Beruflich Dokumente
Kultur Dokumente
Genetic Design
Algorithms
in Conceptual
L. Wells University
June
1996
Abstract
This and report outlines and and research optimization tool, applies outlines the GA accomplishments of rotorcraft. a procedure method in the for using acoustic area the of using the design genetic algorithms (GA) design of
for the
design
It discusses
genetic
algorithm conceptual
as a search
optimization
GA in the
helicopters,
to the
of rotors.
Introduction
design represents reflect Other the of rotor team one bias, decisions systems example experience, made often of this and relies type personal and on partially design preferences of design are based performance as complex, regarding qualitative features. decision-making. of the of the designer blades. for the arbitrary, and field on complicated rotor provides a basis decisions Selection Such and analytical final and made by
Conceptual members section(s) naturally sign team. for the timization design.
of a design
to determine
important
by designers weight,
numerical
or graphical,
can be characterized make other the rarely than final have the decisions
somewhat the
time-consuming. These designers, parameters nor ally included of a rotor in the is acoustic
operating generprocess.
system, analytical
of acoustics,
calculation,
of thumb," design
evaluation
conceptual
Consequently, one of the more important tends to receive little or no attention most. vide genetic technique into other appeared the ability problem[3]. (an integer It seems apparent, means and (GA) then, that design a logical for making highly-complex may provide developed computer engineering discrete, selection
of a rotor system--its phase of design which for but (such a tool which which can also and of such
noise signature-can affect it the can easily not only proThe tool. its way of GA of blades choice is incorporate
exists
goal functions a basis genetic programs[l]. problems[2]. integer, of airfoil and section algorithms
as low-noise mid-1960's
scientists
The tool.
as an optimization
advantages or number
variable)
can be handled
by the
(a continuous variable). Because the genetic algorithm is not calculus-based, as a global optimizer of highly-non-smooth and discontinuous functions. The numerical because optimization of their procedures reliance on the cannot readily handle non-continuous derivatives. computation of numerical
or functions
Genetic
algorithms
Algorithms
mimic the patterns concept methodology presented of natural by selection and and reproduction fittest" has into since an as an been accepted characteristic optimization expanded search by and
Genetic algorithm
of biological Goldberg[5] optimization into binary combination An some these jective high erations. exchange includes and not initial
This The
of the
in work
developed
technique. Design variables form the "genes" strings. These strings are then concatenated of variables generation number is decoded a fitness value are search value The which represent an individual placing of the This problem. be used The string values through is created from the by randomly
of a given design and are mapped to form the "chromosome" for a design "l"s point. and "0"s variables along the chromoof these From to the obwith genprocess designs, a trait in each
design fitness
individuals
to each
is analogous
optimization to survive in GA
a structured
probabilistic
information
among the members of the population of design crossover, where a "child" design inherits traits where a bit in the chromosome in engineering design string
mutation,
is changed, numerical
parent. efforts
employ
optimization
to improve
upon the results of standard design processes. The genetic algorithm does not replace these numerical methods. Instead, it provides a tool for searching a larger and more difficult design a global require complex space than is easily of highly evaluations handled and design by the calculus-based and discontinuous a rapid procedures. functions search they GA continuity. in the can be used it does as not optimizer gradient and non-smooth space, do not the GA search because
for functional
For a highly direction of the a search of a given GA in or when uniform result hard[5]) Crossley[6] a genetic of the design a design design means the direction
multi-modal
global optimum. Because genetic at a local design, can be character, a problem from outlines algorithm global space optimum. therefore, and, prone but
algorithms In fact, a GA
gradients,
complete
guarantee
In addition,
near-optimum is difficult especially a large technique, several can usually [8] describe rotorcraft
individuals. initial
Premature algorithm
sampling
randomly-generated As described the population the in conceptual Once combination genetic from
for resolving
optimum. is large
contains
search
is narrowed,
continuous
along
of improving the performance of the basic algorithm. The remainder of the report will focus on the approachto and results from using GA in designing rotors for low acoustic signature.
3
The
Genetic
design
Algorithms
involves of those both
in the
a means noise
Design
for analyzing In the case hand,
of Low-Noise
a candidate here, based the system performs which variables
Rotor
design the on and the
Systems
a method component known and design to the most of the mutation for
candidates. of the
presented
analysis
level
on the knowledge
other
crossover themselves.
individuals
generation
part, the procedures operate the fitness of each individual selected values for the design
3.1
Acoustic
Fitness
design the phase, system. level the analysis other used design tools used must provide many realistic potential cannot is made blade values design for the points high
In the conceptual performance must simply the because be considered, fitness Predicting
hand,
because
in evaluating point
performance
be too
in determining (includ-
of a given
thickness
planform
ing twist), airfoil section, and tip Mach derived from BVI, demands a description tion the et. that on the rotor blades. loading shown fairly Modern prediction good results practice are followed authors by an aerodynamics have calculation,
number. Loading of the complete, in rotor usually used using noise with the
including that loading distribua wake tool. code. [9] and analysis from Several Spiegel, good, for Results
prediction to a noise
only
conceptual of design
applications must
cannot
be analyzed
a reasonable
philosophy was followed rotor blade. The analysis an attempt levels. The as the Reference ness function
in developing the fitness evaluation procedure for the low-noise utilizes a combination of "exact" and approximate techniques in accurate, noise and performance initial stages of the design as long are correctly predicted.
to obtain reasonable, though not completely reduced level of accuracy is tolerable at the noise levels among the the various rotor designs [11] describes for use with
relative
prediction methodology used in preparing the acoustic fitgenetic algorithm. The thickness and harmonic loading
noise are computed exactly, with the aerodynamicpressurescalculated using a blade element/panel method combination. Blade-vortex interaction (BVI) noise is estimated using a formula introduced by Hardin[12].
3.2
The which
Fitness
actual fitness
Function
of an individual by f = giNs + A2NBv: + _
i
is determined
through
the value
of an objective
function
can be described
C_(max[O,g,(x)])
In the above equation, thickness and loading, coeiTicients penalty it meets any chosen function which
Ns and NBVl represent and by the blade-vortex the noise sources is used constraints. to ensure,
the measures of noise produced interactions, respectively. appropriately. that the The third rotor is not
to weight
term
for example,
imposed
3.3
Genetic
Algorithm
Methodology
crossover on these basic ge-
The genetic algorithm-based and mutation. The code operators netic and with algorithm A tournament
design code utilizes the three basic operators--selection, is customized for the current application with variations order" operators which alter the performance of the
"higher
convergence (as described above). individuals which will contribute to succeeding are selected without replacement from the and that with the better is evaluated in the same
In this
individuals
individuals are evaluated for fitness, for the crossover step. A second pair
manner, and the superior individual from that inal two. The process continues until the new roulette-wheel dividuals entire population. or rank-order rather This avoids selection than problems at a time comparing the with
pair is '_nated" with the better of the origgeneration is filled. Unlike more traditional tournament fitness scaling selection which can compares lead two inthe an is relative fitness of one individual against to premature
methods[5],
convergence. As an additional "elitist" tournament selection retained The gence. and design When it replaces code can the employ
hedge against premature convergence, in which the best individual from the least fit member a further in the best of the scheme individual next to stall is noted generation. the onset after
convera G-bit
no improvement
five generations,
improvement[5] method, a gradient-like current best individual in the population, a time the ment, patterns in the string. in effect, into the forces The single bit change
bitwise improvement approach, the G-bit improvement routine producing of the best the current best fitness string The
is used. For the varies one bit at is entered G-bit new into string
population,
replacing population.
member on the
generation. in order
improve-
individual
to introduce
Generation 1 2 5 10 15
Blades 4 5 5 5 5
Taper .4 .4 .8 .8 .8
Table
1: Evolution
of Low-Total-Noise
The and
To use the genetic algorithm, the seven design variable list contains one discrete variable---the of blades. loading, are The other variables, with length the Better design blade disk continuous, quantities. string a total the resolution
variables are coded airfoil section--and blade of the taper, binary However,
the number
twist,
determining
in these
resolution space
of the continuous
by increasing
for each
variable.
of 27 bits,
of 22T, or 134,000,000
indi,_iduals
represented.
Results
[11] shows results for GA design of a low-thickness-noise rotor, a low-loading-
rotor, a rotor system with low thickness and loading noise, and a low-total-noise Included here are results for the low-total-noise rotor system, which includes the loading terms design and in the variables blade-vortex objective along equal with interaction function the were power for a nominal noise chosen case. required components to make Table to hover in the the BVI 1 shows fitness and the function. the steady of coefficients rotor
thickness, loading
approximately
evolution generations.
for several
(Note that "DL" refers to the disk loading.) that with the lowest noise signature of the generations evolve. According normally tip the velocity strength before to the allows of the any presented the real blade results, design. sections tip vortex. the because of excess CPU time
The individual described in the table population. This case was terminated though the population tip speed by the coefficient, should resulting rotor fact
represents after 15 to
would higher
expect
in an optimized trailing
be driven solution
at lower
reducing to evolve
further
conclusions
5
The design
Conclusions
research
and
Further
the feasibility The
Study
of utilizing genetic a relatively method procedure, new methodology for the discrete, seems
demonstrates rotor
system.
algorithm design
integer,
variables
particularly The paper examines The guaranteed design not the variables noise
appropriate for use in the selection of rotor design and operational parameters. presents solutions for the design of rotors with lowered acoustic signatures, and, the results trade-offs provide to be optimal all of them required several in the are in order low-noise sense "good" that to achieve rotor solutions. blade a low-noise solutions. the very the design. Though lowest none noise of these rotor solution is is lowest-noise in the design a lowterm
it represents
space,
In general,
lowest power solution, although the noise-lowering trends in some of the decrease the power required to hover as well. If it is desired to design low-power rotor, the fitness function can be defined to include a power In this way, the genetic The results presented of a multi-objective the speed fitness of the calculation, GA
and
with an appropriate coefficient. multi-objective design problems. easily rotor incorporate systems. To low noise improve
algorithm is well-suited to handle here indicate that the method will fitness method, in particular under function work in the continues design of the of in the
development
of an improved
harmonic loading noise and the BVI noise. References [6], [7], [8], and [11] were prepared this grant.
all or in part
References [1] Holland, J. H., "Genetic K., algorithms," Scientific robust American, optimization pp. 66-72, tool," July AIAA 1992. 93-0315,
[2] KrishnaKumar, January [3] Lin, and C.-Y. integer 1993. and
"Genetic
algorithms-a
Hajela,
P., "Genetic
algorithms
in optimization vol.
problems
with
discrete 1992.
variables,"
Engineering in Natural
Optimization,
Systems.
Algorithms
Optimization
Learning.
Wesley,
[6] Crossley, W. A., Using Genetic tual Design of Rotorcraft. PhD [7] Crossley, gorithms W. A., 1995. Wells, V. L., design
Algorithms as an Automated Methodology thesis, Arizona State University, August and Laananen, systems," D. H., "The potential
for conceptual
of rotor
Engineering
Optimization,
pp. 221-238,
[8] Crossley, W. A., genetic algorithms Vertical [9] Gallman, interaction Lift
Regulski, J., Wells, V. L., and Laananen, D. H., and sizing codes for conceptual design of rotorcraft," Design Conference, Y. (San Francisco), January
"Incorporating in AHS/NASA
Aircraft
1995.
[10] Spiegel,P. and Rahier, G., "Theoretical study and prediction of bvi noise including closeinteractions," in Proceedings of the Inter_mtional Technical Specialists Meeting
on Rotorcraft Royal Acoustics and Rotor October Fluid Dynamics, American Helicopter Society and Aeronautical Society, 1991. of rotor blades using and Aeroacoustics of
[11] Wells, V. L., Han, A. Y., and Crossley, W. A., "Acoustic design a genetic algorithm," in A GARD Symposium on Aerodynamics Rotorcraft, [12] Hardin, noise," pp. 35.1-35.10, October 1994.
of blade/vortex 1987.
interaction