Heat transfer performance prediction of aluminum slit fin-and-tube heat exchangers via interpretable machine learning

The heat transfer performance of an aluminum slit fin-and-tube heat exchanger is governed by the coupled effects of flow conditions and structural parameters. This study established a research framework integrating experimental validation, computational fluid dynamics, machine learning, and interpretability analysis. First, experimental tests were conducted to validate the numerical simulations, and good agreement was obtained between the numerical and experimental results. Subsequently, using the validated computational fluid dynamics dataset, multiple linear regression, random forest, and extreme gradient boosting models were developed to predict the Nusselt number and friction factor. Shapley additive explanations and Pearson correlation analyses were employed to identify and quantify key influencing factors. Among the three models, extreme gradient boosting exhibited the best predictive performance, yielding independent-test coefficients of determination of 0.9910 for the Nusselt number and 0.9932 for the friction factor, respectively. The interpretability analysis further revealed that the Reynolds number is the dominant factor affecting the Nusselt number, while slit height is the most critical structural parameter. Specifically, the contribution of the Reynolds number to the Nusselt number increases continuously, whereas slit height exhibits a distinctly nonlinear influence. Pearson correlation analysis further identified that the Reynolds number had the highest correlation coefficient with the Nusselt number, reaching 0.81, followed by slit height at 0.54. Overall, these findings enhance the interpretability of data-driven models in heat exchanger performance analysis and provide useful guidance for identifying key parameters and optimizing aluminum slit fin-and-tube heat exchangers.