Spreading dynamics of a droplet impacts on a supercooled substrate: Physical models and neural networks

The phenomenon of droplets impacting substrate exists widely in nature and industrial production. We investigate the maximum spreading time and the maximum spreading factor of a water droplet impacting a supercooled substrate. By collecting data on the spreading dynamics of a water droplet impacting a supercooled substrate, we compared the suitability of four physical models of maximum spreading time and five physical models of maximum spreading factor to the data. Current physical models more accurately predict the maximum spreading time for a water droplet impacting a supercooled substrate, however, the prediction accuracy of the model for the maximum spreading factor is low. We use the Reynolds number, Weber number, Reynolds number, Capillary number and dimensionless subcooled temperature of the droplet impacting on the supercooled substrate as the input layer, and establish different neural network models to predict the droplet spreading dynamics. We establish Convolutional neural network, and use Spreading pattern and Fingering pattern as output layer to classify and predict impact patterns. In addition, we established an optimized BP neural network with the maximum spreading factor as the output layer for regression prediction.