Professional Documents
Culture Documents
Application of Optimization Algorithms To Scramjet Inlet Design
Application of Optimization Algorithms To Scramjet Inlet Design
An automated design optimization process is applied to both single and multi ob-
jective optimization problems of scramjet inlet design. This optimization process inte-
grates together an optimizer with a mesh generator, a flow solver, and an objective
analysis tool into an automated optimization loops because the flow simulation is re-
quired for every step along the line search and finding the feasible direction. This paper
presents the implementation of these new design techniques by the gradient-based
optimizer Sequential Quadratic Programming (SQP) and their application to scramjet
inlet case in flight condition of Mach 8. The performance of scramjet inlets with uniform
inflow is improved, and the optimized functions, that is, the total pressure recovery co-
efficient increases by approximately 10%. The trade-off (also known as the ε-constraint)
method is applied and implemented to find the Pareto optimal set formed by the non-
dominated solutions of the feasible design. The objective functions are the total pressure
loss and the drag, and some solutions are obtained to analyze the relations between the
total pressure loss and the drag.
-1-
American Institute of Aeronautics and Astronautics Paper 2005-3207
Copyright © 2005 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
13th AIAA/CIRA International Space Planes and Hypersonic Systems
and Technologies Conference, 16-20, May, 2005, Capua, Italy
Aerospace Laboratory (NAL) and Na- tation and application of an efficient and
tional Space Development Agency powerful tool to optimize the aerody-
(NASDA) consolidated. JAXA has been namic performance of a two-dimensional
conducting scramjet engine research at scramjet inlet. The results for improved
the flight conditions of Mach 4, 6 and 8 by scramjet inlets are described using the
using the RamJet engine Test Facility single objective optimization. In addition,
(RJTF)1-5 at Space Propulsion Research multi objective optimizations are con-
Center. A free-piston high-enthalpy shock structed, and some solutions are obtained
tunnel (HIEST)1,6, the largest in the by applying the trade-off method.
world, was completed in November 1997.
Scramjet Engine Testing in the range of
Mach 8 to 15 has been conducted. In ad- Geometry Model
dition to the large scale experimental
research, numerical computations of A set of parameters is needed to de-
scramjet engines are conducted on vector fine the two-dimensional scramjet inlet
parallel supercomputers7-10. The com- geometry so that a design space can be
prehensive research activities are sum- formed for the optimization. Fig.1 shows
marized in Ref.1. the model of the two-dimensional scram-
Design optimization in aerodynamics jet inlet geometry which is composed of 6
and propulsion has always been of great variables and 5 fixed parameters given in
interest and has, therefore, received Table 1 and 2. The geometrical model has
much attention11-18. The direct result of two parts23.
improvement in computer technology is The first part is the inlet (compres-
the growing interest in automated design sion system). The inlet shown in Fig.1
techniques of both single and multi ob- has three ramps for multiple compres-
jective optimization19-22 aimed at achiev- sions. The angle of first ramp, second
ing better design in a shorter period of ramp, and third ramp are indicated by
time than classical methods, i.e., by hand θ1, θ2, and θ3 , respectively.
or extensive analysis. The second is the isolator, constant
The design of an efficient compression section area part. Since the flow speed
system for an airframe-integrated decreases and the pressure rises during
scramjet engine is essential for accept- combustion at supersonic speeds - this
able performance of a hypersonic air- phenomenon is reversed at subsonic
breathing vehicle. The goal in the design speeds - a scramjet is fitted with an ad-
of a scramjet inlet is to define a minimum ditional element, the isolator, between
weight geometry that provides an effi- the inlet and the combustion chamber.
cient compression process, and provides The isolator's function is to isolate the
these characteristics over a wide range of combustion chamber before injection. It is
flight and engine operating conditions. supposed to prevent the pressure force,
The design of a scramjet inlet is compli- which rises during supersonic combus-
cated by the many constraints, both tion, from affecting the inlet flow via the
aerodynamic and mechanical, that are bordering walls. This could lead to
imposed on the inlet. Examples of aero- blocking the inlet. A phenomenon called
dynamic constraints include starting "a shock train" forms within the isolator,
limits, boundary layer separation limits, consisting of a changing succession of
and constraints on combustor entrance compression shock waves and expansions.
flow profiles. Examples of mechanical It is caused by the interaction of shock
constraints include variable geometry waves and the boundary layer of the iso-
flexibility and cooling system limits. The lator's wall and causes a (beneficial) fur-
design of scramjet inlets is influenced ther rise in flow pressure.
greatly by vehicle and flight constraints. The geometrical constraints are lin-
This paper focuses on the implemen- ear inequality constraints, and a nonlin-
-2-
American Institute of Aeronautics and Astronautics Paper 2005-3207
13th AIAA/CIRA International Space Planes and Hypersonic Systems
and Technologies Conference, 16-20, May, 2005, Capua, Italy
ear equality constraint. Besides the ob- of Wilcox [Ref.23] to generate the inflow
vious geometrical constraints such as, profile. Program EDDYBL is a program
suitable for computing properties of
0 ≤ θ1 ≤ θ2 ≤ θ3 (radians) laminar, transitional and turbulent
boundary layers. The program embodies
0.2 ≤ X1 ≤ X 2 ≤ 1.25 (m) a wide variety of turbulence models
0.2 ≤ Xe ≤ 1.25 (m) ranging from mixing-length oriented al-
gebraic models to a complete stress
the following constraint is required: transport model. The freestream condi-
tions at the entrance to the inlet are de-
(X1 − 0.2) tan θ1 + (X 2 − X1 ) tan θ2 scribed in Table.3. The condition corre-
sponds to the RJTF experimental condi-
+ (1.25 − X 2 ) tan θ3 = 0.15 (m) tion in the flight condition of Mach 8.
-3-
American Institute of Aeronautics and Astronautics Paper 2005-3207
13th AIAA/CIRA International Space Planes and Hypersonic Systems
and Technologies Conference, 16-20, May, 2005, Capua, Italy
individuals are not valuable. The objec- The control of the optimization loop
tive function is then discontinuous so implemented for designing the scramjet
that the computation of gradient has no inlet is provided by the optimizer. The
real significance. The method used in this optimizer generates candidate inlet de-
research is C code for Feasible Sequential signs described by five geometrical pa-
Quadratic Programming (CFSQP)19. rameters. These parameters are passed
CFSQP is a set of C functions for the to the analysis module which returns the
minimization of the maximum of a set of aerodynamic performance of this candi-
smooth objective functions (possibly a date inlet. Based on these simulation
single one) subject to nonlinear equality results, the objective function and some
and inequality constraints, linear equal- aerodynamic constraints are calculated.
ity and inequality constraints, and simple
bounds on the variables. In addition,
CFSQP contains special provisions for Single Objective Optimization
efficiently handling problems with many
sequentially related objec- In this single optimization program,
tives/constraints, for example discretized the objective function is the total pres-
Semi-Infinite Programming (SIP) prob- sure recovery coefficient η. The total
lems. CFSQP is an implementation of two pressure recovery coefficient η is repre-
algorithms based on Sequential Quad- sentative of the efficiency of the flow de-
ratic Programming (SQP), modified so as celeration performed by the inlet. To
to generate feasible iterates. In the first maximize the total pressure recovery is
one (monotone line search), a certain as same as to maximize the kinetic en-
Armijo type arc search is used with the ergy efficiency. The relation between the
property that the step of one is eventually total pressure recovery and the kinetic
accepted, a requirement for superlinear energy efficiency is given in Ref.25-26.
convergence. In the second one the same The total pressure recovery coefficient η
effect is achieved by means of a non- is obtained numerically as follows:
monotone search along a straight line.
The merit function used in both searches Pt exit
is the maximum of the objective functions η=
Pt 0
if there is no nonlinear equality con-
straints, or an exact penalty function if
nonlinear equality constraints are pre- where Ptexit and Pto are respectively the
sent. averaged total pressure in the exit plane
of the inlet and the free stream total
pressure. Therefore, the problem ad-
Automatic Optimization Methodology dressed in the following can be expressed:
-4-
American Institute of Aeronautics and Astronautics Paper 2005-3207
13th AIAA/CIRA International Space Planes and Hypersonic Systems
and Technologies Conference, 16-20, May, 2005, Capua, Italy
fields are conducted under the condition multi objective optimization algorithms
of Table 3. The initial total pressure re- to scramjet inlet design. The tools and the
covery coefficient η0 is obtained as 0.545. algorithms of our optimization study are
The flowfield is shown in Fig. 4(a). The summarized. Tools are the software
shock waves from the ramp collides with packages and the hardware resources
the cowl, and then the reflected shock required for our analysis. Algorithms are
from the cowl collide around the corner our overall design optimization method-
between the inlet part and the isolator ology and the multi-objective optimiza-
part. The separation region appears due tion methodologies applied in our study.
to the shock collision. Having more than one objective function,
The perturbation used by CFSQP is our problem is a multi-objective optimi-
approximating the gradients by finite zation problem.
difference is called udelta. Different val-
ues of udelta have been investigated, Minimize fi (x) (or Maximiza fi (x) )
showing its important influence on the i=1,…,m
optimization behavior. Moreover, udelta
equally affects the six design variables Optimization problems with more
used to parameterize the inlet geometry than one objective are known as
which are passed to CFSQP, where they multi-objective or multi-criteria deci-
are angle or distance. sion-making problems. Most of the design
The description of the five CFSQP problems in engineering have several
optimizations is provided in Table 4. The objectives. These objectives can conflict
setting parameters and the total pressure with one another. Therefore, rather than
recovery coefficient are also described in finding one single solution, a set of good
this Table. Depending on the runs, from 5 compromises are sought. The decision
to 13 line searches are performed during maker searches for the values of design
these optimizations. The maximum value variables, x= (x1,…,xk ), which optimize
obtained by these computations is ηmax the objective functions, fi, i=1,…,m, si-
=0.599. multaneously. Since a vector of design
The maximum value is higher than the variables that are acceptable by the deci-
initial value by 9.9%. The flow field is sion maker are searched, this type of op-
shown in Fig.4 (b). The improved con- timization problems is also known as
figuration is compared with the initial vector optimization. The set of solutions
configuration. Though the initial con- from which decision maker selects the
figuration has single ramp, the improved desired values of design variables is
configuration has triple ramp. The sepa- called the Pareto optimal set.
ration gets smaller and the flow around A set of points, x*, is Pareto Optimal
the corner of the improved configuration if there exists no feasible vector of deci-
becomes closer to the shock-cancel flow sion variables, x∈F , which would de-
than the flow around the corner of the crease any one of the objective functions
initial configuration. without causing a simultaneous increase
With this low value of udelta, CFSQP in at least one other objective function.
has explored a very limited region of the The vector x* corresponding to the solu-
design space around the starting point, tions included in the Pareto optimal set
which explains the poor result obtained are called nondominated, A design xa is
in terms of the total pressure recovery dominated by a design xb if fi (xb)≦fi (xa),
coefficient. for i=1, …, k, and for at least one value of
i, fi (xb)<fi (xa), where k is the dimension
of vector x. The set of nondominated de-
Multi Objective Optimization signs in the Pareto set. The plot of the
objective functions whose nondominated
The purpose of this study is to apply
-5-
American Institute of Aeronautics and Astronautics Paper 2005-3207
13th AIAA/CIRA International Space Planes and Hypersonic Systems
and Technologies Conference, 16-20, May, 2005, Capua, Italy
vectors are in the Pareto set is called the Therefore, the problem addressed in
Pareto front. the following can be expressed:
Multi-objective optimization problems
can be treated as single-objective prob- Minimize f1=ζ
lems by scalarizing the objective vector Minimize f2=D
into a single objective. There are different s.t. Geometrical Constraints
methods available in the literature for
scalarization of the objective functions. The trade-off (ε-constraint) method is
The trade-off (also known as the applied to solve above multi-objective
ε-constraint) method is applied to solve optimization problem, and the problem is
multi objective optimization problem of scalarized as follows.
scramjet inlet design.
The trade-off (the ε-constraint) Minimize f1= ζ
method is not biased towards convex por- s.t. f2=D≦ε,
tions of the Pareto set. It selects a single and Geometrical Constraints
objective function to minimize. All other
objective functions are restricted with The trade-off method is implemented and
user-specified values. The problem with several optimizations are performed with
many objectives is transferred into a sin- the gradient-based optimizer CFSQP. The
gle objective problem with an additional processes to get optimum solutions are as
set of constraints. The user defined ε same as the process described in previous
values are varied to determine the com- section. Different settings of CFSQP have
plete Pareto set. It is critical to specify a been tried. CFSQP, as a gradient-based
valid range of the ε values to avoid empty optimizer, requires a starting point to
solution sets. This method is able to as- initialize the search process. The initial
sociate with non-convex parts of the set of geometric parameters is the case of
Pareto front. The mathematical formula- single ramp (θ1=θ2=θ3). The initial total
tion is as follows: pressure loss is 0.455 and the initial drag
is 2985N. By repeating the processes to
Minimize fi (x) (or Maximiza fi (x) ) get the optimum solution, some optimum
s.t. fj (x) ≦ ε , j≠i solutions are obtained and shown in Fig.
6.
In order to implement this single ob-
jective design optimization methodology
to the multi-objective optimization prob- Conclusions and Futrure Work
lems, a modified form of the objective
function f(x) is used. In case of optimiza- An automated design of a two dimen-
tion based on the trade-off (ε-constraint) sional scramjet inlet has been achieved.
approach, one of the objective functions is The objective of the single optimization is
minimized while restricting the other. to increase the total pressure recovery
Fig.5 displays a flow chart of the optimi- coefficient while maintaining the geo-
zation algorithm based on the trade-off metrical contraction ratio. Computations
method. of the scramjet inlet are performed and
In this multi-optimization program, an automated optimal design of the
the objective functions are the total scramjet inlet is performed using six de-
pressure loss ζ (=1−η= f1) and the drag D sign variables and five fixed parameters
(=f2). The total pressure loss is introduced by using the gradient-based optimized
to solve the problem for convenience and method, SQP method. The scramjet inlets
the relation between the total pressure with the uniform inflow are improved,
recovery and the total pressure loss is and the optimized functions, that is, the
described above. total pressure recovery coefficient are
-6-
American Institute of Aeronautics and Astronautics Paper 2005-3207
13th AIAA/CIRA International Space Planes and Hypersonic Systems
and Technologies Conference, 16-20, May, 2005, Capua, Italy
fied by the global optimization method K., Tanno, H., and Takahashi, M., “Hy-
such as Genetic Algorithms. pervelocity Aerothermodynamic and
Several additional factors needed to Propulsion Research Using a High En-
be considered in future research for prac- thalpy Shock Tunnel HIEST,” AIAA Pa-
tical application of the results obtained in per 99-4960, Nov. 1999.
this study. These include 1) the effect of 7 Hasegawa, S., Sunami, T., Takaha-
three dimensional condition, 2) the sen- shi, M., Mochizuki, M., Miyajima, H.,
sitivity of the inflow boundary layer, 3) Itoh, K., Vogels, M.E.S., “The Virtual Test
several flight Mach conditions. Bed Environment at NAL-Kakuda Re-
search Center,” Parallel Computational
Fluid Dynamics 1999, pp.233-240, El-
Acknowledgement sevier Science, Amsterdam, 2000.
8 Hasegawa, S., Tani, K. and Sato, S.,
One of the authors (S. Hasegawa) “Numerical Simulation of Scramjet Inlets
stayed at Rutgers University as a visiting on a Vector-Parallel Supercomputer,”
scientist under the financial support by Parallel Computational Fluid Dynamics
JAXA (Japan Aerospace Exploration 2001, pp.159-166, Elsevier Science, Am-
Agency) and MEXT (Ministry of Educa- sterdam, 2002.
tion, Culture, Sports, Science and Tech- 9 Hasegawa, S., Tani, K. and Sato,
nology, Japan). The author would like to S.,“Aerodynamic Computation of a
thank all the people concerned for their Scramjet Engine on Vector-Parallel Su-
assistance. percomputers,” Parallel Computational
Fluid Dynamics 2002, Elsevier Science,
Amsterdam, 2003.
References 10 Hasegawa, S., Tani, K. and Sato,
T. Curran and S.N.B, Murthy, Vol. 189, M., Smith, D., and Knight, D., `` Auto-
Progress in Astronautics and Aeronautics, mated Redesign of the NASA P8 Hyper-
AIAA, Reston, VA, 2001, pp.159-222. sonic Inlet Using Numerical Optimiza-
2 Kanda,T., Hiraiwa,T., Mitani,T.,
tion'', AIAA Paper No. 96-2549, 1996.
Tomioka, S., and Chinzei, N., “Mach 6 12 Zha, G.-C., Smith, D., Schwabacher,
Testing of a Scramjet Engine Model,” M., Rasheed, K., Gelsey, A., and Knight,
Journal of Propulsion and Power, Vol.13, D., ``High Performance Supersonic Mis-
No.4, 1997, pp.543-551. sile Inlet Design Using Automated Opti-
3 Sato S., Izumikawa, M., Tomioka, S.
mization'', AIAA Paper No. 96-4142,
and Mitani, T., “Scramjet Engine Test at 1996.
the Mach 6 Flight Condition,” AIAA Pa- 13 Shukla, V., Gelsey, A., Schwabacher,
per 97-3021, July 1997. M., Smith, D., and Knight, D., ``Auto-
4 Kanda, T., Wakamatsu, Y., Ono, F.,
mated Design Optimization for the P2
-7-
American Institute of Aeronautics and Astronautics Paper 2005-3207
13th AIAA/CIRA International Space Planes and Hypersonic Systems
and Technologies Conference, 16-20, May, 2005, Capua, Italy
K., ``Automated Optimal Design of Two Solve It: Modern Heuristics'', Springer
Dimensional High Speed Missile Inlets'', Verlag, Dec., 1999.
AIAA Paper No. 98-0950, 1998. 22 Deb, K., `` Multi-Objective Optimi-
16 Knight, D., ``Automated Optimal zation using Evolutionary Algorithms'',
Design of Supersonic and Subsonic Dif- John Wiley & Sons, 2001.
fusers Using CFD'', ECCOMAS 2000, 23 Wilcox, D., “Turbulence Modeling
Barcelona, Spain, September 2000. for CFD”, DCW Industries, 2nd edition,
17 Knight, D., ``Application of Genetic Dec., 2002.
Algorithms to High Speed Air Intake De- 24 Kanda, T., Kudo, K., “A Conceptual
sign'', VKI Lecture Series Course on Ge- Study of a Combined Cycle Engine for an
netic Algorithms for Optimisation in Aerospace Plane”, AIAA Paper No.
Aeronautics and Turbomachinery, May 2002-5146, Oct. 2002.
2000. 25 Curran, E.T. Leingang, J., Carreiro,
18 Taskinoglu, E., Jovanovic, V., L., and Petters, D., “Further Studies of
Knight, D. and Elliott, G., Kinetic Energy Methods in High Speed
"Multi-objective Design Optimization and Ramjet Cycle Analysis,” AIAA Paper
Experimental Measurements for a Sub- 92-3805, July. 1992.
merged Inlet", AIAA Paper No. 2004-0025, 26 Heiser, W.H., Pratt, D.T., “Hyper-
-8-
American Institute of Aeronautics and Astronautics Paper 2005-3207
13th AIAA/CIRA International Space Planes and Hypersonic Systems
and Technologies Conference, 16-20, May, 2005, Capua, Italy
-9-
American Institute of Aeronautics and Astronautics Paper 2005-3207
13th AIAA/CIRA International Space Planes and Hypersonic Systems
and Technologies Conference, 16-20, May, 2005, Capua, Italy
Inlet Isolator
200mm
Xe
Air Flow
R3=20mm
R2=50mm
Y R1 =150mm
θ2 θ3
δ θ1
Origin X
X1
X2
L1=1250mm
L2=1250+200=1450mm
Generates Returns
Optimizer
Candidate inlet Aerodynamic Performance
-10-
American Institute of Aeronautics and Astronautics Paper 2005-3207
13th AIAA/CIRA International Space Planes and Hypersonic Systems
and Technologies Conference, 16-20, May, 2005, Capua, Italy
-11-
American Institute of Aeronautics and Astronautics Paper 2005-3207
13th AIAA/CIRA International Space Planes and Hypersonic Systems
and Technologies Conference, 16-20, May, 2005, Capua, Italy
3100
Solutions
3050
3000
2950
2900
0.4 0.42 0.44 0.46 0.48 0.5
-12-
American Institute of Aeronautics and Astronautics Paper 2005-3207