Fluid-Structure Interaction of Thin Structuresin Turbulent Flows
G. De Nayer, A. Kalmbach and M. Breuer
Department of Fluid Mechanics (PfS)
Helmut–Schmidt–University (HSU), Hamburg, Germany
M. Munsch S. Sicklinger, R. Wuchner, K.U. BletzingerInstitute of Fluid Mechanics Chair of Structural Analysis
University of Erlangen–Nurnberg, Germany Technical University of Munich, Germany
SuperMUC Review WorkshopGarching, July 2014 SfP
1 Motivation / Objectives
2 Computational Methodology
3 ValidationDefinition of the Test CasesSimulations and Comparison with Experiments (FSI-PfS-2a)
4 Conclusions
Outline 2
1 Motivation / Objectives
2 Computational Methodology
3 ValidationDefinition of the Test CasesSimulations and Comparison with Experiments (FSI-PfS-2a)
4 Conclusions
Outline 3
Tents / Sun Shades /Mobile Umbrellas
Motivation / Long-term Objectives 4
1 Motivation / Objectives
2 Computational Methodology
3 ValidationDefinition of the Test CasesSimulations and Comparison with Experiments (FSI-PfS-2a)
4 Conclusions
Outline 5
FSI in Turbulent Flows
CFD CSD
p, FτF
PartitionedApproach
∆ x
Publication
Breuer, M., De Nayer, G., Munsch, M., Gallinger, T., Wuchner, R.:
Fluid-Structure Interaction Using a Partitioned Semi-Implicit Predictor-CorrectorCoupling Scheme for the Application of Large-Eddy Simulation.
Journal of Fluids and Structures 29, 107-130, 2012.
Computational Methodology:FSI with LES and Thin Structures
6
Momentum Eq.
Runge−Kutta
Poisson Eq.
u*
u , p
Pre
dic
tor
Co
rre
cto
r
Block−struct. grids
i
i
FASTEST−3D
FVM (2 order)
Predict.−Correct.
LES
ALE
nd
Computational Methodology:FSI with LES and Thin Structures
7
Forceson the
Estimationof structural
displacements
Momentum Eq.Runge−Kutta
Poisson Eq.
structure
Grid
u*
u , p
Pre
dict
orC
orre
ctor
Predict.−Correct.
nd
i
i
adaptation
FASTEST−3DFVM (2 order)
LES
ALE
Block−struct. grids
Computational Methodology:FSI with LES and Thin Structures
7
Forces
on the
Estimationof structural
displacements
Momentum Eq.
Runge−Kutta
Poisson Eq.
structure
step
Grid
Next time
u*
u , p
Pre
dic
tor
Co
rre
cto
r
Block−struct. grids
Predict.−Correct.
nd
i
i
adaptation
yes
no
FASTEST−3D
FVM (2 order)
Carat++
FEM
Generalized
α −method
Membranes
Shells
Index: n
LES
ALE
Partitioned
Approach
Grid
adaptation
Computational Methodology:FSI with LES and Thin Structures
7
Forces
on the
Estimationof structural
displacements
Momentum Eq.
Runge−Kutta
Poisson Eq.
structure
step
Grid
Next time
?
u*
u , p
Pre
dic
tor
Co
rre
cto
r
Block−struct. grids
Predict.−Correct.
nd
i
i
adaptation
FSI convergence
yes
no
FSI−Subiteration Loop
(conservative)
FASTEST−3D
FVM (2 order)
CoMA MPI
Interpolation
Grid−to−GridFluid Force
Grid−to−Grid
(ensure dynamic equilibrium)
Displacement
Interpolation(bilinear)
DisplacementsUnderrelax. of
Carat++
FEM
Generalized
α −method
Membranes
Shells
Index: n
Index: k
LES
ALE
Partitioned
Approach
Grid
adaptation
Computational Methodology:FSI with LES and Thin Structures
7
1 Motivation / Objectives
2 Computational Methodology
3 ValidationDefinition of the Test CasesSimulations and Comparison with Experiments (FSI-PfS-2a)
4 Conclusions
Outline 8
1 Motivation / Objectives
2 Computational Methodology
3 ValidationDefinition of the Test CasesSimulations and Comparison with Experiments (FSI-PfS-2a)
4 Conclusions
Outline 9
FSI-PfS-1a
EPDM−rubber
fixed cylinder
60 m
m
2 mm
22 mm
(flexible)
(rigid)
E = 16 MPa= 0.48
u∞Ø
ν
– Re = 30,470– quasi 2D-deformations– small deformations– first swiveling mode
FSI-PfS-2a
2 mm
fixed cylinder
10 m
m
para−rubber
50 m
m
22 mm
steel (rigid)
(flexible)
(rigid)
E = 4.10 MPa
= 0.48
u∞Ø
ν
– Re = 30,470– 2D-deformations– large deformations– second swiveling mode
FSI Test Cases for Turbulent Flows 10
Publications for FSI-PfS-1a
De Nayer, G., Kalmbach, A. Breuer, M., Sicklinger, S. and Wuchner, R.:
Flow past a cylinder with a flexible splitter plate: a complementaryexperimental-numerical investigation and a new FSI test case (FSI-PfS-1a).
Int. Journal of Computers and Fluids 99, 18-43, 2014.
http://qnet-ercoftac.cfms.org.uk/w/index.php/UFR_2-13
Publications for FSI-PfS-2a
Kalmbach, A. and Breuer, M.:
Experimental PIV/V3V Measurements of Vortex-Induced Fluid-StructureInteraction in Turbulent Flow New Benchmark FSI-PfS-2a.
Journal of Fluids and Structures 42, 369-387, 2013.
De Nayer, G. and Breuer, M.:
Numerical FSI Investigation based on LES: Flow past a cylinder with a flexiblesplitter plate involving large deformations (FSI-PfS-2a).
Submitted.
http://qnet-ercoftac.cfms.org.uk/w/index.php/UFR_2-14
FSI Test Cases for Turbulent Flows 11
1 Motivation / Objectives
2 Computational Methodology
3 ValidationDefinition of the Test CasesSimulations and Comparison with Experiments (FSI-PfS-2a)
4 Conclusions
Outline 12
Axial pump with motor
Flow straightener
Test section Water Channel
+Flow measurement methods (PIV / V3V)
Laser displacement sensor
⇓Experimental Data
High-speed video
(real frequency 11.25 Hz)
FSI-PfS-2a: Experiments 13
g
inflow
subset of the channel
Fluid/CFD:
• wall–resolved LES
• 13.5 million CVs
• 72 CVs in spanwise direction
• periodic boundary conditions
Structure/CSD:
• 7–parameter shell elements
• 30 × 10 quadrilateralfour-node elements
• zero z–deformation vs.periodic b.c.
• (Rayleigh damping)
FSI-PfS-2a: Computational Setup 14
93 cores needed for each simulation13.5 million CVs on 91 blocks → 91 processes for CFD
1 process for CSD
1 process for coupling program
2 seconds physical time computed for each simulation
CPU: 1000 hours wall-clock
RAM: 242 Mbytes per core → 22 Gbytes for the entire simulation
Sensitivity study on FSI-PfS-2a
about 30 simulations with different parameters conducted
FSI-PfS-2a: Computations on SuperMUC 15
t u / D
Uy
/D
800 850 9000.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
OO
EXP: Raw Signal
t u / D
Uy
/D
20 40 60 80 100 1200.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
OO
CFD: Raw Signal
2mm
Monitoring Point(mid plane)
FSI-PfS-2a: Deflection of the Structure 16
t u / D
Uy
/D
800 850 9000.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
OO
EXP: Raw Signal
t u / D
Uy
/D
20 40 60 80 100 1200.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
OO
CFD: Raw Signal
Phase
Uy
/D
0 2 4 60.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8NumExp
AveragedPhase
FSI-PfS-2a: Deflection of the Structure 16
Frequency
St fFSI Error
(Hz) (%)
EXP 0.177 11.25 -
CFD 0.183 11.53 2.49
FSI-PfS-2a: Comparison Experiment/Simulation 17
Frequency
St fFSI Error
(Hz) (%)
EXP 0.177 11.25 -
CFD 0.183 11.53 2.49
Displacements
Uy/D|max Error Uy/D|min Error
(%) (%)
EXP 0.667 - -0.629 -
CFD 0.670 0.5 -0.674 7.2
FSI-PfS-2a: Comparison Experiment/Simulation 17
Streamwise velocity in the midplane
FSI-PfS-2a: Simulated Instantaneous Flow 18
Streamwise velocity Transverse velocity
EXP
CFD
FSI-PfS-2a: Comparison of Phase-averaged Data(t ≈ 1/24 T)
19
Streamwise velocity Transverse velocity
EXP
CFD
FSI-PfS-2a: Comparison of Phase-averaged Data(t ≈ 5/24 T)
20
1 Motivation / Objectives
2 Computational Methodology
3 ValidationDefinition of the Test CasesSimulations and Comparison with Experiments (FSI-PfS-2a)
4 Conclusions
Outline 21
Computational Methodology for FSI and ThinStructures
Each program specialized in its task
Each program parallelized (MPI, OpenMP)
New FSI coupling scheme developed
+ based on explicit time–marching scheme (predictor–corrector),but nevertheless stable and strong FSI algorithm
+ corrector step and structural computation directly connected ina FSI subiteration loop to achieve dynamic equilibrium
Conclusions 22
Computational Methodology for FSI and ThinStructures
Each program specialized in its task
Each program parallelized (MPI, OpenMP)
New FSI coupling scheme developed
+ based on explicit time–marching scheme (predictor–corrector),but nevertheless stable and strong FSI algorithm
+ corrector step and structural computation directly connected ina FSI subiteration loop to achieve dynamic equilibrium
Validation
Methodology validated for laminar flows (not presented here)
Methodology validated for turbulent flows (FSI-PfS-2a,...)
Generation of FSI test cases for the community with experimentaland numerical data available online (ERCOFTAC/QNET wiki)
Conclusions 22
Outlook
New coupling program (EMPIRE) → more flexibility in thecoupling
Reduce the CPU costs with the help of special wall models
Acknowledgments:
- Thanks to Jens Nikolas Wood for his support.
- Financially supported by the Deutsche Forschungsgemeinschaft (BR 1847/12-1)
- Computations on the Top-Level Computer SuperMUC at LRZ Munich.
Outlook 23
Thanks for your attention 24
Top Related