TELKOM
NIKA
, Vol.13, No
.3, Septembe
r 2015, pp. 1
079
~10
8
8
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v13i3.1382
1079
Re
cei
v
ed
De
cem
ber 2
3
, 2014; Re
vi
sed
April 21, 201
5; Acce
pted
May 14, 20
15
Conceptual Design of Multi-agent System for Suramadu
Bridge Structural Health Monitoring System
Seno Adi Putra*
1
, Bamb
a
ng Riy
a
nto
2
, Agung
Har
s
o
y
o
3
, Achmad Imam
4
Schoo
l of Engi
neer
ing a
nd Inf
o
rmatics,
Band
ung Institute of
T
e
chnol
og
y,
Jln. Ganesh
a
No. 10 Ban
d
u
n
g
Indon
esi
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: seno_
ap@
ya
hoo.com
1
, briyanto@lskk.ee.itb.ac.id
2
,
agu
ng.h
a
rso
y
o
@
gmai
l.com
3
, imam@inform
a
tika.org
4
A
b
st
r
a
ct
W
i
reless S
ens
or Netw
ork (W
SN) is s
m
all
e
m
b
e
d
ded
d
e
vic
e
s d
epl
oye
d
i
n
larg
e sca
le
net
w
o
rk w
i
th
capa
bil
i
ty to se
nse, co
mp
ute,
and c
o
mmun
ic
ate. It co
mb
in
e
s
mo
der
n se
nsor, micr
oel
ectr
onic, co
mputati
on,
communic
a
tio
n
,
and distrib
u
te
d process
i
ng t
e
chn
o
lo
gy.
W
S
N has be
en ta
king a
n
import
ant contrib
u
tio
n
in
structural he
alt
h
mo
nitori
ng s
ystem, esp
e
cia
lly in
Sura
mad
u
Bridg
e
, one
of the long
est span br
idg
e
s i
n
Indon
esi
a
co
n
nectin
g
Sur
a
b
a
ya (E
ast Jav
a
) a
nd M
adur
a Isla
nd. D
u
e
to su
bjecte
d
by e
n
viro
n
m
e
n
ta
l
circu
m
stanc
e, it is nec
essar
y
to i
m
pl
e
m
e
n
t intel
lig
ent a
nd a
u
ton
o
m
ou
s W
S
N to monitor th
e br
id
g
e
cond
ition,
dete
c
t the br
idg
e
d
a
mag
e
, a
nd s
end
w
a
rn
in
g
messag
e
to
bri
d
ge
users w
h
en
uns
afe co
nd
iti
o
n
occurs. The multi-a
gent syste
m
is
a pro
m
is
in
g appr
oach to
be i
m
pl
e
m
e
n
te
d on intel
l
i
gent
and a
u
ton
o
mo
us
W
S
N, espec
ial
l
y
in t
he
brid
ge
stru
ctural hea
lth mo
nitori
ng
sy
stem.
In
t
h
is a
ppro
a
ch ag
ent
s
are e
m
p
o
w
e
red
to hav
e sev
e
ra
l inte
lli
ge
nt le
ar
nin
g
ca
pab
iliti
e
s for st
ructural mo
nitori
ng, da
ma
ge detecti
o
n
,
an
d
pr
edicti
on.
This pa
per d
e
scribes
mu
lti-a
gent syste
m
c
once
p
tua
l
des
i
gn that w
ill
be
imple
m
ente
d
as mod
e
l of l
o
ng
span br
idg
e
structural h
ealt
h
mo
nitori
ng syst
em co
nsi
deri
n
g
system archit
e
c
ture and a
g
e
n
t organi
z
a
t
i
o
n
.
Ke
y
w
ords
: wireless sensor network, structu
r
al he
alth
m
o
nitoring system
, m
u
lti-agent sys
tem
Copy
right
©
2015 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Today, se
nsor technol
og
y developme
n
t gro
w
s
ra
pidly with capability not
only for
sen
s
in
g and
sign
al a
c
qui
si
tion, but also
for co
m
putin
g
and
comm
un
icating to
oth
e
r devi
c
e
s
. T
h
is
s
e
ns
or
te
c
hno
lo
g
y
is
c
a
lled
W
i
r
e
le
ss
Se
n
s
o
r
Ne
tw
ork
(W
SN
)
.
It a
l
s
o
us
es
in
te
rn
e
t
te
c
h
no
lo
gy a
s
comm
uni
cati
on media. WSN gives sig
n
i
ficant effect
s
and ch
an
ce
s for further research in several
appli
c
ation d
e
velopme
n
ts
esp
e
ci
ally in brid
g
e
structu
r
al health m
o
nitoring
syste
m
s.
In the ca
se o
f
WSN ap
plication for lon
g
su
spen
sio
n
bridg
e
structu
r
al he
alth mo
nitoring
system,
stru
ctural a
g
ing
a
nd environm
e
n
tal co
ndition
are
su
bje
c
ts
that must b
e
fully monitore
d.
Thus, a
c
cording to the
s
e con
s
id
erations
we n
e
e
d
to monito
r and d
e
tect deficien
c
y or
deform
a
tion stru
ctures du
e to normal
operatio
n
or environ
men
t
al effects such a
s
ambi
ent
temperature
and h
u
midity. In addition,
entire
stru
ctural monito
ring
is ne
ce
ssary t
o
be
con
duct
e
d
after extrem
e co
ndition l
i
ke di
sa
ste
r
. In ord
e
r to
quantify th
e mea
s
u
r
em
ent of struct
ura
l
perfo
rman
ce,
it is necessary to monitor and eval
u
a
te integrity of civil const
r
uctio
n
s. Th
u
s
,
Structu
r
al he
alth monitori
n
g
became a tr
endin
g
metho
d
and re
se
arch topic [1].
Surama
du is the longest spa
n
su
spe
n
s
ion b
r
idge,
with a lot of
cabl
es, in Indone
sia
con
n
e
c
ting Surab
a
ya City and Madu
ra Island. As an interisl
a
nd brid
ge (Java and Ma
dura
Island
), Sura
madu i
s
su
bj
ected
to o
perational
and
e
n
vironm
ental
influen
ce th
at will
affect to
its
perfo
rman
ce.
Therefo
r
e, it need
s a structural heal
th
monitorin
g
system that is respon
sible
for
informatio
n d
i
sseminatio
n
of enviro
n
me
ntal varia
b
le
s affecting
directly to b
r
idg
e
structu
r
e
a
nd
identify the cause of bridg
e
deform
a
tion
.
Curre
n
tly, Surama
du Bri
d
g
e
ha
s b
een
d
eployed
by a
lot of se
nsors perfo
rmin
g specifi
c
role
s and ta
sks. However,
it still uses
cl
ient-serve
r a
ppro
a
ch in which e
a
ch se
nso
r
tran
smit
s its
sen
s
o
r
y data
one h
op fro
m
one
sen
s
o
r
to con
c
entrator o
r
si
nk,
also
call
ed a
s
data
acqui
sition
unit (DA
U
). T
h
is ap
pro
a
ch also follo
ws o
ne se
nsor no
de as p
e
rcep
tor achitectu
r
e descri
bed b
y
[2] in which
all sen
s
o
r
no
des a
c
t as p
e
rcepto
r
and
they are co
ordin
a
ted by
one cent
rali
zed
concentrator. However,
T
he drawba
ck of this architecture i
s
in
scalability in which the sy
st
em
will be
down
if center node fa
ils to
work and unbalance
energ
y
consumption if
we im
plement
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 3, September 20
15 : 1079 – 10
88
1080
multi hop
dat
a tran
smi
ssi
o
n
wh
ere sen
s
or n
ode
s that
is
clo
s
est to
the co
ncentra
tor will t
r
an
smi
t
more
data th
an the
others [3]. In oth
e
r h
and,
o
n
e
of issue
s
th
at must
be
addresse
d
when
impleme
n
ting
WSN i
s
aut
onomo
u
s
op
eration
s
that
has n
o
t bee
n impleme
n
ted in Suram
adu
Bridge
WSN
yet. In this case, auton
omo
u
s me
an
s
tha
t
sensor n
o
d
e
s shoul
d org
anize their o
w
n
netwo
rk coo
peratively using
di
st
ri
bute
d
alg
o
rithm
s
and
n
o
t on
ly sen
d
packets o
r
exe
c
ute
appli
c
ation p
r
ogra
m
s, but it is ac
tively involved in determini
ng de
ci
sion h
o
w n
e
twork is o
p
e
r
a
t
ed,
calle
d in-n
etwork d
a
ta processing [4].
WSN in Su
ra
madu Brid
ge
has n
o
t utilized the se
nso
r
node
cap
a
b
ility in compu
t
ing ye
t
whe
r
e data p
r
ocessin
g
pe
rforme
d in centrali
zed
m
anne
r is shifted to dece
n
tralize mann
e
r
. It
allows a
pie
c
e of
dete
c
tion an
d p
r
edi
ction p
r
o
c
e
s
s to b
e
pe
rforme
d in e
a
c
h
sen
s
o
r
n
ode.
Therefore,
we ne
ed
to
co
nsid
er di
strib
u
ted a
r
ch
ite
c
ture
of
WSN that combi
n
e
s
ce
ntrali
ze
d a
n
d
distrib
u
ted p
r
oce
s
sing.
In this pap
er
we p
r
op
ose a
gent-b
ased a
ppro
a
ch for i
n
-net
wo
rk d
a
ta pro
c
e
s
sing
that will
be im
pleme
n
t
ed a
s
a m
o
del fo
r b
r
id
g
e
st
ru
ctural
health monit
o
ring
system
s,
e
s
pe
cially
in
Surama
du B
r
idge. A
c
cording to [5],
Agent is
co
mputer
syste
m
with
cap
a
b
ility to perf
o
rm
autonom
ou
s
action
in it
s
environ
ment i
n
o
r
de
r to
a
c
hieve its go
al
s. Thi
s
defini
t
ion of
agent
is
conve
n
ient f
o
r
WSN th
at req
u
ire
s
aut
onomo
u
s op
eration
s
. To
desi
gn
WSN based
on m
u
lti-
agent sy
stem
, at least four issu
es
that should b
e
co
n
s
ide
r
ed: sy
st
em archite
c
tu
re, mobile a
g
e
n
t
itinera
r
y plan
ning, middl
e
w
are de
sig
n
, and ha
rd
wa
re de
sig
n
[3]. In this pap
er we de
scri
be
architectu
ral
desi
gn an
d a
gent org
ani
za
tion.
In the agent architectu
ral
desi
gn, we p
r
opo
se
con
c
ept of WSN architectu
re i
n
whi
c
h
netwo
rk
is se
gmented
into clu
s
ters coo
r
dinated
by cl
uster he
ad, al
so
calle
d DA
U (as
mentio
ned
above). In th
is network, we modifie
d
mobile
ag
ent
-ba
s
ed
WSN (MAWSN) d
e
scrib
ed in [6],
hybrid
archite
c
ture
de
scri
b
ed in [
2
], an
d ag
ent o
r
ga
nizatio
n
de
scribed
in [7]. I
n
the
co
ntext of
agent o
r
g
anization, we
con
s
ide
r
a
gent
s
perfo
rming
spesifi
c
tasks
are
deploye
d
on e
a
ch sen
s
or
node an
d co
ordin
a
ted by a manag
er a
gent. We also implement
the intelligent
mobile agen
t for
data agg
reg
a
t
ion.
Whe
n
impl
e
m
enting th
e i
n
telligent m
o
bile ag
en
t, we co
nsi
d
e
r
th
e co
ncept of
itinirerary
planni
ng a
s
p
a
ssive lea
r
ni
ng in o
r
de
r to
make sure t
hat mobile
ag
ent will visit a
ll sen
s
o
r
no
d
e
s
in efficient manne
r. We u
s
e the app
ro
ach of m
obil
e
agent multi
p
le itinera
r
y planni
ng to solve
delay o
r
scal
ability pro
b
le
m [3]. Com
b
ination of
gen
etic alg
o
rithm
and
rei
n
forcement le
arni
ng
prob
ably will
give better re
sult in mobil
e
agent imple
m
entation a
n
d
they are o
u
r
main resea
r
ch
focu
s. Gen
e
t
ic algo
rithm
has
bee
n
proved
to
solve
com
b
inatory p
r
o
b
lems,
whe
r
eas
reinfo
rcement
lea
r
ning,
whi
c
h i
s
suffi
cie
n
t to b
e
impl
emented
on
a sen
s
o
r
n
o
d
e
[8], emp
o
wers
agent to
ma
ke their o
w
n
d
e
ci
sion.
We
will al
so
l
e
verage M
a
rkov
De
cissio
n Proce
s
s (MDP
), as a
part of reinfo
rceme
n
t learni
ng model, an
d Belief-De
si
re-Intention
(BDI) a
s
two ap
proa
ch
es that
can b
e
mapp
ed ea
ch othe
r [9] to create the intelligent
agent de
ploy
ed on a sen
s
or nod
e.
In the context of bridge structural health
m
onitoring
system, we propo
se two m
a
in acto
r
role
s. First, agents d
e
p
loyed on
each se
nsor
performing
spe
s
ific ta
sk, e.g. se
nsin
g
environ
menta
l
variable,
o
u
tlier d
e
tecti
on, re
so
urce
monitori
ng,
sen
s
o
r
n
ode
org
ani
zing,
and
mobile
age
nt transmissio
n. Seco
nd, op
e
r
ators in
moni
toring and co
ntrolling
cent
er will
follo
w up
data that
ha
ve bee
n g
a
there
d
fo
r fu
rther p
r
o
c
e
s
sing
an
d d
e
t
ermine
op
erational
poli
c
i
e
s
according to the brid
ge env
ironm
ental co
ndition.
This p
ape
r d
e
scrib
e
s
con
c
eptu
a
l de
sig
n
of
multi-ag
ent system f
o
r Suramad
u
Bridge
Structu
r
al He
alth Monitori
ng System. It is divided
in
to five sectio
ns. Sectio
n 1
is introd
ucti
on,
Section 2
de
scribe
s
WSN in Suram
a
d
u
Bridg
e
, Section 3
de
scribes
agent o
r
gani
zatio
n
a
n
d
role
s, Section
4 describ
es
agent fram
ework
con
s
id
era
t
ion, and Sect
ion 5 de
scrib
e
s con
c
lu
sion
s.
2.
Wireless Se
nsor Net
w
o
r
k in Suramadu Bridge
Structural health monitoring sy
stem (S
HMS)
i
s
a sy
stem perform
i
ng sensing utilization
and in
-site
a
nd no
n-d
e
st
ructive an
alysis ab
out st
ru
ctural
characteristics,
in
cl
uding
stru
ctu
r
al
respon
se
s, to identify damage dete
c
ti
on, determi
n
e
damag
e lo
cation, predi
ct severity level o
f
damag
e, and
evaluate the effect
s of stru
ctural da
mage. SHM
S
gives grea
t challeng
es for
damag
e dete
c
tion an
d stru
ctural
con
d
ition pre
d
ictio
n
.
There a
r
e
se
veral fun
c
tio
nal re
qui
rem
ents that m
u
st be
con
s
id
ered
wh
en d
e
sig
n
ing
stru
ctural he
alth monitori
ng
system
on Suram
a
d
u
Bridge. Fi
rst, system
must implem
ent
advan
ced te
chniqu
es, go
o
d
perfo
rman
ce, long term
stability, and
econ
omic v
a
lue ratio
nality.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Con
c
e
p
tual Desig
n
of Multi-age
nt Syste
m
for
Suram
adu Bridg
e
Structural… (Se
no Adi Putra)
1081
Secon
d
, syst
em ha
s capa
bility to transmit dat
a, pro
c
e
ss vie
w
, archive d
o
cum
ent, and
sha
r
e
long di
stan
ce
informatio
n.
Third,
syste
m
is
abl
e to
colle
ct data
synchrono
usly
, real-tim
e, lo
ng
term, and hi
erarchi
c
ally. F
ourth,
system
has
capabilit
y
to assess,
control,
and calibrates
itself.
Fifth, system
is able to i
dentify dama
ges a
nd
eva
l
uate structu
r
al health. Fi
nally, system
is
reu
s
abl
e and
upgrada
ble.
In context of applicatio
n pro
c
e
ss, st
ru
ctur
al h
ealth
monitoring
system in Suram
adu
Bridge
ha
s a
b
ility to rep
o
rt environ
men
t
al co
ndi
tion i
n
clu
d
ing
ch
a
nge
s in
wo
rk and l
oad
of
the
bridg
e
, rep
o
rt
strain a
nd d
e
formatio
n st
atus
of bri
d
g
e
main comp
onent
s, re
cord abno
rmal
or
anomaly l
o
a
d
ing
co
nditio
n
(e.g.
sto
r
m, earth
qua
ke, and
overl
oade
d vehi
cl
e), ide
n
tify main
comp
one
nts
damag
e of t
he b
r
idge
an
d failure a
c
cumulation,
send
wa
rning
messa
ge
when
detectin
g
ab
norm
a
l con
d
i
tion, disse
m
i
nate in
form
ation for lo
ad lifting calcul
ation, and
manag
eme
n
t purpo
se e
s
pe
cially ma
intenan
ce
m
anag
ement. Therefore, structu
r
al
h
e
a
l
th
monitori
ng
system in S
u
ra
madu B
r
idg
e
must
co
ns
i
s
t
of se
nsor
sy
stem
s, data
acq
u
isitio
n a
nd
transmissio
n system
s, data pro
c
e
ssi
ng
and cont
ro
l system
s, and
structu
r
al he
alth evaluatio
n
system
s. Th
ese
sub
s
yste
ms m
u
st
pro
v
ide c
apabilit
y to run
op
e
r
ation
auton
o
m
ously
both
in
norm
a
l and a
bnormal con
d
ition. They must able
to
comm
uni
cate
each oth
e
r a
nd ke
ep sy
stem
integrity when a failure
occu
rs. These subsy
s
tem
s
will
be im
plemented using
multi-agent
system
s de
scribed in Se
cti
on 4 of this p
aper.
In ope
ration
a
l
function
pe
rspe
ctive, stru
ct
ural
he
alth
monitori
ng
system in S
u
ra
madu
Bridge i
s
divided into thre
e levels of hi
era
r
chy. Fi
rst
is data colle
ctor level.
In
this level se
n
s
or
system colle
cts input si
g
nals of
se
nsor, data pre
-
pro
c
e
ssi
ng, a
nd data tran
smissio
n
. Secon
d
level is data
pro
c
e
s
sing
and
analy
s
is that is re
sp
onsi
b
le fo
r p
r
ocessin
g
, a
r
chiving, vie
w
i
ng,
storin
g, and
colle
cting all
data gathe
re
d from
sen
s
o
r
nod
es. Th
e
last level is
stru
ctural he
alth
evaluation
s
in
which data a
nalysi
s
and d
o
cum
ent man
ageme
n
t are
perfo
rmed.
Figure 1. Sensor d
eploym
e
nt on t
he middle sp
an of Suram
adu Bri
d
ge
Surama
du
bridge
structu
r
al he
alth m
o
nitoring
invol
v
es a
ppli
c
ati
on d
eploye
d
on
sen
s
o
r
device
s
(see
Figu
re 1
)
. Q
uantity of se
nso
r
n
ode
de
ployed o
n
th
e bri
de m
u
st
cove
rag
e
e
n
t
ire
bridg
e
. In Su
rama
du B
r
idg
e
, there
a
r
e
397
se
nsor
n
ode
s p
e
rfo
r
m
i
ng vari
ou
s roles (Se
e
Ta
ble
1). As co
nseq
uen
ce
s, this system req
u
ir
e
s
accu
rate an
d prop
er d
epl
oyment.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 3, September 20
15 : 1079 – 10
88
1082
Table 1. Lab
e
l
Descri
ption
of Senso
r
De
ployed on Th
e Middle Spa
n
of Suramad
u
Bridge
Sensor T
y
p
e
Label
Qt
y
Bi-Axial Anemometer
B-ANE
2
Tri-Axial Anemo
m
eter
T-ANE
10
Structure Steel T
e
mperatu
r
e Sens
or
SST
24
Structure Co
ncre
te Tempe
r
ature
Sensor
SCT
47
Road Tem
peratu
r
e Sensor
RT
4
AF Tempe
r
atu
r
e
and Relative Hu
midity
Senso
r
AT&RH
8
Corrosion Senso
r
CS
8
Global Positioning S
y
stem
GPS
18
Force Ring
FR
72
Steel Strain Gau
ge
SSG
62
Concrete Str
a
in Gauge
CSG
64
Strain Rosettes
Gauge
SRG
4
Single-Ax
is Acce
lerometer
S-ACC
11
Bi-Ax
i
al Accelerometer
B-ACC
14
T
r
i-
Ax
ial Acceler
o
meter
T
-
ACC
1
Seismic A
ccelero
meter
SE-ACC
2
Bi-Axial Tilt meter
B-TILT
14
Digital Video Camera
DVC
22
Displacement Tr
ansducer
DT
8
Weight-In Motion
S
y
stem
WIM
2
Total Sensor
397
2.1.
Data Ac
quisi
tion and Tra
n
smission Sy
stem (DAT
S)
DATS is
re
spon
sible fo
r
data colle
ctin
g,
sign
al con
d
itioning, dat
a storage, a
nd data
transmissio
n. Wh
en d
e
si
g
n
ing
DATS, i
t
is ne
essa
ry
to und
erstan
d abo
ut sen
s
or type,
sen
s
or
quantity, se
n
s
or lo
cali
zatio
n
, and
mo
nitoring
meth
od
. To
red
u
ce a
nd p
r
eve
n
t di
stortion
of
sig
nal
and el
ectrom
agneti
c
interf
eren
ce, all
da
ta acq
u
is
ition
unit, includi
n
g
se
nsors, m
u
st be
deploy
ed
in
efficie
n
t
m
anne
r
a
nd secu
re area. For effici
en
cy, fewer dat
a a
c
qui
sition
units are b
e
tter.
DATS with proper in
stalle
d
software and
param
et
er configuration for data a
c
qui
sition mu
st abl
e
to colle
ct, pre
-
process, and
store temp
orary data.
Functio
nal d
e
s
ign
of DATS
at least sho
u
ld
follow fo
u
r
re
quirement
s. First, each
DATS
station, calle
d
Data
Acqu
isition Unit (DAU),
w
ill
b
e
pla
c
e
d
on
some
wh
ere
corre
s
p
ondin
g
to
sen
s
o
r
pla
c
e
m
ent (se
e
Fi
gure
2
)
. Se
cond, all
DAT
S
are
config
ured
p
r
op
erly
to e
n
sure
t
hat
system d
e
vel
opment, inte
gration, exe
c
ution,
mainte
nan
ce, and i
m
provem
ent
will be
con
d
u
c
ted
eas
ily. Third,
Sys
t
em is
able to
run its s
e
rv
ic
es for
long
duration, 24 hours a
day, even
when
routine m
a
int
enan
ce i
s
perform
ed o
r
abno
rmal
co
ndition
s occu
rs. The la
st requi
rem
ent is
dynamic topo
logy. It mean
s that
wh
en
o
ne of
syste
m
part i
s
m
a
lfun
ction, the
syst
em reconfigu
r
e
their network
autonom
ou
sl
y.
2.2.
Data Proce
s
s
ing and Co
ntrol Sy
stem (DPCS
)
DPCS in Su
ramadu B
r
idg
e
co
nsi
s
ts of
serve
r
s with
cap
ability to extract, po
st-pro
ce
ss,
archive, an
d
store
la
rge
a
m
ount d
a
ta. I
t
is al
so
re
sp
onsi
b
le to
vie
w
d
a
ta to
use
r
. In thi
s
syst
em,
all data are p
r
oce
s
sed into i
n
formatio
n, and t
hen information is tra
n
s
form
ed into
kno
w
le
dge.
There are at least two DP
CS functio
nal
requi
rem
ents that shoul
d b
e
impleme
n
te
d. First,
DPCS m
ana
ges
co
mmuni
cation
amon
g
all DA
U in
WS
N. In other
hand
s, DP
CS
also
re
ceive
s
a
real
-time
coll
ection
of
coll
ective d
a
ta from
all
DA
U. Seco
nd, di
g
i
tal data th
at ha
s b
een
p
r
e-
pro
c
e
s
sed i
n
DAU a
r
e
se
nt to DP
CS
usin
g inte
rne
t
proto
c
ol th
rough
wi
rele
ss n
e
two
r
k.
T
hen
these d
a
ta wil
l
be store
d
in DPCS for
sho
r
t or long time
duration.
Many
of raw data
and process
ed data will
be store automatica
lly at determined interval
in storage
de
vices. Th
e av
ailable pi
ctu
r
es the
n
w
ill b
e
se
nt to DP
CS user i
n
terface to b
e
vie
w
ed
and co
ntroll
e
d
. From two DPCS units,
operators
or
use
r
s
can re
que
st views
about all cu
rrent
values, se
rvice statu
s
, repo
rts, or notes
ab
out
events,
ch
ange
spe
c
ific p
a
ra
meters,
add/del
ete/ed
it data view at r
outin
e
monitori
ng, chang
e de
scri
ption of
status, ala
r
m, a
n
d
engin
eeri
ng u
n
it, add n
e
w
status of data
view to
sy
st
em, retri
e
ve li
st of data
b
a
s
e and
prog
ra
m,
perfo
rm data
analysi
s
, and
requ
est hi
storical re
po
rts.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Con
c
e
p
tual Desig
n
of Multi-age
nt Syste
m
for
Suram
adu Bridg
e
Structural… (Se
no Adi Putra)
1083
Figure 2. Dat
a
Acqui
sition
Unit (DA
U
) pl
acem
ents o
n
Surama
du Bri
dge
2.3.
Structural Health Ev
alua
tion Sy
stem
(SHES
)
The
core of
stru
ctural he
alth monito
rin
g
sy
st
e
m
is
st
ru
ct
ur
al he
alth evaluatio
n syste
m
(SHES). In the monitori
ng room, sy
stem
evaluate
s
loa
d
cap
a
city
an
d dynamic
wo
rk
s of
st
ru
ct
ur
e
and id
entifies the pote
n
cy
of stru
ctu
r
al
damag
e u
s
in
g onlin
e a
nd
real
-time d
a
ta that have
b
e
e
n
pro
c
e
s
sed by
DPCS. Thu
s
, it is necessary
to implem
ent two types of compute
r
.
First co
mput
er,
calle
d SHES serve
r
, pe
rforms proble
m
solving fo
r
a
nalytical work of limited elements
su
ch
as
static no
n-lin
ear an
alysi
s
, dynamic
a
nal
ysis (wind, seismi
c, and vibration
)
, interactio
n betwe
en
fluid
and stru
cture, and structural
h
ealth
eval
uation. S
e
co
nd
com
p
u
t
er, call
ed SHES-WK (S
HE
S
Works
t
ation), tak
e
s
into acc
o
unt in dat
a grap
hical ana
lysis an
d rep
o
rting.
In the context
of stru
ctural
health mo
nito
ri
ng, Surama
du Bridg
e
WSN take
s into
account
in monito
ring
of highway loadin
g
effect
s, monito
ring
of temperatu
r
e effect
s, m
onitorin
g
of wind
effects, mo
nitoring
of sei
s
mic effect
s,
and mo
nitori
ng of corro
s
i
on. In the co
ntext of stru
ctural
damage detection, WS
N
can be
utilized for vibrat
ion-based damage detecti
on in whi
c
h
we
can
use meth
od
s such a
s
fre
quen
cy cha
n
ges, mod
e
shape chan
ge
s, modal da
mping chan
g
e
s,
freque
ncy
re
spo
n
se fun
c
tion chan
ge
s, mode Sh
ape
Curvatu
r
e
Chang
es, mo
d
a
l Strain e
n
e
r
gy
cha
nge
s, and
flexibility change
s for vibra
t
ion-ba
se
d da
mage dete
c
ti
on.
3.
Agen
t Org
a
n
i
zation an
d Roles
In our re
se
arch, we
wil
l
use a me
thodolo
g
y base
d
on ag
ent-o
riente
d
softwa
r
e
engin
eeri
ng (AOSE) and p
r
opo
se im
ple
m
entation of
intelligent ag
ent cha
r
a
c
teristics de
scrib
e
d
in [5]. AOSE leverages
developm
e
n
t cycle sta
r
ting
from an
alysi
s
to codi
ng b
y
using m
o
d
e
l-
driven en
gine
ering
(MDE
) approa
ch [10]
. There a
r
e
two main a
c
tivities that sho
u
ld be pe
rforme
d
in this meth
odolo
g
y: identifying platform ind
epe
n
den
ce mo
dul
e (PIM) an
d
platform sp
ecific
module
(PS
M
). In PIM
we de
scrib
e
a
gents, th
ei
r
roles, i
n
form
a
t
ion they ex
chang
e, and
t
heir
intera
ction. In
PSM we d
e
scrib
e
d
e
vice
s platform
bot
h ha
rd
ware a
nd softwa
r
e p
l
atform that
will
sup
port multi-agent sy
stem
in WSN.
3.1. Agen
t
Org
a
n
i
zation
There are
four types
of WSN a
r
chite
c
tu
re i
m
pleme
n
ting
multi-age
nt system
orga
nization:
one
sen
s
o
r
n
ode
as
pe
rce
p
tor, on
e a
g
e
n
t ea
ch
sen
s
or n
ode, m
o
b
ile age
nt, an
d
hybrid a
r
chit
ecture [2]. In the case
of Su
ram
adu
Bridge S
H
M, we
will i
m
pleme
n
t hybrid
architectu
re i
n
whi
c
h it co
mbine
s
many
agents
ea
ch
sen
s
o
r
no
de
and mo
bile a
gent (see Fi
g
u
re
3). This
arch
itecture i
s
a
model of a
gent
organi
zation propo
sed in [7] with addition
al our
modificatio
n
.
It introdu
ce
s
agent
s o
n
mi
ddle
w
are lay
e
r
of sen
s
or n
ode
and
sepa
rates a
gent
s i
n
to
two layer of
middle
w
a
r
e: middle an
d a
pplication layer.
In middle
lay
e
r, the
r
e
are
two
age
nts, re
sou
r
ce a
gent
(RA) an
d
sen
s
ing
ag
ent (S
A). RA
is an age
nt with ca
pabilit
y to make de
cisi
on rel
a
ted
to memory and po
wer u
s
age control.
This
agent
dete
r
m
i
nes cost th
at sh
ould
be
p
r
epared
by ot
her age
nts to
pe
rform
its t
a
sks.
Thi
s
a
g
ent
also
ma
ke
s
d
e
ci
sion
ba
se
d on
B
e
lief
-
D
e
sir
e
-I
nt
e
n
tio
n
mo
del
(BDI
mod
e
l) wh
ether to a
c
cept
or
not executing
anothe
r ag
e
n
t requ
est a
c
cording to
co
st
co
nsi
deration.
SA is an
agent that h
a
s
in
te
r
e
s
t
in
phys
i
c
a
l var
i
ab
le
s
an
d
h
a
s
ac
ce
ss
to
s
ens
o
r
co
mp
on
en
ts
(e
.g
. te
mp
e
r
a
t
ur
e
s
ens
o
r
,
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 3, September 20
15 : 1079 – 10
88
1084
accele
rom
e
te
r, light sen
s
o
r
, etc). This a
gent has
re
spon
sibility to
make d
e
ci
sio
n
base
d
on BDI
model for o
u
tlier dete
c
tion
and di
ssem
in
ation of physi
cal varia
b
le
s.
In application
layer, there
are two types
of agent:
manag
er a
g
e
n
t (MgrA
)
an
d mobile
agent
(MA).
MgrA i
s
an
a
gent that i
s
resp
on
si
ble i
n
mana
ging, o
r
gani
zin
g
, an
d neg
otiating
with
other a
gent
s unde
r its
co
ordin
a
tion. M
g
rA al
so
impl
ements B
D
I
model a
nd M
a
rkov De
ci
si
on
Process (M
DP) model to
make de
cision.
MA is an indep
end
ent mobile agent and a
l
so
impleme
n
ts B
D
I and
MDP
model, it pe
rf
orm
s
commu
nicatio
n
with
MgrA to m
a
ke co
ope
ratio
n
.
It
is a sp
eci
a
l a
gent that perf
o
rm
s migratio
n from
one
se
nso
r
nod
e to anothe
r for d
a
ta colle
ction.
Figure 3. Pro
posed multi-a
gent
hybrid a
r
chite
c
ture in
WSN
In the contex
t of mobile a
gent (MA),
we impleme
n
t sin
k
nod
e as a MA dispat
che
r
. It
may be a
sen
s
or
nod
e with
powerful
cap
ability or ot
he
r co
mpute
r
sy
stem
s. In this approa
ch
sin
k
node
send
s
a MA to ta
rg
et are
a
to vi
sit each
sen
s
o
r
no
de i
n
that
are
a
. Data i
n
sen
s
or no
d
e
is
eliminated
an
d collect
ed b
y
the MA. Th
en, MA
with i
t
s colle
cted d
a
ta retu
rn
s to
sin
k
fo
r furth
e
r
data pro
c
e
s
si
ng. This ap
proach will re
du
ce commu
nication tasks a
m
ong sen
s
o
r
node
s.
We impl
eme
n
t multi-ag
en
t system in
WS
N a
nd m
o
dify agent organi
zation p
r
opo
sed in
[12]. Here, we define six d
e
finitions:
1)
DPCS
age
nt is the m
a
in
age
nt of S
H
MS
. It is pl
ace
d
in
DP
CS and
di
spat
che
s
several MAs to aggreg
ate data from
DAU si
nk
n
o
des a
nd co
o
r
dinate
s
sen
s
ory data fu
sion.
DPCS ag
ent is also the co
ordin
a
tor of DAU agent.
2)
DAU a
gent, p
l
ace
d
in DA
U (also called
as
sin
k
),
di
sp
at
che
s
mult
ip
le MA
s t
o
c
o
ll
ect
data from worker
sen
s
o
r
node
s in their clu
s
te
r. DAU can be
called a
s
cl
uster h
ead t
hat
coo
r
din
a
tes
worke
r
sen
s
o
r
unde
r its cl
uster a
r
e
a
. There a
r
e two clu
s
ters co
ordinated by DAU:
Girde
r
DA
U (
G
DAU
)
an
d Env
i
ronm
ent D
A
U (EDA
U).
3)
Wo
rke
r
sen
s
or no
de
s are
respon
sibl
e
fo
r
data colle
cting,
si
gnal con
d
itioning, data
stora
ge, an
d
data tran
smi
s
sion. Mg
rA, RA, and
SA are in
stalle
d on ea
ch
worker
se
nsor n
o
d
e
(se
e
Figu
re 3
)
. It receives and tran
smit
s variou
s mo
bile agent
s di
spat
che
d
by DAU ag
ent. A
worke
r
se
nso
r
n
ode
pe
riod
ically p
e
rfo
r
m
s
se
n
s
in
g p
r
o
c
e
s
s a
r
ou
nd t
heir environm
ent an
d
store
s
sensing data
in local mem
o
ry.
4)
Dire
cto
r
Age
n
t (DA), d
e
p
loyed on
DPCS, is an
entity where
agent
s and
its
cap
abilities
are regi
stered.
The main
role
of DA
is sto
r
i
ng inform
atio
n about MA a
nd age
nt gro
up
profiles with i
t
s capabilities in
its local
database. Wh
en a request
er
requests a
pref
erence, DA
perfo
rms m
a
tchin
g
pro
c
e
s
s to find relat
ed se
nsor no
de ca
pability and send
s re
spo
n
se back
a
set of app
rop
r
iate events t
o
requ
este
r.
5)
Req
u
e
s
ter m
a
y be a huma
n
or othe
r ag
ents who re
q
uest
s
inform
a
t
ion to syste
m
.
6)
Weig
ht in Mot
i
on (WIM)
sin
k
is
a cl
uste
r head
th
at
co
o
r
dinate
s
wo
rker sen
s
o
r
no
d
e
s
measuri
ng th
e axle wei
ght
of passing v
ehicl
e,
veloci
ty of the vehicle
s
, and di
st
ance betwee
n
axles. Here, t
he traffic l
oad
of the bri
dge
can
be m
e
a
s
ured. It al
so
disp
atch
es
a
mobile a
gent
to
colle
ct d
a
ta from its mem
b
ers a
nd
eval
uate n
o
t only
traffic l
oad
o
f
the b
r
idg
e
,
but al
so i
dent
ify
the cau
s
e of
damag
e.
The a
r
chite
c
t
u
ral d
e
si
gn d
e
scrib
ed a
b
o
v
e modi
fies
conceptual a
r
chitecture expl
ained in
[3] that describe
s
two typ
e
s of
a
r
chite
c
ture: a
r
chite
c
ture i
n
hie
r
archical sen
s
or net
work a
nd
architectu
re
in flat sen
s
or
netwo
rk. It also
im
plement
s in
-hiera
rchi
cal
sen
s
o
r
n
e
twork
architectu
re
descri
bed
in
[13] an
d
co
mbine
s
tw
o
types a
r
chite
c
ture
in
flat
sen
s
o
r
n
e
twork:
mobile ag
ent-based dist
rib
u
ted sen
s
o
r
netwo
rk
(MA
D
SN), descri
bed in [14], and mobile ag
ent-
based
WSN (MAWSN), de
scribe
d in [14
]. In the
middle sp
an of Suramadu B
r
idg
e
, there a
r
e fi
ve
DAUs depl
oyed on the bri
dge (See Fi
g
u
re 2
)
, called
DAU1, DAU2, DAU3, DA
U4, and DA
U5
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Con
c
e
p
tual Desig
n
of Multi-age
nt Syste
m
for
Suram
adu Bridg
e
Structural… (Se
no Adi Putra)
1085
respe
c
tively. DAU may b
e
a si
ngle
-
bo
ard compute
r
(SBC), e.g.
Rasp
berry, wit
h
ad
ditional
b
a
se
station, a dev
ice conn
ectin
g
DAU to sen
s
or n
ode
s.
As mention
e
d
in our
six definition
s
, each
DAU
c
o
ordinates
two c
l
us
ters
: EDAU and
G
D
AU
. Ea
ch c
l
us
te
r co
nsis
ts
o
f
w
o
rk
er
se
ns
or
n
ode
s
p
e
r
for
m
ing
e
n
v
iro
n
m
en
ta
l se
ns
in
g
.
A
worke
r
se
nso
r
in
ea
ch
cl
u
s
ter pe
riodi
ca
lly per
fo
rm
s
sen
s
in
g p
r
o
c
ess a
r
o
und
their environ
ment
and c
o
llec
t
s
s
e
ns
ing data in its
loc
a
l database. When detec
ting
an event, worker s
e
nsors
s
end
an explo
r
ato
r
y messag
e to
DAU. T
hen,
DAU
age
nt
di
spat
che
s
m
u
l
t
iple MAs to
collect d
a
ta an
d
accumul
a
te i
t
s si
ze
hop
by hop. After MA re
turn
s to DAU,
DAU ag
ent se
nds
notificati
on
messag
e to
DA, to indica
te that there
are d
a
ta
that
ready to b
e
further
pro
c
essing.
DA then
disp
atch
es M
A
to process
data coll
ecte
d in DA
U.
When a requ
ester req
u
e
s
ts
a prefe
r
en
ce,
DA
perfo
rms m
a
tchin
g
pro
c
e
s
s to find relat
ed DA
U ca
pa
bility, sends
MA to DAU for data g
a
the
r
ing,
and send
s re
spo
n
se ba
ck
a set of appropriate eve
n
ts to requ
este
r. Figure 4
sh
ows an exam
pl
e
of sen
s
or n
e
twork topol
og
y coordi
nated
by DAU1.
One
of the
o
r
gani
zation
go
als th
at mu
st
be a
c
hi
eved i
s
e
n
e
r
gy efficiency to
ma
ke a
lon
g
life WS
N. Strategy-based
comm
uni
cati
on [15]
w
ill
be
considered in thi
s
research. It defines
three aspe
cts
that sho
u
l
d
be co
nsi
d
ered:
i
n
form
ation imp
o
rt
ance comm
unication, which
decrea
s
e
s
to
tal co
mmuni
cation
am
on
g sen
s
o
r
n
o
des,
preventi
ng u
nne
ce
ssary inte
r-se
n
s
or
informatio
n e
x
chan
ge, and
data con
c
at
enation.
Th
e first and
se
cond a
s
pe
ct will be solve
d
by
impleme
n
ting
distribute
d
re
inforceme
n
t learni
ng [
16], descri
bed lat
e
r in this se
cti
on, whe
r
ea
s the
last aspect
will be
solved by implem
enting mo
bil
e
agents that perform
s
synthesi
s
solution,
calle
d sol
u
tion fusion am
o
ng mobile a
g
ents.
Figure 4. An example of sensor net
wo
rk topolo
g
y co
ordin
a
ted by DAU1
3.2.
Mobile Agen
t Migration
One
of imp
o
rtant issue
s
th
at sh
ould
be
con
s
id
ere
d
in
de
signi
ng
m
u
lti-age
nt
system in
WSN i
s
MA
migratio
n pla
nning, al
so
called itine
r
ary
planni
ng. Itinera
r
y is a
rou
t
e that sho
u
ld
be
followe
d by the MA wh
en
performing
migratio
n [1
3
]. There are two mai
n
issu
es that mu
st be
solved
by the
syste
m
a
u
to
nomou
sly
rel
a
ted to
iti
nera
r
y pla
nning:
selectin
g a
set of
sen
s
o
r
n
o
des
that will
be visited
by the
MA and determining
a sequence of sensor n
odes that will be selected
by the MA
co
nsid
erin
g e
n
e
r
gy effici
en
cy [3]. We
will
i
m
pleme
n
t three type
s of
itinera
r
y pl
anni
ng:
static pla
nnin
g
[14], dynamic plan
ning [1
3], and hybrid
plannin
g
[17].
In static
plan
ning, MA mig
r
ation
route i
s
ab
sol
u
tely determi
ned
b
y
sink nod
e b
e
fore M
A
is dispatched
. In our case, sink n
ode is
DAU.
DA
U u
s
e
s
cu
rrent gl
obal net
work
informatio
n a
n
d
obtain
s
effici
ent route b
e
fore
sendi
ng
MA. Powerful
techniq
ue th
at can be
co
nsid
ere
d
in st
atic
itinera
r
y plan
ning i
s
Ge
net
ic alg
o
rithm
(GA). Using
th
is alg
o
rithm,
MA only visits a
sen
s
o
r
n
ode
once a rou
n
d
[18]. In this re
sea
r
ch we use GA
co
nsid
erin
g two
optimization
objective
s: the
distan
ce
bet
wee
n
two
se
nso
r
n
ode
s
(determi
ned b
y
received si
gnal stre
ngth
indication
)
a
n
d
remai
n
ing en
ergy in the next hop sen
s
or nod
e.
In dynamic pla
n
n
i
ng, each MA
determine
s
next
sen
s
o
r
n
ode
that will
be
visited f
r
om
cu
rrent
visit
ed n
ode.
Dy
namic ag
ent
route
al
so
must
con
s
id
er trad
e-off betwe
e
n
sen
s
o
r
no
de dista
n
ce
and re
mainin
g energy. Dynamic pla
n
n
i
ng
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 3, September 20
15 : 1079 – 10
88
1086
approa
ch will
search a se
nso
r
nod
e with enoug
h re
maining e
nergy for dispatching the MA [13].
Finally, in hy
brid pl
anni
ng
, sele
cting a
set of visite
d nod
e is
st
atic whe
r
ea
s determi
ning
a
seq
uen
ce of
visited nod
e is dynami
c
.
In the contex
t of Data fusion pattern, there a
r
e thre
e types of pattern that sh
ould be
determi
ned i
n
the archite
c
t
u
re:
conve
n
tional p
a
ttern,
singl
e MA-ba
s
ed
pattern, and m
u
ltiple
MA-
based p
a
tterns. In
co
nventional p
a
ttern, data
co
lle
cted
by se
nsor n
ode
is t
r
ansmitted
fro
m
s
e
ns
or
to
s
i
nk
. T
h
is
c
o
nc
ep
t is
th
e sa
me
as
c
l
ie
n
t
-s
er
ve
r
p
a
r
a
d
i
gm. In
s
i
ng
le
mo
b
ile
a
g
e
n
t, o
n
l
y
one mobile a
gent is dispat
che
d
to all senso
r
nod
es, whe
r
ea
s in m
u
ltiple mobile
agents, vario
u
s
mobile ag
ent
s are di
sp
atched from
sin
k
to sens
or no
des. We will
use multipl
e
mobile ag
ent
s.
3.3.
Outlier Dete
ction Ag
ent
Agent ca
pabi
lity that shoul
d be defin
ed
in
WSN i
s
o
u
tlier dete
c
tio
n
. When m
o
nitoring
environ
ment, agent
s that are implem
ent
ed on se
ns
or node have to identify
and classify events.
Therefore,
it is necessary to
ensure reli
ability a
nd information quality
in WSN. To avoi
d
transmitting inaccu
rate dat
a, real time d
a
ta analys
i
s
should b
e
co
n
ducte
d in ea
ch sen
s
o
r
nod
e.
Due to u
n
certainty environment, a
sens
or n
ode
may be mal
f
unction o
r
prod
uce
inaccu
rate d
a
t
a. In addition
, reso
urce lim
itation
su
ch a
s
limited p
r
o
c
essing,
stora
ge, ban
dwidt
h
,
and ag
ent a
u
tonomy ma
y cause ina
c
curate ra
w d
a
ta. This is
calle
d an out
lier [21], whi
c
h a
sen
s
o
r
nod
e read
s data th
at deviates from co
mm
on
data. An outlier may be ca
use
d
by noise,
data e
rro
r d
u
e
to ha
rd
ware malfun
ction
,
or mali
ciou
s attack.
To d
e
termin
e a p
r
oper metho
d
that
can
be
imple
m
ented i
n
a
sen
s
o
r
n
ode
with limited
reso
urce, a
te
chni
que
of re
al-time
outlie
r
detectio
n
wa
s introdu
ced. T
h
is tec
hniq
u
e
is perfo
rmed
locally by SA.
3.4.
Agen
t Perfor
ming Reinfo
rcemen
t Lea
r
ning
Another a
g
e
n
t capa
bility that shoul
d
be co
nsi
d
e
r
e
d
whe
n
impl
ementing m
u
lti-agent
system in
WSN is a
b
ility to comm
uni
cate with
its
e
n
vironm
ent without tutor a
ssi
stan
ce [2
2
],
calle
d Rei
n
fo
rce
m
ent L
e
a
r
ning
(RL).
RL i
s
on
th
e top surve
y
that the most ap
prop
riate
techni
que for
WSN o
p
timization [8]. Therefore, RL is t
he tech
niqu
e that will be o
u
r main fo
cu
s to
be implem
ent
ed in MA, MgrA, DA, and DPCS ag
ent.
RL th
at will
b
e
u
s
ed
in
ou
r
WSN is Di
st
ri
buted In
dep
e
ndent
RL
(DI
R
L).
The
aim
of DIRL
is for a
gent
s’
coo
r
din
a
tion
in ord
e
r to
manag
e thei
r own
re
sou
r
ce in di
stribut
ed man
n
e
r
a
n
d
eliminating communi
catio
n
among sen
s
or
n
ode
s
when co
ordina
tion is perfo
rmed [23]. In
our
resea
r
ch, ea
ch ag
ent gro
up deploy on
sen
s
or
n
ode
is selfish an
d implement
DIRL. It means
each agent o
n
sen
s
o
r
nod
e modifies its behavior u
s
i
ng RL alg
o
rit
h
m to allocat
e
tasks sche
dule
autonom
ou
sl
y by lea
r
nin
g
its
utility corresp
ondi
ng to
given state. The re
wa
rd
f
unctio
n
fo
r
e
a
ch
task i
s
descri
bed a
s
com
b
ination betwe
en task outp
u
t and remai
n
ing ene
rgy. In addition, it is
necessa
ry
to ensure
that gl
obal sy
ste
m
behavio
r i
s
th
e colle
ctive ef
fect of i
ndivid
ual a
gent. T
h
i
s
responsibility
belongs to
MA and MgrA cooperat
ion when migration proc
ess is
perform
e
d. MA
and Mg
rA are
also re
sp
on
sible to update
a rewa
rd.
3.5.
BDI and M
D
P Agent
Belief De
si
re
Intention
(B
DI) i
s
rea
s
o
n
i
ng p
r
o
c
e
s
s t
o
dete
r
min
e
action
s th
at
must
be
perfo
rmed
by an age
nt to achi
eve its g
oals. It in
volves two impo
rtant pro
c
e
s
ses: delib
eration
pro
c
e
ss th
at prod
uces a
set of the agent int
entions
or commitme
n
ts and m
e
a
n
s-end
rea
s
o
n
ing
that pre
s
ent
s a seq
uen
ce
of action
s corres
pon
d to sele
cted
co
mmitment. Markov De
ci
si
on
Process
(M
DP) ag
ent i
s
def
ined
a
s
a
state
sp
ace
(S
MK
V
), a se
t of actio
n
s (A
MKV
), a re
w
a
rd
function (R), and a stat
e transitio
n functi
on (T
MK
V
), which defe
nd on
current state
and perfo
rm
ed
action
s.
BDI
age
nts impl
ement
de
scri
ptive app
ro
a
c
h in
whi
c
h
th
ey ma
ke
a d
e
ci
sion
whereas
MDP ag
ents impleme
n
t
pre
s
criptive
approa
ch in
whi
c
h they i
dentify optim
al de
cisi
on.
BDI
agent
s are m
o
re tra
c
tabl
e than MDP [9].
BDI model d
oes n
o
t con
s
i
der sto
c
h
a
sti
c
ac
tio
n
s, BDI assu
me
s that action take
n by an
agent al
ways result desi
r
e
effect, it is po
ssi
ble for inte
ntion plan to
gene
rate u
n
d
e
sirable
effects
and
cau
s
ing
the agent d
e
v
iates from a
seq
uen
ce
of
states that h
a
s b
een pl
an
ned befo
r
e [
9
].
Therefore,
re
con
s
id
eratio
n
of int
entions is re
quired
whe
n
the ag
ent
reali
z
e th
at curre
n
t sta
t
e is
not unde
r its plan. This p
h
enome
non p
r
obably o
ccu
rs wh
en imple
m
enting mo
bile agent dyna
mic
itinera
r
y plan
ning. Thu
s
,
we h
a
ve to
combi
ne B
D
I and M
D
P a
gent so that
we d
o
not
only
develop a
gen
ts to perfo
rm
their task but
also
p
e
rfo
r
m
optimal pe
rforma
nce. We
place B
D
I a
nd
MDP beh
avio
r in MgrA an
d
MA.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Con
c
e
p
tual Desig
n
of Multi-age
nt Syste
m
for
Suram
adu Bridg
e
Structural… (Se
no Adi Putra)
1087
4.
Agen
t Frame
w
o
r
k Consid
eration
The ag
ent pl
atform that
matche
d with
our
desi
r
a
b
l
e
de
sign i
s
Agent Fa
ctory Micro
Edition (AFM
E) [24]. It w
a
s de
sig
n
to adopt BDI para
d
igm in
whi
c
h ag
ents follow se
nse-
delibe
r
ate
-
act
cycle [24].
It describe
s
age
nt thro
ugh Age
n
t Facto
r
y Age
n
t Program
ming
Lang
uage
(A
FAPL) ba
sed
on logi
cal formal of belief
and commitm
ent. This lan
g
uage i
s
u
s
ed
to
encode ag
en
t behavior by determinin
g
rule
s def
ining
condition
wh
en commitm
e
n
t is adopted
.
Unfortu
nately
,
AFME doe
s
not supp
ort
st
rong
mig
r
atio
n. It mean
s th
at AFME only
su
ppo
rt o
b
je
ct
state an
d dat
a migration
(wea
k mi
grati
on), n
o
t co
de
. There
is
a
middle
w
a
r
e p
l
atform that al
lows
agent'
s
co
de
migratio
n (st
r
ong mig
r
ation
)
. This platfo
rm is calle
d Agent Platform
for Sun SPOT
(MASPOT). However, strong migr
ation consum
es more energy
than weak migration [25].
Therefore, in
this re
sea
r
ch,
we focu
s on
wea
k
mig
r
ati
on app
roa
c
h.
Agents
th
at impleme
n
ts AFME
co
nsi
s
ts of
four
comp
one
nts:
perce
ptor,
actuato
r
,
module,
and
se
rvice. Pe
rcepto
r
all
o
ws agent
to se
nse and gen
erate belief. Actuator allo
ws
agent to pe
rform a
c
tion to
its environ
m
ent ac
co
rdin
g to gene
rat
ed belief. Mo
dule represe
n
ts
informatio
n space sha
r
ed
betwe
en
perceptor
and
a
c
tuator. An
ag
e
n
t may
contai
n pe
rcepto
r
a
nd
actuato
r
. Th
e
s
e t
w
o
comp
onent
s
coul
d
not
refe
r e
a
c
h
othe
r di
re
ctly. The
r
efore, mod
u
le i
s
a
spa
c
e tha
n
can be u
s
ed to
sha
r
e inform
ation betwe
e
n
perce
ptor a
nd actu
ator.
Finally, servi
c
e is
informatio
n space sh
are
d
among a
gent
s.
5. Conclu
sion
This
pap
er
d
e
scrib
e
s mult
i-age
nt sy
ste
m
in
WSN i
n
ord
e
r to
ma
nage li
mited
resou
r
ce,
esp
e
ci
ally energy and ba
ndwi
d
th, whi
c
h can su
pp
ort data pro
c
essing in di
stributed man
ner.
Capability of
current
sensor no
de i
n
computing
allows
us to
implem
ent i
n
-net
work data
pro
c
e
ssi
ng
o
n
sen
s
o
r
n
o
d
e
s
usi
n
g
age
nt ori
ented
p
a
radi
gm.
Wh
en im
pleme
n
t
ing multi-ag
ent
system in WSN in orde
r to supp
ort in-netwo
rk
d
a
ta
processin
g
, there are sev
e
ral a
s
pe
cts
that
sho
u
ld
be
co
nsid
er:
archit
ecture
co
nsid
eration,
middl
ewa
r
e, m
obil
e
ag
ent itine
r
ary pl
annin
g
,
and
learni
ng ag
en
t capability.
Implementin
g
distribute
d
WSN a
r
chite
c
ture in
which netwo
rk i
s
segm
ented in
to cluste
rs
coo
r
din
a
ted
b
y
clust
e
r
hea
d is
our prop
ose
d
solu
tion
. In this n
e
twork,
we
can i
m
pleme
n
t ag
ent
orga
nization i
n
whi
c
h ea
ch sen
s
o
r
no
de is de
ploy
ed by agent
s that perform
a spe
c
ific ta
sk.
These a
gent
s are
co
ordi
na
ted by a ma
n
ager age
nt.
In the context of mobile
ag
ent, we h
a
ve
to
con
s
id
er itine
r
ary pl
anni
ng
that en
sures the mobil
e
a
gent visits all
sen
s
o
r
n
ode
s in ea
ch
clu
s
t
e
r
in efficient m
anne
r. Ge
net
ic Algo
rithm
is ou
r ma
i
n
con
s
id
eratio
n
.
The la
st consi
deration
is
cap
ability of
agent
s to le
a
r
n
whe
n
they
determi
ne th
eir o
w
n
de
cision auto
nom
ously. Th
ere
are
many machi
n
e learni
ng techni
que
s that can be
use
d
to impleme
n
t learnin
g
capability on an
agent, but rei
n
forceme
n
t le
arnin
g
is
a te
chni
que th
at is sufficient to
be imple
m
en
ted on a
se
nsor
node. Belief De
sire Intenti
on model a
n
d
Markov
De
cisi
on Pro
c
e
s
s are a
pproa
che
s
that can
be
con
s
id
ere
d
when devel
opi
ng the intellig
ent agent de
p
l
oyed on a se
nso
r
nod
e
.
For future
work,
we are going to
dev
elop agent-ori
ented system
that will
be
impleme
n
ted
on Surama
d
u
Bridge. Th
e next rese
a
r
ch a
gen
da i
n
clu
d
e
s
impl
ementing m
o
bile
agent
cap
abil
i
ty to determi
ne efficie
n
t itinera
r
y pla
nni
ng both
usi
n
g
GA and
RL i
n
whi
c
h B
D
I and
MDP agents
tak
e
into acc
o
unt.
Referen
ces
[1]
You L
i
n
Xu, Y
ong
Xia. Struc
t
ural H
ealt
h
M
onitor
i
ng
of L
o
ng Sp
an
Susp
ensi
on Bri
d
g
e
. Ne
w
Y
o
rk:
Spon Press. 2
012.
[2]
Richar
d T
y
na
n
,
David M
a
rsh
,
Dona
l O’ka
n
e
. T
he Multi-a
gent S
y
stem
Architecture f
o
r W
i
reless
Sensor N
e
t
w
or
ks. Springer-V
erla
g Berli
n
He
idel
ber
g. 200
5: 687-6
94.
[3]
Min
Che
n
, Ser
g
io
Gonz
alez,
and
Victor
C.
M. Leu
ng. A
p
p
l
icatio
n a
n
d
D
e
sign
Issues
for
Mobil
e
A
g
e
n
t
in Wireless S
e
nsor Net
w
o
r
ks.
IEEE Wireless Communic
a
tion
. 2007; 1
4
(6): 20-2
6
.
[4]
Holg
er Karl, A
ndre
a
s Willi
g.
Protocols An
d
Architectures F
o
r W
i
reless Se
nsor Net
w
o
r
ks. Chich
e
ster:
John W
i
l
e
y
an
d Sons. 20
05.
[5]
Micha
e
l W
ool
d
r
idg
e
. Introduct
i
on to Multi-
ag
ent S
y
stem. C
h
ich
e
ster: John
W
ile
y
a
nd So
n
s
. 2002.
[6]
M Chen. Mob
i
l
e
Agent-Bas
ed
W
i
reless Sens
or Net
w
ork.
Jo
urna
l Co
mp
utin
g
. 2006; 1(
1): 14-21.
[7]
Alcid
e
s Monto
y
a, Dia
na C
a
rol
i
na R
e
strepo, D
e
metrio
Artur
o
Ovalle. Artifici
a
l
Intelli
ge
nce fo
r W
i
reless
Sensor N
e
t
w
or
k Enha
nceme
n
t, Smart W
i
reless Sensor N
e
tw
o
r
ks.
In: Yen
Khen
g T
an.
Editors
.
Te
ch
.
201
0: 73-8
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 3, September 20
15 : 1079 – 10
88
1088
[8]
Mar
w
a Sh
ara
w
i, Imane Al
y Sa
roit, Hesham E
l
-Mahd
y, Ei
d E
m
ar
y
.
R
outin
g
W
i
reless Se
ns
or Net
w
orks
Ba
se
d On
So
ft C
o
mpu
t
i
n
g Pa
ra
di
g
m
s: Su
rvey
.
Internati
o
nal
Jour
na
l o
n
Soft Co
mputi
ng, Artifici
a
l
Intelli
genc
e an
d Appl
icatio
ns
(IJSCAI)
. 2013
; 2(4): 21-36.
[9]
Gerardo I Simari, Simon D Parsons. Mark
ov De
cisi
on P
r
ocesses a
nd the Beli
ef-Desi
r
e-Intentio
n
Mode
l. Ne
w
Yo
rk: Springer Bri
e
f in Comp
uter
Science. 20
11
.
[10]
Maria Guij
arro,
Rube
n F
uente
s
-F
ernan
dez, Gonzal
o Paj
a
res. A Multi-ag
ent S
y
stem Ar
chitecture fo
r
Sensor N
e
t
w
or
ks-Model
li
ng, Contro
l, Progr
amming,
Simu
l
a
tions, an
d Ap
plicati
ons. In: T
e
ch. 2011:
23-4
0
.
[11]
Qingq
ua
n Su
n
,
W
e
ihon
g Yu,
Nikol
ai K
o
ch
urov
, Qi Ha
o, F
e
i Hu. A M
u
lti-Ag
ent-Bas
ed Intel
l
i
gen
t
Sensor
an
d Ac
tuator Net
w
o
r
k
Desi
gn fo
r
Smart Hous
e a
nd
Home A
u
tomat
i
on.
Jo
urn
a
l of
Sensor an
d
Actuator Netw
orks
. 2013; 2: 557-
588.
[12]
Khin
Ha
ym
ar
SH, You
n
g
Sik
Ch
oi, Jo
ng
S
ou P
a
rk. T
he
Multi A
gent
S
ystem Sol
u
tio
n
s
for W
i
rel
e
ss
Sensor N
e
t
w
or
k Applic
ations.
Sprin
ger-Ver
la
g Berli
n
Hei
del
berg. 20
08: 45
4-46
3.
[13]
Y Xu, H Qi. M
obil
e
A
gent M
i
gratio
n Mo
del
li
ng A
nd D
e
si
g
n
for T
a
rget T
r
ackin
g
i
n
W
i
rel
e
ss Se
nsor
Net
w
ork.
Ad Hoc Network.
2007;
6(1): 1-16.
[14]
H
Qi,
Y Xu, X W
ang.
Mob
i
l
e
Agent-Bas
ed
C
o
lla
bor
ative S
i
gna
l a
nd Infor
m
ati
on Pr
oces
sing In
Sens
or
Netw
ork
. IEEE
Proceeding. 2003;
91(8): 1172-1183.
[15]
Ahmad S
a
rd
o
u
k, Ran
a
R
ahi
m-Amoud, L
e
il
a Ma
rg
hem-Bo
u
la
hia, D
o
min
i
que Ga
iti. Age
n
t Starte
g
y
Data Gatheri
n
g
f
or Long L
i
fe
W
S
N.
Internationa
l Journ
a
l of
Comput
er N
e
tw
orks and Co
mmu
n
icati
o
n
s
(IJCNC)
. 201
0; 2(5): 71-90.
[16]
Kuna
l Sha
h
, Mario Di F
r
a
n
s
esco, Moha
n
Kuma
r. Distri
buted R
e
so
urc
e
Mana
gem
en
t in W
i
reless
Sensor
Net
w
or
k Usin
g Re
info
rcement L
earn
i
ng.
Spri
ng
er S
c
ienc
e Busi
ne
ss Medi
a
. 20
1
2
; 19: 7
05-
724.
[17]
M Ch
en. Mo
bil
e
Ag
ent-Bas
ed
Direct
ed
Diffu
sion
in
W
i
rel
e
s
s
Sens
or N
e
t
w
ork.
EURASIP
Jour
nal
of
Advanc
es in Si
gna
l Processi
n
g
. 2007.
[18]
Q W
u
. On Computin
g Mob
i
l
e
Age
n
t Ro
ute
for Data F
u
si
on i
n
Distri
but
ed Se
nsor N
e
t
w
o
r
k.
IEEE
T
r
ansactio
n
on
Know
ledg
e an
d Data Eng
i
n
e
e
rin
g
. 200
4; 16
(6): 740-7
53.
[19]
Char
alam
pos
K, Aristides M
p
itzio
pou
los, D
a
mia
nos G, Grammati Pa
ntzi
ou. Effective D
e
termin
a
ti
o
n
of Mobi
le A
g
ent Itinerar
ies
for Data
A
g
greg
ation
on
Sensor
Net
w
o
r
ks.
IEEE Transactions on
Kn
o
w
l
e
dg
e
and
D
a
ta
En
gi
nee
ri
ng
. 201
0; 22
(12): 167
9-1
6
9
3
.
[20]
X W
ang, M Ch
en, T
K
w
on, H
C
Chao
.
Multi
p
le Mobi
le Ag
en
ts Itinerar
y
Pla
nni
ng in W
i
rel
e
ss Sensor
Net
w
ork: Surv
e
y
and Eva
l
u
a
tion.
IET
Communic
a
tions
. 20
10
.
[21]
Paul
o Gil, Am
â
n
cio
Santos,
Al
berto
Card
oso.
De
ali
ng
w
i
th
Outliers i
n
W
i
r
e
less
Sens
or
Net
w
orks: A
n
Oil Re
fi
n
e
r
y
A
pplic
atio
n
.
IEEE Transactions on Cont
rol System
s
Technology
. 20
13;
22(4): 1
589-
159
6.
[22]
Rag
have
ndr
a
V Ku
lkarn
i
,
Ann
a
F
ö
rst
e
r, Ganes
h
Kumar V
ena
yagam
oorth
y. Comp
utation
a
l
Intelli
genc
e i
n
W
i
reless
Sen
s
or Net
w
o
r
ks: A Surve
y
.
IE
EE Co
mmu
n
ic
ations
Surveys
& T
u
toria
l
s
.
201
1; 13(1): 68
-96.
[23]
Kuna
l Sha
h
,
Mario D
i
F
r
an
sesco, Moh
a
n
Kuma
r. Distri
buted
Reso
urc
e
Man
agem
en
t In W
i
reless
Sensor
Net
w
or
k Usin
g Re
info
rcement L
ear
ni
ng.
W
i
rel
e
ss N
e
tw
ork Spring
e
r
Scienc
e Bus
i
ness Me
dia
.
201
2; 19: 705-
724.
[24]
F Aiello, FL Be
llifemi
ne, G Fortino, R Gravin
a, S Galzaran
o. An Agent-B
ased Si
gn
al Pr
ocessi
ng In-
Nod
e
Env
i
ron
m
ent for R
eal
-T
ime Human
Activit
y
M
onit
o
rin
g
Bas
ed
On W
i
reless
Bod
y
S
ens
o
r
Net
w
orks.
Eng
i
neer
ing A
ppl
ic
ations of Arti
fi
ci
al Intell
ige
n
ce
Scienc
e Direct
.
2011; 2
4
: 114
7-11
61.
[25]
Ramo
n Lo
pes,
F
l
avio Assis,
Carlos M
ontez.
MASPOT: A
Mobile Agent
System
for S
u
n SPOT
. 10th
Internatio
na
l Symp
osi
u
m on
Autonom
ous D
e
centra
lize
d
Systems (ISADS
)
. 2011. 25-
31.
Evaluation Warning : The document was created with Spire.PDF for Python.