Académique Documents
Professionnel Documents
Culture Documents
=
Fig 1: Sketch of the diffuser with experimental measurement stations and
reference systems used. Dimensions are in mm.
0
1
=
r r
rUr
Z
U
Z
Fig 2: Radial velocities imposed at the diffuser inlet and resulting
velocity profiles at station S7.
Table 1: Turbulence parameters imposed at the inlet boundary.
ARTIGOS TCNICOS TECHNICAL ARTICLES
sured k profile and Armfield's equation for or, alternatively, using
the combination I=0.05 with Le=10%Di. However, a fundamental
difference exists between these two modelling approaches. The ef-
fective viscosity is the sum of the fluid viscosity and the eddy vis-
cosity, the latter being related to both k and in the following way:
The effective viscosity is therefore modified by the turbulent
quantities imposed at the diffuser inlet. As can be seen in Figure
3b, it is considerably larger in the case using Le = 10%Di, despite
the similarity in the predicted Us profiles. As expected, a higher vis-
cosity tends to damp the fluctuations in the flow as visible in region
A of Figure 3a. The opposite behaviour is also seen in region B
where the fluidity of the flow in the case Le = 0.1%Di allows the
boundary layer to briefly separate at station S4, as showed by the
local negative velocity, but to reattach before station S7.
Despite the good results noted here, using the approximation
Le = 10%Di appears to be a risky operation due to the resulting
damping that tends to makes the flow artificially robust, and be-
cause of the difficulty to determine the correct turbulence length
scale to privilege in other applications. Moreover, the problematic
approximation does not seem to lie in the eddy length scale itself
but rather in the constant turbulence intensity imposed. Looking
back at Armfield's equations presented in Table 1, it can be noted
that this model also needs a fraction of the inlet diameter to char-
acterize the turbulence. It thus seems that the error may mostly be
attributed to the use of a uniform intensity I, which gives a turbu-
lence kinetic energy, profile very different from the measurements.
In this case, imposing Le = 10%Di only hides the induced error and
should not in any case be associated with a reliable modelling pro-
cedure.
TREATMENT OF THE OUTLET CONDITION
The impact of the outlet geometry is determined by comparing
five extensions downstream of the diffuser. The size and geometry
of the discharge tanks added (with solid and/or permeable walls)
are shown on Figure 4, the simplest being no extension at all.
As a first step, boundaries 1 & 2 are specified to be standard no-
slip walls as to represent a physical tank. The effect of varying the
geometry is in most cases negligible for the axial and tangential ve-
locity profiles, but the pressure distribution in the outlet plane is in-
deed changed. On the left hand side of Figure 5, it is seen that the
wall pressure normalized as,
begins to be affected by the extension geometry at 70% of the
diffuser wall length, L. In the outlet plane, the pressure distribution
along r varies considerably between all simulations as shown by
the right hand side of the figure.
The geometry of the extension thus affects the pressure results
near the diffuser's exit. However, the shape of the extension do-
Fig 4: Outlet extensions geometry.
i
atm wall
p
U
P P
C
2 / 1
=
main is not the only parameter be considered to model the test
case properly. The boundary conditions imposed on the extension
also have to be as accurate as possible and this is why two addi-
tional simulations using the medium tank are presented on Figure
5. In both cases, the standard no-slip walls are removed on bound-
ary 1 and 2. In the first case, a zero total pressure is imposed on the
extension's back and top (boundary 1 & 2) to make the best repre-
sentation possible of a large volume of fluid at rest. In the other
case, only the back of the tank (boundary 1) uses this condition
while boundary 2 is a free slip-wall. Both pressure curves are
superposed but differ from the one using standard no-slip walls.
This leads to the conclusion that the treatment of the extension's
boundaries is significant, but in this particular case, the results re-
main unaffected as long as fluid entrainment by the free jet is free
to occur. It is to be noted that the flow topology within the exten-
sion box is considerably altered between the two cases, but this
has no effect on the results inside the diffuser itself since the veloci-
ties are small (see flow fields presented in [5]).
The radial variation of the pressure in the outlet plane is not to
be neglected since it is directly related to the machine's efficiency.
One way to quantify the quality of a diffuser is to look at the amount
of kinetic energy converted into pressure via the recovery coeffi-
cient, defined as
The static pressures P1 and P2 can be determined from an inte-
gration of the values in the entire plane or from an average of the
wall pressure values. These two options were evaluated for the dif-
fuser with extensions using no slip walls and results are presented
in Table 2. The variables wall(1) and wall(2) use the parietal pres-
sure at positions defined in Figure 4. The first of these two coeffi-
cients includes the last portion of the geometry, thus being more af-
fected by the extension than the same measure taken farther in-
side the diffuser. Although the velocity profiles inside the diffuser
are not significantly affected by the extension's shape, it is impor-
tant to keep in mind that it is likely to influence the pressure recov-
ery coefficient, especially when calculated from wall pressure and
close to the outlet.
Fig 5: Pressure evolution along the diffuser wall and
profile in the outlet plane.
2
1 2
2
1
=
ref
A
Q
P P
X
+
=
2
,
2
1
_
It is clearly seen from this example that modifying the inlet tur-
bulence level can, by itself, modify the whole flow field even if the
imposed velocity profiles are unchanged. In the present case, us-
ing 10% of the inlet diameter as the eddy length scale leads to an
obviously too viscous flow and the calculated losses are underesti-
mated of 0.5% (absolute), which could lead to costly penalties in a
contractual context.
FINAL CONSIDERATIONS
Simulating diffusers using computational fluid dynamics has al-
ways been a difficult task due to the unstable nature of the flows un-
der adverse pressure gradients. It has been showed by studying
two test cases the ERCOFTAC's swirling flow in a conical diffuser
and the Chute--la-Savane draft tube that the imposed bound-
ary conditions have an important impact on the calculated flow to-
pology. Among these parameters, the inlet radial velocity was
shown to have the greatest impact on the conical diffuser's flow
since it is directly related to the appearance of wall separation or
core flow recirculation. The second most important parameter is
probably the inlet turbulence and its role is best seen in the draft
tube geometry. Overestimating the incoming turbulent mixing arti-
ficially rises the viscosity and leads to a much more uniform flow
which tends to underestimate the draft tube losses. The third as-
pect to recall is the outlet treatment. The addition of an outlet dis-
charge extension did not influence much the velocity profiles inside
the conical diffuser, but it did modify the outlet static pressure dis-
tribution and consequently altered the calculated recovery coeffi-
cient. It is not excluded that, in some cases, it might even have an
impact on velocity profiles within the diffuser, especially in the last
part of the geometry.
On the other hand, refining the mesh and imposing an
axisymmetric boundary condition did not appear to have a first or-
der influence on the computed results. These two parameters
should however be taken into account as precisely as possible
when suspecting the presence of very sensitive phenomena such
as the efficiency drop near the best efficiency point.
ACKNOWLEDGEMENT
The first author would like to thank the Natural Sciences and En-
gineering Research Council of Canada for its financial support to
this research project.
BIBLIOGRAPHICAL REFERENCES
[1] AVELLAN, F., 2000, Flow Investigation in a Francis Draft
Tube : The FLINDT Project, Proceedings of the Hydraulic Machinery
and Systems, 20th IAHR Symposium, Charlotte, USA.
[2] Turbine-99 website : www.turbine99.org
[3] MAURI, S., 2002, Numerical Simulation and Flow Analysis
of an Elbow Diffuser, Doctoral Thesis, cole Polytechnique de
Lausanne, Switzerland.
Fig. 6 : Axial velocity contours in the draft tube channels with
turbulence calculated from a) upstream components and
b) the approximation I=5% and Le=10%Di.
2
K
t
Fig 3: Effect of the inlet turbulence on axial velocity profiles
and eddy viscosity.
28 29