MODELS OF TURBULENCE

• Mixing Length Models

• One Equation Models Isotropic Turbulence
• Two-Equation Models

• Reynolds Stress Model Anisotropic Effects

• Large Eddy Simulation (LES)
• Direct Numerical Simulation (DNS)

In a turbulent flow, eddies of various sizes exist. For example, in a typical lab scale
experiment, eddies with a size range of microns to centimetres may be observed.
These eddies cause fluctuations, which have time constants of the order of µs to seconds.

1. DIRECT NUMERICAL SIMULATION (DNS)

In this approach, the idea is to capture the flow effects of all the eddies faithfully, by
employing grid sizes < µs. For any practical problem, the number of grid points required may
be of the order of 1012 and several million time steps may be required. Meeting such a heavy
computational demand is not feasible in many institutions. Extremely fast and powerful
parallel computers are needed to perform DNS.

2. LARGE EDDY SIMULATION (LES)

This approach divides the eddies into two categories

large which need to be precisely modelled and small which can
be approximately accounted for. A moderately dense grid (few
million points) is selected which captures large eddies exactly.
For eddies with size less than grid size, approximate sub-grid
scale models are evoked.

3. REYNOLDS STRESS MODEL (RMS)

Transport equations are derived for the Reynolds stress and for turbulent fluxes.

1
Reynolds stresses:

− ρ u ′ 2 , − ρ v ′ 2 , − ρ w′ 2

− ρ u ′ v ′, − ρ u ′w′, − ρ v ′w′
Turbulent Fluxes:
(For Heat Transfer)

− ρC p u ′T ′, − ρC p v ′T ′, − ρC p w′T ′

For each of the above quantities, a transport equation can be obtained. The RMS approach
involves 6 Additional differential equations to be solved for the 6 Reynolds stresses and
more equations for the turbulent fluxes. Although it involves more computations, this model
can capture the anisotropic effects in the turbulent flow. Grid pints used are not as large as
in the cases of DNS and LES. Hence it is computationally more economical than DNS and
LES.

LOWER ORDER MODELS

In these models, the fluctuations due to eddies are not precisely simulated. The Stresses
and Fluxes are represented as:

 ∂u ∂v 
τ xy = − µ  + 
 ∂y ∂x 
∂T
q y = −k etc.
∂x

where
µ = µl + µt µl → Laminar Viscosity µt → Turbulent or Eddy Viscosity

k = kl + kt kl → Laminar Conductivity kt → Turbulent Conductivity

µt, kt arise because of mixing enhancement by eddies. While µl, kl depend on the type of
fluid (and the local temperature, concentration etc.) µt, kt depend on eddy properties.
They vary significantly from location to location

2
 Turbulent BL

In Laminar Sublayer, µt << µl.

Laminar sub layer
Far away from wall, µt >> µl.
In between, µ ≈ µt.

MIXING LENGTH MODEL (PRANDTL)

Turbulent Diffusivity, νt = µt/s. In analolgy with kinetic theory of molecules, which gives
ν l ≅ λ ⋅ v rms , where λ → mean free path, and vrms → root mean square velocity of molecules,
Prandtl Proposed ν t ~ l m ⋅ v ′ , where and v’ the (Root Mean Square) fluctuation velocity of

an eddy and mixing length of an eddy lm is the average distance travelled by an eddy before
it gives up its momentum completely by mixing with main flow.

∂u ∂u
Prandtl Proposed v ′ = l m ⋅ ⇒ ν t = l m ⋅ v ′ = l m2 .
∂y ∂y
Proposing that τ (Shear Stress) ≈ constant and lm α y, Prandtl derived the logarithmic law
2
∂u  ∂u 
τ = µt = const. ⇒ y 2   = const.
∂y  ∂y 

u = c1 ln y + c 2

The logarithmic profile is valid away from wall where νt >> νl. Very close to wall, µ ≈ µl =
const.

3
 ∂u ∂u
µl = const. ⇒ = const.
∂y ∂y

u =c y

ONE EQUATION MODEL

u ′ 2 + v ′ 2 + w′ 2
Define the kinetic energy of the fluctuating motion as k = . If we take the
2
velocity fluctuation, v ′ = k , then ν t = l m k . For kinetic energy of turbulence, a transport
equation is derived in the form

TURBULENT KINETIC ENERGY EQUATION

 
 
∂k  ∂k ∂k ∂k 
+ u +v +w
∂t 
, ∂x ∂y ∂z 
storage



 convective  
 transport of k 
∂  v ∂k  ∂  vt ∂k  ∂  vt ∂k 
= v l + t  + vl +  + vl + + P + Dk
∂x  σ k ∂x  ∂y  σ k ∂y  ∂z  ,k
σ k ∂z  rate of production ,




















 of k rate of dissipatio n of k
diffusive transport of k

The production and dissipation terms are suitably formulated. The figure shows the flow in a
convergent-divergent nozzle.

Nozzle Diffuser
(k decreases and main flow (k increases, main
kinetic energy increases) flow decelerates)

4
k - ε TWO EQUATION MODEL

In addition to the k-equation, a transport equation is derived for the turbulent

dissipation rate, ε. Dimensionally, ε ~ k/to where t0 = turbulent time scale, Let l0 = turbulent
length scale and v0 = turbulent velocity scale
l 02
ν t ~ v0 l0 or
t0

u
t0
l0 t

l 02
ε → k/t0., vo → k , vt → = v02 ⋅ t 0 ., vt → k 2 / ε
t0

k2
vt = c µ ⋅
ε

The transport equation for ε is also similar to that of k. it has a form

TURBULENT DISSIPATION RATE EQUATION

 
 
∂ε  ∂ε ∂ε ∂ε 
+ u +v +w
,∂t  ∂x ∂y ∂z 
storage
L O M ON
 convective transport 
 
∂  vt ∂ε  ∂  vt ∂ε  ∂  vt ∂ε 
= vl +  + vl +  + v l + + Pε + Dε
∂x  σ k ∂x  ∂y  σ ε ∂y  ∂z  σ ε ∂z  rate of , ,
LOOOOOOOOO OMOOOOOOOOOO N production rate of dissipatio n
diffusive transport

νt k 3/ 2
In the k – transport equation, the rate of destruction is ~ − ⋅ k = −C D = −ε . The
l 02 l0
production due to viscous dissipation of mean flow KE is given as (in Cartesian Tensor
Notation):

5
  ∂u ∂u j  ∂u i
P
,k = νt  i + 
 ∂x j ∂xi 
 ∂x j
production rate of turbulent kinetic energy





viscous dissipation of main flow kinetic energy

For the ε equation

ε  ∂u ∂uj  ∂u i ε ε2
Pε = Cε 1 ⋅ ⋅ν t  i +  ; Dε = = Cε 2 .
∂x  ∂x
k  j ∂xi  j t 0 k

In summary, we have

 
∂k ∂k ∂  ν t ∂k   ∂u ∂u  ∂u
j
+ ui = ν l +  +ν t  +  i − ε,
i

∂t ∂xi ∂xi  σ k ∂xi  ∂

 jx ∂x  ∂x
 LOM ONi  j Dk
 Pk 

 
∂ε ∂ε ∂  ν t ∂ε  ε  ∂u ∂u  ∂u ε2
j
+ ui = ν l +  + Cε 1 ⋅ν t  i +  i − Cε 2
∂t ∂xi ∂xi  σ ε ∂xi  k  ∂x j ∂xi  ∂x j LMN k
 LOM ON  Dε
 Pε 

The model has five model constants: σk, σε, Cε1, Cε2, Cµ. The k - ε model is an equilibrium,
anisotropic model. Equilibrium between convective/diffusive transports and rates of
production/destruction is considered. For the above reason it is not valid very close to the
wall, initial parts of jets/shear layers/ wakes etc., where non-equilibrium and anisotropic
effects may be important. The model is also not valid for problems with strong streamline
curvature. Close to wall, (sublayer laminar) the predictions of k-ε model are matched with the
universal velocity profile obtained from the mixing length model. This is known as the
standard wall function approach. In addition to these, at inlet of the flow domain, the inlet
turbulence level has to be specified.