Vous êtes sur la page 1sur 10

16

Int. J. Engineering Systems Modelling and Simulation, Vol. 3, Nos. 1/2, 2011

Hybrid upwind splitting scheme by combining the


approaches of Roe and AUFS for compressible flow
problems
Ghislain Tchuen*
Institut Universitaire de Technologie Fotso Victor,
Universit de Dschang,
LISIE, BP. 134 Bandjoun, Cameroun
Fax: +237-33-01-46-01
E-mail: tchuengse@yahoo.com
*Corresponding author

Ferdinand Fogang
Facult de Sciences,
Universit de Yaound I,
LaMSEBP, BP. 8210, Yaound, Cameroun
E-mail: ferdifogang@yahoo.fr

Yves Burtschell
DME, Technopole de Chateau Gombert,
Universit de Provence,
5 rue Enrico Fermi, 13453 Marseille Cedex 13, France
E-mail: yves.burtschell@polytech.univ-mrs.fr

Paul Woafo
Facult de Sciences,
Universit de Yaound I,
LaMSEBP, BP. 8210, Yaound, Cameroun
E-mail: pwoafo1@yahoo.fr
Abstract: This paper presents a new robust finite volume shock-capturing scheme based on an
improved version of the artificially upstream flux vector splitting (AUFS) scheme. This novel
numerical scheme, named AUFSR, is obtained by combining AUFS scheme and the Roe solver.
The resulting flux functions can be implemented in a very simple manner, in the form of the Roe
solver with modified wave speeds, so that converting an existing AUFS flux into the new fluxes
is an extremely simple task. The method handles difficulties encountered by the AUFS scheme,
when the artificial diffusion is computed with the Steger-Warming approach in the case where
the flux vector does not check the homogenous property. The solution accuracy is further
improved by extension to high-order spatial and temporal scheme. Accuracy, efficiency and other
essential features of the proposed scheme are evaluated by analysing shock propagation
behaviours for both the steady and unsteady compressible flows.
Keywords: hybrid method; numerical instability; shock capturing; artificially upstream flux
vector splitting; AUFS; Roe; AUFSR.
Reference to this paper should be made as follows: Tchuen, G., Fogang, F., Burtschell, Y. and
Woafo, P. (2011) Hybrid upwind splitting scheme by combining the approaches of Roe and
AUFS for compressible flow problems, Int. J. Engineering Systems Modelling and Simulation,
Vol. 3, Nos. 1/2, pp.1625.
Biographical notes: Ghislain Tchuen is a Senior Lecturer at the University of Dschang,
Cameroon. After graduating in Energetic at University of Yaounde I, Cameroon, he received
his PhD in Mechanic-Energetic at University of Provence, France, in 2003. His research interests
are in computational fluid dynamics, aerodynamics, re-entry problems, scramjet, plasma flow,
non-equilibrium flows, radiative flux, combustion phenomena, magneto-hydrodynamic flows,
Riemanns solvers, renewable energy and traditional oven. He has a significant contribution in
the development of the research computer code named CARBUR since 2002.

Copyright 2011 Inderscience Enterprises Ltd.

Hybrid upwind splitting scheme by combining the approaches of Roe and AUFS for compressible flow problems

17

Ferdinand Fogang received his Masters in Fluid Mechanics at University of Yaounde I,


Cameroon in 2008. He started his PhD thesis the same year on numerical simulation of
magneto-hydrodynamic flows.
Yves Burtschell received his PhD thesis in 1990 in Marseille in compressible fluid mechanics.
He obtained his HDR in 2003 and became a University Professor in 2006. He conducts most of
his work research on the study of unsteady gas flows at high speeds. Writing a research computer
code named CARBUR since 1997, he has studied numerically the chemical relaxation and
combustion phenomena, molecule vibrational excitation, ionisation, viscous and turbulence
effects and electrical discharges with the help of eight PhD students. The areas of application
(with experimental comparisons) are hypersonic wind tunnels, re-entry problems, micro-fluidic,
scramjet and fluid-structure interaction. He published approximately 70 archival journals or
meeting papers. Moreover and curiously, in 2006, he created the PolytechMarseille Civil
Engineering Department, where he is currently the Director.
Paul Woafo is a Professor of Physics at the University of Yaounde I, Cameroon. He is a holder of
a Doctorat dEtat obtained in 1997 in the field of mechanics (non-linear dynamics). He is the
Director of the Laboratory of Modelling and Simulation in Engineering and Biological Physics
with strong interests in applied non-linear dynamics, electromechanical devices, control of
vibrations, dynamics of semiconductor lasers, chaos cryptography, biological physics and
biomechanics, and appropriate technologies for development (http://www.lamsebp.org). He is
presently a Co-author of more than 100 papers published in peer-reviewed journals and member
of various scientific organisations at the national and international levels.
This paper is a revised and expanded version of a paper entitled A new hybridised scheme for
improved shock-capturing in compressible flow problems presented at the 45th Symposium of
Applied Aerodynamics PolytechMarseille, 2224 March 2010.

Introduction

In fluid dynamics problems, compressible flows normally


involve complex flow phenomena, such as strong shock
waves, shock-shock interactions and shear layers. In a
progressive effort to solve complex flow phenomena
described by the Euler/Navier-Stokes equations, the
research of maximising both accuracy and efficiency has
been the primary goal for designing an algorithm in
numerical analysis. A number of numerical flux functions
for inviscid fluxes have been devised to solve an
approximate Riemann problem. Among these formulations,
upwind numerical methods have become popular for
solving hyperbolic partial differential equations with
discontinuous solutions. These are usually classified as
either flux difference splitting (FDS) (Godunov, 1959; Roe,
1981; Woodward and Collela, 1984; Harten et al., 1987;
Shu, 1997) or flux vector splitting (FVS) (Sun and
Takayama, 2003; Steger and Warning, 1981; Van Leer,
1977), each method having its advantages and demerits. In
recent years, some blended schemes (Coquel and Liou,
1995; Liou and Steffen, 1993) have appeared and they are
an attempt at combining the two main families in such a
way as to produce a scheme that keeps only the good
properties.
Roes (1981), known as the FDS scheme, is widely used
for solutions due to its quality and mathematical clarity, and
its efficiency for providing solution accuracy with less
computational effort. The Roes approach has the desirable
property of accurately resolving shock waves as well
as contact discontinuities. However, the scheme may
sometimes lead to unphysical flow solutions in certain

problems; admit rarefaction shocks that do not satisfy the


entropy condition, these flaws can be easily handled by
simple entropy fix correction (Harten et al., 1983) in
one-dimensional case. Quirk (1994) also pointed out that to
improve the solution accuracy; the original Roes scheme
should be modified or replaced by other schemes in the
vicinity of a strong shock. The combination of the Roe
solver with another scheme should be explored.
The AUFS method proposed by Sun and
Takayama (2003) for splitting the flux vectors of the Euler
equations is a special FVS scheme. This scheme has
recently been extended (Tchuen et al., 2008) to calculate
two-dimensional hypersonic viscous flow field in
thermochemical non-equilibrium. The AUFS scheme
introduces two-artificial wave speeds into the flux
decomposition. The direction of wave propagation is
adjusted by two-wave speeds. One part of the flux is
numerically obtained by the Steger-Warming approach
which is a FVS method. Steger and Warming (1981) were
the first to use the homogenous property of the governing
equations of gas, and expressed the inviscid flux vectors in
terms of their Jacobian matrices. This method is simple,
accurate and robust for the Euler equations but it is not
applied to non-homogeneous hyperbolic equations such as
e.g., the ideal MHD equations.
The main objective of this paper is to propose a new
scheme (named AUFSR) constructed on the principle of the
hybrid method. The current approach is motivated by the
desire to combine the efficiency of FVS method and the
accuracy of FDS approach. The new numerical flux splitting
scheme is obtained by hybridising of the AUFS flux and
the Roe flux. The idea and procedure is to avoid the

18

G. Tchuen et al.

homogenous condition imposed to the FVS, before the


using of Steger-Warming approach. The present scheme
should be able to solve flows with non-homogeneous
property. The Roes approach with the averaged Jacobian
matrix appears to be simpler and is used to linearise the
Steger-Warming matrix. One critical task in this procedure
is the construction of a Roe matrix. Furthermore, this type
of hybrid method can be easily extended to higher-order of
accuracy through the formal use of MUSCL interpolations.
The present scheme for multidimensional flows introduces
some amount of numerical dissipation to shear waves, as the
Roe splitting does. When the artificial diffusion can be
devised by the Roes formulation, the AUFSR scheme is
not only robust for shock-capturing, but also accurate for
resolving shear layers. This is demonstrated by an extensive
series of numerical experiments with standard finite volume
Euler codes, including various shock instability problems.
This AUFSR solver has high computational efficiency when
compared to AUFS, Roe and exact Riemann solvers.

Depending upon the value and sign of s1 one gets:


F1 =

1 L
P + P R + U and F2 = U u s2 + P
2

(7)

where
L for s1 > 0,
R for s1 0

(8)

The Jacobian matrices are A1,2 = Fi,2/U = A sI, where s


is an artificially introduced wave speed. An excellent merit
of the AUFS scheme is that the eigenvalues can be changed
by varying the scalar value s. The performance of the
scheme should rely on the way to express U which
represents the artificial viscosity. In the AUFS scheme, this
viscosity is determined by the Steger-Warming formula.
Substituting (6) and (7) to (5), the following final intercell
flux is obtained:

F = (1 M ) PL + PR + U + M U u s2 + P (9)

Numerical technique
2.2 The bases of the Roe scheme

Consider the Euler equations in one dimension:

U F
+
=0
t x

(1)

where U and F are conservative vectors quantities and


fluxes respectively, U = (,u, E)T and F = (u, u2 + p,
(E + p)u)T. is the density, u is the velocity component in
the x direction, p is the static pressure. E is the specific total
energy defined as, E = (e + u2/2), and the ideal-gas
equation of state is assumed, p = ( l)e. is the ratio of
specific heats. The flux vector can be rewritten as:
F = uU + P

(2)

where P = (0, p, pu)T. Equation (1) can be integrated on a


cell-wise in 1D space domain, then the discretised
governing equation yields in conservation form:

U in +1 = U in Fi n+1/2 Fi n1/2

n N,

iZ

(3)

where = t/x, t and x is the ratio of time step t to


grid size x. Fi+1/2 is the numerical flux vector defined by
two neighbouring cells or left and right cells:

Fi n+1/2 = F U in , U in+1 = F U L , U R

(4)

2.1 The fundamental idea of the AUFS scheme


The fundamental idea of AUFS scheme is to split flux
vector (2) as follows:
F = (1 M ) ( u s1 )U + P + M ( u s2 )U + P

(5)

where s1 and s2 are two-scalar constants:


M = s1 ( s1 s2 ) and F = (1 M ) F1 + MF2

(6)

The Roes approximate Riemann solver is one of the most


successful approaches for capturing shock waves with high
solution accuracy. The Roes method stands out by its
simplicity and clarity of the underlying physical model. The
numerical flux vector (4) at the cell interface between the
left cell L and the right cell R according to Roes scheme is:
Fi n+1/2 =

1 n
Fi + Fi n+1 kki +1/2 | ki +1/2 | Rki +1/2

(10)

where ki +1/2 is the wave strength of the kth wave, ik+1/ 2 is


the eigenvalue, and Rki +1/2 the corresponding right
eigenvector. The Roes original scheme may sometimes
produce unphysical flow solutions in certain problems.
These admit rarefaction shocks that do not satisfy the
entropy condition. To avoid such unphysical solutions, the
entropy fix corrections (Harten et al., 1983; Lin, 1995;
Pandolfi and dAmbrosio, 2001) have been proposed.
The Hartens entropy fix method is designed to correct
unphysical expansion shock. The one-dimensional entropy
fix was developed by replacing the characteristic speeds of
the acoustic waves (for k = 1 and 4) with:
k
if
k > ,

2
k = 1 k 2 +
2 otherwise,

(11)

= *(| un | + c), where * is a constant between 0 and 0.25,


un is the contravariant velocity normal to the cell interface
and c is the average sound speed.

2.3 The AUFSR scheme


When the non-linear flux vector F(U) is a homogeneous
function of degree one in U, the flux vector F(U) can be

Hybrid upwind splitting scheme by combining the approaches of Roe and AUFS for compressible flow problems
split into subvectors, each one of which is associated with a
tailored set of eigenvalues. The Steger-Warming splitting
approach is not applicable for the investigation of first-order
conservative systems that are non-homogeneous. The basic
idea used here is to linearise the Steger-Warming matrix. In
the present scheme, the artificial viscosity U is determined
using averages of Roe eigenvalues:
1
U =
2

k Rk

The final AUFSR intercell flux F

1
1
F AUFSR = (1 M ) P L + P R
2
2

k Rk

k =1

(13)

1
1
1

R1 = u c ; R2 = u ; R3 = u + c
1 2
H uc
H + uc

2u

(14)

and the characteristics variables satisfy the relations:


1

[ p cu ] ,
2c 2
2 = p / c 2 ,

(15)

[ p + cu ]

H=

L + R

L H L + R H R
L + R

JGG

( F n)
k =1

=0

;;

(22)

The limiter minimum-modulus (MINMOD) function is


used. The method is spatial second-order upwind, and a
time predictor-corrector algorithm is used to obtain
second-order time accuracy.

3
(16)

(21)

In high-order, the scheme can be extended following


Van Leers (1977) MUSCL approach which verified TVD
properties. The fluxes at the interfaces are extrapolated
between neighbouring cells. The values of fluxes are related
to the conserved variables through a high-order generalised
extrapolation approach:

= L R ;
L uL + R uR

U
1
+
t i , j

F U i +1/2, j = F U i , j , U i +1, j

The symbol r in (14), (15) denotes a Roe average for a


variable r. The relevant averages are given as follows (Roe,
1981):

u=

(20)

The numerical scheme employs the finite volume method


for any grid system. After integrating, over an arbitrary and
discrete volume, the system becomes a system of ordinary
differential equation and takes the form:

The eigenvectors are defined by:

2c 2

(19)

u
v

2


uv
u
u + p

; F=
, G=
U=
uv
v2 + p
v


u ( E + p)
E

v( E + p )

U F G
+
+
=0
t x y

is given by:

+ M U u s2 + P

3 =

The spatial and temporal extension of the numerical method


to second and higher-order accuracy provide a way of
superlinearly improving the accuracy of the solution as cell
sizes and time steps are reduced. The time-dependent Euler
equations (1) can be extended in two-dimensions as follows:

with
AUFSR

1 =

2.4 Higher-order extension

(12)

k =1

19

Numerical results and discussion

The performance of the AUFSR scheme is illustrated on the


ID and 2D Euler equations for ideal gas with = 1.4. All 2D
simulations are performed with a CFL number equal to 0.6.

1 2

c = ( 1) H u 2
2

3.1 Results for 1D problem tests

In order to determine completely the numerical fluxes, we


need to provide the values of artificial velocity. Wave speed
s1 is simply set to their Roe average.
s1 = u

(17)

Numerical values of s2 can be computed from:


min ( 0, u L cL , u c ) for s1 > 0,
s2 =
max ( 0, uR cR , u c ) for s1 0

(18)

For 1D, four-tests were solved, which are generally tested


by Toro (1999). The tests are selected to investigate the
performance of the AUFSR solver. In all chosen tests, data
consists of two constant states (, u, p)L and (, u, p)R,
separated by a discontinuity at a position x = x0. The states
are given in Table 1. The exact and numerical solutions are
found in the spatial domain 0 x 1 using 100 cells. The
Courant number coefficient is taken as 0.9. For each test
problem, an initial location of discontinuity, x0, and the
output time are selected; these are stated in the caption of
each figure. Boundary conditions are transmissive. For these

20

G. Tchuen et al.

1D problem tests, the numerical results are compared


with those of five other schemes (exact solution, AUSM,
VanLeer, Roe and AUFS schemes).
Table 1

Data for four tests problems

Test

uL

pL

1.0

0.75

1.0

1.0

2.0

0.4

1.0

0.0

1,000.0

5.99924

19.5975

460.894

uR

pR

0.125

0.0

0.1

1.0

2.0

0.4

1.0

0.0

0.01

5.99242

6.19633

46.0950

Figure 1

Test 1 is the so called Sod test problem (1978); the solution


consists of a right shock wave, a right travelling contact
wave and a left expansion wave with a sonic point inside.
This test is devised to assess the entropy satisfaction
property of numerical methods. Figure 1 shows solution
profiles for pressure and velocity at time t = 0.2 units. The
new hybrid scheme (AUFSR) resolves the sonic rarefaction
point better than the AUFS and Roe schemes. It resolves the
left travelling waves and the contact wave more smoothly
than the others schemes, because it implicitly introduces
some amount of important non-linear viscosity by slightly
over-estimating the artificial wave speed s2. The results
from the Van Leer (1977) scheme are virtually identical to
those of AUFSR method. The AUFSR scheme captures the
right travelling shock sharper than the other schemes.
Figure 2

Solutions of test 2 with X0 = 0.5

Solutions of test 1 with X0 = 0.3

Note: Exact (solid line) and numerical solutions are


compared at the output time 0.15 U.
Note: Exact (solid line) and numerical solutions are
compared at the output time 0.2 U.

Test 2, called the 123 problem, has solution consisting of


two-strong expansion waves and a trivial contact wave of

Hybrid upwind splitting scheme by combining the approaches of Roe and AUFS for compressible flow problems
zero speed; the star region between the non-linear waves is
close to vacuum, which makes this problem a suitable test
for assessing the performance of numerical methods for
low-density flows. The results of this test are shown in
Figure 2. The Roe scheme fails on this test. Density results
from all schemes agree similarly well with the exact
solutions. The AUFSR scheme is less dissipative than the
AUFS scheme. The AUFSR scheme gives more accurate
results than other from approximate Riemann schemes; in
the vicinity of the trivial contact, where density is close to
zero, the results are somewhat erratic, see internal energy
plots.
Figure 3

21

et al., 1991), but the AUFSR scheme is also more accurate


than the results from the Roe scheme. The numerical
solution has an incorrect dip behind the shock wave, which
is more clearly seen in the density plot.
Figure 4

Solutions of test 4 with X0 = 0.4

Solutions of test 3 with X0 = 0.5

Note: Exact (solid line) and numerical solutions are


compared at the output time 0.035 U.

Note: Exact (solid line) and numerical solutions are


compared at the output time 0.012 U.

Test 3 is designed to assess the robustness and accuracy of


numerical methods; its solution consists of a strong running
shock wave of shock Mach number 198, a contact surface
and left rarefaction waves. Figure 3 shows the results for
six schemes. The results from AUFS and AUFSR are
virtually identical to those of Godunov scheme (Einfeldt

Test 4, as Test 3, is also designed to assess the robustness of


numerical methods, its solution consists of three strong
discontinuities travelling to the right. It is an interaction of
two strong impinging shock waves. The left shock wave
moves to the right very slowly, which adds another
difficulty to numerical methods. The results are overall
comparable with those of AUFSR method shown in
Figure 4. The only visible difference is seen near the left
slowly moving shock. The new hybrid scheme resolves the
right-travelling shock slightly sharper than the AUFS and
Roe scheme. AUFSR and AUFS give a wider left-travelling
shock. These two-schemes introduce more artificial
viscosity into the wave moving opposite to the flow
direction.

22

G. Tchuen et al.

Solutions for one-dimensional problems can be used to


check programs in two or three-dimensional. In order to
verify the performance of the AUFSR scheme, a few 2D
problems are solved to illustrate the accuracy and robustness
of this new scheme. Numerical results are compared with
approximate Riemann solvers.

3.2 Quirks test (odd-even grid perturbation


problem)
This is a well-known test first proposed and investigated
by Quirk (1994) to explore the carbuncle phenomenon and
has been later studied by several authors (Pandolfi and
dAmbrosio, 2001; Edwards, 1997; Liou, 2000). It is well
known that many upwind schemes, including the exact
Riemann solver and the approximate ones are afflicted with
the shock instabilities (also called odd-even decoupling). A
single shock wave travels downstream in a straight duct at
MS = 6. The problem is computed on a grid of 800 21
two-dimensional duct with unit spacing. The centreline grid
points are modified by adding (subtracting) a small
perturbation (103) to the odd (even) streamwise stations.
The initial position of the shock wave is at x = 20 as shown
in Figure 5. Figure 5 shows the predicted density contours at
two locations along the duct by the first-order scheme of
Roe, exact Riemann, AUFS and AUFSR respectively. All
figures show the captured shock waves at x = 900. The Roe
flux and the exact Riemann scheme failed to preserve the
initial shock, resulting a carbuncle. They suffer from
numerical instabilities. The AUFSR and AUFS scheme
show that decoupling was completely eliminated, and
successfully preserved the initial shock. It can be seen that
the proposed AUFSR scheme remains to be shock stable for
sufficiently long time.
Figure 5

Mach 6 moving shock along odd-even grid


perturbation, (a) Roe (b) exact Riemann (c) AUFS
(d) AUFSR

3.3 M = 8 flow over a circular cylinder


This is yet another test case for which some Riemann solver
based on FDS scheme suffers from the catastrophic
carbuncle. The free stream Mach number was taken to be
M = 8 flow over a circular cylinder. The computational
grid is 160 80 structured grid. The boundary conditions
used are similar to those used in Pandolfi and dAmbrosio,
(2001). The first-order scheme was used with the AUFSR,
AUFS fluxes and the Roe flux. As expected, the Roe flux
created a distorted solution as shown in Figure 6(a). On the
other hand, the AUFS fluxes do not produced in Figure 6(b)
such kind of shock instability. The AUFSR flux did not
allow the carbuncle to appear and produced a correct
solution [see Figure 6(c)].
Figure 6

(a)
Figure 7

M = 8 flow over a circular cylinder: pressure contours,


(a) Roe (b) AUFS (c) AUFSR

(b)

(c)

M = 20 flow over a circular cylinder: pressure


contours, (a) Roe (b) Riemann exact (c) AUFSR

(a)

(b)

(c)

(a)

(b)

(c)

3.4 M = 20 hypersonic flow over a circular cylinder

(d)

The AUFSR flux is applied here to a more severe case:


M = 20 hypersonic flows over a cylinder. This problem has
the same boundary conditions and the same computational

Hybrid upwind splitting scheme by combining the approaches of Roe and AUFS for compressible flow problems
grid as in the case M = 8. Figure 7 shows the computed
pressure contours and shock profiles when using the
first-order of the Roe, the exact Riemann and the AUFSR
scheme respectively. It is evident that the Roe flux created a
carbuncle phenomenon. It is also observed, to be somewhat
unexpected, that the Exact Riemann scheme created,
although weaker, the oscillation around the stagnation
region. On the other hand, the numerical result of AUFSR
scheme does not exhibit such kind of shock instability and
produced a correct solution with a clean flow field behind
the bow shock.

3.5 shock wedge reflection from wedge


Shock wave phenomena considered here are those arising
from the reflection of a shock wave from a wedge placed
at an angle to the incident shock wave direction. At the
initial time, a shock wave of Mach number Ms, perfectly
perpendicular to the x-direction is placed at a position xs.
The planar shock wave travels in the x-direction and
encounters a wedge that makes an angle with the shock
direction. For boundary conditions, non-slip conditions are
applied for wedge surface and zero derivatives along
freestream are assumed at incoming and downstream
boundaries. To the right of the shock, the flow field is
initialised with the ambient conditions with = 1.225
kg/m3, p = 101,325 Pa, u = v = 0. To the left of the shock,
the flow variables are computed using moving shock
relations.
Figure 8

23

computed on a uniform grid of 800 600 cells. Figure 8 and


Figure 9 show the computed density contours using AUFS
and AUFSR schemes respectively. The wave pattern of the
solution corresponds to single Mach reflection. It is
observed that the overall wave structure of this problem is
computed well with the two schemes.

3.5.2 Ms = 1.2 shock reflection over a wedge with


= 30
Figure 10 and Figure 11 show experimental and
computational results for the case of a Mach number
Ms = 1.2 and wedge angle of = 30. This case also
corresponds to single Mach reflection, but the incident
shock is weak. Weak shocks, and generally weak flow
features, are difficult to compute accurately. Many
approximate Riemann solvers may be sensitive to extreme
conditions and suffer in this type of problem. The flow
feature is captured very well and produced a perfectly
comparable solution to the experimental result presented in
Toro (1999). The qualitative agreement between the
experimental result and the computations is satisfactory as
demonstrated the numerical holographic interferogram
obtained with the second-order of AUFSR scheme.
Figure 10 Ms = 1.2, experimental results, Ms = 1.2 and = 30

Density contours from AUFS solution, Ms = 1.7,


= 25

Figure 11 Ms = 1.2, numerical interferogram from AUFSR


solution, Ms = 1.2 and = 30

Figure 9

Density contours from AUFSR solution, Ms = 1.7,

= 25

3.6 Diffraction of a shock over a 90 corner

3.5.1 Ms = 1.7 shock reflection over a wedge with


= 25
The computational domain is [0, 25.0] [0, 16.5] on the
x-y plane, with the apex of the wedge placed at x = 4.69,
and the shock placed initially at x = 4.0. This problem is

Another expansion shock problem used to evaluate


numerical instability is the diffraction of a plane shock wave
moving around a 90 corner. Quirk (1994) has shown the
complexity of the flow which generates a series of complex
shock diffraction, reflection, and interaction patterns. This
problem is another test case for which many Godunov type
fluxes suffer from the carbuncle. A square computational
domain that is discretised into a 400 400 uniform cells is
used. The corner is located depending of the studied case.
Initially, the flow field to the right of the shock is initialised
with the ambient conditions and to the left of the shock, the

24

G. Tchuen et al.

flow variables are computed using the Rankine-Hugoniot


relations. The top boundary is taken as a wall, and the
right and bottom boundaries are taken as outflow. All
computations were performed by first-order AUFSR scheme
with CFL = 0.6. Results are shown in Figure 12 to
Figure 15. It can be observed the ability of the AUFSR
scheme to detect shock, contact and expansion regions. The
shape of the primary shock wave for each Mach number
(Ms = 5.09, Ms = 2.4, Ms = 1.3) matches its corresponding
experimental results (Quirk, 1994). All the major flow
structure named by Skews (1967) can be easily identified.
The recompression shock, secondary shock, slipstream,
contact surface, incident shock and expansion wave can all
be seen as sharply defined density gradients. No secondary
shock is present for the case Ms = 1.3. Sun and Takayama
(1997) have determined analytically that the threshold
incident shock Mach number Ms required for a secondary
shock wave to form is Ms = 1.346. The visualisation of
vortex is well defined when Ms < 1.5 and can be observed in
the case Ms = 1.3. The pressure contours for the case
Ms = 1.3 are shown in Figure 15. It can be seen that the
shock can be sharply captured with a good resolution with
AUFSR scheme without any anomaly.

Figure 14 Mach 1.3 shock diffraction, density contours with


AUFSR scheme

Figure 15 Mach 1.3 shock diffraction, pressure contours with


AUFSR scheme

Figure 12 Mach 5.09 shock diffraction, density contours with


AUFSR scheme

Figure 13 Mach 2.4 shock diffraction, density contours with


AUFSR scheme

Conclusions

A novel hybrid numerical scheme is proposed for improve


and eliminate the unrealistic carbuncle phenomenon, and
provide accurate shock resolution. The method combines
the AUFS flux with the Roes FDS scheme. The method
was then evaluated on several well-known test cases and
found to eliminate unphysical solution that may arise
from the use of many approximate Riemann solvers. The
numerical results obtained indicate that the new scheme is
accurate, robust for flow calculation across the Mach
number range. Numerical examples demonstrate that the
new hybrid scheme has high computational accuracy when
compared to AUFS, AUSM, Roe and exact Riemann solver.

References
Coquel, F. and Liou, M.S. (1995) Hybrid upwind splitting scheme
by a field decomposition, NASA TM, 106843.
Edwards, J.R. (1997) A low-diffusion flux splitting scheme for
Navier-Stokes calculations, Comput. Fluids, Vol. 26,
pp.635659.

Hybrid upwind splitting scheme by combining the approaches of Roe and AUFS for compressible flow problems
Einfeldt, B., Munz, C.D., Roe, P.L. and Sjogreen, B. (1991) On
Godunov-type methods near low densities, J. Comput. Phys.,
Vol. 92, p.273.
Godunov, S.K. (1959) Finite difference methods for the numerical
computation of discontinuous solutions of the equations of
fluid dynamics, Math. Sb, Vol. 47, p.271.
Harten, A., Lax, P.D. and Van Leer, B. (1983) On upstream
differencing and Godunov-type schemes for hyperbolic
conservation laws, SIAM Reviews, Vol. 25, p.35.
Harten, A., Endquist, B., Order, S. and Chakravarthy, S. (1987)
Uniformly high order accurate essentially non-oscillatory
schemes, IIP, J. Comput. Phys., Vol. 71, p.231.
Lin, H.C. (1995) Dissipation additions to
flux-difference-splitting, J. Comput. Phys., Vol. 117, No. 1,
pp.2027.
Liou, M.S. and Steffen, C.J. (1993) A new flux splitting scheme,
J. Comput. Phys., Vol. 107, pp.2339.
Liou, M-S. (2000) Mass flux schemes and connection to shock
instability, J. Comput. Phys., Vol. 160, pp.638648.
Pandolfi, M. and dAmbrosio, D. (2001) Numerical instabilities
in upwind methods: analysis and cures for the carbuncle
phenomenon, J. Comput. Phys., Vol. 166, pp.271301.
Quirk, J.J. (1994) A contribution to the great Riemann
solver debate, Inter. J. Numer. Meth. Fluids, Vol. 18,
pp.555574.
Roe, P.L. (1981) Approximate Riemann solvers, parameter
vectors, and difference schemes, J. Comput, Phys, Vol. 43,
pp.357372.

25

Shu, C-W. (1997) Essentially non-oscillatory and weighted


essentially non-oscillatory schemes for hyperbolic
conservation laws, ICASE Report 97-65.
Skews, B.W. (1967) The perturbed region behind a diffracting
shock wave, J. Fluid Mech., Vol.29, No. 4, pp.705719.
Sod, G.A. (1978) A survey of several finite difference methods
for systems of nonlinear hyperbolic conservation law,
J. Comput. Phys., Vol. 27, pp.131.
Steger, J.L. and Warning, R.F. (1981) Flux vector splitting of the
inviscid gasdynamic equations with applications to finite
difference methods, J. Comput. Phys., Vol. 40 pp.263293.
Sun, M. and Takayama, K. (1997) The formation of a secondary
shock wave being a shock wave diffracting at a convex
corner, Shock Waves, Vol. 7, pp.287295.
Sun, M. and Takayama, K. (2003) An artificially upstream flux
vector splitting scheme for the Euler equations, J. Comput.
Phys., Vol. 189, pp.305329.
Tchuen, G., Burtschell, Y. and Zeitoun, E.D. (2008) Computation
of nonequilibrium hypersonic flow with artificially upstream
flux vector splitting (AUFS) schemes, Int. J. Comput. Fluid
Dyn., Vol. 22, No. 4, pp.209220.
Toro, E.F. (1999) Riemann Solvers and Numerical Methods for
Fluid Dynamics, 2nd ed., Springer, Berlin.
Van Leer, B. (1977) Towards the ultimate conservative difference
scheme. III. Upstream-centered finite-difference schemes for
ideal compressible flow, J. Comput. Phys., Vol. 23, p.263.
Woodward, P. and Collela, P. (1984) The numerical simulation of
two-dimensional fluid flow with strong shocks, J. Comput.
Phys., Vol. 54, pp.115173.

Vous aimerez peut-être aussi