COMPUTATIONAL FLUID DYNAMICS

LECTURE 5: TURBULENCE MODELING

Sekhar Majumdar

Professor, Department of Mechanical Engineering, NMIT Bangalore

Former Head, CTFD Division, NAL, Bangalore, India

7th Semester (July- December 2019)

Department of Mechanical Engineering
Nitte Meenakshi Institute of Technology, Yelahanka, Bangalore
 TOPICS TO COVER

 PHENOMENON OF TURBULENCE

 CFD AND TURBULENCE MODEL

 TURBULENCE PROBLEM

 LIMITATIONS OF DNS METHODOLOGY

 CONCEPT OF STATISTICAL AVERAGING

 FOUNDATION OF TURBULENCE MODELING

 CLASSIFICATION OF MODELS

 LINEAR EDDY VISCOSITY HYPOTHESIS

 LINEAR EDDY VISCOSITY (LEV) BASED ALGEBRAIC MODELS

MODEL FOR TURBULENT FLOW SYSTEM
Level of Reality :

Concerns the scale of reality considered

Molecular level …… Boltzmann Equations
Continuum Level …. Navier Stokes Equation for Newtonian fluid

Level of Space-Time resolution :

Space and time scales characteristics of the system evolution

Level of Dynamic Description

Relative importance of various forces exerted on the system

Potential Flow / Euler Equations / NS Equations

SM_MNNIT_CFD-2016
 MODEL FOR TURBULENT FLOW SYSTEM
Decomposition of the Energy Spectrum

Scale Separation problem :

Dynamics of all the scales of the

exact solution need not always be Steady RANS
solved directly

Equations couple the interaction

among all the scales. Models can
resolve scales up to certain resolution Unsteady RANS
and the interaction of finer unresolved
scales have to be considered in
statistical terms
Ref: LES for incompressible flows, P. Sagaut (2000)

Large Eddy Simulation

SM_MNNIT_CFD-2016
 REYNOLDS AVERAGING LEADS TO LOSS OF INFO
The challenge of
description lies
in the complexity
Jets

Instantaneous After Reynolds Averaging

Plumes
Important Interaction
info lost due to
Reynolds Averaging

5
(Durbin P.A. and Pettersson Reif. B.A.,2001)
 PHENOMENON OF TURBULENCE
The Fact of Turbulent Flow has been described in terms of sequence of events by
Professor B.E. Launder (1991) as following

 At moderate Reynolds number, the restraining effects of viscosity are too weak to
prevent small, random disturbances in a shear flow from amplifying

 The disturbances grow, become non-linear, interact with neighbouring disturbances

 This mutual interaction leads to a tangling of vorticity filaments

 Eventually the flow reaches a chaotic, non-repeating form describable only in

statistical terms

Grid Turbulence Turbulent jet (Re=20,000)

 CFD AND TURBULENCE MODEL
Computational Fluid Dynamics (CFD)
• A powerful practical tool for analyses of flow in engineering applications
• Most of the practical flows of industrial interest are TURBULENT
Three major elements of CFD analysis
• Grid Generation
• Flow solution Algorithm
• Turbulence Modelling
"… Turbulence still remains the chief outstanding difficulty of our
subject " points out Bradshaw (Experiments in Fluids, 1994)
Physics of Turbulence too complex
Random/Three-dimensional / Rotational / Time-dependent/
Wide range of scales/ Strong Non-Linear interactions/ Dissipative

Mathematical Difficulties
Navier Stokes Equations do contain all the physics of Turbulent flow.
But even Supercomputers inadequate to resolve all the length and time scales
 THE TURBULENCE PROBLEM

Turbulence manifests itself as a complex agglomeration of vortices

(eddies) of different size and strength varying over a wide spectrum

How can such complexity be fathomed?

Analyses of turbulent flow inevitably invoke a statistical description

Statistical Theory is a way to fathom the complexity of turbulence

Mathematical Modeling is a way to predict turbulent flows using the

statistics of the detailed and rapidly varying irregular flow information
 MAJOR FEATURES OF TURBULENT FLOW

Highly random, irregular with wide range of scales and Intermittency

Mixing intensive with high exchange rates

Unsteady and three dimensional process

A non-linear process due to advection; correlations responsible for energy

exchange between different scales of motion

Rotational character - consists of multiple vortex elements of different size

and strength maintained by the kinetic energy of the mean flow

Dissipative nature - Turbulence is generally damped by the fluid viscosity.

Viscous forces effective on the smallest eddies eventually transform kinetic
energy of turbulent motion into internal (heat) energy
LIMITATION OF DIRECT NUMERICAL SIMULATION
Navier Stokes Equation in Cartesian coordinate system

( U j )
Continuity : 0
x j
Momentum :  ( U i )  P
 ( U iU j )     2U i
t x j xi

Four equations for four unknowns (U1, U2, U3 and P) form a closed set

Eddy size may vary from D/Re0.75 to D or even larger where

Re=Reynolds number and D = Characteristic Dimension

Problem of resolution of wide range of time and length scales –

prohibitive disc space and computation time

A simple channel flow problem at Re=106 may need billions of grid

points and perhaps millions of time steps. This may take even years
of computational time even on modern supercomputer
DIFFERENT APPROACHES TO COMPUTE TURBULENT FLOW
Methodologies

Direct Numerical Simulation

Large Eddy Simulation

Discrete Vortex Method

Eddy Viscosity Based Turbulence Models

Reynolds Stress Transport Equation Based Turbulence Models

Desirable Features

Great Width of Applicability

Accuracy

Mathematically simple

Computationally Stable
 CONCEPT OF STATISTICAL AVERAGING
FOR TURBULENT FLOW

Stationary flow Non Stationary flow

Time scales of ensemble averaged

flow are widely separate from the
scales from turbulent fluctuations
1T
     where   lim  dt
T  T 0 ~
      
Time Averaged Coherent Stochastic
   
Ensemble Averaged Stochastic
 FOUNDATION OF ENGINEERING TURBULENCE MODELS
Concept of Reynolds Averaging (Reynolds (1895))
Reynold 's Decomposition of any flow variable gives :

    
Instantane ous Time Averaged Turbulent
Substitution of instantaneous flow variables in Momentum transport equation
by the Reynolds decomposition and averaged over time cycle gives :


U i     (U i  ui )   U i    ui    U i 
t t t t t
zero Averaged


xi
 
U iU j 

xi
 
 (U i  ui )(U j  u j ) 

xi

U iU j  

xi

U i u j 

xi

U j ui  
xi

ui u j  
zero zero
  ρU U     ρu u 
After Time Averaging   i j  
x   x  i j 
i i
Time Averaged
 LINEAR EDDY VISCOSITY (LEV) HYPOTHESIS
(Boussinesq (1877))
Analogy between turbulent transport and molecular transport in laminar flow
Reynolds Stress Tensor proportional to Mean Strain Rate Tensor
Molecular Transport
 U i U j  2 U k
 ij        ij
 x j xi  3 xk
Shear Stress Laminar Viscosity Mean Strain Rate
(Physcial fluid property)
Turbulent Transport
 U i U j  2
  ui u j  t     k ij
 x j xi  3
Reynolds Stress E ddy Viscosity Mean Strain Rate
(Local Turbulent property)
2
k ij ensures the sum of normal Reynolds stress components as 2k
3
Dimensional Analogy gives
Laminar :   (Density)  (Mean Free Path)  (Molecular Velocity)

Turbulent : t  (Density)  (Length Scale)  ( Velocity Scale)

FOUNDATION OF ENGINEERING TURBULENCE MODELS

Momentum Transport Equation in Reynolds Averaged form


  Ui

 
 UiU j  
P
  2U i 


  ui u j 
t x j xi x j
Reynolds Stress Tensor

Scalar Transport Equation in Reynolds Averaged form

    
  
U  S  F   u  
t x x
j  j j
j j

Turbulent Scalar Transport

Turbulence Model expresses the Reynolds stress and/or

Turbulence Scalar Transport in terms of known flow quantities
and close the equation system for solution
FOUNDATION OF ENGINEERING TURBULENCE MODELS
EXACT TRANSPORT EQUATION FOR REYNOLDS STRESS TENSOR

Equation of motion for instantaneous fluctuating velocity component ui

ui  1 p   2ui 

t xk
ui U k  uk U i  ui uk    
 xi  xk xk xk
uiuk   0
Multiply the equation for ui by u j and for u j by ui
add and time average the sum
Convection Diffusion

 uiu j     U k uiu j      uiu juk  pui jk  p u j ik     uiu j 
t xk xk  xk 
U j U i  ui u j  ui u j
  ui uk   u j uk  p    2  x x
xk xk 
 kx  xk k k
Production Pressure Strain Dissipation
Closure at second moment level needs Diffusion, Pressure Strain
Correlation and Dissipation to be modeled

Practical application problems need more simple, cost effective but

reasonably accurate methodology
CLASSIFICATION OF TURBULENCE MODELS USED FOR CFD
Linear Eddy Viscosity based Models (LEVM)

Zero Equation (Algebraic) Model

Turbulence Length and Velocity scales prescribed algebraically

One Equation Model

One Transport equation solved for Velocity scale and Length scale prescribed

Two Equation Model

Two transport equations solved - one for Velocity scale one for Length scale

Non Linear Eddy Viscosity based Models (NLEVM)

Direct or indirect algebraic expression between Reynolds

Stress and the Mean Strain Rate tensor

Reynolds Stress Transport equation-based Models (RSTM)

Solves six transport equations for six components

of symmetric Reynolds Stress Tensor
 LINEAR EDDY VISCOSITY BASED ALGEBRAIC MODELS
Prandtl Mixing Length Model (1925)
U
t   lm lm
y
Mixing Length
(Velocity Scale)
(Length Scale)
lm = d,  = Closure Coefficient
d = Characteristic Layer Thickness
Far Wake :  = 0.180
Plane jet :  = 0.098
Mixing Layer :  = 0.071
Round Jet :  = 0.080

Reasonably good for Thin Shear Flows.

Ref: Wilcox (1993)

Advantages : Simplicity and computation economy
Disadvantages : Insufficient generality, difficult prescription of mixing length for complex flows
LINEAR EDDY VISCOSITY BASED ALGEBRAIC MODELS

Typical Velocity Profile for a Turbulent Boundary Layer [Wilcox (1993)]

Wall coordinate definition : Logarithmic Law of Wall :

w
u   w  Wall Shear Stress, 1 1
 U  log e ( Ey  )  log e ( y  )  B
 
  Fluid Density
  Von Karman Constant  0.41
  Kinematic Fluid Viscosity
yu
and B  5.0
U   U / u ; y  

LINEAR EDDY VISCOSITY BASED ALGEBRAIC MODELS
Key Modifications of Prandtl’s Mixing Length Model (1956)
Van Driest near wall damping functions for the mixing length


lmix  y 1  e y

/ A0
 and A0  26
Improves the description of the Reynolds Stress near the wall. In the limit of small
near wall distance the Van Driest mixing length implies that the Reynolds Stress
goes to zero as y+ approaching the surface

Clauser's modification near the free-stream edge of the shear layer

T 0  U e *
where  is the eddy viscosity in the outer layer and  is the displacement
thickness, Ue is the velocity at the edge of the layer and  is a closure coefficient

Klebanoff's Intermittency factor at turbulent/non-turbulent interface

-1
  y 6 
FKleb  1     where  is the boundary layer thic kness
    
The flow is not always turbulent as one approaches the freestream from the boundary
layer. Eddy viscosity needs to be multiplied by an Intermittency Factor FKleb
TRANSPORT EQUATION FOR TURBULENT KINETIC ENERGY

Exact equation for k  0.5uiui  derived from the Navier Stokes equation
Convection Production
    k 1  Ui ui ui
( k )  ( U k )    
 x 2 k i j
u u u  puk   ik  
t xk xk  k  xk xk xk
Diffusion Dissipation ()

Analogy to laminar flow models diffusion term

  k 1   k

 x   u u u
k i j  pu k      t  k 
xk  k 2  x k xk

Dissipation term remains unaltered

Dissipation  determined from length scale (l) information used directly   k3 2 l
Modeled equation of k

    k     U i U j  2 
 U
( k )  ( U j k )  (   t  k )    t    k ij  i  
t x j x j  x j   
  x j xi  3  x j

 TRANSPORT EQUATION FOR TURBULENT DISSIPATION
Exact equation for  derived from the Navier Stokes equation
Convection Diffusion Production
   U j    ui ui P u j   U i  ui u j uk uk 
   u j  2    2   
t x j x j  xk xk xk xk x j
 x j 
 xk xk xi x j 

2
ui  2U i ui uki ui   2ui 
 2  uk  2  2 
2

x j xk x j xk x j x j  xk x j 
Production Dissipation
Analogy to laminar flow models diffusion term like k equation
 
 Production of   C 1 Production of k 
Analogy to Model k equation  k
 
Destructio n of   C 2 Destructio n of k    C 2 
Modeled equation of   k k
   
  

(  )  ( U j  )   (      ) 
t x j x j 
t
 x j 

   U i U j  2 
 U i 2
 C 1 t   k ij   C 2 
 
Closure constants   x j
k xi  3  x j
 k

t  C k 2  C  0.09 C 1  1.44 C 2  1.92  k  1.0    1.3

LEV BASED TWO-EQUATION TURBULENCE MODELS
Standard k- model (Jones and Launder(1972))
Two equation model, "… the simplest complete model of turbulence" is more or less
the foundation for most of the turbulence model research during the last three
decades
Two transport equations solved to determine the length scale l  Cl k 3 / 2 / 
and the velocity scale k
Most widely used popular two-equation models are the k- and k- models

   k 

 k    U j k   t  k       ij Ui  

t x j x j 
 x j 
 x j

       U i 2
      U j   t        C1  ij
 
 C 2 
t x j x j 
 x j
 k x j k

where t   C k 2 

Good compromise between universality and economy for engineering problems

Needs near wall modification, insensitive to freestream boundary conditions
Poor performance for flows with sudden change of strain rate - curvature, swirl,
rotation, separated flows (1981 AFOSR-HTTM Stanford Olympics II)
 LEV BASED TWO-EQUATION TURBULENCE MODELS
Standard k- model (Wilcox, 1993)

Equation for  = Specific Dissipation of Turbulence Energy =  / k for which the

Dimension is 1/time or frequency ( per second or Hertz)
First proposed by Kolmogorov (1942), followed later by Saffman(1970) and Spalding
(1972) and finally by Wilcox(1988) in his state-of-the-art formulation
 k 

t

x j

 k     U j k  
x j

 
 t k    

 ij
x j 

U i
x j
  *k

 

  

t x j

      U j  
x j
  
 t      
x j 
 
 U i
 ij
x j
  2
 k
where t   k    5/9,   3/40,  *  0.09 ,  k   2.

Equation is valid even in the laminar sublayer and near wall modification not obligatory
Poor performance for flows with sudden change of strain rates. Overpredicts the shear
stress in adverse pressure gradient boundary layers. Spurious sensitivity to free
stream conditions. Unreliable for flow with detached free shear layers
NEAR WALL MODIFICATIONS FOR TWO-EQUATION MODELS
Physical effects of the presence of wall surface in turbulent flow

Molecular viscosity diffuses vorticity and damps turbulence. Viscous

diffusion terms become one of the largest terms to be balanced by the
other terms in the Reynolds Stress transport equation

Inviscid blocking of wall-normal fluctuation leads to near-wall anisotropy

and has an effect at considerably greater distances from the wall.
Isotropic eddy-viscosity based models can not separate this second effect
from viscous damping

Standard k-  model does not automatically allow the eddy viscosity to tend to
zero as the wall is approached. Further assumption of isotropy of eddy viscosity
does not allow the prediction of inviscid effects as near wall anisotropy
NEAR WALL MODIFICATIONS FOR TWO-EQUATION MODELS
Three different ad-hoc near-wall modeling approaches

Standard Wall Function method based on Logarithmic Law of Wall

(Launder-Spalding(1972, 1974),Wilcox (1993))

Strictly valid for zero pressure-gradient flows only ; but

computationally economic for engineering CFD calculation of the
mean flow properties
NEAR WALL MODIFICATIONS FOR TWO-EQUATION MODELS
Damping functions based on asymptotic consistency for normal-to-wall variation of flow
variables (Patel et al(1985) , Rodi(1991), Shih-Mansour(1990) )
Exponential Functions chosen to damp the eddy viscosity and the dissipation source terms
to tune to the DNS / measurement data for simple boundary layer flows
U  ( y ) k  ( y 2 )   (1)
t  ( y 3 ) uv  ( y 3 ) where y is the near wall distance

Turbulence Kinetic Energy Eddy Viscosity profile

profile for channel flow for channel flow

Requirement of fine near wall resolution expensive and often causes numerical instability
NEAR WALL MODIFICATIONS FOR TWO-EQUATION MODELS

Zonal or Multi-layer approach (Chieng-Launder(1980), Amano(1984), Rodi(1991),

Chen-Patel(1988), Lien- Leschziner(1993))

To account for the different scales near wall, uses algebraic prescription of eddy
viscosity and length scale ( hence dissipation) inside layer 1 and solves for
normal k and  in layer 2

Still empirical but economic; easy handling of surface roughness effects

Layer interface matching often causes numerical instability
STIFFNESS OF THE TURBULENCE TRANSPORT EQUATIONS
Demonstration of Stiffness of an ordinary differential equation

Solution of the equation y  100 y is y(t )  Ae10t  Be10t

10 t
if y( 0 )  1 and y (0)  10 the analytical solution is yexact (t )  e
10t
and numerical solution ends up as ynum (t )  e   e10t where   1 ; but
the roundoff and truncation error can excite the second factor leading to
divergence or non-physical values of y

Turbulence scalar equations are often source – dominated

Three different time scales for propagation of boundary conditions

Wave propagation (convection) – L U a

Molecular diffusion scale – L2 

Dissipation time scale – k 
Major cause of stiffness is that the dissipation time scale is often widely
different from the other two scales
BOUNDARY CONDITIONS FOR TURBULENCE SCALARS

Inflow : k, ,  specified from measurement data or kin  f1 0.5U in  2


and  in or  in computed from the relationship; tin  f 2 la minar

f1 depends on the freestream turbulence level

and f 2  1 to 100

Outflow : Streamwise derivatives are set to zero

Symmetry : Normal derivatives of variables and normal velocities are set to zero

Wall :

Standard wall function recommended for relatively coarse grid 30  y nw  100

For fine resolution of near wall zone, integration possible upto the wall node using
damping functions &  wall or  wall are computed at the first near wall grid node

2k 6
 wall  2  wall 
y Cw 2 y 2
 CONCLUDING REMARKS
Turbulence Models are required for estimating the statistical second moment
correlations appearing in the Reynolds-Averaged Navier Stokes Equation

Two important parameters identified to characterise the turbulence effects on the

mean flow are the Length Scale and the Velocity Scale of turbulence

The k- model is the most widely used two-equation model for engineering application
problems. However it is inaccurate for flows with adverse pressure gradient and the
model does not allow integration of the conservation equations through the viscous
sublayer where low Reynolds number corrections are usually recommended

The k-  model is claimed to be accurate for flow with variable pressure gradient. One
weakness of this model is its sensitivity to the free stream boundary conditions for
free-shear flows

Any two-equation model based on the concept of isotropic eddy viscosity, however
fails in flow situations where anisotropy in the normal Reynolds stress components is
significant due to sudden change in mean strain rate