Nusselt number predictive model for Taylor–Couette flow with circumferential grooves under different operating parameters

Abstract

Taylor–Couette flow (TCF) is crucial for rotating machinery in chemical processing, energy systems, and other fields. Previous studies have mainly focused on fluid dynamics characteristics, while the heat transfer mechanism has been insufficiently explored, especially in TCF systems with circumferential grooves. A systematic investigation of the thermal performance with circumferential grooves was conducted through a combination of numerical simulation and machine learning (ML). Key operating parameters (angular velocity, inlet flow rate, and inlet temperature) were selected for a three-factor, five-level experimental design, providing training data for using Polynomial regression, Random forest, and XGBoost models. The comprehensive evaluation results indicate that the Polynomial regression heat transfer predictive model on the test set not only has the highest coefficient of determination (R² = 99.2 %) but also has the lowest mean absolute error (MAE = 6.5) and root mean square error (RMSE = 8.8), demonstrating strong predictive fidelity. The 95 % confidence intervals of the results were determined based on bootstrap sampling, and nine combinations of features, such as angular velocity, significantly affect heat transfer performance based on the Student's t-test. The findings establish a reliable machine learning framework for optimizing TCF heat exchangers while providing a basis for elucidating coupled thermal fluid phenomena in modified Taylor–Couette systems.