TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.5, May 2014, pp
. 3366 ~ 33
8
0
DOI: http://dx.doi.org/10.11591/telkomni
ka.v12i5.4926
3366
Re
cei
v
ed O
c
t
ober 1
8
, 201
3; Revi
se
d Decem
b
e
r
2, 2013; Accepte
d
De
cem
ber
21, 2013
Nonlinear Robust Control Approach Based on
Integrity
Chen Jinli*
1
, Xue Yali
2
, Liu Xingang
3
1
Aircraft Air
w
or
thiness C
e
rtific
ation
D
ep
artme
n
t, Civil
Aviati
o
n
Mana
gem
ent
Institute of China (CAMIC)
No. 3 East Roa
d
Hua
jia
di, Ch
ao
ya
ng District
,
Beijin
g, Chin
a
,
10010
2, Ph./Fax: +
861
0-58
250
63
1/582
50
630
2
State Ke
y
La
b
of Po
w
e
r S
y
st
ems,
T
hermal
Engi
neer
in
g D
epartme
n
t,
T
s
ingh
ua U
n
ivers
i
t
y
No. 30Sh
u
a
n
g
q
in
g Roa
d
, Hai
d
ia
n District, Beiji
ng, Ch
ina, 1
000
84, Ph./F
ax: +
8610-62
79
5
736
3
T
e
chnica
l Dep
a
rtment, Beiji
n
g
Lon
g-March
Lau
nch Ve
hicl
e Equi
pme
n
t
T
e
chn
o
lo
g
y
Co.
L
td
No. 19Bu
ild
in
g W
ang
yu
an
xi
li, Fengtai D
i
strict,
Beijin
g, Chin
a
,
10007
6, Ph./Fax: +
861
0-88
530
54
2
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: chenj
inl
i
@ca
m
ic.cn*
1
, xu
e
y
ali@tsi
n
g
hua.
e
du.cn
2
, liux
g
@lvet.cn
3
A
b
st
r
a
ct
A kind of Non
l
i
near R
obust C
ontrol (NR
C
) a
ppro
a
ch b
a
sed
on integr
ity for multiv
aria
bl
e systems
is prese
n
ted. It uses mod
e
l e
s
timatorw
hic
h
provi
des
the
a
pprox
imate
m
o
del i
n
for
m
ati
o
n
to compens
at
e the
non-
mode
le
d d
y
na
mics, syste
m
u
n
certa
i
nti
e
s
,
and
externa
l
disturb
ances
of
a syste
m
. Firstly, the existe
n
c
e
of NRC w
i
th i
n
tegrity is ex
a
m
ine
d
. Then, st
abl
e re
g
i
o
n
s o
f
each N
R
C
’
s
para
m
eters ar
e calc
ulate
d
, a
n
d
some par
a
m
et
ers are o
b
tain
e
d
by pl
acin
g su
itabl
e clos
ed-l
o
op po
les, for meetin
g
the d
e
si
gn spec
ificati
o
ns
of the whole c
ontrol system
. The proposed
m
e
thod is
applied to twoillustrative
examplesfrom
lit
erature.
Results dem
onstrate that NRC is
feasible and robust for co
mplic
ated m
u
l
t
ivariable
system
s.
Ke
y
w
ords
:
no
nlin
ear ro
bust control (NR
C
),
m
o
del estim
a
tor, integrity
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
Most ind
u
st
ri
al pro
c
e
s
se
s are M
u
lti-In
put
Multi-O
u
tput (MIMO
)
in esse
nce. Due to
intera
ction
s
b
e
twee
n loo
p
s, simply exte
nding th
e de
sign
metho
d
for si
ngle
-
inp
u
t singl
e-outp
u
t
(SISO) control system to
MIMO
syste
m
gene
rally
cau
s
e
s
the
p
e
rform
a
n
c
ed
eterio
ration, and
even system
instability. No
wad
a
ys
, the method
s for tuning MIMO
system a
r
e categori
z
e
d
into
two main g
r
o
ups: multivari
able cont
rol a
nd de
centrali
zed
control.
Gene
rally
sp
eaki
ng, the
p
e
rform
a
n
c
e
o
f
a multiv
ari
a
ble control system
i
s
sup
e
rior
to
th
e
decentrali
ze
d
one.
Ho
wev
e
r, the l
a
tter i
s
ea
sie
r
to d
e
sign a
nd
reali
z
e [1]. It has l
e
ss p
a
ra
mete
rs
to be tune
d, and i
s
conve
n
ient to de
al with
when
so
me loop
s b
r
e
a
k d
o
wn. Becau
s
e
of the
s
e,
decentrali
ze
d
cont
rol h
a
s
be
en
wid
e
ly use
d
in
indu
strial
p
r
ocesse
s. E
s
pe
cially rob
u
st
decentrali
ze
d
control h
a
s a
promi
s
ing fut
u
re in ind
u
st
ri
al pro
c
e
s
ses.
Some re
se
arche
r
s
have p
a
id more atte
ntion
to tune
para
m
eters o
f
decent
rali
ze
d PI/PID
controlle
r in recent years,
such a
s
De
tuningm
etho
d [2],Sequen
tial-clo
se
met
hod [3], Mono-
variable
met
hod [4], Effe
ctive op
en
-lo
op p
r
o
c
e
s
s (EOP) meth
o
d
[5], an
d Ite
rative meth
o
d
[6],
etc. (1)
Detu
ning metho
d
[2]: Each PI/PID c
ontroller is desi
gne
d indep
ende
ntly base
d
on SISO
method, a
nd
then the
co
ntrolle
r pa
ram
e
ters
ar
e
detu
ned to
com
p
ensate the
co
upling
s
b
e
tween
loop
s when
all loop
s
are
clo
s
e
d
. The
gre
a
test
ad
vantage
of this m
e
thod
i
s
simple,
but
the
perfo
rman
ce
and sta
b
ility spe
c
ification
s
are not cl
ear
during the
d
e
tuning, so it is som
e
kin
d
o
f
trial and
erro
r. (2
) Sequ
e
n
tial-cl
o
se m
e
thod [3]: Cl
ose
ea
ch lo
o
p
usi
ng the
SISO metho
d
in
certai
n seq
u
ence.
The m
e
thod
i
s
con
c
eptu
a
lly
sim
p
le, but the
resulted
perf
o
rma
n
ce g
r
e
a
tly
depe
nd
s on t
he clo
s
in
g se
quen
ce. Whe
n
som
e
loop
s brea
k do
wn,
the stability of the re
st can
n
o
t
be g
uarantee
d auto
m
atical
ly. (3)
Mon
o
-v
ariabl
e m
e
th
o
d
[4]: Each
co
ntrolle
r i
s
d
e
signed
by SISO
method und
er ce
rtain constraints
on
stability
a
n
d
pe
rform
a
n
c
e sp
ecifi
c
ati
ons. The en
tire
perfo
rman
ce
may be un
sa
tisfied be
ca
u
s
e of n
egle
c
ting of detaile
d inform
ation
of controlle
rs in
other lo
op
s. (4) Effective
open
-loo
p pr
oce
s
s (EOP
) method [5]: Reg
u
late th
e
i
-th controll
ers
according to
i
-th EOP wh
en all the other loo
p
s a
r
e clo
s
ed. Th
e whol
e perf
o
rma
n
ce ca
n
be
guarantee
d since th
e EO
P desig
n ha
s con
s
id
ere
d
the informatio
n of all othe
r loop
s, but the
cal
c
ulatio
n of
EOP be
com
e
s p
r
o
g
re
ssively com
p
lic
ated with
the i
n
crea
se
of sy
stem dim
e
n
s
i
on
and m
odel
order,
so
doe
s t
he a
pproxima
t
ion mod
e
l e
r
ror. So th
e me
thod i
s
suitabl
e only fo
r lo
w-
dimen
s
ion
a
nd lo
w-ord
e
r system.
(5) Iterative me
thod [6]: It is si
milar to
seq
uential
-
cl
ose
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Nonli
nea
r Ro
bust Control
Appro
a
ch
Based o
n
Integri
t
y (Chen
Jinli)
3367
method, b
u
t tuning
process i
s
carried
o
u
t iterat
ively until
all cont
roller paramet
ers
conve
r
ge.
The
effort on
tu
ning i
s
eno
rmou
s, b
u
t
the rela
tion
s between
tuning
proce
s
s an
d
re
sulted
perfo
rman
ce are
wea
k
.
The
comm
o
n
characte
ri
stics
of ab
ove me
tho
d
s are de
com
p
osin
g
o
r
tra
n
sforming
multivariable system with
n
inputs va
ria
b
les a
nd
n
o
u
tputs vari
abl
es into
n
sing
le-loo
p
sy
st
e
m
s.
They all
hav
e some
kin
d
of co
nserva
tiveness
du
e
to the inte
raction
betwe
en
chan
nel
s, so
overall perfo
rmance coul
d
be
p
o
ssibly optimize
d
fu
rther. Th
e EO
P method i
s
con
s
id
ere
d
a
n
entire
pe
rformance
tool.
Ho
wever, m
odel re
d
u
ctio
n e
rro
r m
a
kes E
O
P met
hod
hard to
be
extendedfo
r
m
u
ltivariable a
nd high
-o
rde
r
system.
In orde
r to improve robu
stne
ss p
e
rfo
r
manc
e for system uncert
a
inties, a
s
well as to
guarantee th
e high integ
r
ity of controlsystem,
this
pape
r propo
ses a ki
nd ofNonli
earRo
bu
st
Control (NRC) app
roa
c
h. B
a
se
d on inte
g
r
ity theor
em a
nd pa
ramete
r tuning metho
d
s, the contro
l
para
m
eters
a
r
e o
b
tained
di
rectly a
nd e
a
s
ily.No m
odel
red
u
ctio
n is
need
ed, so n
o
mod
e
l erro
r is
introdu
ce
d.
This pa
per is orga
nized as follo
ws
. Sectio
n
2 outlines the probl
em and
requi
rem
ents.
Section 3 i
n
trodu
ces NRC approach and stabl
e regi
on calculation method for 1
-
orde
r and 2-orde
r
SISO system.
In
Se
ction 4,the
int
egrity theo
re
m and
refere
nce
sta
b
le
re
gion
for MIMO
co
upling
sy
ste
m
are p
r
e
s
e
n
ted.The
enti
r
e
d
e
si
gn proce
dure
of NRC app
roa
c
h
i
s
prop
osed in
section
5
.Some
simul
a
tion re
sults
are
give
n in
se
ction 6
,
following
wit
h
analy
s
is
an
d
summ
ary. Th
e con
c
lu
sio
n
is dra
w
n in th
e last se
ction.
2. Problem
Des
c
ription
In this pa
pe
r, we a
s
sume
that the pr
o
c
e
ss matrix has diag
onal
domina
n
ce and
the
input-o
utput variable
s
are
paired in the
diagonal
wa
y. If not, a n
e
w input/outp
u
t pairing
sh
ould
be con
s
ide
r
e
d
[6], or a
compen
sato
r
sho
u
ld b
e
int
r
odu
ce
d [7] to minimi
ze i
n
tera
ction of
the
sy
st
em.
We con
s
ide
r
a
stabl
e
n
by
n
co
ntrolle
d proce
s
s, whi
c
h
is de
scrib
ed
by a nomin
al transfe
r
function mat
r
i
x
()
p
Gs
as:
11
12
1
21
2
2
2
12
()
()
(
)
()
()
(
)
()
()
(
)
()
n
n
p
nn
n
n
g
sg
s
g
s
g
sg
s
g
s
Gs
g
sg
s
g
s
(1)
Whe
r
e
()
ij
g
s
rep
r
e
s
ents the tra
n
sfer function from the
j
-th in
put to the
i
-th output.
If we i
gno
re
the
cou
p
lin
g am
ong
th
e sub
s
ystem
s
, the
probl
em of
de
sig
n
ing
a
decentrali
ze
d
controller t
u
rn
s into
n
indep
ende
nt controllers.
Then the
pro
c
e
ss i
s
t
o
be
controlled in
a negative fe
edba
ck co
nfigur
atio
n by the decentrali
zed co
ntrolle
r
()
c
Gs
:
1
2
()
0
0
0(
)
()
0
00
(
)
c
c
c
cn
Gs
Gs
Gs
Gs
(2)
Then, the fina
l dece
n
trali
z
e
d
PI/PID
control syst
em is
as follo
ws:
()
()
()
()
(
)
(
(
)
(
)
)
p
c
Ys
G
s
Us
Us
G
s
R
s
Y
s
(3)
Whe
r
e
12
()
[
(
)
,
(
)
,
,
()
]
T
n
Ys
y
s
y
s
y
s
is a
n
output vecto
r
,
12
()
[
(
)
,
()
,
,
()
]
T
n
Us
u
s
u
s
u
s
is an input
vector, a
nd
12
(
)
[
(
)
,
()
,
,
()
]
T
n
Rs
r
s
r
s
r
s
is a
set-poi
nt
vector. M
a
trix
indicates the
loop failu
re
of
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3366 – 33
80
3368
s
e
ns
or
s or
a
c
tu
a
t
or
s
,
1
{,
,
}
,
{
0
,
1
}
,
1
,
,
ni
di
ag
i
n
. If the
i
-th su
bsyste
m
i
s
workin
g
prop
erly, then
1
i
, els
e
0
i
.
The co
ntrol di
agra
m
is sho
w
n in Figu
re
1.
1
r
2
r
n
r
1
y
2
y
n
y
1
u
2
u
n
u
Figure 1. MIMO Co
ntrol S
y
stem Con
s
id
ering Integ
r
ity
3.
Nonlinear Robust Con
t
r
o
l
(NRC)
Ap
proach Desi
gn
3.1. SISO NRC
The r-order S
I
SO NRC usi
ng a kin
d
of model e
s
tima
tor is de
sign
e
d
as [8]:
-1
1
0
--
r
ii
i
uh
z
d
(4)
Whe
r
e the m
odel e
s
timato
r is:
2
--
-
r
r
dk
z
kk
z
k
u
(5)
Whe
r
e
d
denot
es mo
del inf
o
rmatio
n eve
n
incl
uding
u
n
-mo
dele
d
d
y
namics, un
certai
nties in
system
pa
ra
meters, an
d
disturban
ce
s;
,0
,
,
1
i
hi
r
are
suitabl
e
po
sitive
con
s
tants,
whi
c
h
ensure
polyn
omial
1
01
1
()
rr
r
hs
h
h
s
h
s
s
is
Hu
rwi
w
z an
d satisfy desi
r
e
d
dynami
c
s
of
clo
s
ed
-loo
p system;
k
is a key
controller
param
eter,
whi
c
h determi
ne stability
directly;
,0
,
i
zi
r
are sy
stem st
ates, and
(1
)
,1
,
i
i
zy
i
r
.
Takin
g
th
e L
apla
c
e t
r
an
sformatio
n
of
(4)
and
(5
), it
is
straightforward to
calculate th
e
trans
fer func
tion of NRC:
0
1
()
r
i
ci
i
Gs
K
s
s
(6)
Whe
r
e
00
K
kh
,
1
,1
,
1
ii
i
K
kh
h
i
r
,
1
rr
K
kh
.
It can be cle
a
r
ly seen that t
he stru
ctu
r
e o
f
(6) is si
milar to common
PID controller.
3.2. Stable Region
For 1
-
o
r
de
r a
nd 2
-
order SISO system, t
he sta
b
le
reg
i
ons
of NRCa
ppro
a
ch a
r
e
easy to
cal
c
ulate.
3.2.1. 1-orde
r Plant
We con
s
ide
r
a gene
ral 1
-
o
r
de
r plant:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Nonli
nea
r Ro
bust Control
Appro
a
ch
Based o
n
Integri
t
y (Chen
Jinli)
3369
y
ay
bu
(7)
Whe
r
e
y
is the system outp
u
t,
u
is the cont
rol sig
nal, an
d con
s
tant
s
a
a
nd
b
are b
o
th u
n
kn
own.
The equ
ation
can b
e
re
writt
en as follo
ws:
y
du
(8)
Whe
r
e
(1
)
da
y
b
u
.
The 1-ord
e
r
NRC is d
e
si
g
ned a
s
follows [8]:
0
2
ˆ
ˆ
dk
y
kk
k
u
uh
y
d
y
(9)
The tran
sfe
r
functio
n
of (9)
is
s
i
milar to PI c
ontroller as:
0
0
()
(
)
c
kh
Gs
k
h
s
(10
)
Substituting (9) into (8
), we
obtain:
0
yh
y
d
(11
)
Whe
r
e
ˆ
dd
d
.
Let us no
w compute
r
the time derivative
of
ˆ
d
,
2
ˆ
()
()
dk
y
kd
u
kk
y
kd
ku
ky
kd
(12
)
And the expression of
d
is
:
0
00
ˆˆ
(1
)
(
)
ˆ
)
ˆ
(
ay
b
dd
d
hy
d
d
hb
h
a
y
b
d
(13
)
Takin
g
into
a
c
count
(11
)
a
nd (12), the
d
y
namics of
d
can be
obtain
e
d
by co
mputi
ng the
time derivative of (13).
00
00
0
22
00
0
0
0
ˆ
()
()
(
)
()
(
)
hb
h
a
y
b
d
hb
h
a
h
y
d
b
k
d
ah
bh
h
y
h
b
h
a
b
d
kd
(14
)
D
e
fin
i
ng
ne
w s
t
a
t
e va
r
i
ab
le
ve
c
t
or
~
12
[,
]
[
,
]
TT
yd
, th
e
c
l
os
e-
loo
p
sys
te
m
c
a
n be
descri
bed a
s
follows:
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3366 – 33
80
3370
10
1
2
22
00
0
21
0
2
0
()
(
)
ah
bh
h
h
bh
k
h
ab
(15
)
That is:
z
A
(16
)
Whe
r
e
22
00
0
0
00
1
z
ah
b
h
h
h
bh
a
b
h
k
A
.
System (16
)
is asym
ptotica
lly stable only
if
Z
A
is Hu
rwi
z
e
.
Then we ha
ve:
0
0
0
0
ab
h
b
k
bkh
(17
)
Bec
a
us
e
0
0
h
and
0
b
, the stable condition b
e
co
mes:
0
0
ab
h
b
(18
)
Whe
r
e
/s
g
n
(
)
kb
.
The re
sults
a
r
e sho
w
n
in Figure
2. wit
h
pla
n
t
pa
ra
meters
a
locating [-1
00
100]
and
b
being e
qual t
o
-5,-1, 1,5,re
spe
c
tively. Thestabl
e regi
o
n
is the are
a
offold line up
per-rig
h
t side.
Figure 2. Stable Re
gion of
1-order SIS
O
NRC
3.2.2. 2-orde
r Plant
We con
s
ide
r
a gene
ral 2
-
o
r
de
r plant:
10
ay
a
y
yb
u
(19
)
Whe
r
e con
s
tants
1
a
,
0
a
and
b
are all unkno
wn
.
Usi
ng mod
e
l estimato
r, the 2-order
NRC is desi
gne
d as follo
ws [8]:
1
2
0
ˆ
ˆ
dk
y
uh
y
h
k
d
kk
y
yu
(20
)
-1
00
-80
-6
0
-40
-2
0
0
20
40
60
80
10
0
0
1
2
3
4
5
6
7
8
9
a
my
f
u
n
1
(
a
,5
,1
0
)
b=
-
5
b=
-
1
b=
1
b=
5
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Nonli
nea
r Ro
bust Control
Appro
a
ch
Based o
n
Integri
t
y (Chen
Jinli)
3371
The tran
sfe
r
functio
n
of this 2-
or
de
r
NRC
is
co
mp
u
t
ed
a
s
fo
llow
s
:
0
10
1
()
(
)
(
)
c
hk
Gs
h
k
h
h
k
s
s
(21
)
Who
s
e
stru
ct
ure is
simila
r to PID control
l
er.
The a
nalysi
s
method i
s
si
milar to th
e 1
-
order
pla
n
t. Becau
s
e
of t
he limited
arti
cle l
ength,
we o
m
it man
y
analysi
s
steps. By defin
ing ne
w
stat
e varia
b
le ve
ctor
12
3
[,
,
]
[
,
,
]
TT
yy
d
,
the clo
s
e-l
o
o
p
system
can
be descri
bed
as follows:
z
A
(22
)
Whe
r
e
01
22
10
0
1
0
1
1
1
1
1
0
0
0
1
1
1
01
0
1
z
Ah
h
ah
b
h
h
h
h
b
h
h
ah
h
b
h
a
h
b
h
a
b
k
.
Hen
c
e, Syste
m
(22) i
s
a
s
ymptotically st
able only if:
11
00
1
0
00
1
1
1
0
0
0
0
()
(
)
ab
h
b
k
ab
h
b
h
k
bh
k
ab
h
b
h
k
a
b
h
b
k
b
h
k
(23
)
Bec
a
us
e
0
0
h
,
1
0
h
and
0
b
, the stable con
d
ition be
comes:
11
00
1
00
1
1
1
0
()
(
)
ab
h
b
ab
h
bh
ab
h
b
h
a
b
h
b
b
h
(24
)
Whe
r
e
/s
g
n
(
)
kb
.
In order to
simplify the
tuning
pro
c
e
ss, th
e p
a
ra
meters, which dete
r
min
e
de
sire
d
dynamics a
r
e
sele
cted a
s
1
2
c
h
and
2
0
c
h
. In other words, the
cl
ose
d
-lo
op ro
ots are pla
c
e
d
at
c
.
The results a
r
e sho
w
n in
Figure 3. wit
h
plant pa
ra
meters
0
a
locating [-20
00 1
0
00],
1
a
loc
a
ting [-100 100],
1
b
and
10
c
, where th
e are
a
to the upp
er si
de of the
mesh
su
rface is
the stable reg
i
on.
Figure 3. Stable Re
gion of
2-order SIS
O
NRC
-2
00
0
-
150
0
-1
00
0
-5
00
0
500
100
0
-1
00
-5
0
0
50
10
0
0
20
40
60
80
10
0
12
0
a
0
a
1
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3366 – 33
80
3372
4.
Integrity
Theorem and Stable Region
of MIMO Sy
s
t
em
4.1. Integrit
y
Theorem
The MIMO system has
more
sen
s
o
r
s and a
c
tuat
ors tha
n
SISO one. Because of
intera
ction, e
a
ch el
ement
failure will
ch
ange t
he p
e
rforman
c
e of
rest
sub
s
yste
m, even cau
s
e
instability. The closed-loop system should exhi
bit stability with
accepted dy
namic properties
even if some
elements o
r
chann
els b
r
eak do
wn. T
h
is prope
rty is calle
d integ
r
ity. The system
has hig
h
-inte
g
rity prope
rty if it
is sta
b
le
und
er any
p
o
ssible
loo
p
f
a
ilure
s.
We
d
enote th
e
sta
t
ic
gain matrix of
process
()
p
Gs
as
:
11
0
1
2
0
1
0
21
0
2
20
2
0
0
10
2
0
0
n
n
p
nn
n
n
gg
g
gg
g
G
gg
g
(25
)
Then th
e foll
owin
g lemm
a
on the i
n
teg
r
ity of
decent
ralized PI form co
ntrol
sy
stem i
s
given.
Lemma1
[9]:
Con
s
id
er a stable
li
nea
r system (3
), a
nd a
s
sume t
hat static gai
n matrix
0
0
,
1
,
...,
ii
g
in
. Then a ne
cessary an
d sufficient co
ndi
tion for the existen
c
e of de
centralized PI
controllers that ensure
closed
-loop stability for all possible
is given
by the requi
rement that all
prin
cipal min
o
rs of
0
p
G
are po
sitive.
If the control
system i
s
pre
s
ente
d
as tra
n
sfer
func
tion form (of c
o
ur
se in
cludi
ng
PI form),
we have the f
o
llowin
g
lem
m
a.
Lemma 2
[9
]: Con
s
ide
r
a sta
b
le
line
a
r
system
(3
), and assu
me that decentrali
zed
controlle
rs en
sure th
e
stabi
lity of the
clo
s
ed-lo
op
sy
st
e
m
.
Wh
en i
n
t
e
ract
io
n
de
cre
a
se
s
or
b
r
ea
k
s
down, or so
me loo
p
s are
discon
ne
cte
d
from th
e
sy
stem, if the f
o
llowing
condition is satisfied,
the overall cl
ose
d
-lo
op sy
stem will be
stable and
still remai
n
stabl
e
:
1
p2
()
[
d
i
a
g
|
(
(
)
(
)
)
(
)
|
]
1
ci
i
i
ci
d
s
V
I
Gs
g
s
Gs
(26
)
Whe
r
e:
12
1
21
2
1,
1,
1
0|
(
)
|
|
(
)
|
|(
)
|
0
|(
)
|
|(
)
|
|
(
)
|
0
n
nn
nn
n
gs
g
s
gs
V
g
s
gs
g
s
(27
)
()
p
denote
s
Pe
rro
n-F
r
obe
niu
s
eigenvalu
e
of
matrix
()
;
denot
es ab
sol
u
te matrix, in whi
c
h
all element
s are the ab
sol
u
te value of primary mat
r
ix
()
.
Becau
s
e the
1-ord
e
r
NRC has
simila
r PI fo
rm, we can exami
ne the integrity of a
desi
gne
d NRC usi
ng this two lemma
s. Also, le
mma 1 is u
s
ed
to ensu
r
e t
he existen
c
e
of
decentrali
ze
d
1-o
r
de
r
NRC with
hig
h
-i
ntegrity.
If condition i
n
l
e
mma 1
is
not sati
sfied,
the
pairin
g
s
sh
ou
ld be re-o
rg
a
n
ize
d
, or mu
ch attention sh
ould be
paid
on when a
nd
how to
swit
ch
to
other controll
ers in
ca
se some loop
s di
sconn
ect fro
m
the system
.
If the design
e
d
NRC h
a
s 2
-
order
or hi
gh
er
o
r
de
r, lem
m
a 1 cannot
be used di
re
ctly. We
adopt eq
uival
ent tran
sform
a
tion, whi
c
h repla
c
e
s
()
ii
g
s
with
'
()
ii
g
s
in the calculatio
n.
,
0
,0
,
1
,
0
0,
1
,
'
0,
1
,
11
(
)
()
(
)
(
)
()
1
()
(
)
i
i
j
ji
j
j
ji
i
i
i
i
i
i
j
ii
ii
i
i
Ks
K
sgs
K
K
s
gs
ss
K
K
s
KK
s
g
s
s
(28
)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Nonli
nea
r Ro
bust Control
Appro
a
ch
Based o
n
Integri
t
y (Chen
Jinli)
3373
4.2. Stable Region
As mention
e
d
above, the stable re
gio
n
s of
1-o
r
de
r and 2-o
r
de
r controll
er fo
r SISO
system
can
be figured in
two-di
men
s
i
onal an
d th
ree-di
men
s
ion
a
l sp
ace, re
spectively. Fo
r a
MIMO sy
ste
m
, the sta
b
l
e
re
gion i
s
quite compli
cated.
From system (3
)
,
th
e
ch
ar
ac
te
r
i
s
t
ic
equatio
n of the decentrali
z
ed co
ntrol sy
stem is:
11
1
1
2
1
2
2
1
12
1
1
2
2
2
2
2
11
1
2
2
2
det
{
(
)
(
)
}
1(
)
(
)
(
)
(
)
(
)
(
)
()
(
)
1
(
)
(
)
(
)
(
)
de
t
{
}
(
)
()
()
()
1
(
)
(
)
0
pc
cc
n
n
c
n
cc
n
n
c
n
nc
n
c
n
n
n
c
n
IG
s
G
s
gs
G
s
g
s
G
s
g
s
G
s
gs
G
s
g
s
Gs
g
s
G
s
g
s
G
s
g
sG
s
g
sG
s
(29
)
It can be
se
e
n
that the sta
b
le re
gion
s a
r
e
co
rrelative,
and the va
ri
ety of each
controlle
r
cha
nge
s th
e
stable
region
s of
all
other
controlle
rs.
S
o
it i
s
difficult to exp
r
e
s
s a
nd
cal
c
ulate
th
e
stable regio
n
s
of MIMO NRCs.
In fact, the m
a
in rea
s
on
for cal
c
ulatin
g
stable
regi
on
s
of NRCs i
s
to
provid
e a
r
ea
son
abl
e
sea
r
ch sp
ace
for paramete
r
tuning. Since integrit
y is one of our ob
jectives, and
at the existence
of integ
r
ity property, it impl
ies th
at ea
ch
contro
lle
r
can
stabili
ze
the
corre
s
p
ondin
g
lo
cal fe
edb
a
ck
loop i
nde
pen
dently. For t
he p
u
rpose
of sim
p
licity,
the
union
sets of
sta
b
le
re
gion
of e
a
ch
diago
nal ele
m
ent are u
s
e
d
as NRC parameter
sreferencesta
ble re
gi
on for MIM
O
system.
Mean
while, combi
n
ing d
e
sired
dyn
a
m
ics
an
d
si
mulation, the
paramete
r
s set of
NRC
0
,
1,
1,
{,
,
,
,
}
(
1
,
,
)
ii
i
r
i
kh
h
h
i
n
are dete
r
min
ed finally.
5. Design
Pro
c
edure
To
sum
up, t
he d
e
si
gn
proce
dure of
M
I
MO de
ce
ntralize
d
NRCs
can
be
carrie
d out
as
f
o
llowin
g
st
ep
s:
1) Co
nfirm a
ppro
p
ri
ate struct
ure of co
ntrol syste
m
with
n
SISO suitable o
r
d
e
r NRC
according to
pro
c
e
ss info
rmation.
2) Examin
e t
he integ
r
ity o
f
the de
sign
e
d
c
ontrol
sy
stem u
s
ing
lem
m
a 1, oth
e
rwise, re-
analysi
s
the system or de
compo
s
e it, and desi
gn ne
w control
syste
m
.
3) Calculate
the stable
re
gion of ea
ch
di
ago
nal ele
m
ent as the
para
m
eters referen
c
e
tuning spa
c
e.
4) Determi
n
e
controller
p
a
ram
e
ters co
mbining
abo
ve regio
n
, desired dyn
a
m
ics
and
s
i
mulation.
5) Ch
eck the
integrity of the result
ed cl
o
s
ed
-loo
p syst
em with lem
m
a 2.
6) If s
a
tis
f
ies
,
des
ign is
finis
h
ed. Other
wise, ch
eck the
integrity by simulation.
Rem
a
r
k
1
: In many conditi
ons, the exact relative
deg
ree
s
of controller process
are not
easy to obt
ain be
cau
s
e
of the co
mplexity and unce
r
taintie
s
of model.
We ca
n d
e
sig
n
approp
riate
decentrali
ze
d
control sy
stem, who
s
e diago
nal
ele
m
ents
a
r
e 1
-
order or 2-o
r
de
r
N
RCs
.
Rem
a
r
k
2
: T
he p
r
op
osed
method i
s
b
a
s
ed
on
nomin
al mod
e
l. Un
der
plant u
n
certaintie
s
or
ope
rating
point
variati
on, robu
stne
ss ca
n
al
so
be
en
su
red
be
cau
s
e
NRC with
suitabl
e
para
m
eters h
a
s st
rong di
st
urba
nce rej
e
ction.
Rem
a
r
k
3
:
We ca
n al
so
op
timize the
co
ntrolle
r pa
ra
meters by d
e
fining
spe
c
ific
obje
c
tive
function.
The
para
m
eter o
p
t
imization
can
always
be
ca
rrie
d
o
u
t a
s
l
o
ng a
s
the
obj
ective fun
c
tion
is qua
ntitatively valuable. It
will be di
scussed in o
u
r foll
owin
g pape
r.
6. Illustrativ
e
E
xamples
In this
se
ctio
n, we
demo
n
s
trate th
e su
perio
r
cont
rol
ability and
robu
st pe
rformance to
great gai
n pe
rturb
a
tion
s in cro
s
s-cou
p
lings of
NRC
approa
ch through
six typical multivaria
bl
e
p
r
oc
es
se
s
[5
].
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 3366 – 33
80
3374
6.1. 2×2 Mod
e
ls
W
o
od
an
d
Be
r
r
y
(W
B)
c
o
n
t
r
o
lle
d pr
oce
s
s
is
c
o
n
s
id
ered
firstly. The tran
sfer
function
matrixis
given as
follows
:
3
1
73
12.8
18.9
16.7
1
21
1
()
6.6
19.4
10.
9
1
14.4
1
s
s
s
s
ee
ss
Gs
ee
ss
(30
)
The metho
d
in Section 5
is ca
rri
ed o
u
t to
tune the decentralized NRC. In
orde
r to
simplify an
al
ysisp
r
o
c
e
ss,
the time
-d
elay is not
co
nsi
dered
wh
en
we
desi
gn
su
bsystem
controlle
r. First, two 2
-
o
r
der
sub
s
yste
m cont
rolle
rs are d
e
si
gne
d acco
rdin
g
to the relativ
e
degree
s of co
rre
sp
ondi
ng d
i
agon
al tran
sfer functio
n
,
0,
0,
1
()
(
(
))
,
1
,
2
ci
i
i
i
i
Gs
k
h
k
h
s
i
s
(
3
1
)
Whe
r
e
sgn(
(
)
)
ii
i
i
kg
s
.
Then the
cont
rol for overall system i
s
:
1
2
()
0
()
0(
)
c
c
c
Gs
Gs
Gs
(
3
2
)
By checking the pro
c
e
s
s with lemma 1, it can
be foun
d that all principal min
o
rs of static
gain matrixe
s
are positive,
so it satisfie
s the
sufficie
n
t and necessary conditio
n
s of existing
2-
orde
r de
ce
ntralize
d
NRC
with integrity.
Figure 4. Dyn
a
mic Respon
se of
Dif
f
eren
t Step Inputs (WB mo
del)
The
stabl
e re
gion
of ea
ch
2-o
r
de
r
cont
roller i
s
calcul
ated u
s
in
g th
e meth
od
pre
s
ente
d
in
se
ction 3.2. We get
1
0
and
20
,
2
0.0
515
h
. In the
s
e t
w
o
loop
s, the
d
e
sired
pole
s
are
all
placed at -0.1
, that is,
0,
1
0
,
2
0.1
hh
.
A
l
so
,
we sele
ct
1
0.
35
,
2
0.
2
.
0
20
40
60
80
10
0
0
0.
2
0.
4
0.
6
0.
8
1
1.
2
1.
4
y
1
Ti
m
e
(
s
)
0
20
40
60
80
10
0
-0
.
4
-0
.
2
0
0.
2
0.
4
0.
6
0.
8
1
y
2
Ti
m
e
(
s
)
0
20
40
60
80
10
0
0
0.
05
0.
1
0.
15
0.
2
0.
25
0.
3
0.
35
0.
4
Ti
m
e
(
s
)
y
1
0
20
40
60
80
10
0
0
0.
2
0.
4
0.
6
0.
8
1
1.
2
1.
4
Ti
m
e
(
s
)
y
2
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Nonli
nea
r Ro
bust Control
Appro
a
ch
Based o
n
Integri
t
y (Chen
Jinli)
3375
Whe
n
the
set-point
sig
nal
of two
ch
ann
els
ha
s u
n
it step di
sturb
a
n
c
e
s
respe
c
tively, the
dynamic
responses are illu
strated in Fi
gure 4.
If there are +20% gain pe
rturbation
s
in
cr
o
s
s-co
uplin
gs, ne
w cross-co
uplin
gs b
e
c
ome:
'
12
1
1
2
()
(
)
g
sg
s
(33
)
'
21
2
2
1
()
()
g
sg
s
(34
)
Whe
r
e
1
and
2
both vary ra
n
domly in the
area
of [1, 1.2]. The bla
c
k lines
rep
r
e
s
e
n
t dynamic
respon
se
s i
n
norm
a
l condit
i
on, and
gray
lines
re
pre
s
e
n
t dynami
c
re
spo
n
ses
wh
e
n
pe
rturb
a
tio
n
s
exis
t.
The 2-order d
e
ce
ntrali
zed
NRC for th
re
e mo
re
2×2
p
r
ocesse
s d
e
m
onst
r
ated a
s
Tabl
e 1,
usin
g the sa
me config
uration.
Table 1. Tran
sfer Fu
nctio
n
Matrices of 2
×
2 System
s
Model T
r
ansfer
Functio
n
V
i
nante-Lu
y
b
e
n
model (VL)
1
2
.
9
3
.
4
1
5
.
9
8
.
2
1
7
3
.
1
1
7
2
.
2
)
(
35
.
0
8
.
1
3
.
0
2
s
e
s
e
s
e
s
e
s
G
s
s
s
s
W
a
rdle-W
ood
model (WW)
1
35
12
.
0
)
1
38
(
094
.
0
)
1
48
)(
1
45
(
101
.
0
1
60
126
.
0
)
(
8
8
12
6
3
s
e
s
e
s
s
e
s
e
s
G
s
s
s
s
Ogunnaike-
R
a
y
model (OR2)
1
801
.
1
8
.
5
1
174
.
2
689
.
4
1
807
.
1
64
.
11
1
572
.
4
89
.
22
)
(
4
.
0
2
.
0
4
.
0
2
.
0
4
s
e
s
e
s
e
s
e
s
G
s
s
s
s
The dynami
c
respon
se
s in both norm
a
l con
d
it
ion and
existing +20
%
gain perturbatio
n
s
in cross-couplings
condition
for the three processes
are illu
strated in Figure 5-F
i
gure 7.
Figure 5. Dyn
a
mic Respon
se of
Dif
f
eren
t Step Inputs (VL
Mod
e
l)
0
20
40
60
80
10
0
0
0.
2
0.
4
0.
6
0.
8
1
1.
2
1.
4
Ti
m
e
(
s
)
y
1
0
20
40
60
80
10
0
-0
.1
0
0.
1
0.
2
0.
3
0.
4
Tim
e
(s
)
y
2
0
20
40
60
80
10
0
-0
.
1
0
0.
1
0.
2
0.
3
0.
4
Ti
m
e
(
s
)
y
1
0
20
40
60
80
10
0
0
0.
2
0.
4
0.
6
0.
8
1
1.
2
1.
4
Tim
e
(s
)
y
2
Evaluation Warning : The document was created with Spire.PDF for Python.