Regression analysis of heat transfer in twin slot jet impingement with computational fluid dynamics and machine learning techniques

Abstract

This study proposes a hybrid methodology coupling computational fluid dynamics (CFD) and machine learning to analyze the heat transfer characteristics of unconfined twin slot jets impinging on a heated wall. We develop an accurate numerical model, apply machine learning to identify the key parameters, and subsequently perform a regression analysis across a wide range of operating conditions. Specifically, a high-fidelity two-dimensional CFD model is developed, and the solutions are obtained using the k-" turbulence model. The model is validated by comparing the simulated Nusselt number distributions with experimental results on the impingement surface. A large dataset of 1575 simulations is constructed by systematically varying four operating conditions of the jet: jet velocity, nozzle spacing, wall temperature, and jet height. This study employs machine learning for objective feature selection to facilitate the development of interpretable regression models. Four machine learning algorithms, including multi-layer perceptron, random forest, gradient boosting, and support vector regression, are applied to the dataset, identifying jet velocity and jet height as the key parameters governing heat transfer. Focusing on these key parameters, we identify a critical flow transition at a threshold jet height of 12 times the nozzle width, which separates the flow behavior into near-and far-field regimes. For each regime, separate polynomial and power-law regression models are obtained to calculate the average and maximum Nusselt numbers. This methodological approach effectively bridges the gap between data-driven analysis and practical engineering design, providing accurate and effective predictive tools for thermal analysis.