Vous êtes sur la page 1sur 15

Mechanical Systems

and
Signal Processing
Mechanical Systems and Signal Processing 23 (2009) 3044
Identication of structural non-linearities using describing
functions and the ShermanMorrison method
Mehmet Bu lent O

zer
a
, H. Nevzat O

zgu ven
b,
, Thomas J. Royston
c
a
Baxter Healthcare Corporation, 1620 Waukegan Rd., MPGR-D1, McGaw Park, IL 60085, USA
b
Mechanical Engineering Department, Middle East Technical University, 06531 Ankara, Turkey
c
The Department of Mechanical and Industrial Engineering (M/C 251), University of Illinois at Chicago, 2039 ERF,
842 West Taylor Street, Chicago, IL 60607, USA
Received 27 March 2007; received in revised form 5 October 2007; accepted 17 November 2007
Available online 22 November 2007
Abstract
In this study, a new method for type and parametric identication of a non-linear element in an otherwise linear
structure is introduced. This work is an extension of a previous study in which a method was developed to localize non-
linearity in multi-degree of freedom systems and to identify type and parameters of the non-linear element when it is
located at a ground connection of the system. The method uses a describing function approach for representing the non-
linearity in the structure. The describing function contains only the rst harmonic terms. The ShermanMorrison matrix
inversion method is used in the present study to put the response expression in a form where the non-linearity term
can be isolated. Using measured responses one can calculate the value of the describing function representation of the non-
linear element and thus perform the identication. This new method can be used for type and parametric identication of a
non-linear element between any two coordinates of the system. Case studies are given to demonstrate the applicability of
the method.
r 2007 Elsevier Ltd. All rights reserved.
Keywords: Non-linear structural dynamics; Non-linear identication; Non-linear structural identication
1. Introduction
Engineering structures generally exhibit some degree of non-linearity. Localization and identication of
non-linearity by using experimentally measured response is important for several reasons. For instance, in
model updating, a non-linearity in the system will make any linear model incorrect and model updating would
not correct the problem; essentially, different corrections at different forcing levels would be required.
There has been extensive research on localization and identication of non-linearity in structures. Studies in
this eld can be classied into two groups: time domain and frequency domain techniques. Time domain
techniques generally focus on time series [1] and force state mapping [24]. Some of the frequency domain
techniques in which rst-order or higher-order harmonic responses are used to localize and/or identify
ARTICLE IN PRESS
www.elsevier.com/locate/jnlabr/ymssp
0888-3270/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ymssp.2007.11.014

Corresponding author. Tel.: +90 312 2102549; fax: +90 312 2102536.
E-mail address: ozguven@metu.edu.tr (H.N. O

zgu ven).
non-linearity in structures include the works of Tanrkulu and O

zgu ven [5], Vakakis and Ewins [6], Lin and


Ewins [7], S- anltu rk and Ewins [8], Richards and Singh [9], and O

zer and O

zgu ven [10].


In the work presented here the method developed in an earlier study [10] is used to localize non-linearity in
multi-degree of freedom (MDOF) systems by using measured harmonic responses. The method is then
extended to identify the type and the parametric values of the non-linear element within the structural system
where there is a single non-linear element between any two coordinates of the system. The identication
method presented in the earlier study [10] is restricted to systems with non-linear elements located at ground
connections only. Application of ShermanMorrison matrix inversion method allows the method to be used
to identify a non-linear element located at any place in the system. Some numerical case studies are presented
to demonstrate the application of the method.
2. Theory
The method developed is based on expressing the non-linear forcing vector in a matrix multiplication form
in the differential equation of motion of a MDOF non-linear system with harmonic excitation. This has been
rst achieved by Budak and O

zgu ven [11,12] in their studies analyzing harmonic vibrations of non-linear


MDOF systems, and then the method is generalized for any type of non-linearity by Tanrkulu et al. [13] by
using describing functions. This representation has been used in several other studies on harmonic vibrations
of non-linear MDOF systems. Some of the works in this eld are: Tanrkulu and O

zgu ven [14]; Co mert and


O

zgu ven [15]; Kuran and O

zgu ven [16]; Ferreira and Ewins [17], O

zer [18]; and Platten et al. [19].


2.1. Non-linearity matrix
Consider the differential equation of motion of a non-linear MDOF system with harmonic excitation
M f xg C f _ xg K fxg fNg ffg, (1)
where [M], [C], and [K] are mass, viscous damping and stiffness matrices of the linear system, respectively.
Here, {f} and {N} represent the external forcing and non-linear internal forcing, respectively. Harmonic
excitation is expressed as
ffg fFg e
iot
, (2)
where {F} denotes the amplitude vector of forcing and i represents the unit imaginary number. The assumed
response is
fxg fXg e
iot
, (3)
where {X} is the vector of response amplitudes. Since the response is assumed to be harmonic, the internal
non-linear forces can also be assumed to be harmonic, and then can be written as
fNg fGg e
iot
. (4)
Substituting Eqs. (2)(4) into Eq. (1) yields
o
2
M io C K
_ _
fXg fGg fFg. (5)
The amplitude of the internal forcing vector can be expressed [1012] as
fGg D fXg, (6)
where [D] is the response-dependent, so-called non-linearity matrix. The non-linearity matrix can be
formed by using describing function representations of internal non-linear forces. For details of how to
calculate the non-linearity matrix one should refer to the works of Tanrkulu and O

zgu ven [14] and


Tanrkulu et al. [13]. This representation makes it possible to perform harmonic vibration analysis of non-
linear MDOF systems with non-linearities that have describing function representations, provided that
an iterative solution accurately converges to the actual solution. This is basically the application of the
harmonic balance method and describing functions to non-linear MDOF systems. Moreover, by using this
ARTICLE IN PRESS
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 31


representation it is also possible to apply almost all the techniques developed for linear MDOF systems to
non-linear MDOF systems, provided that the non-linearity is not so strong so that the harmonic response
assumption holds true.
From Eqs. (5) and (6) the relation between harmonic forcing and response can be obtained as
o
2
M io C K D
_ _
fXg fFg. (7)
Then {X} can be written as
fXg h fFg, (8)
where [h] is the pseudo-receptance matrix of the non-linear structure, given by
h o
2
M io C K D
_ _
1
. (9)
Here, [h] is called pseudo-receptance since it is the receptance matrix of the system only for a specic
response level.
The receptance matrix [a] of the linear part of the same system can be written as
a o
2
M io C K
_ _
1
. (10)
From Eqs. (9) and (10), [D] can be obtained as
D h
1
a
1
. (11)
Post multiplying both sides of Eq. (11) by [h] yields
D h I Z h, (12)
where [Z] is the dynamic stiffness matrix of the linear part of the system, and is given by
Z a
1
K o
2
M io C
_ _
. (13)
Eq. (12) is the starting point for determining the harmonic response of a non-linear MDOF system by using
the receptances of the corresponding linear system.
2.2. Evaluation of non-linearity matrix using single harmonic response
Tanrkulu et al. [13] proposed the use of describing functions for the evaluation of [D]. The elements of [D]
can be found from u
rj
as follows:
D
rr
u
rr

n
j1
jar
u
rj
, (14)
D
rj
u
rj
, (15)
where u
rj
is the harmonic describing function representation of the non-linear force and can be evaluated using
the expression below:
u
rj

i
pY
rj
_
2p
0
N
rj
e
ic
dc, (16)
where N
rj
is the non-linear forcing between the rth and jth coordinates and Y
rj
is the amplitude of the inter-
coordinate displacement
Y
rj

if raj; X
r
X
j
if r j; X
r
_ _
(17)
and
c ot. (18)
ARTICLE IN PRESS
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 32


It should be noted that the harmonic describing functions of many systems have already been derived; a
table of describing functions for different types of non-linearities is given by Gelb and Van der Velde [20].
Common types of structural non-linearities are given in Table 1.
If the type of non-linearity is known, using the describing functions one can nd the describing
function representation of the non-linearity and form the [D] matrix using Eqs. (14)(18). It should be noted
that the expressions given above for the elements of [D] are also a function of the response itself. Therefore, in
order to calculate the response one should employ an iterative procedure. The initial guess can start with the
response of the linear system. First, the initial guess is used in the [D] expression (Eqs. (14)(18)) and then one
can solve for the non-linear response (Eq. (9)). In the next iteration step, the calculated responses are used in
obtaining the elements of [D] and non-linear responses are recalculated. This procedure goes on until a
convergence criterion is satised for the non-linear responses. Note, although the fundamental harmonic of
the response is considered here, it is always possible to include the higher harmonics, as explained, for
instance, in [16,17].
3. Identication of structural non-linearities using describing functions and ShermanMorrison method
3.1. Localization of structural non-linearity
O

zer and O

zgu ven [10] introduced a non-linearity number (now called non-linearity indexNLI) that can
be obtained from harmonic vibration measurements, in order to localize any non-linear element in the system.
Let us consider the ith column of Eq. (12), which will give
D fh
i
g fe
i
g Z fh
i
g, (19)
where {e
i
} is a vector of which the ith element is unity while all other elements are zero, and {h
i
} is the ith
column of [y]. The rth row of Eq. (19) gives,
D
r
fh
i
g d
ir
Z
r
fh
i
g, (20)
ARTICLE IN PRESS
Table 1
Describing function representations of common structural non-linear elements (A is the oscillation amplitude)
Type of non-linearity Describing function representation
Cubic stiffness
Desc
3
4
csA
2
Coulomb damping
Desc i
4F
f
pA
Piecewise stiffness
Desc m
1
m
2

2
p
arcsin
d
A
_ _

d
A

1
d
A
_ _
2
_
_ _
m
2
Friction-controlled backlash
Desc
1
2
1
2
p
arcsin 1
b
A
_ _
1
b
A
_ _

1 1
b
A
_ _
2
_
_ _ _ _
i
1
p
2b
A

b
A
_ _
2
_ _
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 33


where [D
r
] and [Z
r
] represent the rth rows of [D] and [Z], respectively, and d
ir
is the Kronecker Delta function.
The term d
ir
Z
r
fh
i
g is called NLI. If we consider the rth coordinate of the system, NLI
r
is an indication
of non-linear element(s) connected to the rth coordinate. This term is zero if the rth row and column of [D]
have all zeros. Any non-zero element in the rth row or column of [D] will yield a non-zero NLI for the rth
coordinate, since NLI
r
can be written from Eq. (20) as
NLI
r
D
r1
y
1i
D
r2
y
2i
D
rn
y
ni
. (21)
Here i can be any coordinate. However, in calculating NLI
r
from experimental measurements, the right-
hand side of Eq. (20) will be used:
NLI
r
d
ir
Z
r1
y
1i
Z
r2
y
2i
Z
rn
y
ni
. (22)
Then, NLI
r
can be calculated from harmonic vibrations measured at all coordinates connected
to the rth coordinate, while the system is harmonically excited at coordinate i. The measurements
are carried out under high forcing amplitudes to obtain the elements of pseudo-receptance vector {h
i
}. The
elements of the pseudo-receptance vector can be obtained as in the case of receptances of a linear system, i.e.,
by using the applied harmonic force and the complex displacement responses measured. However, unlike
linear receptances, pseudo-receptances are response amplitude dependant. On the other hand, [Z
r
] can be
obtained from receptances experimentally determined through response measurements at low forcing levels,
which can be practical only for systems with small number of degrees of freedom (DOFs). For systems with
large number of DOFs, however, [Z
r
] can also be obtained via analytical and numerical (nite element, etc.)
methods.
Eq. (22) implies that measurements from all DOFs are needed to calculate the NLI of a coordinate r.
However in practical applications, impedance matrices are sparse with several zero elements. Since NLI is a
weighted summation of pseudo-receptances (weights being the elements of impedance matrix), the pseudo-
receptances that are multiplied with zero or small impedance matrix elements do not need measurement.
Similarly, the NLI of each coordinate is usually not required. Most of the time structural systems consist of
mainly known linear materials, but have non-linearities due to joints, ground connections, etc. Therefore, for
any coordinate that is in the linear region of the system there is no need to calculate a NLI. The NLI should be
calculated only for coordinates around potential non-linearities, and this will considerably decrease the
number of coordinates that need measurement.
3.2. Analysis of non-linear structures using Sherman Morrison method
In this section, it will be shown that if there is a localized non-linear element in a vibrational system, the type
and parametric identication is possible if the ShermanMorrison matrix inversion formula is used in the
expression for response amplitudes. The ShermanMorrison matrix inversion formula was rst introduced by
Sherman and Morrison [21] and it has been applied to various vibrations and acoustics problems by O

zer and
Royston [2225] and Hong and Kim [26].
The system under consideration is shown in Fig. 1. The non-linearity can be between any of the DOFs of the
system. Let us assume that the non-linearity exists between rth and jth coordinates. The non-linearity matrix
ARTICLE IN PRESS
Fsin(t)
Non-linear elements
k
1
m
1
k
2
k
*
k
n
c
n
m
n
m
3
m
2
c
1 c
2 c
*
Fig. 1. N-degree of freedom non-linear system with a localized non-linearity.
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 34


[D] can be represented as
D
0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0
u u
u u
_ _
0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

0
_

_
_

_
. (23)
Since the non-linearity exists between the rth and jth coordinates the number of non-zero
elements of the [D] matrix is four. One can write the [D] matrix as a multiplication of two matrices as
shown below:
D fd
1
gfd
2
g
T
, (24)
where
fd
1
g
0
.
.
.
u
u
.
.
.
0
_

_
_

_
; fd
2
g
0
.
.
.
1
1
.
.
.
0
_

_
_

_
. (25,26)
It can be seen that only the rth and jth elements of the above vectors are non-zero. Using the above equation
in Eq. (8), which gives the non-linear response, one can obtain the following equation:
fXg Z fd
1
gfd
2
g
T
_
1
fFg, (27)
where the denition of [Z] is given in Eq. (13).
The ShermanMorrison matrix inversion formula is given as
A fugfvg
T
_
1
A
1

A
1
fugfvg
T
A
1
1 fvg
T
A
1
fug
, (28)
where [A] is an NN matrix with full rank and {u} and {v} are N1 vectors. Then, Eq. (27) can be written as
fXg a
afd
1
gfd
2
g
T
a
1 fd
2
g
T
afd
1
g
_ _
fFg. (29)
Recall that [a] is the receptance of the linear system and is given as
a Z
1
. (30)
ARTICLE IN PRESS
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 35


Using Eq. (29) one can write the response of each coordinate as follows:
X
1
.
.
.
X
r
X
j
.
.
.
X
n
_

_
_

X
lin
1
.
.
.
X
lin
r
X
lin
j
.
.
.
X
lin
n
_

_
_

afd
1
gfd
2
g
T
1 fd
2
g
T
afd
1
g
X
lin
1
.
.
.
X
lin
r
X
lin
j
.
.
.
X
lin
n
_

_
_

_
. (31)
The non-linear responses are on the left-hand side of the equation and they are experimentally measured
quantities. The vectors on the right-hand side of the equation are the linear response of the system. They can
be obtained from measured receptances of the system under low-amplitude loading conditions, where non-
linearities would not be excited. The only unknown in this equation is u, which is the describing function
representation of the non-linearity. Eq. (31) can be solved for the value of u, and the values of u at
different amplitudes of response can be plotted, from which the type and the parametric value of the non-
linearity can be identied.
4. Case studies
4.1. Localization of non-linearity in the structure
It was explained in Section 3.1 that the describing function representation of non-linearities and the non-
linearity matrix concept can be used to identify the location of non-linear elements. Unlike the type and
parametric identication methods developed for non-linearity, in this method there is no limitation to the
number of non-linear elements; so the non-linearity can be distributed to many DOFs. The localization
procedure can be used to locate the non-linearity infected coordinates. As a case study, the system shown in
Fig. 2 is investigated. This is a massspringdamper system with three DOFs. There is a non-linear macro-slip
element (Coulomb friction) between the rst and the second DOFs. The numerical values of the system are
given as follows:
M
1
1 kg; M
2
2 kg; M
3
1 kg; k
1
1000 N=m; k
2
1000 N=m; k
3
1000 N=m,
k
4
1000 N=m; c
1
8 Ns=m; c
2
2 Ns=m; c
3
2 Ns=m; c
4
2 Ns=m; F 1 N,
c
f
0:25 N Coulomb friction force.
In this case study, as well as in all others, the experimental results are simulated with the time domain
numerical solution of the differential equation of motion. A fth-order RungeKutta numerical integration
method is used to solve the equations of motion. The steady-state response (amplitude and phase) information
is obtained from the numerical solution. The response after 40 cycles is taken as the steady-state response, and
the time step is determined by dividing each cycle into 400 equal step sizes (i.e. a xed step size is used). Linear
ARTICLE IN PRESS
Fig. 2. The three-degree of freedom system with a macro-slip (Coulomb damping element).
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 36


responses are calculated by transforming the equations into the frequency domain. Using Eq. (22) one can nd
the NLIs for each coordinate. The NLIs are plotted out for frequencies between 40 and 80 rad/s. Figs. 35
show the plots of the NLIs for each DOF.
From Figs. 3 to 5 it can be observed that the NLI has nite values at coordinates 1 and 2, while the value of
the NLI for coordinate 3 is very close to zero almost at all frequencies. Figs. 35 show that the coordinates to
which a non-linear element is connected can be accurately identied using the method proposed in this study.
The primary reason for having a very small, but non-zero NLI at a coordinate that is not connected to a non-
linear element is the numerical inaccuracies in the simulated experiments. Similar results are expected in real
experimental cases due to experimental measurement errors.
ARTICLE IN PRESS
Fig. 3. The value of the non-linearity index at coordinate 1 evaluated at different frequencies.
Fig. 4. The value of the non-linearity index at coordinate 2 evaluated at different frequencies.
Fig. 5. The value of the non-linearity index at coordinate 3 evaluated at different frequencies.
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 37


4.2. Localization of non-linearity in the structure: multiple non-linear elements
Eq. (22) can be used also for localizing multiple non-linear elements. The system in this example is a 15
DOF system where the forcing is applied at the eighth DOF. The system is shown in Fig. 6 and the numerical
data for the system are as follows:
M
1...15
1 kg; k
1...15
20; 000 N=m; c
1...15
30 Ns=m; F 3 N located on the 8th coordinate,
k

6
k
6
3 10
7
x
2
N=m; k

11
k
11
3 10
7
x
2
N=m.
The response of the system is calculated by a fth-order RungeKutta numerical integration method. The
simulation was run for 250 cycles at each frequency to ensure that transients die out. The time step used in the
simulation is set such that each cycle is divided into 650 time steps. The frequency range used during
the simulations is between 2 and 4 Hz with frequency increments of 0.025 Hz. The NLI is calculated for each
coordinate using Eq. (22). Fig. 7 shows the non-linearity infected coordinates correctly. Since the two cubic
stiffness elements are between coordinates 5 and 6, and 10 and 11, these coordinates have higher NLI values as
compared to the rest of the coordinates.
4.3. Type and parametric identication of non-linearity: single degree of freedom system
A single degree of freedom (SDOF) system with cubic type of stiffness non-linearity will be used in this case
study to demonstrate that the type and parametric identication is possible applying the method suggested in
Section 3.2. An experimentally measured response will be simulated with a numerical solution of the
ARTICLE IN PRESS
Fig. 6. The 15 degree of freedom system with two cubic stiffness elements.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
0
2
4
6
8
10
12
14
16
18
Coordinate Number
S
u
m

o
f

N
o
n
-
l
i
n
e
a
r
i
t
y

I
n
d
i
c
e
s
Fig. 7. The summed values of NLI at all frequencies over the whole frequency range for all coordinates.
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 38


differential equation of motion. The numerical data for the system in Fig. 8 are
M 1 kg; k 100 N=m; k

10
4
N=m
3
; c 1 Ns=m.
For a SDOF system the [D] matrix is a scalar, which is equal to u. Therefore one can take {d
1
} to be u and
{d
2
} to be unity in Eq. (29). The nonlinear response X is equal to y times F (the amplitude of the forcing on the
system). One can substitute these scalar values into the right- and left-hand side of Eq. (29) and obtain the
following expression for the describing function of the non-linearity:
u
a y
ya
, (32)
where y is the pseudo-receptance of the non-linear SDOF system.
The receptance of the linear and non-linear system is given in Fig. 9. A typical jump phenomenon can be
observed in the non-linear response. The non-linearity value (D) (which is a scalar in this case) is calculated
using the linear and non-linear receptance data. Delta values for different response amplitudes can be
calculated using Eq. (32). The change of delta values with harmonic vibration amplitude is plotted in Fig. 10.
It can be seen from Fig. 10 that the non-linearity is a polynomial type of non-linearity. The scalar value of
delta is the describing function of the cubic type non-linearity. The describing function of a cubic type of
stiffness non-linearity is given as
u
cubic

3
4
csX
2
, (33)
where cs is the cubic stiffness coefcient and X is the response amplitude of the non-linear system. The solid
line in Fig. 10 is the data t (obtained through non-linear regression) to the delta values calculated from
Eq. (32). As depicted in Fig. 10, the curve t is closely tracing the numerically obtained delta values. Using
Eq. (33) and the quadratic equation of the t as given in Fig. 11, one can identify the cubic stiffness coefcient
to be 9765 N/m
3
, which is very close to the cubic stiffness coefcient of the system: 10,000 N/m
3
. Fig. 11 shows
the response of the original non-linear system and the response of the identied system. It can be seen that the
difference in the cubic stiffness coefcient result does not translate into an appreciable difference in the
frequency domain response.
4.4. Type and parametric identication of non-linearity: multi-degree of freedom system-I
In this case study, type and parametric identication of a MDOF system will be performed. The system that
will be identied is the same system as the one in Section 4.1. Using Eq. (31) one can solve for the unknown
describing function representation of the non-linearity.
ARTICLE IN PRESS
Fig. 8. A single degree of freedom system with cubic stiffness.
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 39


Using Eq. (31) one can write the response of the kth coordinate when the non-linearity is located between
the rth and jth coordinates as follows:
X
k
X
lin
k

a
kr
a
kj
X
lin
r
X
lin
j
_ _
u
1 ua
rr
2a
rj
a
jj

. (34)
From Eq. (34) one can solve for the describing function representation of the non-linearity, u, as shown
below:
u
X
lin
k
X
k
X
k
X
lin
k
_ _
a
rr
2a
rj
a
jj
a
kr
a
kj
X
lin
r
X
lin
j
_ _. (35)
ARTICLE IN PRESS
Fig. 10. The delta values, ( ) calculated values, and (_____) obtained by performing a curve t on calculated data.
Fig. 9. The receptance of the SDOF system case study: (UUUUUUUUUUUUUUUU) linear receptance, (_____) non-linear pseudo-receptance
(increasing frequency), and (

) non-linear pseudo-receptance (decreasing frequency).
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 40


In this case study, the k value (measured coordinate) is taken to be the rst coordinate, the r and j values
(the coordinates to which the non-linear element is attached) are 1 and 2, respectively.
Using the measured linear and non-linear responses of the kth coordinate (which are all simulated) along
with the linear receptance matrix, one can nd the values of the describing function. Note that the linear
response can be obtained as the response of the non-linear system at a low forcing level. If the describing
function of the non-linearity and the amplitude of the response are plotted, behavior of the curve allows us to
identify the type of the non-linearity. If a simple curve tting algorithm is applied, the parametric
identication is also possible.
It can be observed from Fig. 12 that the describing function representation of the non-linearity decreases as
the amplitude of the response increases. Table 1 shows that the only non-linearity that would cause such a
behavior is the Coulomb damping type of non-linearity. Using the describing function representation for
Coulomb damping one can t a curve to the data given in Fig. 10. The curve t (using non-linear regression
ARTICLE IN PRESS
Fig. 11. The receptance of (______) original system, (

) original system (decreasing frequency), and (UUUUUUUUUUUUUUUU) identied
system.
Fig. 12. The plot of the calculated describing function values at different measured response amplitudes.
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 41


analysis) gives a Coulomb friction force of 0.251 N, which means less than 1% error in estimating the
frictional force.
4.5. Type and parametric identication of non-linearity: multi-degree of freedom system-II
The previous two case studies have small numbers of DOFs. In this case study, a system with more DOFs
will be used (15 DOF). The type of non-linearity is coulomb friction as in the previous example. In this case
study, the coulomb friction element is not located in between the system coordinates. It is located between the
eighth DOF and ground. The system is depicted in Fig. 13
M
1...15
1 kg; k
1...16
20; 000 N=m; c
1...16
30 Ns=m; F 3 N; F
f
0:12 N.
Eq. (35) shows the describing function representation when the non-linear element is between the
coordinates r and j. Eq. (35) can be modied for a non-linearity between ground and a certain DOF by
omitting all the terms that contain coordinate j:
u
X
lin
k
X
k
X
k
X
lin
k
_ _
a
rr
a
k
X
lin
r
_ _ . (36)
Since an increased number of DOFs is used in this example, the number of oscillations to reach the steady
state is increased to 250 in the numerical solution, and the responses at 500 time instants are determined for
each cycle. This is done to ensure that all of the coordinates reach steady state. The frequency range for the
ARTICLE IN PRESS
Fig. 13. 15 DOF system with macro-slipMDOF system-II.
Fig. 14. The summed values of NLI at all frequencies over the whole frequency range for all coordinatesMDOF system-II.
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 42


solution is taken to be 3.85 Hz with a frequency increment of 0.025 Hz. This frequency range covers the rst
natural frequency of the 15 DOF system.
In this case study the NLI is calculated at each frequency simulation location. The amplitude of NLI at each
frequency is summed over the whole frequency range for each DOF. The plot of the sum of the amplitudes of
NLI for the simulation frequency range for all DOFs is shown in Fig. 14.
As shown in Fig. 14, the non-linearity infected coordinate is successfully identied as coordinate 8. The
change of the value of describing function for different vibration amplitudes around the rst resonance is
shown in Fig. 15. Using a least-squares curve tting algorithm the frictional force is determined as 0.133 N.
It should be noted that due to increased number of DOFs, the computational time required for
solving equations of motion (for simulating experiments) has increased drastically. Therefore, in the
numerical solutions the response may not have reached steady state and/or it may contain slight contributions
from higher harmonics which are not considered in identication. Using the method described in [16] the rst-
order harmonic response has also been calculated for the same system, and about a 2% difference was
observed in the mean value between rst-order harmonic solution and the steady-state time domain solution.
This error in numerical solution gives some idea on how the method would respond to measurement error.
The effect of this difference can be seen in Fig. 15. This difference leads to a deviation of 10% in the estimated
friction force.
5. Conclusions
A new method is introduced for type and parametric identication of a non-linear element in an otherwise
linear structure. This method is an extension of the method developed by the rst two authors in an earlier
work [10] to localize the non-linearity in a MDOF system and to identify type and parameter values of the
non-linear element when it is located between the system and ground. The non-linearity matrix concept is
used also in this study, and the ShermanMorrison matrix inversion method is applied to put the response
expression in a form where the non-linearity term can be isolated. This made it possible to identify type and
parameters of a non-linear element between any two coordinates of the system. Case studies are given to show
that the method can be used to nd the location of the non-linear element, and also to identify types of non-
linearities successfully. In the case studies given, cubic stiffness and Coulomb damping non-linearities are
considered, and both localization and parametric identication of the non-linearities are demonstrated. In the
last case study it is shown that the method can successfully be applied to large-scale systems. However, it is
observed in this case study that larger measurement errors may yield inaccurate values in parametric
identication. As it is aimed in this paper to introduce an index, which can be used to locate non-linearities in
ARTICLE IN PRESS
Fig. 15. The plot of the calculated describing function values at different measured response amplitudes.
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 43


structural dynamics by using FRF measurements, and give the theory of a new method to parametrically
identify such structural non-linearity, the case studies are given just to verify the theory developed. In order to
investigate further the applicability of the method suggested to real structures, it will be necessary to study the
accuracy of the method when real experimental measurements are used.
References
[1] Q. Chen, G.R. Tomlinson, A new type of time series model for the identication of nonlinear dynamical systems, Mechanical Systems
and Signal Processing 8 (5) (1994) 531549.
[2] E.F. Crawley, A.C. Aubert, Identication of non-linear structural elements by force state mapping, AIAA Journal 24 (1) (1986)
155162.
[3] G. Kerschen, J.C. Golinval, Theoretical and experimental identication of a non-linear beam, Journal of Sound and Vibration 244 (4)
(2001) 597613.
[4] G.E.B. Tan, S. Pellegrino, Non-linear dynamic identication: an application to pre-stresses cable structures, Journal of Sound
Identication 208 (1) (1997) 3345.
[5] O

. Tanrkulu, H.N. O

zgu ven, A new method for the identication of non-linearities in vibrating structures, in: Proceedings of the
Fourth International Conference on Recent Advances in Structural Dynamics, Southampton, England, July 1518, 1991.
[6] A.F. Vakakis, D.J. Ewins, Effects of weak non-linearities on modal analysis, Proceedings of IMAC 10 (1992) 7278.
[7] R.M. Lin, D.J. Ewins, Location of localised stiffness non-linearity using measured data, Mechanical Systems and Signal Processing 9
(3) (1995) 329339.
[8] K.Y. S- anltu rk, D.J. Ewins, Modelling two dimensional friction contact and its application using harmonic balance method, Journal
of Sound and Vibration 193 (2) (1996) 511523.
[9] C.M. Richards, R. Singh, Feasibility of identifying non-linear vibratory systems consisting of unknown polynomial forms, Journal of
Sound and Vibration 220 (3) (1999) 413450.
[10] M.B. O

zer, H.N. O

zgu ven, A new method for localization and identication of non-linearities in structures, in: Proceedings of the
ESDA 2002: Sixth Biennial Conference on Engineering Systems Design and Analysis, ASME, I

stanbul, Turkey, July 811, 2002.


[11] E. Budak, H.N. O

zgu ven, Iterative receptance method for determining harmonic response of structures with symmetrical non-
linearities, Mechanical Systems and Signal Processing 7 (1) (1993) 7587.
[12] E. Budak, H.N. O

zgu ven, A dynamic analysis method for the harmonically excited non-linear structures, Structural Vibration and
Acoustics, ASME 18 (3) (1989) 2329.
[13] O

. Tanrkulu, B. Kuran, H.N. O

zgu ven, M. Imregu n, Forced harmonic response analysis of non-linear structures using describing
functions, AIAA Journal 31 (7) (1993) 13131320.
[14] O

. Tanrkulu, H.N. O

zgu ven, A new method for the identication of non-linearities in vibrating structures, in: Proceedings of the
Fourth International Conference on Recent Advances in Structural Dynamics, Southampton, England, July 1518, 1991.
[15] M.D. Co mert, H.N. O

zgu ven, A method for forced harmonic response substructures coupled by non-linear elements, in: Proceedings
of the 13th International Modal Analysis Conference, Nashville, TN, 1995, pp. 139145.
[16] B. Kuran, H.N. O

zgu ven, A modal superposition method for non-linear structures, Journal of Sound and Vibration 189 (3) (1996)
315339.
[17] J.V. Ferreira, D.J. Ewins, Algebraic nonlinear impedance equation using multi-harmonic describing function, in: Proceedings of the
XV International Modal Analysis Conference-IMAC 96, Orlando, USA, February 1997.
[18] M.B. O

zer, A new method in identication of symmetrical non-linearities in structures, M.Sc. Thesis in Mechanical Engineering,
Middle East Technical University, 1999.
[19] M.F. Platten, J.R. Wright, J.E. Cooer, M. Sarmast, Identication of multi degree of freedom non-linear systems using an extension of
force appropriation, IMAC XX, Los Angeles, 2002.
[20] A. Gelb, W.E. Van der Velde, Multiple-input Describing Functions and Nonlinear System Design, McGraw-Hill, New York, 1968.
[21] J. Sherman, W.J. Morrison, Adjustment of an inverse matrix corresponding to changes in the elements of a given column or a given
row of the original matrix, Journal of Annals of Mathematical Statistics 20 (1949), 621621.
[22] M.B. O

zer, T.J. Royston, Extending Den Hartogs vibration absorber technique to multi degree of freedom systems, Journal of
Vibrations and Acoustics 127 (4) (2005) 341350.
[23] M.B. O

zer, T.J. Royston, Application of ShermanMorrison matrix inversion formula to damped vibration absorbers attached to
multi degree of freedom systems, Journal of Sound and Vibration 283 (35) (2005) 12351249.
[24] M.B. O

zer, T.J. Royston, Passively minimizing structural sound radiation using shunted piezoelectric materials, The Journal of the
Acoustical Society of America 114 (4, Part 1) (2003) 19341946.
[25] M.B. O

zer, New techniques in the design, analysis and optimization of tuned dynamic vibration absorbers, Ph.D. Thesis in
Mechanical Engineering, University of Illinois at Chicago, 2004.
[26] K.L. Hong, J. Kim, New analysis method for general acousticstructural coupled systems, Journal of Sound and Vibration 192 (1996)
465480.
ARTICLE IN PRESS
M.B. O

zer et al. / Mechanical Systems and Signal Processing 23 (2009) 3044 44