See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/290431210

Draft tube calculations

Conference Paper · January 1999

CITATIONS READS
4 285

2 authors, including:

Thomas Staubli
Lucerne University of Applied Sciences and Arts
161 PUBLICATIONS   872 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Effects of suspended sediment on Pelton turbine wear and efficiency View project

Sediment monitoring and hydro-abrasive erosion in hydropower components View project

All content following this page was uploaded by Thomas Staubli on 14 January 2016.

The user has requested enhancement of the downloaded file.

 Turbine 99
Workshop on Draft Tube Flow
Porjus 20.-23. June 1999

DRAFT TUBE CALCULATIONS

Thomas Staubli Daniel Meyer

Fluid Mechanics and Hydromachines

HTA Luzern
CH 6048 Horw
Switzerland

ABSTRACT reattachment point can be found (Hornung, 1985). Whereas the

The flow of the workshop drafttube was calculated using planar separation comprises only diffusive transport of vorticity,
the commercial CFD code CFX-TASCflow. As inlet condition the three-dimensional recirculation is dominated by convective
of the model drafttube the measured velocity distributions were vorticity transport (Ginter, Staubli, 1999).
used with the exception of the radial velocity component. In
order to check the Reynolds number effects additional NOMENCLATURE
calculations were performed on a modified mesh for the The coefficients used in this paper are defined as follows:
prototype geometry. A series of interacting vortical structures • Pressure recovery factor Cpr
are observed in the calculated flow fields. These structures are p out wall − pin wall
visualized by their core streamline and selected streamlines C pr = 2
around that core. The most interesting features observed are the r Q 
 
breaking down of the cone vortex rope and the formation of a 2  Ainlet 
strong vortex in the three-dimensional recirculation zone in the
lower draft tube corner. • Energy loss coefficient ζ
E inlet − E outlet
ζ=
INTRODUCTION E kin −inlet
The physics of the draft tube flow poses a major challenge
 ρ    ρ 2 
∫∫  p + 2 u ∫∫
to numerical flow calculations, since it encompasses a series of
u  dA −  p + 2 u u  dA
2
flow phenomena difficult to model:
Ainlet   Aoutlet   
• the inlet flow is highly three-dimensional and unsteady in =
ρ 
nature,
∫∫  2 u
2
u  dA
• the flow separates at the rotating hub, 
Ainlet
• the diffuser flow may cause local or global recirculation
zones, • Efficiency η
• the bent gives rise to secondary flows and potentially to p out wall − pin wall
η=
ρ  Q  
local recirculation zones, 2 2
 Q 
• at the outlet there is a the possibility of backflow into the  
2  Ainlet  −  A  
draft tube,
   outlet  
• for swirling flows and local recirculation zones the chances
for occurrence of unsteady flow phenomena are high. Note: instead of using pressures averaged along the walls
p in , out wall the pressure recovery factor Cpr and the efficiency η
Separations, or rather recirculation zones, observed in draft
tube flows will be generally of three-dimensional nature, that is, were also calculated using area averaged pressures pin , out .
different streamlines for the separation point and the
Data are given in Tab. 1.

1 ERCOFTAC/IAHR Workshop Porjus 1999

• Kinetic energy correction factors aaxial and aswirl

∫∫ u ∫∫ (u )dA
3 2
axial dA tangential u axial

a axial = A
3
a tangential = A
3
Au axial Au axial

• Momentum correction factor β

∫∫ u
2
dA utangential u
β= A
2
Au axial
• Swirl intensity S

∫∫ (ru u tangential )dA

uaxial
axial
π
S= A
dA
∫∫ u axial dA
2
A
A Fig.2 Discretization and block structure at the inlet.
where r = radial distance from the axis, respectively the
centerline. As inlet condition of the model drafttube the measured
velocity distributions were used. The radial component was
• u, dA = vectors estimated in a way that a linear angle variation in radial
Q direction could be achieved, providing a flow congruent to the
• u axial = hub wall. Such flow angle variations are typically observed in
A
probe measurements at the turbine outlet flows. Inlet turbulence
intensities were chosen to be in agreement with the measured
SIMULATION velocity fluctuations, although these fluctuations arise from
The flow of the workshop drafttube was calculated using blade passage periodic flow and do not correspond to statistic
the commercial CFD code, CFX-TASCflow, version 2.8. A turbulence.
quasi-unsteady prediction of the flow was carried out by solving Velocities close to the walls were extrapolated from the
two iterations per time step. The time step was set to 0.3 s. measured data on the basis of the log law approximation. Since
On the hardware side we had at our disposition a sun work- the integration of the measured data did not meet exactly the
station ultra 60 with a 360MHz CPU and 512 Mbyte RAM. given experimental discharge values of test points T(r) and R(r)
For grid generation we employed ICEM CFD powermesh, the axial velocity vectors were proportionally stretched to yield
version 3.3. The block structured mesh with 32 blocks is the required values.
displayed in Fig. 1. A number of 493 884 elements have been
totally used for discretization of the computational domain. TASCflow parameters (CFX-TASCflow User Documentation,
Grid refinement was done in the inlet section and close to the 1995) set for the simulation:
walls, Fig. 2. The non-dimensional distance to the wall y+ was Discretization control parameters:
adjusted to remain within the range of 11 and 330. The log law pure linear profile scheme ISKEW = 4
was used for boundary layer approximation. physical advection correction LPAC = true
Solution control parameters:
number of iterations per time step kntlin = 2
time step dtime = 0.3 s
Fluid properties:
viscosity viscfl = 0.001 [kg/(ms)]
density rhofl = 0.999 [kg/m3]
specific heat cvfld and cpfld = 4182 [Nm/(kg K)]

In the course of the calculations three major causes were

found to give rise to convergence problems. The first was
connected to the mesh generation. Elements at the block
Fig.1 Block structure of the simulation domain boundaries within the bent having very small angles were found
to be responsible for convergence troubles. A second cause
came from the backflow at the outlet. This problem was

2 ERCOFTAC/IAHR Workshop Porjus 1999

overcome by using "opening" (allowing in- and outflow at a MODEL - PROTOTYPE
constant outlet pressure) as outlet boundary condition until In order to check the Reynolds number effects additional
good convergence was achieved before then changing back to calculations were performed on a modified mesh for the
"outflow" and parameter "i.o. walls = f" as boundary condition prototype geometry which was enlarged by a factor 11. The
(allowing variable pressure at the outlet). Providing radial overall quantity of nodes was equal in both simulations and
velocity components at the inlet also proved to be an amounts to 493 884 nodes. In order to model the relatively
unfavorable measure with respect to convergence and slowed smaller boundary layers of the prototype flow the outer points
down the convergence rate considerably. had to be located closer to the walls for the prototype.
The global quantities varied considerably (see Tab.1) and
GLOBAL DATA RESULTS the results reflect that viscosity effects become smaller for
Tab. 1 gives the data for the coefficients defined in the higher Reynolds numbers. The velocity distribution at the outlet
nomenclature section. Since authors have doubts that the inlet is less uniform for the prototype case, but, in contrast, the swirl
pressure variation resulting from the given velocity distributions component aswirl and the swirl intensity S are less pronounced
are physically correctly modeled they propose to compare also for the prototype.
the area averaged pressures and the accordingly modified At first glance it is astonishing that the prototype ratios at
coefficients Cpr and η. cross section Ia are not equal to one. But since the inlet bound-
Not included in this table are calculations which were ary layers of model and prototype were approximated slightly
performed with reduced inlet turbulence. When reducing this different (extrapolated points close to the walls were introduced
turbulence by a factor of 4 the pressure recovery and efficiency for the prototype) the deviation becomes explainable.
rose both by about 0.01 while the losses were reduced by the
same amount. FLOW FIELDS
A general overview on the draft tube flow downstream
Tab.1 Global results from the bent is given in Fig.3. The velocity vectors show the
case T(r) case R(r) Prototype bep velocity components within the planes and the colors give the
N 595 595 rpm N/NP 4.76 velocity vector magnitudes (blue = small velocities, green =
high velocities). There is no indication for a separation or
Q 0.535 0.554 m3/s Q/QP 0.0036 recirculation along the critical upper wall in the center plane.
H 4.5 4.5 m H/HP 0.1875
Using wall pressures:
Cpr 0.697 0.689 - Cpr/CprP 0.94
η 0.711 0.702 - η/ηP 0.94
Using area averaged pressures:
Cpr 0.561 0.522 - Cpr/CprP 0.89
η 0.571 0.531 - η/ηP 0.89

ζ 0.438 0.474 - ζ/ζP 1.36

cross section Ia
aaxial 1.032 1.030 - aaxial/aaxialP 1.02
aswirl 0.074 0.043 - aswirl/aswirlP 1.07
β 1.106 1.074 - β/βP 1.03
S 0.222 0.163 - S/SP 0.98

cross section III

aaxial 1.812 1.936 - aaxial/aaxialP 0.85
aswirl 0.200 0.258 - aswirl/aswirlP 1.21
β 1.391 1.466 - β/βP 0.95
S 0.274 0.303 - S/SP 1.18
Fig.3 Overview on the draft tube flow: case T(r)

1 (C.S. Ia) 12 (C.S. IVb)

2 (C.S. Ib) 11 (C.S. III) 3 ERCOFTAC/IAHR Workshop Porjus 1999
3 6
Fig.4 Visualization of the tangential velocity distributions in planes along the centerline (looking
4
downstream, shaded areas indicate upstream flow): case T(r)
5
 = overload). The same basic features of vortex formation is
observed in case of both inlet data.
In the portion of the draft tube downstream of the cone the
vortex flow is dominated by the inlet swirl component, which is
rotating in clockwise direction when looking downstream. The
resulting cone vortex core can be clearly followed until the
beginning of the bent where it is breaking down (Fig. 4, plane 1
to 6 and Fig.5). By viscosity this vortex is feeding energy to a
counter rotating vortex forming on the right side in the planes 4,
5, and 6 of Fig.4. This secondary vortex is developing into one
the two dominating bent vortices which are observed until the
draft tube outlet section.
Fig.5 displays the vortex rope starting downstream of the
cone separation. The vortex is visualized by its core and a
selected number of surrounding streamlines. The color coding
of the streamlines shows the time the fluid remains in the draft
tube and reflects the fact that the axial velocities become very
small close to the vortex core.
A second very important vortex formation starts within the
separation region of the lower corner of the draft tube (Fig.6).
In this recirculation zone the vortex is fed by viscous forces
with energy, is then detaching and forming the clockwise
rotating vortex observed in all downstream sections. The color
coding of the tangential velocities in the cutting planes
corresponds again to the magnitude of the velocity vector (blue
= small velocities, green = high velocities).
Fig.7 shows the two vortex cores which dominate the flow
field in the diffuser section of the draft tube downstream of the
bent. The one on the left side rotating clockwise is the vortex of
Fig.6. The counter clockwise rotating vortex observed on the
right side of Fig.7 results from the secondary vortex formed in
the planes 4 to 6 of Fig.4 and is additionally fed by energy by
the bent flow.

Fig.5 Vortex rope downstream of the cone: case T(r)

The calculated draft tube flow fields are clearly dominated

by the interaction of counter rotating vortices. Fig. 4 displays an
overview of the flow development in various planes along the
draft tube centerline for the case T(r). The cutting planes show
the tangential velocity components when looking downstream.
The color coding represents the magnitude of the velocity
vector, that is the speed.
It was decided to show only images of flow visualization
for the case T(r) (top point on the propeller curve = bep) in this
paper because the general flow effects look very similar also in
the case R(r) (point on the right leg of the same propeller curve

4 ERCOFTAC/IAHR Workshop Porjus 1999

Fig.6 Vortex formation in the lower draft tube corner: case T(r)

Fig.7 Counter rotating vortices propagating downstream from the bent: case T(r)

5 ERCOFTAC/IAHR Workshop Porjus 1999

Pressure Fields
As mentioned before, the authors have their doubts with
respect to the correctness of the prediction of the pressure
distributions at the inlet cross section Ia. The lowest pressures
are found at the walls, which results of course in very good
values for the pressure recovery factors and efficiency.
However, this pressure distribution with positive gradients
towards the center as depicted in Fig.8 is in contradiction with
The computed static pressure distribution in the outlet cross
section IV varies only slightly, confirmed by the experimental
wall pressures data which also differ not more than ±100 Pa
from a mean value. Comparing the local wall pressures reveals
however that the distributions have their maxima and minima at
different locations.
looking downstream
1
1 0.5

p [mbar]
0.5 0
500 400 300 200 100 0 -100 -200 -300 -400 -500
pressure distributions typically found in swirling flows. In 0 -0.5

p [mbar]
-0.5 0 -1
-1 -1.5
addition, also the streamwise curvature of the streamlines may -1.5
-2 100
-2
-2.5

not be made responsible for such strong radial pressure gradient -2.5

200
leading to the conclusion that inlet pressures are not correctly 1
1
0.5

p [mbar]
0
predicted. 0.5
0
300 -0.5

p [mbar]
-1
-0.5 -1.5
-1 400 -2
1 -1.5 -2.5
C.S. Ia -2
-2.5 500
2 1
600 0.5
1

p [mbar]
0
0.5 -0.5
3

p [mbar]
0 -1
-0.5 700
-1.5
C.S. Ib -1 -2
4 -1.5
-2 800
-2.5

-2.5

5 900 1
1 0.5

p [mbar]
0.5 0
1000

p [mbar]
0 -0.5
-0.5 -1
-1.5
6 -1
-1.5 -2
-2 -2.5
-2.5 1
1 1
0.5 0.5
0.5
0 measured

p [mbar]

p [m bar]
0

p [mbar]
0
-0.5 -0.5
-0.5
-1 -1 -1 calculated
-1.5 -1.5 -1.5
-2 -2 -2
-2.5 -2.5 -2.5

Fig.9 Distribution of the static pressure in the outlet

1 4
section of the draft tube: case T(r)

2 5
3 6
Fig.8 Development of the static pressure distributions
in the inlet section of the draft tube: case T(r)
CONCLUSIONS
Of course, the final conclusions can only be written after
the detailed comparison with the experimental data, but the
following preliminary conclusion can be drawn beforehand:
namely, that the secondary flow in this draft tube test case is
driven by several interacting vortical structures. Since the
employed kε-solver is not especially adapted to model such
flows the quantitative prediction may be expected to be
moderate. Most interesting features observed in the calculations
are the breaking down of the cone vortex rope and the
formation of a strong vortex (rotating clockwise as the inlet
swirl) in the three-dimensional recirculation zone in the lower
draft tube corner.
A comparison with static pressure measurement along the
lower and upper wall center lines is displayed in Fig. 10. Since
the measured coefficients larger than one are unphysical the
comparison is done with respect to cp -cp out.
Comparing data along the upper wall one sees that the
jump in cp values at 0.11 of the non-dimensional length at the
transition from the conical to the rectangular section is in
perfect correspondence in the calculation and the measurement.
Further downstream the discrepancies become important
indicating that there is a separation on the upper wall
downstream of the bent which is not predicted by the numerical
calculation.

6 ERCOFTAC/IAHR Workshop Porjus 1999

 Along the lower wall center line an oscillation is found in
the numerical calculations being caused by the corner vortex
shown in Fig. 6. Unfortunately the experimental data do not
have good enough spatial resolution to show this effect.
However, since such a vortical structure will be in reality rather
unsteady, static wall pressure measurements will anyhow
provide only a smeared or smoothed picture of such local
pressure variations. In general, discrepancies along the upper
wall are considerably larger than along the lower wall.

0.60

0.40

0.20

0.00
cp -cpout [ ]

-0.20

-0.40

-0.60
cp upper
-0.80 cp upper experiment
cp lower
-1.00
cp lower experiment
-1.20
0 0.2 0.4 0.6 0.8 1

Parameter along wall center line [ ]

Fig.10 Pressure coefficients along the lower and

upper wall center lines: case T(r)

ACKNOWLEDGMENTS
The authors wish to express their appreciation to Dr. Peter
Holbein for his invaluable advice when convergence problems
procured them moments of desperation.

REFERENCES
Hornung H. "Abgelöste Strömungen", ZAMM, Vol 65, 1985
Ginter F., Staubli T. "Performance discontinuity of a shrouded
centrifugal pump impeller - part I, numerical modeling and flow
instability", 3rd European Conference on Turbomachinery,
Fluid Dynamics and Thermodynamics, London 1999
AEA Technology, "CFX-TASCflow User Documentation", AEA
document, 1995

7 ERCOFTAC/IAHR Workshop Porjus 1999

View publication stats