1. Complementary spillway of Salamonde dam
Complementary spillway of Salamonde dam. Physical and 3D numerical modeling
Miguel R. Silva
MSc Student, IST
António N. Pinheiro
Full Professor, IST

2. Introduction
Throughout the design and planning for hydraulic structures as spillways, engineers and researchers are increasingly integrating computational fluid dynamics (CFD) into the process.
Velocity magnitude in a labyrinth weir (Lan, 2010)
Despite reports of success in the past, there is still no comprehensive assessment that assigns ability to CFD models to simulate a wide range of different spillways configurations. In this study, the applicability of CFD model (FLOW-3D) to simulate the flow into a geometry complex spillway was reviewed. 1

3. Case study
Salamonde dam, located in the north of Portugal in river Cávado, Peneda Gerês National Park, is a double-curved arch dam in concrete with maximum height of 75 m from foundations. The present spillway is composed by four surface orifices controlled by gates.
Salamonde Dam (from INAG and CNPGB)
After safety analysis from (EDP, 2006), it was concluded that it would be necessary an additional discharge structure - complementary spillway for Salamonde dam (under construction), which is the case study in this presentation.
Present spillway in Salamonde Dam (from origens.pt) 2

4. Inlet structure (from AQUALOGUS) Inlet structure (from AQUALOGUS)
Case study
The complementary spillway of Salamonde dam (under construction), is a gated spillway, controlled by an ogee crest, followed by a tunnel with rather complex geometry, designed for free surface flow, and a ski jump structure which directs the jet into the river bed.
The ogee crest is divided into two spans controlled by radial gates, 6.5 m wide.
Design characteristics:
Design discharge – m3/s 1
Inlet structure (from AQUALOGUS)
Inlet structure (from AQUALOGUS) 3

5. Case study Tunnel 4 Cross section AA’ Cross section BB’
Inlet sctructure
Cross section AA’
Cross section BB’
Cross section CC’ 4

6. Case study Outlet structure
Cross section BB’
The outlet structure is a ski jump, producing a free jet with impinging region in the center of the river bed.
Along the last 28 m of the outlet structure, the flow section is progressively reduced.
Cross section AA’ 5

7. Objectives
Calibrating the computational model, providing a sensitive analysis for some of the main parameters/models in FLOW-3D such as:
Momentum advection model ( 1st order vs 2nd order with monotonicity preserving)
VOF method (Defaut → One fluid, free surface vs Split Lagragian Method)
Turbulence mixing length (TLEN) parameter
Cell size
Validating the computational model using physical model discharge and flow depths measurements
Assessing the computational model results for design conditions, and compare them with physical model. The assessed results were:
Fluid depth
Velocity
Pressure at outlet structure
Free jet impinging region 6

8. Physical Model
The spillway was primarily tested and developed in a physical model built in Portuguese National Laboratory for Civil Engineering (LNEC), where discharge and flow depth were measured in ten cross-sections for five different gate openings. Additionally, the pressure at some points of the bottom and side walls of the outlet structure were measured.
Inlet structure from upstream view (from LNEC)
Inlet structure from downstream view (from LNEC) 7

9. Physical Model
Location of the ten cross-sections for flow depth measurements (from LNEC) 8

10. Numerical Model Geometry 5 components: Terrain Inlet structure Gates
Tunnel (hole component)
Outlet structure
1 flux surface baffle:
1 flux surface baffle downstream of tunnel (aligned with x axis) for discharge calculation
Initial conditions:
- Hydrostatic pressure in z direction
- Initial fluid elevation at reservoir and
river bed
hole component 9

11. Numerical Model Model setup Numerics: Volume-of-fluid advection:
Default VOF and
Split Lagrangian VOF
Momentum advection:
1st order and
2nd order monotonicity preserving
Physics:
Gravity g = m/s2 in z direction
Viscosity and turbulence: RNG model
No-slip
Main features of the computer used in simulations:
Processor: Intel® Core™ i QM 2.30GHz
8GB DDR Memory 10

12. Numerical Model Meshing Nested block 5 Mesh blocks: Cubic cells
1 Nested block at inlet structure and tunnel
4 auxiliary solids to “block” unnecessary active cells
Boundary Conditions:
Specified Pressure at reservoir mesh block (Y Max, X Max and X Min)
Specified Pressure upstream of the river (X Max)
Outflow downstream of the river (X Min and Y Min)
Symmetry between mesh blocks
auxiliary solids 11

13. Sensitive analysis - Discharge
The model configurations under analysis were:
Momentum advection model ( 1st order vs 2nd order with monotonicity preserving (MP))
VOF method (Default → One fluid, free surface vs Split Lagragian Method)
Turbulence mixing length (TLEN) parameter
Cell size
3 flow conditions were considered by setting the level in the reservoir:
Notes:
Every simulations were considered with a 1,0 m cell size* and for 200 seconds (exception in Cell size analysis).
Results of flow rate presented are calculated as a average value after steady state was reached
Laminar regime was considered until TLEN parameter analysis
VOF default method was considered until VOF method analysis
* Coarsest mesh to start 12

14. Sensitive analysis - Discharge
Momentum advection model (1st order vs 2nd order with MP)
2nd order with monotonicity preserving
VOF method (Default → One fluid, free surface vs Split Lagragian Method)
Default VOF method → One fluid, free surface 13

15. Sensitive analysis - Discharge
Turbulence mixing length (TLEN) parameter
For a 1,0 m cell size, spillway discharge didn’t appear to be very sensitive to TLEN parameter, although it influences the duration and stability of the simulation
Region not relevant to the spillway flow
Dynamically computed TLEN
TLEN ≈ 7% of hydraulic diameter ≈ 0.4 m 14

16. Sensitive analysis - Discharge
Cell size
Computational real time was drastically increased with 0.25 m cell size.
For flow rate results, the increase in computational time didn't justify the increased accuracy of a 0.25 m mesh.
*Restart Simulation of 30 seconds from 1.0m cell size simulation
For somehow accurate flow rate results, 0.50 m cell size is enough 15

17. Sensitive analysis - Fluid depth
The model configurations under analysis were:
Cell size
VOF method (Default → One fluid, free surface vs Split Lagragian Method)
Turbulence mixing length (TLEN) parameter
1 flow condition was considered by enabling the gates solid component for a 2.0 m opening
QPhysical Model = m3/s
Notes:
TLEN parameter considered was 0.4m
The presented results are related to Restart Simulations from steady state conditions 16

18. Sensitive analysis - Fluid depth
Cell size 1 2 10 4 5 6 3 8 7 9
Section 1
Section 3
Section 5
Section 2
Physical Model
1.00 m cell size
0.50 m cell size
0.25 m cell size
Section 4
Section 6 17

19. Sensitive analysis - Fluid depth
Cell size 1 2 10 4 5 6 3 8 7 9
Cell size = 1.00 m:
QFLOW-3D (m3/s) = → -22.2% error
Computational time ≈ 2 hours
Section 7
Section 9
Cell size = 0.50 m:
QFLOW-3D (m3/s) = → -10.4% error
Computational time ≈ 2 hours*
Well rendered gates are crucial for flow rate accuracy
Section 8
Section 10
Cell size = 0.25 m:
QFLOW-3D (m3/s) = → -2.8% error
Computational time ≈ 61 hours*
1.00 m cell size; m cell size 0.25 m cell size
Conclusion: 0.25 m cell size to represent with greater detail the free surface and flow rate accuracy for gate opening conditions
*Restart Simulation of 30 seconds 18

20. Sensitive analysis - Fluid depth
VOF method (Default → One fluid, free surface vs Split Lagragian Method) 1 2 10 4 5 6 3 8 7 9
Default VOF
Split Lagragian VOF
Physical Model
Section 1
Default VOF:
Q (m3/s) = 258.2
Relative error = -3.3 %
Computational time ≈ 61 hours
Split Lagragian VOF:
Q (m3/s) = 258.0
Relative error = -3.4 %
Computational time ≈ 107 hours
Section 4
Section 6
Conclusion: Default VOF method for lower computational time 19

21. Sensitive analysis - Fluid depth
Turbulence mixing length (TLEN) parameter 1 2 10 4 5 6 3 8 7 9
TLEN = 0.4 m
TLEN = 0.8 m
TLEN = 1.1 m
TLEN = dyn. computed
Physical Model
TLEN = dynamically computed:
Q (m3/s) = → -2.4% error
Computational time ≈ 137 hours
TLEN = 0.4 m:
Q (m3/s) = → -3.3% error
Computational time ≈ 61 hours
Section 1
TLEN = 0.8 m:
Q (m3/s) = → -2.8% error
Computational time ≈ 65 hours
Section 4
Section 6
TLEN = 1.1 m:
Q (m3/s) = → -3.5% error
Computational time ≈ 97 hours
Conclusion: TLEN adjusted in live simulation time (values lower than 0.4 m) for lower computational time 20

22. Results for Design Discharge
Design conditions:
Reservoir level = 270,64 m
Q = 1233,0 m3/s
FLOW-3D configurations:
2nd order monotonicity preserving momentum advection
Default VOF (One fluid, free surface)
RNG turbulence model
TLEN adjusted in real time (between 0,4 and 0,1 m)
Cell size 0,50 m and 0,25 m for Restart simulation
Simulations and real computational time:
1st simulation: cell size 0,50 m for 60 seconds -> Real time: 21 hours and 27 minutes
2nd simulation (Restart simulation): cell size 0,25 m for 10 seconds -> Real time: 17 hours and 5 minutes
QFLOW-3D (m3/s) = → -2.15% error
Calculated flow in 60 seconds simulation 21

23. Results for Design Discharge22

24. Results for Design Discharge
Velocity profiles: 1 2 10 4 5 6 3 8 7 9
Section 4:
VDim. (m/s) = 24.01
VFLOW-3D (m/s) = → % error
Section 4*
Section 9*
Section 6:
VDim. (m/s) = 24.62
VFLOW-3D (m/s) = → % error
*Velocity profiles rendered from post-processing software EnSight
Section 9:
VDim. (m/s) = 25.88
VFLOW-3D (m/s) = → % error
Section 6* 23

25. Results for Design Discharge
Pressure diagrams: 1 2 10 4 5 6 3 8 7 9
One of the greatest benefits of CFD models is the possibility to determine the main characteristics of the flow in any region of the computational domain.
Pressure diagrams can be useful for the structural design of hydraulic structures.
Section 5*
*Pressure diagrams rendered from post-processing software EnSight
Section 6*
Section 10* 24

26. Location of the pressure taps in the physical model (from LNEC)
Results for Design Discharge
Pressure diagrams:
The outlet structure has, naturally, higher values ​​of pressure studied in Physical model.
The pressure value was compared for 4 different points: 2 for the bottom wall and 2 for the right side wall
P2F
P1F
The outlet structure has, naturally, higher values ​​of pressure and was, for that reason, studied in physical model. P3 P1
P2F
Pressure diagrams in the outlet structure (for each 5 m distance profile) – Rendered from EnSight
P1F
Location of the pressure taps in the physical model (from LNEC) 25

27. Results for Design Discharge
Pressure diagrams:
Point 2F (higher pressure value in physical model):
PPhysical Modell (Pa) =
PFLOW-3D (Pa) = → 0.4 % error
Point 3:
PPhysical Modell (Pa) =
PFLOW-3D (Pa) = → -8.8 % error P3 P1
P2F
P1F
Point 1F:
PPhysical Modell (Pa) = 97184
PFLOW-3D (Pa) = → -5.8 % error
Pressure diagram at m from downstream outlet structure limit
Pressure diagram at 2.00 m from downstream outlet structure limit
Conclusion: FLOW-3D appears to give accurate results of pressure
Point 1:
PPhysical Modell (Pa) = 95419
PFLOW-3D (Pa) = 73220→ % error 26

28. Results for Design Discharge
Free jet:
One of the main objectives of building the physical model of the Salamonde spillway was analyzing the free jet impinging region at the river bed.
View of free jet from left hill (physical and computational models)
View of free jet from right hill (physical and computational models)
Conclusion: FLOW-3D represents accurately the free jet shape
View of free jet from top (physical and computational models) 27

29. *Rendered with EnSight *Rendered with EnSight
Results for Design Discharge
Free jet:
In the physical model, the Min. and Max. of free jet impinging was registered :
LPhysical Model Min = m
LPhysical Model Max = m
*Rendered with EnSight
*Rendered with EnSight
FLOW-3D results of free jet incidence:
LFLOW-3D Min = m → -1.0 % error
LFLOW-3D Max = m → % error
Conclusion: FLOW-3D appears to be somehow accurate for free jet impinging region 28

30. Results for Design Discharge
Water level fluctuations against the slopes of the valley:
One of the major concern of engineers during the design of spillways with free jet dissipation is the increase of the level in the river hills.
FLOW-3D could help to decide the jet impinging region in the river bed in order to minimize erosion and water level fluctuations against the slopes of the valley.
For Salamonde, a 12 m elevation above the maximum river level was expected . 29

31. Conclusions
A sensitive analysis for some main FLOW-3D configurations is quite important before starting to predict flow characteristics:
2nd order momentum advection model with monotonicity preserving was a good choice for better accuracy without much more computational time
Split Lagragian VOF didn’t appear to be useful in this kind of extreme velocity flow conditions
TLEN parameter has a huge importance in simulation stability and duration
Finer cell size has higher importance if there are some geometric details such as radial gates
In general, FLOW-3D accurately represents the flow along the spillway with rather complex geometry
FLOW-3D could be very useful for structural design of hydraulic structures. Pressure diagrams and velocity profiles could be rendered with a post-processing software as EnSight and could be helpful in that design phase. 30

32. Acknowledgments
Flow Science, Inc. support team : Melissa Carter and Jeff Burnham
Simulaciones y Proyectos, SL : Francisco Garachana and Raúl Martín
Ensight support team, Distene support and ANALISIS-DSC support : Daniel Cuadra
EDP
AQUALOGUS
LNEC: Lúcia Couto, Teresa Viseu and António Muralha
Instituto Superior Técnico: Prof. António Pinheiro 31

33. Any Questions? Miguel R. Silva MSc Student, IST
António N. Pinheiro
Full Professor, IST 32