SCIENTIFIC MACHINE LEARNING METHODS FOR REACTIVE-TRANSPORT AND THERMAL-TRANSPORT PROBLEMS

Simple item page
dc.contributor.advisorNakshatrala, Kalyana Babu
dc.contributor.committeeMemberLiu, Dong
dc.contributor.committeeMemberMo, Yi-Lung
dc.contributor.committeeMemberRao, Jagannatha R.
dc.contributor.committeeMemberKulkarni, Yashashree
dc.contributor.committeeMemberMudunuru, Maruti K.
dc.creatorJagtap, Nimish Vijay
dc.date.accessioned2023-05-28T17:31:13Z
dc.date.createdAugust 2022
dc.date.issued2022-08-09
dc.date.updated2023-05-28T17:31:14Z
dc.description.abstractScientific machine learning (SciML) involves development of machine learning models trained using scientific data. SciML involves confluence of machine learning and scientific computing tools and has accelerated research in a gamut of scientific disciplines. This dissertation presents novel SciML frameworks for reactive-transport and thermal-transport problems. The traditional numerical methods use initial and boundary conditions for tran- sient problems and cannot use the limited time history of solutions that might be available. The framework developed for reactive-transport problems overcomes this shortcoming to improve prediction accuracy using available time-history data. The framework uses convolutional neural networks for capturing spatial patterns and long short-term memory networks for forecasting temporal variations in mixing. The framework was carefully built to ensure non-negativity of the chemical species at all space-time points. The framework was validated for 2D problems and is easily extensible to 3D problems. The time needed to obtain a forecast using the model is a fraction of the time needed to obtain the results using a high-fidelity simulation. Therefore, the proposed framework will be a valuable tool for modeling reactive- transport in a wide range of applications. The framework for thermal-transport utilizes physics-informed neural networks (PINNs) to solve forward and inverse problems for active cooling due to fluid circula- tion through the microvasculatures embedded in thin components. Such components are used in emerging technologies like hypersonic aircraft, space exploration vehicles and batteries for efficient thermal regulation. Modeling is vital during the design and operational phases of such systems. However, what is lacking is an accurate frame- work that (i) captures sharp discontinuity in thermal flux across complex vasculature layouts, (ii) accommodates oblique derivatives of the temperature gradients, (iii) han- dles nonlinearity because of radiative heat transfer, (iv) provides a high-speed forecast for real-time monitoring, and (v) solves inverse problems with a noisy data. A fast, reliable, and accurate framework for vascular thermal regulation is presented that is mesh-less and elegantly overcomes all aforementioned challenges. The framework is valuable for real-time monitoring of thermal regulatory systems because of rapid forecasting. It facilitates systematic inverse modeling studies, e.g., for parameter identification, which is the most significant utility of the framework.
dc.description.departmentMechanical Engineering, Department of
dc.format.digitalOriginborn digital
dc.format.mimetypeapplication/pdf
dc.identifier.citationPortions of this document appear in: N. V. Jagtap, M. K. Mudunuru, and K. B. Nakshatrala. A deep learning modeling framework to capture mixing patterns in reactive-transport systems. Communications in Computational Physics, 90:1302–1309, 2022. DOI: 10.1007/ s10891-017-1687-6.
dc.identifier.urihttps://hdl.handle.net/10657/14321
dc.language.isoeng
dc.rightsThe author of this work is the copyright owner. UH Libraries and the Texas Digital Library have their permission to store and provide access to this work. UH Libraries has secured permission to reproduce any and all previously published materials contained in the work. Further transmission, reproduction, or presentation of this work is prohibited except with permission of the author(s).
dc.subjectScientific Machine Learning
dc.subjectDeep learning
dc.subjectReactive-transport
dc.subjectNon-negative solutions
dc.subjectSpatial-temporal forecasting
dc.subjectPattern recognition
dc.subjectConvolutional neural networks (CNN)
dc.subjectLong short-term memory (LSTM) networks
dc.subjectThermal-transport
dc.subjectPhysics-informed neural networks (PINNs)
dc.subjectThermal regulation
dc.subjectMicrovasculatures
dc.subjectActive cooling
dc.subjectVerification
dc.subjectInverse problems
dc.subjectInformed decision making
dc.titleSCIENTIFIC MACHINE LEARNING METHODS FOR REACTIVE-TRANSPORT AND THERMAL-TRANSPORT PROBLEMS
dc.type.dcmiText
dc.type.genreThesis
dcterms.accessRightsThe full text of this item is not available at this time because the student has placed this item under an embargo for a period of time. The Libraries are not authorized to provide a copy of this work during the embargo period.
local.embargo.lift2024-08-01
local.embargo.terms2024-08-01
thesis.degree.collegeCullen College of Engineering
thesis.degree.departmentMechanical Engineering, Department of
thesis.degree.disciplineMechanical Engineering
thesis.degree.grantorUniversity of Houston
thesis.degree.levelDoctoral
thesis.degree.nameDoctor of Philosophy

Files

License bundle

Now showing 1 - 2 of 2
No Thumbnail Available
Name:
PROQUEST_LICENSE.txt
Size:
4.43 KB
Format:
Plain Text
Description:
Download
No Thumbnail Available
Name:
LICENSE.txt
Size:
1.81 KB
Format:
Plain Text
Description:
Download

Collections

Published ETD Collection