2007-03-24 TPC DD CFDOverviewOct2010 PDF

Tech-Pro Consultants
Turbulent Flow CFD –

An Overview
Dr. K.S.Ravichandran, Head Advanced Engineering

Ingersoll Rand Engineering & Technology Center
October 2010
References
• Analysis of Turbulent Flows by Tuncer Cebeci
• Turbulent Modeling for CFD by David C Wilcox
• Simulation of Turbulent Flows – Stanford Guidelines
Outline
 An Orientation for CFD
 Turbulence: Flow Features
 Turbulence Scales
 Simulation of Turbulence DNS
 Turbulence Modelling
 Computational Grid
 Example Test Case
Warning: Symbols may not be consistent through the presentation
An Orientation for CFD
The Challenges
 Flow Physics: Complex and Unintuitive flow fields,
Flow Features: turbulence (and transition) modeling:
Reynolds number, mixing, heat transfer, pressure gradients,
recirculating flow, vortices, rotation.
Mathematical Model: DNS, LES, RANS, (un)steady,

3D/2D/1D, non-viscous (Euler), (in)compressible, boundary-
layer, potential flow, ...
 Boundary Conditions: inflow, outflow prescription and
extrapolation, far field, wall, symmetries, ...
 Additional Complications: Combustion, multi-phase, non-
Newtonian fluids (polymers, metals), ...
Real Gas Effects: Mach number, temperature, calorically or
thermally perfect gas, chemical reactions, chemical and thermal
non-equilibrium effects, rarefied gas, ...
Computational Grid:
 Resolving Gradients: shocks, boundary layers, free
shear layers, vortices,
 mesh adaptation, ...
 Scale Separation: global scales, viscous scales
(turbulence), molecular scales (shocks), ...
 Grid Topology: stretching, cell orientation and shape,
structured/unstructured/ hybrid grids
 Geometry: approximations of details
Numerical Method: No Universal Method
 Basic concept: FEM, FVM, FDM, Spectral, SPH...
 Mathematical Models: pressure or density based N-S
Solvers, Vorticity-Velocity based methods, Vorticity-
Stream function approaches,
 Numerical Algorithms: multi-grid, implicit, explicit,
AF, steady state relaxation, ...
 spatial schemes: central, upwind, higher order, ...
Computational Resources:
 Computational time: parallelization, number of grid
points, Convergence acceleration, Algorithmic
requirements
 Memory use: number of grid points, number of
equations, model and algorithmic complexity (physical
/ Mathematical)
Accuracy / Quality:
 Numerical errors: grid convergence, different
numerical with schemes
 Data Comparison: experiments, others results
 Assessment of approximations: compare different
models and approximation levels
 CFD code settings
Turbulence: Flow Features
Turbulence
Effect of Reynold’s Number
Turbulent Motion: Experiment, Cylinder
Turbulent Motion: CFD-LES, Cylinder
Ref: Large Eddy Simulation of Flow around

Circular Cylinders, J. Fr¨ohlich et. al.
Turbulent Motion: CFD-LES
Flow Domains
Turbulent Motion on Flat Plate: Velocity Traces
Turbulent Motion: Boundary Layers
Turbulence Intensities in Boundary Layers
Pressure Drop in Laminar and Turbulent Flows
Turbulence: Scales
Turbulence: Basic Concepts
 Turbulence : random fluctuations , 3D, time dependent
 Energy cascade (L and U): generated at the largest scales,
large scale vortices break down to smaller vortices
 dissipates to heat at the smallest viscous scales ,
 balanced cascade (equilibrium turbulence)
 turbulent kinetic energy
Scale Separation
 Large scales L and U related to global scales
 Small viscous scales
 Scale separation
 Reynolds number
Wall Bounded Turbulence
 Wall friction velocity
 Viscous wall scales
 Friction Coefficient
 Viscous Wall Distance
Velocity Distribution in Flat Plate Boundary Layers
Different Reynolds Numbers
 Based on wall skin friction

( is the boundary layer thickness)
 Based on boundary layer thickness
 Based on global scales
 Based on distance from leading edge

(flat plate)
Empirical Relations
 Friction coefficient (Flat Plate)
• turbulent: ; laminar:
 At what value of Re are the two values comparable?
(Re=3150)
 What is ratio of turb to lam skin friction? ~ 0.09 times
(Re)0.3
 Boundary layer thickness
• turbulent: ; laminar:
 At what Re are the values comparable? (~ 5900)
 Ratio of BL thickness (turb/lam)? 0.075 times (Re)0.3
Simulation of Turbulence: DNS
Simulation of Turbulence: DNS
 DNS: Direct Numerical Simulation
 Solution of the 3D time dependent Navier-Stokes equations
 Computational effort scales rapidly with Reynolds number
 Used for low Reynolds number generic flows
 Gives detailed knowledge about turbulence
 Not (yet?) practically useful for “CFD applications”
Computational Effort For Free Turbulence (No Wall): DNS
 Number of grid points
 Number of time steps
 Computational effort
Computational Effort For Wall Bounded Turbulence
 Number of grid points
 Number of time steps
 Computational effort
DNS of Flow Past a Flat Plate
Turbulence Modeling
Turbulence Modeling
Reynolds averaged Navier-Stokes equations (RANS)
Navier-Stokes equations ( denote instantaneous values)
Conservation of mass
Conservation of momentum
Turbulence Modeling
 Reynolds decomposition
Where
 The “mean” is time average, ensemble average or averaging in
homogeneous directions. (Ui(x) may actually vary in time with a time
scale much longer than the turbulent time scale.
Take the mean of the Navier-Stokes equations -> RANS
Turbulence: Reynolds Stresses
 Not “small”
 Significant effects on the flow
 Needs to be modeled in terms of mean flow quantities
 Reduces the problem to steady (or slowly varying)
 2D assumptions possible
Derivation Of Equations For The Reynolds Stresses
 Subtract RANS equations from Navier-Stokes equations
 Take dot product of the equation above with uj and repeat

the operation with i and j interchanged
 Sum the resulting two equations, do the averaging and
contract on the indices i, j to obtain the equation for uj2 .
This results in the equation for the turbulent KE
Reynolds Stress Equations
Equation For Turbulent Kinetic Energy

Eddy-Viscosity Models (EVM)
 Assume: Reynolds stresses related to an “eddy viscosity”, νT
 Based on Boussinesq (1877) Eddy viscosity ~

turbulence velocity V and length scales L: νT ~ VL
Algebraic Models:
 Velocity V and length scale L related to mean flow field
and global geometry (wall distance,wake thickness)
 Works well for attached boundary layers
 Not very general
 Example: Baldwin-Lomax (1978)
One-Equation Models:
 One transport equation for K (turbulent kinetic
energy) or νT.
 Additional information from global conditions
(typically wall distance)
 more general than algebraic models
 Example: Spalart-Allmaras (1992) (a reasonable
and robust model for external aerodynamics)
Two-Equation Models:
 Two transport equations for the turbulent scales, typically K- ε or K- ω.
 Completely determined in terms of local quantities (except near-wall
corrections which may be dependent on wall distance)
 Somewhat more general than algebraic and one-equation models.
 Model transport equations loosely connected to the exact equations.
 Examples:
 Standard K- ε model (Launder & Spalding 1974)
 Wilcox (1988, ...) K- ω models
 Menter (1994) SST K- ω model (performs reasonable well for separated
flows)
Two-Equation Models: (Eddy Viscosity)
 Model equations (standard K- ε model):
 Turbulent viscosity
 Production of K
 Model coefficients calibrated from generic flows (homogeneous

turbulence, boundary layers)
Problems:
 No dependency on rotation or curvature in equations. Whereas real
turbulence is strongly dependent.
 Modeled production proportional to strain rate squared ~S2 . Exact
production ~S. Results in an overestimated production of K in highly
sheared flows (around stagnation points, pressure gradient BLs, separated
flows).
Fixes
 Rotation & curvature corrections
 Yap correction (limit excessive turbulent length scale)
 Menter SST correction (limit excessive νT )
Fixes :
 More general and higher order models may be considered.
 Standard K- ω model:
• Closely related to K- ε through a variable transformation where
• Somewhat better behaviour than K- ε in pressure gradient BLs and

separation
• Unphysical dependency on free stream turbulence level
Near-Wall Treatment
 In the very near wall region (y+ < 50 ):
• large velocity gradient
• large variations of turbulence quantities
 Low-Reynolds number models
• May contain specific near-wall correction terms
• May be dependent on wall distance and/or wall skin friction
• Require high wall-normal grid resolution, typically y1+ (distance to
the first grid point) and 5-10 points within y+ =20 . Typically ~ 30-50
grid points in the boundary layer.
Near Wall Variation of k+
Near Wall Variation of ε+
Wall Boundary Conditions
 zero turbulent fluctuations: , also
 balance in the equation:
 Note that goes to infinity at the wall and there are

different ways of taking care of that in CFD codes (no
major problem).
Wall Function Boundary Conditions
 Reasonable universal wall law for (y+ < 50 ):
 The first grid point must be carefully placed in the lower part of the
log layer (y+ > 20 , y > 0.1 )
 Velocity and turbulence quantities set according to log-law relations
 Theory breaks down at boundary layer separation / reattachment
(the method is still useful).
 Lowers computational cost by one order of magnitude
 “Adaptive” or “scalable” wall-function boundary conditions may work
also for finer grids and will adapt to the low-Reynolds number
solution for sufficient fine grids. Not commonly used yet.
Near Wall Grid for Turbulent Flow
Eddy Viscosity in Pipe Flow
Mean Velocity Distribution in Pipe Flow
Planar Jet: CFD
Round Jet: CFD
Reynolds Stress Models (DRSM, RST, SMC)
 Based on the exact transport equations for derived from the
Navier-Stokes equations
 Terms dependent on higher order correlations and

cannot be closed -> modelling
 More general: Captures effects of rotation, curvature, secondary
motions, nonequilibrium and other forms of complicated strain.
 Computational expensive and often numerical robustness problems
 Example: SSG (Speziale, Sarkar & Gatski 1991)
Algebraic Reynolds Stress Models (ARSM, ASM, EARSM)
 Applying the “weak equilibrium assumption” to the DRSM
equation, results in an purely algebraic relation for .
 Replaces the eddy-viscosity assumption in the K- ε or K- ω
models
 Computational effort and robustness comparable with two-
equation eddy-viscosity models.
 The model behaviour of the DRSM is to some extent retained.
 Examples:
 Gatski & Speziale (1993)
 Wallin & Johansson (2000)
Non-linear Eddy-Viscosity Models (NL-EVM)
 Empirical extension of the eddy-viscosity assumption
 Examples:
 Shih, Zhu & Lumley (1992, 1993)
 Craft, Launder & Suga (1993)
Large Eddy Simulation of Turbulence (LES) and
LES-RANS hybrids
 Simulation of the large scale turbulence and modeling small scales
(compare with DNS, simulation of all scales and RANS-modeling of all
scales)
• Always time dependent and 3D -> expensive
 Wall free turbulence simulations Re independent
 Wall bounded turbulence largely Re dependent
• fully resolved near-wall region very expensive (almost as DNS)
• wall-function or near-wall RANS coupling saves computational cost
• hybrid RANS-LES (RANS in attached BLs and LES in wall-free
separated regions) a very active research field
Large Eddy Simulation of Turbulence (LES) and
LES-RANS hybrids (contd)
 LES in academic research for:
• low Reynolds number generic flows
• complement to DNS for higher Re
• gives detailed knowledge about turbulence
 LES in industrial use for:
• internal flow with complex geometries
• flows around blunt bodies (with large separated regions)
• atmospheric boundary layers (e.g. weather forecasts)
• combustion simulation and other complex flow physics at moderate Re
 Warning: £€$ is extremely expensive in high Re attached and slightly
separated wall-bounded flows, if properly resolved.
Computational Grid and Boundary
Conditions
Computational Grid
 Geometry definition
• Sources of geometry definitions
• CAD definition of the structure (complex “engineering”
geometries)
• Mathematical definition of surfaces (simple geometries)
 Surface grid (previous CFD or other computations)
• Need to be converted for input to grid generation tools
• Preferable in “clean surface definitions” (e.g. splines)
• Cleaning of CAD definitions not at all a trivial task
Basic Approaches - 1
 Cartesian grids
• A regular and cartesian mesh covering the geometry
• Grid points inside of the object are removed
• Efficient solver algorithms
• Problem of capturing geometry details
• Problem of capturing viscous boundary layers
• Not very commonly used
 Body-fitted grids
• Grid lines follow the surfaces
• Geometry details can be captured
• Grid points easily clustered in viscous boundary layer
• Could be structured or unstructured or hybrid
• Most frequently used
 Structured body-fitted grids
• The “traditional” approach
• Regular blocks of hexahedral grid cells
• Efficient solver algorithms
• Multi-block approach for complex geometries
• Grid generation a tedious “art” (complex grids can take
months!)
• No general automatic grid generation algorithm
• Grid points not easily located where they are needed
• Solution of high accuracy on well designed grids
 Unstructured body-fitted grids
• Most common in commercial CFD solvers today
• Grid cells of different types (tetrahedra, hexahedra, prisms and
pyramids)
• Cell connectivity information -> less efficient solver algorithms
• Grid generation can be highly automated
• Grid points locally clustered without need to influence the whole
computational domain
Grid Quality -1
 Sufficiently fine grids
• Grid points must be clustered around regions of large
gradients (e.g. shocks, boundary layers and other thin shear
layers)
• Unstructured grids may be automatically clustered based
on the solution (gradients or error estimates) -> grid
adaptation
 Shape of the cells
• Skewness
• Aspect ratio
Grid Quality -2
 Orientation of cell faces
• Preferable normal to flow gradients
 Spatial distribution of cell sizes
• Smooth changes of cell sizes, max 10-20% changes
between neighboring cells
• Grid quality is particularly important around large gradients
 Grid quality may influence both accuracy and numerical
stability
Grid Quality -3
 Grid topology
• Needs to be carefully chosen for structured grids
• Limited possibilities to control for unstructured grids
 “Prismatic layers” in boundary layers (unstructured
grids)
• Locally “structured” near-wall grid topology
• Improves grid quality (cell shape, cell-face orientation, cell
size distribution)
• Required in high Reynolds number boundary layers
Boundary Conditions -1
 Initial conditions
• A steady-state problem is completely determined by its
boundary conditions
• Also the final stage of a limit-cycle oscillation is
independent of the initial conditions
• Only the transient computation of an initial-value problem
is dependent on the initial conditions
• However, the initial conditions may be important for the
convergence to steady state
 Boundary flow field
• The flow field at the boundary may be strongly dependent on
the flow field in the computational domain. Supersonic inflow and
Supersonic outflow are the only exceptions.
•Subsonic inflow: Most information from boundary to interior,
pressure related information from interior to boundary.
• Subsonic outflow: Only pressure related information from
boundary to interior
• Supersonic inflow: All information from boundary to interior
• Supersonic outflow: No information from boundary to interior
 If interaction between the boundary and interior flow
fields are a problem:
• Extend the boundaries for external flows, typically 10-50 times
the size of the object. Less problem in 3D flows.
• Include the stagnation chamber or the true inflow/outflow
geometries for internal flows
• Make empirical/mathematical corrections on the boundaries
• Be aware of the problem!!!
 If interaction between the boundary and interior flow
fields are a problem:
• Extend the boundaries for external flows, typically 10-50 times
the size of the object. Less problem in 3D flows.
• Include the stagnation chamber or the true inflow/outflow
geometries for internal flows
• Make empirical/mathematical corrections on the boundaries
• Be aware of the problem!!
Exactly what quantities are set on the boundaries or taken
from the interior seriously impacts numerical stability
Inflow Conditions for Turbulence -1
 At inflow boundaries the turbulence quantities must be prescribed
for turbulent computations
 Inflow turbulence levels mostly not fully known
 Solution may be strongly dependent on the inflow turbulence
levels, but mostly only minor dependencies
 Important to prescribe realistic values for solution accuracy and
numerical stability
 Turbulence level
 Turbulent Length Scale
• External flows: T/ 1 to 10 scale is a reasonable guess
• Internal flows: turbulent length scale related to geometry
of geometrical scales
• Check also the length scale on which the advected
turbulence changes. That should be in the order of, or larger
than, typical geometrical scales.
 If problems: move the inflow boundary sufficiently far from
the region of interest.
 The turbulence levels at the inflow boundary are also
applicable as initial conditions
Transition to Turbulence
 Turbulent boundary layers always proceeds by a laminar boundary layer
and transition to turbulence
 The location of the transition point (or region)
• Depends on surface roughness, free stream turbulence, noise, etc.
• No general method to predict
• Hard to measure
 The flow may be dependent on the transition location
• Try to get information from experiments
• Try to estimate (specific empirical relations exist)
• Compute the growth rate of disturbances (a subject as big as CFD)
• Assume the flow fully turbulent (if transition is unimportant)
 Transition location prescribed in CFD by setting laminar or turbulent walls
A Computational Test Case
Transonic Wing Profile (test case)
 RAE 2822, Case 10
 2D wing profile in transonic wind tunnel, slotted walls
 Conditions
Ma = 0.754 α = 2.57 Rec = 6.2 x 106
 Tripped transition at x / c = 0.03 on upper and lower sides
 Reference: Cook, P.H., MacDonald, M.A. & Firmin, M.C.P. 1979
Aerofoil 2822 – Pressure distributions, boundary layer and wake
measurements. AGARD AR 138.
Geometry
 Geometry defined from measurements of the experimental model as a sequence
of points + wind tunnel corrections
• Wing chord, c = 1.0 (normalized geometry)
 2D approximation can be used
• Wind tunnel experiment setup to be 2D
• 3D effects are always present and may be of significance (warning!)
 Free external boundaries
• Wind tunnel walls are “slotted” for simulating free conditions
• Slotted wind tunnel walls hard to model in CFD
 Sharp trailing edge
• The true trailing edge is mostly blunt
• Effect of the trailing edge bluntness mostly local
Test Conditions - 1
 Wind tunnel corrections
 Temperature T∞ = 300 K
 Mach number Ma = U/a = 0.754 , where a is speed of sound.
• Speed of sound a = γRT , air: =1.4, R=287,->U =262 m/s
• Transonic speed: shocks may be present
Test Conditions - 2
Computational Grid
• Structured grid
• C-grid topology
• Far-field distance ~ 10c
• Number of nodes on the profile ~ 200
• Viscous wall scale
Boundary Layer Thickness
 Shock induced boundary-layer separation

 low-Reynolds number model: ->
 Choose: 224 cells on surface, 24 in the wake, 80 in wall
normal direction > 273 x 81 nodes
 Multi grid: number of cell on each boundary multiple of 8 (or
at least 4)
Turbulence (use K-ω type of models) -1
 Turbulence level
• Tu = 0.3% (in external aerodynamic flows)
• Turbulent kinetic energy
Transition: set transition at x / c = 0.03 on upper and lower sides
Turbulence (use K-ω type of models) -2
 Turbulent length scale
• T/ 10 (external flows)
• Dissipation rate
• Specific dissipation rate
• Check turbulent length scale
• Check the length scale on
which the advected turbulence changes
Boundary Conditions
 External boundaries:
• Set to free stream (U,V,P,T,K, ) (or ε∞ if K – ε models)
• Use “Riemann invariant”, or “Characteristic” boundary
conditions
 Wall:
• No-slip conditions
• Adiabatic wall (no heat flux through wall)
 Connectivity:
• Connect the boundaries in the wake
Basic Solver - 1
 Transonic flow
• Density based (compressible) solution method
• Calorically perfect gas
 Steady-state problem
• Solving the time-dependent equation to steady state
• Local time stepping: non-physical time step in each cell considering
only
 stability
• CFL=1.5
Basic Solver - 2
 Multi grid
• Convergence acceleration method
• Iterating on different grid levels
• 3 grid levels
 Spatial discretization
• 2’nd order upwind scheme
• symmetric TVD
• van Leer limiter
• Entropy fix
Initiation
 Initiate the field with the free stream values
 Full multi grid
• Start on the coarsest level
Turbulence model
• Use Menter SST K – ω model
Check the Results
 Convergence
 General flow picture
 y+ in first grid points
 Wall pressure
• Normalized
 Wall skin friction
• Normalized
Sources of Errors - 1
 Grid
• Grid convergence studies
• in first grid points
• stream wise resolution around suction peak and separation
 Geometry
• wind tunnel effects (3D effects, wall effects, “free stream” conditions)
• trailing edge resolution
Sources of Errors – 2
 Turbulence model
• Check sensitivity by computing with other models
 Numerical scheme
• Check influence of scheme
• In principle: sensitivity of scheme decreases on finer grids
• Check convergence rate (iterate more)
For
Any questions
The SST model of Menter [5] combines several desirable elements of
existing
two-equation models. The two major features of this model are a zonal
weighting
of model coefficients and a limitation on the growth of the eddy viscosity in
rapidly strained flows. The zonal modeling uses Wilcox's k-w model near
solid
walls and Launder and Sharma's k-~ model near boundary layer edges and
in
free shear layers. This switching is achieved with a blending function of the
model coefficients. The shear stress transport (SST) modeling also modifies
the
eddy viscosity by forcing the turbulent shear stress to be bounded by a
constant
times the turbulent kinetic energy inside boundary layers. This modification,
which is similar to the basic idea behind the Johnson-King model, improves
the
prediction of flows with strong adverse pressure gradients and separation
Wilcox's model equations have the advantage over the k-~ model that they
can be integrated through the viscous sublayer, without using damping
functions.
At the wall the turbulent kinetic energy k is equal to zero
The Spalart-Allmaras model is a relatively simple one-equation model that solves a
modeled
transport equation for the kinematic eddy (turbulent) viscosity. This embodies a
relatively new class of one-equation models in which it is not necessary to calculate
a
length scale related to the local shear layer thickness. The Spalart-Allmaras model
was
designed specifically for aerospace applications involving wall-bounded flows and
has been
shown to give good results for boundary layers subjected to adverse pressure
gradients.
It is also gaining popularity in the turbomachinery applications
May be low Re model or can be modifed to be used with Wall functions
Including both the rotation and strain tensors reduces the production of eddy
viscosity and consequently reduces the eddy viscosity itself in regions where the
measure of vorticity
exceeds that of strain rate
Instantaneous Velocity in Boundary Layers

2007-03-24 TPC DD CFDOverviewOct2010 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2007-03-24 TPC DD CFDOverviewOct2010 PDF

Uploaded by

Copyright:

Available Formats

Tech-Pro Consultants

Turbulent Flow CFD –

Dr. K.S.Ravichandran, Head Advanced Engineering

• Analysis of Turbulent Flows by Tuncer Cebeci

• Turbulent Modeling for CFD by David C Wilcox

• Simulation of Turbulent Flows – Stanford Guidelines

Warning: Symbols may not be consistent through the presentation

Mathematical Model: DNS, LES, RANS, (un)steady,

Ref: Large Eddy Simulation of Flow around

 Small viscous scales

 Wall friction velocity

 Viscous wall scales

 Viscous Wall Distance

 Based on wall skin friction

 Based on boundary layer thickness

 Based on global scales

 Based on distance from leading edge

 Number of grid points

 Number of time steps

 Number of grid points

 Number of time steps

Navier-Stokes equations ( denote instantaneous values)

 Take dot product of the equation above with uj and repeat

 Based on Boussinesq (1877) Eddy viscosity ~

 Model coefficients calibrated from generic flows (homogeneous

 Menter SST correction (limit excessive νT )

• Somewhat better behaviour than K- ε in pressure gradient BLs and

 balance in the equation:

 Note that goes to infinity at the wall and there are

 Terms dependent on higher order correlations and

 Shock induced boundary-layer separation

Transition: set transition at x / c = 0.03 on upper and lower sides

• Specific dissipation rate

• Check turbulent length scale

• Check the length scale on

which the advected turbulence changes

May be low Re model or can be modifed to be used with Wall functions

You might also like