A novel feed-forward neural network for flow boiling pattern prediction

Abstract

Microscale flow boiling presents a promising solution to emerging cooling requirements in many applications. Predicting flow boiling patterns could play a key role in the development of new engineering design tools for predicting heat transfer rates and pressure drops. A novel feed-forward neural network architecture was developed to classify flow boiling patterns in the microscale, in which each transition boundary was considered with its own Forward Neural Network within the overall architecture. The network was then compared to new flow boiling pattern data using HFE-7100 for heat fluxes and mass fluxes between 3.2-132.4 kW/m² and 100-1000 kg/m²s, respectively.