You are on page 1of 10

Water Management

Volume 167 Issue WM5


A nested grid based computational fluid
dynamics model to predict bridge pier scour
Baranya, Olsen, Stoesser and Sturm

Proceedings of the Institution of Civil Engineers


Water Management 167 May 2014 Issue WM5
Pages 259268 http://dx.doi.org/10.1680/wama.12.00104
Paper 1200104
Received 27/09/2012
Accepted 14/12/2012
Published online 24/04/2013
Keywords: fluid mechanics/hydraulics & hydrodynamics/river engineering
ICE Publishing: All rights reserved

A nested grid based


computational fluid dynamics
model to predict bridge pier
scour
1
j
Sandor Baranya MSc, PhD

3
j
Thorsten Stoesser MSc, PhD

Research Fellow, Water Management Group of the Hungarian


Academy of Sciences, Department of Hydraulic and Water Resources
Engineering, Budapest University of Technology and Economics,
Budapest, Hungary

Professor, Hydro-environmental Research Centre, School of


Engineering, Cardiff University, Cardiff, UK
4
j
Terry W. Sturm MSc, PhD

Professor, School of Civil and Environmental Engineering, Georgia


Institute of Technology, Atlanta, GA, USA

2
j
Nils Reidar B. Olsen MSc, PhD

Professor, Department of Hydraulic and Environmental Engineering,


Norwegian University of Science and Technology, Trondheim,
Norway

1
j

2
j

3
j

4
j

A three-dimensional numerical model that computes hydrodynamics and morphodynamics is applied to the problem
of calculating local scour around cylinders. A nested grid approach is implemented in the numerical scheme in order
to decrease the overall computational demand but at the same time allow for high spatial resolution around the
cylinder to increase numerical accuracy there. The computational grid consists of two structured blocks a coarse
block that discretises the entire domain and a fine block that only discretises the area around the cylinders. A novel
interpolation technique accomplishes communication between the two blocks. Sediment transport in the morphodynamic model is calculated with a bed load transport formula combined with the sediment continuity equation to
reproduce the unsteady scour hole development while the flow field is recalculated at each time step. Experimental
data of flow and the local clear-water scour around single and multiple circular cylinders in a laboratory channel with
a movable bed are obtained and compared with the results of the model. Flow, the maximum scour hole depth and
the general scour hole geometry are reproduced well in all cases. The time evolution of the scouring is also captured
reasonably well.

Notation
a
D
d
d50
d90
Fr
g
h
k
ks
n
p9
Q
qb

weighting factor in discretised equations


pier diameter (m)
grain diameter (m)
mean grain diameter (m)
grain diameter where 90% of the particles are smaller
Froude number
acceleration due to gravity (m/s2 )
flow depth (m)
turbulent kinetic energy (m2 /s2 )
roughness height (m)
unit vector normal to the bed surface
porosity
discharge (m3 /s)
bed load transport rate (kg/(s/m))

r
S0
TE
u
Ui
u
u c


rs
rw

vector from the centre of surface s to the coarse grid cell


bed slope
time to equilibrium scour
velocity in bed cell (m/s)
time-averaged flow velocity (m/s)
bed shear velocity (m/s)
critical bed shear velocity (m/s)
transverse bed slope (8)
dissipation rate of turbulent kinetic energy (m2 /s3 )
longitudinal bed slope (8)
variable in nested grid method (velocity, turbulence,
sediment concentration, etc.)
kinematic viscosity of water (m2 /s)
density of sediment (kg/m3 )
density of water (kg/m3 )
259

Water Management
Volume 167 Issue WM5

A nested grid based computational fluid


dynamics model to predict bridge pier
scour
Baranya, Olsen, Stoesser and Sturm


c

reason for this is that the morphological timescale is several


orders of magnitude greater than the timescale of the turbulence
features that need to be resolved to accurately describe the flow.
This results in extremely long simulation times. For instance, for
the present case of pier scour, vortex shedding occurs at the
timescale of a fraction of a second, whereas morphological
changes occur over hours or even days.

c,S


1.

bed shear stress (N/m2 )


critical bed shear stress for movement of sediment
particles (N/m2 )
critical bed shear stress for movement of sediment
particles according to Shields curve (N/m2 )
angle of repose of sediment particle (8)

Introduction

Local scour is a typical morphological feature in alluvial rivers at


locations where sudden changes in flow conditions occur. It can
be induced by flow obstruction caused by engineered structures
such as groins or bridge piers, but it can also occur naturally at,
for instance, river confluences. Excessive local scouring may
endanger the stability of a hydraulic structure and lead to its
failure. Therefore, investigation of scour development processes
is of primary importance and an intensive research area for
engineers and scientists due to the complexity of the phenomenon
(e.g. Amini et al., 2010; Ghodsian et al., 2012). In addition to
empirical formulas developed to estimate maximum scour depth
and physical model experiments carried out in flumes with a
movable sediment bed, computational modelling has become a
suitable investigation method. Focusing on the latter, the purpose
of this paper is to introduce a numerical morphodynamic model
for the analysis of local scouring phenomena.
As described in the literature, the development of scour at bridge
piers is the result of the interaction between highly threedimensional (3D) flow, which is the result of flow obstruction and
separation leading to the development of the horseshoe and wake
vortices, and the alluvial river bed (e.g. Dargahi, 1989, 1990;
Graf and Istiarto, 2002; Roulund et al., 2005). At such locations,
the complex flow field can produce significant changes in river
bed geometry in the immediate vicinity of a pier during a flood
event. In order to develop a numerical tool capable of predicting
scour with acceptable accuracy not only for laboratory conditions
but also for prototype engineering problems, a numerical model
should be of relatively low computational cost and should include
3D description of flow using an appropriate turbulence model
coupled hydrodynamic, sediment transport and
morphodynamic models
j detailed and accurate prediction of flow and morphological
features in the vicinity of the obstruction.
j
j

Large eddy simulation techniques have been increasingly used to


simulate complex free surface flows with improved description of
the flow details (e.g. Bomminayuni and Stoesser, 2011; Koken
and Constantinescu, 2008; Stoesser et al., 2009, 2010b; van Balen
et al., 2010). Some of these models have dealt with flow analysis
around obstacles (e.g. Catalano et al., 2003; Gobert et al., 2010;
Kirkil et al., 2008; Paik et al., 2010; Palau-Salvador et al., 2010;
Stoesser et al., 2010a; Tseng et al., 2000). However, due to their
extremely high computational demand, the coupling of such
models with sediment transport and morphological models is still
not feasible for large-scale engineering problems. The main
260

One of the very first studies of computational modelling of local


scouring around a cylinder was presented by Olsen and Melaaen
(1993), who used a 3D numerical flow model. Their model solved
the Reynolds averaged NavierStokes (RANS) equations together
with a k turbulence closure. The model used the bed shear
stress to calculate sediment transport and concentration at the bed,
based on the empirical formula of van Rijn (1984). Olsen and
Melaaen found acceptable agreement with experimental results
but the maximum scour depth was not verified. As a continuation
of this work, Olsen and Kjellesvig (1998) applied the same
numerical model to the same case but with increased computer
capacity. They were able to model the evolution of the scour
around the circular cylinder in more detail but the grid resolution
was still relatively coarse. The calculated maximum scour depth
corresponded well with the results of empirical formulas. Moreover, the calculated spatial flow features also showed satisfactory
agreement with observations from physical model studies.
Roulund et al. (2005) carried out numerical and physical experiments of flow and scour around a circular pile. The 3D numerical
model EllipSys3D was used in their work to solve the RANS
equations together with a k turbulence closure. The channel in
their study was discretised on a structured, curvilinear multi-block
grid system. They validated their model with experimental data
and studied the horseshoe vortex and lee-wake processes. Both
steady and unsteady calculations were carried out for flow modelling. As a significant outcome, they pointed out that the timeaveraged bed shear stress obtained from the unsteady calculations
was very close to that obtained from the steady solution, justifying
the use of steady-state simulation for morphodynamic modelling.
They reported 15% under prediction of the scour depth upstream
of the pier and 30% under prediction on the downstream side. The
latter was attributed to the steady-state description of flow.
Nagata et al. (2005) developed a 3D numerical model to study
scouring at hydraulic structures with application to a spur dike and
a bridge pier. The model solved the RANS equations with the k
turbulence closure on a structured, non-orthogonal grid. The flow
model was coupled with a stochastic model for sediment pick-up
and deposition. They validated the model against the experiment
of Melville (1975), who investigated the scour geometry around a
circular pier. The qualitative comparisons of the flow field showed
satisfactory agreement with respect to capturing the typical flow
features for this case. The coupled morphological model resulted
in an overall adequate agreement of the developed scour geometry; however, somewhat smoother edges of the scour hole were
predicted on the upstream side of the pier.

Water Management
Volume 167 Issue WM5

A nested grid based computational fluid


dynamics model to predict bridge pier
scour
Baranya, Olsen, Stoesser and Sturm

The numerical model presented here solves the 3D RANS equations with a k turbulence closure on a combined grid consisting
of coarse and nested fine grids. The basic idea is to use high
spatial resolution in the area of local scour and much coarser
resolution elsewhere to save computational time. The method can
be highly effective in cases where the size of the structures being
studied is several orders of magnitude smaller than the river.
Similarly, the timescale of the scour process is assumed to be
much larger than that of the turbulence structure causing scour,
so that a quasi-steady treatment of the flow field can be coupled
with an unsteady morphodynamic model. For model verification
purposes, a laboratory experiment was conducted in which
morphological changes around single and double cylindrical piers
were quantified. Different arrangements of the bridge piers
(aligned at various angles relative to the main flow direction) and
their effect on scour development were studied.

Discharge Q: m3 /s
Flow depth h: m
Slope S0
Froude number (Fr)
Sand grain diameter d50 : m

The main objective of this paper is to provide a method for


modelling of scour around multiple piers in engineering applications using a 3D RANS model with a nested grid to account for
widely varying spatial scales, and an unsteady morphodynamic
model coupled with a quasi-steady flow field in order to account
for the widely varying timescales. Such a numerical tool yields a
very efficient investigation method for real river engineering
problems due to the fact that it saves considerable computational
time compared with structured grid solvers.

2.

Laboratory experiments

For numerical model validation, laboratory experiments were


conducted in the Hydraulics Laboratory at the Georgia Institute
of Technology. A 2.75 m long test section was located 17 m
downstream of the entrance of a 1.1 m wide by 24 m long tilting
flume. The approach channel bed and the test section consisted of
a movable bed formed from a depth of 0.18 m of uniform sand to
investigate flow velocities and morphological changes around
distinct circular pier layouts. The three model pier arrangements
(cases A, B and C) shown in Figure 1 were chosen for the study.
The pier models in cases A, B and C were identical PVC tubes
having an outside diameter of D 0.0476 m. Uniform flow
conditions were set with the flume tailgate by carefully measuring
water surface profiles over the full length of the flume until
uniform flow was achieved for the given flow and sediment
parameters summarised in Table 1.

Case A

Case B

Case C

4D

4D
30

45
1

Figure 1. Plan view of pier arrangements; flow direction is from


left to right

0.057
0.162
0.000364
0.26
0.0011

Table 1. Laboratory model parameters

At Q 0.057 m3 /s, scour around the piers occurred but no


sediment transport was observed on the bed in the approach flow
section of the flume. The experimental conditions produced clearwater scour with u /u c , 0.9, where u is the bed shear velocity
and u c the critical bed shear velocity. For the multiple-pier cases
(B and C), detailed velocity measurements were carried out in the
vicinity of the piers using acoustic Doppler velocimetry (ADV).
The velocity measurements were done for fixed bed conditions by
applying polyurethane spray to the sediment around the piers to
avoid bed erosion. A 25 Hz sampling frequency was used with a
sampling time of at least 2 min; filtering of the velocity data as
described by Lee and Sturm (2009) was then applied.
Water depths and scour depths near the pier were measured using
a point gauge with an uncertainty of 0.5 mm while most of the
scour depths were measured by pinging the bed with the ADV
with an uncertainty in depth of 1.0 mm, which was determined
by calibration between the required range of point gauge and
ADV bottom elevation measurements. For the ADV bottom
elevation measurements, the flume tailgate was raised at the end
of the experiment and the discharge gradually reduced to zero.
The ADV probe head was set at a fixed elevation on the
instrument carriage and moved very slowly from one horizontal
position to the next so as not to disturb the submerged bed as the
scour hole bathymetry was measured. The distance of the probe
from the scoured bed varied from approximately 6 to 18 cm,
again to avoid any bed disturbance.
As to the morphological study, each case was run for (a very
conservative) 5 days (120 h) until equilibrium scour conditions
were achieved based on the results of Lee and Sturm (2009).
Melville and Chiew (1999) reported that the time to equilibrium
scour (TE ) varies, and depends on grain size, cylinder diameter,
approach velocity and the ratio of approach velocity to critical
velocity (U/Uc ). Their recommended formula using the values of
these parameters corresponding to the experiments in this study
yielded TE  75 h and approximately 96% of the scour depth
reached at t 24 h (Melville and Chiew, 1999). Indeed, visual
checks during all experiments confirmed similar behaviour
(regardless of the set-up) and, after approximately 3 days of
experiments, hardly any sediment was moved by the flow.
Because the scour hole development follows an exponential
function (i.e. most of the scour process occurs immediately after
scour hole initiation (Melville and Chiew, 1999)), the scour hole
geometry was measured not only at equilibrium but also after a
duration of 1 h of the physical experiments, but only for the
261

Water Management
Volume 167 Issue WM5

A nested grid based computational fluid


dynamics model to predict bridge pier
scour
Baranya, Olsen, Stoesser and Sturm

multiple-pier cases. These experiments are referred to as halfscour experiments, because the observed scour depth magnitude
was approximately half the equilibrium scour depth. To measure
the scour hole profiles along the pier centrelines, the discharge
was reduced gradually to zero over a period of 30 min, either at
the end of the experiment for the equilibrium case or after 1 h of
the experiment for the half-scour experiments, respectively. After
the water had drained off the flume, the scour hole profile was
measured along the pier centrelines with a point gauge having an
uncertainty of 0.5 mm.

algorithm finds the four coarse grid cells closest to s, but outside
the fine grid region. To make sure the cells are outside the fine
grid region, a dot product of n (the vector normal to surface s,
pointing out of the fine grid domain) and ri (the vector from the
centre of surface s to the coarse grid cell, i) is used. When the
dot product is positive, the coarse cell is located on the outside of
the fine grid.

3.

Numerical model

The numerical finite-volume model used in this study solves the


3D RANS equations with a k turbulence closure (Olsen, 2010),
discretising the domain on a curvilinear grid system using a
nested grid approach. The flume flow field was discretised on a
relatively coarse grid. In order to gain detailed insights into the
flow features and calculate bed changes accurately around the
pier, this zone near the pier was described with a fine grid placed
inside the coarse one. A space resolution of 0.05 m was used for
the coarse domain, while an average resolution of 0.0015 m was
established for the fine grid (Figure 2). Grid refinement was
needed close to the smooth cylinder wall to capture the flow
separation point accurately. For the flow modelling, fixed bed
conditions were studied with different vertical resolutions. For the
single-pier case, the study used both 12 and 20 uniformly
distributed layers vertically to study the influence of vertical grid
resolution; for the double-pier cases, 12 layers were used. The
variation in the free surface was not calculated because of the
relatively small variation in water surface (small Froude number)
and the relatively large water depth to pier diameter ratio. During
the morphological modelling, the sigma transformation was
applied together with the possibility of generating new hexahedral
cells if the water depth increased (details of the grid generation
are given by Olsen (2003)). For the single-pier case, approximately 200 000 and 330 000 cells were used initially for the
whole domain for the 12 and 20 layer cases, respectively. For the
multiple-pier cases, larger nested grids had to be implemented in
order to reproduce the more complicated flow field while keeping
the same refinement at the cylinder walls as was used for the
single-pier case. Therefore, wider and longer domains were
discretised with the fine grid, resulting in 540 000 and
730 000 cells for cases B and case C (Figure 2), respectively.
The next question is how to couple the fine and the coarse grid.
One-way coupling uses the values from the coarse grid to give
the boundary conditions for the fine grid. Two-way coupling
additionally gives information from the fine grid to the coarse
grid. One-way coupling is described first.
The algorithm needs to provide values for the velocities, turbulence variables and sediment concentrations where the water
flows into the fine grid. The values need to be interpolated to
each boundary surface, s, of the fine grid. First, the coarse grid
cells for the interpolation have to be found. The currently used
262

The value on the boundary at surface s is then computed as a


weighted average of the values in the four coarse grid cells. The
weighting factors are computed from the dot product of the
velocity in the coarse grid cell (uc,i ) and the vector pointing from
the coarse grid to the centre of surface s (nsc,i ). The boundary
value on surface s of one of the parameters (s ) is then computed
from the following formula, where c is the value in the coarse
grid

1:

P4
i1 c,i max(0, nsc,i g uc,i )
s P
4
i1 max(0, nsc,i g uc,i )

The parameter is then velocity, turbulence variables or sediment


concentrations. Note that nsc,i ri. Using the dot product in
Equation 1 means more weight is given to the coarse cells
upstream of the surface. If the dot product is negative from all
the four coarse cells, then the surface has an outflow and zero
gradient boundary conditions are used.
Two-way coupling means that the fine grid values should affect
the results in the coarse grid. This proved to be more complicated, as instabilities often occurred. The most promising method,
which was used in the current study, only affected the coarse grid
cells on the boundary of the fine grid. The sum of the weighting
factors, ap , in the discretised equations was used to modify the
source term (S) for these coarse cells (details of the numerical
solution are given by Olsen (2012)). This was done for the
velocity and the turbulence equations. The following equation
was used for the coarse cell grid p closest to the surface area that
bordered the fine grid cell f

2:

S p S p r

j f

!
 c ap

where * denotes the original values. The j parameter in the


formula is the number of fine grid cells for each coarse cell and
f is the value in the cell of the fine grid. The parameter r is a
relaxation factor, which was set to 0.01.
In order to accurately describe the effect of the rough boundary
at the flume bed a roughness height (ks ) of 0.006 m was defined
based on the formula given by van Rijn (1984): ks 3d90 , where
d90 is the grain diameter for which 90% of the sediment sample

Water Management
Volume 167 Issue WM5

A nested grid based computational fluid


dynamics model to predict bridge pier
scour
Baranya, Olsen, Stoesser and Sturm

10

06

y: m

08

04
02
0
15

10

20

25
x: m
(a)

30

35

40

060

y: m

055

050

045
195

200

205

210
x: m
(b)

215

220

225

070

065

055

y: m

060

050

045

040
205

210

215

220

225

230
x: m
(c)

235

240

245

250

Figure 2. (a) Whole numerical grid for case A. (b) Detail of


nested grid for case A. (c) Nested grid for case C

263

Water Management
Volume 167 Issue WM5

A nested grid based computational fluid


dynamics model to predict bridge pier
scour
Baranya, Olsen, Stoesser and Sturm

is finer by weight. Close to the bed, the well-known log law was
used. At the vertical walls, due to the smooth surfaces there, the
wall laws for smooth boundaries were applied.

In this case,  is not equal to the angle of internal friction, but


depends on the flow direction (Lysne, 1969). In Lysnes study, sand
movement in tunnels was investigated and it was concluded that,
for uphill flow, the value of  varies between 358 and 408 whereas
for downhill flow it can be as high as 528. In this study, therefore,
the value of  was calculated based on the flow direction and
interpolated between 408 and 528 with respect to slope angle

For the morphodynamic modelling, the flow model is coupled with


a sediment transport model together with a movable bed model
consisting of the sediment continuity equation. During the unsteady morphodynamic modelling, the bed shear stress distribution
calculated by the hydrodynamic model of the current time step is
used as input for the sediment transport modelling in the next time
step. The empirical sediment formula of van Rijn (1984) was
applied for modelling bed load transport. According to van Rijns
experiments, the rate of bed load transport can be expressed as

qb

7:

6:

d 1 5 f(rs  rw )g=rw g0 5

nx ux ny uy
:
:
(n2x n2y )0 5 (u2x u2y )0 5





1
1
52
 40 1 
2
2

3:

0:053

(  c )=c 2 1
0:3

d f(rs  rw )g=rw 2 g0 1

where qb is the bed load discharge, d is sediment particle


diameter,  is grain bed shear stress, c is the critical bed shear
stress for movement of sediment particles, rs is density of
sediment, rw is density of water,  is the kinematic viscosity of
water and g is acceleration due to gravity. In the absence of
bedforms,  is the total bed shear stress.

where n x and n y are components in the x and y directions of the


unit vector normal to the sloping bed surface of the given cell
and u x and u y are the velocities in the x and y direction in the
bed cell. In addition, a sand slide algorithm is used to prevent the
bed slopes from being steeper than 328. The algorithm moves
grid intersections vertically down and up to prevent steep slopes.
The method is based on the angle of repose for the material and
the continuity of sediments during the move.

4.

4:

(1  p9)

@qb,y
@q
@zb
 b,x 
@t
@x
@y

where p9 is the bed porosity.


Because the scour hole geometry includes steep slopes, especially
on the upstream part of the scour hole, the effect of a sloping bed
on incipient sediment motion had to be taken into account in this
study. Several researchers (Brooks, 1963; Dey, 2001; Ikeda, 1982;
Kovacs and Parker, 1994; Lane, 1955; Lysne, 1969) studied this
phenomenon by introducing formulas for the modification of the
critical bed shear stress derived from Shields curve (c,S ). In this
study, the empirical expression of Dey (2001) was combined with
the findings of Lysne (1969). Dey (2001) introduced a formula
for the ratio of critical bed shear stress for a sloping bed to that
for a horizontal bed considering transverse bed slope (), longitudinal bed slope () and the angle of repose of the sediment
particle ()

4.1 Single-pier case


Flow simulations of fixed flat-bed conditions were carried out
first in order to produce a fully developed flow field for
morphological modelling. The accuracy of the computed approach flow was confirmed by comparing computed and measured profiles of the streamwise velocity (U1 ) and the turbulent
kinetic energy (k) located a distance of three pier diameters
upstream of the cylinder; acceptable agreement was found (Figure
3). Furthermore, the calculated turbulent kinetic energy at the

z: m

Once the bed load is calculated, the bed level changes (zb ) can be
computed using the sediment mass-balance for the bed cells

Results

016

016

014

014

012

012

010

010

008

008

006

006

004

004

002

002
0

0
0

02 04
U1: m/s
(a)

06

Exp.

5:

264


: 
:
c
0 745
0 372
0:954 1 
1


c,S

00025 00050
k: m2/s2
(b)

CFD

Figure 3. Approach streamwise velocity (a) and turbulent kinetic


energy (b) profiles in three-diameter distance upstream of the pier

A nested grid based computational fluid


dynamics model to predict bridge pier
scour
Baranya, Olsen, Stoesser and Sturm

flume bed corresponded well with the measured value, which


implies an accurate calculation of the bed shear stress in the
approach flow field (since  is calculated based on k at the flume
bed, using the formula  0.3rw k (Rodi, 1980)).

015
010
005

The hydrodynamic conditions around the piers for the different


set-ups were then validated with measured data. Further details of
the hydrodynamic model validation are described in a separate
paper (Baranya et al., 2012).
The morphological simulations were run for 24 h until approximately 96% of the expected equilibrium scour geometry was
achieved. Running the simulation until full equilibrium seemed
unreasonable due to the excessive additional computing time (an
additional 48 h of physical time to produce another 4% of scour).
Figure 4 shows the time evolution of bed changes for case A for
four chosen points along the flume centreline. The first point is
located in front of the pier on the upstream side, where maximum
scour depth occurs; the following three points are on the downstream side of the pier at distances of one, two and three pier
diameters from the downstream pier wall, respectively. As to the
maximum scour depth at the toe of the cylinder, which develops
rapidly in the first few hours, the equilibrium condition was reached
after approximately 1 day. The time evolution of bed elevations at
the downstream portion of the pier indicates the movement of a
dune, as positive values of about 0.02 m of bed elevation changes
appear in the first hour. This phenomenon was also observed in the
laboratory experiments, albeit with a somewhat higher amplitude
(0.020.04 m) and slower motion of the dune.

Bed change: m

Figure 5 shows measured and simulated equilibrium scour


geometry along the centreline of the flume for case A for the
cases of 12 and 20 vertical layers. Streamlines from the flow
simulation are also shown for the 20 layer case. The vertical flow
recirculation at the upstream toe of the pier, a signature of the
horseshoe vortex, can be seen clearly. The computational fluid
dynamics (CFD) model successfully reproduced the steep slope
on the upstream side of the pier irrespective of grid resolution. In
the lee of the pier, however, differences in scour depth can be
seen, in particular for the coarser grid. Apparently, the higher
vertical grid resolution yields better agreement with experimental
results, which is due to a better resolution of near-bed velocity

002 0
0
002
004
006
008
010
012
014

Time: h
10
15

20

25
3D
2D
1D
Max.
scour

Figure 4. Time series of bed changes for case A (20 layers) at the
upstream toe of the pier (Max. scour) and at 1D, 2D and 3D
downstream of the pier respectively (D is pier diameter)

0
005
010

Exp.
CFD 20 layers
CFD 12 layers

195

200

205

210 215
x: m

220

Bed level: m

Water Management
Volume 167 Issue WM5

015
225

Figure 5. Measured and calculated equilibrium scour profiles for


case A; plotted path lines are derived from the 20 layer case

gradients and hence a more accurate description of turbulence


production. However, from an engineering aspect, estimation of
the maximum scour depth, which occurs at the upstream side of
the pier, is predicted quite well by employing only 12 vertical
layers. Since computational demand plays a crucial role when
dealing with real engineering problems, it was considered reasonable to use the lower vertical grid resolution (i.e. 12 layers) for
the double-pier cases. In such a way, computational time could be
significantly reduced (by approximately 60%).
4.2 Double-pier cases
For the multiple-pier cases, scour profile measurements at T 1 h
were also available from the experiment. These results are used in
the present study to validate the time development of the scour
hole as shown in Figure 6. As before, the scour profiles upstream
of the piers show good agreement with measurements, whereas
some deviation is observed in the wake zone. For case B, the
predicted scour profile for pier 1 agrees very well with the
experimental data (Figure 6(a)). For the multiple-pier cases, bed
erosion is not only the result of flow conditions around the
individual piers, but is also due to flow contraction between the
piers, which influences the scouring process. Scouring due to
contraction plays an important role, especially in case B, for which
the area between piers is smaller than for case C. The behaviour of
the scour profile of pier 2 (Figure 6(b)), however, exhibits similar
inaccuracies to those for the single-pier case at 24 h (i.e. underestimation of scour magnitude behind the cylinder). Also worth
noting is that there is an overestimation of downstream erosion
after 1 h. Clearly, this observation indicates that the deposition
zone downstream of the pier moves faster in the numerical model
than in reality, and that equilibrium conditions are reached more
rapidly in the numerical model than in the experiments.
Figure 7 presents a perspective view of the calculated contours of
265

Bed change: m

Water Management
Volume 167 Issue WM5

002 18
0
002
004
006
008
010
012
014

A nested grid based computational fluid


dynamics model to predict bridge pier
scour
Baranya, Olsen, Stoesser and Sturm

x: m
20

22

24

26

28

24

26

28

Bed change: m

(a)
002 18
0
002
004
006
008
010
012
014

x: m
20

22

Exp. 1 h
Exp. eq.
Calc. 1 h
Calc. 24 h
(b)

Figure 6. Measured and calculated half scour, after 1 h and


equilibrium scour profiles for case B along the centreline of (a)
pier 1 and (b) pier 2

of flow contraction in this zone. On the downstream side of both


piers, again, ridges evolve as a typical feature in the wake zone.
However, the axis of the ridge behind pier 1 is not parallel with
the flume centreline, but deviates approximately 208 from it due
to the influence of the flow direction in the contracted flow zone.
The intensive flow between the two piers will eventually result in
the formation of a trough downstream of the piers as well.
Contraction in case C has less effect on bed morphology than for
case B (see Figures 8 and 9). The shapes of the local scour holes
are almost identical for the two piers and no strong interaction is
calculated by the numerical model. In contrast to case B, the
scour profile upstream of pier 2 is very similar to that of pier 1.
The same inaccuracies are observed in this case as for the singlepier case namely, an under prediction of the equilibrium scour
on the downstream side of both piers. Nevertheless, the maximum
scour depths and scour shapes are well reproduced on the
upstream sides. Worth noting are the predictions after 1 h of
scour duration, which are in pretty good agreement with the
measurements. This is particularly encouraging and is an imporx: m

Bed change: m

Ridge
Ridge

Trough

20

22

001
003
005
007
009
011

Figure 7. Perspective view of equilibrium geometry (the


transparent grey plane indicates the initial bed level) for case B;
water flows from left bottom to top right

24

26

28

24

26

28

(a)
x: m

Ridge z: m

Bed change: m

Trough

002 18
0
002
004
006
008
010
012
014

002
0
002
004
006
008
010
012
014

18

20

22

Exp. 1 h
Exp. eq.
Calc. 1 h
Calc. 24 h
(b)

045

08

0
08

z: m
0
002
004
006
008
010
012

0
006
04

050

4
00 006
08
0

010

y: m

060
055

10

08

266

065

The complex equilibrium geometry can be characterised by the


local scour around the individual piers, ridges and troughs. The
shapes of the local scour holes around both piers agree well with
the one around the single cylinder. However, due to the interaction between the two scour holes, a ridge is formed upstream of
pier 2. On the other hand, strong erosion causes the development
of a trough along the centreline between the two piers as a result

Figure 8. Measured and calculated half scour, after 1 h and


equilibrium scour profiles for case C along the centreline of
(a) pier 1 and (b) pier 2

bed elevation changes around the piers for case B (the transparent
grey plane indicates the initial bed level). An individual scour hole
similar in shape and depth to the scour hole formed around a
single pier is formed around each pier. The interaction between
the two piers is apparent from the eroded area observed between
pier 1 and pier 2. Despite the complex bed geometry, the location
(at the upstream toe of the piers) and the magnitude of maximum
scour depth (approximately 0.12 m) is not affected by the contraction but seems to be the same as for the single-pier case.

02

010

040
21

22

23

x: m

24

Figure 9. Calculated bed changes for case C

25

26

Water Management
Volume 167 Issue WM5

Case

Number of
cells

A (12 layers)
A (20 layers)
B
C

200 000
330 000
540 000
730 000

A nested grid based computational fluid


dynamics model to predict bridge pier
scour
Baranya, Olsen, Stoesser and Sturm

CPU Hardware
time: h
16
40
32
40

PC 4 cores; CPU speed 3.0 GHz


PC 4 cores; CPU speed 3.0 GHz
IBM p575+ 16 cores; CPU speed 1.9 GHz
IBM p575+ 16 cores; CPU speed 1.9 GHz

Table 2. Computational time of the morphological modelling.

tant finding for short stormwater flow events (e.g. as a result of a


flash flood) for which morphological changes occur over a short
time period and are critical for the stability of a structure.
However, in the case of investigating the effect of flash floods,
temporal changes of discharge significantly influence scour
development. In the introduced model it is possible to define
discharge time series as inlet boundary conditions, which makes
it more capable of studying real problems.
Overall, the numerical morphological model using the nested grid
approach provided reasonably accurate predictions for the singleand multiple-pier cases, particularly with regard to critical aspects
of local scour such as maximum scour depth. The main reason
for development of the nested grid approach is that it offers a
cost-effective method to calculate local flow and morphodynamic
features around structures in larger rivers where the characteristic
width of the river and the size of structures can differ by several
orders of magnitude. In order to reproduce the flow separation
point precisely in the test cases, a high refinement of the nested
grid at the pier walls was necessary; this in turn required a large
number of cells. As a compromise, despite some inaccuracies at
the downstream side of the piers, only 12 layers were used for the
double-pier cases and this resulted in reasonable computational
times, as shown in Table 2. The CPU times correspond to a
modelling time of approximately 30 h with a time step of 10 s
and a number of inner iterations of 20. The runs of case A were
made on a PC with 4 cores and a CPU speed of 3 GHz. The
double-pier cases were simulated on an IBM p575+ computer
with 16 cores and a processor speed of 1.9 GHz.

5.

Conclusions

The 3D numerical model described in this paper is capable of


reliably predicting local scour around one and two cylinders with
reasonably good accuracy, particularly on the upstream side of the
piers. The use of a nested grid with finer resolution around the
cylinders allowed a high spatial resolution around the piers and
enabled a more accurate computation of the flow field near the
cylinder, including the separation point and the recirculation zone.
At the same time, most of the domain was resolved with only a
coarse grid, thereby saving computational resources. Changes in
bed elevation could be computed by the numerical model using
the empirical equation of van Rijn (1984) to estimate the sediment
pick-up rate. It was necessary to reduce the critical shear stress on
the bed as a function of the bed slope and the direction of the

velocity vector close to the bed in the morphodynamic model.


This feature of the model was implemented by applying the
formula of Dey (2001). The computed scour hole geometries
matched the results of the laboratory experiments well, both at
durations of 1 h and at equilibrium; however, decreasing the
vertical grid resolution from 20 layers to 12 layers resulted in an
underestimation of the scour depth in the lee of the piers.
Nevertheless, for engineering purposes, the scour hole geometry
is fairly well predicted, especially the maximum scour depth in
front of the pier, also with the coarser vertical grid resolution.
This is the case for both a single cylinder and two cylinders in a
staggered arrangement. The introduced numerical method is an
appropriate investigation tool for real river engineering problems
because the nested grid description results in significant computational time savings as compared with structured grid solvers.

Acknowledgements
The authors acknowledge funding for Dr Baranya from the
Frittek programme of the Research Council of Norway. Computational time on the Njord cluster was made available for the
current project from Notur. We also thank Mr Ben Ioppolo for his
work on the laboratory model study.
REFERENCES

Amini SA, Mohammad TA, Aziz AA, Ghazali AH and Huat BBK

(2010) A local scour prediction method for pile caps in


complex piers. Proceedings of the Institution of Civil
Engineers Water Management 164(2): 7380.
Baranya S, Olsen NRB, Stoesser T and Sturm TW (2012) Threedimensional RANS modelling of flow around circular piers
using nested grids. Engineering Applications of
Computational Fluid Mechanics 6(4): 684662.
Bomminayuni S and Stoesser T (2011) Turbulence statistics of
open-channel flow over a rough bed. Journal of Hydraulic
Engineering ASCE 137(11): 13471358.
Brooks HN (1963) Boundary shear stress in curved trapezoidal
channels. Journal of the Hydraulic Division ASCE 89(3):
327333.
Catalano P, Wang M, Iaccarino G and Moin P (2003) Numerical
simulation of the flow around a circular cylinder at high
Reynolds numbers. International Journal of Heat and Fluid
Flow 24(4): 463469.
Dargahi B (1989) The turbulent flow field around a circular
cylinder. Experiments in Fluids 8: 112.
267

Water Management
Volume 167 Issue WM5

A nested grid based computational fluid


dynamics model to predict bridge pier
scour
Baranya, Olsen, Stoesser and Sturm

Dargahi B (1990) The controlling mechanism of local scouring.

Olsen NRB and Kjellesvig HM (1998) Three dimensional numerical

Journal of Hydraulic Engineering ASCE 116(10): 11971214.


Dey S (2001) Experimental studies on incipient motion of
sediment particles on generalized sloping fluvial beds.
Journal of Sediment Research 16(3): 391398.
Ghodsian M, Melville B and Coleman S (2012) Local scour due
to sediment-carrying free-overfall water jet. Proceedings of
the Institution of Civil Engineers Water Management
165(1): 2129.
Gobert C, Link O, Manhart M and Zanke U (2010) Discussion of
coherent structures in the flow field around a circular cylinder
with scour hole by G. Kirkil, S. G. Constaninescu and R.
Ettema. Journal of Hydraulic Engineering 119(1): 8284.
Graf WH and Istiarto I (2002) Flow pattern in the scour hole around
a cylinder. Journal of Hydraulic Research 40(1): 1320.
Ikeda S (1982) Incipient motion of sand particles on side slopes.
Journal of the Hydraulics Division ASCE 108(1): 95114.
Kirkil G, Constantinescu G and Ettema R (2008) Coherent
structures in the flow field around a circular cylinder with
scour hole. Journal of Hydraulic Engineering ASCE 134(5):
572587.
Koken M and Constantinescu G (2008) An investigation of the
flow and scour mechanisms around isolated spur dikes in a
shallow open channel: 2. Conditions corresponding to the
final stages of the erosion and deposition process. Water
Resources Research 44(8): 116.
Kovacs A and Parker G (1994) A new vectorial bedload formulation
and its application to the time evolution of straight river
channels. Journal of Fluid Mechanics 267: 153183.
Lane EW (1955) Design of stable channels. Transactions ASCE
120(1): 12341260.
Lee SO and Sturm TW (2009) Effect of sediment size scaling on
physical modeling of bridge pier scour. Journal of Hydraulic
Engineering 135(10): 793802.
Lysne DK (1969) Movement of sand in tunnels. Journal of the
Hydraulics Division ASCE 95(6): 18351846.
Melville BW (1975) Local Scour at Bridge Sites. University of
Auckland, Auckland, New Zealand, Report 117.
Melville BW and Chiew YM (1999) Time scale for local scour at
bridge piers. Journal of Hydraulic Engineering ASCE 125(1):
5965.
Nagata N, Hosoda T, Nakato T and Muramoto Y (2005) Threedimensional numerical model for flow and bed deformation
around river hydraulic structures. Journal of Hydraulic
Engineering ASCE 131(12): 10741087.
Olsen NRB (2003) 3D CFD modeling of a self-forming
meandering channel. Journal of Hydraulic Engineering ASCE
129(5): 366372.
Olsen NRB (2010) A Three-dimensional Numerical Model for
Simulation of Sediment Movements in Water Intakes with
Multiblock Option Users Manual. Norwegian University of
Science and Technology, Trondheim, Norway.
Olsen NRB (2012) Numerical Modelling and Hydraulics. Lecture
Notes. Norwegian University of Science and Technology,
Trondheim.

flow modeling for estimation of maximum local scour depth.


Journal of Hydraulic Research 36(4): 579590.
Olsen NRB and Melaaen MC (1993) Three-dimensional
calculation of scour around cylinders. Journal of Hydraulic
Engineering 119(9): 10481054.
Paik J, Escauriaza C and Sotiropoulos F (2010) Coherent structure
dynamics in turbulent flows past in-stream structures: some
insights gained via numerical simulation. Journal of
Hydraulic Engineering 136(12): 981993.

268

Palau-Salvador G, Stoesser T, Froehlich J, Kappler M and Rodi W

(2010) Large eddy simulations and experiments of flow


around finite-height cylinders. Journal of Flow Turbulence
and Combustion 84(2): 239275.
Rodi W (1980) Turbulence Models and Their Application in
Hydraulics A State of the Art Review. International
Association for Hydraulics Research, Delft, the Netherlands.
Roulund A, Sumer BM, Fredsoe J and Michelsen J (2005)
Numerical and experimental investigation of flow and scour
around a circular pile. Journal of Fluid Mechanics 534: 351
401.
Stoesser T, Palau-Salvador G, Rodi W and Diplas P (2009) Large
eddy simulation of turbulent flow through submerged
vegetation. Transport in Porous Media 78(3): 347365.
Stoesser T, Kim SJ and Diplas P (2010a) Turbulent flow through
idealized emergent vegetation. Journal of Hydraulic
Engineering ASCE 136(12): 10031017.
Stoesser T, Ruether N and Olsen NRB (2010b) Calculation of
primary and secondary flow and boundary shear stresses in a
meandering channel. Advances in Water Resources 33(2):
158170.
Tseng M, Yen CL and Song CCS (2000) Computation of threedimensional flow around square and circular piers.
International Journal for Numerical Methods in Fluids 34(3):
207227.
van Balen W, Uijttewaal WSJ and Blanckaert K (2010) Largeeddy simulation of a curved open-channel flow over
topography. Physics of Fluids 22(7): 075108.
van Rijn LC (1984) Sediment transport, part I: bed load transport.
Journal of Hydraulic Engineering 110(10): 14311457.

WH AT DO YO U T HI NK?

To discuss this paper, please email up to 500 words to the


editor at journals@ice.org.uk. Your contribution will be
forwarded to the author(s) for a reply and, if considered
appropriate by the editorial panel, will be published as a
discussion in a future issue of the journal.
Proceedings journals rely entirely on contributions sent in
by civil engineering professionals, academics and students.
Papers should be 20005000 words long (briefing papers
should be 10002000 words long), with adequate illustrations and references. You can submit your paper online via
www.icevirtuallibrary.com/content/journals, where you
will also find detailed author guidelines.

You might also like