Velocity estimation in the mixing layer of a subsonic jet using artificial neural networks

Andrew S. Tenney; Mark N. Glauser; Zachary P. Berger

Velocity estimation in the mixing layer of a subsonic jet using artificial neural networks

Andrew S. Tenney, Mark N. Glauser, Zachary P. Berger

Applied Research Laboratory (ARL)

Research output: Contribution to conference › Paper › peer-review

1 Scopus citations

Abstract

Time-resolved Particle Image Velocimetry (PIV, 10 kHz) was used to measure the flow-field of a Mach 0.6 axisymmetric jet. The field was decomposed into its requisite spatial POD eigenfunctions, and the time dependent coefficients were recovered. An Artificial Neural Network (ANN) and a Linear Stochastic Estimation (LSE) model were then trained to estimate the first five time-dependent POD coefficients from five point velocity measurements made by”virtual crosswires” in the mixing layer. We show that the prediction accuracy is strongly dependent on the POD mode number for both models. On average, the ANN-based model is able to predict the velocity fluctuations more accurately than the LSE-based model. Finally, we examine the estimated reduced-order velocity fields and their correlation to analytically reconstructed reduced-order velocity fields. Possible extensions of this method are also discussed.

Original language	English (US)
State	Published - Jan 1 2019
Event	11th International Symposium on Turbulence and Shear Flow Phenomena, TSFP 2019 - Southampton, United Kingdom Duration: Jul 30 2019 → Aug 2 2019

Conference

Conference	11th International Symposium on Turbulence and Shear Flow Phenomena, TSFP 2019
Country/Territory	United Kingdom
City	Southampton
Period	7/30/19 → 8/2/19

All Science Journal Classification (ASJC) codes

Atmospheric Science
Aerospace Engineering

Cite this

@conference{a02efd01ca8740e69fabb8264daa93d2,

title = "Velocity estimation in the mixing layer of a subsonic jet using artificial neural networks",

abstract = "Time-resolved Particle Image Velocimetry (PIV, 10 kHz) was used to measure the flow-field of a Mach 0.6 axisymmetric jet. The field was decomposed into its requisite spatial POD eigenfunctions, and the time dependent coefficients were recovered. An Artificial Neural Network (ANN) and a Linear Stochastic Estimation (LSE) model were then trained to estimate the first five time-dependent POD coefficients from five point velocity measurements made by”virtual crosswires” in the mixing layer. We show that the prediction accuracy is strongly dependent on the POD mode number for both models. On average, the ANN-based model is able to predict the velocity fluctuations more accurately than the LSE-based model. Finally, we examine the estimated reduced-order velocity fields and their correlation to analytically reconstructed reduced-order velocity fields. Possible extensions of this method are also discussed.",

author = "Tenney, {Andrew S.} and Glauser, {Mark N.} and Berger, {Zachary P.}",

year = "2019",

month = jan,

day = "1",

language = "English (US)",

note = "11th International Symposium on Turbulence and Shear Flow Phenomena, TSFP 2019 ; Conference date: 30-07-2019 Through 02-08-2019",

}

TY - CONF

T1 - Velocity estimation in the mixing layer of a subsonic jet using artificial neural networks

AU - Tenney, Andrew S.

AU - Glauser, Mark N.

AU - Berger, Zachary P.

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Time-resolved Particle Image Velocimetry (PIV, 10 kHz) was used to measure the flow-field of a Mach 0.6 axisymmetric jet. The field was decomposed into its requisite spatial POD eigenfunctions, and the time dependent coefficients were recovered. An Artificial Neural Network (ANN) and a Linear Stochastic Estimation (LSE) model were then trained to estimate the first five time-dependent POD coefficients from five point velocity measurements made by”virtual crosswires” in the mixing layer. We show that the prediction accuracy is strongly dependent on the POD mode number for both models. On average, the ANN-based model is able to predict the velocity fluctuations more accurately than the LSE-based model. Finally, we examine the estimated reduced-order velocity fields and their correlation to analytically reconstructed reduced-order velocity fields. Possible extensions of this method are also discussed.

AB - Time-resolved Particle Image Velocimetry (PIV, 10 kHz) was used to measure the flow-field of a Mach 0.6 axisymmetric jet. The field was decomposed into its requisite spatial POD eigenfunctions, and the time dependent coefficients were recovered. An Artificial Neural Network (ANN) and a Linear Stochastic Estimation (LSE) model were then trained to estimate the first five time-dependent POD coefficients from five point velocity measurements made by”virtual crosswires” in the mixing layer. We show that the prediction accuracy is strongly dependent on the POD mode number for both models. On average, the ANN-based model is able to predict the velocity fluctuations more accurately than the LSE-based model. Finally, we examine the estimated reduced-order velocity fields and their correlation to analytically reconstructed reduced-order velocity fields. Possible extensions of this method are also discussed.

UR - http://www.scopus.com/inward/record.url?scp=85073635367&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85073635367&partnerID=8YFLogxK

M3 - Paper

T2 - 11th International Symposium on Turbulence and Shear Flow Phenomena, TSFP 2019

Y2 - 30 July 2019 through 2 August 2019

ER -

Velocity estimation in the mixing layer of a subsonic jet using artificial neural networks

Abstract

Conference

All Science Journal Classification (ASJC) codes

Other files and links

Fingerprint

Cite this