and
Signal Processing
Mechanical Systems and Signal Processing 23 (2009) 3044
Identication of structural nonlinearities 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., MPGRD1, 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 nonlinear 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 multidegree of freedom systems and to identify type and parameters of the nonlinear 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 nonlinearity 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
nonlinear 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: Nonlinear structural dynamics; Nonlinear identication; Nonlinear structural identication
1. Introduction
Engineering structures generally exhibit some degree of nonlinearity. Localization and identication of
nonlinearity by using experimentally measured response is important for several reasons. For instance, in
model updating, a nonlinearity 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 nonlinearity 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 rstorder or higherorder harmonic responses are used to localize and/or identify
ARTICLE IN PRESS
www.elsevier.com/locate/jnlabr/ymssp
08883270/$  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.
Email address: ozguven@metu.edu.tr (H.N. O
zgu ven).
nonlinearity in structures include the works of Tanrkulu and O
zer and O
n
j1
jar
u
rj
, (14)
D
rj
u
rj
, (15)
where u
rj
is the harmonic describing function representation of the nonlinear 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 nonlinear 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 and O
zgu ven [10] introduced a nonlinearity number (now called nonlinearity indexNLI) that can
be obtained from harmonic vibration measurements, in order to localize any nonlinear 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 nonlinear elements (A is the oscillation amplitude)
Type of nonlinearity 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
Frictioncontrolled 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 and
Royston [2225] and Hong and Kim [26].
The system under consideration is shown in Fig. 1. The nonlinearity can be between any of the DOFs of the
system. Let us assume that the nonlinearity exists between rth and jth coordinates. The nonlinearity matrix
ARTICLE IN PRESS
Fsin(t)
Nonlinear 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. Ndegree of freedom nonlinear system with a localized nonlinearity.
M.B. O
0
_
_
_
_
. (23)
Since the nonlinearity exists between the rth and jth coordinates the number of nonzero
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 nonzero. Using the above equation
in Eq. (8), which gives the nonlinear 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
_
_
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 nonlinear responses are on the lefthand side of the equation and they are experimentally measured
quantities. The vectors on the righthand side of the equation are the linear response of the system. They can
be obtained from measured receptances of the system under lowamplitude 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 nonlinearity. 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 nonlinearity in the structure
It was explained in Section 3.1 that the describing function representation of nonlinearities and the non
linearity matrix concept can be used to identify the location of nonlinear elements. Unlike the type and
parametric identication methods developed for nonlinearity, in this method there is no limitation to the
number of nonlinear elements; so the nonlinearity can be distributed to many DOFs. The localization
procedure can be used to locate the nonlinearity 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 nonlinear macroslip
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 fthorder RungeKutta numerical integration
method is used to solve the equations of motion. The steadystate response (amplitude and phase) information
is obtained from the numerical solution. The response after 40 cycles is taken as the steadystate 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 threedegree of freedom system with a macroslip (Coulomb damping element).
M.B. O
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 fthorder 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 nonlinearity 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 nonlinearity: single degree of freedom system
A single degree of freedom (SDOF) system with cubic type of stiffness nonlinearity 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
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 lefthand side of Eq. (29) and obtain the
following expression for the describing function of the nonlinearity:
u
a y
ya
, (32)
where y is the pseudoreceptance of the nonlinear SDOF system.
The receptance of the linear and nonlinear system is given in Fig. 9. A typical jump phenomenon can be
observed in the nonlinear response. The nonlinearity value (D) (which is a scalar in this case) is calculated
using the linear and nonlinear 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 nonlinearity is a polynomial type of nonlinearity. The scalar value of
delta is the describing function of the cubic type nonlinearity. The describing function of a cubic type of
stiffness nonlinearity 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 nonlinear system. The solid
line in Fig. 10 is the data t (obtained through nonlinear 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 nonlinear 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 nonlinearity: multidegree of freedom systemI
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 nonlinearity.
ARTICLE IN PRESS
Fig. 8. A single degree of freedom system with cubic stiffness.
M.B. O
. (34)
From Eq. (34) one can solve for the describing function representation of the nonlinearity, 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, (_____) nonlinear pseudoreceptance
(increasing frequency), and (
) nonlinear pseudoreceptance (decreasing frequency).
M.B. O
. Tanrkulu, H.N. O
zgu ven, A new method for the identication of nonlinearities 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 nonlinearities on modal analysis, Proceedings of IMAC 10 (1992) 7278.
[7] R.M. Lin, D.J. Ewins, Location of localised stiffness nonlinearity 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 nonlinear 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 nonlinearities in structures, in: Proceedings of the
ESDA 2002: Sixth Biennial Conference on Engineering Systems Design and Analysis, ASME, I
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 nonlinear structures, Structural Vibration and
Acoustics, ASME 18 (3) (1989) 2329.
[13] O
zgu ven, M. Imregu n, Forced harmonic response analysis of nonlinear 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 nonlinearities 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 nonlinear 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 nonlinear structures, Journal of Sound and Vibration 189 (3) (1996)
315339.
[17] J.V. Ferreira, D.J. Ewins, Algebraic nonlinear impedance equation using multiharmonic describing function, in: Proceedings of the
XV International Modal Analysis ConferenceIMAC 96, Orlando, USA, February 1997.
[18] M.B. O
zer, A new method in identication of symmetrical nonlinearities 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 nonlinear systems using an extension of
force appropriation, IMAC XX, Los Angeles, 2002.
[20] A. Gelb, W.E. Van der Velde, Multipleinput Describing Functions and Nonlinear System Design, McGrawHill, 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