Machine learning-based prediction of Nusselt number for vertical helical coils in natural convection heat transfer

Abstract

Machine learning algorithms are used to predict the Nusselt Number (Nu) for vertical helical coils. Nusselt Number is important in heat transfer studies, especially in convective situations like vertical helical coils. Rayleigh number (Ra), emissivity (e), coil diameter to wire diameter (D/d), and pitch ratio (p/d) are used to construct reliable Nu prediction models. Data is collected using numerical simulations utilizing the finite-volume approach conducted in the laminar regime for the specified ranges of non-dimensional parameters: Rayleigh number (10⁴ ≤ Ra ≤ 10⁸), surface emissivity of the coil (0 ≤ ɛ ≤ 1), pitch to rod diameter of the coil (3 ≤ p/d ≤ 7.5), and coil height to rod diameter (40 ≤ H/d ≤ 60). Temperature-dependent fluid characteristics have been used to get precise outcomes. 400 samples with a wide variety of parameter values were collected. For model training and evaluation, the dataset was split into training (70 %) and testing (30 %) sets. Machine learning models included Decision Trees, Random Forest Regression, K-Nearest Neighbors, Extreme Gradient Boosting, and Support Vector Regression. Model performance was assessed using MSE, RMSE, MAE, and R-squared scores. All evaluation measures showed that DT predicted the Nu best. This study proves machine learning can anticipate vertical helical coil Nu. The models help engineers and academics make more accurate convective heat transfer coefficient predictions for helical coil heat transfer studies. In engineering applications, this research improves heat transfer process understanding and optimization.