Experimental analysis and multilayer perceptron ANN prediction of local Nusselt number for turbulent air jet impinging upon a plane surface

This study experimentally investigates the local heat transfer characteristics of a circular turbulent air jet impinging perpendicularly on a flat surface, with Reynolds numbers ranging from 5,000 to 30,000 and nozzle-to-plate spacing ratios (S/D) between 2 and 8. The results indicate that the Nusselt number reaches its highest value near the stagnation point (X/D ≈ 0.5), with a secondary peak around X/D ≈ 3 for S/D ≤ 4 and Re ≥ 10,000, followed by a decline in heat transfer at greater radial distances due to reduced radial flow velocity. Increased Reynolds number and reduced S/D enhance heat transfer by intensifying turbulence and disrupting the boundary layer, while the influence of S/D becomes negligible beyond X/D > 6. To model the Nusselt number distribution, a Multilayer Perceptron Artificial Neural Network (MLP-ANN) with eight hidden layers of 20 neurons each was trained using the present experimental data. The ANN demonstrated excellent predictive performance, with correlation coefficients exceeding 0.99, R 2 values above 0.98, and minimal error metrics. The developed model provides an accurate and efficient tool for optimizing impinging jet cooling systems used in applications such as electronics cooling, turbine blade cooling, and advanced manufacturing.