Proceeding of the 18

th
ME-NETT Conference
18-20 October 2004, Khon Kaen
Abstract
A finite element method for solving inviscid high-speed
compressible flow problems is presented. The finite element
equations corresponding to these flow problems were derived
from the governing Euler equations that consist of the
conservation of mass, momentums, and energy using the
characteristic-based split method for the three-nodes triangular
element. An adaptive meshing technique was combined with the
finite element method to improve the solution accuracy and to
reduce the computational time as well as the computer memory.
The efficiency of the combined method is evaluated by the
examples of an oblique shock reflection at a wall, a Mach 2.0 flow
in a channel with compression and expansion ramps, and a
shock-shock interaction on a cylinder.

1. Introduction
High-speed compressible flows normally include complex
flow phenomena, such as shock waves, flow expansions, and
shock-shock interactions [1]. Effects of these phenomena are
critical in the design of high-speed vehicle structures. These
flows are characterized by steep solution gradients that need
robust analyses and computational techniques as well as dense
meshes to obtain good resolution of flow behaviors. In the past
decades, several finite element algorithms were developed to
alleviate the computational effort due to complex flow filed, such
as the Taylor-Galerkin [2], the Petrov-Galerkin [3], the least-
squares [4], the cell-centered upwinding algorithms [5] and the
characteristic-based split algorithm [6].
The characteristic-based split algorithm or the CBS
algorithm is selected for solutions in this paper due to its
capability to provide flow solution accuracy for most of the fluid
dynamics problems. An adaptive meshing technique is combined
with the CBS algorithm to improve the finite element solution
accuracy and to reduce the computational time as well as the
required computer memory. The adaptive meshing technique is
applied to generate small elements in the regions of large change
in the solution gradients to increase the solution accuracy, while
larger elements are generated in the other regions. The paper
starts by explaining the theoretical formulation for inviscid high-
speed compressible flow analysis and the algorithm procedure.
The basic idea behind the adaptive meshing technique is then
described. Finally, the combined procedure is evaluated by
analyzing the three examples of high-speed compressible flows;
an oblique shock reflection at a wall, a Mach 2.0 flow in a channel
with compression and expansion ramps, and a shock-shock
interaction on a cylinder. The predicted solutions are compared
with the exact solutions, and the experimental results.

2. Governing Differential Equations
The Euler equations for inviscid compressible flow are
governed by the conservation of mass, momentums and energy.
These equations, in two dimensions, are written in the
conservation form as,
{ } { } { } 0 =

F
y
E
x
U
t
(1)
The vector { } U contains the conservation variables defined by,
{ }
T
U

v u = (2)
where is the fluid density, u and v are the velocity
components in the x and y directions, respectively, and is the
total energy of the fluid. The vectors { } E and { } F consist of


High-Speed Compressible Flow Solutions by Combined
Characteristic-Based Split Method and Adaptive Meshing Technique

Parinya BOONMARLERT Sutthisak PHONGTHANAPANICH
Pramote DECHAUMPHAI
*
Department of Mechanical Engineering, Faculty of Engineering,
Chulalongkorn University 10330, Thailand
Tel: 0-2218-6621 Fax: 0-2218-6621 E-mail: fmepdc@eng.chula.ac.th

* Professor, Mechanical Engineering Department,
Chulalongkorn University Bangkok 10330

inviscid fluxes in the x and y directions, respectively. These
inviscid flux vectors are given by,
{ }
T
E

u p u v u p u u + + =
2
(3)
and
{ }
T
F

v p v p v v u v + + =
2
(4)
where p is the pressure. The total energy consists of the
internal energy and the kinetic energy defined by,
( )
2 2
2
1
v u e + + = (5)
The internal energy is assumed to satisfy the equation of state
that can be written in the form,
e
( ) 1
=

p
(6)
where is a specific heat ratio.

3. Computational Procedure
The basic concept of the CBS algorithm is to use the
characteristic-Galerkin process to establish recurrence relations
for temporal discretization, and the method of weighted residuals
with Galerkins criteria is used for spatial discretization for deriving
the finite element equations.

3.1 The temporal discretization
The CBS algorithm for the compressible flow analysis
consists of four steps. In the first step, the intermediate values of
conservative variables of the momentum equations are calculated
by omitting the pressure gradient terms. In the second step, the
continuity equation is solved to determine the density changes in
the fluid. In the third step, the conservative variables of the
momentum equations are updated. Finally, the energy equation
is solved for the total energy and the pressure is calculated by
using the equation of state. These four steps in the fully explicit
form can be written as follows [7],

Step 1: Solve the intermediate momentum equations,
*
i
U =
( ) ( )
n
i j
i j
k
k
j
i j
x
p
x
U u
x
u
t
x
U u
t

|
|
.
|

\
|

2
(7)
where
i
U are the mass fluxes.




Step 2: Solve the continuity equation,
=
n
i i i
*
i
i
i
x x
p
t
x
U
x
U
t

2
(8)
where is between 0.5 and 1.

Step 3: Solve the momentum correction equations,
i
U =
n
i
i *
i
x
p
t U

(9)
Step 4: Solve the energy equation,
=
( ) ( )

i
i
i
i
x
p u
x
u
t



( ) ( )
n
i
i
i
i
k
k
x
p u
x
u
x
u
t

|
|
.
|

\
|

+

2
(10)
3.2 The spatial discretization
The three-nodes triangular element is used in this study. The
element assumes linear interpolation for the variable U , E , F
and p as,
U =

U ) y , x ( N (11a)
E =

E ) y , x ( N (11b)
F =

F ) y , x ( N (11c)
p =

p ) y , x ( N (11d)
where = 1,2,3 and

N is the element interpolation functions.

N , and performing
integration by parts using the Gauss theorem to yield the element
equations in the steps below,

Step 1: Solve the intermediate momentum equations,
| |{ }
*
i
U M = | | | | ( ){ } | |{ }

i j u k i j
U u K u
t
U u S C t
2

| |{ } | | | | ( ){ }
n
p p i j u
p T K U u T

+ (12)
Step 2: Solve the continuity equation,
| |{ } M = | | | | ( ){ }

+
*
i i u
U U R D t
| |{ }
n
p K t

(13)

Step 3: Solve the momentum correction equations,
| |{ }
i
U M = | |{ }
*
i
U M | | | | ( ){ }
n
u
p R D t

+ (14)
Step 4: Solve the energy equation,
| |{ } M = | | | | ( ) ( ) { }

+ p u S C t
j

| | | | ( ) ( ) { }
n
j u u k
p u T K u
t

2
(15)
In above equations, the element matrices can be written in the
integral form as,
| | M = { }

d N N (16a)
| | C =

d N
x
N
j
(16b)
| | D =

)
`

d N
x
N
i
(16c)
| | K =

)
`

d
x
N
x
N
i i
(16d)
| |
u
K =

)
`

d
x
N
x
N
j k
(16e)
| |
p
K =

)
`

d
x
N
x
N
i k
(16f)
| | S = { }

d n N N
j
(16g)
| |
u
R = { }

d n N N
i
(16h)
| |
u
T = { }

d n
x
N
N
k
j
(16i)
| |
p
T = { }

d n
x
N
N
k
i
(16j)
An artificial diffusion is also needed in the algorithm to
reduce oscillation of the solutions especially near the shock wave.
The second derivative of pressure [8] is selected to contribute the
artificial diffusion into nodal quantities. These nodal artificial
diffusions are determined from,
| |

)

+ +
t
U U
M
n n
s
1 1
= | |{ }
n
e
e
U K p
p
c V
h C
2 3

+
(17)
where
1 + n
s
U is the modified solution at time step + n 1 after
adding artificial diffusion,
1 + n
U is the solution at time step
+ n 1,
e
C is the user-specified coefficient normally varies
between 0.0 and 2.0, h is the element size, V is the absolute
velocity, c is the speed of sound, p is the average pressure,
e
p
2
is the second derivative of pressure over an element.
The fully explicit form of CBS algorithm is conditionally
stable. The permissible time step is governed by,
t =
c V
h
+
(18)
where is the Courant number (0< 1).

4. Adaptive Meshing Technique
For high-speed compressible flows, the flow properties, such
as the density changes abruptly across the shock waves. Small
elements are thus needed along to shock waves to capture
accurate shock wave resolution. As small elements must be
placed in the region where changes in the density gradients are
large, thus the second derivatives of density at a point with
respect to global coordinates x and y are needed. Using the
concept of principal stresses determination from a given state of
stresses at a point, the principal quantities in the principal
directions X and Y where the cross-derivatives vanish are
determined. The maximum principal quantities are then used to
compute the proper element size
i
h by requiring that the error
should be uniform for all elements,

2
2 2
2
2
2
y y x
y x x

2
2
2
2
0
0
Y
X

(19)
This value is used to compute proper element size from the
condition,
= =
max min i i
h h
2 2
constant (20)
In the above Eq. (20),
max
is the maximum principal quantity
for all elements and
min
h is the minimum element size specified
by users.

5. Results
To demonstrate the capability of the combined adaptive
meshing technique and the characteristic-based split method for
increasing the flow solution accuracy, three simulations of the
steady-state high-speed compressible flows are used. (1) an
oblique shock reflection at a wall, (2) a Mach 2.0 flow in a
channel with compression and expansion ramps, and (3) a shock-
shock interaction on a cylinder.

5.1 Oblique shock reflection at a wall
The problem statement of an oblique reflection at a wall is
described in Fig. 1. The Mach 2.9 and 2.387 flows enter through
the left and the top boundaries of computational domain resulting
in an oblique shock from the top-left corner. This shock incidents
and reflects at a wall as highlighted in the figure. The procedure
starts by creating a relatively uniform mesh as shown in Fig. 2(a)
that consists of 4,920 elements. The fluid analysis is then
performed to generate the corresponding solution such as the
density contours as shown in Fig. 2(b). The figure shows the
computed shock is not sharp because the elements along the
shock lines are not small enough. This flow solution is then used
to generate an adaptive mesh to cluster small elements in the
regions of sharp changes of the density gradients, and at the
same time, to use larger elements on the other regions. The fluid
analysis is then performed again to yield a more accurate
solution. The entire process is repeated to generate the third
adaptive mesh consisting of 19,882 elements and the
corresponding solution as shown in Figs. 3(a)-(b), respectively.
Figures 4a)-(b) show the predicted density and pressure
distributions at y = 0.25 as compared to with the exact solutions.








Fig. 1 Problem statement of an oblique shock reflection at a wall.



(a)




(b)
Fig. 2 An oblique shock reflection at a wall: (a)-(b) Initial mesh
and the corresponding density contours.






(a)




(b)
Fig. 3 An oblique shock reflection at a wall: (a)-(b) Third
adaptive mesh and the corresponding density contours.










(a)










(b)
Fig. 4 Comparative solutions at y = 0.25 for an oblique shock
reflection at a wall: (a) density distribution; (b) pressure
distribution.

5.2 Mach 2.0 flow in a channel with compression and
expansion ramps
The problem statement of a Mach 2.0 in a channel with
compression and expansion ramps and the sketch of flow
behavior are described in Fig. 5. The flow creates an oblique
shock from the compression ramp that impinges at the upper wall
resulting in a reflecting shock. The reflecting shock also
intersects with the Mach waves generated from the expansion
corner. The combined characteristic-based split method and
y
x
0 4.
0 1.
Computational domain
Oblique shock wave
Reflection shock wave
3781 2. M =

94 10.

29

28 23.
9 2. M =
0.0
0.5
1.0
1.5
2.0
2.5
3.0
0 1 2 3 4

Exact
Initial mesh
Adaptive mesh
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0 1 2 3 4
p
x
Exact
Initial mesh
Adaptive mesh
x

adaptive meshing technique starts from generating a relatively
uniform mesh such as that shown in Fig. 6(a) with 5,578
elements. Fig. 6(b) shows the density contours of the
corresponding flow solution obtained from the finite element mesh
in Fig. 6(a). The figure shows that the computed shock waves
are not sharp and mach waves resolution around expansion
corner is not good due to the element sizes in these regions are
too large. With such the solution, a new adaptive mesh was then
constructed and the flow analysis was performed again. The
same process was repeated for three times. The third adaptive
finite element model consisting of 26,628 elements and the
corresponding density contours are shown in Fig. 7(a)-(b),
respectively.







Fig. 5 Problem statement of a Mach 2.0 in a channel with
compression and expansion ramps




(a)



(b)
Fig. 6 Mach 2.0 in a channel with compression and expansion
ramps: (a)-(b) Initial mesh and the corresponding density
contours.




(a)



(b)
Fig. 7 Mach 2.0 in a channel with compression and expansion
ramps: (a)-(b) Third adaptive mesh and the corresponding
density contours.
5.3 Shock-shock interaction on a cylinder
The problem highlights the shock-shock interaction on a
cylinder. The mainstream Mach number is 8.03 and the disturbed
stream Mach number is 5.25 at angle of 12.5
o
. The problem
statement of the flow is shown in Fig. 8. This represents a
situation in which an oblique shock interacts with bow shock in
front of the cylinder. The adaptive finite element model consisting
of 40,359 elements is shown in Fig. 9(a). The corresponding
pressure and Mach contours are presented in Figs. 9(b) and (c),
respectively. Figure. 10 shows the predicted pressure on the
surface of the cylinder comparing to the experimental data [9].
The predicted and experimental pressures are normalized by the
undisturbed flow (no shock interaction) stagnation pressure. The
figure shows good agreement of the pressure distributions and
excellent agreement of the peak pressure locations.

























Fig. 8 Problem statement of a shock-shock interaction on a
cylinder




0 2. M =
0 1.
0 1. 0 1. 0 3.
x
y
Shock wave
Computational domain
Mach wave
03 8. M =
Incident shock
Cylinder
Bow shock
Computational domain

0 3.
5 3.
5 5.
5 3.
03 8. M =
25 5. M =
= 5 12.
5 12.














(a) (b) (c)
Fig. 9 Shock-shock interaction on a cylinder: (a) adaptive mesh
(b)-(c) the corresponding pressure and Mach contours.











, degree
Fig. 10 Comparative pressure distributions along cylinder
surface.

6. Conclusion
The finite element method based on the characteristic-based
split algorithm for analysis of two-dimensional inviscid high-speed
compressible flow was presented. The method was combined
with an adaptive meshing technique to improve the flow solution
accuracy and to reduce the computational time. The technique
generates an entirely new mesh based on the solution obtained
from a previous mesh. The new mesh consists of clustered
elements in the regions with large change in the solution
gradients to provide high solution accuracy. At the same time,
larger elements are generated in the other regions to reduce the
computational time. Three examples of high-speed compressible
flows were presented to assess the effectiveness of the combined
method. These examples are an oblique shock reflection at a
wall, a Mach 2.0 flow in a channel with compression and
expansion ramps, and a shock-shock interaction on a cylinder.
These three examples demonstrate the combined method can
provide high solution accuracy with reduced computational time
and memory for analysis of high-speed compressible flow
problems.

7. Acknowledgments
The authors are pleased to acknowledge the Thailand
Research Fund (TRF) for supporting this research work.

8. References
[1] Anderson, J.D. Jr. Modern Compressible Flow with Historical
Prospective. (2
nd
ed). New York: McGraw Hill.
[2] Donea, J. A Taylor-Galerkin Method for Convective Transport
Problems, International Journal for Numerical Methods in
Engineering, 1984, Vol. 20, pp. 101-119.
[3] Huges, T.J.R. Recent Progress in the Development and
Understanding of SUPG Methods with Special Reference to the
Compressible Euler and Navier-Stokes Equations, International
Journal for Numerical Methods in Fluids, 1987, Vol. 7, pp. 1261-
1275.
[4] Jiang, B.N. and Carey, G.F. A Stable Last-Squares Finite
Element Method for Non-Linear Hyperbolic Problems,
International Journal for Numerical Methods in Fluids, 1988, Vol.
8, pp. 933-942.
[5] Gnoffo, P.A. Application of Program LUARA to Three-
Dimensional AOTV Flow Fields, AIAA Paper 86-0565, 1986.
[6] Zienkiewicz, O. C., Nithiarasu, P., Codina, R., Vazquez, M.
and Ortiz, P. The Characteristic-Based-Split Procedure: An
Efficient and Accurate Algorithm for Fluid Problems, International
Journal for Numerical Methods in Fluids, 1999, Vol. 31, pp. 359-
392.
[7] Satya Sai, B.V.K., Zienkiewicz, O. C., Manzari, M.T., Lyra,
P.R.M. and Morgan, K. General Purpose Versus Special
Algorithms for High-Speed Flows with Shocks, International
Journal for Numerical Methods in Fluids, 1998, Vol. 27, pp. 57-80.
[8] Nithiarasu, P., Zienkiewicz, O.C., Satya Sai, B.V.K., Morgan,
K., Codina, R.and Vazquez, M. Shock* Capturing *Viscosities*
for* the* General* Fluid* Mechanics *Algorithm, International
Journal for Numerical Methods in Fluids, 1998, Vol. 28, pp. 1325-
1353.
[9] Dechaumphai, P. and Wieting, A. R. Couple Fluid-Thermal-
Structural Analysis for Aerodynamically Heated Structures, 7
th

International Conference on Finite Element Methods in Flow
Problems, Alabama, April 3-7, 1989.

0
2
4
6
8
10
-90 -45 0 45 90
0
p
p
Flow
Cylinder

CBS
Reference [9]