Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
13
,
No.
3
,
Ma
rch
201
9
, p
p.
1065
~
1072
IS
S
N: 25
0
2
-
4752, DO
I: 10
.11
591/ijeecs
.v1
3
.i
3
.pp
1065
-
1072
1065
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Compari
ng t
he
l
inear
and
logarith
m n
or
m
alized
ex
tre
m
e
learnin
g machin
e in
flow
curve m
odeling
of
m
ag
n
etorheologi
ca
l
f
lui
d
Irfan Ba
hiud
din
1
, Abdul
Y Abd
Fat
ah
2
, S
aifu
l
A M
az
lan
3
, Mohd I
Shapiai
4
,
Fitria
n
Imadud
din
5
,
Ubaidi
ll
ah
6
, D
ew
i Utami
7
, M
oh
d
N M
uhta
z
arud
din
8
1
,3,4,7
Malay
s
ia
-
Ja
pan
Int
ern
a
ti
ona
l
Instit
ut
e
of
T
echnolog
y
,
Univ
er
siti
T
eknol
og
i
M
al
a
y
si
a, Kua
la Lu
m
pur,
Malay
si
a
1
Depa
rtment of
Mec
hanica
l
Eng
i
nee
ring
,
Voc
at
io
nal
Co
ll
eg
e, Uni
ver
sita
s Gad
ja
h
Mada
,
Yog
y
a
kar
ta
,
Indone
si
a
2,8
Raz
ak
Facu
lty of
T
ec
hn
o
log
y
a
nd
Inform
at
ic
s
,
Univer
siti
Te
kno
logi
Ma
lay
sia
,
K
ual
a
Lumpur,
M
al
a
y
si
a
5,6
Mec
hanica
l
E
ngine
er
ing
Dep
a
rtment,
Fa
cult
y
of
Engi
n
ee
rin
g,
Univer
sita
s Seb
e
la
s Maret,
Surak
art
a
,
Indon
esia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Sep
1
5
, 201
8
Re
v
ise
d
N
ov 3
0,
2018
Accepte
d
Dec
2
2
, 201
8
The
ext
r
eme
learni
ng
m
ac
hine
(EL
M)
play
s
a
n
important
rol
e
to
pre
dict
m
agne
torhe
olog
i
ca
l
(MR)
flui
d
beha
vior
and
to
red
uc
e
the
co
m
puta
ti
onal
flui
d
d
y
nami
cs
(
CF
D)
ca
lc
u
la
t
io
n
cost
whil
e
sim
ula
ti
ng
th
e
MR
f
lui
d
flow
of
an
MR
a
ct
u
at
or
.
Th
is
pap
er
pr
ese
nts
a
log
ari
t
hm
norm
al
iz
ed
m
et
hod
to
enha
nc
e
the
pre
dic
ti
on
of
EL
M
of
the
flow
cur
ve
rep
rese
nt
ing
th
e
MR
flui
d
rhe
ological
prop
ert
i
es.
MRC
C1
L
was
used
to
t
est
the
per
form
a
nce
of
the
proposed
m
et
hod,
and
d
iffe
r
ent
ac
t
iva
t
ion
func
tions
of
EL
Ms
were
chose
n
to
be
the
n
eur
a
l
n
et
works
sett
ing
.
The
Norm
al
i
zed
Root
Mea
n
S
quar
e
Er
ror
(NRM
SE)
was
sele
c
te
d
as
th
e
indi
c
at
or
of
th
e
EL
M
pr
edi
c
ti
o
n
accurac
y
.
NRM
SE
of
the
proposed
m
et
hod
is
found
to
im
prove
the
m
odel
ac
cur
acy
up
to
77.
10
%
for
t
he
pre
di
ction
or
te
sting
ca
se
whi
l
e
compari
ng
wi
t
h
the
li
ne
ar
norm
al
iz
ed
E
L
M.
Ke
yw
or
d
s
:
Extrem
e
l
earn
ing
m
achine
Ma
gn
et
orheo
l
ogic
al
f
lui
d
Neural
n
et
w
orks
Norm
al
iz
ed
m
et
hod
Sh
ea
r
s
tres
s
Copyright
©
201
9
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Abd
ul Y A
bd
Fata
h,
Ra
zak F
ac
ulty
of Tech
nolo
gy
and in
form
at
ics,
Un
i
ver
sit
i Te
knol
og
i M
al
ay
sia
,
Jal
an
S
ultan
Y
ahya Petra
, 54100
K
uala L
um
pu
r,
W
il
ay
ah
Pe
rse
ku
t
uan Kual
a
Lum
pu
r, M
al
ay
sia
.
Em
a
il
:
ya
sser.
kl@
utm
.
m
y
1.
INTROD
U
CTION
A
sm
art
m
at
erial
cal
le
d
m
agn
et
orhe
ologica
l
(MR)
fl
uid
i
s
know
n
for
it
s
capa
bili
ty
to
cha
ng
e
the
viscosity
as
th
e
eff
ect
of
m
a
gn
et
ic
fiel
ds
e
xposure
.
The
changes
a
re
re
ver
si
ble
that
ha
pp
e
n
l
ess
tha
n
10
m
illi
secon
ds.
The
hi
gh
ly
quic
k
respo
ns
ive
natur
e
of
the
m
at
erial
s
has
at
tract
ed
the
interest
of
va
rio
us
researc
hers
to
app
ly
at
var
i
ous
dev
ic
es
s
uc
h
as
dam
per
s
[1]
-
[
3]
,
brakes
[4]
,
cl
utches
[5]
a
nd
oth
e
rs
[
6]
-
[
8]
.
I
n
each
a
pp
li
cat
io
n,
t
he
beh
a
v
io
r
of
the
MR
flu
id
nee
ds
t
o
be
consi
der
e
d,
es
pecial
ly
in
desi
gn
sta
ges
.
T
he
us
ua
l
i
m
po
rtant
pa
ra
m
et
ers
or
var
ia
bles
of
the
MR
fluid
rh
e
ol
og
i
cal
pr
operti
es
f
or
de
sig
n
and
si
m
ulati
on
are
sh
ear
stress,
sh
ea
r
rate,
visco
sit
y
and
yi
el
d
stress
[
9],
[10]
.
The
se
va
riabl
es
can
be
ob
ta
ine
d
f
ro
m
the
char
act
e
rizat
ion
process
us
i
ng
a
rheom
et
er.
O
ne
of
t
he
c
om
m
on
m
easur
e
m
ent
resu
lt
s
f
ro
m
the
r
heo
m
et
er
is
cal
le
d
flo
w
cu
r
ve.
Fl
ow
c
urve
is a d
at
a set co
ntainin
g
s
hear
stress as a fu
nc
ti
on
of s
hear
ra
te
. Th
is d
at
a set can
be
de
rive
d
to
obta
in
yi
el
d
stre
ss,
plasti
c
visc
os
it
y
and
al
s
o
to
ide
ntify
wh
et
her
the
flui
d
is
New
t
on
ia
n
or
non
-
New
t
on
ia
n,
ha
ving s
hear t
hic
ken
i
ng or t
hinn
ing
e
ff
ect
s
, a
nd
oth
e
r rel
eva
nt
char
act
e
risti
cs
[11],
[12]
.
Rheol
og
ic
al
m
od
el
s a
re th
e usual m
e
tho
ds t
o
predict
and
duplica
te
the MR fluid
be
ha
vi
or
in
te
rm
s
of
the
flo
w
cu
r
ve
data.
Rhe
olog
ic
al
m
od
el
s
include
Bi
ng
ha
m
Plast
ic
,
Papan
ast
asi
ou,
He
rsch
el
B
ulk
l
ey
,
and
bi
-
visco
us
m
od
el
[13]
-
[
15]
.
Bi
ngham
Plas
ti
c
is
the
m
os
t
widely
kn
ow
n
m
et
ho
d
to
ob
ta
in
yi
el
d
stress
an
d
plasti
c
viscosity
f
ro
m
a
flo
w
c
urve
be
cause
of
it
s
si
m
pl
ic
it
y
[1
6],
[
17
]
.
A
no
t
her
m
od
el
is
Her
sc
hel
Bulkley
th
at
tr
ie
d
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
13
, N
o.
3
,
Ma
rc
h 201
9
:
1065
–
1072
1066
to
overc
om
e
the
disad
va
ntage
of
Bi
ng
ham
plasti
c
m
od
el
,
wh
ic
h
is
the
hig
h
acc
uracy
for
a
na
rrow
sh
e
ar
rate
range
[
18]
,
[
19]
.
In
gen
e
ral,
a
ll
m
od
el
s
can
on
ly
be
a
ppli
cable
to
one
value
of
the
m
agnet
ic
fiel
ds
[20]
,
[21]
.
If
one
wan
ts
to
co
ver
ot
her
m
agn
et
ic
fiel
ds
,
a
poly
no
m
ia
l
m
od
el
of
yi
el
d
stress
needs
to
be
a
pp
li
ed
[
22]
,
[
23
]
.
The
li
m
it
at
ion
can
be
s
olv
e
d
us
in
g
m
achine
le
arn
in
g,
su
c
h
as
Ar
ti
fici
al
Neural
Net
wor
ks
(
A
NN),
Suppo
rt
Vecto
r
Ma
chi
ne
(SVM)
[
24]
and
Ext
rem
e
L
earn
i
ng
Ma
c
hi
ne
(E
LM)
[25]
.
A
NN
a
nd
S
V
M
hav
e
bee
n
know
n
for
their
li
m
i
tation
s,
w
hich
ar
e
slow
e
r
trai
ni
ng
ti
m
e
and
ha
ve
m
or
e
possi
bili
ty
to
be
trapp
e
d
at
local
m
ini
m
a
or
e
xtrem
a
[26
]
.
Extrem
e
Learn
i
n
g
Ma
c
hin
e
(ELM)
is
known
f
or
it
s
so
l
ution
i
n
te
rm
s
of
the
qu
ic
ker
t
ra
ining,
m
or
e
accuracy
value
a
nd
hi
ghest
possi
bili
t
y
to
gain
acce
pted
gen
e
rali
zat
ion
tha
n
the
cl
assic
m
et
ho
ds.
Th
e
existi
ng
m
et
ho
d
to
pre
dict
MR
fluid
be
ha
vior
us
in
g
E
LM
has
bee
n
pro
po
se
d
i
n
[
20
]
t
o
pr
e
dict
va
rio
us
rh
e
ologica
l
pa
ram
et
ers.
Desp
it
e
the
high
accuracy
at
som
e
values
and
ran
ges,
the
predict
io
n
nee
ds
to
be
i
m
pr
oved
.
The
po
s
sible
im
pr
ov
em
ent
is
the
act
ivati
on
f
unc
ti
on
,
norm
al
iz
a
ti
on
m
et
ho
d
an
d
the
pr
e
-
proc
e
ssing
of
the
in
pu
ts
.
Me
anw
hile,
the
act
ivati
on
functi
on
eff
ect
s
wer
e
inv
e
sti
gated
in
[
20
]
,
[25]
.
H
owev
er,
the
norm
al
iz
a
ti
on
ef
fect
at
th
e
ELM
-
base
d
r
heo
l
og
ic
al
m
od
el
is
only
discu
ssed
separ
at
el
y
a
nd
not
com
pr
ehe
ns
ive
ly
.
Wh
il
e
li
near
norm
al
iz
at
i
on
has
been
ut
il
iz
ed
in
[25],
[27],
[
28
]
,
l
ogarit
hm
no
rm
al
i
zat
ion
has
bee
n
int
rodu
ce
d
i
n
[20]
.
Since
these
previ
ou
s
w
ork
s
f
ocu
s
on
ly
on
the
intr
oduc
ti
on
of
the
ex
trem
e
le
arn
in
g
m
achine
p
la
tfo
rm
f
or
var
i
ous im
po
rtant p
a
ram
et
ers
pr
e
dicti
on
, th
e
log
a
rithm
n
orm
al
iz
ation
in
fluen
c
e
on the m
od
el
pe
rfor
m
ance h
as
not b
ee
n disc
usse
d
i
n detai
l.
Ther
e
f
or
e,
this
pa
per
present
s
a
com
par
at
ive
stu
dy
betw
een
the
li
near
an
d
lo
gar
it
hm
no
rm
alized
extrem
e
le
a
rn
ing
m
achine
in
the
flo
w
curve
m
od
el
ing
of
the
m
agn
et
orhe
ologica
l
fluid
.
The
norm
al
iz
ation
is
app
li
ed
t
o
the
inputs
of
the
e
xtrem
e
le
arn
in
g
m
achine
(E
LM).
T
he
m
od
el
outp
ut
is
sh
ear
st
ress
w
hi
le
the
inputs
are
s
he
ar
rate
an
d
m
agn
et
ic
fiel
d.
The
ge
ne
ral
ste
ps
of
the
m
od
el
ing
m
et
ho
d,
the
norm
al
izati
on
m
et
ho
d,
t
he
da
ta
colle
ct
ion
a
nd
t
he
sim
ulatio
n
set
up
are
de
scribe
d
in
sec
ti
on
2.
T
he
res
ults
on
the
acc
ur
acy
and the c
om
puta
ti
on
al
burde
n are t
hen prese
nted
a
nd
discu
ssed
i
n
sect
io
n 3.
2.
RESEA
R
CH MET
HO
D
2.1.
Model
i
ng
met
ho
d
The
usual
r
heol
og
ic
al
m
od
el
s
ha
ve
a
s
hear
r
at
e
as
input
an
d
s
hear
st
ress
as
outp
ut.
Me
a
nwhile
,
th
e
ELM
based
-
r
he
ologica
l
m
odel
con
sist
s
of
t
he
sam
e
inp
ut
and
ou
t
pu
t
wit
h
the
ad
diti
onal
inp
ut
of
m
a
gn
et
ic
fiel
ds
t
o
pr
e
dic
t
the
s
hear
stre
ss
at
var
i
o
us
m
agn
et
ic
fiel
ds.
T
he
ELM
ba
sed
-
r
heo
l
og
ic
al
m
od
el
is
cal
culat
ed
us
in
g
the
str
uc
ture
of
Sin
gle
-
hidden
Lay
e
r
Feed
forw
a
r
d
Neural
Netw
or
ks
(SLF
Ns).
I
n
ge
ner
al
,
t
he
ou
t
put
m
od
el
o
r
t
he pr
edict
ed
s
hea
r
s
tress fo
r
eac
h
s
a
m
ple (
i
)
set
ca
n be e
xpresse
d as i
n
E
qua
tion
(1) or
(2).
∑
(
,
,
)
=
1
=
,
=
1
,
…
,
(
1
)
=
(
2
)
Ou
t
pu
t
(
)
is
a
f
un
ct
i
on
of
wei
gh
ti
ng
outp
uts
(
)
an
d
hi
dd
e
n
no
de
outp
ut
(
(
,
,
)
).
(
,
,
)
can
al
s
o
be
ca
ll
ed
Acti
vatio
n
functi
on
as
a
functi
on
of
bia
s
(
)
on
-
th
hidden
no
de
w
he
r
e
=
1
,
…
,
,
a
nd
-
th
in
puts
(
)
wh
e
re
=
1
,
…
,
, w
ei
gh
ti
ng
in
pu
t (
w
j
)
on
eac
h
i
th
in
put.
Me
a
nwhile
,
the h
id
de
n
node
outp
uts
f
or
c
om
plete
trai
nin
g
dataset
s
can
be
de
note
d
a
s
H
.
T
he
de
ta
il
m
at
rix
of
H
is
e
xpresse
d
by
Eq
uation
(
3).
The
predict
e
d
sh
ear
stre
ss
m
at
rix
for
al
l
sa
m
ples
is
denot
ed
a
s
that
ca
n
be
ex
presse
d
i
n
Eq
uation
(
4).
=
[
(
1
,
1
,
1
)
…
(
,
,
1
)
⋮
⋯
⋮
(
1
,
1
,
)
…
(
,
,
)
]
×
(
3
)
=
[
1
⋮
]
×
1
(
4
)
The
act
ivati
on
f
unct
io
n
ca
n
util
iz
e
any
co
ntinuo
us
piecewi
se
f
un
ct
io
ns.
T
he
e
xam
ples
of
the
e
xisti
ng
con
ti
nu
ous fu
nc
ti
on
s a
re
:
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Compari
ng the
li
nea
r
an
d
lo
garit
hm
no
r
m
alized extre
me
le
ar
ni
ng ma
c
hine
in
fl
ow
....
(
Ir
fan
Ba
hiud
din
)
1067
a.
Hard Lim
it
(
,
,
)
=
{
1
,
.
−
≥
0
0
,
.
−
<
0
(
5
)
b.
Sigm
oid
(
,
,
)
=
1
(
1
+
(
−
(
.
+
)
)
)
⁄
(
6
)
c.
Sine
(
,
,
)
=
(
.
+
)
(
7
)
d.
Trian
gu
la
r basi
s fun
ct
io
n
(
,
,
)
=
−
1
≤
≤
1
;
=
0
,
ot
he
rwi
se
(
8
)
e.
Ra
dial B
asi
s F
un
ct
io
n
(
,
,
)
=
(
−
2
)
(
9
)
The
er
r
or
form
ulati
on
betwee
n
the
e
xp
e
rim
e
ntal
and
sim
ul
at
ion
res
ults
is
form
ulate
d
us
ing
Eq
uati
on
(10) w
he
re T m
at
rix
is t
he
m
easur
e
d shea
r
s
tress
of
al
l sam
ples as
desc
rib
ed by E
q
uatio
n
(
11)
=
|
|
−
|
|
(
10
)
=
[
1
⋮
]
×
(
11
)
If
e
is
near
ze
r
o, we o
btain
=
.
Re
placi
ng
in E
q
ua
ti
on
(2) wit
h
yi
el
ds
Eq
uatio
n
(12).
=
(
12
)
ELM
al
gorith
m
has
been
pro
ven
to
m
ini
m
ize
e
a
nd
|
|
β
|
|
[26]
.
The
SLF
Ns
m
od
el
is
trai
ne
d
us
in
g
t
he
ELM
al
gorith
m
. I
n
ge
ner
al
,
t
he
ELM
al
gori
thm
can
be des
cribe
d wit
hin
t
hr
ee
steps
, whi
ch
a
re
a.
assigni
ng the
and
rand
om
l
y usin
g
a
ny c
on
ti
nuous acti
vatio
n functi
on,
b.
cal
culat
ing
by
so
lvi
ng
E
q.
(8)
w
he
re
†
is
defi
ned
as
the
M
oore
–
Pe
nrose
ge
ner
al
iz
ed
in
ve
rse
[
34]
.
†
was
cal
c
ulate
d usin
g
Si
ngular
Value
Dec
om
po
sit
ion M
et
hod (S
VD).
=
†
(
13
)
2.2.
No
r
maliz
at
io
n metho
d
The
in
puts
usual
ly
need
t
o
be
no
rm
alized
before
in
pu
tt
ing
t
he
val
ue
of
t
he
Ne
ural
Netw
orks
.
Ther
e
f
or
e,
al
l
inputs
ca
n
be
inputt
ed
int
o
t
he
m
od
el
with
the
sam
e
scal
e,
bet
ween
0
and
1
or
-
1
a
nd
1
dep
e
ndin
g
on
the
desi
gner
de
ci
sion
.
The
norm
al
iz
ation
c
an
be
a
pp
li
ed
in
the
i
nput
or
outp
ut
va
riabl
es
as
sh
ow
n
in
F
igure
1
.
The
si
m
plest
fo
r
m
of
norm
al
i
zat
ion
m
et
ho
d
is
li
near
as
expresse
d
in
t
he
f
ollow
i
ng
equ
at
io
ns,
=
(
−
)
(
−
)
⁄
(
14
)
wh
e
re
x
n
orm
,
x
e
,
x
max
,
x
min
are
the
no
rm
a
li
zed,
ex
per
im
ental
data,
m
axim
u
m
,
and
m
ini
m
u
m
inputs,
resp
ect
ively
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
13
, N
o.
3
,
Ma
rc
h 201
9
:
1065
–
1072
1068
F
igure
1
.
SLFN
of
t
he
E
LM
-
base
d
r
he
ologica
l
m
od
el
Me
anwhil
e,
th
e
norm
al
iz
ed
l
og
a
rithm
is
exp
ress
ed
i
n
Eq
ua
ti
on
(
15).
Althou
gh
norm
al
i
zed
lo
gar
it
hm
in
the
w
ork
of
ANN
f
or
ai
rc
r
aft
cabin
has
be
en
pro
ve
n
to
ob
ta
in
bette
r
pe
rfor
m
ance
co
m
par
ed
to
the
li
near
on
e
[
29
]
,
t
he
a
pp
l
ic
at
io
n
in
MR
fluid
s
houl
d
be
f
ur
the
r
i
nv
e
sti
gated
t
o
gain
m
or
e
c
ompre
he
ns
ive
a
na
ly
sis
at
var
i
ou
s
act
ivat
ion
f
unct
ion
s
.
In
this
w
ork
,
the
norm
al
iz
ati
on
is
ap
plied
in
the
inputs,
e
it
her
one
of
or
bot
h
sh
ear
r
at
e a
nd
m
agn
et
ic
f
ie
lds
.
1
,
=
(
(
)
−
(
)
)
(
(
)
−
(
)
)
⁄
(
15
)
2.3.
Data C
ollec
tio
n
The
fl
ow
cu
rve
is
ob
ta
ine
d
ba
sed
on
t
he
r
ot
at
ion
al
te
st
on
a
par
al
le
l
plate
rh
e
om
et
er
(Anto
n
Paa
r)
as
repor
te
d
in
[
25]
.
T
he
te
ste
d
MR
flui
d
is
m
anu
fact
ur
e
d
by
CK
,
S
outh
K
or
ea
cal
le
d
MR
C
C1L
w
it
h
the
pro
per
ti
es
as
de
scribe
d
in
[
25]
.
The
app
li
ed
m
agn
et
ic
fiel
ds
are
830
m
T,
770
m
T,
42
0
m
T,
31
0
m
T,
2
00
m
T,
and
0
m
T.
The
obser
vatio
n
is
carried
out
fro
m
0.
01
t
o
10
00
s
-
1
.
T
he
s
hear
rate
is
ta
ke
n
ba
sed
on
the
set
ti
ng
of
a
log
arit
hm
ic
ra
m
p
in
Rheoc
om
pass,
the
gr
aph
ic
al
us
e
r
in
te
rf
ace
of
the
par
al
le
l
plate
r
heo
m
et
er
by
An
t
on
Paar.
Th
e
obtai
ned d
at
a a
re sh
own
i
n
Figure
2
that
c
on
ta
in
flo
w
c
urve
data
ab
out
539
m
easurem
ent
po
i
nts.
A
num
ber
of
data
with
t
he
sam
e
app
li
ed
m
agn
et
ic
fiel
d
are
cal
le
d
da
ta
-
set
.
Th
us
,
there
are
7
dataset
s.
T
he
data
is
div
ide
d
int
o
trai
ning
data a
nd pre
dicti
on
.
Fo
r
MR
C C
1L
,
trainin
g data i
s u
se
d for
obta
i
ning the
m
od
el
. I
t i
s a
set o
f d
at
a o
n
m
agn
et
ic
fiel
ds
of
770
m
T,
420
m
T,
200
m
T,
a
nd
0
m
T.
In
a
dd
it
io
n,
in
t
hi
s
w
ork,
the
m
od
e
l
capab
il
it
y
to
predict
outsi
de
the
trai
ning
da
ta
range
was
e
valuated
.
T
he
sel
ect
ed
datase
ts
fo
r
predict
io
n
are
830
a
nd
310
m
T.
Figure
2
.
Obtai
ned fl
ow curve
on vari
ou
s
m
a
gn
et
ic
fiel
d
.
.
.
S
h
e
a
r
R
a
t
e
M
a
g
n
e
t
i
c
F
i
e
l
d
S
h
e
a
r
S
t
r
e
s
s
1
2
(
1
,
1
,
)
(
2
,
2
,
)
(
3
,
3
,
)
(
j
,
j
,
)
o
1
2
j
3
N
o
r
m
a
l
i
z
a
t
i
o
n
N
o
r
m
a
l
i
z
a
t
i
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Compari
ng the
li
nea
r
an
d
lo
garit
hm
no
r
m
alized extre
me
le
ar
ni
ng ma
c
hine
in
fl
ow
....
(
Ir
fan
Ba
hiud
din
)
1069
2.4.
Simul
at
i
on
Se
tu
p
The
sim
ulati
on
platf
or
m
is
Ma
tl
ab®
with
a
PC
ha
ving
a
process
or
of
64
bit
(
3.4GH
z).
T
hen,
the
fitness
or
accu
r
acy
of
the
trai
ni
ng
a
nd
pr
e
dicti
on
pe
rfor
m
ance
will
be
a
sse
ssed
base
d
on
t
he
root
m
ean
sq
ua
re
error
(RMS
E),
no
rm
alized
R
MSE
(N
RM
S
E)
an
d
coe
ff
ic
ie
nt
of
determ
inati
on
(
2
)
with
the
e
qu
at
io
ns
of
Eq
uation
(
16),
(17), a
nd (1
8)
,
resp
ect
ively
.
=
√
∑
(
−
)
2
(
16
)
=
(
−
)
⁄
×
100%
(
17
)
2
=
1
−
∑
(
−
)
2
∑
(
−
)
2
⁄
(
18
)
w
he
re,
τ
e
,
τ
p
,
τ
max
,
a
nd
τ
min
are
m
easure
d,
pr
e
dicte
d,
m
axi
m
u
m
m
e
asur
e
d,
a
nd
m
ini
m
u
m
sh
ear
stress
(k
P
a),
res
pecti
vely
.
The
c
omplet
e
si
m
ulati
on
set
up
is
sho
wn
i
n
T
able
1
.
The
lo
ga
rith
m
no
rm
al
iz
ati
on
is
carried
out
on
l
y for the
s
hear
rat
e.
T
able
1
.
T
he
P
aram
et
er S
et
up for
Sim
ulati
on
Para
m
eter
Valu
e
Hid
d
en
Nod
es
2
0
,
2
0
0
,
5
0
0
,
1
000, 2 5
0
0
,
5
00
0
,
7
00
0
,
1
0
000
Activ
atio
n
Fun
ctio
n
Hard Li
m
it
(H
L)
,
Sig
m
o
id
(Sig),
and
Sinu
so
id
(
Sin
),
T
r
ian
g
u
lar
b
asis
f
u
n
ctio
n
(
Tri
b
as
),
Rad
ial Basis
Fun
ctio
n
(
Rad
b
as
)
Inp
u
t weigh
t and
b
ias d
eter
m
in
atio
n
No
r
m
al
Distribu
tio
n
No
r
m
aliz
atio
n
m
et
h
o
d
Linear a
ll
,
an
d
log
f
o
r
sh
ear
rate
on
ly
3.
RESU
LT
S
A
ND D
I
SCUS
S
ION
3.1.
Effect
on
Acc
urac
y at v
ario
us a
c
tiv
ati
on f
unct
i
on
s
The
c
om
par
iso
n
is
ca
rr
i
e
d
ou
t
by
c
om
par
ing
the
ex
pe
rim
ent
al
resu
lt
s
with
the
pr
oduce
d
s
hear
rate
at
var
i
ou
s
act
i
vation
functi
ons
.
T
able
2
des
c
ribes
the
RM
SE
,
NRMSE
,
an
d
R
2
at
li
near
(lin)
a
nd
lo
ga
rithm
norm
al
iz
a
ti
on
(lo
g)
.
F
or
t
he
trai
ni
ng
case
,
th
e
best
acc
ur
ac
y
is
achieve
d
by
Hardlim
it
(ELM
HL
)
f
or
both
l
og
and
lin
case.
T
hen,
the
seco
nd
-
best
acc
ur
ac
y
is
Tribas
(EL
M
Tribas
)
a
nd
fo
ll
owe
d
by
R
adb
a
s
(E
LM
R
adb
a
s
),
Sinu
s
oi
d
(
EL
M
Sin),
a
nd
Si
gm
oid
(ELM
S
ig).
Both
Lin
a
nd
lo
g
has
t
he
sam
e
or
de
r
in
t
erm
s
of
acc
ur
a
cy
.
For
li
near
case,
t
his
res
ult
is
cons
ist
ent
with
the
resu
lt
f
r
om
the
pr
e
vious
pa
pe
r
[
25
]
.
Me
a
nw
hile,
the
bigge
st
gap
betwee
n
log
a
nd
li
near
is
achi
eved
by
ELM
Sin.
I
n
ot
her
w
ords
,
the
ELM
Sin
gain
s
the
hi
gh
est
im
pr
ov
e
m
ent
com
par
ing
othe
r
act
ivati
on
f
un
ct
io
ns.
I
f
th
e
i
m
pr
ovem
ent
is
cal
culat
ing
usi
ng
pe
rce
ntage
as
sho
w
n
in
Eq
uation
(
19)
w
her
e
RMSE
Li
n
is
R
MSE
at
Li
n
c
ase
an
d
RMSE
L
og
is
RM
SE
at
L
og,
the
highest
va
lue
achieve
d by H
ard
li
m
i
t fo
ll
ow
ed by Tri
bas,
Ra
db
as
, S
i
n,
a
nd
Sig
.
∆
%
=
(
−
)
⁄
×
100%
(
19
)
T
able
2
.
E
LM
Mod
el
Acc
ur
a
cy
f
or T
he
T
rai
ning
Data
Sch
e
m
es
Hid
d
en
No
d
es
Log
(
k
Pa)
Lin
Lin
-
lo
g
RMSE
(kPa)
NRMSE
(%)
R
2
RMSE
(kPa
)
NRMSE
(%)
R
2
RMSE
(kPa)
∆
%
Hard Li
m
it
1
0
0
0
0
0
.00
0
.00
1
0
.13
0
.18
1
.00
0
0
0
.13
1
0
0
.79
Sig
m
o
id
1
0
0
0
0
1
.19
1
.71
0
.99
5
9
1
.34
1
.91
0
.99
4
8
0
.14
1
0
.82
Sin
u
so
id
1
0
0
0
0
0
.38
0
.55
0
.99
9
6
0
.83
1
.18
0
.99
8
0
0
.44
5
3
.74
Tr
ib
as
1
0
0
0
0
0
.07
0
.10
1
.00
0
0
0
.32
0
.46
0
.99
97
0
.25
7
8
.38
Rad
b
as
1
0
0
0
0
0
.32
0
.45
0
.99
9
7
0
.74
1
.06
0
.99
8
4
0
.42
5
7
.18
Hard
li
m
i
t
and
Tri
bas
ca
n
ac
hieve
th
e
best
ac
cu
r
acy
beca
us
e
of
the
sim
plici
ty
and
strai
gh
t
forw
a
r
dn
e
ss
nat
ur
e
of
the
eq
uation.
Me
anwhil
e,
Si
nu
s
oi
d
an
d
Ra
db
a
s
ha
ve
al
m
os
t
the
sam
e
be
hav
i
or,
wh
ic
h
is
os
ci
ll
at
ion
.
T
he
os
c
il
la
ti
on
will
be
sh
ow
n
at
li
near
norm
al
iz
at
i
on
sc
hem
e
and
can
be
el
im
i
nate
d
wh
il
e
us
in
g
l
ogarit
hm
norm
a
li
zat
ion
.
F
or
e
xam
ple,
Fig
ur
e
3
sho
ws
the
vi
su
al
com
par
is
on
bet
ween
E
LM
S
in
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
13
, N
o.
3
,
Ma
rc
h 201
9
:
1065
–
1072
1070
Lin
an
d
E
LM
Sin
L
og.
Wh
il
e
the
osc
il
la
ti
on
nat
ur
e
occ
urs
at
li
n
schem
e,
the
log
sc
hem
e
can
el
im
inate
it
and
m
ake th
e tre
nd
to follo
w
t
he
e
xp
e
rim
ental
r
esults cl
os
el
y.
(a)
(b)
Figure
3
.
Th
e
s
i
m
ulati
on
r
es
ul
ts of ELM
Sin
for (a)
Linear
a
nd (b) L
og
a
rith
m
n
or
m
alizat
io
n
cases
In
te
rm
s
of
th
e
te
sti
ng
acc
uracy
,
w
hile
pr
edict
ing
unkn
own
data
set
s,
the
sigm
oid
has
the
best
accuracy
f
or
lo
g
an
d
li
n
te
st
as
sh
ow
n
in
Ta
ble
3
.
E
LM
Sig
is
al
so
can
be
con
si
der
e
d
a
s
the
m
os
t
gen
erali
zed
or
t
he
le
ast
overf
it
ti
ng
c
om
par
ed
t
o
ot
her
s
.
The
ov
e
rf
it
ti
ng
or
not
of
the
m
od
el
can
be
check
e
d
in
|R
MSEp
-
RM
SET|
wh
e
r
e
RM
SEp
is
RM
SE
for
the
pr
edict
ion
case
a
nd
RM
SE
T
is
RM
SE
for
trai
ning
case.
Als
o
base
d
on
thi
s
crit
eria,
the
m
os
t
ov
e
rfi
tt
ing
m
od
el
is
ELM
HL
f
or
l
og
case
a
nd
E
LM
Tribas
f
or
li
n.
Both
m
od
el
s
are
the
m
os
t
ov
er
f
it
ti
ng
for
lo
g
a
nd
li
n
ca
ses.
A
s
discusse
d
befor
e
,
the
sim
ple
cal
culat
ion
na
ture
m
ay
be
able
to
pro
du
ce
hi
gh
a
ccur
acy
for
tra
ining
case.
H
ow
eve
r,
the
sim
plici
ty
can
al
so
cause
the
m
odel
cannot
acc
urat
el
y
pr
e
dict
the
un
s
een
data.
T
his
natu
re
of
EL
M
HL
al
so
oc
cur
s
i
n
the
pre
vious
works
[
20
]
,
[25]
.
T
he
highes
t
i
m
pr
ovem
ent
values
afte
r
the
app
li
cat
io
n
of
t
he
lo
g
sc
hem
e
are
ac
hieve
d
by
ELM
Triba
s.
Alth
ough
it
ca
n
be
consi
der
e
d
a
s t
he
le
ast
ov
e
rf
it
ti
ng
m
od
el
s,
∆
E
% h
as
sho
wn the
b
est
value
whi
ch
is a
bout
77.
10 %.
Table
3
.
E
LM
Mod
el
Acc
ur
a
cy
f
or
s
hear st
r
ess
Pr
e
dicti
on
of
MR
C C
1L
Sch
e
m
e
Hid
d
en
No
d
es
Log
T
estin
g
Lin Testin
g
Lin
-
Log
RMSE
(kPa)
NRMSE
(%)
|
RMS
Ep
-
R
MSE
T
|
RMSE
(kPa)
NRMSE
(%)
|
RMS
Ep
-
RMSE
T
|
RMSE
(kPa)
∆
%
Hard Li
m
it
1
0
0
0
0
2
.32
3
.31
2
.32
2
.21
3
.16
2
.09
0
.85
3
8
.19
Sig
m
o
id
1
0
0
0
0
1
.13
1
.61
0
.07
1
.43
2
.04
0
.09
0
.91
6
3
.71
Sin
u
so
id
1
0
0
0
0
1
.64
2
.35
1
.26
1
.86
2
.65
1
.03
1
.01
5
4
.36
Tr
ib
as
1
0
0
0
0
1
.66
2
.38
1
.5
9
2
.53
3
.62
2
.21
1
.95
7
7
.10
Rad
b
as
1
0
0
0
0
1
.70
2
.43
1
.38
1
.81
2
.58
1
.06
0
.88
4
8
.73
3.2.
The ef
fect
on
t
he com
pu
tatio
na
l c
os
t
The
c
om
pu
ta
ti
on
al
c
os
t
ca
n
be
analy
zed f
r
om
the
point of v
ie
w
t
he
need for
the
hi
dd
e
n
node
num
ber
to
achie
ve
a
c
ertai
n
accu
rac
y.
The
higher
the
hidde
n
node
num
ber
m
e
ans
that
m
or
e
resou
rce
is
ne
eded
t
o
reserve
the
m
e
m
or
y
to
cal
c
ulate
†
us
in
g
S
VD.
I
n
oth
e
r
words,
t
he
hi
dden
no
de
num
ber
re
pr
ese
nts
the
com
pu
ta
ti
on
al
cost
beca
us
e
t
he
hi
gh
e
r
the
hidden
node
num
ber
,
the
higher
t
he
tim
e
need
ed
f
or
t
he
trai
ning
process
. T
he d
et
ai
le
d
of the e
ff
ic
ie
ncy
of ea
ch
m
od
el
for t
he
trainin
g proc
ess c
an
b
e
obse
rv
e
d
i
n
Figure
4
.
F
or
ELM
HL,
the
ELM
HL
Log
ca
n
reach
le
ss
RM
SE
at
le
ss
hid
den
node
nu
m
ber
com
par
ed
to
E
LM
HL
Lin.
T
he
erro
r
al
so
c
an
be
re
duced
to
alm
os
t
zero
at
10
00
0
hidde
n
node
nu
m
ber.
The
sam
e
ph
en
om
ena
al
so
occ
ur
for
the
ELM
Sin
an
d
Si
g.
T
he
lo
g
sch
em
e
can
re
duce
th
e
requirem
ent
of
th
e
hidden
no
de n
um
ber
to
ac
hie
ve
a ce
rtai
n
ac
cur
acy
or RM
S
E.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Compari
ng the
li
nea
r
an
d
lo
garit
hm
no
r
m
alized extre
me
le
ar
ni
ng ma
c
hine
in
fl
ow
....
(
Ir
fan
Ba
hiud
din
)
1071
(a)
(b)
Figure
4
.
Th
e
a
ccur
acy
at
va
riou
s
h
i
dden
no
de
num
ber
s of E
LM Si
n f
or
(a)
Linear a
nd
(b)
L
ogarit
hm
n
orm
al
iz
a
ti
on
cases
4.
CONCL
US
I
O
N
A
m
od
el
d
evelop
m
ent is carr
ie
d
out t
o
pr
e
di
ct
the sh
ear str
ess o
f
MR
f
lui
d
as a f
unct
ion
of
s
hear
r
at
e
and
m
agn
et
ic
fiel
ds
us
in
g
ELM
.
In
this
work,
a
log
a
rithm
no
rm
ali
zat
ion
m
et
ho
d
is
inv
es
ti
gated
an
d
com
par
ed
wit
h
the
li
near
norm
al
iz
ation
m
et
ho
d.
T
he
evaluati
ons
ar
e
carried
out
at
diff
ere
nt
act
ivati
on
functi
on
set
ti
ng
s,
wh
ic
h
are
Hard
lim
it
,
sin
us
oi
d,
sigm
oid,
tria
ngular
ba
sis,
and
rad
ia
l
basis
f
un
ct
io
ns.
The
eff
ect
of
the
app
li
cat
io
n
log
a
r
it
hm
ic
no
rm
alizat
ion
can
i
m
pro
ve
al
l
the
E
LMs
accuracy,
especial
ly
EL
M
Si
n
wh
ic
h
dec
reas
es
the
RM
SE
about
4
kP
a
.
Me
anwhil
e,
in
te
rm
s
of
pr
e
di
ct
ion
,
the
l
og
norm
al
iz
a
ti
on
schem
e
can
re
du
ce
t
he
accuracy
with
the
highest
at
RM
SE
of
1.9
kP
a
f
or
ELM
Tribas
.
The
lo
gar
it
hm
no
rm
al
iz
at
ion
can
al
so
reduc
e
the
com
pu
ta
ti
on
al
bu
rd
e
n
by
reducin
g
the
hidden
node
num
ber
to
achie
ve
a
certai
n
ac
cur
acy
.
The
norm
al
iz
a
ti
on
m
et
ho
ds
can
sti
ll
be
fu
rt
her
in
vestigat
e
d
by
ap
plyi
ng
the
norm
al
iz
a
tio
n
m
et
ho
d
to
ei
ther
bo
t
h
or
one
of
th
e
inputs,
w
hi
ch
are
m
agn
et
ic
fiel
d
and
s
he
ar
rate
and
ou
t
pu
t.
T
he
eval
ua
ti
on
f
or
each
reg
i
on
range
ca
n
al
s
o
be
ca
rr
ie
d
ou
t
to
acce
ss
the
m
od
el
capab
il
it
y
fo
r
each
lo
w
a
nd
hi
gh
re
gion
of
sh
ea
r
r
at
e
and
m
agn
et
ic
fiel
d.
So
m
e
scenario
of
the
te
sti
ng
cases
al
so
nee
ds
to
be
ca
rr
ie
d
out
to
chec
k
the
m
od
el
capab
il
it
y
to
pr
e
dict
unle
arn
dataset
s,
s
uch
as
inter
po
l
at
ion
case
a
nd
extra
pola
ti
on
case.
T
he
sam
e
m
et
ho
d
ca
n
al
so
be
app
li
ed
to ot
he
r
sam
ples to
c
he
ck
the
consist
ency o
f
t
he
ef
f
ect
o
f
the acti
va
ti
on
a
nd no
rm
al
i
zat
ion
m
et
h
od
s
.
ACKN
OWLE
DGME
NT
Au
t
hors
ac
knowle
dge
Un
i
versi
ti
Teknolo
gi
Ma
la
ysi
a
fo
r
t
he
f
undi
ng
suppo
rt
thr
ough
GUP
G
ra
nt
(Vote
No
:
15J
44)
a
nd P
DRU
G
r
ant
(Vote
N
o: 04E
02).
REFERE
NCE
S
[1]
M.
Cao
,
e
t
a
l.
,
“
Scal
abl
e
and
i
nver
ti
bl
e
PM
NN
m
odel
for
M
agne
toRh
eo
logic
al
f
lui
d
d
ampers,”
JV
C
/J
ournal
of
Vi
bration and C
ontrol
,
vo
l
/i
ss
ue:
14
(
5
)
,
pp.
731
–
751,
2008
.
[2]
I.
Bahi
uddin
,
e
t
al.
,
“
Magne
t
orhe
ological
va
lve
base
d
actu
at
or
for
improvem
ent
of
passively
cont
rol
led
turboc
har
g
er
s
y
s
te
m
,
”
AIP Conf
e
renc
e Proceedin
gs
,
vol
/
issue:
17
17
(
1
)
,
p
p
.
30007
,
2016
.
[3]
D.
Utami,
e
t
al
.
,
“
Mate
ria
l
Char
ac
t
eri
z
at
ion
of
a
Magne
torhe
o
lo
gic
a
l
Fluid
Subj
ec
t
ed
to
Long
-
T
erm
Opera
ti
on
i
n
Dam
per
,
”
Ma
te
r
ial
s
,
vo
l
/i
ss
ue:
11
(
11
)
,
p
p
.
2195
,
2018.
[4]
T.
Le
-
Duc
,
et
a
l.
,
“
Multi
-
obj
ec
t
ive
op
ti
m
al
d
es
ign
of
m
ag
n
et
o
rhe
ological
bra
k
es
for
m
otorcy
c
li
ng
appl
i
ca
t
ion
conside
ring
ther
m
al
eff
ec
t
in
working
proc
ess,”
S
mar
t
Mate
rials a
nd
Struct
ures
,
v
ol
/i
ss
ue:
27
(
7
)
,
p
p.
075060
,
2018
.
[5]
N.
Najmae
i
,
e
t
al.
,
“
Suit
abi
l
ity
of
Sm
al
l
-
Scal
e
Magne
torhe
o
log
ic
a
l
Fluid
-
Based
Clut
ch
es
in
Ha
pti
c
In
te
rf
ac
es
f
or
Im
prove
d
Perfor
m
anc
e,”
IE
EE/A
SME
Tr
ansacti
o
ns on
Me
chat
ron
ic
s
,
vo
l
/i
ss
ue:
20
(
4
)
,
pp
.
1863
–
18
74,
2015
.
[6]
J.
Li,
e
t
al
.
,
“
Force
-
e
lectr
i
ca
l
ch
ara
c
te
rist
ic
s
of
a
novel
m
ini
-
d
amper,
”
Smar
t
Mat
erial
s
and
Struc
tures
,
vol
/
issue:
25
(
10
)
,
p
p
.
1050
09,
2016
.
[7]
M.
A
.
Port
il
lo
,
e
t
al
.
,
“
S
y
ner
g
y
b
et
wee
n
m
agne
to
rhe
ological
flu
id
s
and al
um
inum
foa
m
s:
Pros
pec
ti
ve al
te
rn
ative fo
r
seism
ic
damping
,
”
Journal
o
f
In
t
el
li
g
ent Mat
eria
l
Syste
ms
and
Str
uct
ures
,
vo
l
/i
ss
u
e:
27
(
7
)
,
pp
.
872
–
879,
2015
.
[8]
N.
Cat
er
ino,
et
al.
,
“
Shaking
t
a
ble
t
esti
ng
of
a
stee
l
fr
ame
struct
ure
equi
pped
with
sem
i
-
ac
t
i
ve
MR
dampers:
compari
son of c
ontrol
al
gori
thms
,
”
Smar
t
S
tructur
es
and
Syst
ems
,
vol
/
issue:
15
(
4
)
,
pp
.
963
–
995
,
2015.
[9]
T.
M.
Gurubasa
var
aj
u
,
et
al
.
,
“
E
val
ua
ti
on
of
optim
al
par
amete
rs
of
MR
flui
ds
for
damper
appl
icat
ion
using
par
ticl
e
sw
arm
and
response
surfac
e
opti
m
isat
ion
,
”
J
ournal
of
th
e
Brazili
an
Soc
iet
y
of
Me
cha
nical
Sc
ie
nc
es
an
d
Engi
ne
ering
,
vol
/i
ss
ue:
39
(
9
)
,
pp
.
3683
–
3694,
201
7.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
13
, N
o.
3
,
Ma
rc
h 201
9
:
1065
–
1072
1072
[10]
K.
Kara
koc
,
e
t
a
l.
,
“
Design
consi
der
ations
for
a
n
aut
om
oti
ve
m
ag
net
orhe
o
logi
c
al
bra
ke,”
Me
chat
r
onic
s
,
vol
/i
ss
ue:
18
(
8
)
,
pp
.
434
–
4
47,
2008
.
[11]
N.
M.
W
ere
l
e
y
,
“
Nondim
ensio
nal
Hersch
el
—
Bulkl
e
y
Ana
l
y
s
i
s
of
Magne
torh
eol
ogi
ca
l
and
E
le
c
trorhe
olog
ical
Dam
per
s,”
Jour
nal
of
Intelli
g
ent Mat
eria
l
S
yste
m
s and
Struct
ure
s
,
vol
/i
ss
ue:
19
(
3
)
,
pp.
257
–
268,
20
08.
[12]
I.
D.
Jung,
et
al.
,
“
A
comprehe
ns
ive
viscosity
m
odel
for
m
ic
ro
m
agne
tic
par
ticle
di
spersed
in
sili
co
ne
oil
,
”
Journal
of
Magne
ti
sm
an
d
Magne
t
ic Mat
erial
s
,
vo
l. 404,
pp.
40
–
44
,
2016
.
[13]
M.
As
hti
ani
,
et
a
l.
,
“
A re
vie
w on
t
he
m
agne
torhe
o
logi
c
al
flui
d
pre
par
ation
and
sta
bil
izati
on
,
”
Journal
of
Magne
ti
s
m
and
Magne
t
ic M
ate
rials
,
vol
.
37
4,
pp
.
711
–
715
,
2015.
[14]
J.
Goldasz
and
B.
Sapinski,
“
Veri
ficati
on
of
m
agne
torhe
olog
i
ca
l
shock
absor
ber
m
odel
s
with
var
ious
piston
conf
igurations,
”
Journal
of
Int
el
l
i
gent
Ma
te
rial
Sy
stems and
Structur
es
,
vol
/
issue:
24
(
15
)
,
pp
.
1846
–
1864,
2013
.
[15]
G.
A.
Dim
ock,
et
al.
,
“
Bingha
m
bipl
asti
c
an
aly
s
is
of
shea
r
thi
nning
and
t
hic
ken
ing
in
m
agne
torh
eol
ogi
c
al
dampers,”
Proceedi
ngs of
SPI
E
-
The
Inte
rnat
iona
l
Soc
i
ety for
Opt
ic
al
Engi
n
ee
ring
,
vol
.
3985
,
pp
.
4
44
–
455,
2000
.
[16]
E.
C.
B
ingha
m
,
“
An
Inve
stiga
ti
o
n
of
the
L
aws
of
Plasti
c
Flow,
”
The
Journal
of
h
ygi
en
e
,
vol
/i
ss
ue
:
13
(
1
)
,
pp.
309
–
353,
1916
.
[17]
S.
G.
Sherm
an,
et
al
.
,
“
Relati
ng
Mason
num
ber
to
Bingha
m
nu
m
ber
in
m
agne
t
orhe
ological
f
luids
,
”
Journal
of
Magne
ti
sm
and
Magne
tic
Mat
eri
als
,
vol
.
380
,
pp
.
98
–
104,
2015.
[18]
A.
Ghaffa
ri
,
et
a
l.
,
“
A
rev
ie
w
on
the
sim
ula
t
ion
a
nd
m
odel
ing
of
m
agne
torhe
olog
i
ca
l
flui
ds,
”
Jour
nal
of
In
te
l
li
ge
n
t
Mate
rial
Syste
m
s and
Struct
ures
,
vol
/i
ss
ue:
2
6(
8
)
,
pp.
881
–
904,
20
15.
[19]
A.
Farjoud,
e
t
al
.
,
“
Mathe
m
atical
Model
of
Drum
-
t
y
p
e
MR
Brakes
using
Hersche
l
-
Bulkl
e
y
Shea
r
Model,
”
Journal
of
Int
el
l
ige
nt
Ma
te
rial
Syste
ms
a
nd
Struct
ures
,
v
ol
/i
ss
ue:
19
(
5
)
,
p
p.
565
–
572
,
200
8.
[20]
I.
Bah
iuddi
n,
et
al.
,
“
A
new
con
stit
ut
iv
e
m
odel
of
a
m
agneto
-
rh
eol
ogi
ca
l
flui
d
a
ct
ua
tor
using
an
ext
r
eme
learni
n
g
m
ac
hine
m
e
thod
,
”
S
ensors
and
A
ct
uators A
:
Ph
ysic
al
,
vol
.
281
,
pp
.
209
–
221
,
2018
.
[21]
M.
Cvek,
et
al.
,
“
A
rhe
ologi
ca
l
e
val
ua
ti
on
of
stead
y
shea
r
m
agne
t
orhe
ological
flo
w
beha
vior
using
t
hre
e
-
par
ameter
viscopl
astic
m
od
el
s,”
Journal
of
Rhe
olog
y
,
vol
/
issue:
60
(
4
)
,
pp
.
6
87
–
694,
2016
.
[22]
M.
Cvek,
et
al.
,
“
A
rhe
ologi
ca
l
e
val
ua
ti
on
of
stead
y
shea
r
m
agne
t
orhe
ological
flo
w
beha
vior
using
thre
e
-
par
ameter
viscopl
astic
m
od
el
s,”
Journal
of
Rhe
olog
y
,
vol
/
issue:
60
(
4
)
,
pp
.
6
87
–
694,
2016
.
[23]
F.
Im
aduddi
n,
et
al.
,
“
A
high
per
form
anc
e
m
agne
torhe
olog
i
ca
l
va
lve
with
a
m
ea
nder
ing
f
low
pat
h,
”
Smar
t
Mate
rials and
St
ructures
,
vol
/i
ss
ue:
23
(
6
)
,
p
p
.
65
017,
2014
.
[24]
Y.
Rabba
ni
,
e
t
a
l.
,
“
Appli
ca
t
ion
of
art
if
ic
i
al
n
eur
al
ne
two
rks
and
support
vec
tor
r
egr
ession
m
odeling
in
pre
d
ic
t
ion
of
m
agne
torhe
ol
ogic
a
l
flui
d
rhe
o
m
et
er
y
,
”
Col
loi
d
s
and
Surface
s
A:
Phy
sicoc
h
emical
and
Engi
nee
ri
ng
Aspec
ts
,
vol
.
520,
pp
.
268
–
27
8,
2017
.
[25]
I.
Bahi
uddin
,
e
t
al.
,
“
Stud
y
o
f
ext
reme
learn
ing
m
ac
hine
a
c
ti
va
t
ion
func
t
io
ns
for
m
agne
to
rhe
ological
flu
id
m
odel
li
ng
in
m
e
dic
a
l
device
s
ap
pli
c
at
ion
,
”
2017
Inte
rnational
C
onfe
renc
e
on
Ro
boti
cs,
Aut
omati
on
and
Sci
en
ce
s
(
ICORAS)
,
pp.
1
–
5
,
2017
.
[26]
G.
B.
B
in
Huan
g,
e
t
al
.
,
“
Ext
re
m
e
le
arn
ing
m
achine
:
The
or
y
an
d
appl
i
cations,”
Neurocomputi
ng
,
vol
/i
ss
ue:
70
(
1
–
3
)
,
pp
.
489
–
501
,
2006.
[27]
I.
Bahi
uddin
,
et
al.
,
“
Constit
ut
i
ve
m
odel
s
of
ma
gnet
orh
eol
ogi
c
al
flui
ds
havi
ng
te
m
per
at
ur
e
-
de
pende
nt
pre
d
icti
on
par
amete
r
,
”
Sma
rt M
ate
ria
ls and
Struct
ures
,
vo
l
/is
sue:
27
(
9
)
,
p
p.
95001,
2018
.
[28]
I.
Bah
iud
din,
e
t
al.
,
“
A
Model
o
f
Magne
torh
eol
o
gic
a
l
Grea
s
e
usi
ng
Mac
hin
e
L
earning
Method
,
”
Key
Engi
ne
ering
Mate
rials
,
vol
.
7
75,
pp
.
191
–
197
,
2018.
[29]
T.
hu
Zh
ang,
et
al.
,
“
Com
par
ing
the
li
n
ea
r
and
lo
gar
it
hm
norm
al
i
ze
d
ar
ti
fi
cial
neu
ral
ne
tworks
in
i
nver
se
design
o
f
a
irc
r
aft c
ab
in en
vironment,
”
Bu
ilding Si
mula
ti
on
,
vol
/i
ss
ue:
9
(
6
)
,
pp.
729
–
734
,
20
16
.
Evaluation Warning : The document was created with Spire.PDF for Python.