Session: K8-03: FUNDAMENTALS OF MULTI-PHYSICS TRANSPORT AND MACHINE LEARNING

Paper Number: 142633

142633 - Development of Novel Machine Learning Approaches to Understanding and Predicting Flow Boiling Thermal Transport 

Abstract: 

Developments in many modern applications are encountering rapid escalation in heat dissipation, coupled with a need to decrease the size of thermal management hardware. These developments have spurred unprecedented interest in replacing single-phase hardware with other more efficient configurations including two-phase flow boiling counterparts. However, a lack of a very clear understanding of the physical mechanisms impacting performance parameters in flow boiling configurations limits availability of thermal design tools and their widespread implementation across applications. Traditional methods including correlation development, analytical modeling, computational fluid dynamics, and direct numerical simulations have lacked the ability to be generalizable and suffer from various disadvantages. New techniques in machine learning modeling have the ability to enhance fundamental understanding of phase-change to improve the prediction capability of performance parameters in these systems. In this presentation, I will discuss various novel machine learning approaches to modeling and predicting performance parameters in flow boiling configurations. First, data sciences-based modeling tools will be developed to enhance predicting capability of performance parameters including heat transfer and pressure drop. Second, flow visualization data on void fraction, vapor-liquid voids, interfacial behaviors, and liquid-solid wall wetting front areas will be analyzed using novel machine learning vision tools to systemically develop improved two-phase theoretical models that compute void fractions, heat transfer coefficients, and critical heat flux with higher accuracy. Finally, we will also show development of physics informed machine learning models that can act as a surrogate for removing empiricism from theoretical formulations for phase-change modeling. This work showcases the potential of using new machine learning-based strategies to accelerate scientific discovery through a consolidated approach to capturing data and analyzing data in flow boiling configurations.

Presenting Author: Chirag Kharangate Case Western Reserve Univerisity

Presenting Author Biography: Chirag Kharangate is leading the Two-Phase Flow and Thermal Management Laboratory in Mechanical and Aerospace Engineering at Case Western Reserve University (CWRU). Dr. Kharangate received his Ph.D. in Mechanical Engineering from Purdue University in 2016 and has multiple years of research and industry experience working on projects dealing with thermal management technologies utilizing single-phase and two-phase flows for automotive, computer, and aerospace applications. As a postdoctoral scholar in the Nanoheat Laboratory at Stanford University, he worked on the design and optimization of two-phase embedded microchannel cooling in Si and SiC substrates. At CWRU, Dr. Kharangate is the recipient of the Case School of Engineering Research Award, ASME K-16 Outstanding Early Faculty Career in Thermal Management Award, and the Office of Naval Research Young Investigator Program Award. He has extensive expertise in testing and modeling flow boiling, flow condensation, and evaporation phase change schemes. He complements his experimental and theoretical work with the development of computational fluid dynamics (CFD) as well as novel machine learning tools for predicting phase change phenomena.

Authors:

Chirag Kharangate Case Western Reserve Univerisity