Numerical Fluid Mechanics: Chungen Yin, PH.D

Numerical Fluid Mechanics Numerical Fluid Mechanics
Chungen Yin, Ph.D.
Pontoppidanstræde 101, room 8
Lecture 1 Introduction to CFD
Lecture 2 Finite volume method
Lecture 3 Finite volume method (cont)

Numerical Fluid Mechanics
Lecture 4 Solvers and algorithm
Lecture 5 Error/uncertainty; Discretization vs. CFD accuracy
Lecture 6 Advanced meshing: Tips & tricks
Lecture 7 Transient problems
Lecture 8 Turbulence
Lecture 9 Heat transfer
Lecture 10 Project-related topic

2
2
Numerical Fluid Mechanics Lecture
8
Turbulence modeling
• Turbulence (energy cascade, scales...)
• RANS modeling & Near-wall modeling
Turbulence
· What is turbulence?
· Effect of turbulence on N-S eq.
· Reynolds averaging.
· Reynolds stresses.
Laminar flow vs. Turbulent flow: Animation1

Animation2
4
Turbulence: (1) High Re numbers
Turbulent flows always occur at high Reynolds numbers. They are caused by
the complex interaction between the viscous terms and the inertia terms in the
momentum equations.
Turbulent, high Reynolds

number jet
Laminar, low Reynolds

number free stream flow
5
5
Transition in a jet flow
6
Small Structure Large Structure
Large-scale vs. small-scale structures
External flows: ρUL

where Re L ≡
μ
Rex ≥ 5×10 5 along a surface
L = x, D, Dh, etc.
ReD ≥ 20,000 around an obstacle
Other factors such as free-
stream turbulence, surface
Internal flows: conditions, and disturbances
may cause earlier transition to
Re Dh ≥ 2,200
turbulent flow.
Natural convection:
gβΔTL3ρ
Ra ≥ 10 − 10
8 10
where Ra ≡
μα
Boundary layers:
Re ≥ 105
The Re number thresholds for different flows

8
8
Turbulence: (2) turbulent flows are chaotic
· One characteristic of turbulent flows is their irregularity or randomness. A full

deterministic approach is very difficult. Turbulent flows are usually described
statistically. Turbulent flows are always chaotic. But not all chaotic flows are
9
turbulent.
Turbulence: (3) diffusivity
·
· The diffusivity of turbulence causes rapid mixing and increased rates of

momentum, heat, and mass transfer. A flow that looks random but does not
exhibit the spreading of velocity fluctuations through the surrounding fluid is not
turbulent. If a flow is chaotic, but not diffusive, it is not turbulent.
10
10
Turbulence: (4) dissipation
·
· Turbulent flows are dissipative. Kinetic energy gets converted into heat due to
viscous shear stresses. Turbulent flows die out quickly when no energy is
supplied. Random motions that have insignificant viscous losses, such as
random sound waves, are not turbulent.
11
11
Turbulence: (5) rotation and vorticity
Turbulent flows are rotational; that is, they have non-zero vorticity.
Mechanisms such as the stretching of three-dimensional vortices play a key

role in turbulence.
Vortices
12
12
What’s turbulence — Summary
Numerical Fluid Mechanics · Main characteristics:
▫ One characteristic of turbulent flows is their irregularity or
randomness. A full deterministic approach is very difficult.
Turbulent flows are usually described statistically. Turbulent flows
are always chaotic. But not all chaotic flows are turbulent. For
example, waves in the ocean can be chaotic but are not
necessarily turbulent.
▫ The diffusivity of turbulence causes rapid mixing and increased
rates of momentum, heat, and mass transfer. A flow that looks
random but does not exhibit the spreading of velocity fluctuations
through the surrounding fluid is not turbulent. If a flow is chaotic,
but not diffusive, it is not turbulent. The trail left behind a jet
plane that seems chaotic, but does not diffuse for miles is then
not turbulent.
▫ Turbulent flows always occur at high Reynolds numbers. They
are caused by the complex interaction between the viscous terms
and the inertia terms in the momentum equations.
13
13
What’s turbulence — Summary (cont)
▫ Turbulent flows are rotational; that is, they have non-zero

vorticity. Mechanisms such as the stretching of three-dimensional
vortices play a key role in turbulence.

▫ Turbulent flows are dissipative. Kinetic energy is converted into
heat due to viscous shear stresses. Turbulent flows die out quickly
when no energy is supplied. Random motions that have
insignificant viscous losses, e.g., random sound waves, are not
turbulent.
▫ Turbulence is a continuum phenomenon. Even the smallest
eddies are significantly larger than the molecular scales.
Turbulence is therefore governed by the equations of fluid
mechanics.
▫ Turbulent flows are flows. Turbulence is a feature of fluid flow,
flow
not of the fluid. When the Reynolds number is high enough, most
of the dynamics of turbulence are the same whether the fluid is an
actual fluid or a gas. Most of the dynamics are then independent
of the properties of the fluid.
14
14
Turbulent boundary layer
Top view
Side view
Merging of turbulent spots and transition to turbulence

in a natural flat plate boundary layer.
15
15
Turbulent boundary layer (cont)

Transition in boundary layer flow
16
16
Turbulent boundary layer (cont)
T-S waves Turbulent spots Fully turbulent flow
17
17
Energy Cascade
• Larger, higher-energy eddies, transfer energy to smaller eddies via

vortex stretching
- Larger eddies derive energy from mean flow
- Large eddy size and velocity on order of mean flow
• Smallest eddies convert kinetic energy into thermal energy via viscous
dissipation
- Rate at which energy is dissipated is set by rate at which they receive energy
from the larger eddies at start of cascade
18
18
19
19
Vortex Stretching
· The largest turbulent eddies interact with and extract energy from the
mean flow by a process called vortex stretching
· Existence of eddies implies rotation or vorticity, which is concentrated

along contorted vortex lines or bundles.
· Vortex lines/bundles become distorted from the induced velocities of
the larger eddies. As as end points of a vortex line move randomly
further apart, (1) the vortex line increases in length but decreases in
diameter; (2) vorticity increases because angular momentum is
nearly conserved, and most of the vorticity is contained within the
smallest eddies. Kinetic energy increases at rate equivalent to the
work done by large-scale motion that stretches the bundle.
· Viscous dissipation in the smallest eddies converts kinetic energy into
thermal energy.
· Vortex-stretching cascade process maintains the turbulence and
dissipation is approximately equal to the rate of production of
turbulent kinetic energy.
20
20
Vortex Stretching (cont)
t1 t2 t3
t4 t5 t6
21
(Baldyga and Bourne, 1984)
21
Smallest Scales of Turbulence

• Smallest eddy (Kolmogorov) scales:
- large eddy energy supply rate ~ small eddy energy dissipation rate
→ ε = − dk / dt
k ≡ 1 (u ′2 + v′2 + w′2 ) is specific turbulent kinetic energy [m2/s2];

2
ε is dissipation rate of k , [m2/s3]
• Motion at smallest scales depends on dissipation rate, ε , and kinematic

viscosity, ν [m2/s]
• From dimensional analysis,

η
η = (ν 3 / ε )1 / 4 [m] τ = (ν / ε )1 / 2 [s] Vˆ = = (νε )1 / 4 [m/s]
τ
22
22
Small scales vs. Large scales
• Largest eddy scales
- Assume l is characteristic of larger eddy size;

- Dimensional analysis to estimate order of large eddy supply rate of k as

,k / τ
turnover
- τ turnover is a time scale associated with larger eddies; the order of τ turnover
can be estimated by τ turnover = l / k
1/ 2
• Since ε ~ k / τ turnover : ε ~ k3/ 2 / l or l ~ k3/ 2 / ε
• Compare l with η
3/ 4
l l l × ( k 3 / 2 / l)1 / 4 ⎛ k1 / 2l ⎞ l
= ≈ =⎜ ⎟ = (ReT )3 / 4 >> 1
η (ν 3 / ε )1 / 4 ν 3/ 4 ⎜ ν ⎟ η
⎝ ⎠
k1 / 2l
ReT = Turbulence Reynolds number
23 ν
23
Energy spectrum vs. Wave number (log-log scales)

∞
1 u ′u ′ = ∫ E (κ ) dκ
2 i i 0
(turbulent kinetic energy)
3 1
l ~ k 2 /ε η = (ν 3 / ε ) 4
k wave number ( κ ) l integral scale η Kolmogorov scale

24
24
Implication of scales
Numerical Fluid Mechanics • Consider a mesh fine enough to resolve smallest eddies. An example:
(2D channel flow)
Cells in 3D needed:
N cells ~ (4l / η )3
u H
N cells = (3 Reτ )9 / 4 where Reτ = τ [Moser & Moin 1984;
2ν Kim, Moin & Moser, 1987]
For turbulent flows:

Re H = 30800 Reτ = 800 N cells = 4 × 107
uτ ~ 5%U
25
25
DNS (Direct Numerical Simulation)
• “DNS” is the solution of the time-dependent Navier-Stokes equations

without modeling
⎛ ∂ui ∂u ⎞ ∂p ∂ ⎛⎜ ∂ui ⎞
ρ⎜ +uj i ⎟ = − + μ ⎟
⎜ ∂t ∂x ⎟ ∂x ∂x ⎜ ∂x j ⎟
⎝ j ⎠ i j ⎝ ⎠
- Numerical time step size required Δt ~ τ

0.003 H
- For the 2D channel example Δt 2d Channel ≈ ; number of time steps
~48000 Reτ uτ
- DNS is not suitable for practical industrial CFD
- DNS only for simple geometry and low turbulence Reynolds number;
however, a useful research tool.
26
26
Removing the small scales
• Two methods can be used to eliminate need to resolve small scales:

- Reynolds Averaging: Transport equations for mean flow quantities are

solved; All scales of turbulence are modeled; Transient solution Δt is set by
global unsteadiness;
- Filtering (LES): Transport equations for ‘resolvable scales’; Resolves larger

eddies, models smaller ones; Inherently unsteady, Δt set by small eddies
• Both methods introduce additional terms that must be modeled for

closure.
27
27
28
28
RANS (Reynolds Averaged Navier-Stokes eqns)
29
29
RANS modeling – Ensemble (time) Averaging
• Consider a point in a given flow field,

r r r
ui ( x , t ) = U i ( x , t ) + ui′ ( x , t )
30
30
r 1 N ( n) r
• Ensemble (phase) average: U i ( x , t ) = lim
N →∞ N n =1
ui ( x , t ) ∑
31
31
Deriving RANS equations
• Substitute mean plus fluctuations in the instantaneous NS equations

and average,
⎛ ∂ (U i + ui′ ) ∂ (U i + ui′ ) ⎞⎟ ∂ ( p + p′) ∂ ⎛ ∂ (U i + ui′ ) ⎞

ρ⎜ + (U j + u ′j ) =− + ⎜μ ⎟
⎜ ∂t ∂x j ⎟ ∂xi ∂x j ⎜ ∂ x ⎟
⎝ ⎠ ⎝ j ⎠
• Use some averaging rules, such as
If x = X + x' and y = Y + y'
Then X = X ; x' = 0 ; x Y = XY ; x' Y = 0 ; x y = XY + x' y '; x ' y ' ≠ 0 ;...
End of L9. DIY: derive RANS equations from NS equations.

32
32
RANS equations
Numerical Fluid Mechanics • Reynolds Averaged Navier-Stokes (RANS) equations:
ρ⎜
⎛ ∂U i
+U j
∂U i ⎞⎟
=−
∂p
+
∂ ⎜μ ⎟+
(
⎛ ∂U i ⎞ ∂ − ρ ui ' u j ' )
⎜ ∂t ∂x j ⎟⎠ ∂xi ∂x j ⎜ ∂x j ⎟ ∂x j
⎝ ⎝ ⎠
• Same in form with NS equations
- The variables U i , p are now mean flow quantities
- One new term appears − ρ ui ' u j ' : called Reynolds stresses, to be modeled.
33
33
Turbulence modeling approaches

• Boussinesq approach
- isotropic
- based on dimensional analysis
• Reynolds stress transport models

- no assumption of isotropy
- contains more “physics”
- most complex and computationally expensive
34
34
Boussinesq approach (1)
• Relates the Reynolds stresses to the mean flow by a turbulent (eddy)

viscosity μt
Viscous stresses Turbulent / Reynolds stresses
⎛ ∂ui ∂u j ⎞ Analogy method ⎛ ∂u ∂u j ⎞

τ ij = τ ji = μ ⎜ + ⎟ − ρ ui ' u j ' = μ t ⎜ i + ⎟
⎜ ∂x j ∂xi ⎟ Boussinesq (1877) ⎜ ∂ x j ∂ xi ⎟
⎝ ⎠ ⎝ ⎠
- Assumptions valid at molecular level, not necessarily valid at macroscopic

level
- μt is a scalar
35
35
Boussinesq approach (2): Modeling μt
• All the models under Boussinesq approach focus on modeling μt

- Basic approach: Dimensional analysis / Analogy method
ν t = μt / ρ [m2/s] need (1) a velocity scale Vˆ ; (2) a length scale, l .
Molecular viscosity, ν Turbulent viscosity, ν t

(based on kinetic gas theory)
Analogy method ν t ∝ Vˆ × l
ν ≈ 12 C λ
Vˆ , l: velocity & length
C : mean molecular speed (=1.35Usound) scales, characterizing the
λ : mean free path of molecules large-scale turbulent motions
• Models classified in terms of number of transport equations solved, e.g.
- Zero-equation / One-equation / Two-equation model / ...
36
36
Boussinesq approach (3): Zero equation models
• Prandtl mixing length model

- assume a velocity scale, Vˆ = Vmix ∝ l mix 2Sij Sij
- require that the length scale l mix be prescribed

ν t = 12 Vmix l mix
- economical but too crude; need to estimate “right” l mix for different problems
37
37
Boussinesq approach (4): One equation models
• Traditionally, one-equation models are based on a transport equation

for k ≡ 1 (u ′ + v′ + w′ )
2 2 2
2
- then the velocity scale, Vˆ = k
- Use of transport equations allows ‘history effects’ to be accounted for.
• A very problem-dependent length scale must be specified algebraically

based on the mean flow.
38
38
Boussinesq approach (5): Two equation models
• Two transport equations are solved, giving two independent scales for
νt
calculating
- Virtually all use the transport equation for the turbulent kinetic energy k
- Several transport variables have been proposed, based on dimensional

arguments, and used for second transport equation, e.g.,
Kolmogorov, ω : νt ∝ k /ω
ω is specific dissipation rate, defined in terms of large
eddy scales that define supply rate of k .
Chou, ε : νt ∝ k2 /ε
- Boussinesq relation still used for Reynolds Stresses.
39
39
Boussinesq approach (5): Standard k − ε model

∂k ∂k ⎛ν ⎞
+ uj = ∇ ⋅ ⎜⎜ t ∇k ⎟⎟ + ℜ
12−3 ε Cε 1 = 1.44
∂t ∂x j ⎝ σk ⎠
{ 123 14243 source

transient Cε 2 = 1.92
convection diffusion
σ k = 1.0
∂ε ∂ε ⎛ νt ⎞ ℜ×ε ε2
+ uj ⎜
= ∇⋅⎜ ⎟
∇k ⎟ + Cε 1 − Cε 2 σ ε = 1. 3
∂t
{ ∂x j ⎝ σε ⎠ 1 4 4k42444 3k
transient 123 14243 source
convection diffusion
k 3/ 2
Vˆ = k 1/ 2 l= ν t = C μ × Vˆ × l = C μ k 2 / ε C μ = 0.09
ε
• Strengths
- Robust & economical; Reasonable accuracy for a wide range of flows
• Weaknesses
- overly diffusive for many situations (flows involving strong streamline
curvature, swirl, rotation, separating flows, low-Re flows)
40
40
Boussinesq approach (5):
Numerical Fluid Mechanics RNG & Realizable k − ε models
• RNG
- Improved performance over standard version for rapidly strained flows and
flows with streamline curvature;
• Realizable
- Performance generally exceeds the standard version; Good for complex flows
with large strain rates
• Both still suffers from the inherent limitations of an isotropic eddy-

viscosity model.
41
41
Boussinesq approach (6): Faults in assumption
• Boussinesq Rij = 2 μt Sij

- Is this simple linear relationship sufficient?

Rij is strongly dependent on flow conditions and history
Rij changes at rates not entirely related to mean flow processes
- Rij is not strictly aligned with Sij for flows with: (1) sudden changes in mean
strain rate; (2) extra rates of strain (e.g., strong streamline curvature); (3)
rotating fluids; (4) stress-induced secondary flows.
• Modifications to two-equation models cannot be generalized for

arbitrary flows.
42
42
Reynolds Stress Models (RSM)
• Starting point is the exact transport equations for the transport of
Reynolds stresses, Rij = − ρ ui′u ′j
• The transport equations can be obtained in this way,
ui′ NS (u j ) + u ′j NS (ui ) = 0
- The resulting equations contain several terms that must be modeled.
• Characteristics of RSM
- Effects of curvature, swirl, and rotation are directly accounted for in the
transport equations for the Reynolds stresses Æ When anisotropy of
turbulence significantly affects the mean flow, consider RSM;
- More CPU resources is needed;
- Strong coupling between Reynolds stresses and the mean flow.
43
43
Near-Wall Modeling
• Walls are main source of vorticity and turbulence;

• Accurate near-wall modeling is important for most engineering applications.
44
44
Near-Wall Modeling: (2) Important variables
· Important variables:
▫ Distance from the wall: y.

▫ Wall shear stress: τw. The force exerted on a flat plate is the area
times the wall shear stress.
▫ Density: ρ.
▫ Dynamic viscosity: μ.
▫ Kinematic viscosity: ν.
▫ Velocity at y: U.
▫ The friction velocity: uτ = (τw/ρ)1/2.
· We can define a Reynolds number based on the distance to the

wall using the friction velocity: y+ = yuτ/ν.
· We can also make the velocity at y dimensionless using the
friction velocity: u+ = U/ uτ.
45
45
Near-Wall Modeling: (3) BL structure

u+=y+
y+=1
46
46
Near-Wall Modeling: (4) Standard wall function
· The experimental boundary layer profile can be used to
calculate τw. However, this requires y+ for the cell adjacent to
the wall to be calculated iteratively.

· In order to save calculation time, the following explicit set of
correlations is usually solved instead:
ρ C 1μ/ 4k 1P/ 2 yP y * ≥ 11.225 ⇒ U ∗ = κ1 ln(Ey∗ ) U P C 1μ/ 4k 1P/ 2

y∗ = ⇒ ⇒τw =
μ y * < 11.225 ⇒ U ∗ = y∗ U∗ / ρ
▫ Up is the velocity in the center of the cell adjacent to the wall.

▫ yp is the distance between the wall and the cell center.
▫ kp is the turbulent kinetic energy in the cell center.
▫ κ is the von Karman constant (0.42).
▫ E is an empirical constant that depends on the roughness of the
walls (9.8 for smooth surfaces).
47
47
Near-Wall Modeling: (5) Momentum eqn

· The objective is to take the effects of the boundary layer correctly
into account without having to use a mesh that is so fine that the flow
pattern in the layer can be calculated explicitly.

· Using the no-slip boundary condition at wall, velocities at the nodes at
the wall equal those of the wall.
· The shear stress in the cell adjacent to the wall is calculated using the
correlations shown in the previous slide.
· This allows the first grid point to be placed away from the wall,
typically at 30 < y+ < 300, and the flow in the viscous sublayer and
buffer layer does not have to be resolved.
· This approach is called the “standard wall function” approach.
· The correlations shown in the previous slide are for steady state
(“equilibrium”) flow conditions. Improvements, “non-equilibrium wall
functions,” are available that can give improved predictions for flows
with strong separation and large adverse pressure gradients.
48
48
Near-Wall Modeling: (6) Two-layer zonal model
· A disadvantage of the wall-

function approach is that it
relies on empirical
correlations. Rey > 200
· The two-layer zonal model
does not. It is used for low-Re
flows or flows with complex Rey < 200
near-wall phenomena.
· Zones distinguished by a
wall-distance-based turbulent
Reynolds number:
ρ ky
Re y ≡ μ
· The flow pattern in the boundary layer is calculated explicitly.
· Regular turbulence models are used in the turbulent core region.
· Only k equation is solved in the viscosity-affected region.
· ε is computed using a correlation for the turbulent length scale.
·
49 Zoning is dynamic and solution adaptive.
49
Near-Wall Modeling: (7) Turbulence
· The turbulence structure in the boundary layer is highly

anisotropic.
· ε and k require special treatment at the walls.
· Furthermore, special turbulence models are available for the

low Reynolds number region in the boundary layer.
· These are aptly called “low Reynolds number” models.
· This is still a very active area of research, and we will not

discuss those here in detail.
50
50
Near-Wall Modeling: (8) Comparison
Numerical Fluid Mechanics Approach Strengths Weaknesses
Standard wall- Robust, economical, Empirically based on

functions reasonably accurate simple high- Re flows;
poor for low-Re
effects, massive
transpiration, PGs,
strong body forces,
highly 3D flows
Non-equilibrium Accounts for Poor for low-Re
wall-functions pressure gradient effects, massive
(PG) effects. transpiration (blowing,
Improved predictions suction), severe PGs,
for separation, strong body forces,
reattachment, highly 3D flows
impingement
Two-layer zonal Does not rely on Requires finer mesh
model empirical law-of-the- resolution and
wall relations, good therefore larger cpu
for complex flows, and memory
applicable to low-Re resources
flows
51
51
Near-Wall Modeling: (9) Grid guidelines
Wall Function Two-Layer Zonal

Approach Model Approach
· First grid point in log-law · First grid point at y+ ≈ 1.

region: +
30 ≤ y ≤ 300 · At least ten grid points within
· Gradual expansion in cell size buffer and sublayers.
away from the wall. · Better to use stretched
· Better to use stretched quad/hex cells for economy.
quad/hex cells for economy.
52
52
Near-Wall Modeling: (10) for improved solution
· When very accurate (say 2%) drag, lift, or torque predictions

are required, the boundary layer and flow separation require
accurate modeling.
· The following practices will improve prediction accuracy:
▫ Use boundary layer meshes consisting of quads, hexes, or prisms.
Avoid using pyramid or tetrahedral cells immediately adjacent to
the wall.
▫ After converging the tetrahedral
prism layer
solution, use the surface volume mesh
efficiently
integral reporting option to is generated resolves
check if y+ is in the right automatically
boundary layer
range, and if not refine the
grid using adaption.
▫ For best predictions use the
two-layer zonal model and triangular surface
completely resolve the flow mesh on car body is
in the whole boundary quick and easy to
layer. create
53
53
Summary of Fluent-included RANS-based models

54
54
Summary of Fluent-included RANS-based models
55
55
Discussion / DIY@home:
• derive RANS equations from NS equations;
• try to derive the transport equations for Reynolds stresses
• (if you have time) Computing using different RANS models on a fine
enough mesh, and comparing CFD solutions with measurements
(Driver and Seegmiller, 1985). (or visit
http://cfd.me.umist.ac.uk/ercoftac/ for such kinds of measurements)
Re H = 3.74 × 10 4
α = 0°
56
56

Numerical Fluid Mechanics: Chungen Yin, PH.D

Uploaded by

Copyright:

Available Formats

Numerical Fluid Mechanics: Chungen Yin, PH.D

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

Numerical Fluid Mechanics: Chungen Yin, PH.D

Uploaded by

Copyright:

Available Formats

Numerical Fluid Mechanics Numerical Fluid Mechanics

Chungen Yin, Ph.D.

Pontoppidanstræde 101, room 8

Lecture 1 Introduction to CFD

Lecture 2 Finite volume method

Lecture 3 Finite volume method (cont)

Lecture 4 Solvers and algorithm

Lecture 5 Error/uncertainty; Discretization vs. CFD accuracy

Lecture 6 Advanced meshing: Tips & tricks

Lecture 7 Transient problems

Lecture 9 Heat transfer

Lecture 10 Project-related topic

• Turbulence (energy cascade, scales...)

• RANS modeling & Near-wall modeling

· Effect of turbulence on N-S eq.

Laminar flow vs. Turbulent flow: Animation1

Turbulent, high Reynolds

Laminar, low Reynolds

Transition in a jet flow

Small Structure Large Structure

Large-scale vs. small-scale structures

External flows: ρUL

The Re number thresholds for different flows

· One characteristic of turbulent flows is their irregularity or randomness. A full

Turbulence: (3) diffusivity

· The diffusivity of turbulence causes rapid mixing and increased rates of

Turbulence: (5) rotation and vorticity

Mechanisms such as the stretching of three-dimensional vortices play a key

What’s turbulence — Summary (cont)

▫ Turbulent flows are rotational; that is, they have non-zero

vortices play a key role in turbulence.

Merging of turbulent spots and transition to turbulence

Turbulent boundary layer (cont)

Transition in boundary layer flow

T-S waves Turbulent spots Fully turbulent flow

• Larger, higher-energy eddies, transfer energy to smaller eddies via

· Existence of eddies implies rotation or vorticity, which is concentrated

Smallest Scales of Turbulence

k ≡ 1 (u ′2 + v′2 + w′2 ) is specific turbulent kinetic energy [m2/s2];

• Motion at smallest scales depends on dissipation rate, ε , and kinematic

• From dimensional analysis,

- Assume l is characteristic of larger eddy size;

- Dimensional analysis to estimate order of large eddy supply rate of k as

• Since ε ~ k / τ turnover : ε ~ k3/ 2 / l or l ~ k3/ 2 / ε

Energy spectrum vs. Wave number (log-log scales)

k wave number ( κ ) l integral scale η Kolmogorov scale

(2D channel flow)

For turbulent flows:

DNS (Direct Numerical Simulation)

• “DNS” is the solution of the time-dependent Navier-Stokes equations

- Numerical time step size required Δt ~ τ

- DNS is not suitable for practical industrial CFD

• Two methods can be used to eliminate need to resolve small scales:

- Reynolds Averaging: Transport equations for mean flow quantities are

- Filtering (LES): Transport equations for ‘resolvable scales’; Resolves larger

• Both methods introduce additional terms that must be modeled for

Numerical Fluid Mechanics

RANS modeling – Ensemble (time) Averaging

• Consider a point in a given flow field,