Sei sulla pagina 1di 8

14th Annual (International) Mechanical Engineering Conference May 2006

Isfahan University of Technology, Isfahan, Iran

GENETIC ALGORITHM APPLICATION


IN AIRCRAFT WEIGHT OPTIMAZATION
Hamid R. Saeedipour1 and P. Sathyanarayana 2
School of Aerospace Engineering, University of Science Malaysia (USM), Penang, Malaysia
E-mail: drhamid11@yahoo.co.uk

Abstract
Genetic Algorithm (GA) in aeronautics may be considered as an adaptive search method premised
on the evolutionary ideas of natural selection and genetic. In this paper, the GA concept in aircraft
weight optimization is designed to simulate process in an integrated aircraft system necessary for
minimum gross mass, specifically the one that follow the GA principles of survival of the fittest.
This paper describes the results of a research to broaden the application of an available genetic
algorithm for design optimization named GADO to weight optimization of a high-subsonic civil jet
transport aircraft. It was initially developed for minimizing take-off mass of a supersonic transport
aircraft. This process represents an intelligent exploitation of a random search within a defined
search space to solve the problem of minimizing the aircraft gross weight at take-off (GWTO). The
GA method has been performed well as the population converged to an optimal solution to the
GWTO dilemma. All of the genes have converged when 97% of the population sharing the same
value. Ten random populations of 120 points each were generated, and for each population the GA
is allowed to proceed for 12000 iterations.
Keywords: Genetic Algorithm, Design Optimisation, Aircraft Design, Minimum Weight.

INTRODUCTION
In a general understanding, Genetic algorithm (GA) is
a probabilistic search algorithm which is used to
solve search and optimization problems. In the past,
GAs have been widely studied, experimented and
applied in many fields such as engineering, finance,
economy, information systems, production/operation
and decision making (McCall 2004, Burhaneddin
2000, Davison & Rasheed 2000, Deboeck 1994, and
Whitely 1993).
The basic principles of GA were first laid down by
John Holland in 1975. Since then, fundamental
research and development have been conducted in
different topics in the field of spacecraft and aircraft
design, particularly in the last 15 years. As the
results of these research projects, practical GA
solutions are now applied in aircraft configuration
optimization, fuel-used analysis, engine performance
analysis, flight control systems and scheduling
aircraft landing times. As with the neural network
method, the GA looks highly promising for modeling
design problems in which the driving variables are
1

Corresponding Author, PhD, MSc, MBS, BSc, IT


Lecturer, PhD, MSc, BSc

highly nonlinear, chaotic and changing all the time.


The GA may also be more useful in aircraft design if
applied in conjunction with neural networks
(Saeedipour et al 2004, Baranowski et al 2004,
Saeedipour 2004, Chan & Vincent 2003, Chung et al
2003, Marta 2003, Choi 2003, Rasheed 2000,
Burhaneddin 2000, and Bosa 1998).
This GA procedure is used to find the optimum
design solution for any particular mission of a highsubsonic passenger aircraft by varying the aircraft
conceptual design parameters over a continuous
range of values. The take-off mass includes the
empty (dry) mass which represents a rough
approximation of the cost of making the aircraft, the
fuel mass which represents a rough approximation of
the operating cost of the aircraft and the payload
mass which represents a rough approximation of the
revenue by the aircraft. By taking all these into
consideration, it is noted that not only does the GA
solution provide an alternative method to solving
problems in aircraft design; it outperforms traditional
methods in some of the aircraft-weight dilemmas.

14th Annual (International) Mechanical Engineering Conference May 2006


Isfahan University of Technology, Isfahan, Iran

The current research is an extension to the genetic


algorithm methodology named GADO initially
developed by Rasheed (2000) for the design
optimization of a supersonic transport aircraft. He
introduced a fitness function based on a simulator
analysis of the aircraft take-off mass and a penalty
function to impose legal limits on the permissible
noise of the supersonic aircraft. In the past, the
GADO has been used to minimize the takeoff mass of
supersonic aircraft at the conceptual design stage
only. The GADO has been modified to study the
gross weight optimization of a high-subsonic civil jet
transport aircraft with two-wing-mounted turbofan
engines.
In this paper, the reference aircraft is a typical midsize civil jet transport aircraft designed to meet the
specification: 200 passengers, 5,000km range, a
supercritical wing with thickness to chord ratio of
0.11 and an aspect ratio of 8, a maximum lift
coefficient of 3.0 at landing and 2.4 at take-off, a wing
quarter chord sweep-back of 30o, an external fuselage
of 50m length and 4m diameter, a cruise Mach number
of 0.81, an altitude of 12km at cruise, a swept-back
low-wing and two wing-mounted turbofan engines
with axisymmetric exhaust nozzles. The adapted GA
procedure and the findings are briefly described
throughout this paper.

GA PROCEDURE
Basically, GA iteratively transforms a set (or
population) of parameters, each with associated
fitness values, into a new population of young
objects using for example, the Darwinian principle of
natural selection.
It uses operations that are
patterned after naturally occurring genetic
operations, such as crossover (e.g., sexual
recombination)
and
mutation.
Over
many
generations, natural populations evolve according to
the principles of survival of the fittest. In general, a
combination of natural selection in the nature permits
the development of living species that are highly
adapted to their environments. GAs exclusively
operate on similar principle. By copying such
processes several times, the GA is then able to
develop a solution to, for example, design problem, if
it has been suitably encoded. In general, the GA can
provide an alternative method to solving design
problems and it consistently outperforms some
traditional techniques.
When the GA procedure is applied to the aircraft
design, the standard GA proceeds for minimizing the
aircraft gross weight as follows (also see Conner
2003):

An initial population of individuals is generated at


random or heuristically. This population is a set of
solutions for the given design problem represented
by chromosomes. At every evolutionary step, known
as a generation, the individuals in the current
population are decoded and evaluated according to
some predefined quality criteria. Such criterion is
usually known as fitness function. To form a new
population or the next generation, individuals are
selected according to their suitability and fitness to
the design task. Then some or all of the existing
members of the current design solution pool is
replaced with the newly-created members. This action
motivated by a hope that the new population will be
better than the old one. The creation of new members
is done by operations called crossover and
mutation. The link between these GA concepts is
shown in Figure 1.
MUTATION

GENE
POOL

SELECTION

CROSS
OVER

OBJECTIVE
FUNCTION

Figure 1: Genetic Algorithm Diagram for Minimum


Aircraft Gross Weight.
Many selection procedures are used in the GA
process where individuals are chosen with a
probability proportional to their relative fitness in the
given design task. This ensures that the expected
number of times an individual is chosen is
approximately proportional to its relative performance
in the population. Thus, high-fitness individuals
stand a better chance of reproducing, while lowfitness ones are more likely to disappear gradually
during the computerized GA process of the design
optimization for minimum aircraft mass. This course
of action is repeated until some conditions are fully
satisfied based on a well defined target in number of
populations and improvement of the best design
solution in aircraft weight. If the GA is designed and
performed well, the population should converge to an
optimal solution in minimizing weight.
The
convergence requirement is satisfied when all of the
genes are converged with at least 95% of the
population sharing the same value.
By using the concept of encoding in the genetic
information of natural life, the aircraft as a system is
encoded in the form of genes, grouped into
chromosomes. The genetic algorithm then relies on
the potential solution being encoded into genes as

14th Annual (International) Mechanical Engineering Conference May 2006


Isfahan University of Technology, Isfahan, Iran

well. Each quanta of solution which is named as a bit


or byte, is represented as a gene. An entire possible
solution to the aircraft weight dilemma is introduced
here as a chromosome. The GA uses binary string
which is easier encoding to understand and to deal
with in the data representation. The chromosome
contains the required information about the design
solution which supposes to represent. In binary
encoding every chromosome is a string of bits, 0 or 1.
For example, chromosomes 1 and 2 represent
1101100100110110
and
1101111000011110,
respectively. Each bit in this string refers to a
characteristic of the solution and the whole string
represents a number.

a given number of program iterations. This is because


if a local optimum or a global optimum is not found
and failed to converge to any one point, the program
running can be terminated after certain time to avoid
unproductive computer usage. An alternative
stopping criterion has also been set up to stop the
program running after the best solution has not
changed over a specified number of iterations. This
usually occurs when the first optimum is found
satisfactorily. If non of the aforementioned criteria
have been meet during the process, the last stopping
criteria is activated when the average fitness of the
generation is the same or close to the fitness of the
best solution.

FITNESS EVALUATION

SELECTION

The fitness function in the GA method allows


assignation of a quality measure (i.e., quantitative) to
a specific set of design parameters or a specific
design configuration of the aircraft system. This
function for each individual is written as y = f(xi),
which can tell the designer how good or fit the
individual is, usually in the form of numbers. For
instance; if f(x1) > f(x2), then the design solution x1 is
better or fitter than x2. The fitness distribution is
given in the form of quantitative probability between
no
fitness
to
maximum
fitness
using

In the GA procedure for minimizing aircraft GWTO,


the best chromosomes should survive to create new
generation. There are several methods in selecting
the best chromosomes which can be used to solve
the design problem. The most appropriate ones for
weight optimization are Roulette Wheel Selection,
Rank Selection, Tournament Selection, Elitism, and
Schemas which are briefly described below (for more
information, see Conner 2003):

P( t) =

x 1t ,..., x it ,..., x tn

where n is the total number of design solutions (see


Belavkin 2002).
The fitness function is formulated for the design
problem to be solved. Given a particular chromosome,
the fitness function returns a single numerical fitness,
or figure of merit, which is largely proportional to the
utility or ability of the individual which that
chromosome represents. The design fitness simply is
the value of the fitness function which is used to
decide the probability that the particular chromosome
would be chosen to contribute to the next generation.
The fitness is normalized to generate a cumulative
probability distribution. This is then used in the
crossover process.
A stopping criterion is used in the evaluation
process to determine whether or not the current
generation and the best design solution found so far
are close to the overall optimum value required for
the minimum GWTO. Three stopping criteria are
used to deal with the following possibilities that may
occur during the computer program running to find
the best GWTO solution: a) the optimal (or global)
solution is found, b) the optimal solution is not
found, and c) a local optimum is found. These
standard stopping criterions stop the procedure after

Roulette Wheel Selection - Parents are selected


according to their fitness. The better the
chromosomes are, the higher are their chances to be
selected. All chromosomes in the population are
placed in a roulette wheel (sometimes called pie chart)
where the area size is proportional to its fitness value.
A random number is then generated which decides
the chromosome to be selected. Chromosomes with
bigger fitness values are chosen more often as they
usually occupy more space on the pie. After using
this technique, it is realized that there is a serious
problem as some fitness values are extensively
different from each other in the aircraft weight
optimization.
Rank Selection This method is applied to solve the
problem associated with the usage of the roulette
wheel selection when some fitness values are
extensively different. For example, if the best
chromosome fitness is 90% of the entire roulette
wheel then other chromosomes will have very little
chances to be selected. In the rank selection, it is
important first to rank the population and then every
chromosome receives fitness from the presumed
ranking. The worst chromosome will have fitness 1,
second worst 2, etc. and the best will have fitness N
which is the total number of chromosomes in
population. After this stage, all chromosomes have a
chance to be selected whatever their ranks are.
However, it is found that the rank selection technique

14th Annual (International) Mechanical Engineering Conference May 2006


Isfahan University of Technology, Isfahan, Iran

leads to slower convergence in the aircraft weight


optimization, because the best chromosomes do not
differ so much from other ones.
Tournament Selection - Any two better members of a
population are randomly picked and evaluated
against each other and the best one survives for the
next generation in the design solution. It is found
that it is difficult to apply this technique to the
aircraft weight optimization as some members have
strong multilateral links and interrelating effects on
aircraft empty weigh parts. Therefore, the GA
analysis would not be accurate enough by just
separating two members without taking into account
their side links to the weight of other aircraft
structure components.
Elitism By using this technique, the best
chromosome (or a few best chromosomes in some
cases) is directly copied to new population and the
rest is done in classical way. It is found that the use
of Elitism method rapidly increases the GA
performance in the aircraft weight optimization due to
the fact that it prevents losing the best found
solution.
Schemas - Some components of chromosomes which
are proved to be fit or encode good values are kept
intact for the next generation. Similar to what is
happening on the tournament selection; it is found
not being easy to apply this technique to the aircraft
weight optimization due to the fact that some
components have strong interrelating effects among
chromosomes. Therefore, it is nearly impossible to
separate their effects from each others properly in
order to choose the fittest component in each
chromosome.

CROSSOVER
Selection alone cannot introduce any new individuals
into the population as it cannot find new points in
the search space. These are generated by geneticallyinspired operators, of which the most well known are
crossover (sometimes called recombination) and
mutation. Both crossover and mutation are the most
important parts of the GA in finding the best design
solution as the algorithm performance is widely
influenced by these two operators. The probability
values are selected in a way that the crossover
process becomes the most frequently used, while the
mutation process is being resorted to relatively rare
use. This is because the mutation is considered as a
random operator to introduce diversity in the
population. This operator is applied to each member
of the gene pool by random choice based on the
probability principle.

In general, the crossover takes two individuals, and


cuts their chromosome strings at some randomly
chosen position, to produce two head segments as
well as two tail segments. The tail segments are
then swapped over to produce two new full-length
chromosomes which take over some genes from each
parent. The crossover process is not usually applied
to all pairs of individuals nominated for mating. A
random choice is made, where the possibility of
crossover being applied satisfactorily is between 0.6
and 1.0. If the crossover is not applied, offspring are
produced simply by duplicating the parents. This
gives a chance of passing on its genes without the
crossover disruption. However, only two ways for
doing the crossover in the GA procedure are
considered here for the weight optimization in aircraft
design, known as the 1-point crossover and the 2point crossover.
One-point Crossover: Two chromosomes are
randomly chosen to crossover a point which is
accidentally picked. Then, all genes after that point
are switched accordingly. For example, if the parent
chromosomes are given as x1 = 0111 and x2 = 1100, the
crossover of x1 = 01 | 11 and x2 = 11 | 00 generates
offspring chromosomes of x01 = 0100 and x02 = 1111.
Note that the sign | indicates the crossover point.
Two-point Crossover: In this case, two crossover
points are randomly chosen and the genes are
switched between the two crossover points. For
instance, the above parent chromosomes may
generate offspring chromosomes of x01 = 0101 and
x02 = 1110, if the crossover of x1 = 01 | 1 | 1 and x2 = 11
| 0 | 0.
By using the above crossover description, it is now
considered the GA with chromosomes consisting of
six genes xi = abcdef, and each gene is a number
between 0 and 9. The population of four
chromosomes are given as x1 = 435216, x2 = 173965,
x3 = 248012, and x4 = 908123.
The fitness function is f(x) = (a + c + e) - (b + d + f).
At the beginning of the process, the chromosomes
are sorted by their fitness. The one-point crossover
is conducted in the middle between the first and
second fittest chromosomes, while the two-point
crossover at points 2 and 4 has been applied for the
second and third fittest. The fitness of all the
offsprings is then calculated as follows.
Solution (1) - The fitness function is used; f(x1) =
-1, f(x2) = -11, f(x3) = 5, and f(x4) = 15 with the order
of x4 , x3 , x1 and x2 in a row.

14th Annual (International) Mechanical Engineering Conference May 2006


Isfahan University of Technology, Isfahan, Iran

Solution (2) One-point crossover in the middle


between x4 and x3 is applied; O1 = 908012 and O2 =
248123. Note that the bold numbers in O1 and O2
represent the genes affected by crossover. The twopoint crossover at points 2 and 4 between x3 and x1 is
applied. O1 = 245212 and O2 = 438016.
Solution (3) - The fitness function is applied here
again; f(O1) = 16, f(O2) = 4 , f(O3) = 0 , and
f(O4) = 4.
As a result, the average fitness of the population has
been increased. It is also shows that crossover is a
straightforward
procedure.
Any
two
new
chromosomes are moved into the next population or
generation with the hope that the next generation will
be better than the previous one; because only the
best chromosomes from the last generation were
used to create this new generation. The crossover
process will continue until the new generation is full.
To avoid any duplication, it is important to regularly
check each new chromosome to be sure that it does
not exist in the new generation (see McCall 2004).

MUTATION
Mutation involves only a single parent and result in
the creation of a single offspring. Mutation is applied
to each child individually after any crossover. It
randomly alters each gene with a small probability of
typically 0.001. Mutation provides a small amount of
random search to ensure that no point in the search
space has a zero probability of being examined.
Mutation also helps in adding diversity to the
population and avoiding local maximum. When a
randomly chosen gene is changed to some other
gene, the mutation is completed. For example, the
mutation of genes 3 and 5 in the offspring x1 = 1 0 1 1
0 0 1 results in x2 = 1 0 0 1 1 0 1.
Importance A GA will always be subject to
stochastic errors such as genetic drift. Even in the
absence of any selection pressure (i.e., a constant
fitness function), members of the population may still
converge to some points in the solution space. This
happens simply because of the accumulation of
stochastic errors. If, by chance, a gene becomes
predominant in the population, then it is just as likely
to become more predominant in the next generation
again.
If an increase in predominance is sustained over
several successive generations, and the population is
finite, then a gene may spread to all members of the
population. Once a gene has converged in this way
and becomes fixed; the crossover cannot introduce

new gene values easily. Therefore, the genes are


trapped in a local optimum only.
Due to the randomness of the process, there will be
chromo somes near a local optimum occasionally, but
probably none would be near the global optimum. As
a result, those chromosomes near the local optimum
will likely be chosen to crossover because they have
the better fitness in the local optimization. Then,
there will be very little chance of finding the global
optimum for the aircraft GWTO in this way. It is
found that the rate of genetic drift can be reduced
substantially by increasing the mutation rate (see
Walker & Garrett 2002).

CONVERGENCE
Convergence here is referred to the progression
towards increasing uniformity by genes when at least
95% of the population sharing the same value. The
population is said to have converged when all of the
genes have converged. As the population
converges, the average fitness will approach to the
best individual. When the convergence is achieved,
the GA process for the aircraft weight optimization is
terminated. Then, the fitness of the best and the
average individual in each generation likely rise
towards a global optimum.

STRENGTHS VS. WEAKNESSES


There are several methods in weight optimization in
which Genetic Algorithm and Gradient Method have
been successfully used in aircraft design. For
comparison purposes, the strengths vs. weakness of
these methods are given in Table 1.

GADO MODIFICATIONS
The current research is an extension to the GADO
(Genetic Algorithm for Design Optimization)
procedure which is initially developed by Rasheed
(2000) to minimize the take-off mass of a supersonic
transport aircraft. It is modified for the use in the
weight optimization of a high-subsonic civil transport
aircraft.
Each individual in the GA population represents a
parametric description of the whole aircraft system
and each parameter is taking on a value in a
continuous interval. The fitness of each individual is
based on the sum of a proper measure of merit
computed by the aircraft take-off mass as the
simulator. Three penalty functions are introduced to
impose legal limits on the aircraft permissible noise,
cost limits on the aircraft fuel consumption, and

14th Annual (International) Mechanical Engineering Conference May 2006


Isfahan University of Technology, Isfahan, Iran

structure limits on the aircraft geometry. Operators


are applied to individuals via a rank-based selection
scheme. A steady-state GA model is used in which
existing points in the population are replaced by
newly generated points via a replacement approach.
This replacement used here takes into consideration
both the fitness and the proximity of the points in the
GA population. The main objective was to select for
replacement a point that both has low fitness and is
relatively close to the point being introduced. Ten
random populations of 120 points each were
generated, and for each population the GA is allowed
to proceed for 12000 iterations.
Global parallelism in the GA method refers to a
concept in which each new individual is generated
after the evaluation of the previous one inserted into
the GA population. The global parallelism has the
potential for being close to linear speed-up in
domains with expensive fitness evaluations such as
those which arise in the field of aircraft design
optimization. The global parallelism has had a strong
impact on the behavior of the GADO in the past by
degrading the performance to make it more like a
random search. In contrast to expectations, it is
found that the degradation in performance is quite
limited for minimizing GWTO in aircraft weight
optimization procedure.

Total fuel including reserve and trapped fuel, WF =


233.7kN or 23.8ktons
Weight of one engine, WEN = 41.5kN or 4.2ktons
Engine weight, WE = 83.1kN or 8.5ktons
Standard weight of baggage and cargo, WBC =
41.4kN or 4.2ktons
Wing structural weight, WWING = 107.4kN or
10.9ktons
Tail structural weight (horizontal & vertical), WT =
23.9kN or 2.4ktons
Structural weight of horizontal tail, WHT = 15.8kN
or 1.6ktons
Fuselage structural weight, WFG = 99.6kN or
10.2ktons
Undercarriage weight, WU = 39.8kN or 4.1ktons
Engine weight over take-off weight, WEOWTO =
0.083
Total fuel over take-off weight at lift-off,
WFOWTO = 0.235
The results of the full design program with genetic
algorithm are presented to see the importance of GA
effects on the high-subsonic jet transport aircraft
design. The percentage changes to the weightrelated design parameters of the aircraft using GA
weight optimization technique are given in Table 2.
However, it is not possible to clearly describe in this
short paper the intensive methodology which has
been used to obtain the data in the table and to
compare with GA.

COMPUTER PROGRAM RESULTS


The convergent iterative computer program uses a
GA step-by-step method and the weight-related
output is presented at below after 35 iterations when
there are no further changes:
Wing loading at landing in emergency, WOSL =
4685.4N/m2

Wing loading at take-off, WOSTO = 5685.8N/m2

Wing loading at the beginning of cruise, WOSCR =


5523.6N/m2
Take-off thrust loading at lift-off point on a hot
day, WOFTOHD = 3.620
Take-off thrust loading at lift-off point on a cold
day, WOFTOCD = 3.113
Payload weight, WPL = 196.6kN or 20.0ktons
All-up weight at take-off, WTO = 996.2kN or
101.5ktons
Empty weight (zero fuel), WEM = 718.2kN or
73.2ktons
Structural weight, WST = 482.7kN or 49.2ktons
Landing weight, WEND = 775.7kN or 79.1ktons
Landing weight in emergency, WL = 820.9kN or
83.7ktons
Aircraft weight at the beginning of cruise, WCR =
967.7kN or 98.6ktons
Fuel used, WFUSED = 220.5kN or 22.5ktons
Reserve fuel, WFR = 13.2kN or 1.3ktons

CONCLUSIONS
One of the major advantages of this genetic algorithm
procedure is its flexibility. It makes relatively few
assumptions about the aircraft design problem being
solved. It also can deal with highly nonlinear
problems such as the aircraft structure, nondifferentiable functions such as engine performance
as well as functions with multiple local optima such
as minimum fuel consumption or best range. Due to
the primary drawback of this GA procedure probably
resulting from flexibility, the aircraft designer has to
come up with an encoding scheme that allows the GA
to take advantage of the primary building blocks. It is
important to make sure the evaluation function
assigns meaningful fitness measures to the GA. It is
not always clear how the evaluation function can be
formulated for the GA to produce an optimal solution
to aircraft weight dilemma. The current GA method is
also computationally intensive and convergence is
sometimes a problem.
One of the problems in the current GA method in
aircraft weight optimization is the genes from a few
comparatively highly fit (but not optimal) individuals
which may rapidly come to dominate the population.

14th Annual (International) Mechanical Engineering Conference May 2006


Isfahan University of Technology, Isfahan, Iran

This may cause the solution to converge on a local


maximum. Once the population has converged, the
ability of the GA to continue searching for better
solutions is effectively eliminated. The crossover of
almost identical chromosomes produces little that is
new. Only mutation remains to explore entirely new
ground, and this may perform a slow random search.
In spite of the above problem, it is concluded that
application of the current modified version of the
GADO results in faster optimization, better final
design, lower variance in final design quality and low
sensitivity to internal parameters and set-up. The
authors did not give much detailed analysis about
how to obtain the above conclusions in this paper to
limit the number of pages.

5)

6)

7)

8)
The future milestone in this research work is to find
out how this GA technique can be extended for other
applications in aircraft design and aviation system
management such as design integrity optimization,
best range, minimum fuel used, minimum required
thrust, engine sizing, optimum area and deflection of
aerodynamic control surfaces, best jet deflection in
thrust-vectored aircraft and shorter take-off and
landing runway lengths.

9)

10)
11)
12)

This paper proposes a GA -based design optimization


method for aircraft weight optimization and the
emphasis is put on the GA rather the design
optimization. The authors are aware of the problem
that the description about the solved optimizing
problem might be too curt, but it is all they could do
to summarize their intensive research in such short
paper. Moreover, there is the lack of comparative
experiments to verify the validity of the proposed
GADO and it is hoped this will be done in the further
stages of the research project.

REFERENCES
1)

2)

3)

4)

Baranowski, J., Bigham, B. & Wilson T., Design


Process Evolution, Boeing Commercial Airplane
Group, 2004.
Belavkin, R. V., Genetic Algorithm, Internet
Online Information, Middlesex University, UK,
2002.
Bosa, A. H. W., Aircraft Conceptual Design by
Genetic/Gradient-guided Optimization, Delft
University of Technology, Kluyverweg, Holland,
1998.
Burhaneddin,
S.,
Genetic
Algorithms,
Department of Industrial Engineering, Bilkent
University, Turkey, 2000

13)

14)

15)

16)

17)

18)

Chan. K. & Vincent. A., A General Aviation


Aircraft Conceptual Design using Conceptual
Design using Multidisciplinary Optimization
Multidisciplinary Optimization, MIT Press,
USA, 2003
Choi. S., Speedups for Efficient Genetic
Algorithms : Design optimization of low-boom
supersonic jet using parallel GA and micro-GA
with external memory, Department of
Aeronautics and Astronautics,
Stanford
University, USA, 2003.
Chung, H.S., Choi. S. I. & Alonso. J. J.,
Supersonic Business Jet Design using a
Knowledge-Based Genetic Algorithm with an
Adaptive, Unstructured Grid Methodology,
Stanford University, USA, 2003.
Conner. J, An Introduction to Genetic
Algorithms, The Pennsylvania State University,
USA, 2003.
Davison. B & Rasheed, K., Effect of Global
Parallelism on a Steady State GA, University of
Georgia, USA, 2000.
Deboeck, G. J., Trading On The Edge, Wiley,
USA, 1994.
Holland, J. H., Adaptation in Natural and
Artificial Systems, MIT Press, USA, 1975.
Joanne Walker & Simon Garrett, Genetic
Algorithm, Internet Online Information,
Middlesex University, UK, 2002.
Marta. A, Parametric Study of a Genetic
Algorithm using a Aircraft Design Optimization
Problem,
Dept.
of
Aeronautics
and
Astronautics, Stanford University, 2003
McCall. J., Genetic Algorithms for Modeling and
Optimisation, School of Comp uting, Robert
Gordon University, Aberdeen, Scotland, UK,
2004
Rasheed, K., GADO: A Genetic Algorithm for
Design Optimization, University of Georgia,
USA, 2000.
Saeedipour, H. R. , Md. Azlin, M. S., &
Sathyanarayana P, Data Link Functions &
Attributes of an Unmanned Aerial Vehicle (UAV)
System Using Both Ground Station & Small
Satellite, The 5th IAA Symposium on Small
Satellites for Earth Observation, Berlin, Germany,
4-8 April 2005
Saeedipour, H. R., Digital Design Applications
in Aerospace Industries, PLM User Conference
2004, MAWEA Industries, Shah Alam, Selangor,
Malaysia, 11 Sep. 2004
Whitely, D., A Genetic Algorithm Tutorial,
Technical Report CS-93-103, Colorado State
University, USA, 1993.

14th Annual (International) Mechanical Engineering Conference May 2006


Isfahan University of Technology, Isfahan, Iran

Table 1: Genetic Algorithm vs. Gradient Method in aircraft weight optimization

Genetic Algorithm

Gradient Methods

Task

Strength

Weakness

Using the information


about the gradient of the
function to guide the
direction of search.

Getting very close to local


optima in a relatively small
number of iterations.

If the derivative of the function


cannot be computed due to
discontinuity, the method may fail.
Inadequate for search space with
numerous peaks.
Depending upon the existence of
derivatives or well-defined slope
values. But, aircraft weight
optimization may ends up with a
search space with numerous peaks.

Works with a coding of


the parameter set, not
the parameters
themselves.
Searches from a
population of points in
the problem domain,
not a singular point.
Utilize probabilistic
transition rules based
on fitness rather than
deterministic one.

Having a better chance to


get to global optima in
weight optimization.
Solving problems with
discontinuities in the
solution space.
Solving highly multimodal problems.
Solving problems with
many variables.

Having a tendency to get only


moderately close to the local optima
in a small number of iterations
Non-deterministic; GA will not
always arrive at the same result for
the same input.
Relatively slower than some
optimization methods in aircraft
design.
Not guaranteed to find the absolute
best solution to the aircraft weight
problem.

Table 2: Percentage changes to the weight-related parameters using GA optimization technique.

Aircraft Weight Item

Reduction (%)

Gross Weight at Take-off, W TO

-1.8%

Total Engine Weight, W E

-1.3%

Total Fuel Weight, W F

-1.4%

Fuel Used; W f

-1.4%

Landing Weight after an aborted Take-off, W L

-1.8%

Landing Weight at the end of Flight; W end

-1.9%

Aircraft Weight at Cruise; W C

-1.8%

Wing Weight; W wing

-2.6%

Horizontal Tail Weight; W tail

-2.6%

The Remaining weight, W R

-1.8%

Potrebbero piacerti anche