Machine learning analysis of boiling regimes and spreading dynamics in liquid nitrogen droplet impact on superheated surfaces

The impact of liquid nitrogen droplets on superheated surfaces involves a complex interplay of transient fluid flow, heat transfer, and phase change phenomena, all of which critically influence the efficiency of spray cooling systems. To investigate this behavior, we develop two machine learning models: one for classifying boiling regimes upon droplet impact and another for predicting spreading dynamics across the regimes. A dataset of 5274 experimentally obtained impact images is used to evaluate seven convolutional neural network architectures with ResNet-18 achieving the highest classification accuracy of 99.8%. Based on this classification, a multilayer perceptron model is trained using five key parameters—Reynolds number, Weber number, Ohnesorge number, Capillary number, and surface temperature—to predict the maximum spreading coefficient and spreading time. Compared to traditional empirical methods, our models significantly improve predictive accuracy, reducing errors to below 4.8% and 8.4% for the respective parameters. In addition to deepening the fundamental understanding of cryogenic droplet dynamics, this machine learning-based framework offers valuable potential for optimizing the design and performance of cryogenic cooling systems.