Int
ern
at
i
onal
Journ
al of
R
obot
ic
s
and
Autom
ati
on (I
JRA)
Vo
l.
8
, No
.
3
,
Septem
ber
201
9
, p
p.
155
~
163
IS
S
N:
20
89
-
4856,
DOI: 10
.11
591/
i
jra
.
v8
i
3
.
pp
155
-
163
155
Journ
al h
om
e
page
:
http:
//
ia
escore.c
om/j
ourn
als/i
ndex.
ph
p/IJRA/i
ndex
Interv
al
type
-
2 fuzzy l
ogic c
ontroll
er desi
gn
for dis
tillation
co
lum
n
process
Moham
med
Y
. Hass
an
,
Seb
al S. Ez
z
at
en
Control
and
s
y
s
t
ems
Engi
nee
r
ing
Depa
rtment
,
Un
ive
rsit
y
of Tec
h
nolog
y
,
Ir
aq
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
ul
24
, 2
01
8
Re
vised
Ma
y
25
, 2
01
9
Accepte
d
J
un
22
, 201
9
Distil
lation
co
lu
m
ns
are
the
m
ost
unit
s
used
in
oil
refine
r
ie
s,
an
d
che
m
ic
a
l
fac
tor
ie
s.
Thi
s
is
a
ver
y
d
i
ffic
ul
t
proc
ess
and
non
-
li
ne
a
r.
Th
ere
for
e,
the
dev
el
opm
e
nt
of
int
e
ll
ig
e
nt
cont
rol
s
y
st
ems
for
the
col
um
ns
of
the
d
isti
llat
ion
i
s
ver
y
diff
ic
ul
t.
In
th
is
p
ape
r
,
an
intell
ig
ent
control
stra
te
g
y
using
Mam
dani
t
y
p
e
Interval
T
y
p
e
-
2
PI
Li
k
e
Fuz
z
y
Log
ic
Contro
ll
er
(IT
2FLC)
is
use
d.
Th
e
cont
rol
ler
consists
of
PD
-
Li
ke
FLC
with
integra
t
e
d
output
.
Kern
ek
Mende
l
(KM
)
algorithm
is
used
as
the
t
y
p
e
r
edu
ct
ion
m
et
hod
f
or
the
I
T2FLC.
Thi
s
cont
r
oller
is
appl
i
ed
to
c
ontrol
a
continu
ous
bina
r
y
tray
s
disti
ll
a
ti
on
col
um
n.
The
c
ontrol
ler
has
th
ree
tuna
bl
e
ga
i
ns
to
reach
m
ini
m
um
over
shoot,
m
ini
m
um
err
o
r
and
m
inim
um
sett
li
ng
t
i
m
e
at
least
var
ia
b
le
s
c
an
b
e
cont
rol
le
d
.
Th
e
cont
rol
l
er
is
a
v
ari
ab
le
of
the
m
ola
r
fr
ac
t
ion
of
disti
llat
e
and
the
r
efl
ex
ratio
is
the
m
ani
pu
l
at
ed
v
ariabl
e
.
In
te
gra
l
Ti
m
e
Abs
olut
e
Err
or
(IT
AE)
is
empl
o
y
ed
as
an
obj
ec
t
ive
func
ti
on
to
m
ea
sure
the
improvem
en
t
in
ti
m
e
r
esponse
where
the
er
ror
is
bet
wee
n
desire
d
and
output
produc
t
c
om
positi
on.
The
per
form
anc
e
of
IT2FLC
is
compare
d
with
T
y
p
e
-
1
PI
L
ike
FLC
(T1FLC
).
The
r
esult
s
of
the
sim
ula
t
ions
have
show
n
tha
t
the
proj
ect
of
IT2FLC
works
eff
icientl
y
to
no
-
distu
rba
nce
and
the
eff
e
ct
s
of
dis
turba
nc
e.
Im
prove
av
era
g
e
is
of
85%
for
a
constant
set
-
po
int
without
a
distur
banc
e
and
80%
with
a
d
isturba
n
ce
.
Furthermore,
the
ave
r
age
improvem
ent
fo
r
a
st
ep
set
-
poin
t
is
53%
wi
thou
t
d
isturba
nc
e
an
d
74%
with
disturba
nc
e.
All
result
s
of
th
e
si
m
ula
ti
on
conf
ir
m
ed
the
h
ard
ines
s
and
cont
rol
an
y
consist
ent i
n
ac
cur
ate
with
ob
vious a
dvan
ta
g
e
s for
the IT
2FLC
.
Ke
yw
or
d:
Disti
ll
at
ion
FLC
IT2
Copyright
©
201
9
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Moh
am
m
ed
Y. Hassa
n
,
Con
tr
ol a
nd sy
stem
s En
gi
neeri
ng
De
par
tm
ent,
Un
i
ver
sit
y o
f Te
ch
no
l
og
y
,
Ir
aq
.
Em
a
il
:
60
003@u
otech
nolo
gy
.ed
u.i
q
1.
INTROD
U
CTION
The
c
olu
m
ns
of
the
disti
ll
at
ion
bin
ary
a
re
li
kely
the
m
os
t
of
t
he
c
on
t
ro
ls
in
pet
ro
le
um
and
chem
ic
al
eng
i
neer
i
ng.
T
he
dy
nam
ic
s
of
t
he
va
riabl
es
therm
o
dy
nam
ic
no
n
id
eal
it
y;
s
iz
e,
uncertai
nty,
a
nd
delay
the
m
easur
es
of
the
c
om
po
sit
ion
pu
t
t
he
c
ontrol
of
pr
oce
dur
es
f
or
the
disti
l
la
ti
on
with
a
c
halle
ng
i
ng
pro
blem
s
[1
]
.
T
he
a
udit
ors
re
porte
d
on
fu
zzy
lo
gic
Ty
pe
-
1
that
ca
n
be
us
e
d
s
uccess
fu
ll
y
f
or
a
nu
m
ber
of
proces
ses
of
non
-
li
near
a
nd co
m
plex.
D
es
pi
te
it
s
wide
p
opula
rity
,
it
is
s
how
n
i
n
m
any scie
ntific
re
sea
r
ch
cases
the
project
s
of the Ty
pe
-
1 FLC
(
T
1FLC
) have
d
i
ff
ic
ulti
e
s to
m
ini
m
iz
e t
he
im
pact o
f
t
he
u
ncer
ta
inti
es
in the pla
nt
[
2
].
Ther
e
f
or
e,
the
cho
ic
e
of
t
he
T
1F
LC
does
not
neces
saril
y
th
e
m
os
t
favor
a
bl
e
so
luti
on
to
the
pro
blem
of
c
on
t
ro
l,
w
her
eas
t
he
be
st
al
te
rn
at
ive
is
us
i
ng
Ty
pe
-
2
FLC
(T
2FLC
)
put
t
he
us
e
of
it
s
fu
z
zy
set
s
char
act
e
rized
by
a
c
onseq
ue
nc
e
an
d,
a
la
r
ge
r
num
ber
of
pa
ram
et
ers.
T2FLC
ha
s
bee
n
a
pp
li
ed
in
m
any
fiel
ds
li
ke
veh
ic
le
nonlinea
r,
s
usp
ensio
n
syst
em
s,
r
obot
m
ob
il
e
autonom
ous
con
t
ro
l,
para
ll
el
m
echan
is
m
of
cable
-
dr
iv
en
, b
iochem
ic
al
r
eact
or
c
ontrol
an
d
in
ve
rted pe
ndul
um
m
od
el
.
Pr
at
ishtha
na
nda
an
d
Wo
ra
pr
adya
[
3
]
was
dev
el
op
e
d
the
con
t
ro
l
with
a
real
-
ti
m
e
fo
r
a
col
um
n
of
a
bi
nar
y
disti
llati
on
us
in
g
Ty
pe
-
1
PI
F
uzzy
Lo
gic
C
on
t
rol
le
rs.
A
nonlin
ear
c
on
t
ro
l
a
da
ptive
al
gorith
m
for
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2089
-
4856
IJRA
,
V
ol.
8
,
No.
3
,
Septem
ber 2
019
:
155
–
163
156
a
m
ult
i
com
po
nen
t
batch
dist
il
la
ti
on
colum
ns
wer
e
de
vel
op
e
d
by
Ja
na
and
Mu
rlidh
a
r
[
4
]
.
Ca
nete
et
al
.
[
5
]
app
li
ed
a
neur
ofuzzy
a
dap
ti
ve
co
ntr
ol
us
i
ng
s
of
t
se
nsors
for
c
onti
nuou
s
disti
ll
at
ion
c
olu
m
n.
Jav
adi
an
d
Ho
s
sei
ni
[
6
]
pro
posed
t
he
P
I
co
ntr
oller
f
or
a
bi
nar
y
disti
ll
at
ion
column.
Ba
rc
el
ó
-
Ri
c
o
et
al
.
[
7
]
re
al
iz
ed
the
m
od
el
ing
and
t
he
co
ntr
ol
of
a
co
ntin
uous
disti
ll
at
ion
tower
t
hroug
h
f
uzzy
te
chn
i
qu
e
.
A
fu
zzy
m
od
el
pr
e
dicti
ve
co
nt
ro
l
in
m
ulti
var
ia
ble
co
ntr
ol
of
disti
ll
at
io
n
col
um
n
wa
s
us
e
d
by
Sivakum
ar
et
al
.
[
8
]
Al
-
D
unai
na
wi
an
d
A
bbod
[
9
]
pe
rfo
rm
ed
three
te
c
hniq
ue
s
of
e
voluti
onary
al
gorithm
as
a
tu
ner
to
f
uzzy
log
ic
c
on
t
ro
ll
er
.
Disti
ll
at
ion
co
lum
n,
as
pr
e
vi
ou
sly
desc
rib
ed,
is
a
high
nonlinea
r,
poorl
y
-
de
fine
d
relat
ion
s
hip
betwee
n
the
inputs
a
nd
out
pu
ts
,
cha
racter
iz
ed
by
uncert
ai
nty.
Th
us
;
usi
ng
T1
FLC
di
d
not
ac
hieve
high
accuracy
i
n
c
ontr
olli
ng
a
bi
na
ry
tray
s
disti
ll
at
ion
col
um
n
and
face
d
dif
f
ic
ulti
es
in
re
duci
ng
the
ef
fe
ct
of
disturba
nces a
nd unce
rtai
nties i
n
t
he dist
il
la
t
ion
c
olu
m
n
c
onti
nuous ty
pe
.
The
ai
m
of
thi
s
pa
pe
r
is
t
o
de
sign
IT
2
PI
-
L
ike
FLC
f
or
c
onti
nuous
disti
ll
at
ion
c
olu
m
n
(
IT2FLC)
to
reach
the
set
-
point value
f
or
a
disti
ll
at
e
fr
act
i
on
(
X
d) w
it
h
m
ini
m
u
m
ov
er
sh
oot,
m
ini
m
u
m
err
or
a
nd
m
i
nim
u
m
set
tl
ing
t
i
m
e
d
espite
the
dist
urba
nce.
Re
su
l
ts
will
be
com
par
e
d
with
us
i
ng
Ty
pe
-
1
P
I
-
Like
FLC
ta
ki
ng
i
nto
account t
he
vari
at
ion
s in
syst
em
p
ara
m
et
ers
and d
ist
urban
ce
.
2.
DIS
TIL
LATI
ON COL
U
M
N
P
ROCESS
AND MO
DE
L
It
is
known
th
at
disti
ll
at
ion
colum
ns
are
the
un
it
s
t
hat
m
o
st
us
ed
i
n
oil
refi
ne
ries;
chem
ic
al
and
petr
och
em
ic
al
plants
[
9
]
.
It
is
the
un
it
t
he
operati
on
in
a
uto
m
atic
co
ntr
ol
m
us
t
be
stu
di
ed.
T
her
e
a
re
a
lot
o
f
ty
pes
of
disti
llati
on
c
olu
m
ns
li
ke,
conti
nu
ous or
batch, bi
na
ry or m
ulti
co
m
po
nen
t,
pac
ke
d or t
ray [
1].
2.1.
Pr
ocess
de
scri
pt
ion
The
disti
ll
at
io
n
is
a
process
i
n
a
li
quid
or
va
por
m
ixtur
e
of
tw
o
or
m
or
e
com
po
ne
nts
w
as
sepa
rate
d
into
it
s
f
racti
ons
of
re
qu
i
red
pu
rity
,
by
the
ap
plica
ti
on
a
nd
rem
ov
al
of
heat.
Disti
ll
ation
col
um
n
ge
ner
al
ly
m
ade
up
of
tra
ys
of
c
olu
m
n.
Th
e d
ist
il
la
ti
on
colum
n
co
ns
is
ts
of
m
any
tray
s,
re
boil
er,
c
ondenser
a
nd
a ve
rtic
al
colum
n
f
or
t
he se
par
at
io
n p
urpo
s
e.
The
fee
d
is
th
e
ent
ry
in
the
colum
n.
Disti
llati
on
c
olu
m
n
i
s
se
par
at
e
d
i
nto
t
wo
sect
io
ns.
T
hey
a
re
recti
ficat
ion
se
ct
ion
a
nd
strip
p
in
g
sect
io
n.
Trays
un
der
t
he
feed
tray
are
cal
le
d
the
sec
ti
on
of
st
rip
pin
g
an
d
tray
s
ab
ove
fe
ed
t
ray
are
cal
le
d
recti
ficat
ion
sect
i
on
[
10
]
,
Fig
ur
e
1
Leav
ing
the
Co
nde
ns
er
li
quid
is
e
ntere
d
into
t
he
drum
of
the
ref
le
x.
A
pa
rt
of
this
l
iqu
id
c
olu
m
n
de
al
s
as
a
ref
le
x
an
d
fall
do
wn
by
the
tray
s.
F
inall
y,
the f
l
uid
is
wit
hdra
wn,
drum
as a
purified
pr
oduct.
B
o
i
l
er
F
e
e
d
s
t
ag
e
Nf
R
e
f
l
e
x
D
r
u
m
F
e
e
d
S
t
a
g
e
Nf
T
o
p
P
r
o
d
u
c
t
D
,
xD
C
o
n
d
en
s
er
B
o
t
t
o
m
P
r
o
d
u
c
t
B
,
xB
v
L
F
e
e
d
,
F
,
xF
Fig
ure
1. Sc
he
m
at
ic
i
m
per
so
nation o
f
a
ty
pical
b
ina
ry co
nt
inuous
disti
ll
ation
c
olu
m
n
[
1
1
]
2.2.
M
athem
ati
cal m
od
el
The
nonlinea
r
syst
e
m
is
m
od
el
e
d
with
t
he
m
ass
bala
nce
a
nd
c
ompone
nt
bala
nc
e
eq
uatio
ns
.
The
m
od
el
is si
m
pl
ifie
d
un
der the
fo
ll
owin
g assum
ption
s
[1
2
]:
Evaluation Warning : The document was created with Spire.PDF for Python.
IJRA
IS
S
N:
20
89
-
48
56
In
te
rv
al type
-
2 f
uz
zy l
og
ic
c
ontroll
er
desig
n f
or
disti
ll
ation colu
m
n pr
oces
s
(
Moha
mm
e
d Y.
Ha
s
san
)
157
−
******
*****/
No ch
em
ic
al
r
eact
ion
s
occ
ur
.
−
No ch
a
nge in
(c
onsta
nt) p
ress
ur
e
.
−
Dou
ble (
Bi
na
r
y) m
ixtur
e.
−
No ch
a
nge in
a
(
c
on
sta
nt) rela
ti
ve
vola
ti
li
ty
.
−
No v
a
por
hold
-
up o
cc
urs in
th
e tray
s.
−
Perf
ect
e
quil
ib
rium
an
d
m
ixing f
or v
a
por
-
li
qu
i
d on al
l st
ag
es.
−
Perf
ect
c
ontr
ol level o
n re
bo
il
er a
nd r
e
flu
x d
ru
m
.
The
Ma
ss
b
al
a
nce e
qu
at
i
on for ea
ch
stage is
[
1]
:
−
Con
si
der
sta
ge
i
=
N
+
1,
x
D
=
x
N
+1,
w
he
re
N
is
Nu
m
ber
of
s
ta
ges,
x
D
is
the
disti
ll
at
e
l
iqu
id
m
olar
fr
act
io
n,
x
N
is Li
quid c
oncent
rati
on on
ℎ
sta
ge
is a
Co
nden
ser st
age a
nd is
represe
nted
as
(
1)
:
M
D
=
V
y
i
–
1
–
L x
D
–
D x
D
(1)
wh
e
re
M
D
is
Conde
ns
er
hold
-
up, V
is Vapor
flo
w
rate,
y
i
is
Vapo
r
m
olar
frac
ti
on
o
n
i
th
s
ta
ge,
L
is
Liq
uid
fl
o
w
rate an
d
D
is
d
i
sti
ll
at
e flo
w
rat
e.
−
Stage
i
=
NF
+
1
t
o
i
=
N:
re
pr
e
sents
tray
s
a
bo
ve
fee
d
tray
w
her
e
NF
is
t
he
nu
m
ber
of
the
feed
tray
an
d
i
ts
equ
at
io
n
is
as
f
o
ll
ows
(2)
:
M
i
=
L x
i +1
V
y
(i
–
1)
–
L
X
i
–
V
y
i
(2)
wh
e
re M
i
is
Hold
-
up on
ℎ
sta
ge
and
X
i
is Li
qui
d
m
olar
f
racti
on
on
ℎ
sta
ge.
−
Stage i
=
N
F: re
pr
ese
nts
feed t
r
ay
an
d i
ts eq
ua
ti
on
is
wr
it
te
n
as foll
ows
(3)
:
Mi
=
L X
i + 1
+V
y
(i
-
1)
–
L X
i
–
V
y
i
+F X
f
(3)
wh
e
re
F is Fee
d flo
w
rate a
nd
X
f
is Fee
d
li
qu
id m
olar
fr
act
i
on.
−
Stages i
=
2
to
i
=
NF
–
1:
repre
sent the
stages
belo
w
the
f
ee
d t
ray w
hich
a
re
expresse
d
as
fo
ll
ow
s
(4)
:
Mi
=
(L
+
F)x
i + 1
+ V
y
(i
-
1)
–
L
(
L +
F
)x
i
–
V y
i
(4)
−
S
ta
ge
i
=
1,
x
B
=
x
1
: re
pr
ese
nts
reboil
er s
ta
ge wh
ic
h
is
repres
ented
a
s foll
ow
s
(
5)
:
M
B
=
(L
+
F)x
i
+ 1
+ V
y
B
–
B
x
B
(5)
wh
e
re
M
B
is
Re
bo
il
er
hold
-
up,
x
B
is
Bott
om
s
li
qu
id
m
ol
ar
f
racti
on,
B
is
Bott
om
flow
rate,
an
d
y
B
is
Vapo
r
m
olar
fr
act
ion
on
bott
om
stage.
−
Vapo
r
–
li
quid
equ
il
ib
rium
r
el
at
ion
s
hip
for
e
ach tray
was re
pr
ese
nted
as
(6)
:
y
i
=
1
+
(
−
1
)
(6)
wh
e
re
is R
el
at
i
ve vo
la
ti
li
ty
−
Re
flux r
at
io
is
represe
nted
as
(7)
:
R
=
(7)
wh
e
re R is
Ref
le
x
rati
o.
−
Total
r
e
bo
il
er
m
ass b
al
ance i
s
re
pr
ese
nted
a
s
(
8)
:
B
=
F
–
D
(8)
−
Total
con
dense
r
m
ass b
al
ance
is represe
nted as
(
9)
:
D
=
+
1
(9)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2089
-
4856
IJRA
,
V
ol.
8
,
No.
3
,
Septem
ber 2
019
:
155
–
163
158
2.3.
M
od
el
in
g o
f
t
he dis
till
ati
on
c
olumn
The
eq
uatio
ns
of
the
m
od
el
a
re
enc
rypted
by
the
MATLA
B
env
ir
onm
ent
,
s
-
sc
ript
f
un
ct
i
on
is
cal
le
d
by
the
blo
c
k
i
n
a
Sim
ulink
.
The
reso
l
ution
ordina
ry
di
ff
e
ren
ti
al
eq
uatio
ns
ar
e
us
e
d
O
DE
45.
Fig
ure
2
s
hows
the m
od
el
Sim
ulink o
f
c
onti
nuous ty
pe dist
illat
ion
co
l
um
n
f
ro
m
(
1
t
o 9).
Fig
ure
2. Mo
de
l of co
nt
in
uous t
ype
disti
ll
at
i
on co
l
um
n
(1
-
9)
More
ov
e
r, sim
ulati
on
m
od
el
of the
Disti
ll
at
i
on Col
um
n
bloc
k
is s
how
n
i
n Fi
g
ure
3.
Figure
3. Sim
ulink
m
od
el
of
di
sti
ll
a
ti
on
co
l
um
n
blo
ck (
1
-
5)
wh
e
re
the
blo
c
k
CO
NS
T
ILLs
s
in
the
Sim
uli
nk
of
disti
ll
at
i
on
c
olu
m
n
is
s
-
f
un
ct
io
n
that
con
ta
in
s
the
program
wh
ic
h descri
be
s equati
ons
(1 to
5).
2.3.1
.
Simul
ati
on
re
sponse
of o
pe
n lo
op dis
till
at
ion
co
lu
mn
The
sim
ulati
o
n
was
perform
ed,
to
e
nsure
com
pliance
wi
th
t
he
dynam
ic
w
orki
ng
i
n
t
he
operati
on,
wh
e
re
by
the
pa
ram
et
ers
are
li
ste
d
in
the
T
a
ble
1.
T
he
m
od
el
has
a
non
-
li
near
reacti
on
(r
es
pons
e
).
Fig
ur
e
4
sh
ows
dynam
i
c
res
pons
e
with
t
he
c
ha
ng
e
i
n
t
he
ref
le
x
ra
ti
o
(RR)
as
+1
0%
a
nd
-
10
%
,
w
her
e
the
nom
inal
value o
f
RR
is
2.56 [
1].
Evaluation Warning : The document was created with Spire.PDF for Python.
IJRA
IS
S
N:
20
89
-
48
56
In
te
rv
al type
-
2 f
uz
zy l
og
ic
c
ontroll
er
desig
n f
or
disti
ll
ation colu
m
n pr
oces
s
(
Moha
mm
e
d Y.
Ha
s
san
)
159
Table
1.
Param
et
ers
o
f dist
il
la
ti
on
c
olu
m
n
m
od
el
[
1]
Sy
m
b
o
l
Ty
p
ical
valu
e and
un
it
Sy
m
b
o
l
v
alu
es an
d
un
it
0
.5 k
m
o
l
F
1
k
m
o
l/h
0
.5 k
m
o
l
2
0
.5 k
m
o
l
2
.56
V
1
.78
k
m
o
l/h
N
20
L
1
.28
k
m
o
l/h
NF
11
D
0
.5 k
m
o
l/h
0
.5
B
0
.5 k
m
o
l/h
0
.98
Figure
4. Dy
na
m
ic
r
esponse
to
the
ch
a
nge i
n t
he reflex
r
at
io
(
RR
)
(
i) Res
pons
e
in
the
no
m
inal v
al
ue
(Blue
li
ne
)
,
(
ii
)
Re
s
pons
e
at +1
0%
i
n
R(
Re
d
li
ne)
,
(
ii
i) Resp
onse
at
-
10% i
n
R
(
Yell
ow li
ne)
3.
DESIG
N OF
THE
PROPO
SED
CONTR
OLL
ER
In the
fo
ll
owin
g
sect
io
n, the
de
ta
il
s o
f desig
ni
ng
IT
2
P
I
-
Lik
e FLC a
re e
xp
l
ai
ned
.
3.1.
Des
i
gn o
f
IT2 PI
-
Li
ke
FLC
(IT2
FL
C
)
The
c
once
pt
of
T
2F
LC
w
as
pro
po
se
d
by
Z
adeh
at
1975
a
s
a
sta
te
m
ent
of
Ty
pe
-
1
F
L
C.
F
uzzy
set
s
functi
on
of
th
e
m
e
m
ber
sh
ip
s
are
incl
ud
e
d
in
t
he
Footp
r
int
O
f
U
nc
erta
inty
(FOU
)‚
a
s
sho
wn
in
Fi
gure
5.
They
are
capa
ble
of
m
od
el
in
g
a
nd
ha
nd
li
ng
the
uncertai
nties‚
nonlinea
riti
es
and
li
nguisti
c
associat
ed
with
the
in
puts
a
nd
outp
ut
of
FL
C
by
m
od
el
ing
t
hem
and
m
ini
m
iz
ing
their
ef
fect.
The
refor
e
‚
IT
2F
LC
i
s
us
ed
instea
d
of
t
he
T1F
LC
.
IT
2
f
uzzy
set
s
are
e
xtens
i
on
s
of
th
e
cl
assic
al
fu
z
zy
set
s
that
ad
d
a
n
ext
ra
degree
of
fr
ee
do
m
to
enti
tl
e an
unce
rtai
nty fact
or in
th
e syst
e
m
b
ut th
ey
are no
t easy
co
m
par
ed
w
it
h
Ty
pe
-
1
F
LC
[13].
In
t
his
pap
e
r,
IT2
PI
l
ike
F
LC
of
Ma
m
dan
i
ty
pe
is
desi
gn
e
d
with
se
ve
n
in
puts
-
ou
t
put
tri
an
gula
r
sh
a
ped
m
e
m
b
ersh
i
ps
.
The
un
i
ver
se
of
di
sco
ur
se
ra
ng
e
is
withi
n
[−
1
1]
f
or
e
ach
in
put
a
nd
outp
ut.
The
f
or
m
s
of
MFs
a
re
sho
w
n
i
n
Fig
ure
6.
Ce
nt
ro
i
d
is
t
he
m
et
ho
d
t
ha
t
is
sel
ect
e
d
f
or
the
defuzzif
ic
at
ion
te
chn
iq
ue.
T
he
Rule
-
base
of
a
pro
posed
IT
2
FLC
is
s
how
n
in
Table
2.
KM
al
gorithm
is
us
e
d
to
im
plem
ent
the
ty
pe
re
duct
ion
m
et
ho
d
of
the
IT
2F
LC
.
T
he
in
puts
an
d
ou
t
pu
t
gain
s
of
the
c
on
tr
oller
with
the
r
ules
wer
e
tun
e
d
t
o
reach
the
best
re
spo
nse
to
reac
h
m
i
nim
u
m
ov
ers
hoot,
m
ini
m
u
m
set
tl
ing
ti
m
e
a
nd
m
i
ni
m
u
m
e
rror
in
disti
ll
at
e
m
olar
f
racti
on.
The
f
eat
ur
es
of
us
i
ng
Ma
m
dan
i
ty
pe
a
re:
it
is
m
y
th
ought,
it
s
uit
s
hum
an
in
pu
t
and
it
is com
m
on
ly
accepted.
Figure
5. IT
2 m
e
m
ber
sh
ip
f
unct
ion st
r
uctu
r
e [
1
4
]
Figure
6
.
Input
/
Ou
tp
ut MFs
0
50
100
150
200
250
0
.
8
5
0
.
9
0
.
9
5
1
T
i
m
e
(
h
o
u
r
)
D
i
s
t
i
l
l
a
t
e
f
r
a
c
t
i
o
n
ii
i
iii
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2089
-
4856
IJRA
,
V
ol.
8
,
No.
3
,
Septem
ber 2
019
:
155
–
163
160
Table
2.
R
ule
base
of
IT
2FL
C
e/Δ
e
NB
NM
NS
Z
PS
PM
PB
NB
NB
NB
NB
NB
NM
NS
Z
NM
NB
NB
NB
NM
NS
Z
PS
NS
NB
NB
NM
NS
Z
PS
PM
Z
NB
NM
NS
Z
PS
PM
PB
PS
NM
NS
Z
PS
PM
PB
PB
PM
NS
Z
PS
PM
PB
PB
PB
PM
Z
PS
PM
PB
PB
PB
PB
A
P
I
-
li
ke
I
T
2FLC
pro
ba
bly em
plo
ye
d
to serve as
a PD s
uc
h
as
I
T
2F
LC
. As s
how
n
i
n
(
10)
sho
ws
a PD co
ntr
oller
equati
on r
es
ul
ts
in a
posit
ion
form
‚
howe
ver
(
11)
s
hows
a
PI
c
ontrolle
r
[1
5].
(
)
=
(
)
+KD
(
)
(10)
(
)
=
.(
(
)
+
(
)
)
(11)
wh
e
re
(
)
is
a
sig
nal
of
sa
m
pled
error
‚
∆
(
)
is
the
sa
m
pled
error
si
gn
al
rate
c
hange,
in
de
x
k
re
pr
ese
nt
s
the
present
sa
m
pl
ing
insta
nt,
(
)
is
a
con
tr
ol
act
ion
,
is
an
outp
ut
gain
,
(
)
is
the
previ
ou
s
a
n
con
t
ro
l act
io
n
and
(
)
is der
i
vative c
on
t
ro
l act
i
on.
3.2.
Desi
gn
of
t
he
contr
oller
log
ic
in
accu
r
at
e
type
1 (T1
FLC)
This
arti
cl
e
de
scribes
a
Ma
m
dan
i
T
1
P
I
Like
FLC
[
1]
is
pro
po
se
d
as
a
com
par
ison
wi
th
IT
2F
L
C
wh
e
re
it
s
houl
d
with
t
wo
in
puts
a
nd
one
ou
tpu
t.
The
e
ntries
are
goin
g
t
hro
ugh
a
proces
s
of
fu
zzi
ficat
ion
with
(5)
m
e
m
ber
sh
ip.
T
he
t
rap
ez
oi
dal
m
e
m
ber
sh
ip
was
sel
ect
e
d
in
the
e
xtre
m
es
and
t
rian
gu
la
r
m
e
m
ber
sh
ip
i
n
the
m
idd
le
a
nd
t
he
de
f
uzzif
ic
at
ion
sel
ect
e
d
as
t
he
m
et
h
od
is
Ce
nt
ro
i
d.
Mo
re
detai
ls
can
be
f
ou
nd
i
n
the r
e
fer
e
nce
1.
4.
RESU
LT
S
AND DI
SCUS
S
ION
The
c
ontrolle
d sy
stem
is test
e
d
acc
ordin
g
t
o t
he
f
ollo
wing s
cenar
i
os
:
−
Con
sta
nt D
ist
il
la
te
f
racti
on se
t
-
point (
0.9
8)
−
Con
sta
nt D
ist
il
la
te
f
racti
on se
t
-
point (
0.9
8)
with
disturba
nc
e
−
Step
Disti
ll
at
e
fr
act
io
n
Set
-
po
int
−
Step
Disti
ll
at
e
fr
act
io
n
Set
-
po
int wit
h dist
urb
ance
The
Disti
ll
at
e
fr
act
io
n
(XD)
set
-
point
is
sel
ect
ed
f
ro
m
a
Table
1
as
a
purity
of
98
%
,
the
tim
e
of
the
si
m
ulati
on
is
250
hours
and
the
funct
ion
of
the
c
ost
us
ed
f
or
the
evaluati
on
of
the
perform
a
nce
of
the contr
ol is
I
TAE
:
J=250
∫
|
|
25
0
0
(1
2
)
The
a
ppli
ed
disturbance
is
th
e
Feed
fl
ow
rat
e
(F)
a
nd
Fee
d
li
qu
id
m
olar
frac
ti
on
XF
[
1],
as
sho
wn
i
n
the
F
ig
ur
e
7
.
The
f
eed
flo
w
rate
is
represe
nted
as
a
Sine
wav
e
os
ci
ll
at
ion
with
lo
w
f
re
qu
e
ncy
a
nd
t
he
Feed
li
qu
id m
olar fr
act
ion
is
repres
ented
a
s a sm
oo
th
square
w
a
ve
w
it
h hig
h fr
e
qu
e
nc
y.
(a)
Feed
f
l
ow ra
te
(b)
Fee
d
li
quid
m
olar
fr
act
io
n
Figure
7
.
D
ist
urba
nces
on the
entire fe
ed
of t
he dist
il
la
ti
on
co
lum
n:
(a)
Feed
f
l
ow ra
te
(b)
Fee
d
li
quid m
olar
f
ract
ion
Evaluation Warning : The document was created with Spire.PDF for Python.
IJRA
IS
S
N:
20
89
-
48
56
In
te
rv
al type
-
2 f
uz
zy l
og
ic
c
ontroll
er
desig
n f
or
disti
ll
ation colu
m
n pr
oces
s
(
Moha
mm
e
d Y.
Ha
s
san
)
161
The
sim
ulatio
n
of
a
cl
os
e
d
lo
op
f
or
a
disti
ll
at
ion
process
us
in
g
IT
2F
LC
is
s
how
n
i
n
F
igure
8
a
nd
the sim
ulati
on
o
f
the cl
os
ed
lo
op for a
disti
ll
a
ti
on
c
olu
m
n
pr
ocess usi
ng I
T
1F
LC
is
sho
wn in F
i
gure
9
.
Fig
ure
8
.
Cl
os
e
d
lo
op syst
em
f
or a
disti
ll
at
ion
co
l
um
n
pr
oces
s u
si
ng I
T
2FL
C
Figure
9
.
Cl
os
e
d
lo
op syst
em
f
or a
disti
ll
at
ion
co
lum
n
pr
oces
s u
si
ng I
T
1FL
C
The
gai
ns
of
t
he
pro
portio
na
l
co
ntr
ol
(
KP),
De
rivati
ve
c
ontr
ol
(
K
D)
a
nd
the
outp
ut
gai
n
(Ko
ut)
are
tun
e
d
m
anu
al
ly
for
both
co
nt
ro
ll
ers
to
reac
h
the
best
ti
m
e
respo
ns
e
a
s
s
how
n
i
n
Table
3
t
hen
the
sim
ulati
on
respo
n
se
for
t
he
X
d
c
on
ce
ntr
at
ion
in
both
c
on
sta
nt
set
-
point
an
d
ste
p
set
-
point
with
dis
tur
ban
ce
an
d
without
disturba
nce
f
o
r
a
Ty
pe
-
1
FL
C
are
s
how
n
i
n
Fig
ur
e
10
a
nd
Fi
g
ure
11
r
especti
vely
a
nd
f
or
an
IT2F
LC
are
sh
ow
n
in
Fi
g
ur
e
12 a
nd Fig
ur
e
13
resp
ect
iv
e
ly
.
(a)
Step
set
-
poi
nt only
(b)
Ste
p
set
-
po
int wit
h dist
urb
ance
Figure
11. Tim
e traject
or
y
res
pons
e
of the
sy
stem
X
d
c
onsiderati
on
with
f
uzzy Ty
pe
-
1 f
or
(a)
Step
Set
-
point o
nly (
b)
Step Set
-
point
with
disturba
nce
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2089
-
4856
IJRA
,
V
ol.
8
,
No.
3
,
Septem
ber 2
019
:
155
–
163
162
Table
3.
T
une
d gain
s fo
r
I
T2L
C
Ty
p
es
Gain
s
PI
-
Like
I
T2F
LC
W
ith
con
stan
t set
-
p
o
in
t on
ly
PI
-
Like
I
T2F
LC
W
ith
con
stan
t set
-
p
o
in
t and
d
istu
rban
ce
PI
-
Like
I
T2F
LC
W
ith
step
set
-
p
o
in
t on
ly
PI
-
Like
I
T2F
LC
W
ith
step
set
-
p
o
in
t
an
d
dis
tu
rban
ce
KP
5
1
7
.5
5
8
.17
9
KD
5
9
.98
5
3
.41
9
7
Ku
10
9
.8
30
8
.9
(a)
C
onsta
nt se
t
-
point
on
ly
(b)
C
on
sta
nt se
t
-
point
with
disturbance
Figure
12. Tim
e traject
or
y
res
pons
e
of the
sy
stem
X
d
c
onsiderati
on
with
f
uzzy Ty
pe
-
2 f
or (
a
)
C
onsta
nt s
et
-
po
i
n
t o
nly (
b)
Con
sta
nt set
-
point
with
disturbance
(a)
Step
Set
-
point
only
(b)
Ste
p
Set
-
po
int wit
h dist
urb
ance
Figure
13. Tim
e traject
or
y
res
pons
e
of the
sy
stem
X
d
c
onsiderati
on
with
f
uzzy Ty
pe
-
2
(a)
Step
set
-
poi
nt only
(b)
Ste
p
set
-
point
with
disturba
nce
By
app
ly
ing
12,
t
he
pe
rce
ntage
of
e
nhance
m
ent
betwee
n
the
IT
2F
LC
c
om
par
ed
with
IT1FLC
f
o
r
diff
e
re
nt cases
are li
ste
d
i
n
Ta
ble 4.
Table
4
.
P
erce
ntage o
f
e
nh
a
nc
e
m
ent b
et
wee
n
the
I
T
2FLC
com
par
ed wit
h T1
FLC
Co
n
troller
Co
n
stan
t
Set
-
p
o
in
t
Step
Set
-
p
o
in
t
p
erce
n
tag
e of
enh
an
-
ce
m
en
t
f
o
r
co
n
stan
t Set
-
p
o
in
t
Percentag
e of
en
h
an
ce
m
en
t
Step
in
p
u
t
Ty
p
e
-
2
7
6
.23
286
~8
5
%
~5
3
%
Ty
p
e
-
2
with
d
istu
rban
ce
1
4
1
.1
4
7
0
.6
~8
0
%
~7
4
%
Ty
p
e
-
1
5
0
6
.8
604
Ty
p
e
-
1
with
d
istu
rban
ce
7
3
7
.3
1831
Evaluation Warning : The document was created with Spire.PDF for Python.
IJRA
IS
S
N:
20
89
-
48
56
In
te
rv
al type
-
2 f
uz
zy l
og
ic
c
ontroll
er
desig
n f
or
disti
ll
ation colu
m
n pr
oces
s
(
Moha
mm
e
d Y.
Ha
s
san
)
163
5.
CONCL
US
I
O
N
The
a
ppli
cat
ion
of
I
T2
P
I
-
Like
F
LC
to
the
nonlinea
r
dynam
ic
al
m
o
del
of
a
c
onti
nuous
ty
pe
disti
ll
at
ion
co
l
um
n
dem
on
str
at
es the effect
ivene
ss in reac
hi
ng
a t
op contr
ol p
e
rfor
m
ance
and all
owin
g
a
f
ast
er
and
e
xact
co
nt
ro
l
of
the
proc
ess.
The
distu
r
ban
ce
re
j
ect
ion
and
set
po
i
nt
tracki
ng
is
done
with
le
ss
am
ount
of
ov
e
rs
hoot
as
a
com
par
ed
w
it
h
us
in
g
Ty
pe
-
1
PI
-
Like
F
LC.
Sim
ulati
o
ns
ca
rr
ie
d
out,
showe
d
a
n
a
ver
a
ge
i
m
pr
ovem
ent
m
easur
ed
us
in
g
IT
AE
of
85%
f
or
a
co
ns
t
ant
set
-
point
without
distu
r
ban
ce
a
nd
80%
with
a d
is
tu
rb
a
nce,
for a st
ep
set
-
point t
he
im
pr
ove
m
ent is 53%
w
it
ho
ut
distu
rba
nce a
nd 74%
wi
th d
ist
urba
nce
.
REFERE
NCE
S
[1]
S.
Dia
z
,
J.
R.
Pere
z
-
Corre
a,
A
.
C
ipri
ano
and
M.
Ferna
nd
ez
“
Inte
l
li
gen
t
C
ontrol
Appl
ications
on
a
B
ina
r
y
Distil
lation
Col
um
n,
”
IE
EE
Int
ernati
onal
Conf
ere
nce
de
la
Associat
ion
Chil
en
a
de
Control
A
utomati
c
,
Ta
l
ca,
Chil
e
,
19
-
21
Oct
.
2016
.
[2]
K.
W
ora
pra
d
y
a
and
S.
Pra
ti
shth
ana
nda
“
Rea
l
Tim
e
Control
of
a
Bina
r
y
Dist
il
l
ation
Colum
n
usi
ng
HG
A
Fuzz
y
Supervisor
y
PI
Control
le
rs
,
”
Tencon.
,
pp.
8
14
-
8
18,
2005
.
[3]
G.
M.
Murlidha
r
and
A.
K.
Jan
a
“
Nonline
ar
Ada
pti
ve
Con
trol
Al
gorit
hm
for
a
M
ult
ic
om
ponen
t
B
at
ch
Dist
il
l
at
ion
Colum
n,
“
Chemical
Eng
ineeri
ng
Scie
n
ce 62, pp.
111
-
1124,
2007
.
[4]
J.
Ferna
nde
z
d
a
Cane
t
e,
P.
Sa
z
-
Orozc
o
and
S.
Gonza
lez
-
Pere
z.
“
Applic
ati
on
of
Adapti
v
e
Neur
o
Fuzz
y
Contro
l
using Soft
Senso
rs t
o
Cont
inuous
Distil
l
ation,
”
C
om
pute
r
Aided
Chemica
l
Engi
n
ee
ring
25,
pp.
46
5
-
470,
2008
.
[5]
S.
Java
di
and
J
.
Hos
seini
“
Control
of
a
Bin
ar
y
D
isti
llati
on
Colum
n
using
PI
Contr
oll
er
,
”
9
th
W
SE
AS
Inte
rna
t
iona
l
C
onfe
ren
c
e
on
S
igna
l
Proc
ess
ing
Com
puta
ti
ona
l
Geom
et
r
y
and
A
rti
fi
ci
a
l
Vision,
W
orld
Scie
nti
f
ic
and
Engi
n
ee
r
in
g
Aca
dem
y
and
Societ
y
(W
SEAS
),
pp.
145
-
152,
20
09.
[6]
F.
Barcel
o
-
Ri
co,
J.
M.
Goza
lvez
-
Za
fri
ll
a
and
J.
L.
Di
ez
“
Modeling
and
con
trol
o
f
a
con
ti
nuous
d
isti
llati
on
tower
through
fuz
z
y
technique
s,
”
Ch
e
m
ic
al
Engi
ne
ering Re
sea
r
ch and
Design
89
,
pp
.
107
-
115,
2011
.
[7]
R.
Siv
akumar,
S
.
Man
ic,
V.
Ner
thi
ga
,
R.
Akil
a
and
K
.
B
al
u
“
Applicati
on
of
Fu
zzy
Mod
el
Pred
ic
ti
v
e
Control
in
Multi
var
i
able
C
ontrol
o
f
Dist
il
l
a
ti
on
Co
lum
n,
”
I
nte
rnational
Jou
rnal
of
Chemi
ca
l
Eng
ineering
an
d
Applica
ti
ons
1
,
pp.
38
-
42
,
2010
.
[8]
Y.
AL
-
Dunainaw
i
and
M.
F
.
A
bbod
“
Evol
uti
on
ar
y
Based
Opti
m
iz
at
ion
o
f
Mul
ti
var
i
able
fuz
z
y
cont
rol
S
y
stem
of
a
Bina
r
y
Dist
i
ll
ation
Colum
n,
”
Int
ern
a
ti
ona
l
conf
er
enc
e
o
n
co
m
pute
r
Modeli
ng
and
Sim
ula
ti
on
,
UK
,
pp.
127
-
132
,
20
16.
[9]
M.
Sharm
il
a and
V.
Manga
r
y
a
rka
rasi
“
Modeli
ng
a
nd
Control
of
B
i
nar
y
Distil
la
t
ion Colum
n,
”
Elec
tr
ic
a
l,
El
e
ct
ron
ic
,
and
Instrum
entat
ion
Eng
ine
er
ing, a
n
ISO
C
ert
if
ied Organi
z
at
io
n,
Vol.
3
,
pp
.
105
-
111,
20
14
.
[10]
Y.
AL.
Dunain
a
wi
and
M.
F.
Abbod
“
PSO
-
PD
Fuzz
y
Con
trol
o
f
Distil
lation
Co
lumn,”
SA
I
Intelli
gen
t
S
y
st
em
Confer
ence, L
on
don,
UK
,
pp
.
55
4
-
558,
2015
.
[11]
A.
George
and
R.
M.
Franc
is
“
Model
Ref
ere
n
c
e
Adapt
ive
Con
trol
of
Bin
ar
y
Distil
lation
Col
um
n
Com
positi
on
using MIT
Adap
ti
ve
Mec
han
ism
,
”
Eng
ine
e
ring
R
ese
arc
h
and
T
echnolog
y
4,
pp.
5
55
-
558,
2015
.
[12]
H.
Ying,
“
Fuzz
y
Control a
nd
Modeli
ng
,
Ana
l
y
tic
al
Founda
ti
ons a
nd
Applicati
ons
,
”
W
ie
l
y
-
IEEE
Pr
ess,
US
A,
2000.
[13]
M.
B.
Oze
k
and
Z.
H.
Akpol
at
“
A
Software
Tool
T
y
pe
-
2
Fuz
z
y
L
ogic
Tool
box
,
”
Com
put
-
APPI
-
E
ng
-
Edu,
Vol.
16,
No.
2,
pp.
137
-
1
46,
2008
.
[14]
Bart
olomeo
Cos
enz
a
“
Control
o
f
a
Dist
il
l
at
ion
Colum
n
b
y
T
y
p
e
2
and
T
y
p
e
1
Fuzz
y
Logi
c
PID
Control
l
er,
“
Journal
of
Pro
ces
s Control
le
r
24
,
pp
.
475
-
484
,
2
014.
[15]
Moham
m
ed
Y.
Hass
an
and
Sa
i
f
Faris
Abulha
il
“
Design
of
FP
G
A
Based
Int
erv
a
l
T
y
p
e2
Fuz
z
y
Logi
c
Controller
,
”
M.
Sc.
the
sis,
C
ontrol
and
S
y
s
tem
Engi
nee
r
ing D
epa
rtment at
Univer
sit
y
of Technolog
y
,
Ir
aq, 2
017.
Evaluation Warning : The document was created with Spire.PDF for Python.