Machine learning-based prediction of heat transfer characteristics in helically coiled tubes with corrugated walls

Helically coiled tubes with corrugated inner walls are commonly used in compact heat exchangers due to their superior heat transfer performance, enabled by secondary and swirl flows. However, predicting their thermal and hydraulic behavior across varying geometries and flow conditions is complex. This study employs machine learning techniques to estimate two key performance metrics: the Nusselt number and Darcy friction factor. A dataset was generated via computational fluid dynamics simulations, covering Reynolds numbers from 480 to 6100, curvature ratios from 0.096 to 0.288, relative corrugation depths from 0.104 to 0.174, and pitches from 1.04 to 3.13. Four machine learning models were trained: Multi-Layer Perceptron, Extra Trees Regressor, AdaBoost Regressor, and Lasso Regression. Among them, the Extra Trees Regressor exhibits the best performance, achieving an R 2 of 0.9586 and a mean of deviation (MoD) of 5.33% for the Nusselt number, and an R 2 of 0.9576 with a MoD of 6.76% for the friction factor. Explainable AI using Local Interpretable Model-agnostic Explanations (LIME) highlighted the physical relevance of key features. The results indicate that machine learning approaches can serve as accurate and complementary tools to empirical models for interpolation within the investigated parameter space in designing complex heat exchangers.