A novel pin finned structure-embedded microchannel heat sink: CFD-data driven MLP, MLR, and XGBR machine learning models for thermal and fluid flow prediction

Abstract

Microscale mechanical systems benefit from microchannel heat sinks' heat transfer efficiency. This study uses six new pin fin shapes embedded in microchannel to improve heat dissipation. The study optimizes geometries based on overall thermal performance for Reynolds numbers ranging from 150 to 350. Validation using numerical and experimental data indicates variations under 10 %. Graphical representations show that the proposed pin fin configurations outperform the baseline case. From numerical investigation, the circular perforated fish fin increases Nusselt number by 25 %, whereas the elliptical fin lowers pressure loss by 66.95 %. Elliptical fins enhance thermal performance by 30 %, proving that novel designs and optimization tactics work. Six machine learning models are considered to predict Nu and pressure drop. Keras and sklearn were used for MLP, MLR, and XGBR was imported from XGBoost. Model performance was assessed using the R2 test, MRE, RMSE, and rRMSE. MLP (R2 test = 0.950, MRE (%) = 11, rRMSE = 2.3 %) and XGBR (R2 test = 0.909, MRE (%) = 2.2, rRMSE = 2.9 %) model a good estimation of Nu. XGBR (R2 test = 0.999, MRE (%) = 1.2, rRMSE = 1.6 %) also estimates pressure drop with good accuracy. KFold cross validation was employed to evaluate the mean CV score of the developed model.