
675a nonlinear operator
N
[
x
(
t
),
x˙
(
t
)]:
L
[
x
(
t
)]+
N
[
x
(
t
),
x˙
(
t
)]=
f
(
t
),
L
[
x
(
t
)]=
Mx¨
(
t
)+
Cx˙
(
t
)+
Kx
(
t
),
N
[
x
(
t
),
x˙
(
t
)]=
s
n j
=1
A
j
y
j
(
t
), (1a–c)where
M
,
C
and
K
are the mass, damping and stiﬀness matrices, respectively,
x
(
t
) is thegeneralized displacement vector and
f
(
t
) is the generalized force vector. Also refer to theappendix for the identiﬁcation of symbols. The nonlinear operator
N
[
x
(
t
),
x˙
(
t
)] containsonly the nonlinear terms which describe the localized constraint forces and this operatoris written as the sum of
n
unique nonlinear function vectors
y
j
(
t
) representing each
j
thtype of nonlinearity present (e.g., quadratic, cubic, ﬁfth order, etc.). Considering onlynonlinear elastic forces, each
y
j
(
t
) is deﬁned as
y
j
(
t
)=
{
D
x
k
(
t
)
m
j
}
, where
D
x
k
(
t
) is therelative displacement across the
k
th junction where the
j
th type of nonlinearity exists, and
m
j
is the power of the
j
th type of nonlinearity. These vectors
y
j
(
t
) are column vectors of length
q
j
, where
q
j
is the number of locations the
j
th type of nonlinearity exists. Note thata single physical junction may contain more than one type of nonlinearity (e.g., aquadratic and cubic); therefore, more than one
y
j
(
t
) is necessary to describe the nonlinearconstraint force across that particular junction, as illustrated in the examples to follow.The coeﬃcient matrices
A
j
contain the coeﬃcients of the nonlinear function vectors andare of size
N
by
q
j
. Inserting equations (1b) and (1c) into equation (1a), the nonlinearequations of motion take the form
Mx¨
(
t
)+
Cx˙
(
t
)+
Kx
(
t
)+
s
n j
=1
A
j
y
j
(
t
)=
f
(
t
). (2)From a system identiﬁcation perspective, it is assumed that the types of nonlinearities andtheir physical locations are known. Therefore the
n
nonlinear function vectors
y
j
(
t
) canbe calculated; also, the coeﬃcients of
y
j
(
t
) can be placed in the proper element locationsof the coeﬃcient matrices
A
j
. This assumption renders limitations on the practical use of this method since various types of nonlinearities at each location are not always known.Therefore, research is currently being conducted to alleviate this limitation. However, itshould be noted that this restriction is currently true for any identiﬁcation scheme whenapplied to practical nonlinear systems.Consider several multidegreeoffreedom nonlinear systems as illustrated in Figure 1.The ﬁrst example as shown in Figure 1(a) possesses an asymmetric quadratic–cubicnonlinear stiﬀness element which exists between the second and third masses and aGaussian random excitation is applied to the ﬁrst mass:
f
e
23
(
t
)=
k
2
(
x
2
(
t
)−
x
3
(
t
))+
a
2
(
x
2
(
t
)−
x
3
(
t
))
2
+
b
2
(
x
2
(
t
)−
x
3
(
t
))
3
,
f
(
t
)=[
f
1
(
t
) 0 0]
T
. (3a, b)Assuming that the form of the nonlinear elastic force
f
e
23
(
t
) is known, the nonlinearoperator
N
[
x
(
t
),
x˙
(
t
)], the nonlinear functions (
y
1
(
t
) and
y
2
(
t
)) and their respectivecoeﬃcient matrices (
A
1
and
A
2
) take the form
N
[
x
(
t
),
x˙
(
t
)]=
A
1
y
1
(
t
)+
A
2
y
2
(
t
),
y
1
(
t
)=(
x
2
(
t
)−
x
3
(
t
))
2
,
y
2
(
t
)=(
x
2
(
t
)−
x
3
(
t
))
3
,
A
1
=(0
a
2
−
a
2
)
T
,
A
2
=(0
b
2
−
b
2
)
T
. (4a–e)