1. www.cranfield.ac.uk
The Impact of Unsteadiness on
Uncertainty in Automotive
Aerodynamics Simulation using
OpenFOAM
By: Akshat Srivastava
MSc – Computational Fluid Dynamics
Supervisor: Dr. Panagiotis Tsoutsanis
September 2016

2. 2
Presentation Overview
OpenFOAM
Case structure
Overview of solver
Computational domain
Mesh generation process
Refinements
Mesh dependence study
Frequency spectrum analysis
Instantaneous flows
Mean flow parameters
Mean flow statisticsFuture work and recommendations
Introduction
Turbulence Modeling
Software and Methodology
Computational Methodology
Results and conclusion
Objective

3. 3
Introduction
• Sphere is taken as a test case in this thesis work.
• Even though geometry is straightforward, the flow around sphere is very complex to analyse.
• Mechanism for transition in shear layer, behaviour and quantitative estimation of wake structure
is still rare for the considered Reynolds number of this thesis.
• The vortex shedding or the first mode is identified to the large-scale shedding in the wake.
• Second high frequency mode connected with the small-scale shear layer Kelvin-Helmholtz
instability on the fringe of the recirculation zone.
• Small-scale instability causes transition to turbulence in a long run.
• A frequency lower than large-scale frequency is also present which linked to the shrinkage and
enlargement of recirculation bubble.

4. 4
Objectives
Introduction continue…
• Using an open source numerical tool.
• Turbulence model which can capture the turbulent flow and is less expensive.
• Selection of suitable initial and boundary conditions.
• Total time period selection by considering the low-frequency component.
• Validation of results for instantaneous

5. 5
Turbulence Modelling
• The Improved Delayed Detached Eddy Simulation (IDDES), combines the advantages of the
wall modelled LES (WMLES) and the DDES capabilities.
• IDDES was developed to avoid the log-layer mismatch (LLM), between RANS and LES regime.
• Classical LES
• The computational domain is divided into three subdomains
1. Away from the wall
2. Region close to the wall
3. Region between the above limiting cases
, Simple DES , IDDES

6. 6
OpenFOAM
Software and Methodology
• OpenFOAM refers to Open Field Operation And Manipulation, which is basically a collection of
libraries.
• It provides serval utilities which help in mesh generation, pre processing and post processing.
• The solver used in this thesis work is pimpleFOAM which combines the PISO and SIMPLE
algorithm.

7. 7
Case structure in OpenFOAM
Subdirectory: 0 Contains Initial and Boundary Conditions
P
U
nuSGS
nuTilda
Pressure flow field
Velocity flow field
Subgridscale viscosity
Turbulent variable, SA
Subdirectory: constant Contains Mesh data and transport model
polyMesh
transportProperties
RASProperties
LESProperties
Directory contains mesh information
Transport model and fluid type is selected
Closure model (SA) and constants are set
LES model and constants are selected
Subdirectory: system Contains solver settings, schemes, etc.
controlDict
fvSolution
fvSchemes
Timestep, write control, functions : force
calculation and averaging, sampling, etc.
All solver settings, tolerance, etc.
Numerical schemes are selected
turbulenceProperties Turbulence model used (LES)

8. 8
Start
Initial u, p, F, turb
Update
Turbulence
Momentum
predictor u
Update H
Solve P
Update F and u
Non-orthogonal
Update
PISO
corrector
PIMPLE
corrector
Next time step
NO
NO
NO
YES
YES
YES
U, P, f, turbulence

9. 9
Example of log file in OpenFOAM from simulation

10. 10
Computational Methodology
• The sphere diameter is 1m and situated at
(0, 0, 0) coordinate.
• Computational domain in upward and radial
direction is 4.5D which corresponds to only 0.1%
change in velocity in comparison to free-stream
velocity.
• In downward direction it extends up to 25D
• Time step is selected using convection time which is 50 time
smaller than it to capture low-frequency fluctuation accurately,
it is 0.00029. Total time of simulation is 9.048 seconds

11. 11
Mesh Generation
Computational Methodology continue…
1. Patch geometry creation in Salome.
2. Export of each patch into separated STL files.
3. Add of patch names in each STL file.
4. Concatenation of all STL files into a single one.
5. Conversion of the single STL file into FMS format using the surfaceToFMS command of cfMesh.
6. Modification of patch types (wall, patch, symmetry) in the FMS file.
7. Preparation of the meshDict file.
8. Generation of the mesh using the cartesianMesh command from cfMesh.

12. 12
Computational Methodology continue…
Refinements

13. 13
Computational Methodology continue…
Cartesian mesh results
Structured mesh results

14. 14
Mesh Dependence Study
Computational Methodology continue…
• Three mesh were generated with y+ of 1
by using the additionalRefinmentLevels
dictionary of cfMesh, while global
maximum element size remain
unchanged.
• The table represents the statistics by
averaging 540 tU/D time units since
initial 75 tU/D time units are taken to
pass the transition stage.
• Errors were calculated by considering
the drag coefficient of experimental data
(by Schlichting). Strouhal number and
Separation angle by Achenbach.

15. 15
Computational Methodology continue…

16. 16
Computational Methodology continue…
Coarse Mesh
Medium Mesh
Fine Mesh

17. 17
Frequency Spectrum Analysis
Results and Conclusion
• Large time span provides data to analyse the existence of low-frequency fluctuations.
• Probes are step up using probe utility in OpenFOAM.
• The probes which are presented in thesis are; P1, P2, P9 and P4.
• Frequencies are computed via FFT with Tecplot for cross-stream velocity component.

18. 18
Results and Conclusion continue…
P1 P1
• With increase in
distance from the
sphere, the energy
content decreases
drastically.
• Due to turbulence
occur close to the
sphere at this
Reynolds number
P9 P4

19. 19
Results and Conclusion continue…
P1 P2
P9 P4
• Kelvin-Helmholtz instability which is associated with separating shear layer is recorded as 0.700
• Besides these two frequency, a much lower frequency then large scale vortex shedding is
recorded and is represented by fm here, it value is 0.038
• Low-frequency amplitude of
sphere is much lower then
other bluff bodies recorded
earlier.
• This low-frequency is an
attribute of shrinkage and
enlargement of recirculation
bubble.

20. 20
Results and Conclusion continue…
P2 P9
• Probes P2 and P9 are
selected which are just after
the mean recirculation
region.
• To analyse these two
probes, cross correlation is
performed.
• When P9 observe negative
or close to zero value then it
is indicative of enlargement
of recirculation region.
• The cross-correlation starts
with negative value which
indicates 180 degree phase
difference.

21. 21
Instantaneous flow
Results and Conclusion continue…
X-Y plane
X-Z plane

22. 22
Mean flow parameter
Results and Conclusion continue…

23. 23
Mean flow statistics
Results and Conclusion continue…

24. 24
Results and Conclusion continue…
• Profile of mean streamwise velocity at three
locations.
• At x/D=3, the flow is in recovery zone.
• Only difference is observed at last location.
• Compared with experimental results of Kim &
Durbin.

25. 25
Results and Conclusion continue…
• Mean streamwise and cross-streamwise velocity profiles compared with DNS results at Re=3700

26. 26
Results and Conclusion continue…
Mean streamwise stress Mean cross-streamwise stress
Mean shear stress Mean turbulent kinetic energy

27. 27
Future work and recommendations
• Error encountered during initial set up of simulation is the meshing software, latest version of
OpenFOAM would be preferred.
• Hybrid RANS-LES method is not working in the region of separation accurately, comparison with
other models would be preferred.
• For IDDES simulation, mesh is a crucial part, hence a perfect mesh for DES simulations should
be made with cfMesh or any other meshing software.
• Comparison with other solvers and turbulence models can provide solid verification and
validation of this thesis work.
• Long simulation time is still a problem, some other methods are available in literature which
could reduce the calculation time of such flow statistics.

28. 28

29. 29

30. 30
Miscellaneous slides

31. 31
LES principles
Turbulence Modelling
• LES computes the large scales which contains the
most of energy and models the small scale which
contributes fraction of total energy.
• Energy spectrum is divided into three sub-regions:
1. Energetic scale
2. Inertial subrange
3. Dissipative scale

32. 32
Turbulence Modelling continue….
• Filtering is applied in LES to write any variable as a contribution of large and small scale
• Standard filter applied in OpenFOAM is the implicit
top-hat filter.
• Eddy viscosity model is utilised in LES.
• The filtered Navier-Stokes equation is given by:
• Spalart-Allmaras is used as a turbulence closure
model

33. 33
Hybrid RANS-LES models
Turbulence Modelling continue….
• They are divided into two methods
1. Zonal hybrid method
2. Blended hybrid method
• Thesis used the blended hybrid method where smooth transition is made between different
regions.
• Common type of blended hybrid method is DES (Detached Eddy Simulation), switching
between RANS and LES depends on the local grid resolution.
• Original DES combines the standard Spalart-Allmaras RANS model with it Sub-Grid Scale
(SGS) counter part.

34. 34
Turbulence Modelling continue….
• Even though DES is very promising, it still
encounters mainly two drawbacks.
1. Grid Induces Separation (GIS) : When
wall bounded flows have thick boundary
layer and small separation regions.
2. Modeled Stress Depletion (MSD) : Arises
when SGS model of the DES originates
deep in the boundary layer.
• The Delayed Detached Eddy Simulation (DDES) was formulated in order to avoid the
appearance of the MSD.