TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.7, July 201
4, pp
. 5350 ~ 53
6
1
DOI: 10.115
9
1
/telkomni
ka.
v
12i7.577
8
5350
Re
cei
v
ed Fe
brua
ry 12, 20
14; Re
vised
Ma
rch 24, 20
14; Accepted
April 10, 201
4
Design of Decentralized Controller for Interactive
Processes through Relative Frequency Array
R. Hanum
a Naik*
1
, D.V. Asho
k Kum
a
r
2
, K.S.R. Anjane
y
u
lu
3
*
1
RGM Colle
ge
of Engine
eri
n
g
&T
echnol
og
y,
Nan
d
y
al, AP, Indi
a
2
S
y
ama
l
a
devi I
n
stitute of T
e
chno
lo
gy
for
w
o
men, Nandy
al,
AP, India
3
Ja
w
a
h
a
rla
l
Ne
hru T
e
chnol
ogi
cal Univ
ersit
y
Ananta
pur, An
antap
uram, AP
, India
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: rhnaik.1
7
1
7
@
gmai
l.com
A
b
st
r
a
ct
T
uni
ng of c
o
n
t
rollers for
mul
t
ivaria
ble
proc
ess is
a
difficu
lt task beca
u
s
e
of inter
a
ctio
n
invo
lve
d
amon
g the
vari
abl
es. In this
p
aper, a
si
mp
le
tunin
g
stra
tegy
is us
ed for
de
sign
of a
multi-
loo
p
PI co
ntroll
e
r
to ach
i
ev
e
de
sired
frequ
enc
y resp
ons
e fo
r inter
a
ctiv
e
mu
ltivari
a
b
l
e p
r
ocesses.
T
o
han
dle
the
l
o
op
interacti
on, a
p
a
irin
g of c
ontr
o
lle
d var
i
a
b
le
and
mani
pu
lte
d
vari
abl
e is
d
e
termin
ed
bas
ed o
n
the
inte
grity
obtai
ne
d from effective relat
i
ve ga
in (ERG
) and Ni
ed
erli
nski
’
s
Ind
e
x (
N
I).T
hen the
Equiv
a
le
nt transfer
function
of mo
del is a
ppr
oxi
m
ated by
the
us
e of relativ
e
freque
ncy array (
R
F
A
), Relative
Gain Array (R
GA).
F
i
nally
a
Dec
e
ntrali
z
e
d PI c
o
ntroll
er is
des
i
gne
d fo
r s
u
g
g
e
sted
pair
of v
a
ria
b
les
to
ach
i
eve
des
ired
g
a
in,
phas
e
mar
g
i
n
s
.
T
he p
e
rfor
ma
nce
is ver
i
fied
on th
e i
n
d
u
stri
al
multivar
iab
l
e
proc
esse
mod
e
ls to
show
th
e
effectiveness
o
f
the propos
ed
meth
od. T
he r
e
sults cle
a
rly r
e
vea
l
that, it gives better perf
o
rmanc
e for set
poi
nt chan
ges
and d
i
sturb
anc
e rejecti
on. Si
mu
lati
on res
u
lt
s are incl
ud
ed
to valid
ate the
robustn
ess of the
prese
n
ted a
l
go
rithm.
Ke
y
w
ords
:
nteractive proc
ess,
effective relativ
e
ga
in
array (ERGA), relativ
e
frequency array (
R
FA),
Nied
e
rl
inski
’
s
Index, po
ly
meri
z
a
ti
on re
actor, ind
epe
nd
ent control
l
er
.
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
The
obje
c
tive of the
control en
gine
er i
s
t
uni
ng of controlle
r para
m
eters. Ma
n
y
of the
indu
strial p
r
o
c
e
s
ses a
r
e
consi
s
ting of
multi-inp
u
t an
d multi-outp
u
t
(MIMO) pro
c
e
s
ses. Tracking
of desi
r
ed
pe
rforma
nce for these
kin
d
s
of pro
c
e
s
se
s is
v
e
ry
compl
i
cat
ed co
mpa
r
ed wit
h
singl
e
input/output p
r
ocesse
s be
caus
e of intera
ction
s
existin
g
among the
variable
s
[1].
In the literatu
r
e, the Multi-l
oop PI/PID cont
rol u
s
in
g multiple si
ngl
e input si
ngl
e output
(SISO) PI/PID controll
ers
is commo
nly use
d
for
co
ntrolling MIMO pro
c
e
s
ses wit
h
intera
ction
[
2
-
5]. The
reas
on for us
ing SI
SO PI/PID c
o
ntroller for
M
u
lti-loop control is its
simpl
e
st
ru
ct
ure,
e
a
sy
tuning and
ability to achi
eve most
of
the
expected control objecti
ves. It is the
comm
on
scenario
to extend the controlle
r design met
hod
s
of SISO systems to multi-lo
op system
s, but it affects the
perfo
rman
ce
and
stability of the syste
m
s [6, 7].
Many
method
s [8-11] have b
e
e
n
pro
p
o
s
ed
o
v
er
the peri
od, consi
deri
ng lo
op interactio
n
s
into a
c
cou
n
t
in the multi-l
oop
control d
e
sig
n
. But ea
ch
method h
a
s
its own me
rits an
d dem
erits [12]. Th
e
s
e meth
od
s
con
s
id
ers th
e intera
ction
s
in
seq
uential,
re
quire
minimu
m pro
c
e
s
s in
formati
on, b
u
t
tuning
seq
u
ence ha
s to
be repe
ated f
o
r
corre
c
t sequ
ence if the d
e
sig
n
se
que
n
c
e i
s
not
pro
per [13]. In indep
ende
nt desi
gn meth
ods
SISO cont
roll
ers a
r
e
de
si
gned
ind
epe
ndently by
u
s
ing
the defi
ned bou
nda
ri
es
to
gu
aran
te
e
stability and
perfo
rman
ce
[14-18]. But t
he detail
ed
in
formation
ab
out the
controller dyn
a
mi
cs in
other loo
p
s is not
co
nsi
dered, th
e
re
sulti
ng p
e
rfo
r
ma
n
c
e
may b
e
p
o
or. In thi
s
,
an
indep
ende
nt
PI
controlle
r is d
e
sig
ned fo
r specifi
c
gain
a
nd pha
se
m
a
rgin
s. The int
e
ra
ction am
o
ng the varia
b
l
e
s
are d
e
termi
n
ed usi
ng rela
tive gain arra
y (RGA)
and
effective rel
a
tive gain array (ERGA
)
.Then
the decentral
i
zed
controll
er is de
sig
n
ed for t
he p
a
irs of ma
ni
pulated vari
a
b
le (inp
ut) and
controlled
va
riable
(output
)
sug
geste
d
by ERGA
to
achieve th
e
de
sire
d
perf
o
rma
n
ce of t
he
intera
ctive proce
s
s.
To dete
r
mine
the optimum
setting
s of M
u
ltivar
iable PI
cont
rolle
r, a
con
s
i
s
tent m
e
thod i
s
use
d
, the clo
s
ed lo
op p
e
rf
orma
nce obt
ained in th
i
s
method i
s
co
mpared with
existing meth
ods
and re
sult
s are discu
s
sed.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
De
sign of De
centralized Controlle
r for Intera
ctive Proce
s
se
s thro
ugh… (R. Ha
num
a Naik)
5351
2. Nota
tion
The notation
use
d
throu
g
h
out the pape
r is stated bel
ow.
Indexes:
()
c
Gs
Controlle
r tra
n
sfer fu
nction
in s- dom
ain
()
p
Gs
Process tran
sfer function in
s-do
main
Relative gain
array
E
Effective energy transmi
ssi
on ratio a
rray
Effective relative gain array
(
ERGA
)
m
A
Gain margin
m
Phase m
a
rgi
n
c
K
Propo
rtional controlle
r
gai
n
i
T
Integral time
con
s
tant
i
K
Integral controller gai
n
T
i
me
c
o
n
s
tant
d
t
Time delay in
second
s
N(G) Nied
erlin
ski’s
index
Hadm
ard pro
duct
c
Critical freq
u
ency
e
Deviation b
e
twee
n output
and de
sired i
nput
Array of criti
c
al freque
ncy
Λ
e
ij
effective ene
rgy tran
smi
s
sion
ratio
be
tween
output
variable
an
d input
variable
whe
n
all other loo
p
s are clo
s
e
d
()
g
s
ij
Tran
sfe
r
function of output j to input i in s-dom
ain
3. Problem Formulation
3.1. Interac
t
iv
e
Processe
s
MIMO contro
l problem
s
a
r
e i
nhe
rently
mo
re
com
p
l
e
x than SIS
O
control p
r
oblems
becau
se pro
c
ess
inte
ra
ctio
n
o
c
curs bet
wee
n
controll
ed
va
riable
s
and manip
u
la
ted
vari
able
s
.
In
gene
ral a
cha
nge in ma
nip
u
lated varia
b
l
e
, say u
1
, will affect all of the controlled
variables y
1
, y
2
,
y
3
… y
n
. Because of the
intera
ction
s
,
the sele
ct
ion of the b
e
st pai
ring
of cont
rolle
d a
nd
manipul
ated
variable
for
multi-loo
p
co
ntrol
scheme
can
be
a dif
f
icult task. In
parti
cula
r, fo
r a
control probl
em with n co
ntrolled va
ria
b
les a
nd
n
manipul
ated
variable, the
r
e are n! po
ssible
multi-loop control configurations.
A schemati
c
rep
r
e
s
entati
on of SISO
and MI
MO
control
con
F
ig
ureu
ratio
n
sh
own i
n
Figure 1.
Fo
r co
nvenie
n
ce
, it is
assu
me
d that th
e n
u
m
ber of m
ani
pulated
varia
b
les is eq
ual
to
the nu
mbe
r
o
f
cont
rolled
v
a
riabl
es.
Thi
s
allo
ws pai
rin
g
of
singl
e
co
ntrolled
vari
a
b
le a
nd
a
sin
g
le
manipul
ated
variable
thro
ugh
a fee
dba
ck controller.
For pai
ring
of variabl
es,
two m
e
thod
s are
use
d
namely
RGA and E
R
GA. These two are di
scussed in se
ction
4.
Figure 1. Single-Inp
u
t and
Single Output
(SISO) Process
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5350 – 53
61
5352
Figure 2. Clo
s
ed L
oop Inte
ractive Multiv
ariabl
e Pro
c
e
s
s
Con
s
id
er the clo
s
ed loo
p
stable intera
cti
v
e mu
ltivariable system a
s
sh
own in Fi
gure 1.
Gene
rali
zed t
r
an
sfer fun
c
ti
on matrix of the pro
c
e
s
s is:
g(
s
)
g
(
s
)
.
.
.
g
(
s
)
p11
p12
p1n
g(
s
)
g
(
s
)
.
.
.
g
(
s
)
p21
p22
p2n
G(
s
)
=
p
.
.
.
..
.
.
..
..
.
g(
s
)
g
(
s
)
.
.
.
g
(
s
)
pnn
pn1
pn2
(1)
Whe
r
e the p
r
oce
s
s
G(
s
)
p
ij
is first
orde
r process with delay time (FOP
DT),
i.e.
-
τ
s
K
d
g(
s
)
=
e
pi
j
Ts
+
1
(2)
And the stru
cture of full dimensi
onal
co
ntrolle
r is of the form,
g
(
s)
g
(
s)
...
g
(
s)
c1
1
c
1
2
c1
n
g
(
s)
g
(
s)
...
g
(
s)
c2
1
c
2
2
c2
n
G(
s
)
=
c
..
.
.
..
...
..
.
g
(
s)
g
(
s)
...
g
(
s)
cn
n
cn
1
c
n
2
(3)
Whe
r
e the
co
ntrolle
r is of the form:
1
g(
s
)
=
K
+
+
K
s
c
,
ji
p
,
ji
d
,
ji
Ks
i,
ji
(4)
The co
ntroll
er output and pl
ant output are given by:
u=
G
e
a
n
d
y
=
G
u
cp
ii
i
i
(5)
Whe
r
e,
u
(
i
=
1,
2,
3,
..
...
n
)
,
y
(i
=
1
,
2
,
3
,
.
.
..n)
a
n
d
e
=
y
-
y
ii
i
s
p
i
i
are i
nputs of the
plant, output
of the
plant and e
r
ror sig
nal to the controlle
r re
spe
c
tively.
In practical,
when MIMO
control loop is cl
osed, there exist intera
ction
s
amo
n
g
the loop
s
as a re
sult of
existen
c
e of non-ze
ro di
ag
onal el
em
ent
s in the pro
c
e
ss tran
sfer fu
nction mat
r
ix.
4.
Interaction Anal
y
s
is
4.1. Relativ
e
Gain Arra
y
(
R
GA)
Interactio
n a
nalysi
s
is m
o
st wid
e
ly use
d
tech
niqu
e i
n
co
ntrol
syst
em co
nFig
ureuratio
n
for multivari
a
ble p
r
o
c
e
s
se
s. The
mo
st
comm
only u
s
ed meth
od fo
r inte
ractio
n
measurement
is
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
De
sign of De
centralized Controlle
r for Intera
ctive Proce
s
se
s thro
ugh… (R. Ha
num
a Naik)
5353
RGA. Base
d
on intera
ctio
n matrix obta
i
ned with
RG
A, the pairing
of manipulat
ed variabl
e a
n
d
controlled va
riable is forme
d
.
The
RGA is
defined
as fol
l
ows: Let K=G (0
) be
the
matrix of ste
ady state g
a
i
n
s of th
e
trans
fer func
tion matrix
G(
s
)
p
.
l
i
m
G
(s
)
=
[K
]
p
s0
(6)
Furthe
r let R
be the tran
sp
ose of inverse matrix K.
-T
(0
)
-1
T
R=
[
K
]
=
[
G
]
Therefore, RGA for
nn
sy
st
e
m
s is,
λλ
..
.
λ
11
12
1n
λλ
..
.
λ
-T
21
22
2
n
λ
=
G
(0
)
G
(0
)
=
λλ
..
.
λ
3
1
32
3n
λλ
..
.
λ
nn
n1
n2
(7)
Thus from th
e Equation
(7
) it is p
o
ssibl
e
to de
scrib
e
the level of i
n
tera
ction, la
rge valu
e of
λ
ij
mean
s that there i
s
stro
n
g
inte
ra
ction
betwe
en co
rresp
ondi
ng in
puts
i
and ou
tput
j
. If the
value
of
λ
ij
is greate
r
than 0.5 an
d approa
che
s
towa
rd
s un
ity,
the intera
ction al
so le
ads b
e
twe
en
corre
s
p
ondin
g
pairs. In general
λ
1
ij
is the ideal ca
se
for pairing
and avoidin
g
negative
pairin
g
. The RGA is de
pe
nd upon
stea
dy state
gain
s
and mo
st suitabl
e for n
online
a
r pla
n
t
s
operating aro
und steady state
poin
t. Th
e neg
ative di
agon
al in the
RGA m
a
trix
gives
sufficie
n
t
con
d
ition for i
n
stability. Th
e pairi
ng whi
c
h lea
d
s th
e instability is a
v
oided by usi
ng Nie
derli
nski’s
theore
m
. The
Niede
rlin
ski’
s
index (NI) f
o
r the c
ontrol
stru
cture s a
bove is d
enot
ed by N (G)
and
defined a
s
:
G(
0
)
N
(
G
)
=
i
,
j
=
1
,
2
,
3
,
....
.n
π
g
ij
(8)
Whe
r
e
G(
0
)
denot
es dete
r
min
a
n
t matrix G(0
)
and
π
g
ij
denote
s
produ
ct of diago
nal ele
m
ents of
G(0
)
fo
r a
full
y cent
ralized
control
syste
m
. For a
stabi
lity of com
p
le
x nonlin
ear system, NI
sh
o
u
ld
be gre
a
ter th
an ze
ro. For
dynamical intera
ction stu
d
y
, the effective relative gai
n array (ERG
A) is
use
d
.
4.2. Effec
t
iv
e
RGA
The ene
rgy transmi
ssion
ratio of
g(
s
)
ij
is expressed a
s
:
eg
(
0
)
ω
i
,
j
=
1
,
2
,
...
.....
n
ij
ij
c
,
i
j
Whe
r
e,
ω
c,
ij
is th
e
critical
freq
u
ency
of the t
r
ansfe
r fu
nctio
n
. For the
over
all
system,
the e
n
e
r
gy
transmissio
n ratio ca
n be e
x
presse
d by effectiv
e energy transmi
ssi
on ratio a
rray
and given by
:
ee
.
.
.
e
11
12
1n
ee
.
.
.
e
21
2
2
2n
E=
=
G
(
0
)
Ω
.
.
.
...
.
.
.
.
..
ee
.
.
.
e
nn
n1
n
2
(9)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5350 – 53
61
5354
Whe
r
e,
g
(
0)
g
(
0)
.
.
.
g
(
0)
11
12
1n
g
(
0)
g
(
0)
.
.
.
g
(
0)
21
22
2n
G(
0
)
=
...
...
...
...
g
(
0)
g
(
0)
.
.
.
g
(
0)
nn
n1
n2
And,
ωω
..
.
ω
c1
1
c
1
2
c1
n
ωω
..
.
ω
c2
1
c
2
2
c2
n
Ω
=
..
.
.
.
.
..
.
...
ωω
..
.
ω
cn
n
cn
1
c
n
2
Are the stea
dy state gai
n array and
the criti
c
al
freque
ncy arra
y resp
ectivel
y
. The effective
relative gain
array can b
e
defined a
s
:
e
ij
Φ
=
ij
Λ
e
ij
(10)
Whe
r
e,
Λ
e
ij
is the effective ene
rgy tran
smi
s
sion ratio b
e
tween outp
u
t variabl
e and i
n
put variable
whe
n
all othe
r loop
s are cl
ose
d
. Over al
l ERGA (
Φ
) can be calcula
t
ed as:
-T
Φ
=E
E
(11)
The ERGA b
a
se
d loop p
a
i
r
ing rule
s req
u
ire
s
that ma
nipulate
d
and
controlled va
riable
s
in the main lo
op be pai
red
by those pai
rs wh
ose
ERG
A
and NI values are po
sitive and cl
osest
to
1.0.
The rel
a
tive freque
ncy arra
y can be written as:
-T
=
RFA
(12)
5. Dete
rmine
Equiv
a
lent Trans
f
er Fun
ction for In
te
grit
y
To reve
al the
model
relatio
n
s b
e
twe
en when all lo
op
s
are o
pen
and
all loop
s are
clo
s
ed,
we first d
e
fin
e
the
relative
criti
c
al f
r
equ
ency,
ij
γ
as th
e
ratio
of loo
p
y
i
-u
i
critical freque
nci
e
s
betwe
en whe
n
all other loo
p
s an
d wh
en
other loo
p
s a
r
e clo
s
ed.
c,
i
j
ij
c,
i
j
ω
γ
ˆ
ω
Whe
r
e
c,i
j
ˆ
ω
is the critical freq
uen
cy of l
oop
i-j when oth
e
r
loop
s are cl
ose
d
.
We obtai
n the formula for
cal
c
ulatin
g:
ij
ij
γ
=
λ
ij
(13)
Whe
n
the relative criti
c
al frequ
en
ci
es a
r
e
calculated for al
l the input-o
utput pair o
f
a
multivariable
pro
c
e
ss,
it result
s
in an array
of
the
form, i.e, rel
a
tive criti
c
al
freque
ncy
array
(RCFA) d
e
fin
ed as:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
De
sign of De
centralized Controlle
r for Intera
ctive Proce
s
se
s thro
ugh… (R. Ha
num
a Naik)
5355
11
12
1n
21
2
2
2n
n1
n
2
nn
γγ
...
γ
γγ
...
γ
Γ
=
..
.
...
...
...
γγ
...
γ
And eleme
n
ts can be d
e
termined by:
Γ
=
Λ
(14)
By assi
gning
the equival
e
n
t
transfe
r fun
c
tion (ETF
) wh
en othe
r lo
op
s a
r
e
clo
s
ed t
o
have
the sam
e
st
ructures
as th
e ope
n loop t
r
an
sfer fu
ncti
on, we
ca
n a
pproxim
ate ETFs in te
rm
s
of
relative gain
s
and relative
critical freq
ue
nc
ie
s when th
e control syst
em is clo
s
e
d
as:
ij
ˆ
-
θ
s
0
ij
i
j
ij
ˆˆ
g
(
s
)
=
g
(0)g
(s)e
(15)
Whe
r
e
ij
ˆ
θ
is the time delay of ETF.
The ETF wh
e
n
other loo
p
s
clo
s
ed b
e
co
mes:
ij
-
θ
s
ij
ij
ij
(0
)
ˆ
g(
s
)
=
g
(
0
)
e
λ
ij
ij
g
(16)
6. Contr
o
ller Design
The g
a
in a
n
d
pha
se m
a
rgi
n
are typical
contro
l lo
op
spe
c
ification
s
asso
ciated
with the
freque
ncy
re
spo
n
se analy
s
is. Th
e gai
n
and ph
ase
margi
n
have
alway
s
se
rv
ed a
s
obje
c
ti
ve
measure of robu
st perfo
rmance of
the process. It is kno
w
n from
cl
a
ssi
cal
cont
rol theo
ry that the
pha
se m
a
rgi
n
is
relate
d
to the dam
pi
ng of th
e
sy
stem a
nd
ca
n ther
efore
also se
rve
a
s
perfo
rman
ce
measure. Th
e controll
er d
e
sig
n
satisfyi
ng the
gai
n a
nd p
h
a
s
e m
a
rgin
criteria
is not
new, and thu
s
widely u
s
e
d
in industri
a
l
applicatio
n. In this pape
r a simple PI controlle
r tuni
ng
formula
s
a
r
e
develop
ed fo
r intera
ctive
pro
c
e
s
ses
wi
th time delay to meet desi
r
ed g
a
in ma
rgin
and ph
ase margin
spe
c
ifications.
6.1. PI Contr
o
ller Tuning
PI controlle
r for First ord
e
r
pro
c
e
ss
with time delay (F
OPDT
) is:
1
G(
S
)
=
K
(
1
+
)
c
Ts
i
c
(17)
-
τ
s
K
d
G(
S
)
=
e
p
Ts
+
1
Then
1K
-t
j
ω
di
i
G(
j
ω
)G
(
j
ω
)=
K
(
1
+
)
e
cp
c
ii
ii
Tj
ω
Tj
ω
+1
ii
i
i
i
i
with
ω
gi
i
being foun
d
from
G(
j
ω
)G
(
j
ω
)=
1
ci
i
p
i
i
a
n
d
ω
p
ii
found
from
G(
j
ω
)G
(j
ω
)=
-
π
cp
ii
ii
,
then
22
KK
1
+
ω
T
c
-1
-1
ii
i
G(
j
ω
)G
(j
ω
)=
-
0
.
5
π
+t
a
n
ω
T-
t
a
n
ω
T-
ω
T
cp
ii
ii
ii
i
i
i
i
i
22
ω
T1
+
ω
T
ii
i
The pha
se m
a
rgin:
-1
-1
=
π
-0
.
5
π
+t
a
n
ω
T-
t
a
n
ω
T-
ω
T
m
i
i
g
ii
g
i
i
g
ii
(18)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5350 – 53
61
5356
With
ω
gi
i
given b
y
the solution
of:
22
KK
c
1
+
ω
T
ii
i
ω
==
1
gi
i
22
ω
T1
+
ω
T
ii
i
i
i
(19)
The gain m
a
rgin:
22
ω
T1
+
ω
T
1
ii
i
i
i
A=
=
mi
i
22
G(
j
ω
)G
(j
ω
)
KKc
1
+
ω
T
cp
ii
ii
ii
i
(20)
With
ω
p
ii
given by the solution
:
-1
-1
ω
=-
0
.
5
π
+t
a
n
ω
T-
t
a
n
ω
T-
ω
T=
-
π
p
i
i
p
ii
i
p
ii
p
i
i
(21)
If K
c
and T
i
are desi
gne
d a
s
follows:
aT
K=
ci
i
K
τ
(22)
And,
T=
T
ii
i
(23)
Then Equ
a
tio
n
(21
)
be
com
e
s:
π
-0
.
5
π
-
ωτ
=-
π
i.
e
.
ω
=
pi
i
p
i
i
2
τ
sub
s
tituting into Equation
(20
)
gives:
π
T
A=
a
n
d
K
K
ω
Ti
.
e
.
ω
=K
K
T
mi
i
c
i
i
g
i
i
g
i
i
c
i
i
2K
K
τ
ci
i
And,
=0
.
5
π
-K
K
τ
/T
c
mi
i
.
Then,
π
A=
mi
i
2a
(24)
And,
=0
.
5
π
-a
mi
i
(25)
For conveni
e
n
ce valu
e of a is sel
e
cte
d
as
π
/2,
π
/3,
π
/4 and
π
/6.
Some typical
tuning rul
e
s b
a
se
d on different pha
se an
d gain ma
rgin
s are
sho
w
n i
n
the
Table 1.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
De
sign of De
centralized Controlle
r for Intera
ctive Proce
s
se
s thro
ugh… (R. Ha
num
a Naik)
5357
Table 1. Typi
cal PI Control
l
er Tuni
ng Ru
les for Interactive Proce
s
se
s
a K
c i
i
T
i ii
A
m i
i
φ
m i
i
π
/2
1.57T
ii
/ K
τ
d
T
ii
1.0
0
π
/3
1.047T
ii
/ K
τ
d
T
ii
1.5
π
/6
π
/4
0.785T
ii
/ K
τ
d
T
ii
2.0
π
/4
π
/6
0.524T
ii
/ K
τ
d
T
ii
3.0
π
/3
6.2. Perform
a
nce Inde
x
In the desig
n of PI controller, the performan
ce
crit
erion o
r
obje
c
tive function
is first
defined
ba
se
d on the
de
sire
d spe
c
ifications
su
ch
as fre
que
ncy
domain
and
time integra
l
perfo
rman
ce.
The most commonly u
s
ed time int
egral pe
rform
ance indexe
s
are inte
gra
l
of
absolute
erro
r (IAE), i
n
teg
r
al of the
squ
a
r
e
error
(ISE)
and i
n
tegral o
f
the time
wei
ghted
ab
solut
e
error (ITAE).
M
inimization
of IAE and ISE is co
n
s
ide
r
ed as the o
b
j
e
ctive of pre
s
ent pape
r.
The time inte
gral pe
rforma
nce
crite
r
ia is expresse
d a
s
:
I
A
E
=
(
e
(t
)
+
e
(
t
)
+
e
(t
)
+
.
.
.
.
.
.
+
e
(t
)
)
1
n
23
0
(26)
22
2
2
I
S
E
=
(e
(t
)
+
e
(
t
)
+
e
(t
)
+
.
.
.
.
.
.
+
e
(t
))
n
23
1
0
(27)
7. Simulation Resul
t
s
The differe
nt Interactive m
u
ltivariable p
r
oce
s
se
s have
been u
s
ed to
test the closed loop
perfo
rman
ce
of the prop
osed tuning m
e
thod. All gi
ve very satisfa
c
tory re
spon
ses for
set poi
nt
cha
nge
s a
nd
disturban
ce
rejectio
n. He
re the
simu
lati
on results of i
ndu
strial
scal
e polyme
r
ization
rea
c
tor
and
OR
colu
mn a
r
e p
r
e
s
ente
d
and the
pe
rfo
r
man
c
e i
ndi
ces of a
r
e
give
n in Ta
ble 2,
and
Table 3 respe
c
tively.
Process
1
: Consi
der the in
dustri
a
l scale
polymeri
z
atio
n rea
c
tor p
r
o
c
e
ss give
n by:
2
2
.
8
9
-
11.
64
-
0
.
2
s
-
0.
4s
ee
4.
572s
+
1
1
.
8
07s
+
1
G(
s
)
=
4.
68
9
5
.
8
0
-
0
.
2
s
-
0.
4s
ee
2.
174s
+
1
1
.
8
01s
+
1
Since it is 2×2 pro
c
e
ss, th
e two de
centralize
d
co
ntroll
ers a
r
e
requi
red.
The interactio
n obtaine
d using relative g
a
in array is:
0.
7087
0.
2913
λ
=
0.
.
291
3
0
.
7087
The critical freque
ncy arra
y is:
8.
0554
4.
1888
Ω
=
7.
8540
4.
3036
And Effec
t
ive Relative Gain Array is
:
0
.
71
93
0.
2
8
0
7
=
0.
28
07
0.
71
93
Based
on th
e re
sults
of RGA an
d ERGA, the pair
of u
1
-y
1
, u
2
-y
2
is having
strong i
n
tera
cti
on.
Hen
c
e the
co
ntrolle
r is de
signed for the
s
e two pairs.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5350 – 53
61
5358
The ph
ase
and g
a
in ma
rgin a
r
e
cho
s
en fo
r the
controlle
r de
sign i
s
60
0
and
3.0
respe
c
tively for diag
onal P
I
controlle
rs.
Then PI cont
roller is:
0.2
258
0.
11
44
0.
52
33
+
0
s
G(
s
)
=
c-p
r
op
osed
00
.
4
0
6
7
+
s
The Xiong.et.
a
l-PI cont
rolle
rs a
r
e given a
s
follows:
0.
0479
0.
9
455
0.
2190
+
0
s
G(
s
)
=
c-Xiong
.
e
t
.
a
l
0
0
.
1703
+
s
The Luybe
n-PI controlle
rs
are given a
s
follows:
0.0
929
0.041
1
0.2
1
0
+
0
s
G(
s
)
=
c-L
u
y
b
e
n
.e
t
.
a
l
0
0
.1
75+
s
The sim
u
lati
on re
sult
s of First an
d seco
nd
outp
u
t of indust
r
ial
scale polym
erization
rea
c
tor a
r
e
shown in
Fig
u
r
e
4 a
nd
Fig
u
re
5
re
spe
c
t
i
vely. The p
e
r
forma
n
ce i
n
dice
s i
s
sh
own in
Table 2.
Figure 4. Clo
s
ed L
oop
Re
spo
n
se of First
Output with S
e
t Point Changes
Figure 5. Clo
s
ed L
oop
Re
spo
n
se of Secon
d
Output with S
e
t Point Changes
.
.
Table 2. Performance Indices
for Proc
ess
1
Controller
Input(u)
-
outp
u
t(
y)
IAE
ISE
ITAE
Proposed-PI
u
1
-y
1
u
2
-y
2
2.079
2.904
1.11
1.014
40.4
54.3
Xiong.et.al-PI u
1
-y
1
u
2
-y
2
2.378
1.96
1.44
1.037
31.47
32.87
Lu
y
ben
-PI
u
1
-y
1
u
2
-y
2
2.183
3.427
0.9623
1.274
24.72
71.64
Process
2:
Consider
a bi
nary
ethanol
–water
sy
stem
of a Pilot-plant distillation col
u
mn
with a sid
e
stream as
well a
s
overh
ead a
nd
bottom produ
cts propo
sed by Og
un
naike and
Ra
y,
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
De
sign of De
centralized Controlle
r for Intera
ctive Proce
s
se
s thro
ugh… (R. Ha
num
a Naik)
5359
-2
.
6
s
-
3
.
5
s
-
6
s
0.
66e
-
0
.
6
1
e
-
0
.
0
04
9e
6
.
7s
+
1
8
.
4s
+
1
9.
06s
+
1
-
6
.5
s
-
3
s
-1
.
2
s
1.
1
1
e
-
2.
3
6
e
-
0.
01e
G(
s
)
=
3
.
2
5
s+
1
5
s+
1
7
.
0
9
s
+
1
-9
.
2
s
-
9
.
4
s
-s
-
3
4.
6
8
e
4
6
.
2e
0.
8
7
(
1
1.
6
1
s
+
1)
e
8.
15s
+
1
10
.
9
s
+
1
(
3
.
89
s
+
1
)
(
1
8.
8s
+
1
)
Whe
r
e the o
u
tputs are y1: overhea
d ethanol mo
l
e
fraction, y2: side stream
ethanol mol
e
fraction,
y3:
tray #1
9 tem
peratu
r
e,
de
gree
(co
r
respondi
ng to
b
o
ttoms
com
p
osition
)
, a
n
d
the
inputs a
r
e u1
: reflux flow rate, gpm (m3
/
s), u2: side
strea
m
pro
d
u
c
t flow rate, gpm (m3/
s) a
nd
u3: reboil
e
r
stream p
r
e
s
sure, psig (kPa).
Since it is 3×3 pro
c
e
ss, th
e three de
ce
ntra
lized cont
rolle
rs a
r
e re
quire
d. The i
n
tera
ction
obtaine
d usi
n
g relative gai
n array is:
2
.
0
084
-
0
.
722
0
-
0.
28
64
λ
=
-
0.
64
6
0
1.
82
4
6
-
0
.1
78
6
-
0
.3
624
-
0
.
102
6
1
.
465
0
Energy tran
smissi
on ratio is:
0.09
85
-
0
.0
706
-
0
.00
0
5
E
=
0.34
15
-
0
.4720
-
0
.00
1
4
-
4
.25
5
2
4
.2
385
0.0701
And ERGA is:
2.
4267
-
1
.1
510
-
0
.275
8
Φ
=
-
0
.
824
4
1
.97
4
6
-
0
.
1502
-
0
.
6023
0.176
4
1
.4
259
The ph
ase a
nd gai
n marg
in are
ch
osen
for the controller d
e
si
gn i
s
60
0
a
nd 3.0
respe
c
tively for
diago
nal PI controlle
rs.
Based
on the
results of RGA and ERG
A
, the pair of u
1
-y
1
, u
2
-y
2
,
u
3
-y
3
is havin
g stro
ng
intera
ction. Hence the co
ntrolle
r is de
sig
ned for the
s
e
three pai
rs.
The PI contro
ller is:
0.
30
52
2.
0
45+
0
0
s
0.
0
7
4
G(
s
)
=
0
-
0
.
7
0
0
-
0
c
-
P
r
op
os
e
d
s
0
.
23
82
00
3
.
0
7
6
+
s
The PI contro
ller usi
ng BLT
method is:
0.
0920
1.
5
1
+
0
0
s
0.
0166
G
(
s
)
=
0
-
0
.
3000-
0
c-
B
L
T
s
6.
6100
0
0
2.
63
+
s
The PI controller values by Halveli et.al.:
Evaluation Warning : The document was created with Spire.PDF for Python.