Professional Documents
Culture Documents
Overview
Goals of the testcase:
Corner flow separation often overpredicted by Eddy Viscosity Models Question:
Can EARSM/RSM predict such flows systematically better than Eddy Viscosity models? Are all EARSM/RSM about equal or are there large differences in behavior? What are the reasons for the differences?
SST
Partners
ANS, NTS, NUM,TUD, UniMan
Flow Description
Diffuser geometries Schematic of Flow Separation Zone
Major flow parameters Incompressible fluid Re = 10,000 (inlet channel height and inlet bulk velocity) Fully developed inflow at diffuser inlet
ATAAC final workshop 2012/6/11-12: ST 04 3D Stanford Diffuser; page 3
Experiment
Flow system schematic 3D Magnetic Resonance Velocimetry (MRV)
Measured Velocity
Available Data
Three velocity components (Diffuser 1 and 2) Fluctuations of streamwise component, Urms (Diffuser 1) Pressure coefficient distribution (Diffuser 1)
ATAAC final workshop 2012/6/11-12: ST 04 3D Stanford Diffuser; page 4
Modeling challenges
Flow in a rectangular duct is not unidirectional
secondary flow (Prandtls secondary flow of second kind) due to anisotropic normal stresses
Secondary motion generates vortices in square ducts which drive momentum into the corner
more momentum in the corner allows the flow to overcome stronger pressure gradients than without such secondary features
RANS
LEVM cannot account for secondary flow properly calibrated RSM should perform consistently better
RANS computations
ANSYS
The S-BSL-EARSM using the WJ stress-strain relation has been optimized and documented (Menter et al, 2009, also report available).
NUMECA
S-BSL-EARSM model from ANSYS High-Re Wallin-Johansson EARSM with k-omega model of Hellsten, 2005 (WJ-EARSM)
UniMan
Elliptic-Blending RSM (EBRSM)
NTS
S-BSL-EARSM model from ANSYS
NUMECA
Diffuser 1: 14591121
UniMan
Diffuser 1: 21260180 Diffuser 2: 2206090
NTS
Diffuser 1: 137 x 77 x 135
NTS RANS mesh for Diffuser 1:
ANSYS, UniMan
Fully developed flow from precursor simulations of a periodic 2D duct using the same model as for the entire diffuser
NUMECA
Developed flow, enabled by the upstream development channel being 100 channel heights long
Inlet section
12 8 5 X/H = 2 H
15
Cp line
Pressure coefficient
In general, all RSM models perform better than LEVM (SST) Among all models, EBRSM model of UniMan is superior to all other models tested Reasons for differences can be seen from the streamwise velocity field (next slide)
Data for Diffuser 1
0.7 0.6 0.5 0.4 0.3
Cp
0.2 0.1 0 -0.1 -0.2 Experiment S-BSL-EARSM ANS S-BSL-EARSM NTS S-BSL-EARSM NUM WJ-EARSM NUM EBRSM UniMan 0 0.5 1 1.5 2
X/L
ATAAC final workshop 2012/6/11-12: ST 04 3D Stanford Diffuser; page 11
TUD
Diffuser 1 only RANS/LES: 22462134 SAS-RSM: 15062134
NTS...
ATAAC final workshop 2012/6/11-12: ST 04 3D Stanford Diffuser; page 18
Recycling (periodic conditions) in an additional upstream rectangular channel section with the length L=6H Computational Grids With synthetic inflow: Domain -3 < x < 55; Grid: 414 x 77 x 135 (~4.3M) With recycling: Domain: - 9 < x < 55; Grid: 499 x 77 x 135 (~ 5.2 M)
Sponge layer Recycling
ATAAC, page 19
ATAAC final workshop 2012/6/11-12: ST 04 3D Stanford Diffuser; page 19
Recycling
ATAAC, page 20
ATAAC final workshop 2012/6/11-12: ST 04 3D Stanford Diffuser; page 20
TUD & UniMAN Hybrid RANS/LES methods predict the pressure coefficient very well SAS-RSM model somewhat underestimates Cp.
Cp
0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 Experiment TUD RANS-LES TUD RSM-SAS UniMan RANS-LES
0.5
1.5
X/L
ATAAC final workshop 2012/6/11-12: ST 04 3D Stanford Diffuser; page 23
UniMan RANS/LES
Diffuser 1 Diffuser 2
Consistently with the adequate prediction of Cp value, the UniMan hybrid LES/RANS method correctly reproduces the flow field pattern in both Diffusers.
ATAAC final workshop 2012/6/11-12: ST 04 3D Stanford Diffuser; page 24
TUD SAS-RSM
SAS-RSM somewhat underpredicts the size of the separation zone, as well as maximal values of streamwise velocities and Urms in the region close to the end of the diffuser A spotty behaviour of Urms is due to a small averaging time (7 through-flow times)
Urms/Ubulk100
ATAAC final workshop 2012/6/11-12: ST 04 3D Stanford Diffuser; page 25
Cp
0.5
1.5
X/L
NTS IDDES U
Experiment
IDDES (Recycling) IDDES (Synth. EARSM) IDDES (Synth. SST)
Same rating of the approaches as that based on Cp distributions: Best predictions: IDDES with recycling and IDDES with inflow synthetic turbulence based on S-BSL-EARSM RANS solution Somewhat worse: IDDES with synthetic turbulence based on SST RANS (u2=v2 =w2)
Same rating of the approaches as that based on Cp distributions (cf. prev. slide)
Cp
0.2 0.1 0 -0.1 -0.2
0.5
1.5
X/L
ANSYS IDDES
Experiment
IDDES (Recycling)
Experiment
IDDES (Recycling)
U/Ubulk
Urms/Ubulk
ANSYS WMLES
Experiment
WMLES (Recycling)
Experiment
WMLES (Recycling)
U/Ubulk
Urms/Ubulk