Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
FSI-99-TN48R
THE MODELI NG OF SHALLOW FLOWS
C.W. Hirt and J.E. Richardson
Flow Science, Inc.
INTRODUCTION
While methods exist for modeling complex three-dimensional flow problems,
there are some situations in which full three-dimensional simulations are not efficient
enough to be practical. One class of such problems are those flows characterized as
shallow. A shallow flow is one in which the horizontal extent of the flow is much larger
than the vertical extent. Examples include the ocean, large estuaries, seasonal floods,
liquid coatings, lubricating films, and water on automobile windshields.
If a flow is shallow enough that vertical accelerations can be neglected, then it can
be shown [1] that a good approximation to the flow is to replace all flow variables by
their averages in the vertical direction. The resulting, depth-averaged, three-dimensional
equations of motion then become a two-dimensional representation in the horizontal
plane. The most interesting cases involve a top fluid surface that is free to exhibit wave
phenomena. In some instances non-uniform horizontal boundaries (e.g., a sloping
beach) may also induce small deviations from purely horizontal flow. In this sense the
depth-averaged approximation still includes some three-dimensional effects.
Depth-averaged approximations are probably best known and most widely used in
applications relating to oceans or other large, shallow bodies of water. In this context the
approximation is most often referred to as the shallow-water approximation. There are
excellent discussions of the shallow-water equations and higher order refinements of
those equations in references [2,3].
Our purpose in this paper is to explore a new approach for numerically modeling
shallow flows. The method we propose is based on techniques used in the commercial
fluid dynamics simulation package FLOW-3D
methods. Then we show how these techniques are interpreted when applied in the limit
2
of the shallow-water approximation. This is followed by a discussion of the advantages
offered by these techniques for representing shallow flows over uneven surfaces,
including wetting and drying phenomena.
The paper concludes with a summary of how the depth-averaged model can be
used in the FLOW-3D
) 1 ( 2
1
y y
f u g
y
p
y
v
v
x
v
u
t
v
+ +
0
)] ( [ )] ( [ ) (
y
h h v
x
h h u
t
h h
B B B
4
The two-dimensional momentum equations used in FLOW-3D
are identical to
the shallow water equations if the fluid pressure is defined correctly. In FLOW-3D
the
pressure in an element containing a free surface is defined as p= p
0
+g(h-z/2), where p
o
is the gas pressure on the surface of the liquid and h is the height of the surface above the
bottom of the grid. Except for a constant, this is equivalent to the hydrostatic pressure
used in the shallow water approximation, which means the momentum equations in the
two approaches are equivalent. In practice, for the shallow water option, the pressure is
replaced by p=p
0
+gH, where H is the sum of the fluid and obstacle heights.
The volume/area blockages used in the FAVOR
consider the solid blockage to lie at the bottom of the control volumes. This
is an easy thing to do.
The fluid height in terms of the VOF representation at the centers of control
volumes is h- h
B
=FV
f
z. At the edges of elements, where (A
x
,A
y
) are the open area
fractions at the x and y faces, the fluid height is h- h
B
=FA
f
z. However, since the bottom
contour must vary smoothly to be consistent with the neglect of vertical accelerations,
there are no differences between volume and area fractions, i.e., V
f
=A
f
.
Substituting the fluid height expressions into the shallow water continuity
equation, and using the appropriate area fractions for the x and y directions, we arrive at
This is precisely the equation for F used in the VOF method for flow in a single,
horizontal layer of control volumes provided there is no transport of F in the z direction, a
restriction that is necessary in the shallow-flow model. This completes the demonstration
of the equivalence between the VOF and FAVOR
y
F vA
x
F uA
t
F V
y
x
f
5
The alternative option is to compute the new values F
n+1
using an implicit
formulation of Eqs. 1-2. In this scheme the values of velocity used in Eq. 2 are
considered functions of F
n+1
through the pressure definition p=gF
n+1
. An implicit
equation for the new values of pressure (i.e., F
n+1
) is formed from a combination of Eq. 1-
2 that can be solved by a simple iteration process.
The explicit method is very simple and requires no iteration process. However,
the penalty for this is that there is a limitation on the size of the time step to prevent
gravity waves from moving more than one grid cell in one time step:
Where t is the time step, x is the width of a grid cell, g is the acceleration of gravity
and h is the depth of fluid.
To eliminate this time-step restriction, it is necessary to use the implicit
formulation, which couples the depth of fluid to the velocities causing changes in the
depth. In many cases the increase in time-step size that can be realized by the implicit
approach more than compensates for the extra work required to solve the implicit
equations.
ADVANTAGES OF VOLUME FRACTION METHODS FOR SHALLOW FLOWS
The advantages of using volume fraction techniques to represent bottom contours
and free surfaces are many. One advantage is associated with the fact that fluid moves
through a fixed grid. This mean it is easy to have fluid move into dry regions or to
withdraw from previously wet regions. No re-meshing is necessary to account for
movement of the lateral boundaries of the fluid.
Defining bottom contours with area/volume fractions, i.e., with the FAVOR
program has many additional advantages associated with the numerous user conveniences
and auxiliary physical models already in the commercial program. For example, area and
volume fractions used in the FAVOR
is accomplished by setting
the parameter ISWAT=1 for the explicit option or ISWAT=2 for the implicit option
(Note: the implicit option is only available in versions after 7.6). Definition of the
bottom contour is made by using obstacles to fill the space below the contour.
0
2
,
) 1 ( 4 h
h
where
7
Figure 1. Wave run-up sequence on beach (left) and withdrawal (right). Only a
small portion of computation near the wetting line is shown.
Initial fluid configurations, Coriolis forces, and bottom roughness (i.e., obstacle
roughness) are defined in the same way as for three-dimensional simulations (consult the
manual). A wind shear on the free surface can be imposed in terms of the top wall
tangential velocity components (UBC(6) and VBC(6)) or their time-dependent
counterparts UBCT(n,6) and VBCT(n,6)). A nonzero value for the wind shear
coefficient (WNDSHR) is needed to activate this model and set the strength of the
applied shear stress.
If it is expected that relatively strong hydraulic jumps may develop, the second
order momentum advection option IORDER=3 should be selected because this includes a
correction to momentum advection that gives the proper jump conditions.
SAMPLE APPLICATIONS
In this section we illustrate the shallow flow model by applying it to several simple
problems. The examples selected show a variety of capabilities that could be extended in
many directions with only a small amount of imagination.
A. Wave Run-up and Fall-back on a Sloping Beach
Periodic waves running up on a sloping beach offer a good example of the wetting
and drying capabilities of the combined VOF-FAVOR
2 . A
x
H
g
x
u
u
t
u
10
How can it be that Eq.A2 comes directly from Eq.A1 yet does not give the correct
hydraulic jump condition for momentum? The answer to this question involves several
issues. First, it must be recognized that the depth-averaged equations involve the
assumption that vertical accelerations can be neglected, which cannot be correct at a
hydraulic jump. Nevertheless, it should be possible to satisfy integrated jump conditions
between two uniform states.
As mentioned in the main text, the difference between the conservation of
momentum and conservation of energy jump conditions is second order in jump height.
Could it be that this higher order effect is missed because the depth-averaging
procedure has already omitted some comparable higher order terms? To answer this
question we propose writing difference approximations to Eqs.A1 and then manipulating
them in the same way that lead to Eq.A2. The goal of this exercise is to see how the
difference form of Eq.A2 must be written to maintain the momentum conservation
property contained in Eqs.A1.
In our case the height H is located at the center of a grid cell, while the velocity u
is located at the sides of the cell. An upstream or donor cell scheme is used for the
advection terms to insure numerical stability. These assumptions lead to the difference
equations (using constant cell sizes for simplicity):
Notice that the first term in the second equation has been broken into two terms
that represent the time derivative of u and H respectively. The first equation can be used
to eliminate the H time derivative term to yield:
Now we can divide through by the average H. On the right-hand side we have
taken the liberty to cancel H factors in the numerator and denominator.
( )
n
i
n
i
n
i n
i
n
i
n
i
n
i n
i
n
i
n
i
n
i
n
i
n
i
n
i n
i
n
i
n
i
n
i
n
i
n
i n
i
n
i
n
i n
i
n
i
n
i
n
i
n
i
x
H
Hg x
u u
u H
u u
u H
u
H H H H
u u
H H
t
x
u u
H
u u
H
H H H H
t
x
2 / 1
2 / 1 2 / 1
2 / 1
2 / 1 2 / 3
2 / 1 1
2 / 1
1
1 1
1
2 / 1
1
2 / 1
1 1
1
2 / 1 2 / 1 2 / 1 2 / 3
1
1
1 1
1
2 2
2 2 2
0
2 2 2 2
+
+
+ +
+ +
+
+
+ +
+
+
+
+
+ +
+
+ + +
+
+
+ +
+
,
_
,
_
,
_
+
+
1
1
]
1
,
_
,
_
,
_
+
+
,
_
,
_
,
_
,
_
+
+
1
1
]
1
,
_
,
_
,
_
( ) ( )
n
i
n
i
n
i
n
i
n
i n
i
n
i
n
i
n
i
n
i
x
H
Hg x u u
u u
H u u
H H
t
x
2 / 1
2 / 1 2 / 1
2 / 1 2 / 1
2 / 1
1
2 / 1
1 1
1
2 2
+
+
+
+
+
+
+ +
+
,
_
,
_
+
+
,
_
,
_
( ) ( ) ( )
n
i
n
i
n
i
n
i
n
i
n
i
n
i
n
i
n
i n
i
n
i
H H g u u
u u
H H
H
u u
t
x
,
_
+
+
+
,
_
+ +
+
+ +
+
+
+
+ 1 2 / 1 2 / 1
2 / 1 2 / 1
1 1
1
2 / 1
1
2 / 1
2
2
11
This is the desired result; the equation is the usual non-conservative equation except for
the H factor in the advection term. To put this into practice we must approximate the
H
n+1
terms in the denominator by H
n
values since we cant compute the new H values
until the new velocities have been computed. Thus, to approximate the conservation of
momentum jump condition when using the non-conservative momentum equation we
must multiply the advection term by the factor,
This result is for the case that u is positive. If u is directed negative the factor is,
It should be observed that this factor remains close to unity as long as the fluid
height does not change significantly from one grid cell to the next. It should also be
observed that when making the transition to a continuum equation by taking the limit of
vanishing cell size, this factor reduces to unity and Eq.A2 is recovered.
The inclusion of the H factor adds a higher-order effect necessary for a better
approximation to the momentum jump condition. Physically, the factor may be thought
of as performing a momentum averaging. That is, the velocity change is equal to the
momentum entering the control volume divided by the total mass in the control volume.
With this simple change added to the FLOW-3D
+
+
+ +
+
.
2 2
1
1
1 1
1
1
n
i
n
i
n
i
n
i
n
i
n
i
H H
H
H H
H
+
+
+
+
+ +
+
+
0.0 20.0 40.0 60.0 80.0 100.0
x
0.0
12.0
z
Figure A1. Comparison of results from non-conservative formulation and
modified formulation for a jump generated by a ratio 10 discontinuity.
Modified results have smaller jump height and higher speed.
12
APPENDIX B: SURFACE TENSION AND WALL ADHESION
There are many potential applications for a shallow water model in liquid coating.
For this class of applications, however, it is essential to have surface tension forces
including adhesion to solid surfaces. The usual surface tension model in FLOW-3D
does not apply in the shallow limit because adhesion forces must be nearly horizontal and
cannot be included in an equivalent pressure formulation. To remedy this a new model
for surface tension has been especially developed for the shallow-water limit. This model
has been incorporated into a routine for thermocapillary effects, which allows for force
contributions in any direction.
Surface tension forces may be temperature dependent and adhesion to the
substrate does account for the slope of the substrate. No special input is required to
activate this model other than the usual values for surface tension, contact angle and, if
applicable, the coefficient of the temperature dependence.
The new model is a direct extension of the thermocapillary force computation
done in the program routine tfsten. For thermocapillary effects only force components
associated with a changing surface tension coefficient are included in routine tfsten,
while all the remaining forces are represented by an equivalent normal pressure and are
computed in another routine.
For shallow-water applications we have modified tfsten to compute all the forces.
It is convenient, however, to still use an equivalent pressure for the basic surface tension
forces that includes thermocapillary effects. This is done by first computing the surface
tension forces acting at the four sides of a computational grid cell. Then, the sum of
vertical forces divided by the horizontal area of the cell is defined to be the surface-
tension pressure.
Figure A2. Comparison of results computed for a jump using conservative (solid)
and non-conservative (dashed) formulations. Conservative results have smaller
jump height and higher jump speed.
13
Wall adhesion effects are added to any grid cell having fluid and with one or more
empty neighbor cells. If the substrate is sloped then vertical components of adhesion are
included in the equivalent pressure calculation. All horizontal adhesion forces are
retained as separate horizontal forces in the force arrays used for thermocapillary effects.
Several example problems have been done to test this surface tension addition.
These included a cylindrical liquid rib with a 10 contact angle and a spherical cap also
with a 10 contact angle. In both cases the liquid remained stationary indicating that the
surface tension and adhesion forces are consistent with the initial configurations. When
the contact angles were increased somewhat the fluid in both cases contracted toward the
new angle.
A more dynamic test example is the evolution of a spherical cap running down an
inclined slope. The cap is a portion of a spherical water drop of radius 1.0 cm having a
thickness of 0.25 cm. The contact angle is 60 and the plate is inclined at an angle of 45
to the vertical. Fig. B1 shows the liquid drop and wet trail after 0.2 s. A characteristic
steepening of the drop surface at the front and gentle sloping at the back are easily seen.
The wet trail is a consequence of the no-slip condition used on the plate surface, which is
also a characteristic of real drops. No actual experimental data was available with which
to make detailed comparisons.
Finally, we have computed an example of permanent waves on a sheet of water
flowing down a vertical wall. This problem was originally studied experimentally by
Kapitza and Kapitza (Collected Papers of P.L. Kapitza, ed. D.Ter Haar, Vol.2, 690
Pergamon 1965). Surface tension is an important component affecting the dynamics of
the waves. The example modeled is for an alcohol sheet (=0.79, =22.9, =0.02) of
average thickness 0.0197cm. Periodic boundary conditions were used in the direction of
flow with a grid length equal to the observed disturbance wavelength of 1.77cm. Also, a
roughness was added to the plate in order to adjust the mean flow rate to the proper
theoretical value. The roughness accounts for the fact that the wall shear stress is not
correct because the depth-averaged model does not account for a parabolic velocity
profile. A roughness value of 0.0008cm was found to give the desired flow rate.
The computed surface profile assumes a solitary wave type of behavior having a
crest with a long sloping tail and a steeper leading edge that moved into a few small
amplitude precursor capillary waves, Fig. B2. These features are in excellent agreement
with the experimental observations. What is not as good is the amplitude of the wave,
x
z
x
z
Figure B1. Drop sliding down a 45 tilted surface, leaving a wet trail.
14
which is computed to be higher than what is observed. For example, the ratio of the peak
height to the height of the flat portion was computed to be 3.46. This is higher than the
values of 1.84 to 2.4 obtained from computed solutions based on full Navier-Stokes
formulations (B. Ramaswamy, S. Chippada and S.W. Joo, J. Fluid Mech. 325, 163
(1996)).
0.0 0.354 0.708 1.062 1.416 1.770
x
0.0
0.07
Figure B2. Solitary type wave moving down (to the right) a vertical wall.