Scientific Machine Learning Methods for Reactive-Transport and Thermal-Transport Problems - Dissertation

Aug 5, 2026

Upload by Aria

Scientific machine learning (SciML) develops machine learning models trained on scientific data, combining learning algorithms with scientific computing to accelerate progress across scientific disciplines. This dissertation introduces novel SciML frameworks targeting reactive-transport and thermal-transport problems and emphasizes improved prediction from available time histories. The reactive-transport framework integrates convolutional neural networks and long short-term memory networks while enforcing non-negativity and validating extension from 2D to 3D, reducing forecast cost. The thermal-transport framework applies physics-informed neural networks to address forward and inverse active cooling with microvasculature-dependent modeling.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages119

Reading Time5 h 0 min

AccessFree to download

Acknowledgments

Abstract

Reactive-transport framework

Thermal-transport framework

Problem motivation and application context

SCIENTIFIC MACHINE LEARNING METHODS FOR REACTIVE-TRANSPORT AND THERMAL-TRANSPORT PROBLEMS by Nimish Vijay Jagtap A dissertation submitted to the Department of Mechanical Engineering, University of Houston in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Mechanical Engineering Committee Chair: Dr. Dong Liu Committee Co-Chair: Dr. Kalyana Babu Nakshatrala Committee Member: Dr. Yi-Lung Mo Committee Member: Dr. Jagannatha Rao Committee Member: Dr. Yashashree Kulkarni Committee Member: Dr. Maruti Kumar Mudunuru University of Houston Copyright 2022, Nimish Vijay Jagtap Acknowledgments I would like to thank my thesis advisor Dr. Kalyana Babu Nakshatrala for allowing me to work in his research group and his guidance during the course of research work. He gave me few options of different research topics and suggested that I pursue my research in Machine Learning related topic. Looking back, I am happy that he offered a topic related to Machine Learning because it is being used in more and more engineering applications. I would also like to thank Dr. Maruti Kumar Mudunuru from Pacific Northwest National Laboratory for his constant guidance during my research work. Dr. Mudunuru was practically my co-advisor. We had frequent interactions during the course of the reactive-transport work, which caused that work to progress at a very fast pace. Completion of the reactive-transport work in a timely manner helped maintain my motivation in the otherwise gloomy Covid- 19 pandemic period. I am also thankful to Dr. Mudunuru for facilitating access to ALCF super-computing facility. Special thanks to Dr. Dong Liu who agreed to serve as my advisor from the ME department. Dr. Liu made sure I meet all the ME department requirement necessary for PhD. I am also thankful to the committee members Dr. Jagannatha Rao, Dr. Yashashree Kulkarni and Dr. Yi-Lung Mo for generously agreeing to serve on my dissertation committee. Majority of the programs for physics-informed neural network were run on the the Sabine cluster of Hewlett Packard Enterprise Data Science Institute (HPE DSI) at University of Houston, and I am grateful to HPE DSI for providing the computational resources and timely support. I would also like to thank College of Engineering and Department of Mechanical Engineering for employing great faculty members. The courses I took were challenging and provided depth of knowledge in the respective areas. I worked on my PhD while working full-time and therefore completing the coursework was a challenge because six out of eight courses that I took were in-person and required travelling to the University. I am thankful to my employer Kalsi Engineering, Inc. for their flexible work-hour policy due to which I could attend the classroom courses offered in the late-afternoons and evenings. Since I was not a full-time student, I did not have as much technical conversations and interaction with the peer students as a common full-time student would. However, I would like to thank Rajgopal and Tejasree with whom I interacted frequently during the course of my PhD. Last but the most important, I would like to express my deepest gratitude tomy entire family. Every member of my family contributed for my PhD. Without the support of my wife, this part-time endeavor would not have been possible. I had to carve out the time from my personal space and family responsibilities for completing the PhD coursework and research. My wife took extra responsibility in the family to make up for the deficit due to my focus on PhD. My parents also shared some of responsibility to look after my son for over a year. Similarly, my mother-in-law looked after our newborn for few months. My elder brother was a great mental support during the Covid-19 pandemic period. My younger brother and his wife, who already have a doctorate in engineering, helped me providing feedback during the course of research. Abstract

FAQs

What is scientific machine learning (SciML) in this dissertation?

SciML develops machine learning models trained using scientific data by combining machine learning with scientific computing tools to speed research across disciplines.

How does the reactive-transport framework improve transient prediction?

It leverages available time-history data instead of relying only on initial and boundary conditions, using convolutional neural networks for spatial patterns and long short-term memory networks for temporal forecasting.

What approach is used for the thermal-transport problem, and what tasks does it support?

It uses physics-informed neural networks (PINNs) to solve forward and inverse problems for active cooling driven by fluid circulation through embedded microvasculatures.

Aug 5, 2026

Upload by Aria