Integration of , , and Machine Learning for Turbulent Neutral Jet Flows

This chapter presents an integrated framework combining particle image velocimetry (PIV), computational fluid dynamics (CFD), and machine learning (ML) techniques to investigate turbulence characteristics in neutral jet flows. High-resolution PIV experiments were conducted to capture detailed velocity fields and turbulence statistics, providing essential data for both model validation and ML training. Analytical models based on Gaussian velocity distributions were developed to describe the self-similar behavior of axisymmetric jets and predict Reynolds shear stresses. Complementary CFD simulations employing realizable k - ϵ , SST k - ω , and standard k - ω turbulence models were performed, with the standard k - ω model demonstrating the best agreement with experimental measurements. Ensemble ML methods, including Random Forest, Gradient Boosting, and XGBoost regressors, were trained on combined experimental and simulated datasets to predict turbulent Reynolds stresses. Among these, XGBoost achieved the highest predictive accuracy. A detailed hyperparameter sensitivity analysis revealed the optimal model configuration and emphasized the importance of moderately complex trees and fine-grained feature splitting for turbulence prediction. The results underscore the nonlinear, multiscale nature of turbulent flows and demonstrate the efficacy of integrating experimental, numerical, and data-driven approaches to advance turbulence modeling. The framework developed in this study offers significant insights for enhancing predictive capabilities in fluid dynamics and provides a foundation for future hybrid physical and data-driven modeling strategies.