REDUCTION OF KINETIC MECHANISMS FOR FUEL OXIDATION THROUGH GENETIC ALGORITHMS

roomycankerblossomΤεχνίτη Νοημοσύνη και Ρομποτική

23 Οκτ 2013 (πριν από 4 χρόνια και 20 μέρες)

87 εμφανίσεις

REDUCTION OF KINETIC

MECHANISMS
FOR FUEL OXIDATION
THROUGH
GENETIC
ALGORITHMS


J
.
J
.
Hernández
,
R.

Ballesteros
,
J
.
Sanz
-
Argent

JuanJose.Hernandez@uclm.es, Rosario.Ballesteros@uclm.es, Josep.Sanz@uclm.es
;

Grupo de Combustibles y Motores

,
Universidad de Casti
lla
-
La Mancha
;

Av Camilo José Cela s/n, 13071 Ciudad Real, Spain


Computer modeling has become a very important tool for new internal combustion engine developments since it
permits to save money and time
,
by
reducing the
need of
experimental work in test
bench
. Among all the
phenomena which take place during the fuel combustion (atomization, mixing, turbulent motion of the mixture,
etc.), the chemical kinetics of the fuel oxidation
has

increasing importance due to new concepts of combustion
(HCCI, CAI, etc
)
which have appeared during the last years.

For
both

reasons, the availability of a reliable
chemical model
describing

the oxidation of commercial fuel
s

would be a very useful tool for the design of new
engines. Present chemical kinetic models for fuels i
nclude thousands of species and reactions, size which can be
managed
in case of

simple physical model simulations but
would be
too time consuming
in case of

more detailed
physical models as CFD codes
. For this reason, reduction methodologies are frequently

used to shorten the size
of the model while maintaining its fundamental features. In this work the use of genetic algorithms for the
reduction of skeletal kinetic schemes is presented.

Genetic
algorithms is an optimization technique based on evolutionary
principles.
A possible solution is codified
by

a
chromosome

made of
logical genes, being 1

(true)

and 0

(false)

its possible values.
A initial population of
potential solution
s

for a problem is randomly generated and the goodness of each
individual is esti
mated using a
fitness function. The
higher

the fitness value, the
greater the
possib
ility of an individual
of becoming a parent for
the next generation of solutions. Following populations are obtained from parents of the previous one by using
genetic opera
tors such as crossover, mutation or elitism.

In
our

case
,

a chromosome represents all the reactions of a kinetic model and a
gene
the presence of a
determined reaction in a reduced kinetic mechanism. The size of the reduced kinetic model is kept constant
for a
determined generation (all the chromosomes of a population have the same number of genes with a value of 1).
The
fit
ness

(expression 1)

of each chromosome is evaluated using two macroscopic variables
from

a kinetic
simulation
: the ignition delay (

)

and the temperature of combustion

(Tc)
.
This values are estimated from constant
v
olume

simulations solved using CHEMKIN 4.0 libraries and DVODE solver.



(1)

The crossover operator considers all the genes recessive, this mean
ing

that

a descendant inherits its parent
value
s

only if both parent value
s

are

i
dentical, otherwise, the descendant gene value is obtained in a random
manner, as shown in Figure
1
. Since a crossover operator defined this manner introduces great variability from
o
ne generation to the next one, no mutation operator is included. Operators elitism and superelitism are also
defined to improve convergence to the best solution. Elitism is defined as the number of chromosomes which are
included in the next generation, bei
ng the ones with greater fitness value. Superelitism operator search
es

inside
the population the
genes

that
are true

in every acceptable chromosome (defined as the individual which reach
es

combustion for the defined simulation) and
fixes this gene as true
in every chromosome of the next generation.

A variable size approach is adopted c
onsidering that chromosomes with few number of true genes have higher
probability to produce
non acceptable
solutions. The
proposed methodology has been applied for the reduc
tion of
a methane kinetic model (GRIMECH 3.0) for its application to HCCI modeling and
ha
s

been proved to converge
to an optimal solution.


Figure
1
. Crossover operator