Machine learning-enhanced analysis of liquid nitrogen spreading on superheated walls: An artificial neural network-based investigation

This research investigated the kinematics of liquid nitrogen droplets striking superheated surfaces to enhance heat transfer in cryogenic cooling applications. This process is essential for enhancing energy efficiency and temperature regulation. Experimental data, including wall temperature and Weber number, were evaluated using two artificial neural network models that depict boiling and atomization regimes. Both models used multilayer perceptron architectures, using temperature, Weber number, and dimensionless time as input characteristics to estimate the spreading coefficient. The study reported that increasing Weber number enhances droplet impact kinetic energy, leading to higher maximum spreading coefficients and prolonged maximum spreading times, while higher wall temperatures inhibit spreading due to increased bubble nucleation and viscous dispersion. In particular, when the Weber number was 150, the maximum spreading coefficients reached 4.2, while when the Weber number was 50, the maximum spreading coefficients were limited to 2.8. The models achieved high prediction accuracy, evidenced by coefficient of determination values of 0.99759 and 0.99951, alongside minimum mean squared errors of 2.05E-02 and 3.09E-03, therefore confirming their effectiveness in mimicking droplet spreading behavior. The artificial neural network models effectively represented the intricate relationships between impact kinetic energy and heat transfer pathways. The innovation is in the two-stage modeling of distinct regimes, resulting in superior accuracy compared to prior attempts. This methodology has the potential to improve prediction models in diverse cryogenic cooling applications, hence advancing the comprehension of thermofluid processes in the area.