EXPERIMENTAL NONLINEAR FREQUENCY RESPONSE

DETERMINATION USING THE ARC-LENGTH METHOD

Janito Vaqueiro Ferreira(1) , Alberto Luiz Serpa(2) and Alex Pereira do Prado(3)
State University of Campinas
Faculty of Mechanical Engineering
Computational Mechanics Department
13083-970 Campinas-SP, Brazil
(1) e-mail: janito@fem.unicamp.br
(2) e-mail: serpa@fem.unicamp.br

ABSTRACT
In the last years, several iterative numerical techniques have
been developed to solve nonlinear structural problems. Some of
these methods are capable to trace complex paths in the space
load/displacement. One of those most popular procedures is
the arc-length method of Crisfield, that possesses the capability
to overcome inflection points, without having the necessity of
determining them. This method has been successfully applied
to obtain theoretical curves of the frequency response of nonlinear dynamic systems. Meanwhile, measuring this frequency
response curves is still a hard work. This work consists of the
application of the arc-length method as a control strategy to obtain the measured nonlinear frequency response of a nonlinear
mechanical system.

INTRODUCTION

The frequency domain analysis of linear structures is characterized by a set of unique frequency response functions which
can be defined as frequency-dependent quantities calculated
from the ratio between a harmonic displacement response and
the harmonic driving force. However, for nonlinear structures,
the total response of a nonlinear system in the time domain is
represented by a sequence of frequency response functions in
the frequency domain in contrast to only one function in the
linear case. The stronger the nonlinearities are, the more frequency responses are needed to represent the total response.
On the other hand, for a wide class of nonlinear systems, most
of the dominant effects are contained in the so-called first, second and third order frequency response and these are often
sufficient to characterize the system quite accurately [14, 16].
The frequency response functions for linear structures
(FRFs) have been successfully measured for many years in
a wide variety of modal testing applications using different
types of excitation to drive the test structure [4]. Although
the measured frequency response functions of linear structures
are independent of the choice of the excitation technique, most
engineering structures are often found to exhibit the characteristic that the overall structure response is actually nonlinear and dependent on the level and the kind of excitation.
Therefore, an appropriate excitation technique should be selected in order to study a nonlinear system. It is important to
notice that FRFs correspond to the conventional concept of
Frequency response function of linear systems and FRs correspond to frequency response of nonlinear system.
Sine excitation is one of the most periodic commonlyapplied excitation techniques to obtain frequency response

(3) e-mail: alexp@fem.unicamp.br

functions because of its uniqueness and precision characteristics. This frequency response function is obtained by using
steady-state harmonic excitation. For each frequency, a force
is applied which consists of a constant-amplitude sine wave.
The displacement response is allowed to reach a steady-state
condition and the spectral analysis of the excitation and the
response is calculated.
The great advantage of a sinusoidal excitation is related
with its frequency-selective nature. In this case, the level of
the input force can be accurately controlled, a feature which
becomes very important in the successful evaluation of frequency responses of nonlinear structures [15] due to the fact
that harmonic excitation reveals the distortions of the frequency responses (FRs) in the resonance regions for different
levels of force. Harmonic excitation also reveals the subharmonic and superharmonic responses that are clear manifestations of nonlinear behaviour. Furthermore, the measurements
can be concentrated where they are required, having a different frequency increment in different frequency ranges. For
instance, near resonances and antiresonances, the FRs exhibit
rapid changes and a fine frequency increment is recommended.
On the other hand, away from resonances and antiresonances,
the variation is very slow and a wider frequency gap can be
used instead. In addition, the signal-to-noise ratio is generally
good because once the energy is concentrated at one frequency,
the response in the same frequency and in the harmonics and
intermodulations can be averaged out through an integration
process.
For linear structures, when the input is a sinusoid, the
response is also a sinusoid with the same frequency as the excitation but with a different magnitude and phase. Thus, for
a linear case, just one component of the frequency spectrum is
extracted at each frequency point. However, for nonlinear systems, even when the input is a pure sinusoid, the response is
composed of a number of frequency components, such as harmonics and intermodulation frequencies. In this case, the first
and higher order frequency responses can be calculated using
the ideal mathematical definition related with the Volterra series [8]. Although, in theory, it is possible to calculate the FRs
obtained from the Volterra series, the measurement of these
quantities has proved difficult in practice. Several techniques
have been developed for obtaining the ideal FRs [1, 8], but
when applied to physical structures, the procedures do not
obtain FRFs of good quality [7, 8]. The most promising one
is the NARMAX procedure [1, 2].
Apart from this mathematical definition of FRs obtained
from the Volterra series, there are other definitions based on
experimental measurements that are approximations of the
ideal one [10]. In the current work, the experimental defi-

nition of FRs will be introduced in order to be compatible

with the analytical methods that are based on experimental
FRFs. These non-unique definitions can still contain useful
information about the behaviour of a system.
Two different conditions can be used to obtain frequency
response of nonlinear systems. The first one is by keeping
the amplitude of the force level constant at all the different
excitation frequencies. This technique allows us to observe
the distortion of the frequency response such as the jump
phenomenon as the input level is increased. The second one
is by keeping the amplitude of the response level constant at
different excitation frequencies. Here, the frequency response
function obtained looks like the frequency response function
of a linear system. This allows us to apply standard modal
analysis techniques to obtain some linearized characteristics
of the system.
As discussed above, in the case where the frequency response of nonlinear structures for different vibration response
levels are going to be measured, sinusoidal excitation is strongly
recommended. However, the choice of the excitation is only
the first step to obtain the frequency response. There are
still a number of possible practical problems which need to be
carefully considered in order to achieve a successful frequency
response measurement such as the shaker linearity range and
push-rod dynamic effect both not considered yet in this work.
Usually in order to obtain FRF measurements using sinusoidal excitation require a function generator to generate a
sinusoidal voltage, the attachment of a single electrodynamic
shaker to drive the test structure, a power amplifier that applies the sinusoidal voltage from the function generator to the
shaker, a flexible stinger to transmit the excitation force to the
structure, a force transducer measuring the input force and
multiple accelerometers to measure the structural responses.
The excitation and response signals are acquired and then processed by using a digital spectrum analyser to perform the RF
calculations. In theory, acquiring the RF using the described
measurement system should be straightforward. However, in
practice, to ensure good measurements of structural frequency
response requires consideration of many aspects of the whole
measurement system. There are situations where their success
have been limited, especially in the systems presenting limit
points with snap-throughs and snap-backs behaviour as
shown on figure 1.
50

snapthrough

Magnitude [dB]

55

60

65

snapback
70

and developed by Crisfield [3] and Ramm [12] for dealing with
both cases of limit points. However they seem to be used
rarely applied in dynamics analysis [9], especially in obtaining
nonlinear frequency responses.
This paper applies the arc-length method on dynamic
problems to measure and control the periodic response of a
nonlinear system under periodic excitation obtaining as a result the nonlinear frequency response and also evaluates the
robustness of the method on dynamic problems with localized
nonlinearities.

First-Order Frequency Response

In concept, the first-order frequency response are an extension

of the frequency response functions of linear structures to nonlinear structures. In the case of a pure sinusoidal excitation,
the first-order frequency response of a nonlinear structure is
defined as the spectral ratio of the response xi and the force
fj at the frequency of excitation, , written as:
11
Hij
(, Fj1 ) =

Xi1
Fj1

(1)

where, is the frequency, H 11 is the frequency response (FR)

of the fundamental frequency component of the response and
the excitation. The subindexes i and j are the measuring and
excitation coordinates respectively.
In this case, only the fundamental frequency component
X 1 and F 1 of the response x and of the force f are retained
and all the subharmonics, superharmonics and combinations
of both are ignored.

Nonlinear Frequency Response

using Arc-Length Method

The matrix differential equation of motion for a nonlinear

structure subject to an external excitation {f } having internal
localized nonlinear forces, {f }, can be written as
[M ]{
x} + [C]{x}
+ i[D]{x} + [K]{x} + {f } = {f }

(2)

where, [M ] is the mass matrix, [C] is the viscous damping

matrix, [D] is the hysteretic damping matrix, [K] is the stiffness matrix and {x}, {x}
and {
x} are the displacement, the
velocity and the acceleration vectors respectively.
Assuming now that the excitation is harmonic and that
the response is dominated by his harmonic solution, it is possible to write the matrix differential equation in frequency
domain
([K] 2 [M ] + i[C] + i[D] + []1 ){X 1 }

{Fapp } (3)

75

80
0

0.5

1.5

2.5

3.5

Frequency [Hz]

Figure 1: Snap-through and Snap-back phenomena

The analytical analysis of these sistems has been improved
with the arc-length method initially proposed by Riks [13]

where, []1 is the fist-order describing function matrix, Fapp

is the pure harmonic force applied into the system and X 1
is the first harmonic component of the complete response for
that level of applied force.
For obtaining the experimental nonlinear frequency response it is necessary to choose the variable to be controlled.
The variable to be controlled is the force applied into the system, which in this case depends on the level of the harmonic
signal been sent to the exciter and its respective frequency.

Once defined the variable to be controlled, a general equation in residual form considering a single input and a single
output, relating the force applied into the system and the analytical force to be controlled can be written as

0.8

Force

(V, ) = |Fapp (V, )| Fana

(4)

where V and are the harmonic amplitude voltage and respective frequency of the input sine of the exciter, the scalar
is a load-level parameter, |Fapp (V, )| is the module of
the measured force applied of the structure and Fana is the
harmonic level of the analytical force to be controlled.
The above equation can be solved by many incremental numerical procedures (Newton-Raphson, Newton Raphson modified, Quasi-Newton etc) where force controlled can
be applied. However this methods are not appropriate for solution of problems where equilibrium solution paths need to be
traced beyond limit points. Such situations, typically arising
in structural instability problems, are better dealt by continuation techniques, of which the arc-length method appears to
be the most popular one.
The Arc-Length method is described within the category
of continuation methods and it is applied to obtain solution
paths. Basically, the arc-length method first consider the load
factor , as a variable in the residual equation (4). Then, an
extra new constraint equation is added to the residual equilibrium equation (4) for defining unequivocally the next equilibrium point solution at an intersection between the solution
path and the restriction equation. Finally the nonlinear extend system is then solved using standard iterative techniques
to obtain that equilibrium point solution.
For solving direct both variables of the extended system,
one from the voltage V and the other one from the load factor
variable , using the iterative methods, it is necessary first to
linearize the residual equation, with V and being unknown
variables, together with the relevant arc-length constraint by
the Taylor series.
Considering the constraint equation as the spherical constraint equation proposed by Crisfield [3] in the general format:
a = (V 2 + 2 2 2 ) l2 = 0
(5)
where l is the fixed radius of the desired intersection, the
scalar V and are the incremental voltage and load factor,
and the scalar is a scale parameter.
The extended system collecting the equilibrium equation
(4) and the constraint equation (5) can be rewritten as

{(V, )} =

|Fapp (V, )| Fana

(V 2 + 2 2 2 ) l2

0.6
0.4
0.2
0
0

0.5

1.5

Frequency [Hz]

Figure 2: Solution path and the restriction curves

and the correction phase. The predictor procedures are applied to estimate the first solution to determine the direction
to be followed. The corrector iterative procedures are used for
computing a convergence sequence of estimates to obtain the
converged solution.
After converged to the solution (V, ), the acceleration
and force are measured for calculating one point of the FR.
Then the procedure is repeated until the solution path (FR)
is obtained in the specified frequency range.
It is important to notice that on the linearization proce and ,
dure of the arc-length method [6], both terms, V

must be calculate from the residual equation (4). In this work,

both derivatives were calculated experimentally using finite
difference with epsilon of 5 103 . Also in this work, the
scale parameter was adopted as 1.

Experimental Test Rig

A test rig exhibiting a cubic nonlinear behaviour were selected

to examine the performance of the arc-length method for obtaining the nonlinear frequency response.

4.1

Test Rig Model

The Test Rig was made of a continuous system having a

local nonlinear cubic stiffness, as shown in Figure 3. It consists

= {0}.

(6)

For a known position along the solution path, it is possible

to visualize geometrically the equation (6) for a one degree
of freedom system. Figure (2) shows the intersection of both
curves related with equation (6), highlighting the next possible
solutions.
Equation (6) represents a nonlinear system of two unknowns which can be solved by standard incremental-iterative
procedures to obtain the solution point along the path. These
methods usually requires suitable starting values in order to
the iteration procedure converges to correct solution points.
Thus, the procedure that obtains the next equilibrium point
(V, ) normally consists of two phases, the prediction phase

Figure 3: System configuration

of a uniform beam, A, of 420 mm length with a cross section of
12mm by 8mm, which was clamped at one end and supported
at the other end by two clamped-clamped beams, B1 and B2,

dB ref 1 [ms2/N]

of 380mm length with a cross section of 13mm by 1.5mm.

The stiffness of the beam A is linear and the local nonlinear
cubic stiffness is produced by two clamped-clamped beams,
B1 and B2, due to the increase of the longitudinal tension
under large amplitude of vibrations. One accelerometer and
one force-measuring transducer were attached at 6mm from
the free end of the beam A. Both beams, B1 and B2 were
clamped to the beam A at the place of intersection as shown
in Figure 3. They are bolted together on one side by the
force transducer and on the other side by a block mass with
a thread.

50
45
40
35
30
25
45

55

60

65

60

65

Frequency [Hz]
Degrees []

200

150

100

50

0
45

4.2

50

50

55

Frequency [Hz]

Experimental Setup

11
Figure 5: Frequency Response dB H11

The experimental setup consisted of a B&K 4809 shaker connected via a push-rod to a PCB 208A02 force transducer which
was used to measure the input force to the structure. The response was measured using the PCB 353B68 accelerometer
which were attached to the structure using beeswax.
A virtual analyser, was used to obtain RFs of nonlinear structures whith the force being controled. The virtual
analyser consisted of a Pentium 500MHz computer, the Beran 402 Frequency Response Analyser, the Signal Generator
HP33120A and the arc-length method implemented in the
Intelligent Nonlinear Coupling Analysis software (IN CA++ )
[5, 11]. The block diagram of the nonlinear experimental setup
using the virtual analyser, the signal amplifier Kistler 5134,
the oscilloscope HP54501A and the power amplifier B&K 2706
can be seen in Figure 4.

Real [ms2/N]

200

100

100

200
45

50

55

60

65

60

65

Frequency [Hz]
Imag [ms2/N]

300
250
200
150
100
50
0
45

50

55

Frequency [Hz]
11
Figure 6: Frequency Response Real and Imaginary H11

A three dimension plot is used to visualize the frequency

11
response H11
and is shown on figure 7.

Figure 4: Block diagram of the experimental setup test

rig

4.3

Measured RFs using the test rig

The measurement consists of obtaining the solution path

of the assembled structure of figure 3 with the nonlinear cubic
stiffness elements.
The arc-length method was used to obtain the solution
path and the initial equivalent radius used was 0.07Hz. The
11
amplitude and phase of the frequency response H11
for force
control of 1.95N is shown on figures 5 and 6.

CONCLUDING REMARKS

Application of the arc-length method for obtaining frequency

response curves of nonlinear dynamic problems has been examined in this paper.
The arc length method as a control technique presented
good results in the performed experiments. On the other
hand, the following difficulties can be mentioned. The first
difficult came from obtaining the derivative from the finite difference. There was a necessity to estimate the right amount
of epsilon that will pertube the sistem in order to have the
experimental derivative. The value was chosen by try and error. The second difficult comes from choosing the appropriate
radius of constraint equation. A large radius caused a jump
in the FR, on the other hand, the smaller radius were limited
directly by the precision of the experimental data obtained by
the measurement equipments.
No particular difficulties has been detected after choosing
the appropriate epsilon for the derivatives and radius.

[10] R. Lin. Identification of the dynamic characteristics of

nonlinear structures. PhD thesis, Imperial College of Science, Technology and Medicine, 1991.

250

Imag [ms2/N]

200

[11] A.P. Prado. Theoretical and experimental investigation

of a mechanical transmission gearshift lever considering
nonlinear effects. Masters thesis, State University of
Campinas - UNICAMP, Computational Mechanics Department, 2002.

150
100
50
0

50
200
70

100
65

60
55

100
50

Real [ms2/N]

200

45

Frequency [Hz]

11
Figure 7: Frequency Response H11
- three dimension plot

ACKNOWLEDGEMENTS

The authors would like to express gratitude to Fundaca

o de
Amparo `
a Pesquisa do Estado de S
ao Paulo-FAPESP, Conselho Nacional de Desenvolvimento Cientfico e Tecnol
ogicoCNPQ and Universidade Estadual de Campinas-UNICAMP
for providing the financial support.

[12] E. Ramm. Strategies for tracing the nonlinear response

near limit points. International Journal of Solids &
Structures, (15):529551, 1979.

REFERENCES

[1] S.A. Billings and K.M. Tsang. Spectral analysis for

nonlinear systems, part I: Parametric nonlinear spectral analysis. Mechanical Systems and Signal Processing,
3(4):319339, 1989.
[2] S.A. Billings, K.M. Tsang, and G.R. Tomlinson. Spectral
analysis for nonlinear systems, part III: Case study examples. Mechanical Systems and Signal Processing, 4:321,
1990.
[3] M.A. Crisfield. Nonlinear inite Element Analysis of
Solids and Structures, volume 1,2. John Wiley and Sons,
1997.
[4] D.J. Ewins. Modal Testing: Theory and Practice. Research Studies Press, 1984.
[5] J.V. Ferreira. Dynamic Response Analysis of Structures
with Nonlinear Components. PhD thesis, Imperial College of Science, Technology and Medicine, 1998.
[6] J.V. Ferreira and A. L Serpa. Application of the arclength method in nonlinear frequency response. In 20th
IMAC, pages 13491355, 2002.
[7] A. Frachebourg. Identification of nonlinearities: application of the volterra-model to discrete mdof systems;
possibilities and limitts. In 9th IMAC, pages 10291035,
1991.
[8] S.J. Gifford and G.R. Tomlinson. Recent advances in the
application of functional series to nonlinear structures.
Journal of Sound and Vibration, 135(2):289317, 1989.
[9] G.V. Groll and D.J. Ewins. The harmonic balance
method with arc-length continuation in rotor/stator
contact problems. Journal of Sound and Vibration,
241(2):223233, 2001.

[13] E. Riks. An incremental approach to the solution of snapping and buckling problems. International Journal of
Solids & Structures, (15):529551, 1979.
[14] K.Y. Sanliturk, M. Imregum, and D.J. Ewins. Harmonic
balance vibration analysis of turbine blades with friction
dampers. ASME Journal of Vibrations and Acoustics,
119:96103, 1997.
[15] D.M. Storer and G.R. Tomlinson. Recent developments
in the measurement and interpretation of higher order
transfer functions from nonlinear structures. Mechanical
Systems and Signal Processing, 7(2):173189, 1993.
[16] T. Vinh and H. Liu. Extension of modal analysis to nonlinear systems (possibility, mathematical models, limitation). In 7th IMAC, pages 13791385, 1989.