UE Advanced Numerical Simulations (5EUS5SNA)

Optimization of the
design of a draft tube
for Bulb turbine for
hydroelectric power
plant
Computational Fluid Dynamic (CFD) for the design of energy systems

Guillaume Lafitte
02/02/2021
Guillaume Lafitte 02/02/2021

1. Introduction
In this project, we will study the bulb turbines; which are the turbine used for small head and large flow rate:

For the theoretical questions and demonstrations, we are going to use the article “Analysis of Head Losses in
a Turbine Draft Tube by Means of 3D Unsteady Simulations” written by Sylvia Wilhelm, Guillaume Balarac,
Olivier Métais and Claire Segoufin, in Springer Science+BusinessMedia Dordrecht 2016.

2. Hydraulic efficiency
Hydraulic turbine efficiency can be mainly measured by considering Bernoulli’s principle. The idea is to have
an optimized 𝛼 angle.

3. Project
The design of the tube is, thanks to the symmetrical aspect, half of the normal draft tube. The sections 𝑆2 for
the end of the shape and 𝑆4 for inlet will be considered for the half:

𝑆4 = 1 𝑚2
𝑆2 = 5 × tan(𝛼) + 𝑆4

Page 1 sur 9
Guillaume Lafitte 02/02/2021

3.1. Perform simulations for a Reynolds number based on the inlet bulk velocity and inlet
(normal) length equal to 100.
𝑈×𝑅
We know that 𝑅𝑒 = and 𝑅𝑒 = 100; 𝑈 = 1 𝑚/𝑠; 𝑅 = 1𝑚
𝜂

𝑈×𝑅 1×1
Then: 𝜂 = 𝑅𝑒
= 100
= 0.01

We then create an artificial fluid for the CFD simulation. To compensate the value of 𝑅𝑒 we could also change
the velocity for a much lower value instead of changing the

3.1.1. Discuss the verification of our numerical set-up

We perform the simulation for a fixed angle and for different residuals, the results are in the table below. It
can be seen that for an absolute criteria of 102 the values are very differents from the other residuals and the
number of iterations is too low (approximately 50), we cannot see the convergence on the chart. For a residual
value of 106 the convergence is good.

We can also see that for different angles and the same absolute criteria, the number of iterations is different.
The greater the angle is, the lower the absolute criteria has to be in order to have a good convergence.

To solve those different problem we will uncheck the Convergence Criteria to let the Residual converge on is
own as the image below show:

Page 2 sur 9
Guillaume Lafitte 02/02/2021

Another issue has been found: as the chart below show, when we take a greater angle, the value of Λ is still
increasing. To solve that we change the size of the elements in the mesh, instead of 0.2 we take 0.1. We also
use the inflation around the top of the wall to have more accurate results.

Value of Λ as a fonction of α
1

0,8

0,6

0,4

0,2

0
0 5 10 15 20 25

3.1.2. Find the optimal angle to design the optimal draft tube in this case
In order to find the optimal angle, we use the equation of the head losses coefficient 𝐾42 :

𝑝4 − 𝑝2 𝑣42 − 𝑣22 𝑣42

+ = 𝐾42
𝜌 2 2
Which leads to find the maximum and positive value of the efficiency Λ:

𝑆42
Λ=1− − 𝐾42
𝑆22

We need to compute the integral of the axial velocity in 2 and 4, the integral of the static pressure in 2 and 4.

In order to know the value of 𝑆2 and 𝑆4 we look into the mesh geometry file:

Page 3 sur 9
Guillaume Lafitte 02/02/2021

𝑆4 = 1 𝑚2
𝑆2 = 5 × tan(𝛼) + 𝑆4
We have performed the simulation for different value of 𝛼 in the table below:
α S2 p2 p4 v2 v4 K42 Λ
5 6,491294745 0,23830282 0,00869946 0,48342873 0,99946435 0,306347031 0,459425899
8 9,108110142 0,1045977 -0,30777775 0,34391682 0,99426667 0,046063209 0,83496522
9 10,08760744 0,078712444 -0,35725257 0,31098294 0,99665965 0,024855818 0,878154977
10 11,12296559 0,057788483 -0,38088878 0,28176325 0,99526782 0,034135283 0,886091234
11 12,21575538 0,03916593 -0,3910966 0,25732098 0,99367503 0,06142549 0,872435287
12 13,36770581 0,024370237 -0,38031983 0,23562113 0,99028055 0,118041564 0,826727044
15 17,19836991 -0,000228738 -0,30143346 0,18295049 0,98328956 0,342323186 0,624309198

We see that for an angle of 𝛼 = 10° the value of Λ is maximum and positive. Approximately Λ = 0.886.
Meaning the draft tube design improve the hydraulic efficiency at its best.

Valeur de Λ en fonction de α
1

0,9

0,8

0,7 Re=100 (mesh 0,1)

Re=100 (mesh 0,05)
0,6

0,5

0,4
4 6 8 10 12 14 16

We also can see that if we change the size of the mesh, from 0.1 to 0.05, the values are not very much different
here.

3.1.3. Link the head loss coefficient (and then the Bernoulli’s principle) with the kinetic energy
transport equation integrated in all the draft tube (i.e., in all the volume). Discuss then
the variation of the head loss coefficient with the angle value from the local flow behavior.
The head loss is a global quantity, it doesn’t give locally which are the regions of the flow have a consequence
on the head losses. By expressing the head loss as the volume integration of a local quantity, the analysis of
this local quantity will identify such regions.

We start from the Navier-Stokes equation:

𝜕𝑢𝑖 𝜕 1 𝜕𝑝 𝜕 2 𝑢𝑖
+ (𝑢𝑖 𝑢𝑗 ) = − +𝜈 + 𝑔𝑖
𝜕𝑡 𝜕𝑥𝑗 𝜌 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑗

We then multiply by 𝑢𝑖 on both sides and allows to show the transport equation for 𝐾:

Page 4 sur 9
Guillaume Lafitte 02/02/2021

𝜕𝐾 𝜕𝐾 𝜕 𝑝 𝜕
+ 𝑢𝑗 = −𝑢𝑖 ( ) + 2𝜈 (𝑢 𝛿 ) − 2𝜈𝛿𝑖𝑗 𝛿𝑖𝑗 + 𝑔𝑖 𝑢𝑖
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜌 𝜕𝑥𝑗 𝑖 𝑖𝑗

1 𝜕𝑢𝑖 𝜕𝑢𝑗
𝛿𝑖𝑗 = ( + )
2 𝜕𝑥𝑗 𝜕𝑥𝑖
1
𝐾 = 𝑢𝑖2
2
For a steady laminar flow, we have:
𝜕 𝑝 𝜕
𝑢𝑗 ( + 𝑔𝑧 + 𝐾) = 2𝜈 (𝑢 𝛿 ) − 2𝜈𝛿𝑖𝑗 𝛿𝑖𝑗
𝜕𝑥𝑗 𝜌 𝜕𝑥𝑗 𝑖 𝑖𝑗

Due to the incompressibility, we can write:

𝜕 𝑝 𝜕
(𝑢𝑗 ( + 𝑔𝑧 + 𝐾)) = (2𝜈𝑢𝑖 𝛿𝑖𝑗 ) − 2𝜈𝛿𝑖𝑗 𝛿𝑖𝑗
𝜕𝑥𝑗 𝜌 𝜕𝑥𝑗

We then integrate it in the whole domain (total volume 𝑉 of the draft tube):

𝜕 𝑝 𝜕
∭ (𝑢 ( + 𝑔𝑧 + 𝐾)) 𝑑𝑉 = ∭ (2𝜈𝑢𝑖 𝛿𝑖𝑗 )𝑑𝑉 − ∭ 2𝜈𝛿𝑖𝑗 𝛿𝑖𝑗 𝑑𝑉
𝜕𝑥𝑗 𝑗 𝜌 𝜕𝑥𝑗
𝑉 𝑉 𝑉

Then with 𝑛𝑗 the component of the vector 𝑛⃗ normal to the surface, we can write (𝑆 is the external surface of
the draft tube):

𝑝
∯ 𝑢𝑗 ( + 𝑔𝑧 + 𝐾) 𝑛𝑗 𝑑𝑆 = ∯ 2𝜈𝑢𝑖 𝛿𝑖𝑗 𝑛𝑗 𝑑𝑆 − ∭ 2𝜈𝛿𝑖𝑗 𝛿𝑖𝑗 𝑑𝑉
𝜌
𝑆 𝑆 𝑉

We have on the left side the head flux through the draft tube external surface, this flux is zero through the
lateral surfaces (impermeable). We then make the assumption that for a steady flow with constant flow rate
𝑄 and uniform head 𝐻 on the inlet and outlet (surfaces 𝑆4 and 𝑆2 ).

𝑔𝑄(𝐻4 − 𝐻2 ) = 𝑔𝑄Δ𝐻42 = ∯ 2𝜈𝑢𝑖 𝛿𝑖𝑗 𝑛𝑗 𝑑𝑆 − ∭ 2𝜈𝛿𝑖𝑗 𝛿𝑖𝑗 𝑑𝑉

𝑆 𝑉

Now we know that the head losses in the draft tube are influenced by two terms in the right side: they
correspond to the diffusion and viscous dissipation of kinetic energy. Meaning the speed has an influence on
the head losses.

When wen now clearly see that if the angle 𝛼 increases, the section 𝑆2 also increases. Then the flow rate is
constant, so the speed decreases.

From the data we could have collected from our simulation, we have:

Page 5 sur 9
Guillaume Lafitte 02/02/2021

0,55 0,4

0,5 0,35

0,45 0,3

Head losses
0,4 0,25
Speed

0,35 0,2

0,3 0,15

0,25 v2 0,1
K42
0,2 0,05

0,15 0
4 6 8 10 12 14 16
Angle

We see that the head loses are minimized for a certain speed and for a certain angle as well.

We can also compute the strain rate sensor, which is a local quantity:

1 𝜕𝑢𝑖 𝜕𝑢𝑗
𝛿𝑖𝑗 = ( + )
2 𝜕𝑥𝑗 𝜕𝑥𝑖

2,5 0,4

0,35
2
0,3
Strain rate sensor

0,25
1,5
Head losses

0,2 Strain rate (Surf)

Strain rate (Vol)
1
0,15
K42

0,1
0,5
0,05

0 0
0 2 4 6 8 10 12 14 16
Angle

We see that this local quantity decreases (surface and volume). There should be a minimum, it will mean there
is a link with the head loses coefficient. Further and deeper simulations are needed to see a link.

3.2. Turbulent case: Perform simulations for a Reynolds number based on the inlet bulk
velocity and inlet (normal) length equal to 100 000.
𝑈×𝑅
We know that 𝑅𝑒 = 𝜂
and 𝑅𝑒 = 100000 ; 𝑈 = 1 𝑚/𝑠 ; 𝑅 = 1𝑚

Page 6 sur 9
Guillaume Lafitte 02/02/2021

𝑈×𝑅 1×1
Then: 𝜂 = 𝑅𝑒
= 100000 = 1.10−5 𝑚2 . 𝑠 −1

3.2.1. Discuss the verification of our numerical set-up (models, mesh and convergence)
If we only change the viscosity 𝜂 we have a problem of a floating point on the CFD simulation. Instead, we
change both the velocity 𝑈 = 10 𝑚/𝑠 and the viscosity 𝜂 = 0.0001 𝑚2 . 𝑠 −1 .

We also use a turbulent viscous model: 𝑘 − 𝜖 model (2 equations):

3.2.2. Find the optimal angle to design the optimal draft tube in this case
We perform the exact same kind of simulation as before; the results are in the table below:

α S2 p2 p4 v2 v4 K42 Λ
5 6,491294745 -1,137096 -37,725313 4,48434432 9,986367 0,064593479 0,70117945
6 7,311161106 -1,7049301 -40,284898 4,2986487 9,982499 0,040260495 0,775098975
7 8,183053471 -2,2633785 -42,017624 3,8422417 9,9809367 0,053682329 0,798927349
8 9,108110142 -2,8033223 -43,114426 3,4492755 9,9777395 0,070670046 0,810358382
9 10,08760744 -3,5999382 -43,571255 3,1135513 9,975527 0,099228124 0,803782671
10 11,12296559 -4,2964814 -43,603043 2,8236543 9,9754423 0,129870102 0,790356415
11 12,21575538 -5,0755739 -43,125289 2,5711532 9,9700245 0,167916463 0,765944314
12 13,36770581 -5,5207963 -42,494097 2,3497004 9,9671599 0,200077788 0,74469082
15 17,19836991 -6,2160737 -39,809124 1,8267058 9,9632488 0,289558131 0,677074254

Page 7 sur 9
Guillaume Lafitte 02/02/2021

Evolution of Λ as a function of α
0,83

0,81

0,79

0,77

0,75
Re=1e5 (mesh 0,1)
0,73
Re=1e5 (mesh 0,05)
0,71

0,69

0,67

0,65
4 6 8 10 12 14 16

We have seen that for an optimized angle of 𝛼 we have a maximum hydraulic efficiency. In the non-turbulent
case, the angle was different. The head loss coefficient is the lowest possible: 𝐾42 = 0.07 the turbulence
model allows us to have a more realistic version.

3.2.3. Link transport equation integrated in all the draft tube (i.e., in all the volume). Discuss
then the variation of the head loss coefficient with the angle value from the local flow
behavior. Underline the major role played by the turbulence model in the head loss
coefficient prediction.
As the same way we did when it was a non-turbulent case, with start from the Navier Stokes equation and
then we have the transport equation for 𝐾 as:
𝜕𝐾 𝜕𝐾 𝜕 𝑝 𝜕
+ 𝑢𝑗 = −𝑢𝑖 ( ) + 2𝜈 (𝑢 𝛿 ) − 2𝜈𝛿𝑖𝑗 𝛿𝑖𝑗 + 𝑔𝑖 𝑢𝑖
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜌 𝜕𝑥𝑗 𝑖 𝑖𝑗

1 𝜕𝑢𝑖 𝜕𝑢𝑗
𝛿𝑖𝑗 = ( + )
2 𝜕𝑥𝑗 𝜕𝑥𝑖
1
𝐾 = 𝑢𝑖2
2
For a steady turbulent flow, we have:
𝜕 𝑝 𝜕 𝜕
𝑢𝑗 ( + 𝑔𝑧 + 𝐾) = 2𝜈 (𝑢 𝛿 ) − (𝑢 𝑢′ 𝑢 ′) − 2𝜈𝛿𝑖𝑗 𝛿𝑖𝑗 + 𝑢𝑖′ 𝑢𝑗′ 𝛿𝑖𝑗
𝜕𝑥𝑗 𝜌 𝜕𝑥𝑗 𝑖 𝑖𝑗 𝜕𝑥𝑗 𝑖 𝑖 𝑗

From the laminar case, we have two new terms: one in the diffusion of kinetic energy and the other is
responsible of turbulent energy production.

We integrate it over the total volume 𝑉 of the draft tube, with 𝑆 the external surface and 𝑛𝑗 the component
of the outer-pointing normal vector:

Page 8 sur 9
Guillaume Lafitte 02/02/2021

𝑝
∯ 𝑢𝑗 ( + 𝑔𝑧 + 𝐾) 𝑛𝑗 𝑑𝑆 = ∯ 2𝜈𝑢𝑖 𝛿𝑖𝑗 𝑛𝑗 𝑑𝑆 − ∯ 𝑢𝑖 𝑢𝑖′ 𝑢𝑗 ′𝑑𝑆 − ∭ 2𝜈𝛿𝑖𝑗 𝛿𝑖𝑗 𝑑𝑉 + ∭ 𝑢𝑖′ 𝑢𝑗′ 𝛿𝑖𝑗 𝑑𝑉
𝜌
𝑆 𝑆 𝑆 𝑉 𝑉

The flux on the left side of the equation is obviously zero through the lateral solid surfaces which are
impermeable:

𝑔𝑄Δ𝐻42 = ∯ 2𝜈𝑢𝑖 𝛿𝑖𝑗 𝑛𝑗 𝑑𝑆 − ∯ 𝑢𝑖 𝑢𝑖′ 𝑢𝑗 ′𝑑𝑆 − ∭ 2𝜈𝛿𝑖𝑗 𝛿𝑖𝑗 𝑑𝑉 + ∭ 𝑢𝑖′ 𝑢𝑗′ 𝛿𝑖𝑗 𝑑𝑉
𝑆 𝑆 𝑉 𝑉

On the right-hand side, the first two term correspond to the diffusion of mean kinetic energy. The two last
term contribute to head losses by viscous dissipation of the mean kinetic energy either directly by the mean
flow or indirectly after turbulent kinetic energy production.

Just like in the laminar case, we compute the strain rate sensor, we also see a decreasing, but it is not enough
to say that there is a link with head loses. Further simulations are required to prove this demonstration.

4. Conclusion
We have seen that for a greater Reynolds number, in the turbulent case, the optimal angle was different than
in the laminar case. It shows the importance to use a realistic case, to take into account the local behavior.

More simulations are required to complete this project, to demonstrate the major role of the turbulence
model in the head loses predictions. We should see a variation of the head loses coefficient when the angle
changes.

Page 9 sur 9