Prediction of overall heat transfer coefficient in concentric tube heat exchangers using artificial neural networks: A comparative study with empirical correlations

Abstract

This study presents the development and implementation of an Artificial Neural Network (ANN) model for predicting the $U$-value in a counter-flow concentric tube heat exchanger (CTHE). A dataset comprising 2,700 CFD simulations was generated by varying key parameters, including Reynolds numbers (1,000–20,000), fluid pairings (hot air-cold air, hot air-cold water,hot water-cold air, and hot water-cold water), inner diameters (0.01–0.05 m), diameter ratios (1.25, 1.5, and 3), and heat exchanger lengths (0.4–4 m). The simulations captured both laminar and turbulent flow regimes, providing a robust basis for training the ANN model. The neural network, comprising three hidden layers, L2 regularization and ReLU activation, demonstrated excellent accuracy, with a low mean absolute error (MAE) of 5.503 and mean absolute percentage error (MAPE) of 3.08%, as evaluated on the test dataset. The ANN model demonstrated superior performance compared to traditional empirical correlations from the literature, such as those by Baehr and Stephan, Dittus and Boelter, and Gnielinski, particularly in mixed flow regimes (laminar-turbulent and turbulent-laminar). While existing literature correlations in these regimes often exceeded 20% APE, our ANN model demonstrated a median APE of less than 1%. This illustrates the superiority of the artificial neural network (ANN) in capturing complex heat transport dynamics over empirical models. Furthermore, SHAP feature importance analysis revealed that the cold fluid thermal conductivity, hot fluid Reynolds number, hot fluid dynamic viscosity and inner diameter have the greatest impact on the overall heat transfer coefficient. The ANN model offers a flexible and accurate alternative to empirical correlations, with the potential to be extended to more complex heat exchanger configurations and additional performance metrics such as pressure drop and heat exchanger effectiveness.