CFD-SIMULATION OF A T-JUNCTION -

BORIS HUBER
Institute of Hydraulic and Water Resources Engineering, Department of Hydraulic
Engineering, Vienna University of Technology, Austria

1. INTRODUCTION
In this paper the hydraulic properties of a T-junction are investigated. The T-junction,
planned in the Kops 2 high head power station by the Austrian Vorarlberger Illwerke AG, is
located in the duct system between each turbine- and pump conduit of the three machine units.
The projected pump power-plant Kops 2 is designed to equalize peaks of energy-fluctuations
mainly. Therefore fast regulation processes are necessary and flow conditions are changing.
To determine the flow characteristics and head losses the T-junction was investigated in a
physical model test as well as in a CFD-simulation. The results of the hydraulic model tests
were compared with the numeric calculation in order to assess the results. Then an alternative
design of the junction was simulated to find out which junction has better flow properties or
lower head-losses respectively.

2. INVESTIGATED OPERATION CONDITIONS

Following 5 operating conditions which differ in regard to the direction of flow were
investigated: turbine operation, pumping and 3 special operating conditions for fast regulation
processes in a hydraulic short-circuit flow (coming from the pumps, going through the
turbines and then through the pumps again):
CASE 1 Turbine operation:
Flow approaches the T-junction from the reservoir and leaves
through the turbine; there is no flow in the branch to the pump.
The maximum flow through each of the 3 turbines is 26.67 m/s in
nature.
CASE 2 Pump operation:
Flow comes from the pump and exits to the reservoir while there
is no flow through the turbines. The maximum pump discharge is
19.33 m/s per pump.

res

tu

C1
pu

res

tu

C2
pu

CASE 3 Hydraulic short-circuit

Flow from the pump (19.33 m/s per pump) diverts into the two
remaining branches. The ratio of the split flow is variable.

res

tu

C3
pu

CASE 4 - Hydraulic short-circuit

In this special operating condition flow enters from the pump
and leaves through the turbine; there is no flow in the branch to the
reservoir.

res

tu

C4
pu

CASE 5 - Hydraulic short-circuit

In addition to flow from the pump (max. 19.33 m/s per pump)
there is a combining flow from the reservoir which amounts up to
Qres = - 7.33 m/s for each turbine (flow direction is defined
negative).

res

tu

C5
pu

Fig. 1 Operation Conditions

3. PHYSICAL MODEL
Experiments were conducted with a length-scale of l = 1:9.9. The diameters of the pipes
were 192 mm in the reservoir and turbine branches and 172 mm in the branch coming from
the pump. The T-junction and the adjacent pipes were made of Plexiglas. In nature there is a
90 bending between the pump and the T-junction. In order to reproduce the resulting flowpattern correctly, a 90 bow was used in the experimental setup also (see Fig. 4). The bending
turned out to be of significant influence on the experimental results.

192

192

400

192

590

172

Fig. 2 Sketch and photo of the T-junction in the experimental setup (dimensions in mm)

The experimental conduit was mounted on the laboratory wall at a height of 2.5 m and the
T-junction was tilted with an angle of 16 to the vertical axis corresponding to the prototype.
The complete experimental setup consisted of pipes, valves and measuring instruments:
inductive flow meters (IDM) as well as pressure gauges, a difference pressure transmitter and
piezometric tubes (see Fig. 3 and 4).

Piezometric
tubes

IDM
Pressure gauges

Computer

Difference pressure
sensor
Fig. 3 Hydraulic model in the laboratory
Pressure was measured at 3 sections: M1 and M2 located in the reservoir-turbine branch,
2.08 m from the intersection point of the pipe-axes away and M4 in the branch to the pump
with a distance of 0.61 m from the axes intersection point (see Fig. 4).
200

59

200

208

IDM

outflow (tu)

208
M2

40

M1

inflow (res)
outflow (res)

M4
21

M1

M1b

200

21
M1a

M2

21

M4c
M4a

M4b

M2a

IDM

inflow (pu)

Fig. 4 Sketch of experimental setup (dimensions in cm)

M2b

The head-loss coefficient of the T-junction is determined by:

K ij =

total pressure in branch i total pressure in branch j pipe friction from i to j

2

v m 2g

Generally the loss coefficient is based on the pump-velocity (vpu). However, as there is no
flow in the pump in case 1 here the loss coefficient is based on vres (or vtu which is the same in
case 1).
To obtain the total pressure the local velocity-head (v2/2g) was added to the measured
(static) pressure. The friction head loss of the straight pipe section between two measuring
sections was calculated with a roughness of k=0.003 mm (Plexiglas).
Numerous experiments with different flow and pressure conditions were carried out. In
every case flow was gradually increased within a range where a constant loss coefficient was
obtained. Discharge varied between 30 and 100 l/s.
3.1. EXPERIMENTAL RESULTS
In case 1 the T-junction works like a local expansion. The flow pattern can be compared
with the one of a straight pipe and the head losses are comparatively small the medium head
loss coefficient Kres-tu was 0.141.
In the other cases the flow pattern was significantly different: there is a sudden change of
flow direction combined with strong swirl flow resulting from the junction itself on the one
hand and from the bows in the inlet conduit on the other hand. Due to the asymmetrical
approach flow (caused by the 90 bow at the inlet section) case 2 and case 4 was not mirrorinverted. The head loss between the pump and the reservoir in case 2 (Kpu-res = 0.83)2 was
lower than the head loss between the pump and the turbine in case 4 respectively (Kpu-tu =
1.08)2. The determined head losses for case 1, 2 and 4 - compared with values obtained by
Miller (1978) for a sharp-edged T-junction - are summarized in Table 1. Loss coefficients for
case 3 and 5 are depicted in Fig. 5 and 6.
loss coefficient K (90)
min
case 1 K pu-tu

max mean Miller

loss coefficient K (0)

min

max mean Miller

-0.41 -0.39 -0.40 -1.00 K res-tu 0.15

0.13 0.14 0.05

case 2 K pu-res 0.78 0.89 0.83 0.70 K res-tu -0.15 -0.21 -0.18 -0.33
case 4 K pu-tu

0.99 1.15 1.08 0.70 K tu-res -0.17 -0.30 -0.24 -0.33

Table 1 Loss coefficient for cases 1, 2 and 4 (based on vres in case 1, else on vpu)

based on vres or vtu respectively

Based on vpu

loss-coefficient case 5 (based on v pu)

loss-coefficient case 3 (based on v pu)

2.00

2.00

K pu-res
K tu-res
K pu-tu

1.50

K pu-res case 4
K tu-res case 4
K pu-tu case 4

1.50
1.00

0.50
0.00
-2.00

-1.80

-1.60

-1.40

-1.20

-1.00

-0.80

-0.50

-0.60

-0.40

-0.20

0.00

loss-coefficient K

loss-coefficient K

1.00

0.50
0.00
0

-1.00

-1.00

-1.50

-1.50

-2.00

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

-0.50

K pu-res case 4
K tu-res case 4
K pu-tu case 4

K pu-res
K tu-res
K pu-tu

K pu-res case 2
K tu-res case 2
K pu-tu case 2

-2.00

q=Q(res)/Q(pu)

q=Q(res)/Q(pu)

Fig. 5 and 6 Loss coefficients (based on vpu) for cases 5 (left) and 3(right)
As mentioned before, strong swirl flow was observed in case 2 to 5. To visualize swirl
pressure air was inflated. In the turbine branch swirl was oriented clockwise looking in the
direction of flow.
To quantify swirl, a special measuring-equipment which was developed by our institute
was used. It consists of 2 steel spheres connected with a rod like a bar-bell, pivoted in the
middle. Due to the shape of the spheres fluid forces always run through the center of the
spheres and if there is swirl in the flow, the swirl-meter turns accordingly. With an optical
sensor applied at the pipe outside the passing-by of the spheres is recorded and the tangential
velocity can then be derived from the number of revolutions per minute. The ratio of
tangential and axial velocity ranges from 0.7 to 1.0.

4. CFD-SIMULATION
The numerical simulation (3-dimensional, steady) was conducted with the program
package FLUENT 6.1.22. At first the system was modelled in nature scale and compared with
a simulation in model scale. As there were no significant differences the further simulation
was conducted in model scale, because the simulation in full scale demanded extremely long
computation times.
The simulation was carried out in consideration of gravity (except for case 1). The origin of
the coordinate system was located at the intersection point of the pipe-axes and the z-axis ran
in direction of the axis of the branch to the pump (thus deviating from the gravitational axis
under 16).
4.1 MESH GENERATION
The mesh generation was carried out with GAMBIT. Depending on the particular case,
about 1 - 2 Million cells had been used. Especially in the near-wall region and zones of high
pressure or velocity gradients the mesh had to be very dense and therefore computation time
amounted about 24 hours on a machine with 4 parallel processors. As turbulence models the
Standard-k--Model and the RNG-k--Model (swirl dominated flow) were used.

Fig. 7 (a and b) T-junction with mesh

4.2 MODIFIED T-JUNCTION GEOMETRY

If the results of the CFD-simulation turned out to be satisfactory, a modified T-junction
geometry should be simulated in order to find out which one had better hydraulic properties.
By the use of the numeric simulation instead of another physical model test with the
alternative design it was possible to save costs. The dimensions of the original T-junction and
the variant are shown in Fig. 8.

192
192

172

172

Fig. 8 (a and b) Sketch of original T-junction (left) and variant (right); dimensions in mm
4.3 CFD-SIMULATION CASE 1
In case 1 the Standard-k--Model with standard-wall-function was used. In order to reduce
computational time, the branch to the pump was shortened, as there is no flow in case 1 in this
branch. Additionally, gravity was turned off in case 1 (which had no effect on the results).
The boundary conditions were chosen as follows: constant velocity distribution at the inlet
(res), constant static pressure at the outlet (tu) and a wall at the pump.
To determine loss coefficients mean static pressure was calculated at three cross-sections:
P1 at a location of x=-1.3 m, P2 at x=+1.3 m and P4 at z=+0.4 m. The local velocity head was
added to the static pressure and the friction head-losses of the particular pipe section were
subtracted. Simulations were carried out with different conditions of velocity and pressure,
whereas the loss coefficients remained constant. The loss coefficients of the CFD-simulation
are summarized in Table 2; they matched well with the experimental results. The loss
coefficients for the alternative T-junction design, which were slightly smaller, are also
included in Table 2.
head loss

head loss

head loss

(P4-P1-

(P4-P2-

(P2-P1-

v(res) v (tu) v(pu) pipe frict.) K pu-res pipe frict.) K pu-tu pipe frict.) K tu-res
file
modell_B01_F1
modell_B01_F1-v15

[m/s]

[m/s]

[m/s]

[cm]

(v res)

[cm]

(v res)

[cm]

(v res)

9.50

9.50

0.0

-334.1

-0.73

-219.8

-0.48

-48.58

-0.11

15.00

15.00

0.0

-829.1

-0.72

-556.1

-0.48

-120.90

-0.11

9.50

9.50

0.0

-275.0

-0.60

-174.1

-0.38

-35.27

-0.08

VARIANT:
modell_Var_B01_F1

Table 2: Loss coefficient of the CFD-simulation for case 1 (based on vres)

Contour-plots of velocity in the plane of the pipe axes show a flow pattern which is widely
similar to straight pipe-flow. However, because of the expansion in the T-junction, flow is
slowed down in the near-wall zone and concentrated to the inner zone, slightly deflected to

the top (see Fig. 9). In the dividing branch velocity is approximately zero. The flow pattern of
the variant showed no significant differences.

Fig. 9 Contours of Velocity case 1

4.4 CFD-SIMULATION CASES 2-5
In cases 2-5 there is a sudden change of flow direction. The flow pattern can be compared
with an impinging jet - superposed with strong generation of swirl. It is not expected that the
Standard-k--Model performs well here. For flow with a strong swirl for instance the RNG-k (swirl dominated) or the Realizable-k--Model or even the Reynolds-Stress-Modell (RSM)
is recommended.
The head losses were determined at the same cross sections M1, M2 and M4 where the
measurements took place in the experiments. At first, the Standard-k--Model was used. As
expected, the loss coefficients were clearly lower then those gained from the experiments. A
significant improvement was achieved by using the RNG-model (short description: RNG-k-Model, swirl dominated flow, swirl factor 0.07, Standard wall function, 1 Mill. Cells,
roughness of walls k=3e-6 m). Simulations were conducted until full convergence was
reached which was surveyed by monitors showing the time dependent development of total
pressure in a measuring point.
With the RNG-Model the head losses derived from the CFD-simulation were clearly
smaller than those from the experiments. In particular tangential velocity was significantly
higher in the experiment than in the CFD-simulation. This is most likely due to the number of
bends in the experimental setup. To account for the higher swirl, the boundary condition at
the pump was modified: instead of a constant velocity an additional tangential velocity (half
as much as the orthogonal velocity) was applied.
With this modification the loss coefficients Kpu-res and Ktu-res agreed well with the
experiments (see Fig. 10a and b). The head loss from the pump to the turbine Kpu-tu was
predicted very well in all cases except in case 5 when flow from the reservoir was high
compared to flow from the turbine (q = Qres / Qtu). However, when flow from the reservoir
was high in case 5 (q < -0.5) the CFD-simulation drifted away from the experiments (see Fig.
10 c).

The loss coefficients of the experiments and CFD-simulations of case 2 to 5 are depicted in
Fig. 14 a to c. In these pictures the loss coefficients of the variant are also included. In general,
the pattern of flow of the variant did not differ significantly from the tested T-junction.
Contour-plots of velocity of the T-junction and the variant as well as path lines to visualize
swirl flow are depicted in the following. The loss coefficients of the variant were higher in
case 2 to 4.
Further improvement of the CFD-simulation could possibly be obtained by meshrefinement until the whole boundary layer is resolved and by the use of the RSM-Model. To
obtain this, a huge number of cells and extremely long computation time would be needed.
loss-coefficient K pu-res (based on v pu)
2.00
1.50
1.00

K pu-res

0.50
0.00
-2

-1.5

-1

-0.5

0.5

-0.50
-1.00

experiment
-1.50

CFD-Simulation

-2.00

CFD-simulation variant

-2.50
-3.00

q=Q(res)/Q(pu)

Fig. 10 a) Loss coefficient K pu-res

loss-coefficient K tu-res (based on v pu)
2.00
1.50
1.00

K tu-res

0.50
0.00
-2

-1.5

-1

-0.5

0.5

-0.50
-1.00

experiment
-1.50

CFD-Simulation

-2.00

CFD-simulation variant

-2.50
-3.00

q=Q(res)/Q(pu)

Fig. 10 b) Loss coefficient K tu-res

loss-coefficient K pu-tu (based on v pu)

2.00
1.50
1.00

K pu-tu

0.50
0.00
-2

-1.5

-1

-0.5

0.5

-0.50
-1.00

experiment
-1.50

CFD-Simulation

-2.00

CFD-simulation variant

-2.50
-3.00

q=Q(res)/Q(pu)

Fig. 10 c) Loss coefficient K pu-tu

Fig. 11 a to d) Contours of velocity with details: case 2 left) and case 4 (right)

Fig. 12 Swirl flow in case 4, visualization with path lines

5. CONCLUSIONS
The hydraulic properties of the investigated T-junction proved to be very satisfactory. In
particular the loss coefficients were comparatively small. Due to the asymmetrical approach
flow and the shape of the T-junction considerable swirl was observed in the branch to the
turbine.
The CFD-simulation agreed with the experiments very well. Only in case 5, when flow
from the reservoir compared to flow from the pump was high, there were significant
differences in the loss coefficients.
A variant of the T-junction was also investigated in the CFD-simulation. In case of turbine
operation the head losses of the variant were slightly smaller, but in the other cases higher
than the original T-junction.
REFERENCES
Miller, D.S. (1978). Internal Flow Systems, BHRA Fluid Engineering, Vol. 5
Vorarlberger Illwerke AG, leaflet Kopswerk II, Vorarlberger Illwerke AG