Integration of CFD Simulations and Artificial Neural Networks for Predicting Heat Transfer in Turbulent Incompressible Impinging Jets

Abhary K. A. K., Baş O., Aradağ Çelebioğlu S.

15th International Conference on Computational Heat and Mass Transfer, ICCHMT 2025, Antalya, Turkey, 19 - 22 May 2025, pp.421-432, (Full Text) identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1007/978-3-032-16135-2_39
  • City: Antalya
  • Country: Turkey
  • Page Numbers: pp.421-432
  • Keywords: Artificial Neural Networks (ANN), Computational Fluid Dynamics (CFD), Heat Transfer, Hyperparameter Optimization, Impinging Jets, Nusselt Number, Surrogate Modeling
  • TED University Affiliated: Yes

Abstract

This study integrates Computational Fluid Dynamics (CFD) simulations with Artificial Neural Networks (ANNs) to predict heat transfer characteristics in turbulent incompressible impinging jets. Traditional empirical correlations for impinging jet heat transfer, though useful, exhibit significant limitations and inaccuracies beyond their calibration ranges. By systematically generating data through CFD simulations across diverse nondimensional parameters (Reynolds number, nozzle-to-plate distance, nozzle length-to-diameter ratio, and Prandtl number), this research employs ANN models optimized using hyperparameter tuning (Optuna framework) to capture complex nonlinear relationships. Results demonstrate the ANN’s capability to predict local Nusselt numbers with an average error of less than 2%, significantly outperforming traditional correlations. This ANN-based surrogate model thus offers a rapid and accurate predictive tool suitable for engineering applications involving impinging jet cooling, demonstrating strong potential for generalization to other fluid flow problems.