Using POD and DMD
for comparing CFD and experimental results
in unsteady aerodynamics
A. Guissart, T. Andrianne,
G. Dimitriadis & V.E. Terrapon
University of Liege
November 25, 2014
Motivation
Comparison of results
Experiments (EFD) Simulations (CFD)
3 Global quantities
7 Local quantities
Motivation
Comparison of results
Experiments (EFD) Simulations (CFD)
3 Global quantities
7 Local quantities
POD and DMD used for comparison
POD finds the most persistent spatial structures φ
POD iu px, y , tq “
Nÿ
i “1a
POD ilo
omo
on
amplitude = energyf
POD iptq
loomoon
time evolutionφ
POD ipx , y q
looooomooooon
spatial modeDMD finds single frequency modes φ
DMD iu px, y , tq “
Nÿ
i “1a
DMD ilo
omo
on
amplitudeexp pλ
DMD itq
looooomooooon
time evolutionφ
DMD ipx , y q
looooomooooon
spatial modePOD and DMD used for comparison
POD finds the most persistent spatial structures φ
POD iu px, y , tq “
Nÿ
i “1a
POD ilo
omo
on
amplitude = energyf
POD iptq
loomoon
time evolutionφ
POD ipx , y q
looooomooooon
spatial modeDMD finds single frequency modes φ
DMD iu px, y , tq “
Nÿ
i “1a
DMD ilo
omo
on
amplitudeexp pλ
DMD itq
looooomooooon
time evolutionφ
DMD ipx , y q
looooomooooon
spatial modeMethodology
Data Collection Decomposition Comparison
Static rectangle
‚
Re “ Uc{ν “ 105‚
c “ 4d‚
α “ 50 Pitching rectangle‚
Re “ 4 ¨ 104‚
k “ πfd{U “ 6.85 ¨ 10´2‚
αMAX “ 100Methodology
Data Collection Decomposition Comparison
Static rectangle
‚
Re “ Uc{ν “ 105‚
c “ 4d‚
α “ 50 Pitching rectangle‚
Re “ 4 ¨ 104‚
k “ πfd{U “ 6.85 ¨ 10´2‚
αMAX “ 100Methodology
Data Collection
1. EFD Ñ Tr-PIV
2. CFD Ñ uRANS
Methodology
Data Collection 1. EFD Ñ Tr-PIV 2. CFD Ñ uRANS Decomposition 1. POD: φipx , y q and ai 2. DMD: φipx , y q, ai and λia
DMD iexp
`λ
DMDit
˘
φ
DMD ipx , y q
Methodology
Data Collection 1. EFD Ñ Tr-PIV 2. CFD Ñ uRANS Decomposition 1. POD: φipx , y q and ai 2. DMD: φipx , y q, ai and λi Comparison 1. φipx , y q 2. ai and λi u px, y , tq “ N ÿ i “1 aDMD i loomoon amplitude exp`λDMD i t ˘ looooomooooon time evolution φDMDi px , y q looooomooooon spatial modeModes shapes comparison MAC´φEFDi , φCFDj ¯“ ¨ ˝ φEFDi ¨ φCFDj }φEFD i }}φCFDj } ˛ ‚ 2
Amplitudes and frequencies comparison ERR´aEFDi , aCFDi ¯“}a
EFD
i } ´ }aCFDi } }aEFD
i }
Modal Assurance Criterion
MAC ´ φEFDi , φCFDj ¯ “ ˜ φEFD i ¨ φCFDj }φEFDi }}φCFDj } ¸2 1 2 3 4 1 2 3 4 CFD modes EFD mo des 0 0.2 0.4 0.6 0.8 1
MAC takes values between 0 and 1
Methodology
Data Collection 1. EFD Ñ Tr-PIV 2. CFD Ñ uRANS Decomposition 1. POD: φipx , y q and ai 2. DMD: φipx , y q, ai and λi Comparison 1. φipx , y q 2. ai and λi u px, y , tq “ N ÿ i “1 aDMD i loomoon amplitude exp`λDMD i t ˘ looooomooooon time evolution φDMDi px , y q looooomooooon spatial modeModes shapes comparison MAC´φEFDi , φCFDj ¯“ ¨ ˝ φEFDi ¨ φCFDj }φEFD i }}φCFDj } ˛ ‚ 2
Amplitudes and frequencies comparison ERR´aEFDi , aCFDi ¯“}a
EFD i } ´ }aCFDi }
}aEFD
i }
Modal Assurance Criterion
MAC ´ φEFDi , φCFDj ¯ “ ˜ φEFD i ¨ φCFDj }φEFDi }}φCFDj } ¸2 1 2 3 4 1 2 3 4 CFD modes EFD mo des 0 0.2 0.4 0.6 0.8 1
MAC takes values between 0 and 1
Oscillating rectangle
‚
EFD: rectangle undergoing LCO‚
CFD: pitching frequency andamplitude imposed (same as EFD)
Problem
‚
No other EFD values than velocities‚
CFD data need to be validatedOscillating rectangle
POD results Energy content 5 10 15 20 10−2 10−1 100 101 102 Mode Con tribution to total energy (%) EFD CFD MAC matrix 1 2 3 4 5 6 1 2 3 4 5 6 0.99 0.86 0.76 0.52 0.19 0.19 CFD modes EFD mo des 0 0.2 0.4 0.6 0.8 1Static rectangle
Global values cl cd St CFD 0.83 0.46 0.136 EFD 0.53 0.45 0.152 Problem‚
cl from CFD far from EFD value‚
Why?Static rectangle
DMD results EFD: 2 modes CFD: 2 modes EFD: 3 modes In CFD reconstruction‚
More important reverse flow‚
Discrepancies on the rear part‚
Different dynamic of reattachmentStatic rectangle
DMD results EFD: 2 modes CFD: 2 modes EFD: 3 modes In CFD reconstruction‚
More important reverse flow‚
Discrepancies on the rear part‚
Different dynamic of reattachmentStatic rectangle
DMD results EFD: 2 modes CFD: 2 modes EFD: 3 modes In CFD reconstruction‚
More important reverse flow‚
Discrepancies on the rear partStatic rectangle
DMD results EFD: 2 modes CFD: 2 modes EFD: 3 modes In CFD reconstruction‚
More important reverse flow‚
Discrepancies on the rear part‚
Different dynamic of reattachmentConclusion and future work
DMD and POD are useful
‚
Compare and validate CFD results‚
Highlight and understand potentialdiscrepancies