UNIT V TURBULENCE MODELS AND MESH GENERATION

Turbulence models, mixing length model, Two equation (k-Є) models – High and
low Reynolds number models – Structured Grid generation – Unstructured Grid
generation – Mesh refinement –Adaptive mesh – Software tools.

TURBULENCE MODELS

Reynolds decomposition

The random nature of a turbulent flow precludes an economical description of the motion of all
the fluid particles. Instead the velocity in above graph is decomposed into a steady mean value U
with a fluctuating component u′(t) superimposed on it: u(t) = U + u′(t). This is called the Reynolds
decomposition

 . A turbulent flow can now be characterized in terms of the mean values of flow properties
(U, V, W, P etc.) and some statistical properties of their fluctuations (u′, v′, w′, p′ etc.)
 Even in flows where the mean velocities and pressures vary in only one or two space
dimensions, turbulent fluctuations always have a three dimensional spatial character.
Effective mixing
Particles of fluid which are initially separated by a long distance can be brought close together by
the eddying motions in turbulent flows. As a consequence, heat, mass and momentum are very
effectively exchanged.
For example, a streak of dye which is introduced at a point in a turbulent flow will rapidly
break up and be dispersed right across the flow. Such effective mixing gives rise to high values of
diffusion coefficients for mass, momentum and heat.

Vortex stretching
 The largest turbulent eddies interact with and extract energy from the mean flow by process
called vortex stretching
 The presence of mean velocity gradients in sheared flows distorts the rotational turbulent
eddies. Suitably aligned eddies are stretched because one end is forced to move faster than the
other.

Energy Cascade
 Smaller eddies are themselves stretched strongly by somewhat larger eddies and more weakly
with the mean flow. In this way the kinetic energy is handed down from large eddies to
progressively smaller and smaller eddies in what is termed the energy cascade.

Kolmogorov Microscales
 The smallest scales of motion in a turbulent flow (lengths of 0.1 to 0.01 mm and frequencies
around 10 kHz in typical turbulent engineering flows) are dominated by viscous effects.
 The Reynolds number Reη of the smallest eddies based on their characteristic velocity υ and
characteristic length η is equal to 1, Reη = υη/ν = 1, so the smallest scales present in a turbulent
flow are those for which the inertia and viscous effects are of equal strength. These scales are
named the Kolmogorov microscales
 At these scales work is performed against the action of viscous stresses, so that the energy
associated with small-scale eddy motions is dissipated and converted into thermal internal
energy. This dissipation results in increased energy losses associated with turbulent flows.
 Dimensional analysis can be used to obtain ratios of the length, time and velocity scales of
the small and large eddies.
 The Kolmogorov microscales can be expressed in terms of the rate of energy dissipation of a
turbulent flow and the fluid viscosity, which uses the notion that in every turbulent flow the
rate of production of turbulent energy has to be broadly in balance with its rate of dissipation
to prevent unlimited growth of turbulence energy.

Transition from laminar to turbulent flow

Jet flow: an example of a flow with a point of inflexion

 Flows which possess one or more points of inflexion amplify long wavelength disturbances at
all Reynolds numbers typically above about 10.
 After the flow emerges from the orifice the laminar exit flow produces the rolling up of a vortex
fairly close to the orifice. Subsequent amplification involves the formation of a single vortex
of greater strength through the pairing of vortices.
 A short distance further downstream, three-dimensional disturbances cause the vortices to
become heavily distorted and less distinct. The flow breaks down, generating a large number
of small-scale eddies, and the flow undergoes rapid transition to the fully turbulent regime.
Mixing layers and wakes behind bluff bodies exhibit a similar sequence of events, leading to
transition and turbulent flow.

Characteristics of simple turbulent flows

The characteristics of some simple two-dimensional incompressible turbulent flows with constant
imposed pressure. The following flows will be considered here:
Free turbulent flows
• mixing layer
• jet
• wake
Boundary layers near solid walls
• flat plate boundary layer
• pipe flow
Free turbulent flows
 Among the simplest flows of significant engineering importance are those in the category of
free turbulent flows:
o Mixing Layer
o jets and
o Wakes.

 A mixing layer forms at the interface of two regions: one with fast and the other with slow
moving fluid. In a jet a region of high-speed flow is completely surrounded by stationary fluid.
 A wake is formed behind an object in a flow, so here a slow moving region is surrounded by
fast moving fluid. Figure is a sketch of the development of the mean velocity distribution in
the stream wise direction for these free turbulent flows.
 It is clear that velocity changes across an initially thin layer are important in all three flows.
Transition to turbulence occurs after a very short distance in the flow direction from the point
where the different streams initially meet; the turbulence causes vigorous mixing of adjacent
fluid layers and rapid widening of the region across which the velocity changes take place.
 It is also immediately clear that the turbulent part of the flow contains a wide range of length
scales. Large eddies with a size comparable to the width across the flow are occurring alongside
eddies of very small size.
 The visualization correctly suggests that the flow inside the jet region is fully turbulent, but the
flow in the outer region far away from the jet is smooth and largely unaffected by the
turbulence. The position of the edge of the turbulent zone is determined by the (time-
dependent) passage of individual large eddies. Close to the edge these will occasionally
penetrate into the surrounding region.
 In these formulae Umax and Umin represent the maximum and minimum mean velocity at a
distance x downstream of the source. Hence, if these local mean velocity scales are chosen and
x is large enough, the functions f, g and h are independent of distance x in the flow direction.
Such flows are called self-preserving.

Law of the wall

Close to the wall the flow is influenced by viscous effects and does not depend on free
stream parameters. The mean flow velocity only depends on the distance y from the wall, fluid
density ρ and viscosity μ and the wall shear stress τw. So
U = f ( y, ρ, μ, τw)

Formula (3.16) is called the law of the wall and contains the definitions of two important
dimensionless groups, u+ and y+.

Note: that the appropriate velocity scale is uτ = τw/ρ, the so-called friction velocity.

Velocity-defect law.
Far away from the wall we expect the velocity at a point to be influenced by the retarding
effect of the wall through the value of the wall shear stress, but not by the viscosity itself. The
length scale appropriate to this region is the boundary layer thickness δ.
U = g(y, δ, ρ, τw)

In this region we have the most useful form emerges if we view the wall shear stress as the cause
of a velocity deficit Umax − U which decreases the closer we get to the edge of the boundary layer
or the pipe centre line. Thus
This formula is called the velocity-defect law.

Linear or viscous sub-layer ---(the fluid layer in contact with a smooth wall)
At the solid surface the fluid is stationary and turbulent eddying motions must also stop
very close to the wall and the behavior of the fluid closest to the wall is dominated by viscous
effects. This viscous sub-layer is in practice extremely thin ( y+ < 5) and we may assume that
the shear stress is approximately constant and equal to the wall shear stress τw throughout the
layer. Thus

After integration with respect to y and application of boundary condition U = 0 if y = 0, we

obtain a linear relationship between the mean velocity and the distance from the wall

After some simple algebra and making use of the definitions of u+ and y+ this leads to

Because of the linear relationship between velocity and distance from the wall the fluid layer
adjacent to the wall is also known as the linear sub-layer.

Log-law layer --- the turbulent region close to a smooth wall

 Outside the viscous sublayer (30 < y+ < 500) a region exists where viscous and turbulent effects
are both important.
 The shear stress τ varies slowly with distance from the wall, and within this inner region it is
assumed to be constant and equal to the wall shear stress.

Outer layer --- the inertia-dominated region far from the wall

 Experimental measurements show that the log-law is valid in the region 0.02 < y/δ < 0.2. For
larger values of y the velocity-defect law provides the correct form.
 In the overlap region the log-law and velocity defect law have to be equal.

where A is a constant. This velocity-defect law is often called the law of the wake.

The turbulent boundary layer adjacent to a solid surface is composed of two regions:

 The inner region: 10–20% of the total thickness of the wall layer; the shear stress is (almost)
constant and equal to the wall shear stress τw. Within this region there are three zones. In
order of increasing distance from the wall we have:
– The linear sub-layer: viscous stresses dominate the flow adjacent to surface
– The buffer layer: viscous and turbulent stresses are of similar magnitude
– The log-law layer: turbulent (Reynolds) stresses dominate.

 The outer region or law-of-the-wake layer: inertia-dominated core flow far from wall; free
from direct viscous effects.

Turbulence models
 A turbulence model is a computational procedure to close the system of mean flow equations.
 For most engineering applications it is unnecessary to resolve the details of the turbulent
fluctuations.

 Turbulence models allow the calculation of the mean flow without first calculating the full
time-dependent flow field.

 We only need to know how turbulence affected the mean flow.

 In particular we need expressions for the Reynolds stresses.

For a turbulence model to be useful it:
 must have wide applicability,
 be accurate,
 simple, and economical to run

TURBULENT FLOW CALCULATIONS

Turbulence causes the appearance in the flow of eddies with a wide range of length and time
scales that interact in a dynamically complex way. Given the importance of the avoidance or
promotion of turbulence in engineering applications, it is no surprise that a substantial amount
of research effort is dedicated to the development of numerical methods to capture the
important effects due to turbulence. The methods can be grouped into the following three
categories:

 Turbulence models for Reynolds-averaged Navier–Stokes

(RANS) equations: attention is focused on the mean flow and the effects of turbulence on
mean flow properties. Prior to the application of numerical methods the Navier–Stokes
equations are time averaged (or ensemble averaged in flows with time-dependent boundary
conditions). Extra terms appear in the time-averaged (or Reynolds averaged)flow equations
due to the interactions between various turbulent fluctuations. These extra terms are modelled
with classical turbulence models: among the best known ones are the k–ε model and the
Reynolds stress model. The computing resources required forreasonably accurate flow
computations are modest, so this approach has been the mainstay of engineering flow
calculations over the last three decades.

 Large eddy simulation: this is an intermediate form of turbulence calculations which tracks
the behaviour of the larger eddies. The method involves space filtering of the unsteady Navier–
Stokes equations prior to the computations, which passes the larger eddies and rejects the
smaller eddies. The effects on the resolved flow (mean flow plus large eddies) due to the
smallest, unresolved eddies are included by means of a so-called sub-grid scale model.
Unsteady flow equations must be solved, so the demands on computing resources in terms of
storage and volume of calculations are large, but (at the time of writing) this technique is
starting to address CFD problems with complex geometry.
 Direct numerical simulation (DNS): these simulations compute the mean flow and all
turbulent velocity fluctuations. The unsteady Navier–Stokes equations are solved on spatial
grids that are sufficiently fine that they can resolve the Kolmogorov length scales at which
energy dissipation takes place and with time steps sufficiently small to resolve the period of
the fastest fluctuations. These calculations are highly costly in terms of computing resources,
so the method is not used for industrial flow computations.

Boussinesq hypothesis
 Many turbulence models are based upon the Boussinesq hypothesis.
 It was experimentally observed that turbulence decays unless there is shear in
isothermal incompressible flows.
 Turbulence was found to increase as the mean rate of deformation increases.
 Boussinesq proposed in 1877 that the Reynolds stresses could be linked to the
mean rate of deformation.

 Using the suffix notation where i, j, and k denote the x-, y-, and z-directions respectively,
viscous stresses are given by:

 ui u j 
 ij   eij     
 x j xi
  

 Similarly, link Reynolds stresses to the mean rate of deformation:

 U i U j 
 ij    ui ' u j '  t   
 x j xi
 

Turbulent viscosity
 A new quantity appears: the turbulent viscosity t.

 U i U j 
 ij    ui ' u j '  t   
 x j xi
 

 Its unit is the same as that of the molecular viscosity: Pa.s.

 It is also called the eddy viscosity.

  We can also define a kinematic turbulent viscosity: t = t/. Its unit is m2/s.

 The turbulent viscosity is not homogeneous, i.e. it varies in space.

 It is, however, assumed to be isotropic. It is the same in all directions. This assumption is
valid for many flows, but not for all (e.g. flows with strong separation or swirl).

Reynolds-averaged Navier Stokes equations for incompressible flow

Turbulence models for Reynolds-averaged Navier–Stokes (RANS) equations:

 RSNS equation is focused on the mean flow and the effects of turbulence on mean flow
properties. Prior to the application of numerical methods the Navier–Stokes equations are time
averaged
 Extra terms appear in the time-averaged (or Reynolds averaged) flow equations due to the
interactions between various turbulent fluctuations. These extra terms are modelled with
classical turbulence models: among the best known ones are the k–ε model and the Reynolds
stress model. The computing resources required for reasonably accurate flow computations are
modest, so this approach has been the mainstay of engineering flow calculations over the last
three decades

 The vast majority of turbulent flow computations has been and for the foreseeable future will
continue to be carried out with procedures based on the Reynolds-averaged Navier–Stokes
(RANS) equations.
 In order to be able to compute turbulent flows with the RANS equations it is necessary to
develop turbulence models to predict the Reynolds stresses and the scalar transport terms and
close the system of mean flow equations.
The time-averaging operation is itself an integration. Thus, the order of time averaging and
summation, further integration and/or differentiation can be swapped or commuted, so this is called
the commutative property.
Since div and grad are both differentiations, the above rules can be extended to a fluctuating vector
quantity a = A + a′ and its combinations with a fluctuating scalar ϕ = Φ +φ′:

To start with we consider the instantaneous continuity and Navier–Stokes equations in a Cartesian
co-ordinate system so that the velocity vector u has x-component u, y-component v and z-
component w:

The effects of fluctuations on the mean flow using the Reynolds decomposition in equations (3.23)
and (3.24a–c) and replace the flow variables u (hence also u, v and w) and p by the sum of a mean
and fluctuating component. Thus
It is important to note that the terms (I), (II), (IV) and (V) in (3.26a–c) also appear in the
instantaneous equations (3.24a–c), but the process of time averaging has introduced new terms
(III) in the resulting time-average momentum equations.
The equation set (3.25) and (3.27a–c) is called the Reynolds-averaged Navier–Stokes equations.

The time-average transport equation for scalar φ is

Mixing length model

On dimensional grounds we assume the kinematic turbulent viscosity ν , which has dimensions m2/s, can tbe
expressed as a product of a turbulent velocity scale ϑ (m/s) and a turbulent length scale
(m). If one velocity scale
and one length scale suffice to describe the effects of turbulence, dimensional analysis yields

ν = Cϑ
(3.36)
t

where C is a dimensionless constant of proportionality. Of course the dynamic turbulent viscosity is given
by

µ = Cρϑ

t

Most of the kinetic energy of turbulence is contained in the largest eddies, and turbulence length scale
is therefore
characteristic of these eddies which interact with the mean flow. If we accept that there is a strong
connectionbetween the mean flow and the behaviour of the largest eddies we can attempt to link the characteristic
velocity scale of the eddies with the mean flow properties. This has been found to work well in simple two-
dimensional turbulent flows where the only significant Reynolds stress is τxy = τyx = −ρu′v′and the only
significant mean velocity gradient is ∂U/∂y. For such flows it is at least dimensionally correct to state that, if the
eddy length scale is ,

∂U
ϑ=c
(3.37)
∂y
where c is a dimensionless constant. The absolute value is taken to ensure that the velocity scale is always a
positive quantity irrespective of the sign of the velocity gradient.
Combining (3.36) and (3.37) and absorbing the two constants C and c into a new length scale
m we obtain
This is Prandtl’s mixing length model. Using formula (3.33) and noting that ∂U/∂y is the only significant
mean velocity gradient, the turbulent Reynolds stress is described by

∂U ∂U
τxy = τyx = −ρ u′v′ = ρ
m2 (3.39)
∂y ∂y

Turbulence is a function of the flow, and if the turbulence changes it is

necessary to account for this within the mixing length model by varying
m. For a substantial category of simple
turbulent flows, which includes the free turbulent flows and wall boundary layers discussed in section 3.4, the
turbulence structure is sufficiently simple that
m can be described by means of simple algebraic formulae. Some
examples are given in Table 3.2 (Rodi, 1980).

The mixing length model can also be used to predict turbulent transport of scalar quantities. The only
turbulent transport term which matters in the two-dimensional flows for which the mixing length is useful is
modelled as follows:

∂Φ
−ρ v′ϕ′ = Γt (3.40)
∂y

where Γt = µ /σ and µ = ρνt where ν is found from (3.38). Rodi (1980) recommended values for σ of 0.9 in near-
wall flows, 0.5 for jets and mixing layers and 0.7 in axisymmetric jets.

Table 3.2 Mixing lengths for two-dimensional turbulent flows

Flow Mixing length
m L

Mixing layer 0.07L Layer width

Jet 0.09L Jet half width
Wake 0.16L Wake half width
Axisymmetric jet 0.075L Jet half width
Boundary layer (∂p/∂x = 0)
viscous sub-layer and κ y [1 − exp(− y+/26)]
log-law layer ( y/L ≤ 0.22) Boundary layer
outer layer ( y/L ≥ 0.22) 0.09L thickness
Pipes and channels Pipe radius or
(fully developed flow) L [0.14–0.08(1 − y/L)2 − 0.06(1 − y/L)4] channel half width

The mixing length has been found to be very useful in simple two-dimensional flows with slow changes in
the flow direction. In these cases the production of turbulence is in balance with its dissipation throughout the
flow, and turbulence properties develop in proportion with a mean flow length scale L. This means that in such
flows the mixing length m is pro-portional to L and can be described as a function of position by means of a
simple algebraic formula. The majority of practically important flows, however, involve additional contributions
to the budgets of turbulence properties due to transport, i.e. convection and diffusion. Moreover, the
production and destruction processes may be modified by the flow itself. Consequently, the mixing length
model is not used on its own in general-purpose CFD, but we will find it embedded in many of the more
sophisti-cated turbulence models to describe near-wall flow behaviour as part of the treatment of wall boundary
conditions.

Table 3.3 Mixing length model assessment

Advantages:
• easy to implement and cheap in terms of computing resources
• good predictions for thin shear layers: jets, mixing layers, wakes and boundary layers
• well established

Disadvantages:
• completely incapable of describing flows with separation and recirculation
• only calculates mean flow properties and turbulent shear stress
The k---ε model

In two-dimensional thin shear layers the changes in the flow direction are always so slow that the turbulence
can adjust itself to local conditions. In flows where convection and diffusion cause significant differences
between produc-tion and destruction of turbulence, e.g. in recirculating flows, a compact algebraic
prescription for the mixing length is no longer feasible. The way forward is to consider statements regarding
the dynamics of turbulence. The k–ε model focuses on the mechanisms that affect the turbulent kinetic
energy. Some preliminary definitions are required first. The instantaneous kinetic energy k(t) of a turbulent
flow is the sum of the mean kinetic energy Κ =

–12 (U2 + V 2 + W 2) and the turbulent kinetic energy k = –12 ( u′2 + v′2 + w′2 ):
k(t) = K + k

In the developments below we extensively need to use the rate of deforma-tion and the turbulent stresses.
To facilitate the subsequent calculations it is
common to write the components of the rate of deformation sij and the stresses τij in tensor (matrix) form:

Gsxx sxy sxzJ Gτxx τxy τxzJ

sij = Hsyx syy syzK and τij = Hτyx τyy τyzK
Iszx szy szzL Iτzx τzy τzzL
Decomposition of the rate of deformation of a fluid element in a turbulent flow into a mean and a fluctuating
component, sij(t) = Sij + s′ij, gives the fol-lowing matrix elements:

∂U ∂u′ ∂V ∂ v′
sxx(t) = Sxx + s′xx = + syy(t) = Syy + s′yy = +
∂x ∂x ∂y ∂y
∂W ∂w′
szz(t) = Szz + s′zz = +
∂z ∂z
1 G ∂U ∂V J 1 G ∂ u′ ∂ v′ J
sxy(t) = Sxy + s′xy = syx(t) = Syx + s′yx = H + K + H + K
2 I ∂y ∂x L 2 I ∂ y ∂x L
1 G ∂U ∂W J 1 G ∂ u′ ∂ w′ J
sxz(t) = Sxz + s′xz = szx(t) = Szx + s′zx = H + K + H + K
2 I ∂z ∂x L 2 I ∂z ∂x L
1 G ∂V ∂W J 1 G ∂ v′ ∂ w′ J
syz(t) = Syz + s′yz = szy(t) = Szy + s′zy = H + K + H + K
2 I ∂z ∂y L 2 I ∂z ∂y L
The product of a vector a and a tensor bij is a vector c whose components can be calculated by application of
the ordinary rules of matrix algebra:
Gb11 b12 b13J Ga1b11 + a2b21 + a3b31JT Gc1JT
abij ≡ aibij = [a1 a2 a3] Hb21 b22 b23K = Ha1b12 + a2b22 + a3b32K = Hc2K = cj = c
Ib31 b32 b33L Ia1b13 + a2b23 + a3b33L Ic3L
The scalar product of two tensors aij and bij is evaluated as follows:
aij . bij = a11b11 + a12b12 + a13b13 + a21b21 + a22b22 + a23b23
+ a31b31 + a32b32 + a33b33

We have used the convention of the suffix notation where the x-direction is denoted by subscript 1, the y-
direction by 2 and the z-direction by 3. It can be seen that products are formed by taking the sum over all
possible values of every repeated suffix.

Governing equation for mean flow kinetic energy K

An equation for the mean kinetic energy K can be obtained by multiplying x-component Reynolds
equation (3.27a) by U, y-component equation (3.27b) by V and z-component equation (3.27c) by W.
After adding together the results and a fair amount of algebra it can be shown that the time-average
equation governing the mean kinetic energy of the flow is as follows (Tennekes and Lumley, 1972):
 ∂ (ρK)
+ div(ρKU) = div(−PU + 2µUSij − ρU ui′uj′ ) − 2µSij . Sij + ρ ui′uj′ . Sij (3.41)
∂t
(I) (II) (III) (IV) (V) (VI) (VII)

Or in words

Transport Transport Rate of Rate of destruction

Rate of change Transport Transport
of K by of K by viscous
of mean kinetic + of K by = of K by + + − −
viscous Reynolds dissipation turbulence
energy K convection pressure
stresses stress of K production

The transport terms (III), (IV) and (V) are all characterised by the appearance of div and it is common
practice to place them together inside one pair of brackets. The effects of the viscous stresses on K have
been split into two parts: term (IV), the transport of K due to viscous stresses; and term (VI), the viscous
dissipation of mean kinetic energy K. The two terms that contain the Reynolds stresses −ρ ui′uj′
account for turbulence effects: term (V) is the turbulent transport of K by means of Reynolds stresses and
(VII) is the net decrease of K due to deformation work by Reynolds stresses giving rise to turbulence
production. In high Reynolds number flows the turbulent terms (V) and (VII)VII) are always much larger
than their viscous counterparts (IV) and

Governing equation for turbulent kinetic energy k

Multiplication of each of the instantaneous Navier–Stokes equations (3.24a–c) by the appropriate fluctuating
velocity components (i.e. x-component equa-tion multiplied by u′ etc.) and addition of all the results,
followed by a repeat of this process on the Reynolds equations (3.27a–c), subtraction of the two resulting
equations and very substantial rearrangement, yields the equation for turbulent kinetic energy k (Tennekes
and Lumley, 1972).

∂ (ρk)
+ div(ρkU) = div(− p′u′ + 2µ u′s′ij − ρ –12 ui′ . ui′uj′) − 2µ s′ij . s′ij − ρui′uj′ . Sij (3.42)
∂t
(I) (II) (III) (IV) (V) (VI) (VII)

In words

Rate of change of Transport Transport Transport of Transport of Rate of Rate of

turbulent kinetic + of k by = of k by + k by viscous + k by Reynolds − dissipation + production
energy k convection pressure stresses stress of k of k

Equations (3.41) and (3.42) look very similar in many respects; however, the appearance of primed quantities
on the right hand side of the k-equation shows that changes to the turbulent kinetic energy are mainly governed
by turbulent interactions. Terms (VII) in both equations are equal in magni-tude, but opposite in sign. In two-
dimensional thin shear layers we found (see section 3.4) that the only significant Reynolds stress −ρ u′v′ is
usually positive if the main term of Sij in such a flow, the mean velocity gradient ∂U/∂y, is positive. Hence term
(VII) gives a positive contribution in the k-equation and represents a production term. In the K-equation,
however, the sign is negative, so there the term destroys mean flow kinetic energy. This expresses
mathematically the conversion of mean kinetic energy into turbulent kinetic energy.
The k---εε model equations
It is possible to develop similar transport equations for all other turbulence quantities including the rate of
viscous dissipation ε (see Bradshaw et al., 1981). The exact ε-equation, however, contains many unknown and
unmea-surable terms. The standard k–ε model (Launder and Spalding, 1974) has two model equations, one
for k and one for ε, based on our best understand-ing of the relevant processes causing changes to these
variables.
We use k and ε to define velocity scale ϑ and length scale
representative of the large-scale turbulence as
follows:

k3/2
ϑ = k1/2
=
ε
One might question the validity of using the ‘small eddy’ variable ε to define the ‘large eddy’ scale
. We are
permitted to do this because at high Reynolds numbers the rate at which large eddies extract energy from the
mean flow is broadly matched to the rate of transfer of energy across the energy spectrum to small, dissipating,
eddies if the flow does not change too rapidly. If this was not the case the energy at some scales of turbulence
could grow or diminish without limit. This does not occur in practice and justifies the use of ε in the definition
of
.
Applying dimensional analysis we can specify the eddy viscosity as follows:

k2
µt = Cρϑ
= ρCµ (3.44)
ε

where Cµ is a dimensionless constant.

The standard k–ε model uses the following transport equations for k
and ε:

∂ (ρk) Gµ J
+ div(ρkU) = div H t grad kK + 2µtSij . Sij − ρε (3.45)
∂t I σk L

∂ (ρε) Gµ J ε ε2
+ div(ρεU) = div H t grad εK + C1ε 2µtSij . Sij − C2ε ρ (3.46)
∂t I σε L k k

In words the equations are

Rate of Transport Transport Rate of Rate of

change of + of k or ε by = of k or ε + production − destruction
k or ε convection by diffusion of k or ε of k or ε

The equations contain five adjustable constants: Cµ, σk, σε, C1ε and C2ε. The standard k–ε model employs
values for the constants that are arrived at by comprehensive data fitting for a wide range of turbulent flows:

Cµ = 0.09 σk = 1.00 σε = 1.30 C1ε = 1.44 C2ε = 1.92 (3.47)

To compute the Reynolds stresses we use the familiar Boussinesq relationship:

A ∂U ∂Uj D 2 2
−ρ ui′uj′ = µt B i + E − ρkδij = 2µtSij − ρkδij (3.48)
∂
C jx ∂xiF 3 3
Boundary conditions
The model equations for k and ε are elliptic by virtue of the gradient diffu-sion term. Their behaviour is similar to the
other elliptic flow equations, which gives rise to the need for the following boundary conditions:

• inlet: distributions of k and ε must be given

• outlet, symmetry axis:
• free stream: ∂k/∂n = 0 and ∂ε/∂n = 0
• solid walls: k and ε must be given or ∂k/∂n = 0 and ∂ε/∂n = 0 approach depends on
Reynolds number (see below)

At high Reynolds number the standard k–ε model (Launder and Spalding, 1974) avoids the need to integrate
the model equations right through to the wall by making use of the universal behaviour of near-wall flows discussed
in section 3.4. If y is the co-ordinate direction normal to a
solid wall, the mean velocity at a point at yP with 30 < yP+ < 500 satisfies the log-law (3.19), and measurements of
turbulent kinetic energy budgets indi-cate that the rate of turbulence production equals the rate of dissipation. Using
these assumptions and the eddy viscosity formula (3.44) it is possible to develop the following wall functions, which
relate the local wall shear stress (through uτ) to the mean velocity, turbulence kinetic energy and rate of dissipation:

U 1 uτ2 uτ3
u+ = = ln(EyP+) k= ε= (3.49)
uτ κ Cµ κy

Von Karman’s constant κ = 0.41 and wall roughness parameter E = 9.8 for smooth walls. Schlichting (1979) also
gives values of E that are valid for rough walls.
For heat transfer we can use a wall function based on the universal near-wall temperature distribution valid at
high Reynolds numbers (Launder and Spalding, 1974)

(T − Tw)Cpρuτ A Gσ JD
T+ ≡ − = σT,t B u+ + P H T,l K E (3.50)
qw C I σT,t L F

with TP = temperature at near-wall point yP

Tw = wall temperature
Cp = fluid specific heat at constant pressure
qw = wall heat flux
σT,t = turbulent Prandtl number
σT,l = µCp/ΓT = (laminar or molecular) Prandtl number
ΓT = thermal conductivity

Finally P is the pee-function, a correction function dependent on the ratio of laminar to turbulent Prandtl numbers
(Launder and Spalding, 1974).
At low Reynolds numbers the log-law is not valid, so the above-mentioned boundary conditions cannot be used.
Modifications to thek–ε model to enable it to cope with low Reynolds number flows are reviewed in Patel et al.
(1985). Wall damping needs to be applied to ensure that viscous

and
η(1 − η/η0) k
C 1*ε = C1ε − η= 2Sij . Sij η0 = 4.377 β = 0.012
1 + βη3 ε

Only the constant β is adjustable; the above value is calculated from near-wall turbulence data. All other constants are
explicitly computed as part of the RNG process.
The ε-equation has long been suspected as one of the main sources of accuracy limitations for the standard version
of the k–ε model and the RSM in flows that experience large rates of deformation. It is, therefore, interest-ing to note
that the model contains a strain-dependent correction term in the
constant C1ε of the production term in the RNG model ε-equation (it can also be presented as a correction to the sink
term).
11.5 Block-structured
grids To overcome the problems associated with structured grid generation for
complex geometries, block-structured CFD methods have been developed. In a
block-structured grid, the domain is sub-divided into regions, each of which
has a structured mesh. The mesh structure in each region can be different, and
it is even possible to use different co-ordinate systems. Such meshes are more
flexible than (‘single block’) structured meshes described in the previous
sections. The block-structured approach allows the use of fine grids in regions
where greater resolution is required. The interfaces of adjacent blocks may have
grids on either side that are matching or non-matching, but, either way, they
must be properly treated in a fully conserva-tive manner. In some codes the
solvers are applied in a block-wise manner (block by block with overall final
iterations to unify boundary conditions) and local refinement is possible block-
wise. Block-structured grids with overlapping regions are called composite grids
or chimera grids. Figure 11.9 shows a Cartesian block-structured grid used for
the calculation of flow over an aerofoil. The resulting grid structure combines
the advantages of Cartesian grids – easy to generate, equations simple to
discretise and solve –with the ability of curvilinear grids to accommodate curved
complex bound-aries (see Courier and Powell, 1996).

Figure 11.9 Block-structured mesh

for a transonic aerofoil. Inset shows
cut cells near aerofoil surface. Also
note additional grid refinement in
the flow region
to capture a shock above the
aerofoil
Source: Haselbacher (1999)

Block-structured meshes are extremely useful in handling complex

geometries that consist of several geometrical sub-components such as the IC
engine pent-roof cylinder and inlet port geometry. Figure 11.10 demon-strates
the improvement of grid quality that was achieved by applying the block-
structured meshing in the engine code KIVA-3V to define separate blocks for
mesh inlet ports, valve regions and the engine cylinder (generated using the
pre-processor of Kiva 3V: see Amsden, 1997).
 11.6 Unstructured
grids An unstructured grid can be thought of as a limiting case of a multi-block grid
where each individual cell is treated as a block. The advantage of such an
arrangement is that no implicit structure of co-ordinate lines is imposed by the
grid – hence the name unstructured – and the mesh can be easily concentrated
where necessary without wasting computer storage. Moreover, control volumes
may have any shape, and there are no restrictions on the number of adjacent cells
meeting at a point (2D) or along a line (3D). In prac-tical CFD, triangles or
quadrilaterals are most often used for 2D problems and tetrahedral or hexahedral
elements in 3D ones. Figure 11.11 shows a tri-angular unstructured mesh for the
calculation of a 2D flow over an aerofoil.

Figure 11.11 A triangular grid

for a three-element aerofoil
Source: Haselbacher (1999)

In unstructured grid arrangements we are not restricted to one particular cell type, but it is possible to use a mixture of
cell shapes. In 2D a mixture of triangular and quadrilateral elements can be used to construct the grid. In 3D flow
calculations combinations of tetrahedral and hexahedral elements are frequently used. Such grids are called hybrid
meshes. Figure 11.12 shows an example of a hybrid unstructured grid for the calculation of flow in a tube bank where
quadrilateral cells have been used near solid walls to provide better resolution of the viscous effects in the boundary layers
and an expanding triangular mesh structure elsewhere to utilise the resources efficiently.

Figure 11.12 An example of an

unstructured mesh with mixed
elements

The most attractive feature of the unstructured mesh is that it allows the calculation of flows in or around geometrical
features of arbitrary complexity without having to spend a long time on mesh generation and mapping. Grid generation
is fairly straightforward (especially with triangu-lar and tetrahedral grids), and automatic generation techniques,
originally developed for finite element methods, are now widely available. Further-more, mesh refinement and adaption
(semi-automatic mesh refinement to improve resolution in regions with large gradients) are much easier in unstructured
meshes. In the sections to follow we explore the unstructured methodology in more detail as it is now the most popular
technique and is included in all commercial CFD codes on the market today.
 11.7 Discretisation in
unstructured Unstructured grids are the most general form of grid arrangement for
grids most complex geometries. Here we present a brief outline of discretisation
techniques for unstructured grids with arbitrary cell shapes, which may be
bounded by any number of control surfaces. We limit ourselves to the devel-
opment of the main ideas; interested readers should consult the literature for
further details of the methodology.
There are two ways of defining control volumes in unstructured meshes:(i) i)
cell-centred control volumes and (ii) vertex-centred control volum These two
variants are illustrated in Figure 11.13 for a 2D problem.
In the cell-centred method the nodes are placed at the centroid of the
control volume, as shown in Figure 11.13a. In the vertex-centred method the
nodes are placed on the vertices of the grid. This is followed by a process known
as median-dual tessellation, whereby sub-volumes are formed by joining
centroids of the elements and midpoints of the edges, as illustrated in

Figure 11.13 Control volume

construction in 2D unstructured
meshes: (a) cell-centred control
volumes; (b) vertex-based control
volumes

Figure 11.13b. The sub-volume surrounding a node then forms the control volume for discretisation. Both cell-centred and
vertex-centred methods are used in practice. We develop the ideas of discretisation in unstructured grids for the cell-centred
method, which is simpler to understand, and, since a control volume always has more vertices than centroids, it has slightly
lower storage requirements than the vertex-centred method.
Principles of Grid
Generation

Prior to the numerical solution of the governing equations, we have to discretise

the surfaces of all boundaries and to generate a volume grid inside the flow
domain. As we discussed at the beginning of Chapter 3 (Section 3.1), we can
choose basically between:
• structured, and
• unstructured
grids. An example of structured grid for a civil aircraft [l], [2] is presented in
Fig. 11.1 and 11.2. For comparison, an unstructured surface and volume grid
for a similar configuration [3]-[6] is displayed in Fig. 11.3 and 11.4.
The structured as well as the unstructured grids have their spedfic acivan­
tages and shortcomings, which we mentioned in Section 3.1. However, regardless
of the grid type, the main bottleneck is currently the quality of data being im­
ported from a CAD (Computer Aided Design) system into the grid generation
program. The surface description is usually transfered via a standard format like
IGES [7]. A direct transfer of CAD native data is rather rate [8]. The experi­
ence shows that this process can impair the accuracy of the data. Furthermore,
the surface representation in the CAD system itself is often imprecise. This
leads mostly to gaps, overlaps or discontinuities between neighbouring surface
patches. Such errors have to be eliminated before the surfaces can be discretised.
We speak in this respect of "CAD repair" [9], [10].
In the following, we shall present the basic methodologies applied to generate
structured (Section 11.1) as well as unstructured (Section 11.2) grids. Due to
the restricted space, we can provide here only a brief description. We refer the
reader to [11] or [12] for a deeper discussion of surface modelling, structured
and unstructured surface and volume grid generation.
 Chapter 11

Figure 11.1: Structured surface and volume )-\riu of a wiu)-\-liody co11fi.gmat.io11.

(Courtesy 0. Brodersm, DLH., Germany).

Figure 11.2: Structured surface and volume griu of a wing-body configurat.ion

-- <let.ail of the pylou and the cugiuc uacclle. (Courtesy 0. Brodersen, DLR,
G(•rmanv).
Figure 11.3: Unstructured surface grid of a wing-body configuration. (Cour­
tesy D. ?vlavriplis, ICASE, USA).

Figure 11.4: Unstructured grid of a wing-body configuration - detail of the

slats and rough-cut through the volume grid. (Courtesy D. Jvlavriplis, ICASE,
USA).
11.1 Structured Grids
The distinguishing feature of structured grids is that the grid points in the
physical space are mapped in an unique way onto a continuous set of three
integers i,j, k (one for each coordinate direction). The set of integers defines
what is called the computational space (see Fig. 3.2). The coordinates�' 1/,( in
the computational space are related to i, j, k as follows

{ = i/imax, i = 0, 1, 2, ·· ·, im ax
TJ=i/imax, j=0,1,2,··•,jmax (11.1)
(=k/kmax, k=0, 1, 2, · · ·, km ax

This mapping implies that 0 ;S � ;S 1, 0 ;ST/ ;S 1, and 0 ;S ( ;S 1. Neighbouring

grid points can be connected to form cubes in the computational and hexahcdra
(quadrilaterals in 2D) in the physical space. Structured grid generation systems
discretise the boundary surfaces of the flow domain using quadrilaterals - termed
the surface grid - and fill the interior with hexahedra. The grid inside the
domain is named the volume grid.
The generation of a structured grid starts by distributing grid points along
boundary curves (boundaries of surface patches). The usual procedure is to
place the nodes more dense in regions with high curvature. Using the point
distribution on boundary curves, the surface grid can be generated. Based on
the surface grids which enclose the physical domain, we can finally construct
the volume grid. Thus, the common problem is t,o generate a grid inside the
domain based on known boundary discretisation. This can be solved by two
different approaches:
• algebraic grid generation, or
• grid generation using partial differential equations (PDE's).
The application of PDE's requires a valid initial grid (surface or volume), which
is mostly generated algebraically. Two different types of PDE's are common:
• elliptic equations,and
• hyperbolic equations.
The algebraic grid generation (Subsection 11.1.2) employs a direct functional
description of the coordinate transformation between the computational and
the physical space. The most widely used algebraic technique is the so-called
Transfinite Interpolation (TFI). Given the point distribution on all boundaries,
it generates the grid points inside the physical domain by interpolation. Particu­
lar formulations of the TFI method allow for angle and grid spacing control at
the boundaries.
The methodology based on elliptic PDE's (Subsection 11.1.3) is the most
popular one. It allows the user to prescribe the angle between a grid line and
boundary and to control the grid spacing and the expansion ratio near surfaces.
Elliptic grid generation also guarantees a smooth grid in the entire domain.
Thus, high quality, boundary orthogonal grids can be generated. The downside
of the elliptic method is a much longer computing time as compared to alge­
braic or hyperbolic grid generation approaches. The mP.thod also suffers from
numerical difficulties.
Hyperbolic PDE's can also be utilised for the grid generation (Subsection
11.1.4). This technique generates the volume grid by marching between two sur­
faces in the direction of one particular computational coordinate. The marching
procedure is explicit, i.e., based on a known surface point distribution a new
layer is generated. A natural restriction of the hyperbolic grid generation is
that the shape of the outer grid boundary cannot be fully controlled. However,
this may represent no real problem like in the case of external flows. The hy­
perbolic grid generation can provide approximate grid orthogonality over the
entire domain. It is also computationally inexpensive.

11.1.1 C-, H-, and 0-Grid Topology

Before we can start to generate any grid, we have to think about its topology.
This means, we have to decide how many grid blocks are necessary and how the
blocks should be ordered with respect to each other (by the way, this work may
take weeks to months in the case of a complex geometry). For each grid block,
we have to assign boundaries (or their parts) in the computational domain to
particular boundaries in the physical space (e.g., solid wall, farfield , etc.). The
appearance of the grid in the physical space will depend strongly on this assign­
ment. In practice, three standard single-block grid topologies are established.
They are named as the C-, H-, or the O-grid topology because in a plane view
the grid lines resemble the corresponding capital letter. A grid in 3D can be
described as a combination of two topologies. For example, the grid around a
wing usually consists of a C-grid in the flow direction (cf. Fig. 11.2) and of an
O-grid (or an H-grid) in the spanwise direction. In this case, we speak of a
C-O-grid. In the following, we shall discuss all three topologies in more detail.

C-Grid Topology
In the case of the C-topology the aerodynamic body is enclosed by one family of
grid lines, which also form the wake region (if present). The situation is sketched
in Fig. 11.5. As we can see, the lines T/ = const. start at the farfield ({ = 0),
follow the wake, pass the trailing edge (node b), wrap in clockwise direction
round the body, and finally continue to the farfield again ({ = 1). The other
family of grid lines ({ = const.) emanates in normal direction from the body and
the wake. The part (segment) a-b of the grid line T/ = 0 represents a coordinate
cut. This means that the segment a-b in the physical space is mapped onto
two segments in the computational space, namely a � { � b and b' � { � a'.
Hence, the nodes on the upper (b1 -a1) and the lower part (a-b) of the cut are
hold separately in the computer memory. The appropriate boundary condition
was presented in Section 8.6.
 Chapter 11

TJ= l ri
f
d e
-
f

- --- -- ,-
- --- - - •f -
__ ,_
e
- -- - - -
ri=O
- - - -- .

L: (a) physi<.:al space

d
a
-/~~
~a 1//,,1//'.i i t
~ );%
~),:~

(b) computational space

. ~- ;·--~
a s

Figure 11.5: C-grid topolop;y iu 2D .

Figure 1 1.6 : Partial view of a C -grid around NACA 0012 a irfoil.

H-Grid Topology
The H-grid topology is quite often employed in turbomachinery for grid genera­
tion in the bladed flowpath. The topology is displayed in Fig. 11. 7. As one can
observe, the surface of the aerodynamic body is described here by two different
grid lines, i.e., TJ = 0 and T/ = l. On contrary to the C-grid, one family of grid
lines (TJ = const.) closely follows the streamlines (inlet located at ( = 0, outlet
at ( = 1).
At the first sight, there is no obvious coordinate cut. However, in turbo­
machinery the segments a-b and e- f are periodic (rotationally periodic in 3D)
to each other. The same is true for the segments c-d and g-h. This type of
boundary condition is treated in Section 8.7. Figure 11.8 shows an example of
a non-orthogonal H-grid between turbine blades.

0-Grid Topology
We can see from the rendering of the O-topology in Fig. 11.9 that one family of
grid lines (ri = const.) forms dosed curves around the aerodynamic body. The
second family of grid lines (( = canst.) is spanned in radial direction between
the body and the outer boundary (farfield in this case). The complete boundary
line T/ = 0 represents the contour of the body (from a to a'). The coordinate
cut is defined by the boundaries ( = 0 (nodes a-c) and ( = 1 (nodes a'-c') in
the computational space. The example in Fig. 11.10 shows a standard O-grid
used to simulate inviscid flow past an airfoil. A disadvantage of the O-topology
is the poor grid quality at a sharp trailing edge.

11.1.2 Algebraic Grid Generation

The most widely used algebraic techniques for surface or volume grid generation
from prescribed boundary point distribution is the Transfinite Interpolation
(TFI) method. It was first described by Gordon and Hall in 1973 (13]. The TFI
scheme utilises 1-D univariate interpolations in each of the coordinate directions
in the computational space. The general form of the univariate interpolation
functions reads

1
Adaptive grid
An adaptive grid is a grid network that automatically clusters grid points in
regions of high flow-field gradients; it uses the solution of the flow-field properties
to locate the grid points in the physical plane. An adaptive grid can be visualized as
one which evolves in steps of time in conjunction with a time-dependent solution of
the governing flow-field equations, which computes the flow-field variables in steps
of time. During the course of the solution, the grid points in the physical plane move in
such a fashion to "adapt" to regions of large flow-field gradients as these
gradients evolve with time. Hence, the actual grid points in the physical plane are
constantly in motion during the solution of the flow field and become stationary
only when the flow solution approaches a steady state. Therefore, unlike the
stretched grid discussed in Sec. 5.6 and the elliptic grid generation discussed in Sec.
5.7, where the generation of the grid is completely separate from the flow-field
solution, an adaptive grid is intimately linked to the flow-field solution and changes
as the flow field changes. The hoped-for advantages of an adaptive grid are
associated with the grid points being automatically clustered in regions where the
"action" is occurring. These advantages are (1) increased accuracy for a fixed
number of grid points or (2) for a given accuracy, fewer grid points are needed.
Adaptive grids are still very new in CFD, and whether or not these advantages are
always achieved is not well established.
An example of a simple adaptive grid is that used by Corda (Ref. 35) for the
solution of viscous supersonic flow over a rearward-facing step. Here, the trans­
formation is expressed in the form
2 and 16 Marks