Comparative machine learning prediction study of hybrid nanofluid flow in a magnetized dimpled tube


Gürdal M., Tan M., GÜRSOY E., ARSLAN K., Gedik E.

Applied Thermal Engineering, vol.281, 2025 (SCI-Expanded) identifier

  • Publication Type: Article / Article
  • Volume: 281
  • Publication Date: 2025
  • Doi Number: 10.1016/j.applthermaleng.2025.128569
  • Journal Name: Applied Thermal Engineering
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, PASCAL, Aerospace Database, Business Source Elite, Business Source Premier, Communication Abstracts, Compendex, INSPEC, Metadex, DIALNET, Civil Engineering Abstracts
  • Keywords: Constant magnetic field, Dimpled tube, Machine learning approach, Nanofluid, Polynomial regression, XGBoost
  • Ankara Yıldırım Beyazıt University Affiliated: Yes

Abstract

This study experimentally examines thermo-hydraulic performance of mono and hybrid nanofluids (Fe3O4/H2O, Cu/H2O, and Fe3O4–Cu/H2O) flowing through smooth (ST) and dimpled tubes (DT) under laminar conditions (Re = 1131–2102) with constant heat flux. A total of 95 cases were tested while a constant direct magnetic field (MF = 0.03, 0.16, 0.3 T) was applied via twin coils; performance was assessed using the Heat Convection Ratio (HCR), Pressure Ratio (PR), and Performance Evaluation Criterion (PEC). Baseline validation against Shah–London and Hagen–Poiseuille correlations showed deviations ≤5.85% (Nu) and ≤4.11% (f). DTs enhanced heat transfer substantially: with Fe3O4/H2O, HCR in DT exceeded ST by up to 43.2% at Re = 2102, while pressure penalties remained moderate. MF strength critically shaped outcomes: 0.16 T consistently improved HCR and yielded the best thermo-hydraulic balance (higher PEC), whereas 0.3 T increased PR and could depress PEC below unity, especially in ST. Data-driven models (Linear, Polynomial, XGBoost, ANN) were trained to predict HCR, PR, and PEC. Polynomial Regression achieved the highest accuracy for HCR and PR on the test set (R2 ≈ 0.99), while XGBoost provided slightly superior PEC predictions. SHAP analyses identified MF strength and dimple geometry as the dominant drivers across targets, with velocity/Re effects modulating performance. The results demonstrate that DTs combined with low-to-moderate MF intensities and Fe3O4-based nanofluids deliver practical heat-transfer gains with acceptable pumping costs; the accompanying predictive models furnish design-ready surrogates for rapid optimization of magnetically assisted compact heat exchangers.