Prediction of experimental condensation heat transfer characteristics of helically coiled tube-in-shell type heat exchangers by Artificial Neural Network and XGBoost

This study addresses the inherent limitations of empirical correlations in accurately predicting condensation heat transfer and pressure drop within helical tubes, a crucial area given their superior thermal performance compared to straight configurations. A novel methodology was presented utilizing advanced machine learning algorithms, specifically ANN and XGBoost, to develop broad applicable and robust predictive models. A comprehensive dataset comprising 369 experimental entries from diverse literature sources, incorporating various pipe types and refrigerants, was structured with seven input and two output parameters. This dataset was judiciously split, with 70% allocated for model training and 30% for testing, randomly. Unlike other studies, the whole data of smooth straight, two types of smooth helical, dimpled helical, and microfinned helical tubes were used together to achieve the most difficult task of predicting the outputs. The results affirm the high predictive power of the developed models. The ANNs achieved a MSE of 0.00415 and a correlation coefficient of 0.99075, indicating strong agreement with actual values. Concurrently, XGBoost demonstrated exceptional performance, yielding a correlation coefficient value of 0.92 for the average condensation heat transfer coefficient and 0.86 for the average friction pressure drop, both with MSE values of 0.00. Further analysis revealed that enthalpy of vaporization and pipe length are the most influential parameters for heat transfer, while mass flux and enthalpy of vaporization predominantly govern friction pressure drop. This research provides highly accurate, data-driven tools that significantly advance the design and optimization of specific helical heat exchangers investigated in present work.