Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
Vol
.
4
,
No
. 3,
J
une
2
0
1
4
,
pp
. 36
6~
37
1
I
S
SN
: 208
8-8
7
0
8
3
66
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
Performance Comparison between Classic and Intelligent
Meth
ods for Position Con
t
rol of DC Motor
N
a
v
i
d Mo
sht
a
g
h
i Y
a
zdani*
, A
rezo
o
Ya
zdani
Seqerloo*
*
* Departement o
f
Mechatronics
Engi
neering,
Un
iversity
of Tehran
**Departement
of Computer
En
gineer
ing, University
of
Tehr
an
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Dec 14, 2013
Rev
i
sed
Mar
24
, 20
14
Accepte
d Apr 6, 2014
Controlling
DC m
o
tors is m
a
inly don
e b
y
contr
o
lling
eith
er vol
tage or
fi
el
d
of their armatur
e
. Numerous methods have been proposed so f
a
r for this
purpose. Som
e
intel
ligen
t m
e
th
ods su
ch as
XCS
R
and m
achine le
arning
s
y
stems are used to contro
l po
sition
of a s
e
p
a
rat
e
l
y
ex
ci
ted
DC m
o
tor.
Having set outp
u
t position of the motor to its basic position, voltage of
arm
a
ture becom
e
s
zero and the m
o
tor s
t
ops working. Chara
c
ter
i
s
tic fea
t
ures
of the methods in this paper are re
sistance ag
ainst chang
i
ng friction and
m
o
m
e
nt of ine
r
tia
. Meanwhile
, tim
e to
reach
stabilit
y in th
is t
y
pe o
f
controll
ers is considerabl
y
low
e
r than that of
PID controller
with no
oscillations b
e
in
g obser
ved
in
th
e responses.
Keyword:
DC m
o
tor
Machine lea
r
ni
ng
PI
D con
t
ro
ller
XCSR
Copyright ©
201
4 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Navi
d
M
o
sht
a
ghi
Yaz
d
ani
,
Depa
rtem
ent of
M
ech
atroni
cs
,
University
of
Tehran
,
Teh
r
an
, Ir
an.
Em
a
il: n
a
v
i
d
.
m
o
sh
tag
h
i
@u
t.ac.ir
1.
INTRODUCTION
DC electric
m
o
tors ope
r
ate usi
ng
ba
sic conce
p
ts of electromagnetic su
c
h
that a classic DC
m
o
tor ha
s
a wind
ing
in
ro
t
o
r and
a
perm
an
en
t m
a
g
n
et in
stator.
A ro
tary switch
called
co
mm
u
t
a
t
o
r
rev
e
rses th
e
direction of ele
c
tric current t
w
ice duri
ng eac
h cycle. There
by, a flow
of c
u
rre
nt is created in the a
r
m
a
ture and
the electrom
a
gnet attracts a
n
d
repuls
es t
h
e
perm
anent
m
a
gnet
o
u
t
of
t
h
e
m
o
t
o
r. S
p
ee
d
of
DC
m
o
t
o
r
d
e
pen
d
s
on a
set
o
f
v
o
l
t
ages an
d c
u
r
r
e
nt
s pas
s
i
n
g t
h
ro
u
gh
wi
n
d
i
n
g
s
of t
h
e m
o
t
o
r
as wel
l
as m
o
t
o
r l
o
a
d
or
bra
k
i
n
g
t
o
r
que
. I
n
ot
h
e
r w
o
r
d
s
,
sp
ee
d o
f
t
h
e
DC
m
o
t
o
r i
s
depe
nde
nt
o
n
v
o
l
t
a
ge w
h
i
l
e
i
t
s
t
o
rq
ue i
s
depe
n
d
e
nt
o
n
cu
rren
t. Sp
eed is usu
a
lly contro
lled
u
s
i
n
g a v
a
riab
le
v
o
l
t
a
ge
whi
c
h i
s
ge
nerat
e
d
by
c
u
r
r
ent
passi
ng
t
h
ro
u
g
h
m
o
t
o
r wi
ndi
ng
or by
a va
ri
abl
e
sou
r
ce o
f
v
o
l
t
age. It
i
s
b
ecau
s
e th
is typ
e
of
m
o
to
r
can produce relatively
great
torque at low
spee
ds. T
h
e pe
rm
anent
m
a
gnets in the out
e
r
stator are replaced with electrom
a
gnets in othe
r
m
odel
s
of DC
m
o
t
o
rs kn
o
w
n as sol
e
n
o
i
d
m
o
t
o
rs. The ra
t
i
o
of spee
d t
o
t
o
rq
ue can be
al
t
e
red by
changi
ng
cu
rren
t of th
e so
leno
id
on
th
e
electro
m
a
g
n
e
t. If th
e so
le
no
id is co
n
n
ected
series with
th
e arm
a
tu
re win
d
i
n
g
, a
lo
w-sp
eed
h
i
gh
-t
o
r
q
u
e
m
o
tor will b
e
ob
tain
ed. On
th
e o
t
h
e
r
h
a
nd
, a
h
i
g
h
-sp
eed
low-to
rqu
e
m
o
to
r will b
e
achi
e
ve
d i
f
t
h
e
sol
e
noi
d i
s
c
o
nnect
e
d
i
n
pa
ra
l
l
e
l
.
It
i
s
eve
n
pos
si
bl
e t
o
rea
c
h
hi
g
h
er
spee
ds
by
re
d
u
ct
i
o
n
of t
h
e
field current t
h
ough at the
expe
nse
of a
sm
a
ller torque
. Application
of t
h
is technique ca
n lead t
o
m
a
ke
equi
pm
ents of
a m
echanical gearbox unn
ece
ssary.
Uni
v
ers
a
l DC m
o
tors a
r
e able
t
o
work with
eith
er
d
i
rect or
al
t
e
rnat
i
ng c
u
r
r
ent
.
T
h
ey
are
desi
g
n
ed
bas
e
d o
n
t
h
e f
o
l
l
o
wi
ng
pri
n
ci
pl
e:
When a s
o
l
e
noi
d DC
m
o
tor i
s
connected to
AC, the curre
nt change
s si
m
u
lt
aneou
s
l
y
at
bot
h sol
e
no
i
d
and arm
a
t
u
re wi
n
d
i
n
g. T
hus t
h
e
created m
echanical force
wil
l
always
rem
a
i
n
u
n
c
h
an
ge
d.
That
i
s
w
h
y
im
pedance
an
d
rel
u
ct
ance
m
u
st
be
taken int
o
account in
design to m
a
ke co
m
p
atibility with
alte
rnating curre
nt
. The fi
nally produce
d
m
o
tor ofte
n
has a
per
f
o
rm
ance l
o
wer
t
h
a
n
an e
qui
val
e
nt
DC
m
o
t
o
r.
Ad
vant
a
g
e
of t
h
e
uni
versal
m
o
t
o
rs i
s
t
h
at
AC
s
o
u
r
ce
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
4, No
. 3,
J
u
ne 2
0
1
4
:
36
6 – 3
7
1
36
7
can be use
d
on m
o
tors with typical specifications of
DC
m
o
tors, especially since these m
o
tors have a
considera
b
ly
high
operation t
o
rque as
well
as a
very
co
mp
act
design
f
o
r
h
i
gh
-
s
p
e
ed
op
er
ation
s
.
Th
e
on
ly
d
i
sadv
an
tag
e
of th
ese m
o
to
rs is related to t
h
eir
rep
a
ir and m
a
in
ten
a
n
ce
an
d also reliabilit
y issu
es
which
are
cause
d by
exi
s
t
e
nce of t
h
e c
o
m
m
u
t
a
t
o
r. The
r
ef
ore
,
t
h
ese m
o
t
o
rs can be
ra
rel
y
seen i
n
i
ndust
r
i
a
l
appl
i
c
at
i
ons.
Series-wo
und
m
o
to
r is th
e
m
o
st d
e
sirab
l
e altern
ativ
e
b
e
lo
w 1
KW
power with
fu
ll
lo
ad
sp
eed
of 4
000
to
1
0
,000
rp
m
.
Th
e series m
o
tor with
t
h
e cap
ab
ility o
f
u
s
i
n
g
DC or AC source show a
h
i
gh
sp
eed
at large rang
e
and a
hi
g
h
st
ar
t
i
ng t
o
r
q
ue (a
p
p
r
o
xi
m
a
t
e
l
y
500% o
f
t
h
e
nom
i
n
al
val
u
e
)
. It
i
s
t
hus acc
ou
nt
ed f
o
r a
n
i
d
eal
dri
v
e
i
n
di
f
f
e
r
ent
a
p
pl
i
cat
i
ons
ha
vi
ng
p
o
w
er
o
f
se
veral
t
o
se
veral
h
u
n
d
re
d
wat
t
s
. M
a
xi
m
u
m
sh
o
r
t-term
to
rq
u
e
li
m
i
ts
po
we
r of t
h
i
s
m
o
t
o
r t
o
400
% of t
h
e n
o
m
i
nal
val
u
e. Al
t
h
o
u
gh t
h
i
s
m
o
t
o
r i
s
very
si
m
i
l
a
r t
o
shunt
-w
ou
n
d
m
o
tor, its armature a
n
d field are c
o
nn
ected in
series and
n
o
t
i
n
p
a
rallel with
th
e li
n
e
.
Th
is feat
u
r
e allo
ws the
seri
es-
w
o
u
nd
m
o
t
o
r t
o
be desi
g
n
e
d
fo
r bei
n
g o
p
erat
e
d
wi
t
h
al
t
e
rn
at
i
ng o
r
di
rec
t
curre
nt
. M
o
reo
v
er
,
co
m
p
o
u
n
d
-wou
nd
m
o
to
r is used
in
so
m
e
ap
p
lication
s
, wh
ich
has bo
th
p
a
rallel an
d
series ex
cited
wi
n
d
i
n
g
s.
Wh
en
d
i
rection
of th
e series field
is o
ppo
site to
th
at o
f
p
a
rallel field
,
th
e m
o
to
r is called
ste
p
down
co
m
p
o
u
n
d
-wou
nd
m
o
to
r. Th
is k
i
nd
of m
o
to
r do
es no
t
h
a
ve co
mm
o
n
applicatio
n
s
du
e t
o
its sp
eed in
st
ab
ility
(weak
e
n
i
ng
of lo
ad
ov
erflow will redu
ce sp
eed sign
ific
an
tly). Mech
an
i
cal lo
ad
wou
l
d
n
e
ed
larg
er
to
rq
u
e
s
whe
n
e
v
er t
h
e
spee
d is raise
d
. The a
s
cending charact
eri
s
t
i
c of
spee
d-t
o
r
que i
n
dam
p
i
ng com
p
o
u
n
d
-
wo
u
n
d
m
o
to
rs will ca
u
s
e in
stab
ility
o
n
ce t
h
ey are attach
ed
to
su
ch
lo
ad
s. In
th
is case, sp
eed
will ad
d
to
to
rq
ue and
to
rq
u
e
will in
crease sp
eed
.
On
th
e o
t
h
e
r h
a
n
d
, if a series
ex
cited
field
co
n
t
ribu
tes to
a p
a
rallel ex
cited
field
,
th
e m
o
to
r will b
e
called
step
u
p
co
m
p
o
und
-woun
d
m
o
to
r.
It
is ev
id
en
t that th
e sp
ecificatio
n
s
of su
ch
a
m
o
to
r
are s
o
m
e
how
bet
w
ee
n t
h
ose
of
s
h
u
n
t
a
n
d
seri
es m
o
t
o
rs
.
Whe
n
t
h
e
m
o
t
o
r i
s
fe
d
by
co
nst
a
nt
v
o
l
t
a
ge
,
in
d
i
v
i
d
u
a
l ex
ci
ted
and
sh
un
t ex
cited
sp
ecificatio
n
s
o
f
th
is
m
o
to
r can
no
t be d
i
stin
gu
ish
e
d. At bo
th
con
d
i
tio
n
s
,
the excitem
e
nt winding is fe
d
by a voltage i
n
depe
ndent
from
the curre
n
t received
by
arm
a
ture.
The
r
efore, one
sou
r
ce
w
oul
d
be s
u
f
f
i
c
i
e
nt
t
o
be
use
d
fo
r
b
o
t
h
exci
t
e
m
e
nt
a
n
d
arm
a
ture
. Pe
rm
anen
t
m
a
gnet
DC
m
o
t
o
r
bene
fi
t
s
f
r
om
a pe
rm
anent
m
a
gnet
t
o
gene
rat
e
p
o
we
rs
u
p
t
o
20
0
h
p
fo
r
vari
ous
i
n
d
u
st
ri
es.
One m
a
jo
r a
d
vant
a
g
e
of
per
m
anent
m
a
gnet
DC
m
o
t
o
rs i
s
t
h
at
t
h
ey
d
o
not
nee
d
an
exci
t
e
m
e
nt
cu
rren
t. Th
is
will lead
to
energ
y
sav
i
ng
in co
m
p
aris
o
n
with
eq
u
i
v
a
len
t
m
o
to
rs h
a
v
i
n
g
wou
n
d
p
o
l
es
d
u
ring
typical lifetime of the m
ach
i
n
e. P
o
l
e
o
v
er
fl
ow ca
n
not
be
cont
rol
l
e
d i
n
these m
achines, so their s
p
ee
d a
nd
torque are c
ont
rolled by great
curre
nts of the ar
m
a
ture.
In
m
o
st cases, application of armature circuit cont
rail
i
s
adva
nt
age
o
u
s
ove
r exci
t
e
m
e
nt
ci
rcui
t
co
nt
rol
eve
n
i
n
m
a
ch
in
es
with
a woun
d
p
o
l
e. As a resu
lt, selectio
n
of
perm
anent
m
a
gnet
m
o
t
o
rs f
o
r i
n
d
u
st
ri
al
appl
i
cat
i
o
ns, w
h
i
c
h
need acc
urat
e co
nt
r
o
l
,
doe
s n
o
t
det
e
r
i
orat
e
anything else because rem
oving th
e indi
vidual source
of excitem
e
nt c
u
rrent is ofte
n known as a great
adva
ntage
.
T
h
e effect of constant
ov
er
fl
o
w
on op
erat
i
o
n char
acte
r
istics of a
DC m
achine is ve
ry simple. In
fact, th
e
o
p
e
rat
i
o
n
of
p
e
rm
an
en
t m
a
g
n
e
t DC
m
o
to
rs is very
si
m
ilar to
th
at
o
f
shun
t m
ach
in
es. App
licatio
n
o
f
p
e
rm
an
en
t m
a
g
n
e
t t
o
g
e
n
e
rate ex
cite
m
e
n
t
in
DC m
o
to
rs
with
in
sp
ecific rang
e
o
f
d
i
men
s
ion
s
en
tails so
m
e
econom
i
c benefits. In sm
all
m
o
tors
(bel
ow 70
mm diameters)
structure of the electrom
a
gnet is unable to
com
p
ete with
perm
anent m
a
gnets i
n
term
s of
price.
Ho
weve
r, i
n
l
a
r
g
e m
o
t
o
rs (
ove
r 1
5
0
m
m
di
am
et
ers)
econom
i
c analyses could rec
o
mmend t
o
us
e m
o
tors
with
electrom
a
gnets [1]. T
o
rque-s
peed c
h
aracteristic of
p
e
rm
an
en
t m
a
g
n
e
t
DC m
o
tor is an
alm
o
st straig
h
t
lin
e
between two
p
o
in
ts,
n
a
m
e
l
y
n
u
ll lo
ad
sp
eed
o
n
the
v
e
rtical ax
is and
static to
rqu
e
o
n
th
e
ho
rizo
n
t
al ax
is
. Perm
a
n
en
t m
a
g
n
e
t mo
tors are po
tentiall
y
m
o
re efficien
t
than electrom
a
gnet
m
o
tors si
nce t
h
ey
do
no
t sh
ow
an
y
f
i
eld
lo
ss.
Fur
t
her
m
o
r
e,
p
e
r
m
a
n
en
t
m
a
g
n
e
t m
o
to
r
s
provide great efficiency in a large
rang
e. Stru
cture of small p
e
rm
an
en
t
mag
n
e
t
m
o
to
rs u
p
to
sev
e
ral KW
i
s
co
m
p
letely d
i
fferen
t
fro
m
th
at o
f
p
a
rallel m
o
to
rs. Pe
rm
a
n
en
t m
a
g
n
e
ts
with
d
i
rectional ferrite g
r
ai
ns are
devise
d in the
s
e
m
achines. They are m
a
g
n
etized be
fo
re
being
placed
in the stator.
For a gi
ven
nom
i
na
l
powe
r, a
r
m
a
tu
re
of
suc
h
m
o
tor m
u
st be usually co
nside
r
ed a little larger tha
n
shunt
-wound m
o
tor
because
fl
u
x
de
nsi
t
y
of
t
h
e ai
r ga
p ac
hi
eve
d
by
a
fe
rri
t
e
m
a
gnet
is co
nsid
erab
ly smaller th
an
th
at o
f
woun
d
po
les. A
30
% re
d
u
ct
i
o
n
i
n
wei
g
ht
o
f
t
h
e m
achi
n
e ca
n
be
obt
ai
ne
d
by
re
pl
aci
n
g
e
x
ci
t
e
m
e
nt
coi
l
s
o
f
t
h
e
m
o
t
o
r wi
t
h
perm
anent m
a
gnets
instead
of
wound
p
o
l
es. Meanwh
ile, step
up
co
mp
oun
d-
woun
d
m
o
to
r
is u
s
ed wh
ere
charact
e
r
i
s
t
i
c
s of a se
ri
es m
o
t
o
r i
s
re
qui
red
but
t
h
e m
o
t
o
r
i
s
not
i
n
hi
bi
t
e
d di
sc
ret
e
l
y
by
rem
ovi
ng t
h
e
l
o
ad
suc
h
as lathes whic
h experie
n
ce no loa
d
conditions
during each working
peri
od a
nd
t
h
e
n
are e
x
pose
d
to full
load agai
n. St
ep down co
m
p
ound-wound
m
o
tors are use
d
where a
n
al
m
o
st
const
a
nt
spee
d i
s
neede
d
, i
.
e.
lo
ad
s sm
aller t
h
an no
m
i
n
a
l lo
ad. Thu
s
, th
is typ
e
o
f
m
o
to
rs is often used in
lab
o
ra
tories to
prov
id
e co
n
s
tan
t
rot
a
t
i
o
n s
p
eed
.
Vari
o
u
s i
n
d
u
st
ri
es m
a
i
n
l
y
use ad
di
t
i
onal
c
o
m
poun
d m
o
t
o
r
i
n
st
ead
o
f
se
ri
es m
o
t
o
r t
h
ou
g
h
t
h
ei
r
d
e
sign
s are v
e
ry si
m
ilar. Tech
n
i
q
u
e
s proposed
to
con
t
ro
l
p
o
s
ition
o
f
DC
m
o
to
rs are
g
e
n
e
rally d
i
v
i
ded
in
t
o
t
h
ree cat
e
g
o
r
i
e
s:
C
l
assi
c
m
e
t
hods
l
i
k
e
usi
n
g
PID
co
nt
r
o
l
l
e
r
s
[
3
,
4]
;
m
oder
n
m
e
t
hods
l
i
k
e
co
nf
orm
i
ng,
o
p
t
i
m
a
l
an
d
o
t
h
e
r m
e
t
h
od
s [5
,
6
]
; an
d
in
tellig
en
t meth
od
s lik
e u
s
i
n
g
fu
zzy th
eo
ry an
d
n
e
u
r
al network
[7
, 8
]
. So
m
e
m
e
thods a
r
e propose
d
in this
pape
r to intelligently cont
rol the position
of separately exc
ited DC m
o
tor usi
ng
XC
SR
, i
m
prov
ed XC
SR
a
nd
m
achi
n
e l
earni
ng sy
st
em
s. The su
g
g
est
e
d
m
e
t
hods a
r
e e
x
ecut
e
d o
n
M
A
TL
AB
soft
ware
i
n
SI
M
U
LI
NK
en
vi
ro
nm
ent
by
si
m
u
l
a
t
i
on o
f
a
DC m
o
to
r
with its v
a
ri
o
u
s states b
e
i
n
g ev
al
uated
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Perfo
r
man
ce C
o
mpa
r
ison
b
e
t
w
een
Cla
ssic
an
d In
tellig
en
t
Meth
o
d
s
f
o
r Po
sitio
n Con
t
ro
l
… (Na
v
id
M.Y.)
36
8
2.
METHO
D
S F
O
R CO
NTR
O
LLING
S
P
EED
O
F
DC M
O
TOR
2.
1.
Contro
lling
El
ectric Re
sistance of
Arma
ture
It is the ol
dest
m
e
thod
use
d
to control spe
e
d wh
ich
still
has s
o
m
e
applications in se
ries m
o
tors
.
Term
i
n
al
vol
t
a
ge, e
x
ci
t
e
m
e
nt an
d/
o
r
ove
rfl
o
w
c
u
r
r
ent
are co
nstan
t
, with
th
e
con
t
ro
l b
e
in
g
don
e b
y
alteri
ng
electric resistance of the armature.
Whe
n
a rh
eo
stat is co
nn
ected
to
th
e arm
a
tu
re circu
it in
series, in
d
e
ed
to
tal
resistan
ce of th
e circu
it will
b
e
in
creased
.
Equ
a
tio
n
ab
ove sh
ows th
at th
is in
creased
resistan
ce wou
l
d red
u
ce
sp
eed
of th
e st
ab
le state ex
cep
t
for id
eal
n
o
lo
ad
co
nd
itio
ns. Electric resistan
ce of th
e rheo
stat can
b
e
ad
ju
sted
suc
h
t
h
at
v
a
ri
ous
spee
ds
(f
r
o
m
0 t
o
base
spee
d) a
r
e o
b
t
a
i
n
ed at
c
o
n
s
t
a
nt
t
o
r
q
ue (c
o
n
st
ant
c
u
r
r
ent
of t
h
e
arm
a
tu
re). Alth
oug
h
th
is speed
con
t
ro
l syste
m
is
re
lativ
ely si
m
p
le
an
d
in
exp
e
n
s
iv
e, it su
ffers from th
e
di
sad
v
a
n
t
a
ges bel
o
w:
1)
Spee
d is al
ways dec
r
ease
d
a
n
d it ne
ve
r e
x
ce
eds t
h
e
base s
p
eed;
2)
This m
e
thod is
alm
o
st ineffe
c
tive on
no loa
d
condition;
3)
The m
o
tor l
o
se
s its “constant
spee
d” feat
ure;
4)
The m
a
xim
u
m
po
we
r
gene
rat
e
d i
s
decrea
sed
i
n
pr
o
p
o
r
t
i
o
n
w
i
t
h
spee
d
red
u
c
t
i
on;
5)
A
great
deal
of ene
r
gy is l
o
st i
n
the
rhe
o
stat.
Power l
o
ss is
directly co
rrelat
e
d
with
sp
eed
redu
ctio
n. C
u
r
r
e
nt
o
f
t
h
e
arm
a
t
u
re i
s
not
c
h
a
nge
d
u
nde
r
co
nstan
t
to
rque co
nd
itio
n
s
,
wh
ile inp
u
t
power
o
f
th
e m
o
to
rs also
rem
a
i
n
s con
s
tan
t
. Th
is rheo
stat meth
od
is
u
s
ually ap
p
lied in
co
nd
ition
s
wh
ere t
h
e m
o
to
r is con
tinu
ously tu
rn
ed
on
an
d off,
o
r
wh
en
th
e low sp
eed
s
are
neede
d
only for a
while.
2.
2.
Contro
lling
Vo
lta
g
e
of Arma
ture
Th
e secon
d
meth
od
to con
t
ro
l sp
eed of the m
o
to
r is
chang
i
ng
v
o
ltage
of th
e arm
a
tu
re. Th
is m
e
th
o
d
is
m
a
in
ly u
s
ed
for sep
a
rately an
d
serially ex
cite
d
m
o
tors
. Excitem
e
nt ove
rfl
ow or current and
electric
resistance
of the a
r
m
a
ture are ke
pt consta
nt in this
s
p
ee
d control system
. Current
of
the arm
a
ture is ke
pt
u
n
c
h
a
n
g
e
d
at its n
o
m
in
al v
a
lu
e in
p
r
actical ap
p
lication
s
for a b
e
tter
u
tilizatio
n
of th
e m
o
tor wh
en
ev
er
sp
eed
i
s
al
t
e
red by
chan
gi
n
g
t
h
e
ba
se vol
t
a
ge
. Thi
s
t
echni
q
u
e is
actually the same as controlli
ng electric resi
stance
of t
h
e arm
a
t
u
r
e
whi
c
h e
n
abl
e
s o
p
erat
i
o
n at
l
o
we
r s
p
eed
s w
i
t
hout
di
sad
v
a
n
t
a
ges
of
i
t
.
Sp
eed c
ont
r
o
l
i
s
usu
a
l
l
y
do
ne at
c
o
nst
a
nt
cu
rre
nt
a
n
d
ove
rfl
ow
o
f
t
h
e arm
a
t
u
re. T
h
ereby
,
t
h
e c
o
n
s
t
a
nt
t
o
r
q
ue i
s
m
e
t
befo
re t
h
e
bas
e
spee
d. T
h
e i
n
put power
from
source t
o
m
o
to
r is also
ch
anged
lin
early in
propo
rtion
with
th
e sp
eed
.
Th
is k
i
nd
of
ope
ration until reaching the
base spee
d is
called ope
ra
tion at “constant torque-v
aria
ble powe
r”. Controlling
v
o
ltag
e
of th
e
arm
a
tu
re g
e
n
tly ad
ju
sts th
e sp
eed
and
a
lters it su
ch
th
at the sp
eed
ca
n
be
gra
dual
l
y
i
n
cr
eased
fro
m
0
to
its
b
a
se v
a
l
u
e.
Slop
e of th
e to
rqu
e
-sp
eed ind
e
x is
no
t ch
ang
e
d and no
p
o
wer is lost in
th
is m
e
th
od
.
3.
MODELING OF DC
MOTOR
Consi
d
eri
n
g the
differe
n
t t
y
pes
of DC
m
o
tors
and
num
e
rous m
e
thods
for c
ont
rolling them
,
sep
a
rately ex
cited
DC m
o
to
r is selected
in
th
is c
o
n
t
ri
bu
tio
n. Th
en,
d
i
rection
of its ro
tation
is set b
y
co
n
t
ro
lling
th
e
so
urce
vo
ltag
e
. Th
e gov
ern
i
ng equ
a
tio
ns in this reg
a
rd
are g
i
v
e
n b
e
l
o
w:
ܸ
௧
ൌܮ
ௗ
ೌ
ௗ௧
ܴ
ܫ
ܧ
(1
)
ܧ
ൌܭ
߱
(2
)
ܬ
ௗ
మ
ఏ
ௗ௧
మ
ܤ
ௗఏ
ௗ௧
െܶ
ൌܭ
݅
(3
)
ω
ൌ
ௗఏ
ௗ௧
(4
)
Schem
a
t
i
c
m
o
del
o
f
a
sepa
rat
e
l
y
exci
t
e
d
DC
m
o
t
o
r f
o
r
p
o
si
t
i
on c
ont
rol
ha
s bee
n
de
pi
ct
ed i
n
Fi
g
u
r
e
1.
Fig
u
re
1
.
Sch
e
matic
m
o
d
e
l o
f
a sep
a
rately ex
cited
m
o
to
r
with
con
t
ro
ller
blo
c
k
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
4, No
. 3,
J
u
ne 2
0
1
4
:
36
6 – 3
7
1
36
9
Whe
r
e,
ߠ
ௌ௨
and
ߠ
ா௫௧
repre
s
ent
ba
se a
n
d
out
pu
t
p
o
si
t
i
ons
,
r
e
spect
i
v
el
y
.
M
eanw
h
i
l
e
,
sp
ecification
s
an
d
p
a
ram
e
ters of t
h
e
DC
m
o
to
r sim
u
lat
e
d
i
n
th
is p
a
per are th
e same for all in
t
e
llig
en
t
cont
rollers
as l
i
sted bel
o
w:
electric resistance
of arm
a
ture=1
Ω
in
du
ctan
ce of
ar
m
a
tu
r
e
=0
.46
current
of arm
a
ture=
2
0 A
v
o
ltag
e
of armatu
re=1
10
V
ro
tation
a
l in
ertia o
f
m
o
to
r=0
.
0
93
friction coefficient at
ax
is of
m
o
to
r
=
0
.
0
08
4.
INTELLIGENT X
C
SR METHOD AN
D MACHINE
LEARN
ING SYSTEMS
The
fi
rst
Lea
r
ni
n
g
C
l
assi
fi
er
Sy
st
em
(LC
S
)
pr
op
ose
d
by
H
o
l
l
a
nd
wa
s
desi
g
n
e
d
t
o
w
o
r
k
b
o
t
h
fo
r
i
ndi
vi
dual
a
nd
cont
i
n
u
o
u
s
p
r
o
b
l
e
m
s
. Thi
s
cl
assi
fi
er l
ear
ni
n
g
sy
st
em
i
s
an exam
pl
e of m
a
chi
n
e l
ear
ni
n
g
whi
c
h
co
m
b
in
es ti
m
e
d
i
fferen
ces and
learn
i
ng
superv
ision
s
with
g
e
n
e
tic algo
rit
h
m
an
d
th
en
ad
op
ts to
so
lv
e
si
m
p
le
and
com
p
l
i
cat
ed
pr
obl
em
s. B
a
sed
on
t
h
e s
u
per
v
i
s
i
o
n i
n
t
r
o
duce
d
by
H
o
l
l
and
,
t
h
e
LC
S s
y
st
em
em
pl
oy
s a u
n
i
t
property for each of the classifiers which are ca
lled powe
r. Power is a divider which indicates
i
m
p
r
ession
ab
il
ity o
f
th
e d
i
v
i
d
e
r and
is ex
clu
s
iv
ely d
e
termin
ed
b
y
th
e p
e
rcen
tag
e
o
f
an
swers related
to
th
e
expecte
d
respons
es. T
h
ese
criteria are i
d
entified
with
ex
istin
g prin
ci
p
l
es in
sup
e
rviso
r
y trai
n
i
ng
. After
i
n
t
r
o
d
u
ct
i
on
of
t
h
e LC
S, som
e
ot
her t
y
pes
of i
t
ha
ve bee
n
de
vel
o
pe
d s
u
ch as
XC
S.
B
e
fo
re 1
9
9
5
whe
n
t
h
e
ex
tend
ed classificatio
n
system was no
t in
t
r
odu
ced
yet,
ab
ility o
f
a classifier to
g
e
t
p
r
op
er an
swers in
th
e
reinforcem
ent
syste
m
of these classifiers ha
d a key role
.
There
b
y, basic and sim
p
le classifier syste
m
s were
gra
d
ual
l
y
t
r
ansfo
r
m
e
d i
n
t
o
m
o
re acc
urat
e de
ci
si
on m
a
ker fact
ors.
No
w, i
t
i
s
wel
l
know
n
t
h
at
t
h
e XC
S i
s
abl
e
to
so
l
v
e ev
en
m
o
re co
m
p
lica
t
ed
prob
lem
s
with
n
o
n
e
ed
t
o
ad
ju
st th
e p
a
ram
e
ters. Th
at is wh
y th
is sy
ste
m
is
known as the
m
o
st successful syste
m
for learni
ng. XCSR is the sa
m
e
as
XCS and works with real num
b
ers.
Havi
ng int
r
oduced the classi
fier syst
e
m
wit
h
con
tin
uou
s variab
les (XCSR), so
m
e
in
tri
n
sic d
r
awb
acks o
f
th
e
b
i
n
a
ry classifier system
s in
clu
d
i
ng
i
n
ab
ility to
d
e
fi
n
e
sp
ecific rang
e
o
f
v
a
lu
es for
th
eir variab
les were
eliminated to a
large
exte
nt. T
h
ese
system
s
are now
acc
ount
ed for one
of
t
h
e m
o
st succes
sful lea
r
ni
ng a
g
ents
.
Accord
ing
to
t
h
e co
mm
o
n
ap
p
r
o
a
ch
fo
r t
r
ain
i
ng
XC
SR, fi
tn
ess is on
ly in
creased
for
ru
les wh
ich
correctly
resp
o
nd t
h
e l
e
a
r
ni
ng
dat
a
. T
h
i
s
m
eans t
h
at
chance
o
f
eac
h
rul
e
fo
r
not
bei
ng
del
e
t
e
d a
n
d
t
a
ki
n
g
pa
rt
i
c
i
p
at
i
on
in
th
e p
r
o
cess
o
f
n
e
w ru
les produ
ctio
n, are
d
i
rectly co
rrel
a
ted
to
ho
w they resp
ond
to
th
e train
i
ng
d
a
ta. A
realistic d
e
termin
atio
n
of th
i
s
ch
an
ce req
u
i
res ap
p
li
cat
i
o
n
of s
o
m
e
t
r
ai
ni
ng dat
a
. T
w
o m
e
t
h
o
d
s are m
e
nt
i
one
d
in
th
is sectio
n fo
r learn
i
n
g
with
teach
e
r
(l
earn
i
n
g
ru
le fro
m
train
i
n
g
data): “sep
aratio
n and
so
l
u
tion
”
an
d
“decision tree”
. One
rule is le
arnt at
eac
h
phase of the
form
er, then t
h
e
dat
a
co
vere
d
by
t
h
at
r
u
l
e
are
sep
a
rat
e
d
and
t
h
i
s
p
r
oce
ss i
s
c
ont
i
n
ue
d
o
n
t
h
e r
e
m
a
i
n
i
ng
sam
p
l
e
s. K
-
N
N
al
go
ri
t
h
m
o
r
nearest
nei
g
h
b
o
r
h
oo
d i
s
one
o
f
t
h
e l
ear
ni
n
g
a
l
go
ri
t
h
m
s
based
on
sam
p
l
e
whi
c
h
j
u
st
st
o
r
es t
h
e t
r
ai
ni
n
g
sam
p
l
e
s at
l
earni
ng
p
h
ase
.
Thi
s
alg
o
rith
m
calc
u
lates th
e
d
i
stan
ce of th
is
sam
p
le with
o
t
he
r traini
ng sam
p
les to s
p
ecify
the class of a
sam
p
le
data. The m
o
st comm
on crite
rion for cal
cul
a
tion of suc
h
distance is Euclid
ean cri
t
e
ri
on
al
t
hou
g
h
som
e
ot
he
r
criteria are also
u
tilized
fo
r th
is
p
u
rpo
s
e inclu
d
i
ng
Ma
nh
attan
-
Min
kowski. Hav
i
ng
m
easu
r
ed
t
h
e
d
i
stan
ce, a
m
a
jori
t
y
vot
i
n
g was hel
d
am
on
g
K
train
i
ng sa
m
p
les n
earest to
th
e cu
rrent test
sa
m
p
le.
Th
e m
a
j
o
rity l
a
b
e
l o
f
th
is sam
p
le is
assign
ed
t
o
t
h
e test sam
p
le th
en.
K
is
a p
a
ram
e
ter wh
ich
is
d
e
termin
ed
th
e
u
s
er. Su
ch
algorithm
s
are
also called lazy lear
ni
ng al
g
o
r
i
t
h
m
s
si
nce t
h
ey
do n
o
t
h
i
n
g
du
ring the learning phase and just
sto
r
e th
e test
sa
m
p
les. C4.5 alg
o
rith
m
is o
n
e
of the m
o
st
fa
m
ous al
go
ri
t
h
m
s
for
bui
l
d
i
n
g t
h
e
deci
si
o
n
t
r
ee
[2]
.
Thi
s
al
g
o
r
i
t
h
m
was
p
r
o
p
o
se
d
i
n
1
9
9
3
a
n
d
i
s
i
n
fact
a
d
e
vel
ope
d
ve
rsi
o
n
o
f
I
D
3 al
go
ri
t
h
m
.
Every
m
i
d no
de i
n
the created dec
i
sion tree
repre
s
ents a test on
the val
u
es
of a
property with
each br
a
n
chi
n
g being indicative
of
one al
l
o
we
d va
l
u
es o
f
i
t
.
The
cri
t
e
ri
on
use
d
f
o
r sel
ect
i
o
n o
f
an ap
pr
o
p
ri
at
e
pr
o
p
ert
y
fo
r a
no
de i
s
in
forma
tio
n
gai
n
wh
ich
lead
s to create a bias in
favo
r of
th
e pro
p
e
rties
o
f
v
a
riou
s
v
a
lues.
Ga
i
n
ra
tio
cri
t
e
ri
on
i
s
em
pl
oy
e
d
to
so
l
v
e th
is
p
r
ob
lem
.
I
D
3
alg
o
r
ith
m
j
u
st sup
p
o
r
ts
d
i
stin
ct pr
op
er
ties, w
h
er
eas C
4
.5 h
a
n
d
l
es con
tin
uous
p
r
op
erties in
ad
d
ition
to
t
h
e
d
i
stin
ct on
es.
Mo
reo
v
e
r, m
a
n
a
g
e
m
e
n
t
o
f
the p
r
op
erties
with
un
sp
ecific valu
es i
s
anot
her a
dva
nt
age o
f
C
4
.
5
o
v
er I
D
3. I
n
o
r
der t
o
av
oi
d excessi
ve
pr
o
p
o
rt
i
o
n i
n
t
h
e cl
assi
fi
cat
i
on m
odel
fo
rm
ed, t
echni
que
s o
f
deci
si
on
t
r
ee
pr
u
n
i
n
g are
use
d
.
T
h
e excessi
ve
pr
op
o
r
t
i
on
p
h
en
om
enon
occ
u
r
s
w
h
e
n
accuracy
of the
de
veloped m
odel is ve
ry
high
on the trai
ni
ng
data though i
n
sufficie
n
t on t
h
e test data
. In
othe
r
wo
rd
s, t
h
e cl
as
si
fi
cat
i
on m
odel
havi
n
g
ex
cessiv
e
propo
rtion
with
th
e trai
n
i
ng data has
been c
r
eated a
nd thi
s
h
i
gh
pr
opo
r
tion
do
es
n
o
t
n
ecessar
ily l
ead to greater acc
ura
c
y of classifica
ti
o
n
on
th
e test
d
a
ta set. Th
ere are
t
w
o m
a
jor deci
si
on t
r
ee
pr
uni
ng t
ech
ni
ques
.
In t
h
e
fi
rst
t
echni
que
whi
c
h i
s
cal
l
e
d pre-
p
r
uni
ng
, g
r
o
w
t
h
of t
h
e
tree is term
in
ated
i
n
so
m
e
p
a
th
s
b
e
fore co
mp
letio
n of
t
h
e t
r
ee.
In
th
e seco
nd
techn
i
qu
e
en
titled
po
st-p
ru
n
i
n
g
,
the tree is
grown com
p
letely
first a
n
d the
n
, som
e
sub-tree
s are
re
pl
aced with
a
leaf node.
The
obtained tre
e
can
b
e
t
r
ansfo
r
med
in
to
a set
o
f
equ
a
l classificatio
n
ru
les wh
ich
is
p
o
s
sib
l
e to
prun
e its ru
les b
y
d
e
letin
g
so
m
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Perfo
r
man
ce C
o
mpa
r
ison
b
e
t
w
een
Cla
ssic
an
d In
tellig
en
t
Meth
o
d
s
f
o
r Po
sitio
n Con
t
ro
l
… (Na
v
id
M.Y.)
37
0
of t
h
e p
r
ere
q
ui
si
t
e
s. Lim
i
t
e
d trai
ni
n
g
e
x
am
ples are a
ppl
i
e
d t
o
t
h
e
ge
nerat
e
d
rul
e
s a
f
t
e
r
pre
p
arat
i
o
n
of t
h
e
dat
a
.
Th
ereb
y,
p
a
rameters o
f
th
e ru
les will b
e
upd
ated
.
Gen
e
tic
mech
an
ism
a
l
so
con
t
ribu
tes to
m
a
k
e
n
e
w rules. At
the end
of the
training
phase,
som
e
acceptable res
u
lts can be
obtained using these trai
ne
d rules.
5.
INTRODUCTION OF
T
H
E PROP
OSE
D
METHO
D
Major drawba
ck of PID controller is diffic
u
lt adju
stm
e
nt of the pa
rameters to reach
the desire
d
an
swers as
well as th
e requ
ired
m
o
d
i
ficatio
ns d
u
e
t
o
th
e v
a
riab
le op
eration
cond
itio
n
s
o
f
th
e
m
o
to
r wh
i
c
h
is
im
possible during
ope
ration in practice.
Thus, it is sugges
t
ed to use i
n
te
lligent controlling m
e
thods t
o
solve
th
is p
r
ob
lem
.
A li
m
ited
set
o
f
train
i
n
g
data is g
e
n
e
rall
y u
s
ed
to
m
o
d
i
fy prop
erties o
f
th
e
ru
les (i.e.
“pre
di
ct
i
on”
, “
p
re
di
ct
i
o
n
er
r
o
r” a
n
d
“fi
t
n
e
s
s
”
) i
n
t
h
e
p
r
op
ose
d
m
e
t
hod.
Thi
s
i
s
d
o
n
e
usi
n
g t
h
e
fol
l
o
wi
n
g
equat
i
o
ns:
Up
dat
i
n
g
pre
d
i
c
t
i
on a
n
d
p
r
e
d
i
c
t
i
on e
r
r
o
r:
If
ex
p
i
< 1/
β
t
h
en Pi =Pi + (R-Pi) / exp
i
,
ε
i=
ε
i+(|R-Pi|-
ε
i) / exp
i
If
ex
p
i
≥
1/
β
th
en
Pi
=Pi
+
β
(R-Pi
)
,
ε
i=
ε
i+
β
(|R-P
i
|-
ε
i)
Upd
a
ting
fitn
ess:
If
ε
i<
ε
0 th
en
k
i
=1
If
ε
i
≥
ε
0 the
n
ki=
β
(
ε
i/
ε
0)
–
γ
Fi = fi+
β
[(
ki/
∑
kj
) –
fi
]
Whe
r
e,
ߚ
,
ߛ
and
ߝ
den
o
t
e
rat
e
of l
earni
ng
, p
o
w
er
of
rul
e
acc
ura
c
y
and
pre
d
i
c
t
i
on e
r
r
o
r, re
spe
c
t
i
v
el
y
.
Meanwhile,
݁ݔ
,
ܲ
and
ܴ
represent
rule e
x
perienc
e
, rule pre
d
ictio
n
and receive
d gain from
environm
ent,
and
݇
and
݂
are
rule a
ccuracy a
n
d
fitness, re
sp
ectively. Nu
m
b
er of th
e ru
le is
al
so dem
onst
r
at
e
d
by
t
h
e
s
u
b
s
c
r
i
p
t
݅
.
In t
h
e ne
xt
l
e
vel
,
“
ra
ndo
m
selectio
n
with
rema
i
n
d
e
r
” i
s
use
d
t
o
sel
ect
num
ero
u
s pai
r
s as pare
nt
s
am
ong stri
ngs available in dis
p
layer of
data conditions.
T
h
e new section
of
data condition is form
ed using the
m
i
d cross over
m
e
thod which is applied
on these pa
re
nt
strings
. The value of eac
h conditional vari
able is
gi
ve
n by
t
h
e eq
uat
i
o
n
bel
o
w:
a
i
=
α
(a
i
F
)+(
1
-
α
)(
a
i
M
)
Whe
r
e,
ܽ
de
not
es t
h
e val
u
e o
f
i
th
co
nd
itio
n
a
l
v
a
riab
le in
th
e n
e
w
d
a
ta,
ܽ
ி
and
ܽ
ெ
are the val
u
es of
i
th
con
d
i
t
i
onal
vari
a
b
l
e
i
n
t
h
e
fi
rst
a
n
d sec
o
nd
pa
re
nt
s (
f
at
her
an
d m
o
t
h
e
r),
res
p
ect
i
v
el
y
.
ߙ
is p
a
rticip
atio
n
co
efficien
t o
f
th
e
paren
t
s wh
ich
is
d
e
termin
ed
ad
ap
tiv
el
y. Th
e op
eratio
n section
o
f
th
e new d
a
ta i
s
also
gene
rat
e
d
usi
n
g a
no
nl
i
n
ear
m
a
ppi
n
g
f
r
o
m
t
h
e space
o
f
t
h
e co
n
d
i
t
i
onal
vari
a
b
l
e
s t
o
t
h
at
o
f
exi
s
t
i
n
g
dat
a
.
Div
e
rsificatio
n o
f
t
h
e ex
isting
d
a
ta is co
n
t
i
n
u
e
d
u
n
til
th
e
termin
atio
n
con
d
ition
(e.g
. percen
tag
e
of co
rrect
answers
gi
ven
for the
test dat
a
reac
hes a
pre
d
efi
n
ed
th
resho
l
d) is satisfied u
s
i
n
g th
e co
mp
leted
d
a
ta.
The
ne
xt
sect
i
o
n
deal
s
wi
t
h
i
n
t
r
od
uci
n
g s
o
m
e
com
m
on algo
ri
t
h
m
s
for
l
earni
ng
wi
t
h
t
e
a
c
her
b
r
i
e
fl
y
.
The res
u
l
t
s
o
b
t
ai
ned f
r
om
t
h
e pr
op
ose
d
m
e
t
h
o
d
are e
v
al
u
a
t
e
d f
o
r c
ont
r
o
l
l
i
ng p
o
si
t
i
on
of
DC
m
o
t
o
r as wa
s
pre
v
i
o
usl
y
di
sc
usse
d i
n
sect
i
o
n
2.
respectively.
ߙ
i
s
p
a
rticip
atio
n co
efficien
t
of th
e p
a
ren
t
s
wh
ich
is
d
e
term
in
ed
ad
ap
tiv
ely. Th
e
ope
rat
i
o
n sect
i
on
o
f
t
h
e
ne
w dat
a
i
s
al
s
o
ge
ne
rat
e
d
u
s
i
ng a
n
onl
i
n
e
a
r m
a
ppi
ng
f
r
o
m
t
h
e space
of t
h
e
co
nd
itio
n
a
l
v
a
riab
les to th
at
o
f
ex
isting
data. Div
e
rsificatio
n
o
f
t
h
e ex
istin
g
d
a
ta is con
tin
u
e
d
until th
e
termination condition (e.g.
perce
n
tage
of correct answers gi
ven
for
the
test data reaches a
pre
d
efi
n
e
d
th
resh
o
l
d
)
is satisfied
using
the co
m
p
leted
d
a
ta.
The ne
xt
secti
o
n
deals with introdu
cing s
o
me comm
on algorithm
s
for l
earni
ng with t
eacher briefly. The
results obtaine
d
from
the propose
d
m
e
thod are eval
uated for controlling
positi
on of DC
m
o
tor as
was
pre
v
i
o
usl
y
di
sc
usse
d i
n
sect
i
o
n
2.
6.
RESULTS OF RUNNING
INTELLI
GENT CONT
ROL METHODS
Main purpose
of designing a cont
roller is to incr
ease stability and decrease ti
me to reach the desire
d
state u
p
o
n
app
l
icatio
n
o
f
a
d
i
stu
r
b
a
n
ce in
a
pro
cess. PI
D con
t
ro
ller was u
t
i
lized
in
ad
d
ition
to
co
m
p
arison
of
answ
ers
fr
om
t
h
e
pr
o
pose
d
m
e
t
h
o
d
s i
n
t
h
i
s
p
a
per
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
4, No
. 3,
J
u
ne 2
0
1
4
:
36
6 – 3
7
1
37
1
Par
a
m
e
ter
XCSR
im
p
r
oved XCSR
PI
D
m
a
chine lear
ning
Adaptation ti
m
e
for
increased base
2.
3 1.
5
3.
7
3.
1
Adaptation ti
m
e
for
decreased base
2.
3 1.
5
3.
7
3.
1
Over
shoot per
centage
f
o
r increased base
0 0
0
0
Under
s
hoot per
centage
f
o
r decreased base
0 0
0
0
Adaptation ti
m
e
for
incr
eased base upon
10% change in
J
and B
2.
7 1.
5
4.
5
3.
8
Adaptation ti
m
e
for
decreased base
2.
7 1.
5
4.
5
3.
8
Over
shoot per
centage
f
o
r 10% increased base
0 0
2%
0
Under
s
hoot per
centage
f
o
r 10% increased base
0 0
14%
0
7.
CO
NCL
USI
O
N
A num
b
er of i
n
telligent m
e
thods
were use
d
in this
pa
per to control the
positi
on of sepa
ra
tely excited
DC m
o
to
r. It
was
ob
serv
ed
t
h
at th
e PID con
t
ro
ller
h
a
s l
o
st its p
e
rfo
rm
an
ce b
e
i
n
g un
ab
le to
co
m
p
letel
y
d
e
lete
th
e ov
ershoo
t su
ch
th
at, it
was n
ecessary to
d
e
sign
an
d
m
o
d
i
fy its p
a
ra
m
e
ters o
n
c
e
ag
ain
.
However, the
suggeste
d m
e
t
h
ods
were ne
e
d
less of redesign as they we
re self-c
orrecting. Additio
nal
l
y, ti
me to reach the
fi
nal
a
n
swe
r
i
s
si
gni
fi
ca
nt
l
y
sh
ort
e
r
i
n
t
h
ese
m
e
t
hods
.
REFERE
NC
ES
[1]
Ele
c
tri
c
M
a
chin
es
and Di
rec
t
Cu
rrent, Ghaem Pu
b., (In Persian).
[2]
Quinlan,
J.R.
“C4.5: Programs for Machine Lear
ning”,
Morgan-
K
aufmann, San
Mateo, CA. 199
3.
[3]
Ogatha, K. H.
, “
Control Engin
e
ering
”,
Trans
l
ate
d
b
y
: Di
ani
,
M
.
,
Nas
s
P
ub., 4
th
Edition, 2002
, (In
Persian).
[4]
KworkL. Tang,
Robort J. Mulholland
, “Comparing Fu
zzy
Log
i
c
With Classical
Controller Desig
n
s”.
IEEE Trans.
SMC
, Vol.17, No. 6
.
Nov./Dec.1
987,pp.1085-10
87.
[5]
Mokhtari, M., “
Applica
tions of
MATLAB and SI
MULINK in Eng
i
neering
”,
Trans
l
at
ed b
y
: S
a
m
a
di
, V.,
Khoras
a P
u
b.,
1
st
Ed
ition
,
(In
Persian).
[6]
Menhaj, B., “
Ca
lculational Intelligence,
Basics
of Neural N
e
tworks
”, H
e
s
a
b
i
P
ub.
,
(In P
e
rs
ian)
.
[7]
R. Rico M
a
rtin
ez, J.S. Anders
on and I. G
.
Kebrekidis, “
Continuous-time nonlinear signal pr
ocessing: a neu
r
al
network based
approach for g
r
iy box iden
tification
”,
in Pro
c
. I
EEE Workshop on neur
al
network for sig
n
al
Processing, pp
.
596-605, Oct. 19
94.
[8]
Y.F. La, C
.
C. Lau, “Developmen
t of
Fuzzy
Algor
ithms for Servo
S
y
stems”.
1988
IEEE Internatio
nal Confer
ence
on
Robotics
and automation
, Philadelphir, PA, April
24-29, PP.65-71.
Evaluation Warning : The document was created with Spire.PDF for Python.