Thermohydraulic performance of MXene-based nanofluid-cooled micro-porous heat sink for electronics cooling: A numerical and machine learning optimization approach

The current study is based on computational analysis and machine learning (ML) optimization of micro-porousheat sink cooled with water and MXene-based nanofluid. A Eulerian-Eulerian multiphase model is adopted toexplore the effect of different volume fraction (φ) of Ti3C2Tx nanoparticles dispersed in water-based nanofluidunder the Reynold number (Re) range from 300 to 2100. The porosity (ε) and pore density (PPI) are varied inrange of 0.50 ≤ ε ≤ 0.90 and 10 ≤ PPI ≤ 50, respectively, of micro-porous medium to investigate the heattransfer and fluid flow characteristics and optimized using Random Forest Regressor (RFR) and genetic algorithm(GA) framework. The RFR effectively captured the complex, non-linear thermofluid interactions within the heatsink, enabling an accurate prediction of key performance indicators. Nusselt number (Nuavg), log mean temperaturedifference (LMTD), pumping power (PP), and performance evaluation criteria (PEC) are calculated andused to estimate the thermohydraulic performance of the heat sink along with the isothermal streamlines, velocity,and pressure contours to commit on the temperature and flow field distribution across different planes ofthe heat sink. Multi-objective optimization is performed for maximizing cooling performance Nuavg and minimizingPP, using GA. Multi-objective NSGA-II optimization yields a Pareto front with three operational zones.The high-Nuavg zone (Re = 2100, ε = 0.50, PPI = 10, Ti3C2Tx nanofluid at φ = 0.5 vol%) achieves Nuavg = 76.5and PP = 7.79 mW. The energy-efficient low-PP zone (Re ≈ 300–410, ε = 0.50–0.90, PPI = 10, pure water)delivers PP as low as 0.036 mW at Nuavg = 15–37. A balanced trade-off zone (Re ≈ 530–860, ε = 0.50–0.61, PPI= 10) provides Nuavg = 46–67 at PP = 0.19–0.93 mW. The preferred TOPSIS-ranked solution is identified in thetrade-off zone at Re ≈ 861, ε = 0.50, PPI = 10, φ = 0.5 vol% (nanofluid), delivering Nuavg = 66.8 at PP = 0.93mW. The combination of ML and GA offers easy and less computationally expensive solutions compared to CFDonce trained with appropriate data and helps engineers to provide quicker and optimized solutions for applicationslike electronics cooling, renewable energy, and advanced thermal management.