Back to search

University of Waterloo

Unsupervised learning for coherent structure identification in turbulent channel flow

Abstract

dc:description.abstract

Coherent structures (CS), i.e., regions of flow exhibiting significant spatio-temporal coherence, have long been observed in turbulent fluid flow. These CS offer an opportunity to gain insights on fluid behaviour by bypassing the non-linear complexities associated with turbulent flows. Historically, the identification of CS in turbulent flows has involved using manual thresholds to label regions of interest. More recently, work towards more objective threshold selection have used percolation analysis; yet, particular situations can leave the method vulnerable to human bias. This work takes further steps towards pruning human subjectivity from the CS detection process, where an unsupervised learning framework that uses a clustered self-organizing map is used to automatically organize salient regions of flow within a turbulent channel into distinct clusters. The CS identified and analyzed throughout the study include quasi-streamwise coherent vortices, ejections, and sweeps. Structures pertaining to the near-wall region (𝑦+⪅60), inner region (𝑦+⪅100), and entire wall-normal domain are investigated. Structures are found to agree qualitatively with dynamic expectations, i.e., near-wall vortex structures are quasi-streamwise, and ejection and sweep regions flank vortices. Quadrant distributions of the ejection and sweep structures show larger sweep strength in the lower buffer region (𝑦+⪅15) and larger ejection strength above the buffer region (𝑦+⪆15), both characterized by large fluctuating streamwise velocity, whereas streamwise and wall-normal fluctuations in ejections and sweeps that populate the outer layer are more balanced; vorticity component distributions within vortices indicate counter streamwise rotating vortices in the buffer region; orientation statistics of vortices show preference for streamwise orientation in the near-wall, transverse orientation in the log-layer, and no preferred orientation in the outer layer; and the distribution of vorticity transport components, i.e., stretching and tilting, within vortex clusters demonstrate dominant streamwise vortex stretching within buffer layer vortices. Evidence is found of outer layer structures that resemble outsized counterparts of the ejection--vortex--sweep structures found in the near-wall, reinforcing the notion that a hiearchichal self-sustaining process exists in channel flow turbulence.

Degree

thesis:*
Grantor dc:publisher
University of Waterloo
Year dc:date.issued
2023

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Lyne, John
Advisor dc:contributor.advisor
  • Scott, Katherine Andrea

Subjects

dc:subject × 4

Rights

Language dc:language.iso
en

Identifiers

dc:identifier.*
Handle dc:identifier.uri
http://hdl.handle.net/10012/19032

Chain of custody

source
Harvested from
University of Waterloo
Base URL
dspacemainprd01.lib.uwaterloo.ca/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Lyne, John. Unsupervised learning for coherent structure identification in turbulent channel flow. University of Waterloo, 2023. http://hdl.handle.net/10012/19032