TELKOM
NIKA
, Vol.12, No
.4, Dece
mbe
r
2014, pp. 82
9~8
3
8
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v12i4.293
829
Re
cei
v
ed Au
gust 24, 20
14
; Revi
sed O
c
t
ober 1
0
, 201
4; Acce
pted
No
vem
ber 5,
2014
System Identification and LMI Based Robust PID
Control of a Two-Link Flexible Manipulator
M. Khairudin
1
, Z. Mohamed
2
, A.R. Hus
a
in
2
1
Department o
f
Electrical Eng
i
ne
erin
g, Univ
e
r
sitas Neg
e
ri Y
o
g
y
ak
arta, Ind
ones
ia
No 1. Jln Co
lo
mbo Yog
y
a
k
art
a
, telp. +
62-27
4-54
816
1
2
F
a
cult
y
of Ele
c
trical Eng
i
ne
e
r
ing, Un
iversiti
T
e
knologi Ma
la
ysi
a
, Mala
ys
ia
e-mail: mo
h_k
hair
udi
n@
ya
ho
o.com
A
b
st
r
a
ct
T
h
is pap
er pr
esents i
n
vesti
gatio
ns into th
e dev
elo
p
m
ent
of a lin
ear
matrix ine
q
u
a
liti
e
s (LMI)
base
d
ro
bust P
I
D control
of a
non
lin
ear T
w
o-
Link F
l
ex
ib
le M
ani
pul
ator (T
LF
M) incor
porati
n
g pay
lo
ad. A s
e
t
of lin
ear
mo
de
l
s
of a TLFM i
s
obtai
ne
d by
usin
g syst
e
m
i
dentific
atio
n method in
w
h
ic
h
the li
ne
ar mo
del
repres
ents the operating r
anges
of the
dy
nam
i
c syst
em
.
Thus, th
e LM
I constraints permit to robus
tly
guar
ante
e
a ce
rtain pertur
bati
on rej
e
ctio
n lev
e
l an
d a re
gi
o
n
of pole l
o
cati
o
n
. T
o
study the effectiven
ess
of
the contro
ll
er, i
n
itial
l
y a PID c
ontro
l
is dev
el
o
ped for T
L
F
M
w
i
th varying
pa
yloa
ds. T
he p
e
r
forma
n
ces
of the
controllers are assessed
in
term
s of
the input tracking
c
o
ntroller c
a
pabilit
y of the system
as c
o
m
par
ed to
the res
pons
e
w
i
th PID contr
o
l. More
over, t
he r
obustn
ess
of the
LMI bas
ed r
obust PID
control
sche
m
es i
s
discuss
ed. F
i
n
a
lly, a co
mp
ara
t
ive assess
me
nt
of the contro
l strategies is p
r
esente
d
.
Ke
y
w
ords
: LM
I, PID, system identific
atio
n,
tw
o-link flexi
b
le
ma
ni
pul
ator
1. Introduc
tion
Flexible ma
ni
pulator
ro
bot
s a
r
e u
s
ed i
n
a wi
de
spe
c
trum
of appl
ication
s
starti
ng from
simple pi
ck and place ope
rations of an i
ndu
strial
ro
bo
t to micro-surgery, mainten
ance of nucle
ar
plants an
d
spa
c
e
ro
boti
c
s [1]. More
over, the
dy
namic
be
hav
iour of
the
manipul
ator is
signifi
cantly a
ffected by pa
yload variatio
ns. If
the adv
antage
s a
s
so
ciated
with li
ghtne
ss
are
not
to be sa
crificed, accurate model
s and e
fficient
cont
ro
llers fo
r a TL
FM have to be develop
ed.
The m
a
in g
o
a
l of mo
dellin
g of a
TLFM i
s
to a
c
hi
eve
an a
c
curate
model
rep
r
e
s
enting the
actual
syste
m
behavio
ur.
A good ag
reement
b
e
tween mo
dellin
g and exp
e
ri
ments h
a
s
b
een
achi
eved [2].
Zho
u
et.al
[3] pre
s
e
n
ts the
neu
ral
netwo
rk o
n
line m
odellin
g
tech
nolo
g
y
to
approximate the system u
n
ce
rt
ain mod
e
l a spa
c
e
manipul
ator. Dog
an and I
s
tefano
pulo
s
[4]
have devel
op
ed the finite
e
l
ement mo
del
s to d
e
scribe
the defle
ction
of a pla
n
a
r
two-li
nk flexible
robot ma
nipul
ator. De L
u
ca and Sicilia
n
o
[5]
have utilised the AM
M to derive a
dynamic mo
del
of multilink flexible robot
arm
s
limiting
to the
ca
se
of pla
nar m
anipul
ators
with no to
rsi
o
nal
effects. Su
bu
dhi a
nd M
o
rris [6] h
a
ve a
l
so
pre
s
e
n
te
d a
syste
m
at
ic a
pproa
ch f
o
r d
e
rivin
g
t
h
e
dynamic e
q
u
a
tions for n
-
li
nk mani
pulat
or wh
ere two
-
hom
oge
nou
s transfo
rmati
on matri
c
e
s
are
use
d
to describe the rigid a
nd flexible motions respe
c
tively.
Ne
wly eme
r
g
i
ng techniqu
e
for optimi
s
in
g the c
ontroll
er pa
ram
e
ters is th
e u
s
e
at linear
matrix inequalities (LMI).
The
works
in formul
atin
g set of L
M
Is to ove
r
co
me the effe
ct o
n
mismat
che
d
uncertaintie
s
in dynamic
sy
stem ha
s
al
so su
rfaced in
the literatures [7]. Since LMIs
can b
e
solve
d
efficiently by standa
rd nu
meri
cal
alg
o
ri
thms, this ha
s prompted a
great num
be
r o
f
resea
r
chers t
o
de
scribe
dif
f
erent
control
pro
b
le
m
s
in
terms
of LMI
s
[8]. Bevrani
and
Hiyama
[9]
pre
s
ente
d
an
LMI based robu
st control
to mainta
in the rob
u
st pe
rforma
nce an
d minimize t
he
effect of disturbance and
specified unce
rtainties of power
sy
stem stabilizers.
On the
othe
r
hand, th
e im
portant
feature of
L
M
I ba
sed
robu
st PI
D d
e
si
gn
app
roa
c
h
i
s
that the deriv
ative term at the
cont
rolle
r app
ears in
su
ch a fo
rm
that enable
s
to con
s
ide
r
t
he
model
un
ce
rtainties, to
b
e
co
nsi
d
e
r
ed
i
n
the
de
sig
n
. Assumin
g
th
e st
ru
cture
d
f
eedb
ack
mat
r
ix,
this app
roa
c
h
is app
rop
r
iat
e
for de
centralize
d
PID co
ntrol de
sign.
The gua
ra
nte
ed co
st co
ntrol
pre
s
ente
d
with a
ne
w qu
a
d
ratic cost fu
nction
in
cludi
ng the
de
riva
tive term fo
r
state ve
ctor a
s
a
tool to influence the overshoot
and re
spon
se rate [1
0]. Using LMI
appro
a
ch to desi
gn a rob
u
st
PID co
ntrolle
r p
r
e
s
ented
[11],[12],[13]. On oth
e
r
ha
nd, Lian
g
et.
a
l
[14] imple
m
ented
a fu
zzy
adaptive PID controller
whose duty is to make
su
re the un
ce
rtainty and n
online
a
ritie
s
of
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 829
– 838
830
hydrauli
c
ere
c
ting me
chan
ism. Usi
ng
th
e
LQ
R
to
solve flexible
link
robu
stn
e
ss
and
i
n
p
u
t
tracking
cap
ability of hu
b angul
ar po
sition [15].
Ho
wever, m
o
st of the
pu
blish
ed
wo
rk on
robu
st P
I
D de
sign
via LMI
wa
s b
a
se
d on
simulation exerci
ses
with limited possibility fo
r
experim
ental v
a
lidation
due to compli
cated
controlle
r st
ru
cture. M
o
reo
v
er, not mu
ch wo
rk
on
L
M
I robu
st
co
ntrolle
r of a T
L
FM with
pay
load
been
rep
o
rte
d
. This is
a chall
engin
g
task for
a MI
MO syste
m
and the
syst
em beh
aviou
r
is
affected by several facto
r
s. This pape
r pre
s
ent
s the desi
gn and d
e
velopme
n
t of a robust PID
control b
a
se
d on
LMI for a no
n-lin
ea
r two-li
nk fl
exible ma
nipula
t
or. It is fou
n
d that the
L
M
I
approa
ch ha
s not been explore
d
for co
ntrol of a tw
o-link flexible
manipul
ator
whe
r
e the sy
stem
dynamics ha
ve unce
r
taint
i
es du
e to the variati
on
of payload
s. Usin
g t
he robu
st control
l
er,
identified PI
D gai
ns
ca
n
be u
s
ed
fo
r all p
a
yload
s with
satisf
actory
re
spo
n
se
s. Thi
s
i
s
an
advantag
e a
s
com
pared to
Ziegle
r
-Ni
c
h
o
ls
(ZN)
tun
e
d
PID control
whi
c
h
need
s to be
re-tu
n
ed
for different p
a
yload
s. Subseq
uently, the dynamic
m
odel is represented into co
nvex formulat
ion
whi
c
h lea
d
s to the form
ulation of sy
stem r
equi
re
ment into L
M
Is re
pre
s
e
n
tation that ca
n
accomm
odat
e the convex
model. A set of robu
st PI
D gain
s
is th
en
obtaine
d by
solving th
e L
M
Is
with d
e
si
red
spe
c
ification
s
. For pe
rform
ance a
s
se
ssment, ZN-PI
D
a
n
d
LMI-PI
D
cont
rolle
rs
are
comp
ared to
control of the
manipul
ator i
n
terms
of input trackin
g
, deflectio
n re
d
u
ction level
a
n
d
robu
stne
ss to
payload va
ri
ations
of both
links. Ex
peri
m
ental results sho
w
that b
e
tter ro
bu
stn
e
ss
and sy
stem p
e
rform
a
n
c
e a
r
e a
c
hieved
with LMI-PI
D
controlle
r de
spite usin
g a single set of PID
gain
s
.
2. Rese
arch
Metho
d
The
physi
cal
pa
ramete
rs
of the T
L
FM
syste
m
c
o
ns
id
er
e
d
in
th
is
s
t
u
d
y
ar
e
s
h
ow
n in
Table 1.
M
h2
is the
ma
ss
consi
dered
at the
se
con
d
m
o
tor
whi
c
h i
s
l
o
cate
d in
bet
wee
n
b
o
th lin
ks,
J
hi
is the
ine
r
tia of the
i
th m
o
tor a
nd h
ub.
The in
put torque,
)
t
(
i
is appli
ed at ea
ch
m
o
tor a
nd
G
i
is the gea
r ratio for the
i
th motor. Both links and
motors are consi
dered to
have the sa
me
d
i
me
ns
io
ns
.
Table 1. Para
meters of a TLFM
S
y
m
b
ol
Parameter
Link-
1
Link-
2
Unit
M
L1
, M
L2
Mass of link
0.08
0.05
kg
ρ
Mass density
2666.67
2684.56
kgm
-
1
EI
Flexur
al rigidit
y
1768.80
597.87
Nm
2
J
h
Motor and h
ub in
ertia
1.46 x1
0
-
3
0.60 x1
0
-
3
kgm
2
M
p
Pa
y
l
oad mass m
a
x
-
0.1
kg
J
p
Pa
y
l
oad inertia
max
-
0.05 x 1
0
-
3
kgm
2
l
Length of link
0.5
0.5
m
Width of link
0.03
0.025
m
Thickness of link
2 x10
-
3
1.49
x1
0
-
3
m
J
o
Moment of inerti
a
5 x10
-
3
3.125
x10
-
3
kgm
2
M
h2
Mass of the centr
e
rotor
-
0.155
kg
A nonline
a
r
TLFM is
a d
i
stribute
d
-paramet
er
syste
m
that can
be de
scrib
e
d
by an
infinite-dim
en
sion
al mathe
m
atical m
o
d
e
l. In pra
c
tice, the red
u
ced-o
r
d
e
r mo
del is u
s
e
d
to
confo
r
m to
computation
a
l
limitations [16]. This sy
stem identifica
t
ion to o
b
tain
a
set of li
n
ear
model
s of th
e TLFM i
n
which th
e line
a
r
mod
e
l
repre
s
ent
s the o
p
e
rating
ra
nge
s of the
dyna
mic
sy
st
em.
The metho
d
construct
s
the system ide
n
tificati
on for a n
online
a
r TLF
M
sele
ctively. In this
se
ction, it i
s
indi
cated
th
at the id
entificatio
n s
y
s
t
em is co
ns
is
te
d
o
f
the
pro
g
r
am ba
se
d o
n
Matlab. The interface of the ident
ificatio
n system ap
p
lication u
s
e
s
identificatio
n tools in Matla
b
.
The recognit
i
on sy
stem
descri
bed i
n
this pap
er is the lea
s
t sq
uare o
ffline para
m
etric
identificatio
n
system. Th
e multisine
sign
al pr
o
d
u
c
e
s
sin
u
soi
d
s of differen
t
amplitudes and
freque
nci
e
s,
whi
c
h are su
mmed to co
n
s
titute a persi
stently excitin
g
sign
al
for the identification
p
r
oc
es
s
.
After the iden
tification re
su
lts are
obtain
ed, it
also
ne
eds to ve
rify wheth
e
r thi
s
model i
s
appli
c
able.
M
odel vali
datio
n an
d
simul
a
tion con
s
ist
o
f
com
pari
ng t
he p
r
edi
cted
output
with t
h
e
measured ou
tput, che
cki
n
g
the tran
sie
n
t respon
se
usin
g a ste
p
respon
se
plot
for the estim
a
ted
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
System
Identification and LMI
Based Robust PID Control of a
Two-Link .... (M. Khairudin)
831
model, an
d checkin
g
pole
s
an
d zero
s.
The result
shows that th
e wavefo
rm
s of the fore
cast
output and th
e actu
al outp
u
t are b
a
si
ca
lly the
same,
and the mat
c
hin
g
deg
re
e
is abo
ut 94
%.
The pa
ramet
e
rs of the tra
n
sfer fun
c
tio
n
model
a
r
e
obtaine
d fro
m
least-sq
ua
res e
s
timatio
n
. A
sixth-o
r
de
r id
entified mode
l G(s) that ha
s a goo
d match with the first two mod
e
s is obtaine
d:
2.028e009
+
s
7.246e007
-
s
2.989e004
+
s
2.29e004
+
s
1492
+
s
21.72
+
s
0.700
+
s
8.72e007
-
s
3.126e006
+
s
4.518e005
-
s
919
-
s
5153
-
(s)
G
2
3
4
5
6
2
3
4
5
11
For the sa
me
processe
s, a m
odel of link-2 witho
u
t load, a
model
of system with load
0.05 kg a
nd 0
.
1 kg will be o
b
tained.
Simulation re
sults
of
L
M
I based rob
u
st
PID c
ont
rol
of the T
L
FM
are
presente
d
in th
e
time and
freq
uen
cy dom
ai
ns. Th
e
step
s that
are
ne
ce
ssary fo
r t
he d
e
sig
n
i
s
as foll
ow:
Ste
p
1
:
system ide
n
tification of a nonlin
ear TL
FM to obt
ain a sets of linear mod
e
l of a TLFM (eq
.
2).
Step 2
: co
nst
r
uct the lin
ear model in stat
esp
a
ce form.
Step 3
: polytopic m
odel
s
of TLFM with
out
payload, loa
d
0.05 and 0.
1
kg (eq.6).
Step 4
: set the
uppe
r bo
und
spe
c
ification (eq. 13).
Step 5
:
find
X
su
ch th
at the ineq
ual
itie
s are satisf
ied (e
q.11
).
Step 6
:
con
s
truct LMI
s
re
gi
on (
ρ
,
θ
,
α
) u
s
i
ng
eq.(17
)
, eq.(1
8) an
d eq. (1
9).
Step 7
: u
s
ing the re
sult
of
X
form the previou
s
step
, calcul
ate gai
n
K
(eq.14
).
Step 8
: appli
e
d
the gain
K
t
o
the LMI ba
sed
rob
u
st P
I
D co
ntrolle
r
for a no
nline
a
r
TLFM.
Step 9
: che
ck the
output and re
peat from st
e
p6 to prod
uce
the desired o
u
tput.
In this work,
Linea
r Qua
d
ratic Re
gulato
r
(LQ
R
) app
roach is con
s
i
dere
d
as a b
a
si
s for
tuning the
co
ntrolle
r gai
n since thi
s
app
roach can
gi
ve nice robu
st
ness a
nd it
can be
form
ul
ated
in term
of p
e
rform
a
n
c
e
based
optimi
z
ation
pr
oble
m
whi
c
h
can
be
solved
usin
g nu
meri
cal
techni
que [12
].
Proposition 1
:
Sch
u
r Co
mpliment
s to determi
ne ma
trix inequality [19]:
0
)
x
(
Z
)
X
(
S
)
x
(
S
)
x
(
V
T
(1)
This al
so kno
w
n a
s
Schu
r
compl
e
me
nt in whi
c
h this p
r
ope
rty is very useful to co
st the
imposed cons
traint in to LMIs
s
e
t
s
.
3. Robus
t PID Con
t
roller
Design
Con
s
id
er an
uncertain
sixth orde
r of the
system for li
nk-1
6
5
2
4
3
3
4
2
5
1
6
6
5
2
4
3
3
4
2
5
1
1
)
(
d
S
d
S
d
S
d
S
d
S
d
S
n
S
n
S
n
S
n
S
n
S
n
s
G
(2)
and simil
a
rly the sam
e
orde
r of the syste
m
for lin
k-2, whe
r
e the pa
rameter vary i
n
intervals:
]
,
[
]...
,
[
],
,
[
],...
,
[
6
6
6
1
1
1
6
6
6
1
1
1
n
n
n
n
n
n
d
d
d
d
d
d
(3)
whe
r
e
i
d
,
i
d
and
i
n
,
i
n
are lo
wer and
up
per
boun
ds for the u
n
c
ertai
n
pa
ra
met
e
rs
denum
erator and nume
r
at
or
of
the syst
em
respe
c
tive
ly. Accordi
n
g to A
s
trom
a
nd
Hag
g
lund
[20],
a PID controll
er with the
structure:
s
K
s
K
K
s
C
d
i
p
)
(
(4)
is ad
equ
ate f
o
r
su
ch
a sy
stem. Fo
r a
TLFM b
o
th
t
w
o in
puts, th
e re
sulta
n
t n
egative feed
b
a
ck
sy
st
em.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 829
– 838
832
The obj
ectiv
e
of PID con
t
roller
de
sign
is to dete
r
m
i
ne PID settings to m
eet
variou
s
desi
gn spe
c
ification
s
. In this pap
er, the
PID cont
roller is desi
gne
d in the state sp
ace
setting
s for
the ea
se
of
usin
g LMI
ap
proa
ch.
The
feedba
ck
system
can be expre
s
s
ed i
n
the
state
sp
ace
descri
p
tion:
Bu
Ax
x
Cx
y
(5)
whe
r
e
y
a
r
e
the syste
m
o
u
tput and th
e
referen
c
e in
put re
spe
c
tiv
e
ly,
x=
[
x
1
x
2
x
3
x
4
x
5
x
6
]
T
the
state with vari
able
s
.
6
6
x
R
A
,
1
6
x
R
B
,
6
1
x
R
C
.
The tran
sfer f
unctio
n
mode
ls (eq
uation 3
)
conve
r
t to a state-spa
c
e
model. In the state-
spa
c
e
mod
e
l
,
the PID
co
ntrolle
r d
e
si
g
n
be
com
e
s a
static state
feedba
ck
co
ntrolle
r, an
d
the
static fee
dba
ck
gain
K=
[
K
p1
K
d1
K
i1
K
p2
K
d2
K
i2
] simply contai
ns
all the PID
controller p
a
ramet
e
rs.
Note
also th
at there
a
r
e
several u
n
ce
rtain p
a
ramet
e
rs in
(4
) a
n
d
the
polytop
ic u
n
certain
set
redu
ce
s t
o
32
32
31
31
22
22
21
21
12
12
11
11
B
,
A
,
B
,
A
,
B
,
A
,
B
,
A
,
B
,
A
,
B
,
A
Cov
(6)
whe
r
e the ve
rtex matri
c
e
s
[
A
ij
,B
ij
] are de
termine
d
ba
sed on th
e sy
stem identifica
t
ion re
sults,
i
is
varying load
and
j
is link.
3.1. LMI Bas
e
d robus
t PID Con
t
roller
This
sectio
n pre
s
ent
s
th
e
co
ncept of LMI
an
d p
r
e
s
ent
s the
co
nstrai
nts u
s
e
d
in
the
controller synthesi
s
probl
em whi
c
h
will
be used fo
r t
he robust PID controller design. In several
control p
r
obl
ems, it i
s
wel
l
motivated to
ba
se th
e de
sign
on th
e L
Q
R
co
ntrol th
eory fo
r its ni
ce
robu
stne
ss [17]. The sta
ndard LQ
R probl
em is to determin
e
the signal
control
u
wh
ich
minimizes the
quadratic co
st:
0
T
T
dt
)
Ru
u
Qx
x
(
)
u
(
J
(7)
for an i
n
itial
state x(0
)
,
whe
r
e
Q
an
d
R
a
r
e
sy
mmetric po
si
tive semi-def
inite matrix
and
symmetri
c
positive definite ma
trix, res
p
ec
tively, i.e.
Q
T
≥
0
and
R=
R
T
>
0. A
s
sume th
at (
A,B
)
are controlla
ble
and (
Q
1/2
, A
) are ob
servable. It turns
out that the sol
u
tion
u
*
to this optimal
control proble
m
can be exp
r
esse
d
[17] in the state feedba
ck form:
Xx
B
R
Kx
u
T
1
*
whe
r
e
X
is th
e symmetri
c
positive defini
t
e solution of
the
algeb
rai
c
Riccati eq
uati
o
n
(ARE
):
0
Q
X
B
XBR
XA
X
A
T
1
T
(8)
and the mini
mum qua
drati
c
co
st [12] is given by
)
0
(
Xx
)
0
(
x
J
T
min
(9)
Thus, the
so
lution to the
LQR p
r
o
b
le
m relie
s on
solving the
ARE (8
). An efficient
alternative fo
r this proble
m
is the LMI te
chniqu
e that h
a
s em
erged
rece
ntly as a
powerful d
e
si
gn
utility for a va
riety of control probl
ems d
ue to it
s conv
exity [18]. By the LMI technique, the LQ
R
probl
em can
be rep
h
rased
as an optimi
z
ation p
r
obl
e
m
over
X
and
Y
:
)
0
(
x
X
)
0
(
x
min
1
T
Y
,
X
(10
)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
System
Identification and LMI
Based Robust PID Control of a
Two-Link .... (M. Khairudin)
833
subj
ect to
0
X
,
0
R
0
Y
0
Q
X
Y
X
B
Y
BY
XA
AX
1
1
T
T
T
T
(11
)
whe
r
e
K
X
Y
. In s
e
veral prac
tic
a
l s
i
tuations
, t
he obje
c
tive (1
0) is represen
ted as:
)
0
(
x
X
)
0
(
x
1
T
(12)
whe
r
e
is the spe
c
ified up
per bo
und. T
he above in
e
quality can al
so be exp
r
e
s
sed a
s
LMI:
0
)
0
(
)
0
(
X
x
x
T
(13
)
Con
s
e
quently
, the optimization pro
b
le
m in (10) an
d (11) in co
nverted to seeki
ng a sol
u
tion
)
,
(
*
*
Y
X
that satisfies a set of LMIs in (11) a
nd (13) an
d the st
ate feedba
ck gain is give
n by
1
*
*
)
(
X
Y
K
(14)
The sy
st
e
m
mat
r
ix
B
A
,
is usually not pre
c
isely kno
w
n
in practi
ce.
Assum
e
tha
t
[
A
ij
,B
ij
], is uncertai
n
but be
long
s to a pol
ytopic set:
32
32
31
31
22
22
21
21
12
12
11
11
,
,
,
,
,
,
,
,
,
,
,
B
A
B
A
B
A
B
A
B
A
B
A
Cov
(15)
whe
r
e
cov
ref
e
rs to a
conv
ex hull, or
B
A
,
if
ij
ij
B
,
A
)
u
,
x
(
w
B
,
A
whe
r
e
i
and
j
is varying lo
ad and lin
k resp
ectively, and
w
the weighting functi
on co
nstraine
d
betwe
en 0 an
d 1.
3.2. Pole Pla
cement L
M
Is
Chilali an
d G
ahinet [19] prese
n
ted a re
gion of the complex plan
e
S(
α
,
ρ
,
θ
)
wh
ere
α
,
ρ
,
and
θ
are min
i
mum de
cay rate, the disk
of radiu
s
and
inner a
ngle, i
n
the form
x +
jy
sat
i
sf
y
x
)
cot(
y
,
jy
x
,
0
x
(16
)
whe
r
e
α
,
ρ
a
r
e the minim
u
m de
cay rate
and the
disk
of radiu
s
re
spectively,
θ
i
s
the
se
ctor
of
the cente
r
ed
at the origin a
nd inne
r angl
e.
Proposition
2
: The clo
s
e
d
-
loop
pole
s
of
the
system
with a state
-
feedb
ack
u=
Kx
are in
sid
e
the
regio
n
S(
α
,
ρ
,
θ
)
if there exists a symm
et
ric definite positive matrix
X
and a matri
x
Y
such t
hat
0
X
2
B
Y
BY
XA
AX
T
T
T
(17)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 829
– 838
834
0
X
BY
AX
B
Y
XA
X
T
T
T
(18)
0
)
B
Y
BY
XA
AX
(
cos
)
B
Y
BY
XA
AX
(
sin
)
B
Y
BY
XA
AX
(
sin
)
B
Y
BY
XA
AX
(
cos
T
T
T
T
T
T
T
T
T
T
T
T
(19)
and
K=
YX
-1
is the state fe
edba
ck gain.
4. Results a
nd Discu
ssi
on
A step signal
with amplitud
e of ±0.5 rad is us
e
d
as an
input positio
n in radian ap
plied at
the hu
b of li
n
k
-1
of th
e ma
nipulato
r
. Th
e
sa
me fo
rm of
sig
nal with a
m
plitude of
±0.35 rad
is
u
s
ed
as the inp
u
t sign
al for lin
k-2. T
w
o sy
stem
respon
ses nam
ely the hub an
gul
ar po
sition
s and
deflectio
ns
at 10 cm from t
he hu
bs
of b
o
th links
with
the freq
uen
cy respon
se of
the defle
ctio
ns
are
obtain
e
d
and
evalu
a
ted. Mo
reo
v
er, the
eff
e
cts of varying payloa
d
on
co
ntroller
perfo
rman
ce
s are
al
so
stu
d
ied. Fo
r th
e
s
e i
n
vest
ig
ations, th
e
syst
em with
out p
a
yload, a
nd t
he
system
with p
a
yload
s of 0.05 kg a
nd 0.1
kg are
con
s
i
dere
d
.
To dem
on
strate the p
e
rfo
r
man
c
e
of th
e LMI
ba
se
d robu
st
PID controlle
r with
the
pol
e
placement pa
ramete
rs
α
,
ρ
are
the mi
ni
mum d
e
cay rate
α
<
-1
a
nd
the di
sk
of ra
dius
ρ
=2
π
/10T
s
r
e
spec
tively.
θ
is the se
ct
or of the cen
t
ered at the
origin
and in
ner a
ngle
θ
=25
deg
re
e. Inner
angle i
s
de
si
gned fo
r covering
un
certa
i
nties pa
ram
e
ter of seve
ral
conditio
n
s i
n
this study
wi
th
varying paylo
ad. As com
p
aring the
performan
ce
of
LMI based robu
st
PID controlle
r, a PID
controlle
r i
s
desi
gne
d u
s
i
ng Zi
egle
r
Ni
chol
s
metho
d
for
co
ntrol
o
f
a T
L
FM. A
block
diag
ra
m is
utilised to obt
ain the propo
rtional gai
n,
K
p
, integral gain
K
i
, and the derivative
gain,
K
d
. In t
h
is
study, the task of the co
ntrolle
r is for input
tracki
ng ca
pability of the system. The angu
lar
positio
n of li
n
k
-1
a
nd lin
k-2 a
r
e fe
d b
a
c
k to
cont
rol
of a T
L
FM
with varying
p
a
yload.
Tabl
e 2
summ
arie
s th
e PID controll
er gain u
s
in
g Z-N
PID for t
he TLFM
with
varying payload
s.
On the oth
e
r
hand, by utili
sing th
e LMI
based r
obu
st
PID co
ntroll
er that
wa
s d
e
sig
ned
based o
n
the
dynamic
beh
aviour of the
TLFM with
va
rying paylo
ad
s. The p
a
ram
e
ters
of the L
M
I
based
rob
u
st
PID controll
er fo
r a T
L
FM with va
rying payloa
d
s
a
r
e
K
p1
=0.13,
K
i1
=0.03
43,
K
d1
=0.05 and
K
p2
=0.090,
K
i2
=0.037,
K
d2
=0.05 for link-1
and link-2 re
spe
c
tively.
Table 2. PID para
m
eters u
s
ing Zie
g
le
r Nichol
s
No Pa
y
l
oad
Link-1 Link-2
K
p
1
K
i1
K
d1
K
p2
K
i
2
K
d
2
1
0
0.58
0.07 0.16
0.09 0.12
0.01
2
50
g
0.59
0.07 0.18
0.09 0.13
0.05
3
100
g
0.61
0.07 0.18
0.10 0.15
0.05
Sy
stem
w
i
th
out an
d
w
i
th
Pa
y
l
oad
. Figure
1 sho
w
s the ang
ular
positio
ns
of the TLF
M
without p
a
ylo
ad for both li
nks. Both u
s
i
ng LMI
ba
se
d ro
bu
st PID cont
rol a
nd
Z-N PID
cont
rol
results
show
similar results for lin
k-1
a
nd lin
k-2, wh
ere
stea
dy st
at
e an
gula
r
p
o
sition
levels
of -
0.5 rad an
d 0.35 rad
were achi
eved resp
ectively.
The tran
sie
n
t resp
on
se sp
ecification
s
of the
angul
ar p
o
siti
on for
both li
nks with
out a
nd with
paylo
ad a
r
e
summ
arised in
Tab
l
e 3. Using
L
M
I
based ro
bu
st PID control,
the sy
stem
exhibits lo
we
r settling ti
m
e
s an
d small
e
r oversh
oot
s for
both links co
mpared u
s
ing
Z-N PID.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
System
Identification and LMI
Based Robust PID Control of a
Two-Link .... (M. Khairudin)
835
(a) Li
nk
-1
(b) Li
nk
-2
Figure 1. Angular p
o
sitio
n
of the system.
Table 3. Rel
a
tion betwe
en
payload
s an
d
spe
c
ificatio
n
s
of angul
ar p
o
sition
Pa
y
l
oad
Link-1 Link-2
Settling time (s)
Overshoot
(%)
Settling time (s)
Overshoot
(%)
LMI PID
Z-N PID
LMI PID
Z-N PID
LMI PID
Z-N PID
LMI PID
Z-N PID
0
1.54 1.70 3.07
7.40
1.18 1.64 0.00 4.86
50
g
1.70 1.77 3.08
11.74
1.29 1.66 0.00 5.91
100
g
1.75 1.81 3.10
12.40
1.35 1.74 0.01 9.40
Figure 2
sh
o
w
s re
sult
s of
the defle
ction
re
sp
o
n
ses o
f
link-1 a
nd li
nk-2. It is
not
ed that
the ma
gnitud
e
s
of vibratio
n of th
e d
e
fle
c
tion
re
sp
on
ses
de
crea
se f
o
r
both li
nks
usin
g L
M
I ba
sed
robu
st PID co
ntrol
com
pare
d
with
PID
co
ntrol.
With L
M
I ba
sed
ro
b
u
st PID control, the m
a
ximum
magnitud
e
s o
f
the resp
on
ses were 3.30
mm and 1.9
2
mm for link-1 an
d link-2
respe
c
tively. On
the other
han
d with PID
co
ntrol, the ma
ximu
m magni
tudes
we
re 7.
02 mm a
nd 4
.
45 mm. Figu
re
3 sho
w
s the
freque
ncy re
spo
n
ses of th
e deflectio
n resp
on
se
s obt
ained
with L
M
I base
d
rob
u
st
PID control and PID control exerci
se
s. These we
re
obtaine
d by
tran
sformi
ng the time resp
onse
into the freq
uen
cy dom
ai
n u
s
ing
Fa
st Fou
r
ier
Tra
n
sform. The
re
sults sho
w
that
cont
roller
perfo
rman
ce
s are characte
rize
d by the first two mo
de
s of vibration
s
. With LMI b
a
se
d rob
u
st
PID
control
sho
w
that the vibra
t
ion occu
rs at
5.88
Hz
and
15.69
Hz, a
nd 3.93
Hz a
nd 25.4
9
Hz
for
link-1 an
d lin
k-2
re
sp
ectiv
e
ly. Otherwise, the re
so
na
nce f
r
equ
en
ci
es fo
r lin
k-1
and lin
k-2 we
re
obtaine
d at 7.84 Hz and 2
3
.
52 Hz, an
d 6
.
05 Hz a
nd 2
8
.49 Hz re
sp
ectively usin
g
Z-N PID.
(a) Li
nk
-1
(b) Li
nk
-2
Figure 2. Deflection respo
n
se of the sy
stem
0
1
2
3
4
5
-0
.
6
-0
.
5
-0
.
4
-0
.
3
-0
.
2
-0
.
1
0
Ti
m
e
(
s
)
L
i
nk
-
1
pos
i
t
i
on (
r
ad)
R
o
bus
t
Z-
N
0
1
2
3
4
5
0
0.
1
0.
2
0.
3
0.
4
0.
5
0.
6
Ti
m
e
(
s
)
Li
nk
-
2
pos
i
t
i
on (
r
a
d
)
Robus
t
Z-
N
0
1
2
3
4
5
-12
-8
-4
0
4
8
12
Ti
m
e
(
s
)
Li
nk
-
1
def
l
e
c
t
i
on (
m
m
)
Robus
t
Z-
N
0
1
2
3
4
5
-8
-6
-4
-2
0
2
4
6
8
Ti
m
e
(
s
)
Li
nk
-
2
def
l
ec
t
i
on
(
m
m
)
Ro
b
u
s
t
Z-
N
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 829
– 838
836
(a) Li
nk
-1
(b) Li
nk
-2
Figure 3. Fre
q
. resp
on
se o
f
deflection of
the system
To investigat
e the effects of payload on the
dyna
mic characte
ristic
s of the
system, a
TLFM
with v
a
riou
s paylo
ads was ex
amined.
The
time
respo
n
se
spe
c
ifica
t
ions
of an
g
u
lar
positio
ns hav
e
shown sign
ificant chan
g
e
s with
th
e
variation
s
of p
a
yload
s. It is noted
with
L
M
I
based robu
st
PID cont
rolle
r, the
system
exhibits lo
we
r settling
time
s an
d sm
aller overshoot
s for
both links co
mpared Z-N PID controll
e
r
. The re
sult
s also sh
ow th
at the transie
nt resp
on
se
s of
the system a
r
e affected by t
he variation
s
of payload.
It is noted with increa
si
ng payload
s,
the m
agnitu
des of vibrat
ion of the deflection
increa
se for
both links. Howeve
r, the magnitud
e
s
of vibration
of the deflection re
spon
ses
decrea
s
e
for
both lin
ks wit
h
LMI b
a
sed
robu
st PI
D control com
p
a
r
ed with
PID control.
Ta
ble
4
summ
ari
z
e
s
t
he maximum
magnitud
e
s
o
f
the re
spo
n
ses fo
r lin
k-1
and lin
k-2 a
c
hieved
with L
M
I
based ro
bu
st PID control a
nd PID cont
ro
l.
In this work,
the frequen
cy respon
se
s of t
he deflection is utilised to investigate the
effects
of pay
load o
n
the
d
y
namic
beh
a
v
iour of
th
e system in th
e frequ
en
cy do
main. It is n
o
ted
that the
re
so
nan
ce
mod
e
s of vibrat
ion
of the
syste
m
shift
s
to
lower f
r
equ
en
cie
s
with i
n
crea
sing
payload
s. Fo
r a payloa
d
betwe
en 0.0
5
to 0.1
kg u
s
ing L
M
I based ro
bu
st PID the re
so
na
nce
freque
nci
e
s f
o
r link-1 shifted from 3.92
Hz and 13.
7
4
Hz to 1.96 Hz an
d 13.7
1
Hz for the first
two mod
e
s
o
f
vibration re
spe
c
tively. On the
othe
r h
and, the resonan
ce frequ
enci
e
s of lin
k-2
usin
g LMI ba
sed
rob
u
st PID shifted fro
m
2.02
Hz an
d 17.65
Hz to
1.98 Hz and
15.69 Hz. Ta
ble
5 summ
ari
s
e
s
re
son
a
n
c
e
freque
nci
e
s o
f
the deflection re
spo
n
ses with payload
s for link-1 a
n
d
link-2 with
L
M
I base
d
ro
bust PID
con
t
rol and PI
D control. Thi
s
implies th
at the manip
u
l
a
tor
oscillate
s at lowe
r freq
uen
cy ra
te
s than
those
without
payload.
0
10
20
30
40
50
60
70
80
90
10
0
0
1
2
3
4
5
6
F
r
eq
ue
nc
y
(
H
z
)
M
a
gn
i
t
ud
e (m
m
.
m
m
/
H
z
)
R
o
bus
t
Z-
N
0
10
20
30
40
50
60
70
80
90
10
0
0
0.
5
1
1.
5
2
2.
5
3
F
r
eq
ue
nc
y
(
H
z
)
M
agni
t
u
d
e
(
m
m
.
m
m
/
H
z
)
Rob
u
s
t
Z-
N
Table 4. Effects of payloa
d
s on maxim
u
m m
agnitu
d
e
s of the defl
e
ction re
sp
o
n
se
s
Pa
y
l
oad
(g)
Link-1 Link-2
Robust PID (mm
)
Z-N PID
(mm)
LMI PID (mm
)
Z-N PID
(mm)
0
-3.30
2.41
-7.02
4.93
-1.92
1.47 -4.45
2.43
50
g
-3.33
2.42
-7.09
7.37
-2.05
1.50 -5.93
4.81
100
g
-3.39
2.94
-7.11
7.69
-3.50
1.62 -7.24
4.85
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
System
Identification and LMI
Based Robust PID Control of a
Two-Link .... (M. Khairudin)
837
Table 5. Rel
a
tion betwe
en
payload
s an
d
reso
nan
ce
freque
nci
e
s of
the flexible manipul
ator
Pa
y
l
oad
Link-1 Link-2
Robust PID
Z-N PID
Robust PID
Z-N PID
Mode-1
(Hz
)
Mode-2
(Hz
)
Mode-1
(Hz
)
Mode-2
(Hz
)
Mode-1
(Hz
)
Mode-2
(Hz
)
Mode-1
(Hz
)
Mode-2
(Hz
)
0
5.88 15.69
7.84 23.52
3.93 25.49
6.05 28.49
50
g
3.92 13.74
5.88 23.49
2.02 17.65
5.89 19.61
100
g
1.96 13.71
5.86 15.68
1.98 15.69
5.83 17.65
5. Conclusio
n
The
develop
ment of
dyna
mic m
odel
an
d robu
st
cont
rol of
a T
L
FM
with va
rying
payload
has b
een p
r
ese
n
ted. A set of linear
model ha
s b
een devel
op
ed by taking
throug
h system
identificatio
n of
a nonlin
ea
r
TL
FM app
roach.
A
PID
controlle
r h
a
s, initially, bee
n devel
ope
d
for
control
of a TLFM with
v
a
rying payloa
d
s.
Th
e
L
M
I is
u
n
iversal a
nd can
be a
dapted
for an
y
a
nonlin
ear sy
stem. It can
b
e
extende
d b
y
inco
rpo
r
atin
g othe
r d
e
sig
n
re
quireme
n
t
su
ch that it
is
rep
r
e
s
entin
g in LMI form. A LMI robu
st PID controlle
r has be
en im
plemente
d
for input tra
c
kin
g
control of th
e
TLFM. Pe
rfo
r
man
c
e
s
of the cont
rol
sch
e
mes have
b
een
evaluate
d
in te
rms of
the
input tra
cki
ng
capa
bility of the system
wi
th comp
are
d
PID controlle
r. Simulations
of the dynamic
model
and
L
M
I based
ro
bust PID co
ntrol h
a
ve b
een
ca
rrie
d
out in the
time an
d fre
q
uen
cy
domain
s
whe
r
e the sy
stem
resp
on
se
s in
cludi
ng ang
ul
ar po
sition
s a
nd defle
ction
are st
udied. In
term of i
nput
tracking,
LMI
ba
sed
robu
st PID
ha
s
be
en
sho
w
n
to
be m
o
re
effe
ctive techniq
ue.
These result
s will be verify on the hardwar
e experimental work for f
u
ture work.
Referen
ces
[1
]
D
w
i
v
edy
SK,
Eb
e
r
ha
rd
P
.
D
y
n
a
mic
ana
l
y
si
s of fle
x
ib
le m
ani
pul
ators, a l
i
terature r
e
vie
w
.
Mec
h
anis
m
and Mac
h
in
e T
heory.
20
06; 4
1
: 749-7
7
7
.
[2]
Martins JM, Moham
ed Z
,
T
o
khi MO, Sa da
Costa J, Botto MA. Approac
h
e
s for d
y
nam
ic
model
lin
g of
flexibl
e
man
i
p
u
l
ator s
y
st
ems
. IEE Proc-Contr
o
l T
heory a
nd
Appl
icatio
n.
20
03; 150: 4
01-4
11.
[3]
Z
hou Sh
uhu
a, Ye Xia
opi
ng, Ji
Xi
aomi
ng, Z
h
a
ng W
e
n
hui. Ad
aptive c
ontrol
o
f
space rob
o
t mani
pul
ator
s
w
i
t
h
task spac
e base o
n
ne
ur
al net
w
o
rk.
TEL
K
OMNIKA
. 2014; 12(
2): 349
-356.
[4]
Dog
an M,
Istefanop
ul
os Y.
Optimal
no
nli
n
ear c
ont
rol
l
er
desi
gn f
o
r fle
x
ible
ro
bot m
a
n
i
pul
ators
w
i
t
h
ada
ptive i
n
tern
al mod
e
l.
IET
Contro
l T
heory
and App
lic
atio
ns
. 2007; 1(
3): 770-
778.
[5]
De L
u
ca
A, S
i
cili
an
o B. Cl
o
s
ed-form d
y
n
a
m
ic
mod
e
l
of
pla
nar m
u
lti-l
i
n
k
lig
ht
w
e
ight r
obots.
IEEE
T
r
ansactio
n
s o
n
Systems, Ma
n, and Cyb
e
rn
etics
. 1991; 2
1
: 826-8
39.
[6]
Subu
dhi
B, Mo
rris AS. D
y
na
mic mod
e
ll
ing,
simul
a
ti
on
an
d contro
l of a
mani
pul
ator
w
i
t
h
fle
x
i
b
le
li
nk
s
and j
o
ints
. Ro
b
o
tics and Aut
o
no
mo
us System
. 20
02; 41: 2
57-2
70.
[7]
Sell
ami A, Arzelier D, Mhir
i R,
Z
r
ida, J. A
slidi
ng mo
de c
ontrol a
ppro
a
c
h
for s
y
stem subj
ected to a
norm-bounded uncertainties.
Internatio
na
l J
ourn
a
l of R
o
b
u
st and
Non
lin
ear Co
ntrol
. 2
007; 1
7
: 32
7-
346.
[8
]
Ga
h
i
ne
t P, Apka
ri
an
. A l
i
n
ear ma
trix
i
n
e
qua
l
i
t
i
e
s ap
p
r
oa
ch
to
h
∞
control
.
Internati
ona
l
Journ
a
l Of
Rob
u
st And No
noli
e
n
a
r Co
ntrol
. 199
4; 4: 42
1-44
8.
[9]
Breava
n
i H, Hi
yam
a
T
.
Robust design of power system
stabili
z
e
r: an lm
i approach.
Proc
eed
ing of th
e
IAST
ED Internation
a
l Co
nfere
n
ce of Ener
g
y
and Po
w
e
r S
y
s
t
em.
T
hailan
d
. 200
6: 70-7
5
.
[10]
Rosinova D,
Vesely
V.
Robust dec
entra
li
sed
co
ntrol
l
er desi
gn usin
g
l
m
i.
Internati
o
n
a
l Jo
urn
a
l
of
Co
mp
uters Co
mmu
n
icati
ons
and C
ontrol
. 2
007; 2(2): 1
95-
204.
[11]
Goncalv
e
s EN
, Palhares R
M
,
T
a
kahashi
HC. A novel
appr
oach for
H
2
and H-i
n
fin
i
t
y
ro
bust PI
D
s
y
nt
hesis for u
n
certai
n s
y
ste
m
s
.
Journal of
Process Co
ntrol
. 200
8; 18: 1
9–2
6.
[12]
Ge M, Ciu MS, Wang QG. Robust
PID contr
o
ller
des
ign v
i
a LMI appr
oac
h
. Journ
a
l of Pro
c
ess Co
ntrol
.
200
2; 12: 3-13.
[13]
Olalla
C,
Le
yv
a R, E
l
Aro
u
d
i
A, Garce´s P
,
Quein
nec
I. LMI rob
u
st co
ntrol
desi
g
n
fo
r bo
ost p
w
m
converters.
IET Power Electron
. 2010; 3(
1): 75–8
5.
[14]
Lia
ng
Li, J
i
an
Xie, W
e
i L
i
.
F
u
zz
y
ad
aptiv
e PI
D c
ontro
l
of a
n
e
w
h
y
drau
lic
erectin
g
mec
han
ism.
TEL
K
OMNIKA
. 2013; 1
1
(4): 7
15-7
24.
[15]
Mohamm
ad K, Mohame
d
Z
,
Husain AR.
D
y
namic
mo
d
e
l an
d robust control of fle
x
i
b
le li
nk robo
t
mani
pul
ator.
TEL
K
OMNIKA
. 201
1; 9(2) : 279 – 286
[16]
Ho MT
,
T
u
YW
. Positio
n
co
ntrol of
a s
i
ng
le-l
i
n
k
fle
x
i
b
l
e
ma
n
i
pul
ator
usin
g
H-infin
i
t
y
b
a
se
d PID c
ontrol
.
IEE Proc.-Control T
heory Ap
pl
. 2006; 15
3(5).
[17]
Le
w
i
s F
L
, S
y
r
m
os VL.
Optimal Control
. Joh
n
W
ile
y
& S
ons
. 1995.
[18]
Bo
yd SP, Ghaoui LE, F
e
ron
E, Ba
lkrishn
an V. Linear matri
x
in
eq
ual
it
ies i
n
s
y
stems an
d control the
o
r
y
.
SIAM, Philade
l
phi
a. 199
4.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 829
– 838
838
[19]
Chil
ial
i
M., Gahin
e
t P., H design
w
i
t
h
po
le
p
l
acem
ent const
r
aints an
LMI appr
oach. IEE
E
T
r
ans, Om
Auto. Control.
199
6; 41: 358-
367.
[20]
Astrom KJ, Hagg
lun
d
T
.
PID Contr
o
ll
er
. 2nd
editi
on. In
strument of so
ciet
y
of Ameri
c
a. Rese
arc
h
triang
le park, N
o
rth Caro
lin
a. 199
5.
Evaluation Warning : The document was created with Spire.PDF for Python.