International Journal of Non-Linear Mechanics 40 (2005) 27 – 48

www.elsevier.com/locate/nlm

Detection and description of non-linear phenomena in

experimental modal analysis via linearity plots
Dennis Gögea,∗ , Michael Sinapiusa , Ulrich Füllekruga , Michael Linkb
a Institute of Aeroelasticity German Aerospace Center, Deutsches Zentrum für Luft- und Raumfahrt, Bunsenstr. 10,
D-37073 Göttingen, Germany
b Department of Civil Engineering, University of Kassel, Mönchebergstraße 7, D-34109 Kassel, Germany

Received 29 September 2003; received in revised form 11 May 2004; accepted 13 May 2004

Abstract
Ground vibration tests (GVTs) on aircraft prototypes are mainly performed to experimentally identify the structural dynamic
behaviour in terms of a modal model. This assumes a linear dynamic behaviour of the structure. However, in the practice of
ground vibration testing it is often observed that structures do not behave in a perfectly linear manner. Non-linearities can
be determined, for example, by free play in junctions, hydraulic systems in control surfaces, or friction. This paper compiles
measured, typical, non-linear phenomena from various GVTs on large aircraft. The standard procedure in GVTs nowadays
is the application of the Harmonic Balance method which linearizes the dynamic behaviour on the level of excitation. The
procedure requires a harmonic excitation of the structure which is usually performed during phase resonance testing. The
non-linear behaviour is investigated in terms of linearity plots in which the resonance frequency of a mode is plotted as a
function of the excitation level. The experimental data is then compatible with all post-processing procedures for the measured
results, e.g. updating of the finite element model or flutter calculations. This paper shows measured linearity plots for some
typical non-linear phenomena. In the second part of the paper analytical linearity plots for different non-linear stiffness and
damping models are considered in order to investigate whether the type of non-linearity can be identified from measured
linearity plots. The analytical linearity plots are discussed with respect to their application limits. The analytical linearity plots
are used to interpret the experimental linearity plots stemming from various GVTs on different aircraft prototypes. Finally,
the observability of non-linear stiffness and non-linear damping characteristics via linearity plots is assessed.
䉷 2004 Elsevier Ltd. All rights reserved.

Keywords: Ground vibration testing; Linearity plot; Harmonic balance

1. Introduction
∗ Corresponding author. Tel.: +49-551-709-2321; fax: +49-551-
709-2862.
In the practice of modal testing of large aircraft (test
E-mail addresses: [email protected] (D. Göge),
[email protected] (M. Sinapius), [email protected] set-up in Fig. 1) it is frequently observed that structures
(U. Füllekrug), [email protected] (M. Link). do not behave linearly as mathematically modelized.

0020-7462/$ - see front matter 䉷 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijnonlinmec.2004.05.011
28 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48

2. Experimental check of linearity

Weak non-linearities can be described in the equa-

tion of motion by means of a non-linear restoring func-
tion
k (u) and a non-linear damping function
c (u̇)

mü +
c (u̇) +
k (u) = p(t), (1)

written here for a single-degree-of-freedom system.

The Harmonic Balance, which assumes that most of
the response energy of a harmonically excited struc-
ture is concentrated in the frequency of excitation,
looks for an equivalent linear stiffness keq and equiv-
alent linear viscous damping ceq
Fig. 1. Airbus aircraft during ground vibration testing.
keq u(t) =
k (u), (2)

However, linearization is very common in experimen- ceq u̇(t) =
c (u̇). (3)

tal structural dynamics identification. Linearization in
The assumption of the Harmonic Balance is that the
this case is the determination of the modal parame-
harmonically excited structure mainly responds with
ters for a certain load level. Mainly two linearization
the frequency of excitation
alternatives are commonly applied.
Firstly, linearity can be checked by means of the u(t) ≈ uamp · sin t. (4)
estimated frequency response functions (FRFs). The
FRFs must be independent on the level and type of This leads to a non-linear restoring function
excitation in the case of linearity. However, the FRFs
k (uamp · sin t) and non-linear damping function
vary with the load level in the presence of non-linear
c (uamp ·  cos t), respectively, which can be ex-
structural behaviour. The FRFs also depend on the panded as a Fourier series. An equivalent linear
waveform of excitation in this case [1]. The violation stiffness and damping, depending on the amplitude
of homogeneity and reciprocity of the FRFs are typical of vibration uamp , is derived from the Fourier series.
indications of non-linearities. Now equivalent modal parameters can be defined for
Secondly, the investigation of the structural re- each level of vibration
sponses to harmonic excitation is common in the

ground vibration tests of large aircraft. Harmonic r,eq = keq /m, (5)
distortion of the responses indicates the presence r,eq = ceq /(2r,eq m), (6)
of non-linearities. A detailed investigation of the
non-linearities during a GVT today is based on the with r,eq as the eigen angular frequency and r,eq as
Harmonic Balance method. The Harmonic Balance the viscous modal damping value. These amplitude-
method [2] is technically realized by the phase reso- dependent parameters can be represented in so-called
nance method (sine dwell testing) where the structure linearity plots where the vibration amplitude uamp is
is harmonically excited by means of an excitation plotted as a function of the excitation force and the
force pattern appropriated to a single mode of vi- resonance frequency fr,eq is a function of the vibration
bration. The structure is excited on different force amplitude uamp . The amplitude versus force plot is re-
levels, which results in linearity plots. These plots ferred to as amplitude linearity in this paper, whereas
exhibit the dependency of the investigated resonance the resonance frequency versus amplitude plots is re-
frequency on the load level. This procedure provides ferred to as frequency linearity. The extension from
more detailed information about the non-linear char- the single-degree-of-freedom system to multi-degree-
acteristics, whereas a linearity check by means of FRF of-freedom systems is possible as long as a force ap-
estimation is able to detect structural non-linearities. propriation to a single mode of vibration is possible,
 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48 29

40 1
random
sweep
35 0.8

30
0.6

H/H max
f [Hz]

25
0.4
20

0.2
15

10 0
0 0.5 1 1.5 2 1 1.5 2
u [mm] f/f r,min
amp,max

Fig. 3. FRF from engine excitation.

Fig. 2. Linearity plot of Pilatus PC-21 aircraft.

which at least can be assumed for weak non-linearities.

formed in the past decade. The non-linear behaviour
The resonance frequency is plotted in this case as a
that can be observed in the various linearity plots and
function of the maximum vibration amplitude. The
measured FRFs presented in this section occur only
maximum vibration amplitude, in turn, is plotted as a
because of structural non-linearities and not because
function of the excitation energy, which is represented
of test inaccuracies. Fig. 3 exibits two measured FRFs
by the generalized force
from an engine excitation. Two different types of exci-
nx
 tation signals are compared in the figure: random and
pamp,gen = pamp,i · uamp,i /uamp,max , (7) swept-sine. The FRFs correlate quite well for the first
i=1 two resonance peaks, but differ significantly for range
where nx is the number of exciters. Relating the vibra- 1.3 < f/fr,min < 1.8. The two peaks in this range shift
tion amplitude to the generalized force enables modal downwards with the excitation level, which is signif-
data from different excitation force patterns to be com- icantly higher in swept-sine excitation. A detailed in-
pared and combined in linearity plots. A method for vestigation by means of harmonic excitation yields
deriving linearity plots from swept-sine testing was re- more information about the type of non-linearity re-
cently proposed [3,4]. Fig. 2 shows typical deviations lated to a certain mode shape.
from structural dynamics linearity for specific modes Fig. 4 investigates the linearity of the engine yaw
for the Pilatus PC-21 aircraft [5]. Here, the resonance motion in an engine configuration without the thrust
frequency is plotted as a function of the amplitude. It reverser being installed. The mode was appropriated
can be observed that most of the modes behave lin- on four different load levels that are related to the
early or nearly linearly (dotted curves). Nevertheless, maximum load in the plots. The deviation from linear-
some modes show a significant deviation from struc- ity is accentuated in the lower part of the plot by the
tural dynamics linearity (solid curves). dashed linear curve. The upper part of Fig. 4 shows
the dependency of the resonance frequency on the
vibration amplitude. The resonance frequency varies
3. Typical deviations from structural dynamics with the load level by 5%, a significant decrease from
linearity a low force level and, again, an increase in higher load
levels.
This section compiles typical, measured, non-linear A drop of the resonance frequency with the load
phenomena from various GVTs on large aircraft per- level down to a saturation can frequently be observed
30 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48

1 1

0.98

f/f max
f/fmax

0.95
0.96

0.94 0.9
0 0.5 1 0 0.5 1
uamp/uamp,max uamp /uamp,max
1
1
uamp /uamp,max
amp,max

0.5
0.5
/u
amp
u

0
0 0 0.5 1
0.5 1 pamp /pamp,max
p /p
amp amp,max
Fig. 5. Linearity plot for an engine roll mode.
Fig. 4. Linearity plot for an engine yaw mode.

for non-linear phenomena in the vibration testing of examples are shown in Figs. 6 and 7 that exhibit the
aircraft. This is emphasized for an engine roll mode linearity plots for the rudder rotation and elevator ro-
in Fig. 5 which is appropriated on ten different load tation from different ground vibration tests. Both res-
levels. The main difference to the non-linearity shown onance frequencies vary with the load level by nearly
in Fig. 4 can be discovered in the force amplitude 15% (upper part of the figures), which is pretty much.
plot (shown in the lower part of the figures). The vi- The lower parts of the plots, which depict the depen-
bration amplitude varies cubically with the load level dency of the vibration amplitude on the load level,
in Fig. 4 whereas the variation for the roll mode can show conspicuous differences between the deviation
be characterized as quadratic (Fig. 5). The non-linear from the linearity of both modes. Additional investi-
effects of both engine modes investigated here stem gation, which is not included in the Harmonic Balance
from the engine attachment. However, it should be measurement, is needed to explain the different non-
noted here that the Harmonic Balance does not di- linearities. A completely different mechanism causes
rectly reveal the non-linear restoring function
k (u) non-linearities in the yaw mode of the horizontal
in Eq. (1). tail plane. Fig. 8 depicts the measured linearity plots
A different source of non-linearities is known and evaluated from seven different load levels. Again, the
frequently investigated for control surface modes. Two deviation from linearity is accentuated in the lower
 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48 31

1 1

0.95
0.95

f/fmax
f/fmax

0.9
0.9

0.85
0 0.5 1 0.85
uamp /uamp,max 0 0.5 1
uamp /uamp,max
1 1
uamp /uamp,max

uamp /uamp,max

0.5
0.5

0
0 0.5 1 0
0 0.5 1
pamp /pamp,max
pamp /pamp,max

Fig. 6. Linearity plot for a rudder rotation mode.

Fig. 7. Linearity plot for an elevator rotation mode.

part of the plot by the dashed curve. The non-linearity was hardly affected by the vibration amplitude.
is mainly characterized by a jump phenomenon that This is illustrated in Fig. 9 which shows a slight
causes two different vibration amplitudes at nearly the drop of the resonance in the upper part. The
same load level of about 0.25. This jump phenomenon mode shape itself consists of the in-plane move-
is also conspicuously visible in the upper part of ment x and a bending contribution z of the
the linearity plot. It indicates a sudden change in wing. The lower part of the plot depicts the ra-
stiffness. tio x /z of both components for the five dif-
All the examples of non-linear phenomena in ferent load levels given in the scaled generalized
ground vibration testing on large aircraft that have forces p.
been expounded so far are characterized by resonance A second example of mode shape dependency is
frequencies that vary with the load level. However, given in Fig. 10. The linearity of a lateral engine mode
structural non-linearities may also effect the mode shape is investigated in the plot. The mode itself is
shapes. This kind of non-linearity is elucidated by characterized by unsymmetric amplitudes on the left
two examples. and right side of the aircraft. Again, the resonance fre-
The in-plane bending mode shape of the wing quency is hardly affected by the vibration amplitude.
changed significantly with the load level dur- This is visible in the upper part of the plot. The mode
ing a GVT whereas the resonance frequency shape is traced by means of the amplitudes of the
32 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48

z

x

1 1

0.99

f/fmax
0.98 0.95
f/fmax

0.97

0.96 0.9
0 0.5 1 0 0.5 1
uamp /uamp,max p amp /p amp,max
1 1
uamp /uamp,max

0.8
φx /φz

0.5
0.6

0 0.4
0 0.5 1 0 0.5 1
pamp /pamp,max p amp /p amp,max

Fig. 8. Linearity plot for a HTP yaw mode. Fig. 9. Mode shape dependency of an in-plane wing bending mode.

wing tips on left side (l ) and the right side (r ). The Linearity plots are calculated and assessed with re-
lower part exhibits the ratio of the wing tip amplitudes spect to their limits of application [6].
l /r , indicating a considerably changing unsymme-
try with the load level. 4.1. Harmonic balance and direct integration in the
Consequently, the generalized masses change with time domain
the changing mode shapes in both cases.
A single degree-of-freedom (SDOF) system with
different non-linear spring characteristics and non-
4. Analytical characterization of non-linear linear damping functions is used to create analytical
phenomena linearity plots. The physical parameters of the sys-
tem are: m = 1.5 kg (mass) and c = 0.8 Ns/m in the
Measured linearity plots do not directly reveal the case of viscous damping. For the system with elasto
non-linear function, which in turn, leads to a physical slip spring the parameters are: m = 1.0 kg (mass)
interpretation of the source of the non-linearity. Thus, and c = 1.0 Ns/m in the case of viscous damping.
different non-linear restoring functions and non-linear The parameters are chosen in such a manner that
damping models are taken from the literature in or- the typical characteristics of the respective non-linear
der to reproduce the measured non-linear phenomena. function can be observed via linearity plots. It is im-
 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48 33

nance frequencies and amplitudes for all other cases.

l Moreover, the direct integration in time domain yields
more detailed insight into the acting mechanisms. The
maximum amplitude of the response is used to create
the analytical linearity plots. The responses of the
non-linear system are obtained from very slow nar-
rowband swept-sine excitation, which evokes nearly
r the same maximum amplitudes as those for harmonic
excitation as long as the system is lightly damped [6].
In order not to miss maximum amplitudes in presence
1 of the jump phenomena, upward and downward sweep
is investigated in each case. The excitation frequency
of the maximum response is used as the estimation
f/fmax

0.95 for the resonance frequency of the linearized system

whenever an analytical Harmonic Balance solution
is not available. The maximum amplitude in the time
0.9 domain is calculated by
0.5 1
pamp/pamp,max
uamp,max = 0.5 · (| max(u(t))| + | min(u(t))|). (9)
0.3

0.2
Fig. 12 shows the typical displacement response to
a narrowbanded swept-sine excitation with a hard-
φ l /φ r

0.1
ening cubic stiffness (so-called Duffing Oscillator).
The SDOF system with different non-linear spring
0 characteristics and non-linear damping elements is in-
0.5 1 vestigated next with respect to its observability in the
pamp/pamp,max
linearity plots. The plots are scaled to the maximum
Fig. 10. Mode shape dependency of an unsymmetric engine mode.
values of amplitude, force, and resonance frequency.

4.2. Simulation of linearity plots for piecewise linear

portant not to oversize the non-linear parameters in restoring functions
order to avoid a chaotic behaviour of the simulated
system. The first step from the linear to the non-linear de-
Some of the most common models of non-linearities scription of the restoring function is the piecewise lin-
are compiled in Fig. 11. Table 1 lists the non-linear ear characterization of the stiffness. A sketch of the
restoring functions and non-linear damping func- restoring function is shown in Fig. 13. The restoring
tions. The table includes the parameters that are used function of each linear segment having the stiffness ki
for the determination of all the analytical linearity is described by
plots shown in this paper. The Harmonic Balance
solution is taken in all cases where it is available
k,i = pi + ki · u (10)
analytically. For those cases the equivalent forces are
calculated by and contributes to the equivalent linear stiffness by
 i
pamp,eq = (−m2 + i ceq + keq ) · uamp . (8) 1
keq,i = (pi + ki · uamp sin )
uamp i
Data from the direct integration in the time domain of
the system are used for the determination of the reso- ×sin  d. (11)
34 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48

Cubic stiffness Cubic stiffness Piecewise Pre-loaded piecewise

(softening) (hardening) linear stiffness linear stiffness

Force Force Force Force k3

k2
k2 k1
k1 g
-b b b

Displacement Displacement Displacement Displacement

k1
k1
k2

Off - center Bilinear stiffness

Clearance Bilinear stiffness with offset
clearance

Force Force Force Force

k2 k2 k2 k2

-b k1 k1 -b k1
b g b
Displacement Displacement Displacement Displacement
k1
k2 k1
k2
k1

Arctan stiffness Arctan stiffness Non-linear damping

Saturation (positive amplification factor α) (negative amplification factor α) (here: quadratic damping)

Force Force Force Force

k2 β >>0
β >>0
k1
-b b β>0 β>0

Displacement Displacement Displacement Velocity

k1
k2

Static friction model Dynamic friction model

Coulomb friction (here: elasto slip model) (here: Valanis model)
Et
Force Force Force
σ0
k0+k1
E0
-u+ u+
Displacement Displacement
Velocity
k0+k1 κ< 0.99

κ=0.99
E0
k0

Fig. 11. Idealized types of structural non-linearities.

 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48 35

Table 1
Investigated restoring and damping forces for the SDOF system

Non-linearity
c (u̇)
k (u) Parameters

Piecewise linear stiffness

Piecewise cu̇ k2 u + (k1 − k2 )b, u > b |u| < 0.04 m :
linear stiffness k1 u, |u| < b k1 = 5 × 103 N/m
k2 u − (k1 − k2 )b, u < − b |u| > 0.04 m :
k2 = 7 × 103 N/m
Pre-loaded piecewise cu̇ k2 u + (k1 − k2 )b, b < u < g b = 0.05 m;g = 0.10 m
linear stiffness k1 u, u < b k1 = k3 = 9 × 103 N/m
k3 u + (k1 − k2 )b + (k2 − k3 )g, u > g k2 = 5 × 103 N/m
Bilinear stiffness cu̇ k1 u, u < 0 k1 = 9 × 103 N/m
k2 u, u > 0 k2 = 6 × 103 N/m
Clearance cu̇ k2 u − k2 b, u > b u > 0.2 m ∨ u < − 0.2 m :
k = 0, |u| < b k2 = 6 × 103 N/m
k2 u + k2 b, u < − b −0.2 m < u < 0.2 m :
Off-center clearance cu̇ k2 u − k2 g, u > g u > 0.2 m ∨ u < − 0.4 m :
k = 0, u < g ∨ u > − b k2 = 6 × 103 N/m
k2 u + k2 b, u < − b −0.4 m < u < 0.2 m :
Bilinear stiffness cu̇ k1 u, u < b k1 = 9 × 103 N/m
with offset k2 u + (k1 − k2 )b, u > b k2 = 12 × 103 N/m
b = 0.05 m
Saturation cu̇ k1 u, |u| < b k1 = 5 × 103 N/m
k1 b + k2 u, |u| > b k2 = 0 N/m
b = 0.04 m

Polynomial and trignometrical stiffness

Cubic stiffness, cu̇ ku + knolin u3 k = 6 × 103 N/m
softening knolin = −4 × 106 N/m3
Cubic stiffness, cu̇ ku + knolin u3 k = 6 × 103 N/m
hardening knolin = 7 × 106 N/m3
Arctan stiffness, cu̇ ku + · arctan( · u) k = 6 × 103 N/m3
pos. = 1; = 40
Arctan stiffness, cu̇ ku − · arctan( · u) k = 6 × 103 N/m3
neg. = −1; = 40

Nonlinear damping
Quadratic damping cu̇ + cnolin u̇|u̇| ku k = 6 × 103 N/m
cnolin = 8 Ns/m2
Coulomb friction cu̇ + Nf sgn(u̇) ku k = 6 × 103 N/m
Nf = 0.85 N
Elasto slip model cu̇ Fges = k0 u + dF es
dt (Eq. (29)) k0 = 100 N/m; = 0.3
k1 = 80 N/m; s = 10; Nf = 40 N
dFV
Valanis model cu̇ dt (Eq. (30)) k0 = 6 × 105 N/m; E0 = 2.2 × 106 N/m
Et = 1.4 × 105 N/m; 0 = 180 N; = 0.7

pi is the preload defined by all previous linear seg-  and  are angles derived from the boundaries of the
ments linear segment.

pi
i = sin−1 (ai /uamp ), 0
|i | < 2,
i−1

= (kn − ki )(bn − an ), ∀ (bi · an > 0). (12)
n=1 i = sin−1 (bi /uamp ), 0 < |i |
2. (13)
36 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48

SDOF-system with hardening cubic stiffness 1

0.02

0.015 0.95

f/fmax
0.01
0.9
0.005
0.85
u [m]

0 0 0.2 0.4 0.6 0.8 1

uamp /uamp,max
-0.005 1

uamp /uamp,max
-0.01

-0.015 0.5

-0.02
0 200 400 600 800
t [sec] 0
0 0.2 0.4 0.6 0.8 1
pamp /pamp,max
Fig. 12. Displacement response of the SDOF system.

Fig. 14. Linearity plot of the SDOF system with piecewise linear
stiffness.
1
4.2.1. Symmetric piecewise linear stiffness
ki The symmetric piecewise linear system having
pi two different stiffnesses is taken as a first example.
The spring characteristic of the system is shown in
Fig. 11. The linearized equivalent stiffness evaluated
by the Harmonic Balance is [7]
0    
−1 b
keq =k1 1 − 2 sin−1
 uamp
b  2
+ 2 uamp − b2 (15)
uamp

-1 which is valid for amplitudes where the system be-

-1 0 ai 1
bi comes non-linear, i.e. the transition from the stiff-
ness k1 to the stiffness k2 = k1 . The piecewise linear
Fig. 13. Restoring function of system with piecewise linearity.
stiffness is easily observable via linearity plots, see
Fig. 14. The stiffness characteristic is quite visible in
the frequency linearity. There is no change for the res-
The equivalent linear stiffness, which depends on the onance frequency with increasing amplitudes for low
vibration amplitude uamp , consists of a sum of all N excitation forces (k1 ). The system behaves linearly. If
linear segments the amplitudes of the system comes in the region of
stiffness k2 (high excitation forces), the resonance fre-
1 N  quency increases in a nonlinear manner. This effect
keq = k i u2amp (i − i )
2u2amp occurs because the amplitudes pass through both stiff-
i=1
  nesses k1 and k2 .
2pi  2
+ ai + uamp − ai2
ki 4.2.2. Pre-loaded piecewise linear stiffness
  
2pi  2 An example that is important for practical appli-
− bi + uamp − bi .
2 (14)
ki cation is the system with pre-loaded piecewise linear
 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48 37

1 1

0.98
f/fmax

f/fmax
0.95
0.96

0.94
0 0.2 0.4 0.6 0.8 1 0.9
uamp /uamp,max 0.2 0.4 0.6 0.8 1
uamp /uamp,max
1
1
uamp /uamp,max

uamp /uamp,max
0.5
0.5

0
0 0.2 0.4 0.6 0.8 1
pamp /pamp,max 0
0 0.2 0.4 0.6 0.8 1
pamp /pamp,max
Fig. 15. Linearity plot of the SDOF system with pre-loaded piece-
wise linear stiffness.
Fig. 17. Linearity plot of the SDOF system with bilinear stiffness.

SDOF-system with pre-loaded piecewise linear stiffness

phenomenon is that the displacements reside in the
0.8 p/p
max
= 0.75 region of the stiffnesses k2 (p/pmax = 0.31), where
p/p = 0.31
0.6 p/p
max
max
= 0.025 the stiffness of the system is lower than in the stiff-
ness regime k1 . When increasing the excitation forces
0.4
(p/pmax = 0.75), the system vibrates anew in the
0.2 stiffness regime k3 = k1 with higher stiffness, which
u [m]

0 ends in increasing resonance frequencies.

-0.2
4.2.3. Bilinear stiffness
-0.4
The bilinear stiffness non-linearity is an exceptional
-0.6 case of the piecewise linear stiffness. The stiffness
-0.8 changes in the zero displacement position. If k1 is the
120 140 160 180 200 220 240 260 stiffness for negative displacements and k2 the stiffness
t [sec] for the positive displacements, the linearized equiva-
lent stiffness for the Harmonic Balance simply repre-
Fig. 16. Displacement responses of the SDOF system with sents the average stiffness of the system [1]:
pre-loaded piecewise linear stiffness.
keq = 0.5 · (k1 + k2 ). (16)

stiffness, whose spring characteristic is shown in The system has an effective natural frequency and a
Fig. 11. As for the system with piecewise linear stiff- FRF which is independent of the response amplitude
ness, the pre-loaded piecewise linear stiffness is easily and the excitation force level. This can clearly be seen
observable via linearity plots, see Fig. 15. For low in the linearity plots for the system in Fig. 17. The
excitation forces the resonance frequency is constant non-linear system seems to be linear. However, the
because the system is linear (u(t) < b, p/pmax = system behaves non-linearly which can be seen in the
0.025), see Fig. 16. The resonance frequency dwin- displacement response in the time domain as shown in
dles with increasing displacements for higher ex- Fig. 18. The displacement response is higher for pos-
citation forces (b < u(t) < g). The reason for this itive displacements than for negative displacements.
38 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48

SDOF-system with bilinear stiffness 1

0.08

f/fmax
0.06
0.5
0.04

0.02
0
0 0.2 0.4 0.6 0.8 1
u [m]

0 uamp /uamp,max
-0.02 1

uamp /uamp,max
-0.04
0.5
-0.06

-0.08
0 25 50 75 100 125 150 0
t [sec] 0 0.2 0.4 0.6 0.8 1
uamp /uamp,max
Fig. 18. Displacement response of the SDOF system with bilinear
stiffness. Fig. 20. Linearity plot of the SDOF system with clearance.

1 (point b in Fig. 11). There are no changes in the reso-

nance frequency if the amplitude is high (far beyond
amplitude b). The system behaves linearly with an av-
f/fmax

0.95 erage stiffness like in the case for the system with bi-
linear stiffness.
0.9
0 0.2 0.4 0.6 0.8 1 4.2.5. Clearance or off-center clearance
u /u
amp amp,max A second exceptional case of the piecewise linear
1 stiffness is represented by the clearance type non-
uamp /uamp,max

linearity (normal and off-center). Both restoring func-

tions are visualized in Fig. 11. The linearity plots of
0.5
the system with clearance and off-center clearance
(Figs. 20 and 21) look the same. When increasing the
0 excitation force, the amplitude rises linearly. The am-
0 0.2 0.4 0.6 0.8 1 plitude linearity looks nearly the same as for a lin-
pamp/pamp,max
ear system except for the starting point of the curve
which is different. The resonances rise for increasing
Fig. 19. Linearity plot of the SDOF system with offset bilinear
stiffness. amplitudes. However, it is not possible to differen-
tiate between both non-linearities via linearity plots.
The experimental analysis with the Harmonic Balance
4.2.4. Bilinear stiffness with offset method cannot reveal the off-center clearance non-
The force deflection curve of the system with bi- linearity because the response of the system with off-
linear stiffness with offset can be observed quite well center clearance does not oscillate around the center
in the linearity plots (Fig. 19) in contrast to the bilin- axis, which can be seen in Fig. 22 where it exhibits
ear stiffness. In the case of low amplitudes the system the time domain response.
behaves linearly. The linear raising of the amplitudes
with the force level does not affect the resonance fre- 4.2.6. Stiffness saturation
quency. The resonance frequency of the system rises The third case of piecewise linearity is the sys-
in the region of the stiffness transition from k1 to k2 tem with stiffness saturation, see Fig. 11, which is a
 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48 39

1 1

0.9

f/fmax
f/fmax

0.5
0.8

0 0.7
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
uamp /uamp,max uamp /uamp,max
1 1
uamp /uamp,max

uamp /uamp,max
0.5 0.5

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
uamp /uamp,max pamp /pamp,max

Fig. 21. Linearity plot of the SDOF system with off-center clear- Fig. 23. Linearity plot of the SDOF system with saturation.
ance.

SDOF-system with saturation (up-sweep)

SDOF-system with off-center clearance 0.1
0.8
p/pmax = 1.0
0.6 p/pmax = 0.2
p/pmax = 0.04
0.4 0.05

0.2
u [m]

u [m]

0 0
-0.2

-0.4
-0.05
-0.6

-0.8
-0.1
-1 0 100 200 300 400 500 600
0 100 200 300 400
t [sec]
t [sec]
Fig. 24. Displacement responses of the SDOF system with satu-
Fig. 22. Displacement response of the SDOF system with off-center
ration.
clearance.

more theoretical than practical example. For low ex- placements now reside in the stiffness regime k2 = 0
citation forces the amplitudes of the system reside in (|u| > b, p/pmax = 0.2). The responses significantly
the stiffness regime k1 (|u| < b). This leads to a lin- change in their characteristics, see Fig. 24. The higher
ear behaviour of the system, which can be observed the excitation force, the earlier the change is from a
in the frequency linearity in Fig. 23. For rising dis- “normal” response peak in time domain to a distorted
placements the resonance frequency remains constant response peak (p/pmax = 1.0). For high forces the
(p/pmax = 0.04). The amplitude rises with increas- system goes out of the linear stiffness regime k1 very
ing excitation force, see Fig. 23. Simultaneously, the quickly. The non-linear response of the system decays
resonance frequency of the system dwindles. The dis- when reaching the stiffness regime k2 = 0.
40 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48

resonance frequency dwindles in the case of soften-

1.2 ing cubic stiffness. In case of hardening cubic stiffness
the amplitude rises with the force level and, simulta-
hardening
1 neously, the resonance frequency rises as well. This
f/fmax

leads to the conclusion that the cubic or polynomial

0.8 stiffnesses are observable via linearity plots.
softening
0.6 4.3.2. Arctan stiffness
0.05 The arctan stiffness has an even more practical rel-
kn 0
evance, as is reported in [8]. The arctan spring has
l /k -0.05 0.5 1
0 a softening stiffness characteristic (Fig. 11) and is
uamp /uamp,max defined by

1
k = k · u + · arctan( · u). (18)

uamp /uamp,max

The factor is an amplification factor and a com-

softening
0.5 pression factor. The difference to the cubic stiffness
hardening is that the stiffness of the arctan spring converges to
± · /2 for large amplitudes, whereas the softening
cubic stiffness spring does not converge. The system
0 is investigated with a linear stiffness in parallel to an
-0.1 arctan stiffness having a positive and negative ampli-
0 fication factor . The equivalent stiffness of the arctan
k nl

0.1 0 0.5 1 stiffness is [9]

/k

/p
pamp am p, m ax
2 
keq = k + u + · ( b2 u2amp + 1 − 1). (19)
Fig. 25. Linearity plot for SDOF system with cubic stiffness. uamp
2

The amplitude linearity is the same for both cases,

4.3. Simulation of linearity plots for polynomial and see Fig. 26. The amplitudes rises linearly with the
trignometric restoring functions force level. The difference between both cases can
be observed in the frequency linearity. The resonance
4.3.1. Polynomial stiffness, cubic stiffness frequency rises to a constant value for the system with
The classical Duffing Oscillator with softening and negative whereas the resonance frequency dwindles
hardening behaviour (Fig. 11) is investigated. This sys- to a constant value for the system with positive .
tem with cubic stiffness is used to represent systems This behaviour is caused by convergence of the force
with so-called polynomial stiffnesses. For real struc- deflection curve of the arctan stiffness to ± · /2.
tures the non-linear function includes not only a cubic
term but also higher-order terms in which the cubic
term may dominate. The equivalent stiffness of the 4.4. Simulation of linearity plots for non-linear
polynomial cubic stiffness is damping
3
keq = k + knolin · u2amp . (17) This section focuses on damping non-linearities.
4 Four different kinds of non-linear damping
The linearity plots for the Duffing Oscillator are shown mechanisms are selected, two of which are based on
in Fig. 25. It can be observed that the amplitude de- velocity-dependent damping. The others are described
creases with the force level and, simultaneously, the by hysteresis displacements.
 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48 41

1.1 1

1.05 HB

f/fmax
hardening Time
0.99
f/fmax

0.95
0.98
0.4 0.6 0.8 1
0.9 softening uamp/uamp,max
0.01 1
0

uamp/uamp,max
k
nl /k 0.01 0.5 1
0
uamp /uamp,max 0.5
HB
Time
1
0
uamp /uamp,max

0.2 0.4 0.6 0.8 1

softening pamp/pamp,max
0.5
hardening Fig. 27. Linearity plot of the SDOF system with quadratic damping.

0
0.02 4.4.2. Coulomb friction
k 0 The Coulomb friction model is a static friction
nl /
k 1 model and is frequently used for the description
0.02 0 0.5
pamp /pamp,max of nonlinearities. The damping model visualized in
Fig. 11 is a signum friction model [10,11]
Fig. 26. Linearity plot of the SDOF system with arctan stiffness.

c (u̇) = Nf sgn(u̇), (22)

where is the friction coefficient and Nf is the nor-

4.4.1. Polynomial damping, quadratic damping
mal load. This type of non-linearity represents any
The most common model of the polynomial non-
situation with interfacial motion and often occurs in
linear damping function is quadratic damping (Fig.
combination with clearance non-linearities. It is most
11) defined by
evident at small vibration amplitudes. The equivalent

c (u̇) = c · u̇ + cnolin (u̇)|(u̇)|, (20) linear damping according to the Harmonic Balance
method is
where cnolin is the non-linear damping coefficient and 4cnolin
u̇ is the structural velocity. The absolute value term is ceq = . (23)
uamp 
introduced in order to ensure that the force is always
opposed to the instantaneous velocity. The equivalent Consequently, the Coulomb friction can only be ob-
linear damping according to the Harmonic Balance served in linearity plots exhibiting the damped reso-
method is nances if the excitation forces are not very high.
8
ceq = c + cnolin uamp . (21) 4.4.3. Static friction model, elasto slip spring
3
A more sophisticated static friction model is the
Consequently, the frequency linearity plot (Harmonic elasto slip model [11] (Fig. 11). For example, the
Balance) shows no deviation of linearity, see Fig. 27. elasto slip model can be used to describe frictional
However, the damped resonance frequency varies with joints. The elasto slip model considers the deforma-
the load level in the time domain, which is shown tion that may occur prior to interfacial slip in contrast
in Fig. 27. The quadratic damping characteristic can to the signum-based friction laws which do not con-
clearly be seen in the amplitude linearity. sider this phenomenon. Fig. 28 depicts the generalized
42 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48

k0 SDOF-system with elasto slip spring

60
p/pmax = 0.5
40 p/pmax = 0.13
p/pmax = 0.003
µ Nf 20
k1 Fes

Fges [N]
0

-20
c
-40

-60
u -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
u [m]

Fig. 28. Generalized elasto slip model. Fig. 29. Force deflection curve of the elasto slip element.

k1
elasto slip model. The friction interface is modeled by ceq =c0 + · (1 − cos 2
an elastic spring in series with an ideal Coulomb el- 2
−1
ement with the threshold force · Nf . The restoring ×(cos (1 − 2 · · N/k1 · uamp ))). (28)
force, which includes energy dissipation in this case, For the sake of simplicity this elasto slip model can
is given by be approximated by the following differential
equation [13]:

k (u, u̇) = k0 · u + R, (24)
  
*Fes 1  · Fes
where k0 is a spring parallel to the elasto slip element = k1 u̇ − · tan , (29)
and *t s 2 · · Nf

 k · (u − u+ ) + · N · sgn(u̇+ ) where s is a constant, which controls the closeness
 1 of the approximation to the signum-function. is the
if |R| < · N,
R= + (25) friction coefficient and Nf is the normal load. The lin-
 N · sgn(u̇ )
 ·
earity plots for the elasto slip model (Fig. 30) are gen-
if u̇ = 0.
erated using the Harmonic Balance method. For very
k1 denotes a spring in series to the elasto slip element low excitation forces the system behaves linearly. The
and u+ the displacement prior to velocity reversal. For reason for this is that for low excitation forces no hys-
low amplitudes (( · N )/k1 · uamp > 1) the restoring teresis is visible in the force deflection plot in Fig. 29,
force of the system is simply which is generated using the approximated differen-
tial equation. The system behaviour is nearly linear.

k (u, u̇) = (k0 + k1 ) · u (26) The hysteresis can be clearly seen in Fig. 29 when in-
creasing the excitation force. The system now behaves
which effects a linear behaviour of the system. The
non-linearly. The amplitudes rises with the excitation
equivalent stiffness and damping of the elasto slip el-
forces in a softening as well as hardening manner. The
ement for high amplitudes is [12]
resonance frequency dwindles simultaneously.
  
k1 2· ·N
keq =k0 + · cos−1 1 − 4.4.4. Dynamic friction model, Valanis model
 k1 · uamp
 The Valanis model, which is a dynamic friction
−1
sin 2(cos (1 − 2 · · N/k1 · uamp ))
− , (27) model, is investigated next [13,14] (Fig. 11). In con-
2 trast to the static friction models, dynamic friction
 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48 43

models are capable of describing the transition be- 1

tween microslip and macroslip. The Valanis model is
described by the following differential equation in- 0.9

f/fmax
cluding an elastic spring (k0 ) parallel to the Valanis
model: 0.8
*FV 0 0.2 0.4 0.6 0.8 1
*t uamp /uamp,max
E0 · u̇ · [1 + E0 · |u̇u̇| · (Et · u − FV )] 1
=k0 · u + , (30)

uamp /uamp,max

1+ · E0 · |u̇u̇| · (Et · u − FV )
0.5
where

• FV is force in the complete system, 0

• is microslip parameter, 0 0.2 0.4 0.6 0.8 1
pamp /pamp,max
•  is a material parameter,
• 0 is an equivalent slip limit, Fig. 30. Linearity plot of the SDOF system with elasto slip spring.
• E0 is the stiffness modulus of stick conditions,
• Et is the tangent modulus for the slope of slip
motion, SDOF-system with Valanis model
• and 300

p/pmax = 1.0
200
E0 p/pmax = 0.33
= Et
. (31) p/pmax = 0.002
0 (1 − · E0 ) 100
[N]

Fig. 11 shows the physical interpretation of the pa-

0
rameters in the description of the Valanis model. The
V
F

linearity plots for the SDOF system with the Valanis

-100
model are shown in Fig. 32. Like in the case of hard-
ening cubic stiffness (Fig. 25) it can be observed that
-200
the amplitude rises with the force level in a polyno-
mial manner and, simultaneously, the resonance fre- -300
quency dwindles in a polynomial manner. There is no 1 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1
u [m]
linear part in the linearity plots for the Valanis model
in contrast to the elasto slip model (Fig. 30). The rea- Fig. 31. Force deflection curve of the Valanis model.
son for this is the choice of parameter being set to
0.7 which effects a non-linear increase of the restor-
ing force from the point of origin, see Fig. 11. The of the structure. Some of the non-linear restoring and
choice of parameters differs from the elasto-slip ele- damping functions presented earlier are combined in
ment, which effects different linearity plots. The force the following section in order to investigate the effect
deflection curve is plotted in Fig. 31. of the combination on the linearity plots. All linearity
plots (Fig. 32) for the combined non-linearities are
4.5. Combining damping and stiffness nonlinearities created using the direct time-integrated SDOF system.

Different kinds of non-linearities may occur simul- 4.5.1. Quadratic damping and cubic stiffness
taneously in engineering practice. The most common (softening)
damping and stiffness nonlinearities appear in the First, the cubic stiffness (softening) and the
joints and connections between different components quadratic damping are combined. The linearity plots
44 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48

1 1

0.98 0.9
f/fmax

f/fmax
0.8
0.96
0.7
0.94
0 0.2 0.4 0.6 0.8 1
uamp/uamp,max 0.6
0 0.2 0.4 0.6 0.8 1
uamp/uamp,max
1
1
uamp /uamp,max

uamp /uamp,max
0.5
0.5

0
0 0.2 0.4 0.6 0.8 1
pamp/pamp,max
0
0 0.2 0.4 0.6 0.8 1
Fig. 32. Linearity plot of the Valanis model. pamp /pamp,max

Fig. 34. Linearity plot of the SDOF system with elasto slip spring
and arctan stiffness.
1

0.995 4.5.2. Elasto slip model and arctan stiffness (pos. )

f/fmax

Finally, a combination of an elasto slip spring with

0.99
an arctan stiffness having a positive amplification fac-
0.985 tor is investigated. The linearity plots for this system
0.4 0.6 0.8 1 are shown in Fig. 34. The elasto slip spring dominates
uamp /uamp,max
in the force amplitude diagram, because the arctan
1
stiffness leads to a linear behaviour in the force am-
uamp /uamp,max

0.8 plitude diagram, see Fig. 26. Consequently, the force

0.6 amplitude plot for the system with elasto slip spring
0.4
and arctan stiffness looks like the plot for the SDOF
system with only an elasto slip spring, see Fig. 30. In
0.2
0.2 0.4 0.6 0.8 1 contrast to the force amplitude diagram, the curve
pamp /pamp,max characteristic in the amplitude resonance frequency
diagram comprises both types of non-linearities. A
Fig. 33. Linearity plot of the SDOF system with quadratic damping change for the resonance frequency can be observed
and cubic stiffness (softening). for low excitation forces in comparison to the sys-
tem with pure elasto slip spring. The reason for this
behaviour is the arctan stiffness. For high excitation
for the system shown in Fig. 33 exhibit the same char- forces the curve in the amplitude resonance frequency
acteristics as the diagram for the system with pure diagram converges, which is caused by the force de-
cubic stiffness (Fig. 25). The amplitude rises with flection curve of the arctan stiffness in Fig. 11.
the force level and, simultaneously, the resonance
frequency dwindles. The polynomial damping non- 4.6. Assessment of observability via linearity plots
linearity has no influence on the resonance frequency.
The stiffness non-linearity dominates the frequency Obviously, it is not possible to clearly distin-
linearity. guish between stiffness non-linearity and damping
 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48 45

Table 2
Observability of non-linearities in linearity plots

Type of non-linearity Observation in linearity plots

Cubic stiffness (softening and hardening) possible / possible
Piecewise linear stiffness possible
Pre-loaded piecewise linear stiffness possible
Clearance (normal and off-center) possible / not possible
Bilinear stiffness (normal and with offset) not possible / not possible
Saturation possible
Arctan stiffness (pos. and neg. ) possible / possible
Quadratic damping possible
Coulomb friction (low and high forces) possible / not possible
Elasto slip model possible
Valanis model possible
Quadratic damping and cubic stiffness (soft. and hard.) not possible / not possible
Elasto slip model and arctan stiffness not possible

non-linearity from linearity plots. Table 2 shows an with bilinear stiffness with offset is also assessed with
assessment of the observability of non-linearities via “not possible” because the force amplitude and the
analytical linearity plots. If a specific type of non- amplitude resonance frequency plot look nearly the
linearity is visible in the analytical linearity plot, it same as the system with piecewise linear stiffness.
is rated with “possible”. If the non-linearity is not Consequently, it is not possible to clearly distinguish
visible in the analytical linearity plot, it is then rated between both of these non-linearities. The observ-
with “not possible”. ability of saturation non-linearity via linearity plots
Systems with polynomial stiffness or damping lead is assessed with “possible”. An important example
to linearity plots where the non-linearity of the sys- for practical application is the arctan stiffness non-
tem is visible. Their observability is assessed with linearity. Its characteristic can be observed very well
“possible”. However, it is not possible to identify the in linearity plots. As a conclusion, the observability
order of the polynom. The system with piecewise lin- of this non-linearity via linearity plots is also assessed
ear stiffness and pre-loaded piecewise linear stiffness with “possible”. The assessment of the system with
can be observed via linearity plots. Both systems can Coulomb friction is inconsistent. The non-linearity is
be clearly observed via the force amplitude diagram. only visible if the excitation forces are low. For high
The restoring force (in the positive area) is represented excitation forces the system with Coulomb friction
in the force amplitude plot. The bilinear stiffness behaves nearly linearly. This is the reason why the ob-
non-linearity is an exceptional case of the piecewise servability of the Coulomb friction via linearity plots
linear stiffness. Its observability via linearity plots is is rated with “possible” as well as “not possible”.
assessed as “not possible” because the analytical lin- The elasto slip model can be observed via linearity
earity plots look like the plot of a linear system. The plots and is assessed with “possible”. The linearity
clearance non-linearity (normal and off-center) can plots of the Valanis model look very similar to the
be observed via analytical linearity plots, presuming plots of the system with saturation. There is only one
that the excitation forces are low enough. A problem significant difference: For low excitation forces the
is that the linearity plots of both systems (with normal system behaviour of the system with saturation is
or off-center clearance) look quite similar. The off- linear, whereas the behaviour of the Valanis model is
center clearance system shows the same behaviour in completely non-linear. Thus, the observability of the
the linearity plot as that of the system with normal non-linearity is rated with “possible”.
clearance. This is the reason why the observability When combining damping and stiffness non-
of the off-center clearance non-linearity is assessed linearities the stiffness non-linearity dominates the
with “not possible”. The observability of the system linearity plots in most cases. This is due to the order
46 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48

of magnitude of the restoring force, which is much

higher than the damping force in most cases. A typical
example is the combination of quadratic damping and
cubic stiffness. Both diagrams look quite similar to
the system with cubic stiffness only. Thus, both non-
linearities are “not possible” to observe in the plots.
An exact statement about the non-linearities can be
made neither for the system with quadratic damping
and piecewise linear stiffness nor for the combination
of the elasto spring element with the arctan stiffness.
Again, the restoring force dominates the structural
dynamics behaviour.
From these assessments it can be concluded that
combined damping and stiffness non-linearities can
hardly be distinguished via linearity plots. Fig. 35. Detail of engine attachment in the aft bearing.

5. Interpretation of the experimental linearity

plots from GVT cause friction may also play an important role in the
attachment. Friction can have only a small effect on
The experimental linearity plots presented in Sec- the resonances but the responses are influenced signif-
tion 3 are interpreted next by comparing them with icantly, especially for low excitation forces. Thus the
the simulated analytical linearity plots. linearity plots may have another shape so that analyt-
ical and experimental linearity plots are not directly
comparable.
5.1. Engine yaw mode

The first example that is investigated is the linearity 5.2. Engine roll mode and the rudder rotation mode
plot of the engine yaw mode in Fig. 4. The engine yaw
mode is severely affected by the engine pylon attach- The linearity plots of both the engine roll mode
ment. The engine is fixed in the middle at the pylon (Fig. 5) and the rudder rotation mode (Fig. 6) look
struts and supported by hinges in the aft part (ball and quite similar, although the source of non-linearity is
socket joint, see Fig. 35). The non-linearity occurs as completely different. The dynamic behaviour of the
a result of the connection elements between pylon and actuator hydraulics is responsible for the non-linear
engine at the aft bearing that have a tolerance (free effects of the rudder rotation mode, whereas the con-
play). The experimental linearity plots of the engine nection elements between pylon and engine at the aft
yaw mode can be compared relatively well with the bearing are responsible for the non-linear characteris-
system that contains pre-loaded piecewise linear stiff- tic of the engine roll mode. The linearity plots of the
ness, Fig. 15. The reason for this is that all excitation modes may be interpreted with a static or dynamic
forces are relatively high. In reality there is no zero friction model (e.g. elasto-slip model). It is clear that
stiffness even for very low excitation forces. The res- especially for the rudder rotation mode the hydraulic
onance frequency for low excitation forces dwindles. systems cause a non-linear damping effect for this
Increasing the force level leads to rising resonance mode. The linearity plots for both modes can be com-
frequencies. This is due to the fact that at low exci- pared quite well with the linearity plots of the elasto-
tation levels the motion of the engine occurs near the slip element, see Fig. 30. The linear behaviour in the
range of the free play (low stiffness). Increasing the analytical linearity plots (uamp /uamp,max < 0.2) cannot
force leads to higher displacements which exceed the be seen in the experimental linearity plots because the
clearance (high stiffness). However, a correct state- excitation forces are relatively high, even for the low
ment on the type of non-linearity cannot be made be- force levels.
 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48 47

5.3. Elevator rotation mode

Like for the rudder rotation mode, the dynamic be-

haviour of the actuator hydraulics is responsible for
the non-linear effects of the elevator rotation mode.
The linearity plots of the elevator rotation mode
(Fig. 7) can also be compared quite well with the
analytical linearity plots from the SDOF system with
elasto slip element, see Fig. 30. The analytical and
experimental force amplitude diagrams show a good
agreement. The amplitudes increase for increasing
excitation forces. Both curves show a softening and
a hardening effect. The resonances dwindle down for
increasing amplitudes in a polynomial manner for
both cases.

5.4. HTP yaw mode

The HTP yaw mode linearity plot Fig. 8 correlates

quite well with the linearity plot of the system with
pre-loaded piecewise linear stiffness, Fig. 15. This is
due to the connection between the HTP and the fuse-
lage, see Fig. 36. A rubber is installed between the
HTP and fuselage. The motion occurs in the toler-
ance region (rubber = low stiffness) if the HTP is ex-
cited in-plane by low forces. If the force increases,
the HTP goes out of the tolerance region and actuates
against the fuselage (rubber+aluminium = high stiff-
ness). This transaction can be observed in the experi-
mental and analytical amplitude resonance frequency
linearity plot. The resonance frequency dwindles as Fig. 36. Detail of HTP fuselage connection.
long as the excitation is low. If the motion of the HTP
exceeds the region of tolerance (high forces), the reso- (Fig. 27) cause these changes in the mode shape. The
nances increase. The same phenomenon can be seen in unsymmetries are often a consequence of material
the force amplitude plot. For low excitation forces the variances. Non-linear stiffness can also lead to cross
amplitudes increase linearly (low stiffness). A sudden coupling terms between investigated modes shapes.
change in stiffness has the effect that the amplitudes Stiffness non-linearities can have a significant effect
rise linearly for high force levels but with a different on the frequencies. Thus, different modes can be
gradient. shifted together so that motions of different modes
are present in a single mode of vibration.
5.5. Mode shape dependencies

Finally, the in-plane bending mode shape in 6. Consequences for future work and
Fig. 9 and the unsymmetric engine mode in Fig. 10 are investigations
investigated. The mode shapes change significantly
with increasing force levels. Simultaneously, the reso- The most common types of non-linearities found in
nance frequency dwindles just a bit. Non-proportional the literature and which are listed in Fig. 11 can only
or non-linear damping like quadratic damping give an idea of what type of non-linearity is present in
48 D. Göge et al. / International Journal of Non-Linear Mechanics 40 (2005) 27 – 48

a specific experimental mode of vibration. The exper- [2] N.N. Bogoljubov, J.A. Mitropolsky, Asymptotische Methoden
imental linearity plots give helpful information about in der Theorie der nichtlinearen Schwingungen, Akademie
deviations from linearity, but there is the need for a Verlag, Berlin, 1965.
[3] G. Gloth, M. Sinapius, Swept-sine excitation during modal
method which directly leads to the characteristic of the identification of large aerospace structures, Forschungsbericht
non-linear functions. A new kind of linearity plot that DLR-FB 2002-18, DLR-lnstitut für Aeroelastik,
is able to reveal the different kinds of non-linearities 2002.
is being developed. This approach is based on [15]. [4] G. Gloth, M. Sinapius, Detection of non-linearities in swept-
sine measurements, in: Proceedings of 21st Int’l Modal
Analysis Conference (IMAC),2003, p. 127.
[5] A. Vollan, M. Degener, G. Gloth, Validation testing
7. Summary and conclusions for flutter certification of the new high-performance
trainer aircraft pc-21, in: Proceedings of the International
It is often observed in the practice of GVT that the Forum on Aeroelasticity and Structural Dynamics (IFASD),
structure does not behave as linearly as is assumed 2003.
in the methods of experimental modal analysis. This [6] D. Göge, Interpretation von aus standschwingungsversuchen
ermittelten impedanzplots, Internal Report DLR-IB 232-2002
paper reviews the state of the art of detecting non- J 04, DLR-lnstitut für Aeroelastik, 2002.
linearities during GVT. It compiles typical non-linear [7] G.R. Tomlinson, J. Lam, Frequency response characteristics
phenomena from various GVT. Analytical linearity of structures with single and multiple clearance-type non-
plots are investigated in order to characterize and linearity, J. Sound Vibration 96 (1) (1984) 111–125.
identify the type of the measured non-linearties. The [8] S. Meyer, M. Link, Local non-linear softening behaviour:
modelling approach and updating of linear and non-
analytical linearity plots are created using a single- linear parameters using frequency response residuals, in:
degree-of-freedom (SDOF) system with different Proceedings of 21st Int’l Modal Analysis Conference
types of non-linear restoring functions and non-linear IMAC,2003.
damping models (Harmonic Balance and/or direct [9] S. Meyer, Modellbildung und Identifikation von lokalen
time domain data is used). An assessment of the lin- nichtlinearen Steifigkeits- und Dämpfungseigenschaften in
komplexen strukturdynamischen Finite Elemente Modellen,
earity plots with respect to their application limits is Ph.D. Thesis, University Press Kassel, 2003.
given. The analytical linearity plots are then used to [10] C.A. Coulomb, Theorie des Machines Simples, Memoirs
interpret the experimental linearity plots from GVTs de Mathematique et de Physics de I’Academic Royale,
on different aircraft. Some experimental linearity 1785.
plots could be interpreted using the analytical ones. [11] L. Gaul, R. Nitsche, The role of friction in mechanical joints,
Appl. Mech. Rev. 54 (2) (2001) 93–106.
This paper concludes from this comparison that the [12] L. Gaul, R. Nitsche, Dynamics of structures with joint
linearity plots simulated with the most common types connections, in: D.J. Ewins, D.J. Inman (Eds.), Structural
of non-linearities can only give an idea which type Dynamics @ 2000—Current Status and Further Directions,
of non-linearity is present in a specific experimental Research Studies Press 2001, pp. 29–48.
mode shape. Further work is needed to better interpret [13] E.P. Petrov, D.J. Ewins, Models of friction damping with
variable normal load for time-domain analysis of vibrations,
and distinguish the different non-linear phenomena in: Proceedings of the Int’l Conference on Noise and Vibration
detected during the GVTs of large aircraft. Engineering (ISMA), ISMA,2001.
[14] J. Lenz, L. Gaul, The influence of microslip on the dynamic
behaviour of bolted joints, in: Proceedings of 13th Int’l Modal
References Analysis Conference (IMAC-XIII), SEM,1995.
[15] S.F. Masri, T.K. Caughey, A non-parametric identification
[1] K. Worden, G.R. Tomlinson, Nonlinearity in Structural technique for nonlinear dynamic problems, J. Appl. Mech.
Dynamics, Institute of Physics Publishing, 2001. 46 (1979) 433–447.