Session: K8-01: FUNDAMENTALS OF MACHINE LEARNING IN HEAT TRANSFER

Paper Number: 131476

131476 - Machine Learning Algorithms for Predicting Condensation Pressure Drop in Mini/Micro Channels 

Abstract: 

Recent decades have witnessed a substantial rise in heat dissipation demands within compact systems such as mini/microchannels, rendering single-phase heat transfer inadequate. Consequently, two-phase heat transfer, with its superior dissipation rates due to latent heat during phase changes between liquid and vapor, has become essential. Yet, two-phase systems incur a higher pressure drop than single-phase systems, adversely affecting overall efficiency. Accurate prediction of this pressure drop is thus critical for system efficiency and reliability. Existing models for two-phase flow pressure drop prediction fall short in accuracy and applicability. Our research addresses these shortcomings through machine learning techniques, employing Generalized Additive Models (GAM), Random Forest algorithms, and Artificial Neural Networks (ANN). We amassed a dataset of over 7000 data points, featuring 50 distinct input features reflecting a wide spectrum of thermophysical properties and flow dynamics, aggregated from 45 contemporary studies. For feature selection, we utilized Pearson, Spearman, and Kendall correlation methods alongside multivariate techniques such as Lasso, Ridge Regression, and Recursive Feature Elimination. Our results indicate that non-linear machine learning techniques surpass traditional linear models in efficacy. Notably, ANNs demonstrate exceptional predictive capabilities, especially in discerning vapor quality trends, unveiling complex patterns not fully captured by other models like Random Forest and GAM. Our study confirms the significant advantages of non-linear modeling over traditional linear approaches in thermal system analysis. By accurately predicting two-phase pressure drops, our models contribute to the development of more efficient and reliable thermal management solutions, meeting the pressing demands of modern industrial systems.

Presenting Author: Farshad Barghi Golezani Case Western Reserve University

Presenting Author Biography: He is a PhD Student in Case Western Reserve University.

Authors:

Farshad Barghi Golezani Case Western Reserve University
Jiayuan Li Case Western resereve University
Logan Pirnstill Case Western Reserve University
Chirag Kharangate Case Western Reserve University