See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/262173573

Preliminary Aircraft Design Optimization Using Genetic Algorithms

Article  in  Journal of Scientific and Industrial Research · May 2015

CITATIONS READS
7 2,228

2 authors:

Imran Ali Chaudhry Ali Ahmed

University of Hail SIMAD University
40 PUBLICATIONS   419 CITATIONS    27 PUBLICATIONS   51 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Case Base Reasoning Application for Structural Analysis View project

Integrated Process Planning & Scheduling View project

All content following this page was uploaded by Imran Ali Chaudhry on 09 May 2014.

The user has requested enhancement of the downloaded file.

Journal of Scientific & Industrial Research
Vol. 73, May 2014, pp. 302-307

Preliminary Aircraft Design Optimization Using Genetic Algorithms

Imran Ali Chaudhry* and Ali Ahmed
National University of Sciences & Technology, Sector H-12, Islamabad, Pakistan

Received 06 August 2012; revised 09 October 2013; accepted 27 February 2014

Aircraft design is a highly nonlinear problem and inherently multidisciplinary activity that involves a large number of
design variables and different models and tools for various aspects of design. A spreadsheet based genetic algorithm (GA)
approach is presented to optimize the preliminary design of an aircraft. A domain independent general purpose genetic
algorithm is proposed to implement the optimization routine. Breguet range equation is used as the objective function for the
design evaluation. A total of sixteen design variables are considered in the optimization process. It has also been
demonstrated that the proposed approach can be adapted to any objective function without changing the optimization
routine. The model is applicable to commercial airliner as well as a multirole jet fighter. The proposed model has been
validated against known configurations of various aircraft.

Keyword: Genetic Algorithm (GA), Aircraft Design, Breguet Range equation, Spreadsheet

Introduction approach has been implemented in a spreadsheet

Aircraft design is a tedious and prolonged exercise environment using proprietary software as an add-in
involving complex interdependence of a wide range to the Microsoft Excel™ software.
of variables. The optimized values of these variables
or their best possible combination only yield an Design Process
effective, reliable and cost-effective aeroplane. The People involved in design never seem to agree
most efficient, reliable, fastest, lightest and cost where the design process begins. The designer thinks
effective aeroplane can be termed as an ideal aircraft, it starts with a new airplane concept. The sizing
however, aircraft design is a compromise of different specialist knows that nothing can begin until an initial
aspects because maximizing one capability would estimate of the weight is made. The customers,
render another to an undesired degree. Therefore, a whether civilian or military, feel that the design
healthy compromise between all the desired qualities begins with their requirements which are set by prior
is the ultimate goal of a designer. The constraints design trade studies. Thus, the concepts are developed
dictate the values of the design variables so their to meet requirements. So is the case of other two
ranges have to be kept in the realistic domain. parameters of design wheel. There are three major
Aircraft design is considered to be a separate phases of aircraft design1 conceptual design followed
discipline of aeronautical engineering which is by preliminary design then detailed design. The three
different from the other analytical disciplines such as major phases along with requirement of each phase
propulsion, aerodynamics, controls, and structures. are depicted in Figure 1.
An aircraft designer should be well versed in these The conceptual design of the aircraft starts with the
and many other specialties. Design is not only the study of many feasible configurations in some detail,
actual layout, but also analytical processes that are with the aim of achieving the mission requirements of
used to determine what is to be designed and how the the new aircraft; considering certain safety and
design should be modified to meet the requirements. operational criteria2-3. Conceptual design is subjected
This paper attempts to use genetic algorithms (GA) to an optimization process called the preliminary
to optimize preliminary aircraft design parameters to design. As a result one concept is finally chosen as the
maximize the range of the aircraft. The proposed best compromise for all requirements and specifications.
——————
Preliminary design process, also called ‘frozen
*
Author for correspondence configuration’, goes through somewhat complete
Email: [email protected] aerodynamic, flight mechanic and structural studies.
 CHAUDHRY & AHMED: PRELIMINARY AIRCRAFT DESIGN USING GENETIC ALGORITHMS 303

Fielding4 described that the most important stage of by combining several estimates described in Kroo7,
the aircraft design process is to define the correct set using the design following design variables: take-off
of requirements for future design; these requirements weight, wing span, horizontal tail span, vertical tail
are called design specifications. These inputs require span, mach no, seating capacity, aspect ratio, sfc,
inputs from a variety of disciplines and are dependent altitude, fuselage length, fuselage diameter, wing
upon various design / airworthiness standards. The sweep, angle of attack, ultimate load factor, thickness
design process evolves through various information by chord ratio of wing, taper ratio wing, thickness by
which include; airframe dimensions and shapes, chord ratio of horizontal tail, thickness by chord ratio
performance parameters, static and dynamic loads, of vertical tail, taper ratio horizontal tail, taper ratio
quality standards, certification criteria, and cost vertical tail. Upper and lower range of each variable is
constraints, etc.5. The various variable and constraints pre-defined according to the mission profile.
in the design process are interdependent thus for Although aircraft design depends on a large
efficient design workflow the relations ship between number of variables and essentially falls under the
various success of information must be known and the domain of multi-disciplinary optimisation. The
feedback loops are to be built into the design process6. fundamental aircraft parameters that are determined
Jayabalan et al. stated that aircraft design process during the preliminary design phase are: Aircraft
includes finding an aerofoil shape by testing, do a maximum take-off weight and wing reference area.
sizing and performance optimization and integrating it Based on Raymer1, it was determined that, in
together with the other parts of the aircraft, i.e. preliminary design, sixteen and fourteen design
payloads, propulsion systems, controls etc. variables respectively can be chosen for commercial
The aircraft design itself is an example of airliner and fighter aircraft to determine the wing
multidisciplinary design optimization (MDO) process reference area and resultantly the range. Based on
with a strong interaction between aerodynamic design these design variables and twenty three constants
and structural design. Literature review reveals that detailed spreadsheet model was built using drag
the design methodology though may not be very (including skin friction drag, form factor and
much innovative or novel but are not discussed in wetted areas), sizing and weight equations.
detail to provide leakage of idea or technology. The Torenbeek8 defines independent and dependent
only detailed methodology available pertains mostly design variables.
to conceptual design studies. Two different aircraft types i.e., a commercial
airliner and a multirole jet fighter, are considered in the
Problem Statement paper. Hence, two different models, one for each type
The problem statement for preliminary aircraft
design would be: determine the values of restricted
design variables such that the range of the aircraft R,
as given by Breguet range equation is maximized. The
Breguet range equation is given by7:
V L  Wi 
R ln
SFC D  W f  … (1)
Where
V = Cruise velocity
L = Airplane lift
SFC = Specific fuel consumption at cruise
speed and altitude
D = Airplane drag
Wi = Initial airplane weight
Wf = Final airplane weight
The objective is to maximize the range of the
aircraft corresponding to maximum value of L/D. The
right-hand side terms in equation 1 can be estimated Fig. 1—Aircraft design Phases
304 J SCI IND RES VOL 73 MAY 2014

is built for optimization. The chosen design variables Chromosome Representation

mentioned above make possible to compute all the Direct representation is used for the representation
necessary characteristics to evaluate equation 1. of the chromosome where each gene represents a
particular design variable. Thus for a commercial
Genetic Algorithms airliner the chromosome length would be of
Genetic Algorithms (GAs) belong to a class of twenty genes, which is actually equal to the number
search methods that are especially suited for solving of design variables. Similarly, for multi-role jet
complex optimization problems. GAs were first fighter the chromosome length would be of fourteen
introduced by Holland9. They transpose the notions of genes. Thus for each of the gene a number is
natural evolution to the world of computers, and generated between the defined range to find the
imitate natural evolution. A GA functions by best possible combination of values that gives the
generating a large number of possible solutions to a maximum value for the objective function given
given problem. Each solution is then evaluated in equation 1.
against a “fitness value” to determine the parents.
These solutions after crossover and mutation breed Reproduction / Selection
new solutions. Fitter solutions are more likely to In this research steady state reproduction as
reproduce as compared to less "fit". In successive reported in GENITOR GA22 is used, thus in each
iterations, best solutions (parents) are allowed to iteration only one worst performing organism is
produce new solutions (children). The worst members replaced instead of replacing the whole generation. In
of the population die off to make way for the fitter case of a steady state reproduction, all the genes are
individuals. A detailed introduction to GA’s is given not lost, as is the case in generational replacement
in Goldberg10. GAs have successfully been used in the where after replacement, many of the best individuals
aircraft design11-20. may not produce at all and their genes may be lost.
Steady-state reproduction is a better model of what
GA Implementation
happens in longer lived species in nature. This allows
In the present study GA is applied to the
parents to nurture and teach their offspring, but also
preliminary aircraft design in a spreadsheet
gives rise to competition between them. The value of
environment. The model was made on the basis of the
the objective function for a particular chromosome is
conceptual design by Raymer1. For genetic algorithm
a measure of its fitness.
implementation, we employ a commercially available
GA namely Evolver™21, that functions as an add-in to
the spreadsheet environment i.e., Microsoft Excel™. Crossover Operator
Uniform crossover is performed by the GA routine.
The aircraft design optimization model is developed
This means that instead of chopping the list
using spreadsheet’s built in functions. Figure 2 shows
of variables in a given scenario at some point
the spreadsheet-GA integration.
and dealing with each of the two blocks
The fitness/objective function value is passed on to
the GA component as a single cell value for the
evaluation of the design. Two models were made
essentially for a transport airliner and a multirole jet
fighter. The models were developed separately
according to the respective equations of both types.
The model follows a methodological approach where
a segment of the flight path or the mission profile is
taken i.e. the cruise segment. For this segment as
stated earlier, the range was taken as the objective
function to be the basic entity to be optimized. All the
relationships were built to compute the equation for
range, the Breguet Equation. Keeping in view the
historical trends, a total of 16 and 14 design variables
are used to develop the models for commercial
airliner and a multirole jet fighter respectively. Fig. 2—Spreadsheet-GA Integration
 CHAUDHRY & AHMED: PRELIMINARY AIRCRAFT DESIGN USING GENETIC ALGORITHMS 305

(called “single-point” or “double-point” crossover), Computational Results

two groups are formed by randomly selecting items to The simulations have been run on a Dual Core
be in one group or another. Traditional x-point 2.1 GHz computer having 1 GB RAM. For each
crossovers may bias the search with the irrelevant of the run, the following parameters have been used:
position of the variables, whereas the uniform population size = 65, crossover rate = 0.65, mutation
crossover method is considered better at preserving rate = 0.01, and stopping criteria = 80,000 trials,
schema, and can generate any schema from the which corresponds to approximately 1 min on a
two parents. In uniform crossover, instead of Dual Core 2.1 GHz computer having 1 GB RAM.
chopping the list of variables in a given scenario The initial model was run with restricted variables.
at some point and dealing with each of the two In the initial model, only eight variables namely:
blocks, two groups are formed by randomly selecting take-off Weight (Wo), wing Span (b), horizontal tail
items to be in one group or another23. Figure 3 shows span (bht), vertical tail span (bvt), mach no (M),
uniform crossover. altitude (h), aspect ratio (AR) and seating capacity (n)
In uniform crossover operator, mixing ratio or were considered for optimization. The model
crossover rate decides which parent will was verified against known configurations of
contribute each of the gene values in the offspring various existing commercial airliners. In the second
chromosome. This allows the parent chromosomes phase the model was revised to include additional
to be mixed at the gene level rather than the variables, thus increasing the number of variable
segment level. to 16. The additional variables were: fuselage
length, fuselage diameter, specific fuel consumption,
Consider the two parents in Fig. 3 which have wing sweep, angle of attack (AOA), ultimate load
been selected for crossover. Parent P1 has been factor, wing thickness to chord ratio and wing
coloured green while parent P2 yellow. A random taper ratio.
mask is generated corresponding to the crossover The actual range value and that calculated from the
rate. If the crossover rate is 0.5, approximately proposed model are quite close. The accuracy would
half of the genes in the offspring will come from increase as the number of design variables is
P1 and the other half will come from P2. Below increased as some of the values in the model have
the second parent is the random mask generated
been assumed to be constants for a particular type
corresponding to the crossover rate. The child is
of an aircraft. Table 1 gives the range and %age error
produced by taking bit from P1 if the corresponding
for different aircraft. After validation of the model,
mask bit is 1 or the bit from P2 if the corresponding
the simulations were run to find the optimized values
bit is 0. The colour of the child chromosome
of design variables. The corresponding value of range
represents the mixing of genes if the crossover
of the aircraft was 2876 nm. The model was then
rate is 0.5.
run for 30 runs and average value was calculated.
The average value of range after 30 runs was 2878
Mutation operator
nm. Similar exercise was carried out for a multirole
The purpose of the mutation is to ensure that
jet fighter. Optimized values for the range for the
diversity is maintained in the population. It gives
fighter aircraft calculated after one run was 817 nm,
random movement about the search space, thus
while the average range after 30 runs was 821.7 nm
preventing the GA becoming trapped in “blind
corners” or “local optima” during the search. The GA
in this research performs mutation by looking at each
variable individually. A random number between
0 and 1 is generated for each of the variables in the
organism, and if a variable gets a number that is less
than or equal to the mutation rate (for example, 0.06),
then that variable is mutated. The amount and
nature of the mutation is automatically determined
by a proprietary algorithm. Mutating a variable
involves replacing it with a randomly generated
value (within its valid min-max range). Fig. 3—Uniform crossover
306 J SCI IND RES VOL 73 MAY 2014

Table 1—Results of verification – Initial model

S. No Aircraft Actual Range (Nm) Model Calculated Range (Nm) %age Error
1 Airbus A318-100 3250 3059 6%
2 Boeing 747-100 5300 5170 4%
3 DC 8-32 4116 3630 12%
4 DC 8-63 CF 1913 1695 11%
5 Boeing 737-700 1585 1681 8%
6 Airbus A320-200 3000 2878 4%

Table 2—Comparison with Airbus A320-200 and Mirage 2000

Airbus A320-200 Mirage 2000
S No Variable Name Calculated Values Actual Value Calculated Values Actual Value
1 Take-off Weight 164109 lb 170000 lb 29024 lb 30420 lb
2 Wing Span 124.7 ft 111 ft 27.13 ft 29 ft
3 HT Span 50.2 ft 57 ft - -
4 VT Span 20.03 ft 22 ft 6.1 ft 7 ft
5 Mach No 0.79 0.8 2.2 2.2
6 Seating Capacity 195 180 - -
7 Aspect Ratio 10 9.8 1.8 1.9
8 SFC 0.00023 /hr 0.00024 /hr 0.00025 /hr 0.00025 /hr
9 Altitude 36840 ft 37000 ft - -
10 Fuselage Length 120 ft 123 ft 50 ft 47 ft
11 Fuselage Diameter 15 ft 13 ft 7 ft 7.6 ft
12 Wing Sweep 25º 25 º - -
Range 2878 nm 3000 nm 821.7 Nm 837 Nm

Comparison of Results Even with small number of design variables,

The optimized values found by the model were the results produced in this research were very close
compared with different available configurations. to the already available configurations of aircraft.
The closest configuration of a functional aircrafts to The spreadsheet-GA implementation has been found
the values obtained by the proposed was Airbus to be easy to implement and customizable to
A320-200 for the commercial airliner and Mirage any condition without changing the GA routine,
2000 for the multirole jet fighter. Table 2 gives which makes it a domain-independent approach.
the comparison of different variables with Airbus Furthermore, spreadsheet environment also enables
A320-200 and Mirage 2000 against the calculated and carrying out of what-if analysis. The approach is
actual values respectively. not a customization of the GA logic rather it only
modifies the model in spreadsheet without changing
Conclusion the actual GA routine.
This paper has attempted to use GAs for optimizing
the aircraft range for preliminary aircraft design for a
commercial airliner and a multirole jet fighter. In References
1 Raymer D P, Aircraft Design: A conceptual approach,
preliminary aircraft design problem we define only 4th edn, Washington D.C.: American Institute of Aeronautics
the major aircraft characteristics. After this we move and Astronautics, 2006.
to the detailed aircraft design. Results in the current 2 Martinez-Val R & Hernandez C, Preliminary design of
research indicate that by increasing the number of a low speed, long endurance remote piloted vehicles
variables, we can increase the accuracy of the model. (RPV) for civil applications. Aircr Des; 2 (3) (1999)
167-182.
The approach has demonstrated that it is very easy to
3 Jayabalan N, Horng L J & Leng G, Reverse engineering and
customize the solution for any objective function aerodynamic analysis of a flying wing UAV, Republic of
without disturbing the logic of the GA routine, thus Singapore Air Force’s Aerospace Technology Seminar,
making it a general purpose solution approach. Singapore, Feb 2005.
 CHAUDHRY & AHMED: PRELIMINARY AIRCRAFT DESIGN USING GENETIC ALGORITHMS 307

4 Fielding J P, Introduction to aircraft design. Cambridge GECCO–99: Proc Gen Evol Computat Conf (Banzhaf W et
University Press UK, 1999. al. eds), Orlando, Florida, USA, Morgan Kaufmann, July
5 Gundlach J F, Multidisciplinary design optimization of 1999, 1657-1665.
subsonic fixed-wing unmanned aerial vehicles projected 15 Jun C L & Song A H, Conceptual / Preliminary aircraft design
through 2025, PhD Dissertation, Virginia Polytechnic using genetic algorithm and fuzzy mathematics, In 37th AIAA
Institute and State University, Blacksburg, Virginia, 2004. Aer Sci M&E, AIAA-1999-0113, 11-14 January 1999.
6 Sobester A & Keane A J, Multidisciplinary design optimization 16 Ng T T H & Leng G S B, Application of genetic algorithms
of UAV airframes, in 47th AIAA/ASME/ASCE/AHS/ASC Str to conceptual design of a micro-air vehicle, Eng Appl Artif
Struct Dy Mat Conf, Newport, Rhode Island, USA, AIAA 2006- Intel, 15 (2002) 439-445.
1612, 1-4 May 2006, 13 pages. 17 Bandte O & Malinchik S, A broad and narrow approach to
7 Kroo I, Aircraft Design: synthesis and analysis, Desktop interactive evolutionary design – an aircraft design example,
Aeronautics Inc, Stanford, CA, 2003. in Proc Gen Evol Computat Conf - GECCO 2004 (DebK et
8 Torenbeek E, Advanced Aircraft Design: Conceptual Design, al. (eds.)), Lecture Notes in Computer Science Volume,
Technology and Optimization of Subsonic Civil Airplanes, 3103, 2004, 883-895.
John Wiley & Sons, 2013 18 Kroo I, Aeronautical applications of evolutionary design, in
9 Holland J, Adaptation in natural and artificial Systems, VKI Lecture Series on Optimization Methods & Tools for
University of Michigan Press Ann Arbor, 1975. Multi-Criteria / Multi-Disciplinary Des, November 15-19,
10 Goldberg D E, Genetic algorithms in search, optimization 2004, 11 pages.
and machine learning, Addison-Wesley Publishing 19 Ghorbany A & Malek M B, Airplane Conceptual Design
Company, 1989. Based on Genetic Algorithm, Mech & Aerospace Eng J, 1(1)
11 Rasheed K, Hirsh H & Gelsey A, A genetic algorithm for (2005) 101-114.
continuous design space search, Artif Intell Eng, 11 (1997) 20 Marta A, Parametric study of a genetic algorithm: using
295-305. an aircraft design optimization problem, Report
12 Bos A H W, Aircraft conceptual design by genetic/gradient- Stanford University, Department of Aeronautics and
guided optimization, Eng Appl Artif Intel, 11(3) (1998) 377-382. Astronautics, 2008.
13 Obayashi S, Multidisciplinary design optimization of aircraft 21 Evolver: the genetic algorithm super solver, Palisade
wing planform based on evolutionary algorithms, in Proc of Corporation, New York, USA, 1998
the 1998 IEEE Int Conf Syst Man Cyb, La Jolla, California, 22 Whitley D, GENITOR: a different genetic algorithm, in Proc
USA, 11-14 October 1998, 3148-3153. Rocky Mountain Conf Artif Intel, Denver, Colorado, 1988.
14 Parmee I C & Watson A H, Preliminary airframe design 23 Davis L, Handbook of genetic algorithms, New York, Van
using co-evolutionary multi objective genetic algorithms, in Nostrand Reinhold, 1991.

View publication stats