Thermal performance of an experimentally observed shear-thinning MWCNT-Fe3O4-EG hybrid nanofluid under coupled shear and buoyancy effects: A CFD-machine learning study

This study investigates the thermal transport efficiency of an experimentally characterized non-Newtonian hybrid nanofluid (HNF) comprising Multi-Walled Carbon Nanotubes (MWCNT), Iron Oxide (Fe 3 O 4 ), and Ethylene Glycol (EG). The problem addresses how rheological transitions–from Newtonian to shear-thinning behavior with increasing nanoparticle volume fraction ( ϕ )–influence heat transfer (HT) under coupled shear and buoyancy forces, which is crucial for advanced cooling and energy systems. A finite-volume-based CFD model was employed to simulate mixed convection in a square enclosure containing a bottom heat source, varying the Richardson number ( R i = 0.01 − 100 ), nanoparticle fraction ( ϕ = 0 − 1.8 % ), and source height. Results reveal that the maximum average Nusselt number ( Nu avg ) occurs for shear-thinning fluids ( ϕ = 1.8 % ) under dominant shear conditions ( R i = 0.01 ), while Nu avg increases by 18–40% as the heat-source height decreases. Within mixed convection ( R i = 1.0 ), Nu avg consistently rises with ϕ , confirming synergistic shear-buoyancy effects. Entropy generation ( S avg ) was analyzed to derive a thermodynamic performance efficiency ( T P E = N u a v g / S a v g ), establishing optimal conditions for thermal system design. Additionally, machine-learning surrogate models (RF, DT, XGB) were developed to predict convective performance, where Random Forest achieved superior accuracy ( R 2 > 0.97). The novelty lies in integrating experimentally validated rheology, coupled shear-buoyancy modeling, and explainable CFD-ML frameworks, which provide physically interpretable, rapid, and reliable predictions for practical applications such as compact heat exchangers, electronic cooling, and renewable energy systems.