantly reducing the need for extensive CFD runs.