Abstract
Recent years have seen a big increase in the amount of heat that needs to be dissipated within small systems. This means that single-phase heat transfer is no longer good enough. Because of this, two-phase heat transfer, which can get rid of more heat since liquid changing to vapor takes up latent heat, has become crucial. However, two-phase systems have higher pressure drops than single-phase ones, which decreases overall efficiency. So being able to accurately predict the pressure drop is really important for efficiency and reliability. Currently available models for two-phase flow pressure drop prediction are not very accurate or helpful. Our research uses machine learning techniques, including Polynomial Linear regression, Random Forest algorithms, and Neural Networks, to tackle this problem. We looked at a dataset of 4337 data points with 45 features from 25 studies and used feature selection methods like Kendall correlation and techniques like Lasso and RFE to select features. Our findings show nonlinear machine learning ways, especially Neural Networks, are better at predicting pressure drops than traditional models. The Neural Network model performed great, with a Mean Absolute Percentage Error of 13.89% and an adjusted R2 of 0.9941. This shows state-of-the-art neural nets can be highly effective in thermal system analysis.
