Ensemble and traditional machine learning models for double pipe heat exchanger with vaned twisted tapes

Abstract

This study investigates the predictive performance of traditional machine learning algorithms and ensemble learning techniques for evaluating the thermal–hydraulic behavior of a double pipe heat exchanger fitted with vane-modified twisted tapes. Experimental data comprising Reynolds number, Nusselt number, friction factor, and performance evaluation criterion were analyzed using exploratory visualization methods to establish baseline trends and parameter interdependencies. Subsequently, models including Linear Regression, support vector regression, k-nearest neighbors, and decision tree were compared with ensemble approaches such as random forest, gradient boosting, AdaBoost, and XGBoost. Performance was evaluated using R², mean absolute error (MAE), and root mean squared error (RMSE) metrics. Results revealed that gradient boosting achieved the highest predictive accuracy (R² = 0.975, MAE = 0.018, RMSE = 0.025), outperforming both traditional models and other ensembles, gradient boosting excelled in predicting performance evaluation criterion. Although the decision tree showed competitive accuracy, its tendency to overfit limited generalization. Regularization techniques did not significantly improve results, underscoring the importance of ensemble-based approaches in capturing nonlinear interactions. The novelty lies in applying ensemble learning to model the complex nonlinear interdependencies in vane-angled twisted tape inserts, achieving superior predictive accuracy over traditional machine learning and offering insights for efficient double pipe heat exchanger optimization. Overall, the study demonstrates that ensemble learning methods, particularly Gradient Boosting, provide robust and reliable predictive frameworks for modeling double pipe heat exchanger performance, offering valuable insights for optimizing heat exchanger design.