eddy viscosity distribution on the longitudina1 flow velocity Report nr. 14-88
G.J.G.M Hoffmans
Faculty of Givi1 Engineering Hydrau1ic Engineering Delft University of Techno1ogy
2.1
2. Flow velocity and eddy viscosity distributions (2-D-vertical) 7
Logarithmic flow velocity profile Eddy viscosity (parabolic) profile
7 7 9 10 10 12 13 13 14 Standard concept 2.1.1 2.1.2 2.2 Coles concept 2.2.1 2.2.2
Coles flow velocity profile Eddy viscosity (Coles) profile 2.3 Exponential concept (eddy viscosity)
2.3.1 2.3.2
Exponential form of the eddy viscosity Flow velocity profile (exponential)
3. Eddy viscosity at a free surface (uniform flow) 3.1 General
3.2 Classical theories of Prandtl and Boussinesq 3.3 k-equation
3.4 k-equation and Boussinesq 3.5 ODYSSEE (k-e-model)
17 17 17 17 20 20 4. Assesments
Undisturbed outer flow Disturbed outer flow
22 22 22 26 26 26 29 4.1 4.2 4.3
Open channel flow Pipe flow
Internal boundary layer 4.3.1 4.3.2 5. Conclusions References Appendices 4.3.2.1 Shear-stress distribution 31
a B C 11 c,..,. D g i k 1 m L n m q Q Re R* R u,n t u w x y flow depth
width (of flurne)
turbulence constant
local friction coefficient
turbulence constant
constant in turbulence model (k-f)
internal diameter
acceleration of gravity
energy slope
turbulent kinetic energy per unit mass
mixing length (Prandtl) length scale of turbulence
integer, index
integer, index
pressure fluctuation
stress production of the turbulent energy discharge per unit width
discharge
Reynolds nurnber
u*a/1I ; Reynolds nurnber correction term
time
local time-averaged longitudinal flow velocity longitudinal flow velocity fluctuation
friction or wall- or bed-shear velocity local shear velocity
longitudinal free stream velocity
depth-averaged longitudinal flow velocity
transverse flow velocity fluctuation (y-direction) local time-averaged vertical flow velocity
vertical flow velocity fluctuation longitudinal coordinate transverse coordinate (L) (L) (
-
) (-
) (-
) (-
) (L) 2 (LT ) (-) 2 2 (L T ) (L) (L) (-
) (-
) 1 2 (ML T ) 2 3 (L T ) 2 1 (L T ) 3 1 (L T ) (-
) (-
) (-
) (T) 1 (LT ) 1 (LT ) 1 (LT ) (LT ) 1 (LT ) 1 (LT ) 1 (LT ) 1 (LT ) 1 (LT ) (L) (L)z a w f IC 1/ 1/ t P Ot o 1/ r t "0 Ó w
n
'i1 w( ) zero-velocity level exponential parameterrate of energy dissipation per unit mass by turbulence Von Karman's universal constant
kinematic molecular viscosity coefficient
eddy viscosity fluid density
turbulent normal stress
standard deviation (eddy viscosity)
turbulent shear stress
wallor bed-shear stress
(new wall-) boundary layer thickness
Coles's profile parameter
scalar product wake function Subscripts C Coles E Exponential 1/ t max maximum
ref at reference height
s at surface
w at outer edge new wall-boundary layer
(L) (-) 2 3 (LT ) (
-
) 2 1 (L T ) 2 1 (L T ) 3 (ML ) 1 2 (ML T ) (-
) 1 2 (ML T ) 1 2 (ML T ) (L) (-
) ( - ) (-
)1. Introducti~n
The general purpose of this research project is to model mathematica11y
the loca1 scour downstream of a structure (2-D). The model has to
simu1ate the deve10pment of the scour as a function of the time,
(figure 1). Basica11y two mode1s are necessary name1y a flow model and a
morphologica1 model. The latter model has to describe the bed- and
suspended load and the eros ion of the bed.
In the present study a function (prescription) for the eddy viscosity
with a variable parameter, which is a parameter in the exponential
function, is discussed. A sensitivity study has been made in order to
fit the parameter mentioned above for several flow types.
In this study the hypothesis of Boussinesq and a 1inear shear stress
distribution are applied.
sz
>sz
--uniform flow ...
-
relaxation zonemixing layer
figure 1 Definition sketch
Using this eddy viscosity concept, it follows, that the profile of the
longitudinal flow velocity is a logarithmic profile with a correction
term. The differences between the traditional logarithmic flow velocity profile and the Coles flow velocity profile are small near the wall. A large difference occurs at the surface of a uniform flow or at the
center of a pipe flow. There the gradient of the longitudinal flow
regardless of the value of the parameter in the exponential function for the eddy viscosity.
In section 4.0 some applications are given for a uniform flow, a pipe
flow and a new wall-boundary layer with a disturbed and an undisturbed
outer flow (no turbulence). In paragraph 4.3.2 a function for the eddy
viscosity is discussed, which describes the eddy viscosity in a new
2. Flow velocity and eddy viscosity distributions (2-D-vertical)
2.1 Standard concept
2.1.1 Logarithmic flow velocity profile
In the (turbulent) boundary layer two regions may distinguished. A region adjacent to the wall, in which the flow is directly affected by the condition at the wall. This condition is expressed by the wall shear stress. This region is referred to as the 'wa11' or 'inner' region. Beyond this wall region there is an another region, where the flow is only indirectly affected by the wall through its wal1 shear stress. This second region is usually referred to as the 'outer' region.
In the inner region there is a thin layer at the wall where the flow is (essentially) viscous. Outside this thin viscous sublayer inertial effects become more and more important compared to the viscous effects, until at some distance from the wall the flow becomes fully turbulent with dominating inertial effects and negligible direct viscous effects on the main flow. The region between the fully turbulent flow and the viscous sub-layer, where the viscous stresses and Reynolds stresses are of the same magnitude, is called the transition or 'buffer' region.
The momentum equation over the boundary layer thickness can be written kinematically in terms of local shear velocity u*z as (Coleman, 1984):
aü
u2 = v +
*z az (1)
in which:
u*z local shear velocity
v kinematic molecular viscosity coefficient ulocal time-averaged longitudinal flow velocity
K 0.4; Von Karman's universal constant z vertical coordinate
Using an asymptopic approach, the boundary conditions at the bottom (z = 0) and for z = ó are respectively:
Z = 0: u = 0 and (2)
Z = ó :
w u ü(ó w) and
o
(3)in which:
ó boundary layer thickness w
u* friction velocity or wall-shear velocity
At the origin of z viscous stresses predominate. Dropping the turbulent
.
aü
stress term
(~z)2{az)2
equation (1) reduces to:Equation (4) can only be integrated if an expression for u*z is
available. To scale
ü
by the friction velocity u*, assuming that u*z is equal to u*, equation (4) is integrated over z. This gives:(5)
v
which is, of course, the viscous sublayer equation.
Moving outward in the flow, the turbulent stresses predominate, so eliminating the viscous stress term, equation (1) changes into:
(6)
As with equation (4), equation (6) can be integrated only if an expression for u*z is available. Again to scale
ü
by the friction velocity u*, taking u*z = u*, yields:1
ln ~~ Zo
(7)
The assumption u*z = u* was made in order to perform the integration; in short, it is an asymptotic approximation at Z
«
ó , where the shearw
stress is approximately equal to the shear stress at the wall. In equation (7), Zo is an intercept at which
ü
calculated from the logarithmic function is zero; this is a boundary condition on theintegral and does not indicate that a real velocity profile goes to zero at a finite heigth above the origin of z.
Generally equation (7) can be applied to flows over both rough and smooth boundaries.
2.1.2
Eddy viscosity (parabolic) profileConsidering a uniform flow in an open channel, the flow is driven by the component of the acceleration of gravity in the longitudinal direction. Balance of forces and stresses supplies a linear shear stress
distribution (no external forces):
T = pga(l - ~)i t,xz a (8) in which: T = shear stress t,xz p fluid density g acceleration of gravity a = flow depth i energy slope
The definition of the bed-shear stress (TO) is given by:
pu2
*
(9)Conversion of (8) and (9) leads to:
T = pu; (1 - -az)
Originally Boussinesq assumed that the turbulence stresses act like the
viscous stresses, that is, that they are directly proportional to the
velocity gradient. However, in a boundary layer, for instance, the eddy viscosity is not constant near the wall. The hypothesis of Boussinesq reads (vt is a function of z):
T
t,xz (11)
in which:
vt eddy viscosity
w local time-averaged vertical flow velocity
x longitudinal coordinate
Using a logarithmic flow velocity distribution (7) and applying the
hypothesis of Boussinesq (11), with ~:
«
~~
and the assumptionmentioned above (linear shear stress), the dimensionless eddy viscosity reads:
~ (1 - ~)
a a (12)
which is the parabolic eddy viscosity distribution.
2.2 Coles concept
2.2.1 Coles flow velocity profile
The Coles flow velocity profile, which is also known as the law of the wake, over both geometrically smooth and geometrically rough boundaries
can be written as (Hinze 1975, p. 697):
1
[ln ~ K Ó W - II{2 - w(~ ))] w(~ »1)
Zo (13)in which:
local time-averaged longitudial flow velocity (Coles)
longitudial free stream velocity
Coles's profile parameter 1 + sin(2~ - 1)~ - 2sin2{~ ~
Ó 2 2 Ó
w w
wake function (14)
Z
The wake function w(S ), which is a positive function, is a correction w
to the logarithmic flow velocity profile. Experimenta1 evidence suggests that this law has a universal character.
In order to quantify the longitudinal flow velocity for Z = ó and to
w
indicate the differences between the (pure) logarithmic flow velocity profile and the one according to Coles, the following proceeding has been carried out.
Using the approximation of (2.1.1) and substituting z (13), [üC(zo)= OJ, this gives:
Zo into equation
1 Zo Zo
; [In
s- -
rr{2
- w(s-)}J
w w
(15)
Substitution of (15) into (13), the Co1es flow velocity profile can be written as: z z
1
[Inrr { (
)
K zo+ w s-W (16) Awhere term Arepresents the 'Coles' correction to the logarithmic profile.
The parameter
rr
is a profile parameter and does not depend on z. It is re1ated to the loca1-friction coefficient Cf = 2(U*/U)2. In a boundarylayer
rr
is a weak function of x.The Co1es parameter can be approximated by integrating equation (16) to z (appendix A) :
ó . - ln ...1!
+
1Zo
(17)
Here Ü represents the depth-averaged longitudinal flow velocity.
For a boundary layer with zero pressure-gradient Coles assumed
rr
to be independent of x. He obtainedrr -
0.55 with ~ - 0.4.For an open channel flow
rr
is found to depend on the Reynoldsnumber R* (-u*a/v) up to a certain value of this number: Below R*=
500, the Co1es parameterrr
is nearly zero, then it increases with R* and assumes a constant va1ue of approximate1y 0.2 for R*>
2.000, (~~ 0.419), (Nezu and Rodi, 1986).It can be noted, that if the Coles parameter
(rr)
is equal to zero equation (16) reduces to (7), (figure 2).d
z
~ t i :li
" d QJI 0.. 1.1 1.2 x - exponential;rr
= 0.00 ct - 1.50 wo -
Co1es;figure 2 Coles and 'exponential' flow velocity profile
2.2.2 Eddy viscosity (Coles) profile
Applying the Coles flow ve10city profile, the hypothesis of Boussinesq and a linear shear stress distribution the eddy viscosity can be written as:
{l
-
~ } ~ ~s
w u*ów = K,s
w [1 + 11" "6rrz . ( z)
1
S1n 11""6 w w (18)If the Coles parameter rr is equal to zero, equation (18) reduces to (12),
[ó
-
al
.
w
2.3 Exponential concept
2.3.1 Exponential form of the eddy viscosity
Since the calculations by the turbulence model ODYSSEE (Hoffmans, 1988a) show an exponential character for the eddy viscosity distribution in a number of local scour holes, in 2.3 an analysis is given of the
longitudinal flow velocity using an exponential function for the eddy viscosity.
exponential: K, "'6
z
exp -ow ó(
~ )
w w
(19)
in which:
o = 'exponential' parameter
w
Examination of equation (19) gives the following result. For 0 < 0 the
w
eddy viscosity increases progressively, that is, if dv /dz and d2v /dz2
t t
are positive and increase with z. For 0 = 0 a linear function will be
w
obtained. For 0
<
0<
1 the eddy viscosity increases degressively, that wis, if dvt/dz is positive and increases with z, while d2vt/dz2 is
smaller than zero. For Ow
>
1 a local maximum occurs for z = ó~ow. Byan increase of 0 not only z (z-coordinate where vt is at maximum)
w max
but also the absoLut;evalue of tit will decrease, (figure 3).
Close to the bottom the eddy viscosity is approximately:
z
«
óAssuming a logarithmic flow velocity profile near the wa11 region, this is in agreement with the theorem of Prandtl.
Prandtl: T t,xz (21) in which: 1 - ~z; mixing 1ength m 1.0 0.' 0.6 (lw> I
•
6w
o.~ O.l O<Clw<1 Clw< 00 O.O~ o.oa O.ll 0.16 O.lO O.l~ 0.21 o.n 0.35 o.~ VI
u.liw
figure 3 Eddy viscosity as a function of Ct
w
2.3.2 Flow velocity profile (exponential)
Using the exponentia1 eddy viscosity function and applying the hypothesis of Boussinesq and a 1inear shear stress distribution,
integration of the horizontal velocity gradient from Zo to z results in a logarithmic flow velocity profile with a correction.
aw
aü
Assuming that 8x
«
8z' conversion of (10) and (11), (a - ów)' leads to:(22)
Z) ( ~)
"6
exp aw 6w w
(23)
Using the Taylor-series to the exponential function integration of (23)
from Zo to z, [ü(z) - üE(z)], results in (appendix B):
ÜE(z) z co z Zo
1
[In + l: Ru n{ (6"}n {6"}n)] (24) u* K. Zo n=l ' w w B in which:local time-averaged longitudinal flow velocity (exponential vt)
n-l a a __J!_ [__J!_ _ 1] n = index (25) n! n R u,n
Term B, equation (24), is a correction-term to the logarithmic flow
velocity profile. The higher order terms converge to a limit value,
which is equal to zero, (appendix C). It appears that after 30 terms
(n = 1,...,30), which is satisfactory to converge the sum of the higher
order terms to zero, (appendix D), R is approximately equal to zero
u,n
for a ::::1.50, (1.50286), (figure 4).
w
Difference between the Coles and the exponential concept (velocities) Using this eddy viscosity concept the agreement with a (pure)
logarithmic flow velocity profile is fair in the near wall region,
however, a relatively large difference occurs in the outer region
(z = 6 ), (figure 2). There the gradient of the longitudinal flow
w
velocity is equal to zero, while applying a (pure) logarithmic or the
Coles profile the gradient has a small value.
Differentiation of (24) to z yields: co
1
[
1
+1
l: K. Z 6 w n=l R u,n (26)Substitution of z = 6 into (26) gives:
aÜE(ó
w) co1
_1_ [1 +L
R n1 - 0u*
az
1(15w n-1 u,nco
because
L
R n ...-1, (appendixE).
n=l u,n (27) (28) number of terms n - 30 11
..
3 2 L R u,n 0r
-1 -2 -3 -4 -2.
-o 2 4. ... Ct Wfigure 4 L Ru,n as function of the parameter Ctw
In order to write the longitudina1 flow velocity as a velocity defect using an exponentia1 eddy viscosity, the fol1owing pr-ocedu'r'ehas been carried out.
Substitution of
z -
15w into (24),üE(ó)
can be written as:üE(ó
w) 15 co Zo1
[in ~+
~ R {1- (';5)
n}1
u* I( Zo n-l u,n w
(29)
Subtracting (24) from (29) leads to:
co
1
[ln z.,I( 15
w
3. Eddy viscosity at a free surface (uniform flow)
3.1 General
In this section an overview is given of the different values, which the eddy viscosity can have at a free (plane) surface in a uniform flow starting from different theories. The classical theories of Prandtl and Boussinesq yield a zero eddy viscosity at the surface, while stàrting from the k-equation and the k-e-model, the eddy viscosity does not reduce to zero there.
3.2 Classica1 theories of Prandtl and Boussinesq
Using a logarithmic or the Coles flow velocity profile, the mixing
length must be equal to zero at the surface according to the theorem of
Prandtl (21), since T = 0 (no external
t,xz
[ aw
aÜ]
hypothesis of Boussinesq (11),
ax«
az '
the eddy viscosity reads:
forces). Conversion of the and the theorem of Prandtl
(31)
Because 1 equals zero, the eddy viscosity reduces also to zero.
m
Using the longitudinal flow velocity profile, which follows from the exponential eddy viscosity concept, the shear stress equals zero at the
surface, since the gradient ~~ = 0, however, the eddy viscosity does not
reduce to zero (19).
3.3 k-equation
Considering a steady two-dimensional flow with a constant width and a
plane water surface, the turbulent energy equation reads (Kay and
Dk Dt (32) turbulent diffusion p -t xxa
au
+ -t zxrau
+ -t xzraw
+ _atzzaw
p'a
x
p'az
p'ax
p'az
(33) f = (34) in which:k turbulent kinetic energy per unit mass presssure fluctuation
stress production of the turbulent energy time
t
longitudinal flow velocity fluctuation
transverse flow velocity fluctuation (in y-direction) vertical flow velocity fluctuation
rate of energy dissipation per unit mass by turbulence turbulent normal stress
scalar product
Since there are no changes in the longitudinal direction (uniform flow), all the gradient types with ~x are equal to zero and because the
vertical flow velocity
(w)
and the fluctuating vertical velocity (w1) reduce to zero near the surface (free water surface), thus the normalstress
[a
= _pw1W1] reduces also to zero, the production term Pt,zz reads:
p -tr zx
au
p' Bz (35)
In case there are no external forces at the surface, the production by the shear stress equals zero. Then equation (32) reduces to:
The turbulent diffusion term in (36) and the dissipation are modelled (Rodi, 1980):
a
n1w1 -{~+kwl}aZ
p (37) (38) in which: turbu1ence constant1ength sca1e of turbu1ence
The definition of the kinetic energy k reads:
(39)
The kinetic energy does not reduce to zero, since UI and vI do not equa1
zero at the surface. Based on measurements in an open channe1 flow UI
and VI are modelled by (Nezu, 1977):
z 2.30 exp
(-'6 )
w (40) z 1.27 exp(-'6 )
w (41)ak
a
2kThen
az
andaz
2 are a1so not reducing to zero. Since there is kineticenergy, there is a1so dissipation (38). Consequently the diffusion term (diffusion of the kinetic energy, equation 36) has a certain va1ue at
the surface. In case vt is modelled by (Rodi, 1980):
c
Wk
v (42)
Cv is a turbulence constant, it follows, that if L ~ 0, vt is not
reducing to zero. In case L equals zero the eddy viscosity also equals
zero according (42), however, then the dissipation (38) should tend to
infinity, which is hardly conceivable. It is more 1ikely to assume that
3.4 k-eguation and Boussinesg
Substitution of (11) into (35) yie1ds [8W« 8Ü]:
8x 8z
p (43)
If the production due to the shear stress is equa1 to zero (no external forces), the eddy viscosity or the vertica1 gradient of the longitudinal flow velocity has to reduce to zero. Using the logarithmic or the Coles flow velocity profile in a uniform flow the eddy viscosity has to be
. 8ü. 1 h f
zero, S1nce 8z 1S not equa to zero, at t e sur ace.
3.5 ODYSSEE (k-f-model)
In the turbu1ence model used in ODYSSEE the eddy viscosity is modelled by [conversion of (38) and (42) with c~ ~ cDcv]:
(44)
where c is a constant in the k-f-mode1. ~
The boundary conditions at the water-surface are modelled (Delft Hydraulics, 1987):
8k 8z
II
8z
o
(45)It appears that in case the grid distance is re1ative1y large the turbulence parameters k and f are also relative1y large (Hoffmans,
1988a). Because k does not reduce to zero the eddy viscosity has a certain va1ue at the surface.
4. Assesments
4.1 Open channel flow
Delft Hydraulics (1988) carried out turbulence measurements in a sand flume with a Laser Doppier anemometer. The mean velocity distribution and turbulence characteristics were obtained. The experiments (Tl to T4) were conducted in a sand flume with a flat bed. The width of the flume measured 1.5 mand the discharge per experiment was constant. The most important hydraulic data of the experiment (T2), which were used to fit the parameter ow' were:
3 Q 0.040 m
Is
B 1.5 m 2 q 0.0294 mIs
a 0.0773 m u 0.436mis
s _3 i 0.084*10 u a 3 Re ~ __s_ z30*10 v u* z0.057 u s 6 2 V = 10 mIs
discharge width of flumedischarge (integrated velocity profile) (averaged) flow depth
velocity at surface energy slope
Reynolds number bed-shear velocity
kinematic molecular coeficient (water)
Figure 5 shows the values of the computed and measured velocity defect, [ü - ü(z)]/u*, (van Mierlo and de Ruiter 1988, experiment T2), while
max
figure 6 shows the values of the computed and 'measured' (dimensionless) eddy viscosity [v I(u*ó )] as a function of the relative depth [z/ó ].
t w w
The agreement between the computed (0 = 1.50) and the measured velocity
w
defect respectively eddy viscosity is satisfactory.
4.2 Pipe flow
Measurements of turbulence parameters were made as early as 1954 by Laufer. They were executed in a test tube, which was a straight,
Mean-velocity dï"stributions and turbulence parameters were measured even
into the viscous sublayer.
:
z
d ~ ! ~ ~..
.
'! 0 V"_lty .... ct (-)o
van Mierlo (B.2.1) exponential a = 1.5 wfigure 5 Computed (30) and 'measured' velocity defect (uniform flow)
D D :!:
.
..
!
D .!, D ~.
..
D D D :I D D.02 Q.04 0.06 o.oe 0.1 0.12 iUdJ'wlec..&b- (-)o
van Mierlo (B.2.l) exponential a - 1.5 wThe most important hydrau1ic data of the experiment, which were used to
calibrate the parameter ow' were:
u 30 m/s maximum velocity (center pipe) (air)
max
D 0.247 m interna1 diameter
_6
11 15.1*10 m2/s kinematic m01ecu1ar coefficient (air)
umax D 5 Re - ~ 5*10 Reyn01ds number 11 u* ~ 0.035 wa11-shear velocity u max
The eddy viscosity is obtained directly from the turbu1ence-shear-stress
distribution:
11 (z) ..
-t
(46)
The eddy viscosity computed in this way is shown in figure 7. Not only
Laufer's data for the mean-ve10city distribution have been used but a1so
0.01 + + 0 0.07 0 0.06 + 0 0 :1: 0.05 ~; 0.04 ~ -; ~v 0.03 '" 0.02 0.01 0 0 0.2 0.4 0.11 0.1 R.IoIlv. é.pUl (-) 0 Nunner + Laufer exponentia1 a
-
2.0 wthe data obtained by Nunner (Hinze 1975, p. 730). In Nunner's
experiments the wa11-shear velocity amounted to u*/ü
=
0.045. The maxdrawn 1ine represents the exponentia1 distribution of the eddy viscosity
(19) for a - 2.00. w
The 'measured' eddy viscosity first increases 1inear1y, then reaches a
maximum at about.2z/D - 0.3, then decreases slight1y, and attains a
near1y constant va1ue beyond 2z/D - 0.5. The same course of the
distribution curve was obtained by Reichardt (Hinze 1975, p. 730) in a two-dimensional channel flow, although the maximum there occurred at about 2z/D ~ 0.4 to 0.5.
Figure 8 shows the velocity defect distribution; the agreement between computed and measured distributions is very satisfactory indeed in the region 2z/D
>
0.2. 6 ~ ~ .!-.! 4 ~ D ~ 3 } ~ 2 o 0.2 0.4 0.6 0.8 R.loU.. a.pUl (-)o
Laufer exponentia1 a - 2.0 wfigure 8 Computed (30) and 'measured' velocity defect (pipe flow)
Using a logarithmic flow velocity profile or the Coles profile, a discontinuity in the velocity gradient (8ü/8z) occurs in the center of the pipe, which is in conflict with nature. Due to this the eddy
viscosity reduces to zero for the shear stress has to be zero. Then a correction has to be made for the eddy viscosity or for the longitudinal flow velocity.
Applying the exponential eddy viscosity concept the agreement between the computed and 'measured' eddy viscosity is much better, while the boundary condition for the velocity (aü/az) is also much better in the
center of the pipe (aü/az = Ol).
4.3 Internal boundary layer
4.3.1 Undisturbed outer flow
The most discussed type of boundary layer is the two-dimensional boundary layer flow along a plane surface (plate) with zero pressure
gradient. In front of the leading edge of the plate the velocity of the
undisturbed flow is uniform (aü/az = 0). In case the flow in the outer
region is frictionless, the eddy viscosity reduces to zero. The retarded
layer along the plate is called the boundary layer. With increasing
distance from the leading edge in the downstream direction the thickness of the retarded layer increases continuously. Frictional forces retard the motion of the fluid in a thin layer near the wall. In that thin layer the velocity of the fluid increases from zero at the wall (no slip) to its full value which corresponds to the frictionless flow
(undisturbed outer flow).
Equation (19) describes the eddy viscosity reasonably in case of a well
chosen value of Q . In the lower half of the new wall-boundary layer the
w
modelled v is somewhat smaller, while in the upper half the modelled
t
one is somewhat larger compared with measurements of Klebanoff and Townsend (Hinze 1975, p. 645), figure 9.
4.3.2 Disturbed outer flow
In case the outer flow is disturbed, for instance caused by a mixing layer upstream, the distribution of the eddy viscosity in the boundary layer or the new wall-boundary layer differs from the distribution mentioned in 4.3.1 (wall-boundary layer with undisturbed outer flow),
especially in the outer region, which is caused by the relative high
turbulence degree in the outer flow (relaxation zone).
I.D ,,
,,
,,
0.8,
'-'-,
'-0.6 '-\•
\ 6;, 0.4'.
0.20 O.DI OD2 ODJ 0.D4 0.05 0.06 0.07 O.D'
VI
u.ow
4
0 From Klebanoff's data: Re ::::: 7.5 * 10 u*/U
-
0.0374
X From Townsend's data: Re ::::: (3 to 4) * 10 u*/U - 0.044 according to Coles (18) ; TI - 0.55
according to equation (19); Q - 2.25
w
figure 9 Eddy viscosity: boundary layer with undisturbed outer flow
Measurements above artificial dunes (van Mierlo and de Ruiter, 1988; experiment T5 and T6) show that the eddy viscosity is affected by relative large length scales (vt :: L) in the upper part of the new wall-boundary layer, while in the lower part it is influenced by the wall, where L is relatively small. The following suggestion has been made to model the eddy viscosity:
z 0<-Ó w z
<
ref Ó w (47) z ref v - v Ó Ó 2 [ t,w Ó t,ref]exp{ _[_w~_...:w~]} u*w
°v
zz
z
ref<
Ó Ów
W<
1 (48) in which:v - eddy viscosity at outer edge of boundary layer t,w
vt,ref eddy viscosity at reference heigth
zref reference heigth
a = standard deviation (eddy viscosity)
v
For z smaller than the reference heigth (z f) the eddy viscosity is re
influenced by the wall, while for z larger than Zref it is affected by the óuter flow.
The flow above a dune is characterised by a deceleration and an acceleration zone. Af ter reattachment in the acceleration zone a boundary layer appears. The modelling of the eddy viscosity is satisfactory for, figure 10:
Q 2.5 to 2.8 w
a 0.4 to 0.5· v
zre
f/ó
w-
0.2 to 0.3In figure 10 the eddy viscosity as well as the thickness of the wall-boundary layer are made dimensionless
{v I(u*ó ), z/ó
t w wJ
.
It should be pointed out that the magnitude of the eddy viscosity vt,w
"
"
~ ""
::!: D!
.~d.
D ~ D D :l D D i:P D 0.02 0.060 0.1 0.14 0.'. CL22 0.26...,.
....
_
(-)o
van Mierlo according to (47) and (48) Q - 2.60,z
fiS
-
0.2, a - 0.52 w re w vis dependent on the developed mixing layer downstream (Hoffmans, 1988b).
Considering a uniform flow the eddy viscosity is proportional to the bed-shear velocity, however, in a new wall-boundary layer with a disturbed outer flow, the bed-shear velocity begins to grow af ter the reattachment point, while simultaneously the eddy viscosity (at the outer edge) will decrease in the longitudinal direction. From equation
(48), it follows, that the eddy viscosity is a 'weak' function of the Z
b de -s ear ve oc~tyh 1 · [vt,ref - ~u*zrefexp{-Qw rsef}].
w
4.3.2.1 Shear-stress distribution
Assuming a logarithmic flow velocity profile and applying the hypothesis of Boussinesq (11) and the eddy viscosity concept mentioned in 4.3.2 the
shear stress in the new wall-boundary layer reads:
T _L.g = exp(-Q ~ ) TO W S W Z Z ref T v v v t.ref]ex { S S 2 t.XZ ~ [~
-
[ w w]
} TO ~u*z ~u*z ~u z p a * v Z 0<-S w zref<
Ó w (49) Z Z ref Ó<
Ó w w < 1 (50)Figure 11 shows various profiles for the shear stress in a new wall-boundary layer where the outer flow is disturbed. The shear stress is
made dimensionless with the bed- or wall-shear stress. From equation
(49), it follows, that for z smaller than z f the shear stress
re
decreases (becomes smaller than TO)' while above Z f it is strong1y
re
dependent on the value of the eddy viscosity at the outer edge of the
boundary 1ayer (v). In case v is decreasing or in case the bed
-t,w t,w
shear velocity as well as the thickness of the new wall-boundary 1ayer
are increasing the shear stress will decrease at the outer edge.
If ó is at maximum or if the boundary layer reattaches the surface, the
w
eddy viscosity shou1d tend to zero there. Then the shear-stress is not
equal to zero according to (50),however, the value of the shear stress
In case equilibrium flow conditions are dominating the distribution of
the shear stress is not quite correct, since it is not linear.
0.4 0.8 1.2 1.1 2.4
o
= 0.01 + - 0.1 l:>. - 15. Conclusions
Uniform flow
According to the classical theories of Prandtl and Boussinesq and using a logarithmic flow velocity profile the eddy viscosity equals zero at the surface.
Considering the two-dimensional k-equation and the k-f-model the eddy viscosity does not reduce to zero, because the eddy viscosity is related
to the kinetic energy and the kinetic energy is not equal to zero at the surface.
Using the Coles flow velocity profile with TI= 0.20 (Nezu and Rodi, 1986) the velocities are approximated in a better way especially in the outer region (wake function!) than using a pure logarithmic flow or using the velocity profile which follows from an exponential eddy
viscosity.
Pipe flow
It appears that the agreement between a mathematical exponential function for the eddy viscosity (equation 19) and the corresponding
values based on measurements is satisfactory for a = 2.0. In the center
w
of the pipe the stress production of the turbulent energy is equal to zero, since aü/az ~ 0, however, the eddy viscosity is not, because there the kinetic energy (k), the dissipation (f) and the diffusion of the
kinetic energy are not reducing to zero (Hinze 1975, p. 725 and 737).
Using a logarithmic flow velocity profile or the Coles flow velocity profile the boundary condition in the center of the pipe is not satisfactory, because there the gradient of the longitudinal flow velocity is discontinue. Also the eddy viscosity will not be equal to
zero there. Then a correction has to be made on the eddy viscosity
distribution which follows from the Coles flow velocity profile (linear
shear stress).
Internal boundary layer
The Coles flow velocity profile using TI= 0.55 describes the
viscosity very well in a new wall-boundary layer with an undisturbed
outer flow. However, the description is poor in case the outer flow is
disturbed. Using the equations (47) and (48) the modelling of the eddy
References
Coleman, N.L. (1984), Velocity profiles with suspended sediment, J. of Hydraulic Research, Vol. 22, No. 4, p. 263-289.
Delft Hydraulics, Delft (1987), Documentation to the computerprogram ODYSSEE, part 3b: Mathematical and numerical description of the code Ulysse, 2e version.
Hinze, J.O. (1975), Turbulence, Second edition, Mc. Graw HilI Baak Co., New York.
Hoffmans, G.J.C.M. (1988a), Flow simulation by the 2-D turbulence model ODYSSEE, No. 2-88, Delft University of Technology, Dept. of Civil Eng.
Hoffmans, G.J.C.M. (1988b), Damping of turbulence parameters in relaxation zone, No. 10-88, Delft University of Technology,
Dept. of Civil Eng.
Hoffmans, G.J.C.M. (1988c), Flow model with prescribed eddy viscosity, No. 11-88, Delft University of Technology, Dept. of Civil Eng.
Kay, J.M. and R.M. Nedderman (1985), Fluid mechanics and transfer processes, Cambridge University Press.
Mierlo, M.C.L.M. and J.C.C. de Ruiter (1988), Turbulence measurements ab ave dunes, Report Q789, Vol. 1 and 11, Delft Hydraulics.
Nezu, I. (1977), Turbulent structure in open-channel flows (Translation of doctoral dissertation published in Japanese, Kyoto
University, Dept. of Civile Eng.
Nezu, I. and W. Rodi (1986), Open-channel flow measurements with a laser dopp1er anemometer, Journalof hydrau1ic Engineering, Vol. 112, No. 5, p. 335-355.
Rodi, W. (1980), Turbulence models and their application in hydraulics,
In appendix A an elaboration is given of the integration of equation (16) from Zo to Ów in order to write TIas a function of K, Ü, u*, Ów and Zo (for the meaning of the symbols, see notation).
The Coles velocity profile reads:
(Al) in which: w(~ ) = 1 + sin(2~ w w 2sin2{lrL} . 2 ó ' w wake function
Integration of (Al) from Zo to
s
gives: w Ós
s
w w wI
üC(z)dz u*[I
ln - dzZ + TII
{W("6)Z W("6)}Zo dZ] K Zo Zo Zo Zo w w (A3) 1 2 3 Term 1: q (A4) óws
ZoTerm 2: [z ln ~
-
z]I
s
ln _!tI-
Zo ln --
(ó - zo)Zo w Zo Zo w Zo Ó Zo ó [ln _!tI
-
1 + -] (A5) w Zo ów ZoAssuming (-)
«
1 -+ term 2 may reduce to:s
w Ó ó [ln_!tI-
1] (A5a) w Zos
s
w wI
{l + ZI
Zo sin(2"6 - 11" 2sin2{lr-} dz Term 3: 1)2}dz 2 ó Zo w Zo w Ó Ó Ó Z W Zo w{z _!tI cos(2"6 _ l)lr
11
Z 2sin2{lr "611
11" 2 2
Zo 1 Zo ów[
1
- --
ó + -~ cos(2
s-- 1
)
2~
w w Zo If (--)«
1 ó w (1 - --) 2sin2{~ --}s
w 2s
wterm 3 may reduce to:
s
w(A6) (A6a)
written as:
Substitution of (A4), (A5) and (A6) into (A3) (q - Üó ), TI can be
w
:Q
s
K, in ....::t!. + 1 u* Zo TI= Zo1
Zo 1)~ 1s
+ cos(2s-~ 2 w w Zo Assuming-- «
s
1,n
reads:n
ws
w Zo Zo (1 - --) 2sin2l~ --} ó 2 ó w ws
1n....::t!.+ 1 Zo (A7) (AB)In appendix B an e1aboration is given of equation (23), where the
horizontal flow velocity gradient to Z is integrated from Zo to z. Using the Tay1or-series to the exponentia1 function e1aboration of (BI)
resu1ts in (for the meaning of the symbo1s, see notation):
1 aü L{l ~ } exp(a ~
)
u* az ICZ 6w w 6w L{1 ~ }{1 + a ~ +2'
1 ~ )2 1 ~ )3 1 ~ )4 (B1) (a + 3' (a +4'
(a +.
) ICZ 6 w 6 . w 6 . w 6 . w 6 w w w w wConversion of (BI) leads to:
+ 12,(a ~)2 + -31(a ~6)3 + -41,(a ~6)4 + ... ) + . w 6 ! w . w w w w + a w (~ )2 +
1
a2 (~)3 1 3 (~ )4 ) 6 2! w 6 + 3!aw 6 + ... W W W L{l 1)~ 1 Z )2 1 1 (2) (~ )3 + (a-
+ (- a2-
a )(- + (- a3-
+ ICZ W 6 2! w w 6 3! w 2! w 6 w w w 1 1 (3) (~ )4...
) + (- a4-
+ 4! w 3! w 6 w 1 {1 + (a-
1)1 + (- a1 2 a ) ~ 1 1 ~2 62 + (- a3 - (2) 63 + IC Z w 6 2! w w 3! w 2! w w w w 1 1 ~3 + (- a4 - (3) +...
} 4! w 3! w 64 w (B2)Integration of (B2) from Zo to z gives (a is independent of z): w z z - Zo
1
{1n + (a - 1)( ) IC Zo w 6w Z2 - z~ 1)( 62 ) + w + - z~ 64 ) + ... } wZ3 -
zg
1 Ow Z4 -z~
- 1) ( ) + - 03 (- - 1) ( ) + } 63 4! w 4 6···· w w 1 Ow + - 02(-3! w 3 (B3)In general farm equatian (B3) can be written as [üE(z) = ü(z)]:
(B4) in which: n-l
°
°
Rw
[w
_
1] u,n n! n (BS)In appendix C an argumentation is given of the convergence of the correction-term B.
The correction-term B reads for n ~ ~:
(Cl) in which: n-l Q Q R ~ [~- 1] u,n n! n (appendix B) (C2) Because Zo Z Ó
s
Ós
1 w w estimated by: Zo (_}n]<
R equa l.onti Cl Ó u,n w can be lim n-l Q Q.s:
[~-
1] = n! n n~~ lim n Q __ w_ n n! lim n-l Q W (C3) n~~ n -+eo n(n-l)! Since, n n n Q Q Q~
<
~ and lim ~ ~ 0 (standard limit) andn n! n! n!
n-+~
ri-T n-l m
Q Q Q
W w
and lim ~ = 0 (standard limit, n-l)
n(n-l)!
<
(n-l)! m! mm -+eo
In appendix D an argumentation is given of the convergence of the ~ of
the higher order terms, which are part of the correction-term B in
equation (24).
The correction-term B reads:
(Dl) in which: n-l a a R ...Y!. [...Y!. - 1] u,n n! n (appendix B) (D2) Because Zo Z Ó :S Ó :S 1 w w estimated by: equation Dl can be
z
n=l n-1 a a ...Y!. [...Y!. _ 1] n! n n n-1 eo a eo a :l: ...Y!. :l: ...Y!. n=l n n! n-1 n! n n-1 <Xl a eo a :l: ...Y!. :l: w n=l n n! n=l n(n-1)! n n-1 eo a a :l:1
[n7 (n~l)! ] n=l n n m eo a eo _l_ a :l:1
...Y!. :l: ...Y!. [m=n-1] n=1 n n! m=O m+1 m! n aO m eo a eo a :l:1
...Y!.[
1
...Y!. + :l: _L m7] n=l n n! 1 O! m=l m+1n=l n-1 n co Ct ....:tI
[1
1 L n+1 ] 1 n-1 n! n n co Ct L ....:tI [n + 1 - n] 1 n-1 n! n(n + 1) n co Ct ....:tI 1 L n! [n(n + 1)] 1 (D3) n-1 Since, n n co Ct co Ct1 ....:tI ....:tI [standard limi tJ
L [n(n + 1)]
<
L - exp(Ct )n=l n! n=l n! w
In appendix E an argumentation is given of the following power-series: ~ n=l R n ~ -1 u,n (El) in which: n-l 0 0
R
.s:
[....!tI.-
1] (appendix B) (E2)u,n n! n
Substitution of (E2) into (El) gives:
n-l co 0 0 ~
.s.
[....!tI.-
l]n n-l n! n n-l n-l co 0 co 0 ~ ....!tI. ~ w n! 0 (n-l)! n=l w n=l n n-l co 0 co 0 ~.s:
~ w n=l n! n=l (n-l)! n n co 0 co 0 ~ ....!tI. ~ ....!tI. n=l n! n=O n! n 0° n 0° co 0 co 0~