9th International Conference on Urban Drainage Modelling

Belgrade 2012

Surface roughness effect on near bed Turbulent

Kinetic Energy in a large stormwater detention basin
Hexiang Yan, Gislain Lipeme Kouyi*, Jean-Luc Bertrand-Krajewski
University of Lyon, INSA Lyon, LGCIE (Laboratory of Civil and Environmental Engineering)
F-69621, Villeurbanne, France
*Corresponding author, e-mail: gislain.lipeme-kouyi@insa-lyon.fr

ABSTRACT
The effect of surface roughness on near bed turbulence is of great importance for
sediments deposition and entrainment in stormwater detention basin. In this paper,
effects of surface roughness in sedimentation processes have been investigated by
means of RANS approach (CFD technique based on Reynolds Averaged Navier
Stokes equations). Previous work has showed the ability of CFD modelling to
allow the identification of the preferential deposition zones in a large detention
basin. The new boundary condition on the bottom was based on the interaction
between near bed turbulence and particle settling characteristics such as V80 settling
velocity. In order to further clearly understand the effects of surface roughness on
sedimentation, simulations with different surface roughness have been carried out
and measurements of sediment thickness and distribution have been performed in
the Django Reinhardt large stormwater detention basin in Chassieu (close to Lyon,
France). Analysis of simulated results compared to measurement data reveals that
the contour of preferential deposition zones linked to near-bed turbulent kinetic
energy distribution is sensitive to surface roughness. The maximum value of the
near-bed turbulent kinetic energy distribution in deposition zones is sensitive to
surface roughness and is lower than V802 .

KEYWORDS
Detention basin, surface roughness, turbulent kinetic energy, sediment distribution

INTRODUCTION

Turbulent boundary layer over roughness elements is of great interest in hydraulic engineering. Effects
of surface roughness on sediments distribution are of great importance as reminded by Papanicolaou et
al. (2001). Turbulent boundary layer over a rough surface contains a roughness sublayer within which
the flow is directly influenced by the individual roughness elements and is not spatially homogenous.
The height of this sub-layer presumably depends on the height of the roughness elements.
Experimental and numerical studies have been performed to investigate the effects of surface
1

roughness on wall bounded turbulence to get a clearer picture of the impact of roughness in turbulent
boundary layer. Perry et al. (1987) performed experiments using both three dimensional diamondshaped mesh roughness and streamwise wave length. They observed that smooth and rough wall
boundary layer have quite different structures. Papanicolaou et al. (2001) provided experimental
evidence that various packing density configuration encountered in natural gravel bed streams affect
the turbulence characteristics of the flow. Shafi and Antonia (1997) measured the root mean square
(rms) vorticity fluctuations normalized by the friction velocity and boundary layer thickness. They
found that the effects of roughness on the vorticity is less pronounced than on the Reynolds stress,
which conflicts with the traditional picture of wall similarity.
Recently computational fluid dynamics (CFD) is of increasing interest for many engineering fields. In
urban drainage systems, CFD technique could be used to evaluate actual water treatment facilities and
to improve the design procedure (Wood et al., 1998; Krishnappan and Marsalek, 2002; Hribersek et
al., 2011). In traditional hydraulics the influence of roughness has been cataloged in the form of a
roughness coefficient based on data obtained from a wide range of field and laboratory observations
(Souders and Hirt, 2002). Typically roughness coefficients are defined in one of the three ways: (1)
Chzy resistance coefficient C, (2) Mannings coefficient n, or (3) equivalent sand grain roughness
(ks). When a turbulence transport model (e.g., k - model) is used in CFD simulations, there must be a
suitable boundary condition for the turbulence quantities (e.g., turbulent kinetic energy k and turbulent
dissipation ) only at the inlet and outlet boundary of the computational domain. Regarding the
influence of the wall roughness, the so-called wall functions are widely used for that purpose. It is
assumed that a logarithmic velocity profile (called log-law of wall) exists near a wall, and a modified
law of wall is often used to account for the roughness effect on the wall.
A detailed modelling of fluid flow and pollutants transport in a full scale stormwater detention tank is
a difficult task due to the complex geometry and variable time-dependent processes (deposition, nearbed transport, resuspension, etc) as well as limitation in computational resources in the case of a large
structure (due to a lot of computational mesh elements). Preliminary work on modelling of
hydrodynamic and solid transport in a large stormwater detention basin was undertaken in order to
reproduce the preferential sediment zone and efficiency (Yan et al., 2011). A new boundary condition
was proposed to determine particle state (deposited or suspended) near the bottom using Lagrangian
particle tracking approach to simulate solid transport in fluid flow. This boundary condition was based
on the interaction between the near bed turbulent kinetic energy and near bed particles energy thanks
to settling velocity. In fact, Bagnold (1966) proposed one of the first criteria to estimate the threshold
condition for particle suspension. This was based on the assumption that particles sustain in
suspension by the fraction of turbulent energy. From this point of view, the near bed turbulent kinetic
energy plays an important role in the settling processes. Unfortunately, few studies have been carried
out in order to further highlight effects of roughness on sedimentation processes which are strongly
impacted by near bed turbulence characteristics. Therefore the present paper aims to study the
sensitivity of the near bed turbulence quantities (particularly Turbulent Kinetic Energy TKE) to the
bed surface roughness in the Django Reinhardt large stormwater basin in Chassieu (Rhne, France),
which is a part of OTHU programme site (Field Observatory in Urban Hydrology www.othu.org).

2
2.1

METHODOLOGY
Experimental field site

The Django Reinhardt detention-infiltration facility was built in 1975 in order to store stormwater
from a 185 ha industrial catchments during wet weather time. The facility consists of two sub-basins
2

connected by a 60 cm diameter pipe. The first one is a detention and settling basin (Figure. 1a) where
the stormwater is stored before being discharged downstream to the infiltration sub-basin. The bottom
of the detention basin is sealed with bitumen and is equipped with a low flow trapezoidal channel
collecting and guiding the dry weather flow towards outlets. The sides of the detention basin are
covered with the plastic liner. Stormwater enters the detention basin via two 1.6 m diameter circular
concrete pipes (labelled as inlet 1 and inlet 2 in Figure.1a). In order to improve the settling process, a 1
m high detention wall was built in 2004. There are three 19 cm
diameter outletSAMPLING
orifices (labelled o1, o2
SEDIMENT
and o3 in Figure. 1a) and an overflow weir in the detention wall. The stormwater outflow towards the
infiltration sub-basin is limited to 350 L/s by a regulator (Hydroslide gate).

outlet

(a)

inlet

(b)

Figure 1. (a) Sketch of the Django Reinhardt basin (Bardin and Barraud, 2004) and (b) Sediment trap
locations in the basin (Torres, 2008)
JLBK, INSA de Lyon, 24 Nov. 2006

The inlet and outlet discharges are calculated from simultaneous measurements of water depth and
velocity in the pipes. Three water depth sensors are located on the bottom of the basin (labelled h 1, h2
and h3 in Figure 1a). All variables are recorded with a 2 minute time step to a S50 Sofrel data logger.
Twelve sediment traps installed on the bottom of the basin collect settled sediments during storm
events (Figure. 1b). The sediment traps are numbered according to their altitude, from 1 (lowest) to 12
(highest). After a storm event, samples made up of a mixture of water and sediments are transported as
quickly as possible to the laboratory, where the particles settling velocity and particle size distributions
are measured respectively according to the VICAS protocol and Laser Particle Sizer (LPS) technique
(Torres, 2008).
2.2

Sediments measurement

Lot of sediments has been accumulated in the detention basin since the rehabilitation in 2004. In order
to analyze the spatial distribution of the cumulate sediment in the basin, the profile of preferential
sediments zone and discrete point sediment thickness were measured. The layout of spatial interval of
measurement point is 5 m for the preferential sediment zone (left part of basin, Figure 2). Refined
spacing of measurement points has been used in the centre of the basin and measurement points have
been distributeded more sparsely close to the inlet of the basin (shown in Figure 2).

Figure 2. Layout of the sediments thickness measurement points

52

2.3

Numerical model

The purpose of this part is to highlight the used numerical models integrated into the CFD software
package Ansys Fluent (version 14.0). The key models employed are based on the Reynolds Averaged
Navier-Stokes (RANS) equations. The k- renormalization group (RNG) model was employed to
involve the turbulent phenomena. It is better than standard k- model since more features are included
in this model. While the standard k- model is a high-Reynolds number model, the k- RNG model
provides an analytically derived differential formula for effective viscosity that accounts for lowReynolds number effects as well. These features make the k- RNG model more accurate and reliable
for a wider class of flows than the standard k- model. Compared to the standard k- model and the k RNG model, RSM (Reynolds Stress Model) is more time consuming as it needs to solve more
equations. Dufresne et al. (2009) and Mignot et al. (2011) tested the standard k- model, the k- RNG
model and the RSM in various urban drainage structures (three combined sewer overflow chambers
for solid separation and open channel junctions). They found that the RSM model did not give
significant improvement compared to the results deriving from the k- RNG model.
The surface roughness effects on turbulence quantities and bed shear stress is considered through the
modified law-of-the-wall for roughness. The standard wall functions in Ansys Fluent are based on the
Launder and Spalding (1974) work and have been most widely used in industrial flows (ANSYS, 2011;
Souders and Hirt , 2002). Experiments in rough pipes and channels indicate that the mean velocity
distribution is influenced by the near rough walls and in the usual semi-logarithmic scale, has the same
slope (1/) but a different intercept (Tachie et al., 2004; Akinlade et al., 2004). Thus, the law-of-thewall for mean velocity impacted by rough wall has the form (ANSYS, 2011):

u pu*
where

ln( E

u* y p
) B

(1)

u* C1 4 k 1 2

= Von Krmn constant (= 0.4187)

E= empirical constant (= 9.793)
up= mean velocity of the fluid at the near-wall node P
k= turbulent kinetic energy at the near-wall node P
yp= distance from point to the wall
= dynamic viscosity of the fluid
B is a roughness function that quantifies the shift of the intercept due to roughness
effects. B depends on the type (uniform sand, rivets, threads, ribs, mesh-wire, etc.) and size of the
roughness. There is no universal roughness function valid for all types of roughness. However, for a
sand-grain roughness and similar types of uniform roughness elements B has been found to be wellcorrelated with the non-dimensional roughness height, K s K su* / , where Ks is the physical
roughness height.
The turbulent flow regime is subdivided into three regimes, and the formulas proposed by Cebeci and
Bradshaw (1977) based on Nikuradse's data are adopted to compute the roughness function, B , for
each regime (ANSYS, 2011).
For the smooth regime (Ks+ < 2.25): B 0
For the transitional regime (2.25 < Ks+ < 90):

(2)

K 2.25

ln s
Cs K s sin 0.4258 ln K s 0.811
87.75

where Cs is a roughness constant, and depends on the type of the roughness.

(3)

In the fully rough regime (Ks+ > 90):

2.4

ln 1 Cs K s

(4)

CFD model setup

The flow regime in a detention basin can be described as transient due to the complex geometry of
basin, rough surface conditions (concrete and vegetated sediments at the bottom), variable inflow and
outflow rate, etc. Preliminary tests have been carried out and they proved that a steady state simulation
was able to predict the preferential sediment zone and evaluate the basin efficiency (Yan et al., 2011).
Figure 3 shows a curve of water depth h1 against inflow rate in the basin for the reference storm event
on 31/5/2007 (Torres, 2008). The measured water depth was approximately kept in a same level for a
period of time during the rainfall event (Figure 3). Hence the corresponding water level was chosen as
free surface level. Meanwhile, the inflow rate of 350L/s was chosen since the outflow rate is limited
by a regulator which enables fixing the outflow rate to 350 L/s. A quasi-steady state condition can be
accepted under this situation. As shown in Figure 1(a),it was observed that there was no sediments in
the second compartment close to the outlet. Thus, in order to reduce the number of computational
meshes and then to save computation time, this second part of the basin is cut in order to simplify
geometry and boundary conditions as shown in Figure 4.

Figure 4. Sketch of the simplified geometry of the

Django Reinhardt basin

Figure 3. water depths h1 versus inflow rates

The mesh-independent tests have been carried out. Three different mesh resolutions have been
established and tested: one coarse mesh size with 650 000 cells, a medium mesh size of 850 000 cells
and a refined mesh size of 1000 000 cells. Fairly similar results were obtained with the medium and
refined resolution mesh. Thus the medium resolution mesh with 850 000 cells was used for all
simulations.
The following boundary Conditions have been used. Uniform velocity distribution was set at the inlet
cross-section, corresponding to the inflow rate of 0.35m3/s (corresponding to the outflow rate limited
by a regulator valve). Three orifices and overflow weir were set as pressure-outlet. A specific pressure
was set at orifice 2 as it was completely submerged under free water surface. For the free water
surface, a symmetry condition was set instead of Volume-of-Fluid (VOF) model for air-water
interface capturing in order to reduce the computational time. The symmetry condition means that zero
normal velocity and zero normal gradients of all variables at the symmetry plane. It can be used to
model zero-shear slip walls in viscous flows (Dufresne, 2008; Stovin et al., 2008). The surrounding
wall sides were considered as smooth since they were covered with a plastic liner and were set as no
slip condition with zero roughness height. Lastly, the bottom of basin was considered as rough surface
and was set as no slip wall condition with non-zero roughness height. Initially the bottom was covered
by concrete and later a part of it was covered by accumulated sediments with or without vegetation.
5

However, it was difficult to determine equivalent roughness height for vegetated sediment surface.
Therefore, the bottom was considered to be covered by the concrete and the equivalent sand grain
roughness height was estimated by means of equation (5) proposed by Hager (2010). Table 1 shows a
series of Strickler coefficients K and equivalent sand grain roughness heights ks for concrete (Graf and
Altinakar, 2000).
1

K k s 6 6.5 g

(5)

where g is the gravity acceleration.

Table 1. Strickler roughness coefficient K and equivalent sand grain roughness height ks.
Cases
ks1
ks2
ks3
ks4
K(m1/3/s)
75
65
55
50
ks (m)
0.00040
0.00094
0.0026
0.0045

3
3.1

RESULTS AND DISCUSSION

Spatial sediments distribution

A lot of sediment settled during the storm event and accumulated in the basin since the rehabilitation
of basin in 2004. Figure 5 shows evidently the top view of sediments in the bottom of detention basin
in 2007. Figure 6 shows colour contours of sediment thickness according to the measurement data in
2011, which was processed using Matlab software. The blank part means almost no sediment in the
basin or very thin. As shown in Figure 6, the important amount of sediments was located in the centre
zone of the basin. Initially, the bottom of the basin was covered by bare concrete. Later, part of the
bottom of basin was covered by the settled sediment. Vegetation was also observed on the thick
sediment layer. It is expected that the effect of surface roughness existed due to a rough bare concrete
and cumulate sediment layer with or without vegetation.

Concrete area

vegetation

Figure 5. Photo of detention basin (2007) from

Google Earth

Figure 6. Contour of observed sediment

thickness(m) measurement in 2011

3.2

Effect of surface roughness on near bed turbulence

Four simulation cases have been carried out with the same configuration except for varying the
parameter of bottom surface roughness height. Different concrete resistance coefficients were tested.

ks1

ks2

ks3

ks4

Top view of sediment in 2007

Figure 7. Similarity between observed deposition zones and simulated Bed TKE distribution according
to ksi (i=1, 2, 3, 4) values related to different surface roughness. The simulated deposition zones
correspond to bottom area where TKE is lower than kc=Vs2 .Vs is the settling velocity and
represents a coefficient which enables to account for uncertainties on settling velocity assessment and
the cohesion of particles, etc.
Dufresne (2008) reveals that the distribution of bed turbulent kinetic energy (BTKE) under a critical
value (0.00010-0.00030m2/s2) is very similar to the deposition zone in a pilot tank. Previous work in
the same investigation (Yan et al., 2011) suggested that particle settling velocity (Vs measured with
VICAS protocol, Torres, 2008) could be used to estimate the critical value to determine the particle
state (e.g., deposited or re-suspended) based on the assumption that near bed turbulent kinetic energy
has a significant influence on the motion of particle close to the bed. According to the previous study,
the V80(=23.5m/h) settling velocity was used to estimate the critical value. Then kc=V802 is the critical

turbulent kinetic energy. The bed turbulent kinetic energy (BTKE) distributions of simulations with
different bed surface roughness height were shown in Figure 7 (ks1, ks2, ks3 and ks4). Regarding the
observed spatial sediment distribution in 2011 and the top view of sediment observation in 2007, all
the simulated BTKE distributions showed similar outer contour of preferential sediment zone except at
the downstream corner near the overflow weir (Figure 7). It suggests that the outer contour was
affected slightly by the surface roughness. It also reveals that the BTKE with an appropriate critical
value could enable to identify the preferential sediment outer contour in the full scale detention basin.
Among all cases, contour simulated with ks1 is similar to the observed one if one just focuses on
upstream blank part (shown in Figure 7 with black trapezoid). The simulation with ks4 shows the best
accordance if one just focuses on central thickness of sediments (shown in Figure 7 with red polygon).
With the assumption that the low BTKE corresponds to the thicker sediment layer, this might suggest
that the sand grain roughness height for sediment layer should be higher than that for bare concrete.
Difference of the zone of outer contour indeed existed but the global shape was similar. For example,
all the cases represent almost the same low value BTKE zone in front of orifice 1 (see Figure 4
labelled with o1) which corresponded to the observation.

Checked points

p2

p1

p4
p3
Figure 8. Layout of the checked points and vertical TKE distribution according to different roughness
8

In order to further understand the effects of surface roughness on the near bed turbulence, local
vertical TKE distribution analysis were carried out at several specific locations. The position of these
local points is shown in Figure 8 (the upper). As shown in Figure 8 (p1, p2, p3 and p4) the near bed
turbulent kinetic energy is affected by the surface roughness and hence is sensitive to the surface
roughness. Basically, the vertical turbulent kinetic energy distribution shows similar profile at almost
all points and for all rough surfaces the maximum bed turbulent kinetic energy is obtained close to the
bed due to the influence of the rough element on the bed (see p1, p3 and p4 in Figure 8). The similar
vertical profile was also obtained with experimental investigation carried out by Dey et al. (2011). But
this is not always the same (see p2 in Figure 8). At p2, simulated secondary currents were observed.
This might be the reason for the difference of vertical TKE profile between p2 and the others. In fact
more vertical secondary currents above the gutter were observed looking into the vertical plane
regarding the gutter stream. They were sensitive to the surface roughness.

kc

(a)

(b)

Figure 9. (a) Global max BTKE at the inlet point versus different rough bed surface and (b) peak
values of TKE according to different rough surface
Figure 9a shows evidently that the global maximum BTKE increases with the increase of the sand
grain roughness height. The global maximum BTKE is often observed near the inlet zone. Figure 9b
reveals that the peak near-bed TKE values show a slight downward trend with the increasing of the
non-dimensional roughness Ks+. The BTKE values remain lower than the threshold kc for points p2, p3
and p4 where sediment depositions are always observed. However, no clear relation has been found in
our case between BTKE values and surface roughness height.

CONCLUSIONS

With the attempt to clearly understand the effect of surface roughness on the near bed turbulence
quantities in a large detention basin, simulations with different surface roughness have been carried
out. Measurement of sediment thickness and distribution were conducted in field site. Comparison
between simulated and observed contours of sediments zones was performed. Analysis of simulated
results compared to observed data reveals that the near bed turbulent kinetic energy distribution could
be used in order to estimate the outer contour of preferential sediment zone. The outer contour is
slightly sensitive to the surface roughness. The maximum value of the near-bed turbulent kinetic
energy distribution in deposition zones is sensitive to surface roughness and is lower than kc. However,
no clear relation has been found between BTKE values and surface roughness height. Different

surface roughness should be set for bare concrete surface and sediment layer as well as other inlet flow
rates.

ACKNOWLEDGEMENTS

The authors thank the OTHU (Field Observatory in Urban Hydrology) at Lyon France and IMU for
scientific support and for financing, ANR (CABRRES Project CESA programme) for financing this
project and the Chinese Scholarship Council for PhD funding for CFD modelling of hydrodynamics
and sedimentation in stormwater detention basin.

REFERENCES

ANSYS Inc. (2011). ANSYS FLUENT Theory guide release 14.0. Canonsburg. U.S.
Akinlade O.G.,Bergstrom D.J., Tachie M.F. and Castillo L.(2004). Outer flow scaling of smooth and
rough wall turbulent boundary layers. Experiments in Fluids, 37, 604-612.
Bagnold R.A.(1966). An approach to the sediment transport problem for general. GEOLOGICAL
SURVEY PROFESSIONAL PAPER 422-I, U.S., Geological Surgey, Washington, D.C.
Bardin J.P. and Barraud S. (2004). Aide au diagnostic et la restructurqtion du bassin de rtention de
Chassieu(Diagnostic and restructuration aid of the retention basin in Chassieu). Rapports pour le
compte de la Direction de lEau du Grand Lyon. INSA de Lyon. Villeurbanne, France.
Cebeci T. And Bradshaw M.G.(1977). Momentum Transfer in Boundary layers. Hemisphere
Publishing Corporation, New York.
Dey S., Sarkar S. and Solari L.(2011). Near-bed turbulence characteristics at the entrainment threshold
of sediment beds. Journal of Hydraulic Engineering, 137(9), 945-958.
Dufresne M., Vazquez J., Terfous A., Ghenaim A., and Poulet J.-B. (2009). CFD modelling of solid
separation in three combined sewer overflow chamber. Journal of Environmental Engineering,
135(9), 776-787.
Dufresne M.(2008). La modlisation 3D du transport solide dans les bassins en assainissement : du
pilote exprimental la louvrage rel (Three-dimensional modeling of sediment transport in
sewer detention tanks : physical model and real-life applicaiton). PhD thesis, INSA de
Strasbourg, Strasbourg, France.
Graf W.H. and Altinakar M.S.(2000).Hydraulique Fluviale: coulement et phnomnes de transport
dans les canaux gomtrie simple(Fluvial Hydraulics: Flow and transport processes in
channels with simple geometry). Trait de Gnie Civil, Volume 16, Lcole polytechnique
fdrale de Lausanne, Publi sous la direction de Ren Walther, Presses polytechniques
Romandes.
Hribersek M., Zajdela B., Hribernik A. and Zadravec M. (2011). Experimental and numerical
investigations of sedimentation of porous wastewater sludge flocs. Water Research, 45, 17291735.
Hager W.H.(2010). Wastewater Hydraulics: theory and practice. 2nd edn, Springer, Switzerland.
Krishnappan B.G.and Marsalek J.(2002). Modelling of flocculation and transport of cohesive seiment
from an on-stream stormwater detention pond. Water Research, 36, 3849-3859.
Launder B.E. and Spalding D.B. (1974). The numerical computation of turbulent flows. Comp.
Methods Appl. Mech. Eng., 3, 269-289.
10

Mignot E., Bonakdari H., Knothe P., Lipeme Kouyi G., Bessette A., Rivire N., Bertrand-Krajewski
J.-L. (2011). Experiments and 3D simulations of flow structures in junctions and of their
influence on location of flowmeters. 12th International Conference on Urban Drainage, Porto
Alegre, Brazil, 11-16 September 2011.
Papanicolaou A.N., Diplas P., Dancey C.L., and Blakarishnan M. (2001). Surface roughness effects in
near-bed turbulence: implication to sediment entrainment. Journal of Engineering Mechanics,
127(3), 211-218.
Perry A.E., Lim K.L. and Henbest S.M.(1987). An experimental study of turbulence structure in
smooth- and rough-wall boundary layers. Journal of fluid Mechanics, 177, 437-466.
Shafi H. S. and Antonia R. A. (1997). Small-scale characteristics of a turbulent boundary layer over a
rough wall. Journal of Fluid Mechanics, 342, 263-293.
Souders D.T. and Hirt C.W. (2002). Modeling Roughness Effects in Open Channel Flows Flow.
Science Report FSI-02-TN60, Flow Science Inc., Santa Fe, N.M.
Stovin V.R, Grimm J.P. and Lau S-T.D.(2008). Solute Transport Modeling for Urban Drainage
Structures. Journal of Environmental Engineering, 134(8),640-650.
Tachie M.F., Dergstrom D.J. and Balachandar R.(2004). Roughness effects on the mixing properties in
open channel turbulent boundary layers. Journal of Fluids Engineering, 126, 1025-1032.
Torres A. (2008). Dcantation des eaux pluviales dans un ouvrage rel de grande taille : lments de
rflexion pour le suivi et la modlisation (Stormwater settling process within a full-scale
sedimentation system: elements of reflection for monitoring and modeling). PhD thesis,
INSA de Lyon, Villeurbanne, France.
Wood M.G., Howes T., Keller J., Johns M.R.(1998).Two dimensional computational fluid dynamic
models for waste stabilization ponds. Water Research, 33, 958-963.
Yan H., Lipeme Kouyi G., Bertrand-Krajewski J.-L. (2011). 3D modelling of flow, solid transport and
settling processes in a large stormwater detention basin. Proceedings of the 12th International
Conference on Urban Drainage, 11-16 September 2011, Porto Alegre, Brazil.

11

9th International Conference on Urban Drainage Modelling

Belgrade 2012

Effects of computational meshes on hydrodynamics of

an open channel junction flow using CFD technique
Adrien Momplot1, Hossein Bonakdari1,2*, Emmanuel Mignot3, Gislain
Lipeme Kouyi1, Nicolas Rivire3, Jean-Luc Bertrand-Krajewski1
1

Universit de Lyon, INSA Lyon, LGCIE - Laboratory of Civil & Environmental Engineering, F-69621
Villeurbanne cedex, France
2
Department of Civil Engineering, Razi University, Kermanshah, Iran
3
LMFA, CNRS-Universit de Lyon, INSA de Lyon, Bat. Joseph Jacquard, 20 avenue A. Einstein,
F-69621 Villeurbanne cedex, France
*Corresponding author, e-mail: bonakdari@yahoo.com

ABSTRACT
This paper deals with numerical calculation of flow structures in open-channel
junction flows which are typical singularities encountered in urban drainage
systems. The objective is to evaluate the impact of the mesh shape, mesh
refinement, and free-surface modelling approach on the simulation. The ability of
CFD strategy (appropriate numerical options, particularly: computational meshes,
discretization schemes, turbulence models linked to wall treatment functions and
air-water interface capturing approach) is determined by comparing simulated
results against experimental one obtained on laboratory scaled open-channel
junction where PIV measurement technic was set-up. Comparisons emphasis on
simulated and measured horizontal velocity field, mixing interface between both
inflows and recirculation zone at two elevations, near the free-surface and close to
the bottom of the channel.
The mesh was refined until no significant changes are observed on the main flow
structures. Nevertheless, whatever the mesh refinement, CFD modelling has some
trouble to simulate accurately the recirculation zone extension near the bottom. The
mesh shape has no influence on the velocity field even if hexahedral meshes give
better representation of the mixing interface in the junction than tetrahedral
meshes. Nonetheless, both mesh types do not enable to represent properly the
recirculation in the near-bottom region. Finally, the use of the VOF model leads to
a similar velocity field, but seems to give less ability to represent the mixing
interface in the junction and the extension of recirculation zone.

KEYWORDS
Computational Fluid Dynamics, Hydrodynamics, Junction, Mesh size

INTRODUCTION

Computational Fluid Dynamics (CFD) offers a suitable understanding of open channel flow pattern
and enables to study flow characteristics in different hydraulic and physical situations, including urban
drainage systems. It is known that CFD solution is strongly impacted by the selected modelling
parameters such as grid, boundary conditions, turbulence model, under-relaxation factors, interface
capturing model, spatial and time discretization schemes, etc. One of the main steps in CFD approach
is the spatial discretization of the fluid body into cells or elements forming a computational grid. Then,
CFD codes compute the solution of mass continuity, energy and momentum equations in each cell of
the defined grid.
The grid quality (shape, size and distribution, density of meshes, number of nodes per length of edge,
etc) is among the most important parameters affecting the accuracy and convergence of a finite
volume calculation. Decreasing the mesh size leads to increase the number of cells and the number of
equations and unknowns. This causes longer iterations of the calculation before reaching convergence
or increases the round-off error and sometimes causes problem of divergence (due to numerical
diffusion). This is especially important in three dimensional modeling of two-phase turbulent flow in
open channel (Bonakdari and Lipeme Kouyi, 2010).
The paper aims to investigate the impact of the mesh characteristics on CFD solution when open
channel junction flows encountered in urban drainage systems are simulated. Indeed, such geometrical
singularities exhibit a complex flow pattern sketched in figure 1 (see Weber et al., 2001) with i) an
interface plane (shear plane) between both inflows where mixing occurs if one inflow contains a
given concentration of passive or solid material, ii) a recirculation region in the downstream branch
with lower velocities where solid material should deposit and passive material should be trapped and
iii) an acceleration region facing the recirculation region where high velocities are encountered and
thus where erosion of previously deposited material should occur. The present paper is thus focused on
the impact of the computational meshes on the CFD prediction of the extension and velocity
magnitude of the mixing interface and recirculation region.

Figure 1. Scheme of the main flow structures (from Weber et al., 2001).

METHODS

2.1

Experimental data

The experiments are performed in the channel intersection facility at the Laboratoire de Mcanique
des Fluides et dAcoustique (LMFA) at the University of Lyon (INSA-Lyon, France), sketched in Fig.
2 (see Rivire et al., 2011). The facility consists of three horizontal glass channels of L=2m length and
b=0.3m width each. The channels intersect at 90 with two inlet branches, labeled the main branch
along x axis with the flow rate Qxi and the lateral branch along y axis with the flow rate Qyi and one
downstream branch along x axis aligned with the main branch. Each inlet branch is connected to a
large storage tank. The water passes through a honeycomb to stabilize and straighten each inflow,
collides in the junction, flows through the downstream branch and is collected by the downstream
tank. A sharp crest weir ends the downstream branch which total length is finally 2.6m. The water
circulation is maintained by pumping from the downstream tank to the inlet tanks. Both inlet flowrates are measured in the pumping loops using electromagnetic flow-meters. The three parameters
which govern the flow configuration are: the inlet flow-rates Qxi= Qyi=2L/s and the water depth at the
downstream end of the downstream branch hd=12 cm which is controlled by the weir. Due to
negligible wall friction, the water depth is almost the same in the whole flow and only a slight water
depth decrease appears (maximum of 3% of hd from the upstream to the downstream tank).
Qxi=2L/s

PIV area
Qxi

y x
L=2m

hd

Upstream
Tank

Downstream
Tank

L=2m

Pump

b=30cm
Qyi

Qyi=2L/s
Lateral
Tank

Figure 2. Scheme of the experimental set up.

Velocity fields are measured using a horizontal PIV technique at two elevations (z = 3 cm and 9 cm).
Polyamid particles (50 m diameter) are added to the water. A white light generator along with a
simple slot of less than 1 mm opening layer is used to create a 5 mm thick light sheet at the desired
elevation in the channel junction. A 1280x1920 pixel CCD-camera connected to a PC computer is
located above the free surface at an elevation of about 1.5 m. Inserting the whole set-up in the dark
finally permits to record the particle motion at the lightened elevation at a fixed frame-rate of 30Hz
during 133s with a horizontal resolution of about 0.5 mm. The commercial software Davis (from
Lavision) permits to correct the optical distortions, to subtract the background and to compute the
mean velocity field. Repeatability, time convergence and spatial coherency were verified and the
resulting measured velocity accuracy is estimated to about 5mm/s.

2.2

Numerical study

The 3D numerical modelling is carried out by means of the commercial Ansys - CFX and FLUENT
(version 14) CFD software packages using a steady-state formulation which solves the threedimensional fundamental flow equations. As for most sewer flows, the experimental junction flow
shows an appropriate Reynolds number (Re ~ 15000 in the upstream branches and 30000 in the
downstream branches). A relevant turbulence model is thus of great importance in order to obtain
accurate numerical results. The key equations for fluid motion in the whole domain are: (1) the
continuity equation for the incompressible fluid in Eulerian approach:

ui
0
xi

(1)

and (2) the Reynolds time-averaged Navier-Stokes (RANS) equations for an incompressible turbulent
fluid flow for steady flow condition:

uj

u i
z 1 P

x j
xi xi x j

u i

ui' u 'j
x

(2)

with i and j = 1, 2 and 3, where xi represents the three coordinate axes, u i the time-averaged velocity
along axis i, z the vertical free-surface elevation, P the pressure, the fluid density and u 'i u ' j the
Reynolds stresses with the prime sign referring to time fluctuations. Solving Eqs. 1 and 2 requires a
turbulence model to set the Reynolds stresses. Among the proposed turbulence models, Bradbrook et
al. (1998), Shakibainia et al. (2010) and Mignot et al. (2011) have shown that the RNG (ReNormalization Group) form of the k- model (initially introduced by Yakhot et al., 1992) accurately
computes the 3D behaviour of a junction flow. This turbulence model is then used for the present
work.
Since Eq. 2 is elliptic, boundary conditions are required. Uniform velocity distributions are set at each
inlet cross-section U = Q/A (A being the wet cross-section) with hydrostatic pressure distribution and
outlet extremity water depth h = hd = 12 cm is specified with zero longitudinal gradient of all flow
characteristics. A sufficient length (10 meters) is provided upstream each inlet branch in order to
obtain a fully developed turbulent velocity profile close to the junction. At the outlet, a hydrostatic
pressure distribution is maintained across the entire cross-section. The rigid walls are considered
smooth with no slip condition and the standard wall function method (proposed by Launder and
Spalding, 1974) (see Table 1). Due to limited water depth variations observed in the experiments, the
free surface is modelled either as a single-phase by rigid lid ( the free surface is considered as a wall,
with free-slip conditions i.e. zero shear stress at the free surface as expected in experiments) or as an
interface using volume of fluid (VOF) method (see Table 1).
2.3

Computational strategy

The construction of a suitable mesh is an important task for the numerical simulations because the
accuracy and the calculation speed depend on the mesh resolution. In this study, different structured
meshes with various cell concentrations are tested. Nine quite uniform rectangular meshes with
decreasing cell sizes (see Fig. 3) are considered. These meshes are organized in three groups: coarse
(meshes 1 to 3), medium (meshes 4 to 6) and refined (meshes 7 to 9). For these meshes, the rigid lid
approach is used to model the free surface. Simulations have been performed under CFX (meshes 1 to
9) and FLUENT (meshes 10 and 11).

Besides, two additional calculations are performed with meshes 10 and 11(see Table 1) in order to
investigate the influence of the mesh shape (mesh 10: tetrahedral mesh) and the possible influence of
VOF model (mesh 11: Cartesian mesh with VOF approach) on CFD solutions. For meshes 10 and 11,
the mesh type and size, the boundary conditions and computational strategy are summarized in Table
1.

Figure 3. Computational meshes 1 to 9. Cell sizes are indicated in millimetres for each mesh, using the
following convention: x*y*z.
Table 1. Description of the computational conditions, the discretization scheme, meshes shape and
density (cell size) for meshes 10 and 11. Hexahedral meshes are detailed as x*y*z.
Mesh shape

Cell size

Discretization scheme/wall function

VOF model

Mesh 10

tetrahedral

Distance between
nodes around 10 mm

Power law/standard wall function

No

Mesh 11

hexahedral
cartesian

20*20*10

Power law/standard wall function

Yes

3
3.1

RESULTS
Velocity fields

Five simulated runs (among the 11 available) are selected and for each selected case, the velocity
fields at two elevations z = 3 cm and z = 9 cm are plotted. Then, comparison against measurements
are done (see Fig. 4). No measurements are available for 1.1<x/b<1.8 in the upstream part of the
downstream channel, due to measurement limitations.
It is observed that for the coarsened meshes (mesh 1) CFD solution cannot represent properly the
recirculation zone and the disturbed velocity profile after the junction (x/b 2.5), as shown on figure
4. For medium meshes (mesh 5), the recirculation zone is represented - however the velocity profile
after the junction is not in agreement with measurements. For refined meshes (mesh 8), the
recirculation zone is better represented and disturbed velocity profile in the downstream branch fit
experimental data even though some differences remain (see Fig. 4). We also observed (not shown
here) that the refinement of mesh 9 does not lead to a significant improvement of the CFD solution
compared to that derived from CFD modelling performed with mesh 8. Comparisons of simulated
results obtained with mesh 8 (cartesian mesh with rigid lid approach) and mesh 11 (same density of
cartesian mesh 8 using VOF model see Table 1) against experimental one highlight the influence of
VOF model on the computed velocity field in the junction (see Fig. 4). The VOF model does not seem
5

to influence significantly the velocity field even though flow recovery downstream the recirculation
region (x/b 3) for z=9cm is computed with better accuracy with VOF model.

Experiments

Coarse mesh (Mesh 1)

Medium mesh (Mesh 5)

Refined mesh (Mesh 8)

Tetrahedral fine mesh (Mesh 10)

Refined mesh with VOF (Mesh 11)

U (m/s)
Figure 4. Experimental velocity profiles (Exp., top line) and simulated one (Num., lines 2 to 6) at z = 3
cm (left column) and z = 9 cm (right column). Arrows are the velocity field; colours are the velocity
magnitude. Numerical results are plotted at each node location.
6

Comparisons between simulated velocity field obtained with mesh 8 (cartesian) and mesh 10
(tetrahedral with similar mesh density) demonstrates the influence of mesh shape on the velocity
profile in the junction (see Fig. 4). One can note that the comparison is more difficult in this case, due
to the non-regular distribution of the nodes in the tetrahedral mesh. It seems that the mesh shape does
not influence significantly the velocity field. Flow recovery downstream the recirculation region (x/b
3.5) for z=9cm is different regarding all calculations but it is not obvious which resemble the most to
experimental data.
3.2

Mixing interface

For all simulated and measured velocity fields, the mixing interface at each elevation is defined as the
streamline which originates near the upstream corner of the junction: x/b~y/b~0 (see Fig. 5).
Fig.5 reveals that the mesh 1 cells arrangement (coarsen group) is sufficient to represent the main
tendency of the mixing interface even though this solution has more difficulties to represent this
interface near the bottom (z=3cm). CFD solution obtained with mesh 6 (medium group) does not
improve the precision and the shape of the interface at z = 3 cm compared to the ones obtained with
mesh 1. Finally, for mesh 9 (refine group) this interface calculation is in fair agreement with
measurements.

(a)

(b)

(c)

Figure 5. Measured (in black) and modeled (in red) mixing interface between the two inflows in the
junction, at z = 3 cm (solid line) and z = 9 cm (dashes) for coarse meshes (mesh 1) (a), medium
meshes (mesh 6) (b) and refined meshes (mesh 9) (c).
Fig. 6 shows the influence of mesh shape and VOF model. Comparison between mesh 9 and mesh 10
(see figure 6 (a) and (b) respectively) shows the impact of the mesh shape. The tetrahedral shape has a
strong impact on the position of the mixing zone in the near bottom region. Comparison between mesh
9 and mesh 11 (see figure 6 (a) and (c) respectively) shows the impact of the VOF model. We can
observe that the mixing interface is not well captured in the near bottom region when using VOF
model.

(a)

(c)

(b)
(b)

(a)

(c)

Figure 6. Measured (in black) and modelled (in red) mixing interface between the two inflows in the
junction, at z = 3 cm (solid line) and z = 9 cm (dashes) for mesh 9 (cartesian without VOF model) (a),
mesh 10 (tetrahedral without VOF model) (b) and mesh 11 (cartesian with VOF model) (c).
Differences between the VOF solutions and others can be explained by the impact of the recirculation
zone. This zone induces centrifugal effects which influence the pressure field and water depths, even if
this impact is low (for example, the water depths gradient equals about 2 mm along the length, and 0.2
mm along the width in the recirculation zone). Indeed VOF model allows the free surface to be
disturbed by these effects, whereas the rigid lid boundary condition doesnt. Additionally, even if the
rigid lid condition seems to give better results, it needs to know a priori the heights field to have a
good representation of the velocity field.
3.3

Extensions of the recirculation region

For both elevations, the length of the recirculation region L is defined as the location where the mean
streamwise velocity (along x axis) of the measurement or CFD solution point closest to the side wall
(y~0) changes sign from negative (towards the junction) to positive (towards downstream). Moreover,
the width of the recirculation region is defined as the domain in which the net flow discharge (along x
axis) computed from the side wall (y=0) to the recirculation width equals 0. The maximum width,
noted B herein, is simply the maximum measured or simulated recirculation width, occurring between
x/b=0 and x/b=1+L/b.
L and B are evaluated for each CFD solution and are compared to the measured one and data available
in the literature (Gurram et al., 1997, Borghei et al., 2003, Best and Reid, 1984). Results of
comparison are presented in Table 2 for L and Table 3 for B. It should be noted that B could not be
evaluated from CFD solution obtained with meshes 1 to 4.
Table 2. Normalized length L/b of the recirculation zone for each mesh at two elevations: z = 3 cm and
z = 9 cm.
Measured Mesh 1

Mesh 6

Mesh 8

Mesh 9

Mesh 10 Mesh 11 Gurram

Borghei

Best

z = 3 cm 1.25

1.33

2.33

2.00

2.05

2.37

1.66

z = 9 cm 1.97

2.13

2.67

2.77

2.37

2.50

1.88

2.53

1.70 1.87

Table 3. Normalized maximum width B/b of the recirculation zone for each mesh at two elevations: z
= 3 cm and z = 9 cm.
Measured Mesh 1

Mesh 6

Mesh 8

Mesh 9

Mesh 10 Mesh 11 Gurram

Borghei

Best

z = 3 cm 0.19

NaN

0.20

0.27

0.28

0.38

0.2

z = 9 cm 0.33

NaN

0.20

0.33

0.35

0.30

0.13

0.37

0.47

0.37

Tables 2 and 3 reveal that CFD approach mostly overestimates the length of the recirculation zone L
compared to measurements except for mesh 11 case. The tetrahedral shape and the VOF model do not
improve the estimation of the recirculation length at z=3cm and z = 9cm. Concerning the maximum
recirculation width B, neither VOF model nor tetrahedral shape achieve an accurate estimation at all
elevations while a clear tendency is observed from measurement data.

CONCLUSIONS

The aim of the present paper was to investigate the ability of CFD codes to simulate subcritical open
channel junction flows - with a specific attention on the impact of the mesh characteristics using
prismatic and tetrahedral cells and of the free-surface modelling approach on the CFD solution. The
ability of CFD strategy is assessed through comparisons against laboratory experimental data for
which velocity fields in the junction are measured using a horizontal PIV technique.
This comparison is focused on three criteria: the simulation at two elevations of i) the general flow
pattern, ii) the mixing interface between both inflows in the junction and iii) the length and width of
the recirculation zone in the downstream branch. Among all simulated runs, meshes 8 and 9 (using
refined quite uniform rectangular cells aligned with the main flow and a rigid-lid approach for the
free-surface representation) appeared to be the most efficient solution based on these criteria.
Nevertheless, these two simulations do not reproduce with enough accuracy all criteria: the location of
the mixing interface in the near-bottom region differs between measurements and simulations and the
length of the recirculation zone is severely overestimated by the numerical models. In addition, the
refinement of the mesh 9 compared to mesh 8 appears not to improve significantly CFD solution.
Among all simulations, CFD solution performed with mesh 8 seems to have further capability to
simulate hydrodynamics across subcritical open-channel junction.
Finally, when simple and widely used CFD approaches (RANS, k- model) are selected, the
mixing-interface near the free-surface is accurately computed by very coarse meshes (mesh 1 for
instance) - however the mixing-interface curve near the bottom and the recirculation extension in the
downstream branch are simulated with poor accuracy even for the most elaborated CFD strategy (for
example solution with mesh 9). Due to the importance of these flow structures for pollutants mixing or
sediment transport, additional work is obviously required before reaching a satisfactory CFD strategy
which enables to better simulate passive and solid transport and dispersion in such structures.

ACKNOWLEDGEMENTS

The research was prepared in the frame of OTHU and IMU, Lyon. It was funded by the INSA-LyonBQR program, the French INSU-EC2CO-Cytrix-2011 project No 231 and ANR-11-ECOTECH-007MENTOR projects.

REFERENCES

Akoz, M.S., Kirkgoz, M.S., Oner, A.A. (2009). Experimental and numerical modeling of a sluice gate
flow. J. Hydraulic Research. 47(2), 167-176.
Best, J. L. and Reid, I. (1984). Separation zone at open-channel junctions. Journal of Hydraulic
Engineering, 110(11): pp 1588-1594.
Bonakdari H., Lipeme Kouyi G. (2010). The study of geometric and distribution effects of 3D mesh
on CFD Modeling of two-phased turbulent flow. Hydraulics, Water Resources, Coastal and
Environmental Engineering Conference, Hydro 2010, Ambala, India, 16-18 December 2010.
Borghei S. M., Beranghi A. and Daemi A. R. (2003). Open channel flow junction with different
channel width. XXX IAHR Congress, Thessaloniki-Greece.
Bradbrook K.F., Biron P.M., Lane S.N., Richards K.S. and Roy A.G. (1998). Investigation of controls
on secondary circulation in a simple confluence geometry using a three-dimensional numerical
model. Hydr. Processes, 12(8), 13711396.
Gurram S. K., Karki K. S. and Hager W. H. (1997). Subcritical junction flows. Journal of Hydraulic
Engineering, 123(5): pp 447-455.
Launder B.E. and Spalding D.B. (1974). The numerical computation of turbulent flows. Comp.
Methods Appl. Mech. Eng., 3, 269-289.
Mignot E., Bonakdari H., Knothe P., Lipeme Kouyi G., Bessette A., Rivire N., Bertrand-Krajewski
J.-L. (2011). Experiments and 3D simulations of flow structures in junctions and of their
influence on location of flowmeters. 12th International Conference on Urban Drainage, Porto
Alegre, Brazil, 11-16 September 2011.
Rivire, N., G. Travin, and R. J. Perkins (2011), Subcritical open channel flows in four branch
intersections, Water Resour. Res., 47, W10517.
Shakibainia A., Tabatabai M.R.M. and Zarrati A.R. (2010). Three dimensional numerical study of
flow structure in channel confluences. Can. J. Civ. Eng., 37, 772-781.
Weber L., Schumate E. and Mawer N. (2001). Experiments on Flow at a 90 Open-Channel Junction.
J. Hydr. Engin., 127(5), 340-350
Yakhot V., Orszag S.A., Thangam S., Gatski T.B. and Speziale C.G. (1992). Development of
turbulence models for shear flows by a double expansion technique. Phys. Fluids, 4(7), 1510
1520.

10