Using a neural network in a thermophysical problem

This current work examines the use of an artificial neural network to predict the minimum film boiling temperature of a subcooled liquid at atmospheric and elevated pressures. To train the neural network, it was used more than 2000 different experimental data points were obtained during experiments on the experimental facility. In the current work, several models with different input parameters were considered. Liquid subcooling $\Delta$Tsub, thermophysical properties of the wall, thermophysical properties of liquid, Prandtl number Pr, Grashof number Gr, the linear scale of the cooled body D are used as input parameters. The output of the neural network is wall overheating $\Delta$ Tw. For training, 3 hidden layers with the number of neurons (15,10,3) are used. On average, after training, the prediction error for each model is ±30K, the root-mean-square error is about 10%.