NUMERICAL AND EXPERIMENTAL SIMULATION OF AN

AXISYMMETRIC SUPERSONIC INLET

Mohammad Reza Soltani1, Mohammad Farahani2, and Javad Sepahi Younsi3

1, 2, 3: Department of Aerospace Engineering,
Sharif University of Technology, Tehran, Iran
E-mail: 1. [email protected], 2. [email protected], 3. [email protected]
ABSTRACT
A numerical code for inlet design with proper simulation of normal and/or oblique shocks, boundary layer
development, interaction of the shock and the boundary layer and flow separation can be of great help to the
designers. To solve the inner and outer flow field of the inlet, the numerical methods have been used and
various experimental tests for the code validation have been conducted. The inlet is an axisymmetric exter-
nal compression one. To compute the convective fluxes Roe scheme and new schemes, AUSM+-up for all
speed and AUSMPW, have been used. The original version of the AUSMPW scheme has serious problems
concerning the stability and accuracy. In this investigation for the first time some modifications inspired
from the AUSM+-up for all speed scheme have been implemented on the original version of the AUSMPW
scheme that result in a better stability and a better accuracy. Turbulent viscosity coefficient has been com-
puted using the Baldwin-Lomax algebraic model. Results show that the numerical code compares well with
the experimental data. The Roe scheme has the best accuracy and the AUSM+-up for all speed is numeri-
cally very efficient.

Keywords: supersonic inlet/Roe scheme/AUSM+-up for all speed scheme/AUSMPW scheme

1. INTRODUCTION Investigation of the flow field of the supersonic

Supersonic inlet is often considered to be the most inlet has been the subject of many researches. For
crucial element in the air-breathing engine of a example NASA has done many investigations in
flying vehicle designed to operate at supersonic this field. However, most of them were experimen-
speeds. The function of the inlet is to provide sub- tal [1-4] with a few numerical-experimental studies
sonic flow at the desired velocity to the [5-7]. The numerical methodology uses in the new
downstream components of the engine with mini- researches in this field [8-10] is similar to our re-
mum loss in the total pressure, flow uniformity, and search, however, in this study the new convective
flow stability, all of which are important to the flux schemes for the inlet flow field as well as the
overall engine efficiency. Accordingly proper code validation have been compared with the ex-
simulation of the inlet is one of the most important perimental data.
steps in the designing of the propulsion system of Today, upwind schemes are the main spatial discre-
any flying vehicle particularly flying at supersonic tization schemes. Upwind schemes are usually
speeds. In this research a numerical code was de- categorized as either FDS (Flux Differencing Split-
veloped to solve the Navier-Stokes equations with ting) or FVS (Flux Vector Splitting). FDS uses the
explicit finite volume discretization scheme around solution of the local Riemann problem. Roe FDS is
an axisymmetric supersonic external compression the most popular method among the FDS schemes
inlet. To validate the results, wind tunnel tests for a because of its accuracy and efficiency. However, it
few cases were conducted. The discretized govern- predicts non-physical expansion shock waves and
ing equations were applied in a structured grid carbuncle phenomena. FVS splits the flux accord-
generated with the elliptic grid generator. Since of ing to the sign of its eigenvalues. These schemes
the flow is steady, the time step has been calculated are however, more robust and efficient than the
with the local time stepping method to accelerate FDS schemes. FVS is based on the scalar calcula-
the convergence. In order to increase the spatial tions while FDS is based on the matrix
accuracy of discretization, the MUSCL approach calculations, thus FVS is more efficient than FDS.
has been used. Viscous fluxes with a finite volume Most F'VS such as Van Leer FVS, however, has
method, molecular viscosity coefficient using the excessive numerical dissipation in the shear layer
Sutherland law and the turbulent viscosity coeffi- regions that gives inaccurate results in those re-
cient using the Baldwin-Lomax algebraic model gions. Liou and Steffen proposed AUSM
have been computed. (Advection Upstream Splitting Method) where the
cell interface advection Mach number is appropri-
ately defined to determine the upwind extrapolation
for the convective quantities. As a result of vanish-
ing excessive numerical dissipation, AUSM is very Figure 1: geometry of the inlet
accurate and shows exact results similar to the FDS
in the shear layer regions. Besides, AUSM is as
simple and robust as Van Leer FVS. However,
AUSM has some serious problems: carbuncle phe-
nomena and numerical overshoots behind strong
shock waves and numerical oscillations near re-
gions of small convection velocity or small
pressure gradient, such as around a wedge or near a
wall [11].
To overcome the mentioned difficulties and gener-
alize the scheme for low Mach number cases, Liou
proposed a new scheme, AUSM+-up for all speed Figure 2: grid around the inlet
in 2006 [12]. In addition, to overcome the carbun- ∆t 4

cle phenomena and the numerical oscillations of Qin, +j 1 = Qin, j − i , j [ ( Fc ) k ∆sk ]i , j − α∆ti , jVi , j +
Ai , j k =1
AUSM family schemes simultaneously, Kim et al. (3)
proposed AUSMPW scheme in 1998 that uses two ∆ti , j 4
[ ( Fv ) k ∆sk ]i , j + α∆ti , jVvi , j ,
pressure based weight functions [11]. AUSM+-up Ai , j k =1

for all speed and AUSMPW schemes are examined where

in this study and their results are compared with the ρq 0
Roe scheme.
ρ uq + nx P nxτ xx + n yτ xy
Fc = , Fv = ,
2. NUMERICAL METHODOLOGY ρ vq + n y P nxτ xy + n yτ yy
The geometry of the inlet and the grid around it are ρ Hq nx Θ x + n y Θ y
(4)
presented respectively in Fig. 1 and Fig. 2. The ∂T
Free stream Mach number is 2. The grid is struc- Θ x = uτ xx + vτ xy + k ,
∂x
tured and was generated with an elliptic grid
generator. The governing equations that are discre- ∂T
Θ y = uτ xy + vτ yy + k .
tized in this grid are Navier-Stokes equations in ∂y
conservative two dimensional form: In the above equations, H is the total enthalpy, A is
∂Q ∂T ∂U ∂T ∂U v the area of the cell, s is the length of the cell face
+ + + αV = v + + αV v , (1)
∂t ∂x ∂y ∂x ∂y and q is the velocity component normal to the cell
where face.
ρ ρu ρv To accelerate the convergence, the time step in
ρu P + ρu 2
ρ uv equation (3), ti,j has been calculated using the
Q= ,T = ,U = , local time stepping:
ρv ρ uv P + ρv2
ρE ρ uH ρ vH
A ij
∆t ij = CFL m , (5)
ρv 0 (Λ + Λ ) + C (ΛvI + ΛvJ )ij
I
c
J
c ij

1 ρ uv τ xx where
V= , Tv = , (2)
y ρv2 τ xy Λ cI = ( q I + c)∆s I ,
ρ vH uτ xx + vτ xy − qx
4 γ µ L µT (∆s I ) 2 (6)
0 0 Λ vI = max( , )( + ) .
3ρ ρ PrL PrT A
τ xy 1 τ xy
Uv = ,Vv = . In the above equation c is the speed of sound and
τ yy y τ yy − τ θθ
uτ xy + vτ yy − q y uτ xy + vτ yy − q y
for J direction similar equations are used. T is the
turbulent viscosity coefficient that is calculated by
If =0 the equations are 2D planar and for =1 they the Baldwin-Lomax algebraic model.
are 2D axisymmetric. By applying an explicit finite The MUSCL approach with the van Albada limiter
volume discretization and after a few manipula- is used to increase the accuracy of the spatial dis-
tions equation (1) becomes: cretization:
 ε Kp PR − PL
Q iL+1/2 = Q i + s i (1 − s i k )∆i− + (1 + s i k )∆i+ , Mp =− max(1 − σ M 2 , 0) ,
4 fa ρ1/2ci2+1/2, j
ε
(7) (15)
Q iR+1/2 = Q i +1 − s i +1 (1 − s i +1k ) ∆i++1 + (1 + s i +1k )∆i−+1 , ρL + ρR
4 σ ≤ 1, ρ1/ 2 = , 0 ≤ K p ≤ 1,
where 2
2∆ i+ ∆ i− + η where
si = , f a ( M O ) = M O (2 − M O ),
(∆ i+ ) 2 + (∆ i− )2 + η
(8) M O2 = min(1, max( M 2 , M co2 )),
∆ i− = Qi − Qi −1 ,
1 (16)
∆ i+ = Qi +1 − Qi . M2 = ( M L2 + M R2 ),
-12 2
=1*10 to prevent the denominator to become
1
zero. If =0 the discretization is with the 1st order M co = min( M ∞ ,1).
2

2
accuracy and if =1 the discretization is with the
higher order accuracy. If k=-1 the discretization is fa is used to scale the pressure diffusion term for
2nd upwind and if k=1/3 the discretization is 3rd the low Mach number cases.
order. The pressure at the cell interface is defined as:
As a first step common to all AUSM schemes, the Pi +1/ 2, j = ( n+ ) ( M L )PL + ( n− ) (M R ) PR + Pu , (17)
inviscid flux is explicitly split into convective and where
pressure fluxes: Pu = − Ku (5)+ ( M L ) (5)− ( M R )( ρ L + ρ R ) ×
(F )
c
i +1/2, j
= m i +1/2, jψ L / R + p i +1/2, j , (9) ×ci +1/2, j ( qR − qL ), 0 ≤ K u ≤ 1.
(18)

where ±
are polynomial functions of degree n (= 1,3,
(n )
ψL if m i +1/2, j > 0 5). The fifth degree polynomials are preferred be-
ψ L /R = ,
ψR otherwise (10) cause they are found to give better accuracy:
1 ±
T if M ≥ 1
p i +1/2, j = Pi +1/2, j 0 n x ny 0 . ±
(M ) = M
(1)
. (19)
(5)

m i +1/2, j and Pi +1/ 2, j are mass and pressure fluxes

±
(2) ( ±2 − M ) 16α M (2) otherwise

respectively that are defined differently in various Velocity diffusion term, Pu, also helps to introduce
versions of the AUSM schemes. a scheme for all speed and relate the pressure field
In the AUSM+-up for all speed scheme some modi- to the velocity field [13]. However, in this research
fications are applied to overcome the difficulties comparison of the results with the experimental
with the low Mach number flows and introduce a data show that Ku=0 is the best value.
scheme for all speed with an acceptable level of In the above equations =1/8 and for all speed cal-
accuracy and efficiency. In this scheme the mass culations:
flux at the cell interface has the form of [12]: 3
α= (−4 + 5 f a2 ) (20)
m i +1/2, j = q i +1/2, j ρ L / R = c i +1/ 2, j M i +1/ 2, j ρ L / R , (11) 16
Due to the entropy-satisfying property, the speed of
where sound at the cell interface is defined as:
ρL if q i +1/ 2, j > 0 ci +1/2, j = min(cˆL , cˆR ),
ρL / R = . (12) (21)
ρR otherwise where
The cell interface Mach number is defined as: cˆL = (c* ) 2 / max(c* , qL ),
M i +1/2, j = (+m ) ( M L ) + (−m ) ( M R ) + M p , (13) cˆR = (c* ) 2 / max(c* , − qR ),
± (22)
where (m) are polynomial functions of degree m 2(γ − 1)
(c ) = * 2
H.
(= 1,2, 4) defined as: γ +1
± 1 The main feature of the AUSMPW scheme is re-
(1)(M ) = (M ± M ),
2 moval of the oscillations of AUSM+ scheme near
1
±
(2) ( M ) = ± (M ± 1)2 , the wall or across a strong shock wave by introduc-
4 (14)
±
ing pressure-based weight functions. AUSMPW
if M ≥ 1
±
(4) (M ) =
(1)
. scheme uses the pressure-based weight function, f,
±
(2) (1 16β (2) ) otherwise to remove the oscillations near the wall and to
The pressure diffusion term Mp, introduced to en- remove the oscillation across a strong shock wave.
hance calculations of the low Mach number or The starting point of AUSMPW scheme is to ob-
multi-phase flow cases, is defined as: serve the fact that AUSM+ and AUSMD schemes
are complementary to each other. AUSM+ scheme we apply the all speed modifications of AUSM+-up
has no carbuncle phenomena but shows numerical for all speed scheme to improve the stability and
oscillations near a wall while AUSMD scheme has accuracy of the original AUSMPW scheme. The
no numerical oscillations near the wall but shows modifications that were applied to the original
carbuncle phenomena [14]. AUSMPW scheme were: 1) addition of the pres-
Combining the AUSM+ and AUSMD schemes the sure diffusion term, Mp, according to equation (15)
convective flux in the AUSMPW scheme is calcu- to the cell interface Mach number in the AUSMPW
lated as: scheme, m i +1/ 2, j , 2) calculating the as a function
( Fc )i +1/2, j = M L+ ci+1/ 2, j Φ L + of fa (according to equation (20)), it is not constant
(23) anymore, and 3) calculating the speed of sound
+ M R− ci +1/ 2, j Φ R + pi +1/2, j ,
according to equations (21) and (22). The effect
where and role of these modifications will be discussed
ρ 0 later.
ρu nx
Φ= , pi +1/2, j = PS , 3. EXPERIMENTAL SIMULATION
ρv ny To increase the reliability of the numerical code it
ρH 0 must be examined in different flow patterns. Three
+ − different mass flow ratio ( : the ratio of the mass
PS = ( M L ) PL + ( M R ) PR ,
(5) (5)
flow in the inlet to the mass flow of the free stream
M L+ = +
(4) (M L ) + −
(4) (M R ) − entering the inlet allowing some spillages) are
tested. In all tests M=2 and the angle of attack was
−
(4) ( M R ) × ω × (1 + f R ) + zero. In each test four series of experimental data
mi +1/2, j ≥ 0 : (24)
(
+ fL +
(4) (M L ) + f R −
(4) (M R ) ) were obtained (Fig. 3).

M R− = −
(4) ( M R ) × ω × (1 + f R )
M L+ = +
(4) ( M L ) × ω × (1 + f L )
M R− = +
(4) (M L ) + −
(4) (M R ) −
mi +1/2, j < 0 : +
(4) ( M L ) × ω × (1 + f L ) +
Figure 3: the locations of pressure measurement
+ fL ( +
(4) (M L ) + f R −
(4) (M R ) . )
4. RESULTS AND DISCUSSION
The speed of sound is calculated as: One goal of this research was to validate the nu-
ci +1/2, j = min(cˆL , cˆR ), merical code results. In this regard Fig’s 4-7
(25)
cˆ = (c* ) 2 / max(c* , q ). compare the code results with the experimental
data. As seen from those Fig’s, acceptable level of
Other quantities in equation (24) are:
+ − accuracy has been obtained. For other flow patterns
mi +1/2, j = (M L ) + ( M R ),
(4) (4)
and schemes similar results has been obtained.
PL PR 3 The Schlieren pictures confirm the code results too,
ω ( PL , PR ) = 1 − min( , ),
PR PL Fig. 8.
PL , R ±
( − 1) pl ( PL, R , PR , L ) (4) β = 0
PS
vL , R
f L , R = × min(1, ( )3 ) ; M L , R ≤ 1 ,
ci +1/ 2, j (26)
0 ; M L, R > 1

x y 3 x y
4 min( , ) − 3; ≤ min( , ) < 1
y x 4 y x
pl ( x, y ) = .
x y 3
0 ; 0 ≤ min( , ) < Figure 4: total pressure distribution in the outer rake
y x 4 for =0.9021and AUSMPW scheme
In this scheme: =3/16, =1/8
It is found in this research that the AUSMPW
scheme has stability problem even with very small
CFL numbers. So, for the first time in this research
 function of fa have stabilized the scheme and calcu-
lating the speed of sound according to equations
(21) and (22) has increased the accuracy. The ef-
fect of changing the speed of sound is presented in
Fig's 9-10.

Figure 5: static pressure distribution over the cowl

for =0.8180 and AUSM+-up scheme

Figure 9: static pressure distribution over the spike

for =0.8254 and AUSMPW scheme

Figure 6: static pressure distribution over the spike

for =0.8254 and Roe scheme

Figure 10: total pressure distribution in the inner

rake for =0.8254 and AUSMPW scheme
Adding the pressure diffusion term increase the
numerical dissipation. Numerical dissipation
Figure 7: total pressure distribution in the inner rake smooths the discontinuities and prevents solution
for =0.8254 and Roe scheme
divergence. By allowing to vary as a function of
fa the calculations for low Mach number regions
has been enhanced. Calculating the speed of sound
according to equations (21) and (22) considers the
flow direction and satisfies the entropy condition.
With the speed of sound in equation (25) expansion
fans can not be modeled properly that will result in
inaccurate solution.
Comparison of the flux schemes can be done with
considering some factors such as stability, numeri-
cal efficiency, and accuracy. Al three schemes were
used in this research have the same stability treat-
ment. But the AUSMPW scheme (with the
mentioned modifications) requires small CFL num-
ber to converge.
High numerical efficiency consists of two factors:
rapid calculations and small number of iterations
for convergence. In this regard the AUSM+-up for
all speed scheme is the best followed by the
AUSMPW and Roe schemes. High efficiency of
Figure 8: comparison of shock pattern between nu- the AUSM+-up for all speed scheme is due to the
merical code and experimental tests rapid scalar calculations, while Roe scheme is
Next we examined the effect of modifications on based on matrix calculations and need to satisfy an
the original AUSMPW scheme. Adding the pres- extra condition for the entropy. AUSMPW scheme
sure diffusion term and allowing to vary as a also needs to compute the functions f and .
 According to Fig's 11-12 the accuracy of the Roe Internal Contraction on Design, NASA, TM X-
scheme is more than the two other schemes. The 3042, 1974.
Roe scheme has high accuracy because it uses the 3. Allen, J.L. and Beke, A., Performance Compari-
waves to find the discretization direction and ap- son at Supersonic Speeds of Inlets Spilling
ples them in the calculation, while other schemes Excess Flow by Means of Bow Shock, Conical
use the waves just to find the discretization direc- Shock, or Bypass, NASA, RM-E53H11, 1953.
tion. This results in applying the physics into the 4. Blackaby, J.R., Lyman, E.G., and Altermann,
numerical scheme, thus getting high accuracy. Also J.A., Inlet Performance Characteristics from
the accuracy of AUSM+-up for all speed scheme is Wind Tunnel Test of a 0.1-Scale Air-Induction
more than the AUSMPW scheme. Modifications of System Model of the YF-108A Airplanes at
the original AUSMPW scheme result in more accu- Mach Numbers of 2.50, 2.76, and 3.00, NASA,
rate results for some situations as shown in Fig. 11. MEMO 7-18-59A, 1959.
5. Hawkins, J.E., Kirkland, F.P., and Turner, R.L.,
Inlet Spillage Drag Tests and Numerical Floe
Field Analysis at Subsonic and Transonic
Speeds of a 1/8-Scale, Two-Dimensional, Ex-
ternal Compression, Variable Geometry,
Supersonic Inlet Configuration, NASA, CR-
2680, 1976.
6. Kumar, A., Numerical Simulation of Scramjet
Inlet Flow Fields, NASA, Technical Paper 2517,
1986.
Figure 11: static pressure distribution over the spike 7. Smith, C.F. and Smith, G.E., Two Stage Super-
for =0.8254
sonic Inlet (TSSI): 10-inch Model Calculations,
NASA, CR-2005-213287, 2005.
8. Hirt, M.S., Tacina, K.M., Conners, T.R., Mer-
ret, J.M., and Howe, D.C., CFD Results for an
Axisymmetric Isentropic Relaxed Compression
Inlet, 46th Aerospace Sciences Meeting and Ex-
hibit, Reno, Nevada, January 7–10, 2008.
9. Gokhale, S.S., and Kumar, V.R., Numerical
Computations of Supersonic Inlet Flow,
INTERNATIONAL JOURNAL FOR
Figure 12: total pressure distribution in the inner NUMERICAL METHODS IN FLUIDS, Vol. 36,
rake for =0.8254 pp. 597-617, 2001.
10. Akbarzadeh, M., and Kermani, M.J., Numerical
5. CONCLUSION Computation of Supersonic-Subsonic Ramjet
The numerical code can capture the experimental Inlets; a Design Procedure, 15th Annual (Inter-
data with high accuracy and numerical efficiency. national) Conference on Mechanical
So it can be used to reduce the cost and time for Engineering-ISME2007, Amirkabir University
designing a supersonic inlet and other similar prob- of Technology, Tehran, Iran, May 15-17, 2007.
lems. Modifications of this study on the original 11. Kim, K.H., Lee, J.H., and Rho, O.H., An Im-
AUSMPW scheme result in a stable and accurate provement of AUSM Schemes by Introducing
scheme. The Roe scheme is very accurate but it is the Pressure-based Weight Functions, Com-
not very efficient, while the AUSM+-up for all puters & Fluids, Vol. 27, No. 3, pp 311-346,
speed scheme has acceptable accuracy and is more 1998.
efficient. 12. Liou, M.S., A sequel to AUSM, Part II:
AUSM+-up for all speeds, J. of Computational
6. REFERENCES Physics 214, pp 137-170, 2006.
1. Koncsek, J.L. and Syberg, J., Transonic and 13. Liou, M.S., Ten Years in the Making–AUSM-
Supersonic Test of a Mach 2.65 Mixed Com- family, NASA, TM-2001-210977, 2001.
pression Axisymmetric Intake, NASA, CR-1977, 14. Kim, K.H., Kim, C., and Rho, O.H., Methods
1972. for Accurate Computations of Hypersonic
2. Woollett, R.R., Meleason, E.T., and Choby, Flows, . AUSMPW+ scheme, J. of Computa-
D.A., Transonic Off-Design Drag and Perform- tional Physics 174, pp 38-80, 2001.
ance of an Axisymmetric Inlet with 40-Percent