HYDRO PAGE
Hydro Equations and Boundary Conditions
Initial Conditions
Cartesian (x,y) or cylindrical (z,r) coordinates, depending
on the geometry of the rigid body, are introduced to treat the case
of a planar shock interacting with a spherical or cylindrical
obstacle centered in x (or z)=2 and y (or r)=0 and a radius R
normalized to 1.
The rigid body is located in front of the incident shock
and its radius R is a free parameter that selects the
length scale of every single model.
Our units are such that the density and the sound velocity
in the ambient medium are set equal to 1
so that pressure becomes p=1/gamma.
The mass and time scales can be easily deduced from the physical
quantities previously defined.
The incident planar shock
(e.g. in cartesian coordinates), is perpendicular to the x axis
and is moving from (x=0, y=0) (at the initial time tini=0)
towards the positive x direction.
For all cases, we actually compute using different Mach numbers. We
stop the computation after a time tfin=f\tau,
where tau=2*R/us is the crossing time of the rigid body
by the incident shock, and us is the shock velocity. We chose,
for the factor f, f=20,
so that the final time is reached for all computations after
approximately 18 crossing times.
The Mach number of the incident shock is defined as
M=us/ca, where us is the shock velocity. The shocked quantities
are fixed by the Rankine-Hugoniot conditions:
and the y-component of the shock velocity is equal to zero.
Numerical Grid
The calculation around the rigid body is made using a uniform grid.
this approach has several advantages:
a) the mesh generation is straightforward and
easily computed , b) the grid refinement is easy to do and c) the
grid effects arising from non-uniformities in the mesh are eliminated.
The numerical grid is set up as a
rectangular domain with a uniform "active" part, surrounded
by an "expanded" one, in which the mesh points are not equidistant, but
designed to permit the free development of the bow
shock through the upper side of the active grid.
The active part of the grid
extends from x=0 to x=9 and from y=0 to y=3, and selects the
region of the numerical computation with uniform mesh of size Dy=Dx, depending
on the resolution of the problem. Non uniform expanding zones,
from -10 to 0 in x and from 3 to 20 in y, define the expanded region.
Boundary Conditions
The primary difficulty with this kind of grid is, of course, represented
by the boundary conditions.
Flow-in (x=-10) boundary conditions are fixed on
the left side of the x-y grid,in order to make sure that
the shocked values are
incoming. Flow-out (x=9) and (y=20) boundary conditions are
imposed on the right and top sides of the x-y grid, in order to allow
the outflow of the fluid quantities.
An outflow condition is accomplished by maintaining
a "zero gradient" in all fluid variables .
In the next sections, we will take care only of the numerical results
in the active (uniform) part of the grid.
We set reflecting boundary conditions at the bottom
(y=0) of the x-y grid, enforcing an axial symmetry with respect to the
plane containing the x axis and perpendicular to the y axis.
We should note here, that the fact of adopting an axial symmetry, while
loosing -of course- some spatial detail and forcing some geometrical
"unnatural" behaviour in the numerically simulated turbulence, has the advantage
of using a remarkably smaller amount of computing time. Moreover,
as we will show later, the actual real flows are almost axisymmetric, except
along the axis of the motion, and the latter fact does not affect the flow
macroscopically.
In this regard, we note that
the choice of axial symmetry has been used also in numerical
experiments for viscous flows, using a deterministic vortex method
(e.g. Chang and Chern 1991), in order to reproduce the experimental
flow visualization performed by Ta Phouc Loc and Bouard (1985). However,
the time evolution of viscous flow around a rigid body (see also
Bouard and Coutanceau (1980), BG61 and Van Dyke (1988)) shows a strong
symmetry around the main axis. Therefore, the reflecting condition does not
seem to limit the development of the vortex structures behind the object even
for Mach number >= 1.
Boundary Wall on the Surface of the Object
We paid particular attention to the calculations on the boundary wall
between the object and the gas (see figure 1), represented as an impenetrable
series of "stair steps". As the resolution increases, the uniform grid
approaches very well the "natural" boundary of the body. For the cases
of low resolution, a new treatment has been used in order to determine
better approximated fluxes at the zone interfaces.
In order to do this, Like in Edgar and Woodward (1992) we allow
"fractional" steps, producing a smoother boundary for a given computational
mesh. This correction technique
yields a sensible improvement in the comparison with
the experimental data, especially for the case of the sphere.