Session: 10-01 Interactive Presentations

Paper Number: 97377

97377 - Multimodal Prediction for Flow Boiling Heat Transfer 

 In recent decades, the miniaturization of electronic have promoted a rapid increase in heat fluxes within a small area. As one of the effective cooling technologies, flow boiling i­­n microchannel heat sinks is widely adopted in the electronics owing to high heat dissipation capability. The advantages of flow boiling are derived from reliance on both latent and sensible heat content of a working fluid and can bring several-fold enhancement in heat removal performance compared to single-phase cooling. Due to the characteristics of flow boiling, in which the heat transfer depends on flow regime, it is difficult to accurately predict the heat transfer coefficient, even though there have been various experimental studies available. In this study, based on flow boiling experiment in a microchannel, a model which predicts thermal performance according to flow regime is developed with an aid of machine learning. Multimodal learning combining CNN and ANN was performed, which was used to analyze the characteristics of flow image and to determine the impact of experimental conditions. The experimental data was divided into training, verification, and test sections, and the best model was selected through hyperparameter optimization. Through this model, we can predict the thermal performance of flow boiling. In particular, the local heat transfer performance related to bubble behavior can be characterized, solving the problem of difficulty in predicting heat transfer performance in two-phase flow.

Presenting Author: Haeun Lee Chung Ang University