You are on page 1of 8

SPH Modeling of One-Dimensional Nonrectangular and

Nonprismatic Channel Flows with Open Boundaries

Downloaded from ascelibrary.org by Florida International University on 10/19/13. Copyright ASCE. For personal use only; all rights reserved.

Tsang-Jung Chang 1 and Kao-Hua Chang 2

Abstract: In this study, the authors solve the shallow water equations (SWE) with smoothed particle hydrodynamics (SPH) for onedimensional (1D) nonrectangular and nonprismatic channel flows with open boundaries. To date, 1D SPH-SWE has been only developed
to simulate rectangular prismatic channel flows with closed and open boundaries. However, for practical hydraulic problems, channel cross
sections are not always rectangular and prismatic. A general approach is proposed in this study to extend the engineering application range
of 1D SPH-SWE to nonrectangular and nonprismatic channels with open boundaries by introducing the wetted cross-sectional area and the
water discharge in SWE and combining the method of specified time interval with the inflow/outflow algorithm. Three benchmark study
cases, aiming at testing various steady flow regimes in nonrectangular and nonprismatic channels, are adopted to validate the newly proposed
approach. Through the investigation of the convergence analysis and numerical accuracy test of the study cases, the results show that the
present SPH-SWE approach is capable to model 1D nonrectangular and nonprismatic channel flows with open boundaries. DOI: 10.1061/
(ASCE)HY.1943-7900.0000782. 2013 American Society of Civil Engineers.
CE Database subject headings: Hydrodynamics; Shallow water; Channel flow.
Author keywords: Smoothed particle hydrodynamics; Shallow water equations; Method of specified time interval; Open boundaries;
Nonrectangular and nonprismatic channel.

Introduction
The shallow water equations (SWE) are widely used to mathematically describe a wide variety of free surface flows in rivers,
estuaries, and coasts (Chaudhry 1993; Cunge et al. 1980). Many
Eulerian mesh-based methods are currently available to solve
SWE in hydraulic engineering applications. Recently, a pure
Lagrangian meshless method, smoothed particle hydrodynamics
(SPH) (Monaghan 2005; Gomez-Gesteira et al. 2010), is widely
applied to the numerical formulation of SWE (Wang and Shen
1999; Ata and Soulaimani 2005; Rodriguez-Paz and Bonet 2005;
De Leffe et al. 2010; Vacondio et al. 2012a; Chang et al. 2011; Kao
and Chang 2012). Compared with Eulerian mesh-based methods,
SPH has the following advantages: (1) no convective term (nonlinear term), which often causes numerical oscillations, exists in the
governing equations (Liu and Liu 2003); (2) the wave propagation
of free surface can be intrinsically tracked (De Leffe et al. 2010);
and (3) the wet-dry interface can be automatically described without any special treatment (Vacondio et al. 2012a). Hence, SPHSWE has recently been receiving more and more attention. Some
significant achievements are summarized. Wang and Shen (1999)
firstly investigated inviscid dam-break flows using SPH-SWE. Ata
and Soulaimani (2005) derived a new artificial viscosity for the
dam-break problem with wet bed. Rodriguez-Paz and Bonet (2005)
1
Professor, Dept. of Bioenvironmental Systems Engineering, National
Taiwan Univ., Taipei 10617 Taiwan. E-mail: tjchang@ntu.edu.tw
2
Postdoctoral Researcher, Dept. of Bioenvironmental Systems
Engineering, National Taiwan Univ., Taipei 10617 Taiwan (corresponding
author). E-mail: f94622026@ntu.edu.tw
Note. This manuscript was submitted on October 1, 2012; approved on
May 23, 2013; published online on May 25, 2013. Discussion period open
until April 1, 2014; separate discussions must be submitted for individual
papers. This paper is part of the Journal of Hydraulic Engineering,
Vol. 139, No. 11, November 1, 2013. ASCE, ISSN 0733-9429/2013/
11-1142-1149/$25.00.

proposed a variational SPH-SWE formulation to maintain the mass


and momentum conservation. De Leffe et al. (2010) adopted an
anisotropic kernel with variable smoothing length and performed
SPH-SWE modeling of shallow-water coastal flows. Vacondio et al.
(2012a) presented the modified virtual boundary particle method to
deal with various complex shapes of solid boundaries. Chang et al.
(2011) and Kao and Chang (2012) extended SPH-SWE to investigate shallow-water dam-break flows in realistic open channels.
Additionally, to improve the SPH-SWE solutions, Vacondio et al.
(2012) developed a particle splitting procedure to enhance the resolution in small-depth problems and studied dam break cases with
analytic and experimental results. Vacondio et al. (2013) applied
two corrections for balancing discontinuous bed slopes including
conservative and nonconservative approaches to handle bottom discontinuity problems.
The aforementioned references of SPH-SWE are all related to
closed boundary conditions. The implementation of SPH-SWE in
open boundary conditions remains difficult because of the Lagrangian nature of SPH. Vacondio et al. (2012b) pioneeringly introduced the characteristic boundary method into SPH-SWE to
simulate rectangular prismatic channel flows with open boundaries.
They adopted the Riemann invariants to determine proper boundary
conditions to overcome the well-posed problems (Anderson 1995).
Their work enabled 1D SPH-SWE to deal with open boundaries in
rectangular prismatic channels. However, for natural channels, the
shape of channel cross sections is not always rectangular and prismatic. The Riemann invariants cannot be formulated in the cases
of nonrectangular channels except triangular channels (Sanders
2001). It is also not suitable to simulate nonprismatic channel flows
with the water depth and the water velocity in SWE (Vacondio
et al. 2012b). To extend the engineering application range of 1D
SPH-SWE, the authors adopt the wetted cross-sectional area and
the water discharge in SWE to reflect irregular channel shape
in nonprismatic channels. The authors also solve the characteristic equations and to establish the open boundary conditions in

1142 / JOURNAL OF HYDRAULIC ENGINEERING ASCE / NOVEMBER 2013

J. Hydraul. Eng. 2013.139:1142-1149.

Downloaded from ascelibrary.org by Florida International University on 10/19/13. Copyright ASCE. For personal use only; all rights reserved.

nonrectangular channels with the method of specified time intervals


(Chaudhry 1993; Sturm 2010).
This paper is organized as follows: the core of SPH and
SPH-SWE formulation are briefly introduced. Then the newly
proposed approach for open boundaries is presented. Finally, three
benchmark study cases, representing various steady flows in
nonrectangular and nonprismatic channels, are adopted to validate
the newly proposed approach. In these three study cases, three
combinations of inflow/outflow boundary conditions and channel
cross sections illustrate the ability of the newly proposed approach
on various channel cross sections and flow conditions.

SPH-SWE Methodology
SPH Formulation
SPH is a pure interpolation method. Any physical quantity of
particle i (i ) is approximated by the weighted summation as
i

X j
mj W i jxi xj j; hi
Aj
j

where mj ( Aj V j Aj x0 ) is the mass of particle j; V j ( x0


at the initial state) is the volume of particle j; Aj is the wetted crosssection area of particle j; x0 is the initial particle spacing; xi is the
position of particle i; rij ( jxi xj j) is the distance between particle i and particle j; W i jxi xj j; hi ( W i ) is the kernel function
of particle I; and the cubic spline function (Liu and Liu 2003;
Chang et al. 2012b) is chosen as the kernel function, and hi is
the smoothing length of particle i ( 1.3x0 ).
Two SPH formulas of the operation of the gradient or divergence used in common are listed in the following (Liu and Liu
2003). Eq. (2) is symmetric with respect to i and j to be utilized
to perform the divergence operation. The gradient operation is performed by Eq. (3), which is asymmetric with respect to i and j:
i

1X
m i W i jxi xj j; hi
Ai j j j

i Ai

X  i j 
mj 2 2 W i jxi xj j; hi
Ai Aj
j

section and A dw (B mdw ) for a trapezoidal cross section];


B is the bottom width, m is the side slope; S0 is the bed slope;
Sf is the friction slope ( n2 Q2 =A2 R4=3 ); n is the Manning roughness coefficient; R is the hydraulic radius; and g is the gravitational
acceleration.
SPH-SWE Implementation
Evaluation of Wetted Cross-Sectional Area
Commonly, two numerical approaches can solve the wetted crosssectional area in SPH-SWE. One is solving the continuity equation
of Eq. (4), and another is using the weighted summation formula as
shown in Eq. (1). The present model chooses the latter approach to
calculate the wetted cross-section area of each particle i. A variable
smoothing length scheme is applied to obtain better accuracy of
SWE solutions. Thus, the smoothing length of particle i is connected to the wetted cross-section area (Ata and Soulaimani 2005;
Rodriguez-Paz and Bonet 2005; De Leffe et al. 2010; Vacondio
et al. 2012a) with
 1=D
m
A
hi h0;i 0;i
6
Ai
where A0;i and h0;i are the initial wetted cross-sectional area and
smoothing length for particle i, respectively; and Dm is the number
of space dimensions (Dm 1 in this study).
Because of using the variable smoothing length scheme, the
Newton-Raphson iterations are thus performed to solve the wetted
cross-sectional area of particle i. The iterative procedure (Ata and
Soulaimani 2005; Rodriguez-Paz and Bonet 2005; De Leffe et al.
2010; Vacondio et al. 2012a) is as follows:


Reski Dm
k1
k
Ai Ai 1
7
Reski Dm ki
with

Reski Aki

mj W i jxi xj j; hki

where Reski is the residual of particle i at the kth iteration; and ki is


defined as
X
dW i
ki mj rij
drij
j

Shallow Water Equations


The traditional SPH-SWE solves the water depth and the water
velocity to provide flow simulations, which is only suitable for
rectangular and prismatic channel flows. The wetted cross-sectional
area and the water discharge are introduced in SPH-SWE herein to
extend it to nonrectangular and nonprismatic channel flows. The
Lagrangian form of SWE in terms of the wetted cross section
and the water discharge can be written in Eq. (4) (the continuity
equation) and Eq. (5) (the momentum equation):
DA
Q=A
A
Dt
x

DQ
Q=A
d
Q
gA w gAS0 Sf
x
Dt
x

where D=Dt is the total time derivative [D=D =t u=2x;


Q is the water discharge; u is the water velocity ( Q=A); A is the
wetted cross-sectional area; dw is the water depth and it is generally
a function of A [specifically, A Bdw for a rectangular cross

The Newton-Raphson iterative procedure will be stopped as


Resk1
i
1010
Aki

10

Discretized Momentum Equation. The discretized SPH form


of the momentum equation [Eq. (5)] is thus given in the following
to evaluate the rate of water discharge of each particle i:


Qj Qi
DQi
Qi X

W i
Dt
Ai j j Aj Ai
X  di dj 
gA2i
mj 2 2 W i gAi S0;i Sf;i 11
Ai Aj
j
where W i 1=2W i jxi xj j; hi W j jxi xj j; hj .
Because the smoothing length is variable, the expression of the
gradient of a kernel function (W i ) is a hybrid combination of

JOURNAL OF HYDRAULIC ENGINEERING ASCE / NOVEMBER 2013 / 1143

J. Hydraul. Eng. 2013.139:1142-1149.

Downloaded from ascelibrary.org by Florida International University on 10/19/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 1. Sketch of the computational domain

the scatter andthe gather interpretations herein (Hernquist and


Katz 1989).
Fig. 1 shows the computational domain. In this study, the computational domain is divided into three zones, i.e., the inflow zone
(between the inlet boundary and the inflow boundary), the fluid
zone (between the inflow boundary and the outflow boundary),
and the outflow zone (between the outflow boundary and the outlet
boundary). In addition, four types of particles are used, including
inflow particles, inner particles, outflow particles, and virtual bed
particles (i.e., rectangular points, circle points, diamond points, and
triangular points of Fig. 1). The former three types are, respectively,
within the flow zone, the fluid zone, and the outflow zone. The last
type is arranged in the computational domain.
As to the bed gradient term S0;i in the right-hand side of
Eq. (11), a virtual bed particle that has one specified value of bed
slope is introduced (Vacondio et al. 2012a). The volume of a virtual
bed particle equals to x0 . Then the bed slope of the inflow/inner/
outflow particle i is computed by the weighted summation of the
bed slope of each virtual bed particle surrounding the inflow/inner/
outflow particle i as shown in Eq. (12):
X
vb ~
vb
vb
S0;i
V vb
12
j S0;j W j jxi xj j; hj
j

where the Shepard filter scheme is used (Randles and Libersky


~ means the corrected kernel
1996; Chang et al. 2012a), W
function as
vb
W j jxi xvb
j j; hj

vb
~ j jxi xvb
W
j j; hj P
j

vb
vb
V vb
j W j jxi xj j; hj

13

The superscript vb indicates the virtual bed particle and


hvb
j 1.3x0 .
Additionally, for the aim of enabling the numerical method
stable, an artificial viscosity is introduced (Ata and Soulaimani
2005) as
Y
ij

c ij uij xij
A ij q
x2ij 2


t 0.25 min


15

Open Boundary Treatment


Because the Riemann invariants can only be formulated for rectangular and triangular channels (Sanders 2001), the use of the
Riemann invariants at the inflow/outflow boundaries to seek open
boundary conditions (Vacondio et al. 2012b) has limited applications. As a result, a more general approach, which adopts the wetted cross-sectional area and the water discharge in SWE and
combines the method of specified time interval with the inflow/
outflow algorithm, is proposed to model 1D nonrectangular and
nonprismatic channels with open boundaries. The supercritical
flow problems require two variables (the wetted cross-sectional
area and the water discharge) at the inflow boundaries while the
subcritical flow problems need only one variable (the wetted
cross-sectional area or the water discharge) at both inflow and outflow boundaries. Hence, the method of specified time interval is
used to solve the characteristic equations and obtain the remaining
variable at the boundaries. The time step is specified according to
the CFL condition.
Characteristic Equations
The characteristic equations (Chaudhry 1993; Cunge et al. 1980)
are classified into the positive and negative forms based on the directions of characteristic lines. Eqs. (16) and (17) are the positive
characteristic equations and Eqs. (18) and (19) are the negative
characteristic equations:
C

Ddw c Du

cS0 Sf
Dt
g Dt

16

dx
uc
dt

17

along
C

14

p
where c is the celerity ( gH d ); Hd is the hydraulic depth;
Aij Ai Aj =2; c ij ci cj =2; uij ui uj ; xij xi xj ;
and is a constant ( 106 ).
Time-Marching Scheme. To update particle positions and
velocities in time, the leap-frog time-marching scheme (RodriguezPaz and Bonet 2005; Vacondio et al. 2012a) is herein used. Because
the SPH is an explicit method, the time step (t) has to satisfy the
CFL condition as

x0
p
ui gH d;i

and
C

Ddw c Du

cS0 Sf
Dt
g Dt

18

along

1144 / JOURNAL OF HYDRAULIC ENGINEERING ASCE / NOVEMBER 2013

J. Hydraul. Eng. 2013.139:1142-1149.

dx
uc
dt

19

Downloaded from ascelibrary.org by Florida International University on 10/19/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 2. Sketch of the method of specified time interval

Method of Specified Time Interval


The method of specified time interval (Chaudhry 1993; Sturm
2010) is often used to evaluate the boundary conditions for explicit
numerical methods. Fig. 2 shows the sketch of the method of specified time interval. The characteristic equations of Eqs. (16) through
(19) can be discretized by the finite difference approximations of
the derivatives to obtain the discretized characteristic equations,
i.e., Eqs. (20) through (23). In this method, Eqs. (20) through (23)
are solved to seek the unknown values of the variables at the inflow/
outflow boundaries:
C uS uR

g
d dw;R gS0;R Sf;R t
cR w;S
x xR
C S
uR c R
t

C uP uL

xP xL
uL c L
t

22

23

where the subscripts P and S represent at the inflow and outflow


boundaries, respectively, and the subscripts L and R both represent
inside the fluid zone.
As the flow is subcritical at the inflow/outflow boundaries, the
values of the water velocity and the water depth are prescribed
at the inflow and outflow boundaries, respectively. Hence, Eq. (20)
is solved together with Eq. (21) to provide the value of the
water depth at the inflow boundary. Similarly, the value of the water
velocity at the outflow boundary is given by solving Eqs. (22)
through (23). The detailed procedure of solving the discretized
characteristic equations is presented as follows.
Point S at the outflow boundary is on the trajectory of the positive characteristic line RS as shown as Fig. 2, the water depth and
the water velocity at point S can be evaluated by solving Eq. (20)
only if all of the variables at point R are known. To obtain the values
of all the variables at point R, the computational domain is
separated into ghost cells whose length equals to the initial particle
spacing (x0 ) (Fig. 2). The linear interpolation of the water velocity between point C and point D is performed and that combines
Eq. (21) to derive Eq. (24):
uD uR xS xR uR cR t

uD uC
x0
x0

25

dw;D dw;R uR cR t

dw;D dw;C
x0

26

To solve Eqs. (24) and (25) together, the water velocity uR and
the celerity cR can be found, and then the water depth dw;R can also
be determined by Eq. (26).
In the same way, for point P of Fig. 2 on the trajectory of the
negative characteristic line LP, the linear interpolations between
point A and point B are performed and the discretized negative
characteristic equations of Eq. (22) and (23) are utilized to give
the water depth and the water velocity at point P.
Inflow/Outflow Boundary Conditions
Four kinds of boundary conditions can be specified according to
the local flow conditions (the Froude number) at the inflow/outflow
boundaries:
1. Subcritical outflow condition: As the subcritical flow occurs at
the outflow boundary (Line RS of the Fig. 2), the water depth
is prescribed, and the water discharge is determined by using
Eq. (28), which is derived from combining Eq. (20) with the
continuity equation of Eq. (27):
QS AS uS

21

g
d dw;L gS0;L Sf;L t
cL w;P
C

20

cD cR uR cR t

x0
cD cC

24

In a similar interpolation, the linear relationships of the celerity


and the water depth between point C and point D are presented in
Eqs. (24) and (25):

g
QS AS uR dw;S dw;R gS0;R Sf;R t
cR

27

28

2. Subcritical inflow condition: When the subcritical flow occurs


at the inflow boundary (Line LP of the Fig. 2), the water discharge is prescribed, and the water depth is calculated through
solving Eq. (30) using the Newton-Raphson iterations.
Eq. (30) is derived from combining Eq. (22) with the continuity equation of Eq. (29):
Q P A P uP

g
fdw;P AP uL dw;P dw;L
cL

gS0;L Sf;L t 0

29

30

3. Supercritical outflow condition: If the supercritical flow occurs at the outflow boundary, there is not necessary to specify
the boundary condition there. Hence, the water depth and the
water discharge are extrapolated from the fluid zone.
4. Supercritical inflow condition: Asthe supercritical flow occurs
at the inflow boundary, the water depth and the water
discharge are prescribed.
Inflow/Outflow Algorithms
The inflow/outflow algorithm developed by Federico et al. (2012)
is applied to solve open boundary problems in this study. The detailed procedure includes the following steps: (1) When an inflow
particle (rectangular points of Fig. 1) moves across the inflow
boundary, it will become an inner particle (circle points of Fig. 1).
The momentum equation then controls the behavior of an inner
particle. At the same time, the new inflow particle will be produced
in the inflow zone and the inflow boundary conditions such as
the water discharge (Qp ) and the water depth (dw;p ) at the inflow

JOURNAL OF HYDRAULIC ENGINEERING ASCE / NOVEMBER 2013 / 1145

J. Hydraul. Eng. 2013.139:1142-1149.

boundary are assigned to the inflow particle. (2) An inner particle is


recognized as an outflow particle (diamond points of Fig. 1) as it
moves across the outflow boundary. In addition, the outflow boundary conditions such as the water discharge (Qs ) and the water
depth (dw;s ) at the outflow boundary are assigned to the outflow
particle. An outflow particle will be deleted as it crosses the outlet
boundary. Therefore, the total number of particles within the
computation process will remain a constant as flows reach the
steady state.

Downloaded from ascelibrary.org by Florida International University on 10/19/13. Copyright ASCE. For personal use only; all rights reserved.

Results and Discussion


In this section, three benchmark study cases (Macdonald 1995;
Macdonald et al. 1997; Khalifa 1980) validate the newly proposed
approach. Three combinations of inflow/outflow open boundary
conditions and channel cross sections illustrate the ability of the
proposed approach on various channel shapes and flow conditions.
To investigate the convergence analysis and numerical accuracy
test, the L2 relative error norm based on the variable , as shown
in Eq. (31), is introduced:
v
uP
u simulated analytic=measured 2
i
i
L2 t
31
P analytic=measured 2

i
and analytic=measured
are the simulated physical
where simulated
i
i
quantity and the analytic or measured data at the ith particle,
respectively.
Rectangular Prismatic Channel
The purpose of the first study case is to investigate the performance of the proposed approach on rectangular prismatic channel
flows given by MacDonald (1995). The channel is 100-m long
and 10-m wide and the Manning roughness coefficient of the
channel is 0.03. The bed slope is nonuniform and the bed elevation profile of the channel is plotted in Fig. 3(a). This study case
has the analytic solution. The flow is supercritical at the inflow
boundary and at the outflow boundary, with the given water discharge 20 m3 s1 and the water depth 0.7506 mat the inflow
boundary.
Convergence Analysis
In this study case, four initial particle numbers of 100, 200, 500,
1,000 in the computational domain (i.e., the initial particle spacings
(x0 ) are 1, 0.5, 0.2, and 0.1 m, respectively) are considered to
conduct the convergence analysis. Through Eq. (31), the values
of L2 dw (L2 relative error norm based on the water depth) and
L2 Q (L2 relative error norm based on the water discharge) are
summarized in Table 1. Both L2 dw and L2 Q decrease as the
initial particle number increases. Thus, the proposed approach
can converge to the analytic solution. In addition, the convergence
rates are 1.19 and 1.5, respectively, for the water depth and the
water discharge. A convergence criterion is defined as the difference in L2 dw or L2 Q between two initial particle numbers less
than 0.005. The initial particle number of 500 is adequate herein
and the simulated results of using 500 particles are given in this
study case.
Numerical Accuracy
Fig. 3(b) compares the numerical accuracy of the simulated result
and the analytic solution of the water depth profile in the channel.
Fig. 3(c) demonstrates the spatial variation of the simulated Froude
number of the channel and Fig. 3(d) further shows the simulated
and analytic water discharges of the channel. From Figs. 3(a and c),

Fig. 3. (a) Bed elevation profile of the rectangular prismatic channel;


(b) profiles of the simulated and analytic water depths; (c) spatial variation of the simulated Froude number of the channel; (d) simulated and
analytic water discharges of the channel

Table 1. L2 dw and L2 Q of Various Initial Particle Numbers (N) in the


Rectangular Prismatic Channel
N (x0 )
(1)
(2)
(3)
(4)

100 (1.0 m)
200 (0.5 m)
500 (0.2 m)
1000 (0.1 m)

L2 dw

Difference (dw )

L2 Q

Difference (Q)

0.026
0.020
0.013
0.011

1 2 0.006
2 3 0.007
3 4 0.002

0.0040
0.0036
0.0035
0.0034

1 2 0.0004
2 3 0.0001
3 4 0.0001

the flow is supercritical at the inflow boundary and changes to


subcritical since the channel bed slope reduced, which results in a
hydraulic jump occurring at x 100=3 m as shown as Fig. 3(b).
After the hydraulic jump, the flow returns to supercritical due to the
effect of sharp variation in the channel bed slope.
For the numerical accuracy test, as shown in Fig. 3(b), the
simulated and analytic water depth profiles of the entire channel
are well consistent. In Table 1, L2 dw is only 0.013. Furthermore,
Fig. 3(d) shows the simulated and analytic water discharges of the
channel. L2 Q is about 0.0035 (Table 1). An oscillation near the
hydraulic jump is detected in Fig. 3(d). This is the main source of

1146 / JOURNAL OF HYDRAULIC ENGINEERING ASCE / NOVEMBER 2013

J. Hydraul. Eng. 2013.139:1142-1149.

L2 Q in this case. Such a discontinuity in water depths near the


hydraulic jump results in disordered particles that lead to the water
discharge oscillation. Overall, the simulated results of water discharges are accurately predicted.

Downloaded from ascelibrary.org by Florida International University on 10/19/13. Copyright ASCE. For personal use only; all rights reserved.

Trapezoidal Prismatic Channel


To test the performance of the newly proposed approach on nonrectangular channel flows, the trapezoidal prismatic channel problem given by MacDonald et al. (1997) was selected as the second
study case. This study case aims to demonstrate that as dealing with
inflow/outflow boundary conditions, the proposed approach can
improve the limitation of the Riemann invariants that can only
be formulated for rectangular and triangular channels in the characteristic boundary method given by Vacondio et al. (2012b). The
nonuniform bed-slope channel is 1,000-m
p long and its width and
perimeter are 10 2dw m and 10 2 2dw m, respectively. The
Manning roughness coefficient of the channel is 0.02. The flows
at the inflow and outflow boundaries are both subcritical, and the
water discharge 20 m3 s1 and the depth 1.35 m are specified as the
inflow and outflow boundary conditions, respectively.
Convergence Analysis
In the convergence analysis, four initial particle numbers of 100,
200, 500, 1,000 in the computational domain [i.e., the initial particle
spacings (x0 ) are 10, 5, 2, and 1 m, respectively] are adopted
in this case. Table 2 shows the values of L2 dw and L2 Q for the
four initial particle numbers. The simulated results approach to the
analytic solution as the initial particle number increases. The convergence rates for the water depth and the water discharge are 0.78
and 1.77, respectively. The differences in L2 dw and L2 Q are
both less than 0.005 between 500 particles and 1,000 particles.
Thus, the simulated results of using 500 particles are given in this
study case.
Numerical Accuracy
Figs. 4(a and c) depict the bed elevation and simulated Froude number profiles of the channel, respectively. The bed slope is nonuniform and the flow changes from subcritical to supercritical, and
back to subcritical. A hydraulic jump occurs at the location of
x 600 m. The simulated and analytic water depth profiles of
the channel are both shown in Fig. 4(b). The agreement between
the simulated and analytic results is quite satisfactory. For the
numerical accuracy test, L2 dw is only 0.009 (Table 2). In addition, as shown in Fig. 4(d), the simulated water discharges are compared with the analytic solution. In Table 2, L2 Q is 0.0035. The
main source of L2 Q is also from an oscillation occurs near the
hydraulic jump in Fig. 4(d). To sum up, the proposed approach is
significantly capable of modeling nonrectangular channel flows.
Rectangular Nonprismatic Channel
The third study case is to investigate the performance of the proposed approach on nonprismatic channel flows. The experimental
model of flows in a rectangular divergent channel conducted by

Table 2. L2 dw and L2 Q of Various Initial Particle Numbers (N) in the


Trapezoidal Prismatic Channel
N (x0 )
(1) 100 (10 m)
(2) 200 (5 m)
(3) 500 (2 m)
(4)1000 (1 m)

L2 dw

Difference (dw )

L2 Q

Difference (Q)

0.020
0.014
0.009
0.007

1 2 0.006
2 3 0.005
3 4 0.002

0.0059
0.0041
0.0035
0.0034

1 2 0.0018
2 3 0.0006
3 4 0.0001

Fig. 4. (a) Bed elevation profile of the trapezoidal prismatic channel;


(b) profiles of the simulated and analytic water depths; (c) spatial variation of the simulated Froude number of the channel; (d) simulated and
analytic water discharges of the channel

Khalifa (1980) is adopted. The channel is horizontal, 2.5-m long


and its width is variable as shown as Fig. 5. The flow is supercritical
at the inflow boundary, and it returns to subcritical at the outflow
boundary because of the channel expansion. As to the inflow/
outflow boundary conditions, the discharge 0.0263 m3 s1 and
the depth 0.088 m are given as the inflow boundary conditions
and the depth 0.195 m is specified as the outflow boundary
condition. The Manning roughness coefficient is set as 0.015.
Convergence Analysis
In this case, three initial particle numbers, including 25
(x0 0.1 m), 50 (x0 0.05 m), and 100 (x0 0.025 m)
particles, are given in the computational domain to perform the
convergence analysis. The values of L2 dw and L2 Q for the three
initial particle numbers in the rectangular nonprismatic channel are
reported in Table 3. No obvious difference in L2 dw exists as the
initial particle number is more than 50. However, L2 Q decreases
as the initial particle number increases. The proposed approach obviously can be convergent in the rectangular nonprismatic channel.
Based on the differences in L2 dw and L2 Q of Table 3, the
simulated results of the initial particle number of 50 are appropriate
to describe the flow phenomena.

JOURNAL OF HYDRAULIC ENGINEERING ASCE / NOVEMBER 2013 / 1147

J. Hydraul. Eng. 2013.139:1142-1149.

jump. The result shows that the proposed approach can deal with
nonprismatic channel flows as well.

Downloaded from ascelibrary.org by Florida International University on 10/19/13. Copyright ASCE. For personal use only; all rights reserved.

Conclusions

Fig. 5. Channel width profile of the entire rectangular nonprismatic


channel

Table 3. L2 dw and L2 Q of Various Initial Particle Numbers (N) in the


Rectangular Nonprismatic Channel
N (x0 )
(1) 25 (0.1 m)
(2) 50 (0.05 m)
(3) 100 (0.025 m)

L2 dw Difference (dw ) L2 Q
0.073
0.068
0.068

Difference (dw )

1 2 0.005 0.0385 1 2 0.0186


2 3 0.000 0.0199 2 3 0.0040

0.0159

Numerical Accuracy
The channel width profile of the entire rectangular nonprismatic
channel is described in Fig. 5. As shown in Fig. 5, the channel
expansion starts at x 0.65 m, resulting in a transition from supercritical flow to subcritical flow. Thus, a hydraulic jump occurs
around x 1 m. Fig. 6(a) presents the comparison between the
simulated and measured water depth profiles. In Table 3, L2 dw
is about 0.068. The simulated and measured water depths are quite
consistent. Fig. 6(b) shows the comparison between the simulated
and measured water discharges. L2 Q is 0.0199 (Table 3). Again,
an oscillation in the simulated water discharges near the hydraulic
jump is found in Fig. 6(b). The simulated water discharges are influenced by the rapid variation of water depth in the hydraulic

Fig. 6. (a) Comparison between the simulated and measured water


depth profiles; (b) comparison between the simulated and measured
water discharges of the nonprismatic channel

The authors propose a new SPH-SWE approach that combines the


method of specified time interval with the inflow/outflow algorithm
to model 1 D nonrectangular and nonprismatic channels with open
boundaries. Unlike the traditional 1 D SPH-SWE approach, the
new approach adopts the wetted cross-section area and the water
discharge in SWE to simulate flows in prismatic and nonprismatic
channels. The authors introduce the method of specified time interval to extend the SPH-SWE application to 1 D nonrectangular
channel flows. Three study cases, aiming at various steady flow
regimes in nonrectangular and nonprismatic channels, are adopted
to validate the newly proposed approach. The convergence analysis
is performed to study the appropriate initial particle number in each
case, and the numerical accuracy test is next conducted. The simulated results show good agreement with the analytic and measured
results. Although retaining the water discharges is difficult for locations near the water depth discontinuity, the simulated results still
give satisfactory results of water discharges. Thus, the present SPHSWE approach has been proved its capability in modeling 1 D nonrectangular and nonprismatic channel flows with open boundaries.

References
Anderson, J. D. (1995). Computational fluid dynamics: The basics with
applications, McGraw-Hill, New York.
Ata, R., and Soulaimani, A. (2005). A stabilized SPH method for inviscid
shallow water flows. Int. J. Numer. Meth. Fluid., 47(2), 139159.
Cunge, J. A., Holly, F. M., and Verwey, A. (1980). Practical aspects of
computational river hydraulics, Pitman Advanced Publishing Program,
London.
Chang, K. H., Kao, H. M., and Chang, T. J. (2012b). Lagrangian modeling
of particle concentration distribution in indoor environment with different kernel functions and particle search algorithms. Build. Environ.,
57, 8187.
Chang, T. J., Chang, K. H., Kao, H. M., and Chang, Y. S. (2012a).
Comparison of a new kernel method and a sampling volume method
for estimating indoor particulate matter concentration with Lagrangian
modeling. Build. Environ., 54, 2028.
Chang, T. J., Kao, H. M., Chang, K. H., and Hsu, M. H. (2011). Numerical
simulation of shallow-water dam break flows in open channels using
smoothed particle hydrodynamics. J. Hydrol., 408(12), 7890.
Chaudhry, M. H. (1993). Open-channel flow, Prentice-Hall, New Jersey.
De Leffe, M., Le Touze, D., and Alessandrini, B. (2010). SPH modeling of
shallow-water coastal flows. J. Hydraul. Res., 48(supp 1), 118125.
Federico, I., Marrone, S., Colagrossi, A., Aristodemo, F., and Antuono, M.
(2012). Simulating 2D open-channel flows through an SPH model.
Eur. J. Mech. B Fluids, 34, 3546.
Gomez-Gesteira, M., Rogers, B. D., Dalrymple, R. A., and Crespo, A. J. C.
(2010). State-of-the-art of classical SPH for free-surface flows.
J. Hydraul. Res., 48(supp 1), 627.
Hernquist, L., and Katz, N. (1989). TREESPH: A unification of SPH with
the hierarchical tree method. Astrophys. J. Suppl. Ser., 70, 419446.
Khalifa, A. M. (1980). Theoretical and experimental study of the radial
hydraulic jump. Ph.D. thesis, Univ. of Windsor, Windsor, ON.
Kao, H. M., and Chang, T. J. (2012). Numerical modeling of dambreakinduced flood and inundation using smoothed particle hydrodynamics.
J. Hydrol., 448449, 232244.
Liu, G. R., and Liu, M. B. (2003). Smoothed particle hydrodynamics:
A meshfree particle method, World Scientific, Singapore.
Macdonald, I. (1995). Test problems with analytic solutions for steady
open channel flow. Numerical Analysis Rep. 6/94, Univ. of Reading,
Dept. of Mathematics, Reading, UK.

1148 / JOURNAL OF HYDRAULIC ENGINEERING ASCE / NOVEMBER 2013

J. Hydraul. Eng. 2013.139:1142-1149.

Downloaded from ascelibrary.org by Florida International University on 10/19/13. Copyright ASCE. For personal use only; all rights reserved.

Macdonald, I., Baines, M. J., Nichols, N. K., and Samuels, P. K. (1997).


Analytic benchmark solutions for open channel flows. J. Hydraul.
Eng., 123(11), 10411045.
Monaghan, J. J. (2005). Smoothed particle hydrodynamics. Rep. Prog.
Phys., 68(8), 17031759.
Randles, P. W., and Libersky, L. D. (1996). Smoothed particle hydrodynamics: Some recent improvements and applications. Comput.
Method. Appl. M., 139(14), 375408.
Rodriguez-Paz, M., and Bonet, J. (2005). A corrected smooth particle
hydrodynamics formulation of the shallow-water equations. Comput.
Struct., 83(1718), 13961410.
Sanders, B. F. (2001). High-resolution and non-oscillatory solution of
the St. Venant equations in non-rectangular and non-prismatic
channels. J. Hydraul. Res., 39(3), 321330.
Sturm, T. W. (2010). Open channel hydraulics, McGraw-Hill, New York.

Vacondio, R., Rogers, B. D., and Stansby, P. K. (2012). Accurate particle


splitting for smoothed particle hydrodynamics in shallow water with
shock capturing. Int. J. Numer. Methods Fluids, 69(8), 13771410.
Vacondio, R., Rogers, B. D., and Stansby, P. K. (2012a). Smoothed
particle hydrodynamics: approximate zero-consistent 2-D boundary
conditions and still shallow-water tests. Int. J. Numer. Methods Fluids,
69(1), 226253.
Vacondio, R., Rogers, B. D., Stansby, P. K., and Mignosa, P. (2012b). SPH
modeling of shallow flow with open boundaries for practical flood
simulation. J. Hydraul. Eng., 138(6), 530541.
Vacondio, R., Rogers, B. D., Stansby, P. K., and Mignosa, P. (2013). A
correction for balancing discontinuous bed slopes in two-dimensional
smoothed particle hydrodynamics shallow water modeling. Int. J.
Numer. Methods Fluids, 71(7), 850872.
Wang, Z., and Shen, H. T. (1999). Lagrangian simulation of onedimensional dam-break flow. J. Hydraul. Eng., 125(11), 12171220.

JOURNAL OF HYDRAULIC ENGINEERING ASCE / NOVEMBER 2013 / 1149

J. Hydraul. Eng. 2013.139:1142-1149.

You might also like