An intelligent strategy for phase change heat and mass transfer: Application of machine learning

Khodakarami, Siavash

Permalink

https://hdl.handle.net/2142/127310 Copy

Description

Title

An intelligent strategy for phase change heat and mass transfer: Application of machine learning

Author(s)

Khodakarami, Siavash

Issue Date

2024-08-01

Director of Research (if dissertation) or Advisor (if thesis)

Miljkovic, Nenad

Doctoral Committee Chair(s)

Miljkovic, Nenad

Committee Member(s)

• Jacobi, Anthony
• Feng, Jie
• Stillwell, Andrew

Department of Study

Mechanical Sci & Engineering

Discipline

Mechanical Engineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Condensation
• Machine Learning, Infrared
• Thermography
• Generative Adversarial Network

Abstract

Reliable and cost-effective measurement and characterization of phase change processes have always been challenging and expensive. Likewise, due to the complex nature of these processes, the fundamental understanding of processes such as boiling, and condensation remains limited. Therefore, a need exists in the phase change heat and mass transfer research community to develop new techniques which can achieve both more accurate and simpler heat transfer measurements. Furthermore, a need exists to develop a better understanding of the relevant physical mechanisms governing these processes. Conventional methods for measuring and characterizing phase change heat transfer are often complex and lead to high measurement uncertainty, and their use is limited to narrow conditions. However, in the past decade, the field of engineering has seen a surge in the application of machine learning and computer vision techniques in various areas such as material science, biomedical, manufacturing, and autonomous driving. Recently, these techniques have shown promising results in the field of thermofluidic sciences. This study aims to propose new methodologies based on imaging data and deep learning to predict heat transfer related parameters in vapor-to-liquid phase change processes. This study is intended to provide a strong argument for the need for new characterization techniques in phase change processes and why machine learning has the potential to augment or replace other methods. The new condensation characterization frameworks developed in this work allow for a more reliable, easier, cost-effective, and more generalizable condensation heat transfer measurement techniques and also opens new capabilities for studying the fundamentals of condensation processes including droplet shedding dynamics, droplet shedding frequency, droplet residence time, and transient temperature distribution during condensation on hydrophilic, hydrophobic, and superhydrophobic surfaces. This work provides an insight into the application of imaging data including high-speed imaging, regular optical imaging, and infrared imaging data along with deep learning based computer vision models to develop new predictive and generative models in the domain of phase change heat transfer. Specifically, three different characterization frameworks based on deep learning were developed. The first framework predicts condensation heat flux based on optical imaging data at low frequency of 30 frames per second. Application of this model for local heat flux measurement is demonstrated. The second framework consists of a self-supervised model for detection of shedding droplets during condensation without any manual data annotation step based on high-speed imaging data and a supervised data-driven model to predict shedding droplet dynamics based on tube size, droplet contact angle, and condensation heat flux. The third framework includes high-speed infrared thermography of tube surfaces during dropwise condensation to study temperature distribution at high spatial and temporal resolutions. The infrared imaging data are then used to develop a physics-informed generative model based on conditional generative adversarial network to map 2D grayscale images into temperature maps during condensation. The methodologies and results of this work may be used for better understanding of complex phase change processes as well as developing more accurate design criteria for more durable and efficient engineered surfaces for enhanced vapor-to-liquid phase change heat and mass transfer.

Graduation Semester

2024-12

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/127310

Copyright and License Information

Copyright 2024 Siavash Khodakarami

Owning Collections