This study experimentally investigates the thermal and hydrodynamic performance of water jet impingement cooling for high heat-flux electronic chips. To efficiently examine multiple factors with a limited number of tests, the Taguchi design method is employed. The control parameters included heat flux, nozzle diameter, flow rate, nozzle-to-surface distance, and inlet fluid temperature, while the response variables are the convective heat transfer coefficient, Nusselt number, pumping power, and surface temperature. Experiments are conducted at three levels of heat flux (40, 60, and 80 W/cm2). The results show that decreasing the nozzle diameter from 1.0 to 0.6 mm and increasing the flow rate from 100 to 180 ml/min enhanced the heat transfer coefficient by more than 30% (up to 2.8 W/cm2⋅K), reduced surface temperature by over 10◦C, and increased pumping power proportionally. Nozzle height exhibited a non-monotonic influence, with optimal performance at an intermediate spacing. Raising the inlet fluid temperature from 20 to 30◦C increased the heat transfer coefficient by approximately 7%, but also raised the surface temperature by nearly 8◦C, while having negligible impact on pumping power. Additionally, five machine learning models are trained to predict Nusselt numbers beyond the experimental dataset (27 samples). Among these, the multilayer perceptron (MLP) model achieved the highest predictive accuracy with an R2 of 0.894, the lowest MAE (0.995) and MSE (1.586), outperforming the other algorithms. These results demonstrate the combined effectiveness of Taguchi design and ML-based prediction in optimizing and extending jet impingement cooling performance for electronic chips.

Keywords