ST04: Stanford 3D Diffuser - University of Manchestercfd.mace.manchester.ac.uk/twiki/pub/ATAAC/TestCase004Diffuser3D/... · • Corner flow separation often over-predictedby ... –

ST04: Stanford 3D DiffuserATAAC final workshop 2012-6-11/12 in GöttingenJochen Schütze, Pavel Smirnov, Florian Menter

ANSYS Germany GmbH

Motivation, Goal of Testcase

• Corner flow separationoften over-predicted byLinear Eddy Viscosity Models

• EARSM / DRSM systematicallybetter than LEVMs?

ATAAC final workshop 2012-6-11/12: ST 04 3D Stanford Diffuser; page 2

better than LEVMs?

• Significant differences betweendifferent EARSM / DRSM?– Reasons for such differences?

• Partners– ANS, NTS, NUM, TUD, UniMAN

SST

Asymmetric rectangular Diffuser

Separation Zone* View

Fully developedrect. duct flow


• Incompressible fluid (water)

• Re = 10’000– inlet channel height & bulk velocity

• Fully developed inflow*actually in the opposite corner (diverging walls)

View

Modeling challenges

• Anisotropic normal stresses � secondary flow– (Prandtl’s secondary flow of second kind)

• Generates vortices in square ducts, whichdrive momentum into the corner

� Flow can overcome stronger adverse pressure


� Flow can overcome stronger adverse pressure gradients without separating from the wall

• RANS:– LEVM cannot account for secondary flow– properly calibrated RSM should perform

consistently better• Turbulence resolving methods:

– must capture anisotropic turbulence

Experiment: 3D NMR velocimetry

3D Magnetic Resonance Velocimetry (MRV)


• Three velocity components: U, V, W (Diff. 1 and 2)• Fluctuations of streamwise component Urms (Diff. 1)• Pressure coefficient (Cp) along bottom line (Diff. 1)

Measured Velocity

Locations for cross-comparison:

Planes for cross-comparison of streamwise velocity:• <U>

Line for Cp cross-comparison:


58

1215

X/H = 2H

Cp line

• <U>• Urms

X/H = 0

RANS computations• ANSYS

– S-BSL-EARSM using the Wallin-Johansson stress-strain relation, optimized and documented by ANSYS (Menter et al, 2009).

• NTS– S-BSL-EARSM model from ANSYS


– S-BSL-EARSM model from ANSYS

• 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)

RANS Computational Grids

Medium mesh for Diffuser 1: used by ANSYS and NUMECA

• ANSYS– Diffuser 1 and 2: 145×91×121

• NUMECA– Diffuser 1: 145×91×121


NTS RANS mesh for Diffuser 1:

• UniMAN– Diffuser 1: 212×60×180– Diffuser 2: 220×60×90

• NTS– Diffuser 1: 137 x 77 x 135

Inflow conditions for RANScomputations

• Experiment– Fully (?) developed flow: development

channel, 62.9 channel heights long

• ANSYS, UniMAN– Fully developed: precursor simulation:


Inlet section

– Fully developed: precursor simulation:periodic “2D” duct using the sameturbulence model as for diffuser

• NUMECA– Developed flow: upstream development

channel, 100 channel heights long

Footnote: Assume experimental inflow to be fully developped?

Periodic flow � inlet: || block profiles � inlet:


FVM Numerics for RANS• ANSYS

– Momentum eqs: bounded second order upwind scheme– Turbulence eqs: first order upwind

• NUMECA– Momentum eqs: Jameson central scheme with

scalar dissipation


scalar dissipation– Turbulence eqs: first order upwind

• UniMAN– Momentum eqs: second order centered scheme– Turbulence eqs: second order centered scheme

• NTS– Momentum eqs: fourth (adv .) / second (diff .) order

centered scheme

Locations for cross-comparison:

Planes for cross-comparison of streamwise velocity:• <U>

Line for Cp cross-comparison:


58

1215

X/H = 2H

Cp line

• <U>• Urms

X/H = 0

Cp

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Experiment

LEVM: SST results

• Cp curve:Very earlyseparation , reciculation, pressure loss

Experiment


X/L0 0.5 1 1.5 2

-0.2

-0.1

0

ExperimentANSYS SST

pressure loss

• Cross sections :Largelyoverpredictedseparation fromcorner and sidewall

0.3

0.4

0.5

0.6

0.7

RANS: Pressure coefficient

• All Re-Stress models better than LEVM

• EBRSM (UniMAN) superior to the EARSM’s tested


X/L

Cp

0 0.5 1 1.5 2-0.2

-0.1

0

0.1

0.2

0.3

ExperimentS-BSL-EARSM ANSS-BSL-EARSM NTSS-BSL-EARSM NUMWJ-EARSM NUMEBRSM UniMan

EARSM’s tested

Reasons for differencesto be seen in thestreamwise velocity field…

(Data for Diffuser 1)X/L

Cp

0 0.5 1 1.5 2-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

ExperimentANSYS SST

RANS: Streamwise velocity

• S-BSL-EARSM (NUMECA & ANSYS):consistently has…

– too much uniform& too strong earlyflow profile

– too strong early

ExperimentWJ-EARSM

NUM

S-BSL-EARSM

NUM

S-BSL-EARSM

ANS

EBRSM

UniMAN


– too strong early recirculation (in top-right corner)

• EBRSM (UniMAN):– more realistic

early flow profile– too late / little

separation & recirculation

RANS: Velocity fluctuations

• EARSMs:– too low fluct .

in core flow, – too high in

earlyseparation

ExperimentWJ-EARSM

NUM

S-BSL-EARSM

NUM

S-BSL-EARSM

ANS

EBRSM

UniMAN


separationzone

• EBRSM:– vice versa

Urms / Ubulk × 100

RANS: Velocities at diagonalsX

= -

3 H

u

0 0.5 1 1.5 2 2.5 3 3.50

0.2

0.4

0.6

0.8

1

1.2

1.4

S-BSL EARSM ANSS-BSL-EARSM NTSS-BSL-EARSM NUMWJ-EARSM NUM

v

0 0.5 1 1.5 2 2.5 3 3.5-0.006

-0.004

-0.002

0

0.002

0.004

0.006

w

0 0.5 1 1.5 2 2.5 3 3.5-0.015

-0.01

-0.005

0

0.005

0.01

0.015


X =

0 H

z0 0.5 1 1.5 2 2.5 3 3.50

z0 0.5 1 1.5 2 2.5 3 3.5-0.006

z0 0.5 1 1.5 2 2.5 3 3.5-0.015

z

u

0 0.5 1 1.5 2 2.5 3 3.50

0.2

0.4

0.6

0.8

1

1.2

1.4

z

v

0 0.5 1 1.5 2 2.5 3 3.5-0.02

-0.01

0

0.01

0.02

0.03

0.04

0.05

0.06

z

w

0 0.5 1 1.5 2 2.5 3 3.5-0.1

-0.08

-0.06

-0.04

-0.02

0

0.02

0.04

RANS: Streamwise velocity, Diffuser 2

• Conclusions similar to thosefor Diffuser 1

• Both…

ExperimentS-BSL-EARSM

ANS

EBRSM

UniMAN


• Both…S-BSL-EARSMandEBRSMcapture the velocity field well

RANS results: Conclusions

• Accounting for stress anisotropy � drastic improvement of the results for this case.– Flow topology matches much better the Experiment

• Not all details of velocity profiles matched by any method– Wall pressure (Cp) distribution improves significantly


– Wall pressure (Cp) distribution improves significantly• Cp-distribution best captured by EBRSM (UniMAN).

• Wall pressure (Cp) distribution alone isnot sufficient to assess the models’ capabilityto correctly predict corner flow separation …– all curves manually gathered in first exp. data point...

• Importance of correct inflow conditions

Turbulence -resolving computations: TUD, UniMAN

• UniMAN : RANS / LES– Two-Velocity hybrid RANS / LES scheme with

underlying v2f (RANS) turbulence model– Inflow : fluctuating flow from Synthetic Eddy Method

• (Jarrin et al.; based on precursor EBRSM calculation of a


• (Jarrin et al.; based on precursor EBRSM calculation of a periodic “2D” duct with inlet dimensions)

• TUD: RANS / LES– another zonal , two-layer hybrid approach

• RANS model for near-wall and LES in the remainder– Inflow : unsteady prec. sim. of fully developed duct flow

• TUD: SAS-RSM– Inflow: Vorton Method , at x / H = -0.6 to avoid decay...

Turbulence-resolving computations (II): NTS, ANSYS

• NTS: IDDES-SST– “NTS synthetic turbulence ” (Adamian & Travin, 2011)

…based on RANS solutions:• SST• WJ-BSL-EARSM


– “Recycling ” (concurrent unsteady periodic ductflow simulation (L = 6H)) [like TUD]

• ANSYS: IDDES-SST– “Recycling” (as for NTS above)

• ANSYS: (algebraic) WMLES– “Recycling” (as for NTS above)

Grids for TUD and UniMANturbulence-resolving simulations

• UniMAN– Diffuser 1: 212×60×180

• TUD– RANS/LES: 224×62×134– SAS-RSM: 150×62×134


Sponge layer

Recycling

– SAS-RSM: 150×62×134

• NTS– 414×77×135 -- + 85×77×135 “Recycling”

• ANSYS– 450×77×135

Numerics forturbulence-resolving simulations• UniMAN

– Code Saturne (unstructured collocated FVM code)– SIMPLEC pressure correction algorithm– Mom. & Turb. eqs: second order centered scheme

• TUD


• TUD– FASTEST (block-structured FVM) / OpenFOAM (SAS)– SIMPLEC / SIMPLE with geometric multi-grid scheme– Momentum eqs: second-order, central differencing /

blended first- / second-order scheme (SAS)– Turbulence eqs: ‘‘flux blending ” technique � limited

upwinding

Numerics (II) forturbulence-resolving simulations• NTS

– Incompressible NTS code (Rogers & Kwak scheme)– 4th order centered approximation of inviscid fluxes– 2nd order centered approximation for viscous fluxes– Implicit, 2nd order (three-layer) time-integration


– Implicit, 2nd order (three-layer) time-integration

• ANSYS– FLUENT, unstructured collocated FVM code – SIMPLEC pressure correction algorithm – Momentum eqs: second order centered scheme– Turbulence eqs: second order upwind– Implicit, 2nd order (three-layer) time-integration

Turb.-resolv.: Pressure profile: TUD, UniMAN

• Both hybrid RANS/LESmethods predict the pressure coefficient

Data for Diffuser 1

0.4

0.5

0.6

0.7


coefficientvery well

• SAS-RSMunderestimatesCp from the beginning...

X/L

Cp

0 0.5 1 1.5 2-0.2

-0.1

0

0.1

0.2

0.3

0.4

ExperimentTUD RANS-LESTUD RSM-SASUniMan RANS-LES

UniMAN LES/RANS,TUD SAS-RSM

• LES/RANS :–too much early

recirculationw/ high Urms

X/L

Cp

0 0.5 1 1.5 2-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

ExperimentTUD RANS-LESTUD RSM-SASUniMan RANS-LES


rms

• SAS-RSM:– too little

separation!–too much

Urms alongall top wallearly on

Turb.-resolv.: Pressure profile: NTS, ANSYS IDDES, WMLES

• “Recycling ” & syntheticturbulence based onS-BSL-EARSM RANSsolution: “perfect”

Cp

0.1

0.2

0.3

0.4

0.5

0.6

0.7

ExperimentIDDES, recycling

NTS


• Synthetic turbulencebased on SST RANS(isotropic) solution:significantly affected …

• Alg. WMLES verysimilar to IDDES (its origin…)

X/L0 0.5 1 1.5 2

-0.2

-0.1

0

IDDES, recyclingIDDES, synth. turb., SSTIDDES, synth. turb., EARSM

X/L

Cp

0 0.5 1 1.5 2-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

ExperimentIDDESWMLES

ANSYS

NTS IDDES – U, Urms• Same “rating” as Cp

– Synth. from SST poorer• Recycling : too

weak fluctuations (no harm)IDDES

(Synth. EARSM)IDDES

(Recycling)IDDES

(Synth. SST)Experimen tIDDES

(Synth. EARSM)IDDES

(Recycling)IDDES

(Synth. SST)Experimen t

X /L

Cp

0 0 .5 1 1 .5 2- 0 .2

- 0 .1

0

0 .1

0 .2

0 .3

0 .4

0 .5

0 .6

0 .7

E x p e r im e n tID D E S , r e c y c l i n gID D E S , s y n t h . t u r b ., S S TID D E S , s y n t h . t u r b ., E A R S M


Urms / Ubulk × 100

Turb.-resolv.: Pressure profile: ANSYS IDDES & alg. WMLES

• Very similar results

• IDDES and (algebraic)WMLES with recyclingoverestimate Cp C

p

0.2

0.3

0.4

0.5

0.6

0.7


overestimate Cpdownstream of X/L = 0.5

• Grid sensitivity has tobe checked– Same mesh as used by NTS, but NTS code has

higher-order discretisation of advective fluxes

X/L0 0.5 1 1.5 2

-0.2

-0.1

0

0.1

0.2

ExperimentIDDESWMLES

NTS, ANSYS: IDDES,ANSYS: alg. WMLES

IDDES(Recycling)

IDDES(Recycling)

WMLES(Recycling)

WMLES (Recycling)

NTS – ANSYS . NTS – ANSYS .IDDES

(Recycling)Experimen tIDDES

(Recycling)Experimen t

X /L

Cp

0 0 .5 1 1 .5 2-0 .2

-0 .1

0

0 .1

0 .2

0 .3

0 .4

0 .5

0 .6

0 .7

E x p e r im e n tID D E SW M L E S


Urms / Ubulk × 100

Velocities at diagonals, NTS IDDES (??)

X =

-3

H

u

0

0.2

0.4

0.6

0.8

1

1.2

1.4 RecyclingSynth. turb, EARSM

v

-0.006

-0.004

-0.002

0

0.002

0.004

0.006

w

-0.015

-0.01

-0.005

0

0.005

0.01

0.015


X =

0 H

z0 0.5 1 1.5 2 2.5 3 3.50

z0 0.5 1 1.5 2 2.5 3 3.5-0.006

z0 0.5 1 1.5 2 2.5 3 3.5-0.015

z

u

0 0.5 1 1.5 2 2.5 3 3.50

0.2

0.4

0.6

0.8

1

1.2

1.4

z

v

0 0.5 1 1.5 2 2.5 3 3.5-0.02

-0.01

0

0.01

0.02

0.03

0.04

0.05

0.06

z

w

0 0.5 1 1.5 2 2.5 3 3.5-0.1

-0.08

-0.06

-0.04

-0.02

0

0.02

0.04

Transient Calculations: Conclusions

• Zonal hybrid RANS (near-wall) / LES (bulk): generally successful– SAS-RSM implementation of TUD: weaker…

• issues with tendency to fall back into RANS behaviour

• Inlet fluctuations: need high -quality synthetics or


• Inlet fluctuations: need high -quality synthetics or (“concurrent precursor”) “recycling” periodic calc.

• IDDES differences NTS—ANSYS: numerics…?!

• Limited over- / underestimation of fluctuations (Urms) doesn’t seem to do any harm

ST04: Stanford 3D DiffuserATAAC final workshop 2012-6-11/12 in GöttingenJochen Schütze, Pavel Smirnov, Florian Menter

ANSYS Germany GmbH

Documents

ST04: Stanford 3D Diffuser - University of Manchestercfd.mace.manchester.ac.uk/twiki/pub/ATAAC/TestCase004Diffuser3D/... · • Corner flow separation often over-predictedby ... –