MECH3780 Fluid Mechanics 2 and CFD

MECH3780
Fluid Mechanics 2 and CFD
Computation Fluid Dynamics

(CFD)
Lecture 8 – Turbulence
Modelling
Reference: CFD Lecture Notes by Assoc Prof Dr Kamarul Arifin Ahmad
Objectives
• To introduce the turbulence modelling

• To give overview about the turbulence
modelling
Tools to Approach turbulence
 The equations describing fluid flow are very
complex and are very difficult to solve.
CFD
Direct Numerical Simulations (DNS)
mathematical
hierarchy The Small Eddies are
being modeled
Large Eddy Simulations (LES)

The Small and large Eddies are
being modeled
Reynolds Averaged Navier Stokes (RANS)

Why the Need for Turbulence Models?
Turbulent energy is extracted from the mean flow by the large scale
structures (eddies).
Turbulent energy is dissipated by the smallest structures, at the
Kolmogorov microscale.
To resolve all of the dynamics of this process, the law of Wilcox
estimates
NDNS > Ret2.25
e.g. for Re = 106, Ret ≈ 103 and NDNS ≈ 107

for Re = 25×106, Ret ≈ 104 and NDNS ≈ 109
Turbulence modelling techniques
DNS
• The most accurate
• To calculate total number of grid, ∆
3
L 3
L
~ Re , ∆ =
4
η L x
η
• For RL=1000, at least 10009/4 ~5.6 million cells
• It’s too expensive
Direct Numerical Simulations
• Resolve ALL the scale (eddies)
DNS
-Only for low Re and simple geometries
-The details are not required for the
design purpose
-Large scale application is enough
for application
Flow over a back step
Courtesy of Stanford College

LES
• Less accurate than DNS but more accurate than RANS
• Reason
– Only large eddies influenced by the mean flow but not the small scale
• Use filtering method
– Turbulence in a small region of flow will be modelled
– Expanding the choices of turbulence model
• But still very expensive
• Available in FluentINC
RANS
• The most popular
• Model all the turbulence scales
– Hence coarser grid can be applied
– Less computation resources are needed
• Time averaging
– A lot of research about turbulence models helped the RANS method
to be accepted as a universal tool
• But care must be taken as each turbulence model is designed for a specific
application and never for general purpose
• Available in Fluent
Mathematical model.
N-S Equations
RANS
Viscous-inviscid
approximation
Thin shear layer Parabolised

Boundary layer
approximation Navier Stokes 3D Euler equations approximations
equations
Potential Flow Models

Approximation
level Linearised Potential Flows
RANS
u u′
u 0
= +
t
0 0
t t
Time-dependent Time-averaged Fluctuation

(instantaneous)
Turbulence modelling
• Boussinesq hypothesis
– - simple relationship between Reynolds stresses and velocity
– gradients through the eddy viscosity (similar to molecular viscosity)
– isotropic (eddy viscosity is a scalar!)
• Reynolds stress transport models
– equations derived directly manipulating the NS equations
– still contain unknown (undetermined) quantities
– no assumption of isotropy
– very complicated and expensive to solve
• Non-linear Eddy viscosity models (Algebraic Reynolds stress)
• Model directly the divergence of the Reynolds Stresses
Reynolds Averaged Navier Stokes
(RANS)
The picture can't be displayed.
Momentum transport equations (NS)
∂ui ∂ui u j 1 ∂p ∂  µ ∂ui 

+ =− +  
∂t ∂x j ρ ∂xi ∂x j  ρ ∂u 
 j 
ui = U i + ui'
∂U i
+U j
∂U i
=−
1 ∂p
+
∂  µ ∂ui
 +
(
 ∂ − ui' u 'j )
∂t ∂x j ρ ∂xi ∂x j  ρ ∂u  ∂x j
 j 
Needs modelling
The Boussinesq Assumption
Molecular viscosity is the ratio of shear stress to rate of strain:
τ
µ=
du dy
By analogy, a turbulent viscosity exists which is the ratio of the

Reynolds stress to the mean rate of strain:
 ∂u ∂u  2
− ρu u = µ 
/ /
 ∂x + i j
 − ρδ k
 3
∂x
i j t ij
 j i 
1
k =
uu /
i
/
i
2
The simplest task of turbulence modelling is to estimate νt.
Options of Boussinesq method
• Hundreds of them
– zero-equation/algebraic models:
• Mixing Length, Cebeci-Smith, Baldwin-Lomax, etc
– one-equation models:
• Wolfstein, Baldwin-Barth, Spalart-Allmaras, k-model, etc
– two-equation models:
• k-ε, k-ω, k-τ, k-L, etc.
– three-equation models:
• k-ε-A
– four-equation models:
• v2-f model
Mixing length model
Advantages
 Easy to implement and computationally inexpensive
 Good predictions for well behaved shear layers
 Well established
Disadvantages
Incapable of describing flows with separations or recirculation
Spallart Almarast (One equation model)
Advantages
 Not so new turbulence model
 Because of only 1 additional equation to be solved, the
computation resources needed is very low
 Developed originally for external aerodynamics flows
 Self-adaptive depending on the local y+
Disadvantages
Incapable of describing complex flows
The k-ε Model
Turbulent kinetic energy, k, is particularly important since
υ≈ k
k is destroyed by dissipation – the process is usually in equilibrium

Rate of dissipation, ε, scales like kυ , i.e.
l
3
k 2
ε≈
l
Hence, the mixing length model suggests that

k2
ν t = Cµ
ε
The k-ε Model
It remains to model k and ε.
But… Turbulence has a history
 transport by convection
 transport by diffusion
 production
 destruction
This can be modelled by equations of the form:
Convective Diffusive Rate of Rate of
Rate of change
of k or ε + transport of k = transport + production − destruction
or ε of k or ε of k or ε of k or ε
The k-ε Model
Convective Diffusive Rate of Rate of

Rate of change
of k or ε + transport of k = transport + production − destruction
or ε of k or ε of k or ε of k or ε
∂ρk µ 
+ ∇ ⋅ (ρkU) = ∇ ⋅  t ∇k  + 2µ t E ij E ij − ρε
∂t  σk 
∂ρε µ  ε ε
+ ∇ ⋅ (ρεU) = ∇ ⋅  t ∇ε  + C1ε 2µ t E ij E ij − C2ε ρε
∂t  σε  k k
The k-ε Model
Boundary conditions…
Very large changes in k and ε occur adjacent to solid walls
Computationally expensive to resolve effects down to y+ ≈ 1
Often preferred to use wall functions and integrate to y+ ≈ 30
= ln(Ey + )
u 1
Solve u + ≡ for uτ
uτ κ
u τ2 u τ3
Then k = and ε=
Cµ κy
The k-ε Model
Advantages
 Well established, validated and widely used
 Only initial and boundary conditions are required
 Good performance for many industrial flows
Disadvantages
Computationally expensive (2 additional PDEs)
Poor performance in flows with additional sources of strain,
especially swirling flows
Physics of the ε equation cannot be experimentally verified
The k-ω Model
Advantages
 Designed to overcome k-ε shortcomings
 It has less stiffness than the k- ε model when integrated into
the shear layer
 Due to simplify boundary conditions, this model is more stable
 Also a good performance for many industrial flows
Disadvantages
Also computationally expensive (2 additional PDEs)
The Reynolds Stress Model
6 transport equations exist for the 6 Reynolds stresses.
Solution also involves equation for dissipation rate.
Therefore 7 additional PDEs to be solved.
Advantages
 Most general of turbulence models
 Good predictions (including Reynolds stresses) for many
classical test cases
Disadvantages
Computationally expensive
Not widely used or validated
Performance is sometimes no better than that of k-ε model
Near wall treatment
• The near-wall region can be analysed in two
different ways, namely by wall functions or a near-
wall model
Use additional The turbulence

equations model is used up
until to the wall
Recommendation 1:
Modelling procedure
•Determine relevant Reynolds number to estimate if the flow is turbulent
•Select a turbulence model option and a near-wall treatment
•Estimate the physical dimension of the first grid point off the wall (y+)
•Generate the grid
•Perform the simulation
•REALITY CHECK (experiments, theory, mesh independency, time step
independency)
Recommendation 2: First grid point
off the wall
τw
y = ρuτ y / µ
+
uτ =
ρ
τw
cf =
1
ρU 2
2
Skin friction can be estimated using relationships below
cf
•Flat plate ~ 0.052(Re x ) −0.142
2
cf
•Pipe ~ 0.046(Re x ) −0.2
2
Exercise
• Estimate the distance of the first node from
the flat plate surface. Given that U=20m/s,
and desired y+ = 1 and y+=30. Take air as an
ideal gas.
Recommendation 3: Modelling
procedure
For k-ε simulations
•Two layer is preferable over wall functions (accuracy and grid dependency)
•rk- ε is preferable for separated flows
For k-ω simulations
•SST is better than standard
•Two layer method is more preferable
SA is usually a good compromise between accuracy and simplicity. It
has a very good convergence characteristics
Reynolds stress model is expensive and it require a good initial guess
(typically a k- ε type simulations)

MECH3780 Fluid Mechanics 2 and CFD

Uploaded by

Copyright:

Available Formats

MECH3780 Fluid Mechanics 2 and CFD

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

MECH3780 Fluid Mechanics 2 and CFD

Uploaded by

Copyright:

Available Formats

MECH3780

Fluid Mechanics 2 and CFD

Computation Fluid Dynamics

• To introduce the turbulence modelling

Large Eddy Simulations (LES)

Reynolds Averaged Navier Stokes (RANS)

e.g. for Re = 106, Ret ≈ 103 and NDNS ≈ 107

Courtesy of Stanford College

Thin shear layer Parabolised

Potential Flow Models

Time-dependent Time-averaged Fluctuation

∂ui ∂ui u j 1 ∂p ∂  µ ∂ui 

By analogy, a turbulent viscosity exists which is the ratio of the

k is destroyed by dissipation – the process is usually in equilibrium

Hence, the mixing length model suggests that

Convective Diffusive Rate of Rate of

Use additional The turbulence

You might also like