Date:__04/14/06________
It is entitled:
Study of the effect of Mass Distribution, Path
of Energy and Dynamic Coupling on Combined
Coherence (A Non-linerarity Detection Method)
MASTER OF SCIENCE
2006
by
Almost all practical systems are non-linear to some extent with the non-linearity being
introduced in the data analysis and are observed during the modal tests of a structure. For
example, high forcing levels may cause the frequency response function estimates to
show non-coherent behavior over certain frequency bands. A new coherence function
Thomas Roscher [1] applied the combined coherence formulation to theoretical data
generated from a lumped parameter (M, K, C) with static coupling. The results showed
improvement in the combined coherence function over the ordinary coherence, but when
Doug Coombs [2] applied combined coherence to a real world structure, it did not show
theoretical data generated from a lumped mass model with dynamic coupling. The effects
of mass distribution, spatial density, forcing level, location of forcing function, path of
energy and the dynamic coupling on the combined coherence are studied. The testing
cases include SIMO and MIMO cases for a MDOF simulink model with a cubic
calculated for a non-linear model and effects on the combined coherence are studied for
I would like to express my gratitude towards all who were involved in the
completion of this thesis. First of all, I would like to thank Dr. Randall Allemang for
providing me with this opportunity to work under his guidance. I would also like to thank
Dr. Allyn Phillips who helped me out through the research. Dr. Randy and Dr. Allyn
have always been a source of support and encouragement. Their inputs and advice have
I would like to thank Dr. Ronald Huston for serving as member on my thesis
committee.
Laboratory for their helpful discussions in various matters during the course of this work.
I would also like to thank all those who helped me with the style and grammar of the
writing.
Finally, I would like to thank my parents and family for constantly supporting my
academic pursuits.
Table of Contents
1. Introduction………………………………………………………………………1
2. Theoretical Background…………………………………………………………3
I
4.7 Effect of Spatial Density of Masses on Combined Coherence………………74
5. Conclusions……………………………………………………………………...89
6. Future Work…………..………………………………………………………...92
7. References…………………………………………………………………….....93
8. Appendix………………………………………………………………………..95
8.1 Simulink Model when the non-linearity is between DOFs 1 and 2………....95
8.2 Simulink Model when the non-linearity is between DOFs 1 and 3………….96
8.3 Simulink Model when the non-linearity is between DOFs 1 and 4………....97
8.4 Simulink Model when the non-linearity is between DOFs 2 and 3………....98
8.5 Simulink Model when the non-linearity is between DOFs 2 and 4………….99
8.6 Simulink Model when the non-linearity is between DOFs 3 and 4………...100
II
LIST OF FIGURES
Figure 4-3: FRF’s and Coherences of Case 4.2.1……...….…. .…. . .….…..….……… .35
III
Figure 4-9: FRF’s, Coherences and MCCOH for Case 4.2.6……………...…………….41
IV
Figure 4-31: FRF’s, Coherences and MCCOH of Case 4.6.5………………………….. 72
V
LIST OF TABLES
Table 1-1: Sample test cases of combined coherence applied to Roscher model………..19
VI
NOMENCLATURE
NOTATION
m Mass
M Mass Matrix
k Stiffness
K Stiffness Matrix
c Viscous Damping
C Damping Matrix
q Input Location
p Output Location
Hpq Frequency Response Function at output p and input q
..
x(t ) Acceleration
.
x(t ) Velocity
x(t ) Displacement
F Force input in frequency domain
f (t ) Force input in time domain
λ1, 2 Eigen Value
ω Circular Frequency
η Noise on output
ν Noise on input
X`(ω) Measured input of the system
F`(ω) Measured output of the system
γ pq
2
(ω ) Coherence Function
VII
j Rotary Inertia
r Radius
∆t Sample time
ABBREVIATION
DOF Degree of Freedom
SIMO Single Input Multiple Output
MIMO Multiple Input Multiple Output
SDOF Single Degree of Freedom
COH Coherence Function
MCOH Multiple Coherence Function
CCOH Combined Coherence
MCCOH Multiple Combined Coherence
VIII
1. Introduction
Experimental modal analysis is often used for checking the accuracy of an analytical
approach such as finite element analysis and verification/correction of the results of the
analytical approach (model updating). During the modal analysis procedure, there are
four basic assumptions (linearity, time invariance, reciprocity and observability) made
concerning any structure. Because these assumptions are assumed to be valid, errors
accumulate at the modal parameter estimation phase. Among these errors are the errors
due to nonlinearities in the structure and the errors due to digital signal processing. The
errors due to nonlinearities are visible in the measured data as slight distortions in the
frequency response function (FRF) plots, but they are also responsible for significant
discrepancies in the modal analysis process. Some of the algorithms used to extract
modal parameters can be surprisingly sensitive to the small deviations (from linear
structures. Understanding these effects and detecting their presence, means that
alternative test procedures can be used so that the nonlinear effects are not only prevented
from contaminating the measurement and analysis processes but can actually be
quantified and included in the model. In this thesis, a further study is done on the
between degrees of freedom by separating the errors due to digital signal processing and
between the degrees of freedom from the data taken during the modal test. Thomas
1
Roscher applied combined coherence to the theoretical data generated from lumped mass
model [1] and Doug Coombs applied combined coherence to the data measured from a
practical nonlinear structure [2]. Combined coherence was able to locate the
nonlinearities in the first case for theoretical lumped model whereas in the second case it
could not locate the nonlinearities spatially. In this thesis, further study is done on
theoretical data generated from a lumped model with dynamic coupling similar to the real
world system used by Doug Coombs. A study is done on how different parameters such
as mass distribution, spatial density, forcing level, location of forcing function, path of
energy and the dynamic coupling effects the combined coherence. The second chapter in
this thesis gives an introduction to non-linear vibration and methods in detecting the
and the previous work of Roscher and Coombs. In Chapter 4 combined coherence is
applied to a non-linear model and effects on combined coherence are discussed for the
following cases.
2
2. Theoretical Background
the concept of modal analysis. The number of degrees of freedom is the minimum
There exist six degrees of freedom at each point, the motion in each direction and the
rotational motion of each axis. A mechanical system has an infinite number of degrees of
freedom, because the system is continuous. The observed degrees of freedom are in
reality, of course, a finite number, limited by different physical causes. The following
parameters reduce the effective number of degrees of freedom: the frequency range of
There are four assumptions made during the modal analysis procedure [11]. The first
basic assumption is that the structure is linear. This means that the structure obeys the
superposition principle, which states that the response of the system to a combination of
forces applied simultaneously is equal to the sum of the responses due to the individual
forces. The second assumption is that the structure is time invariant. This means that the
properties of the system such as mass, stiffness and damping do not change with time
(i.e., they remain unchanged for any two different testing times). The third assumption is
that the structure obeys Maxwell reciprocity. This principle states that the response of the
function at a degree of freedom “q” due to the input at “p” is equal to the response at “p”
due to the input at “q” i.e., Hpq = Hqp. The fourth assumption is that the structure is
3
observable. The response points are chosen such that the complete structure is observed.
For example, structures with loose components, which have degrees of freedom that
single direction. These are referred as single degree of freedom (SDOF) systems. A
below.
.. .
m x(t ) + c x(t ) + kx(t ) = f (t ) (2.1)
4
This equation has two solutions, a transient solution and a steady state solution. The
equation can be solved using the Laplace transform, assuming initial conditions to be
and finally the homogeneous equation is solved using f (t) = 0 which yields the natural
frequencies as
c c k
λ1, 2 = − ± ( )2 − (2.4)
2m 2m m
The relation between the input to the system and its response is determined by the
frequency response of the system, which is a characteristic feature of the system. The
function.
5
Consider a single input system shown below:
F (ω) = Frequency Domain information of the input signal with no noise on signal
X (ω) = Frequency Domain information of the output signal with no noise on the
signal
But, due to measurement errors, the actual frequency response function is given by
6
The three most common types of frequency response algorithms are based on the least
squares model: the H1 algorithm which minimizes the noise on the output, the H2
algorithm which minimizes the noise on the input and Hv algorithm which minimizes the
noise on both the input and output. In this thesis, the H1 algorithm is used.
This function is frequency dependent and is a real value between zero and one. The value
1 indicates that the measured response power is totally correlated with the measured input
power. The value zero indicates the output is totally correlated with the sources other
than the measured input. A coherence value less than unity at any frequency is due to
variance and bias errors. The low coherence due to a variance error like random noise can
indicator, the more ensembles averaged, the more reliable is the result (smaller standard
deviation). The bias errors can be broadly classified into two categories, digital signal
processing errors and the errors due to nonlinearities. All errors causing drops in the
coherence fall into one of these two categories. The frequencies where the coherence is
low are often the same frequencies where the FRF is maxima in magnitude (resonance) or
7
in coherence at any other frequency is more clearly due to other errors such as noise or
nonlinearities. Multiple inputs are often desired during testing so that the energy is more
evenly distributed throughout a structure and as a result the vibratory amplitudes across
the structure will be more uniform, with a consequent decrease in the effect of
input and output when there is more than one input. The multiple coherence function
(MCOH) that determines the linear dependency of input and output is computed as [12]
Ni Ni
H pq (ω )GFFqt (ω ) H *pt (ω )
MCOH p (ω ) = ∑∑ (2.6)
q =1 t =1 GXX pp (ω )
The value of MCOH varies between zero and one. A value of one indicates an output is
correlated with all known inputs, while a value less than unity indicates unknown
For a linear system the dynamic characteristics will not vary according to the choice of
the excitation technique used to measure them. However, the effects of most kinds of
nonlinearities, encountered in structural dynamics are generally found to vary with the
the type of excitation so that the nonlinearity is exposed and identified. There are
currently many types of excitation methods widely used in vibration study practice.
These excitation techniques are broadly classified as sinusoidal, transient and random
excitation. Sinusoidal excitation is widely regarded as the best excitation technique for
8
to accurately control the input signal level and hence, enables a high input force to be fed
into the structure. However, the drawback of this type of excitation is, it is relatively slow
compared to many of the other techniques used in practice. Since the excitation is
performed frequency by frequency and at each step, time is required for the system to
settle to its steady-state value, sinusoidal methods are very time consuming. On the other
hand, with the random excitation technique, the system can be excited at every frequency
simultaneously within the range of interest. This wide frequency band excitation enables
it to be much faster than the sinusoidal excitation. Also, random excitation in general
linearizes the nonlinear structure due to randomness of input force amplitude. This
technique is the best match for modal analysis, as most of the modal parameter estimation
methods are based on linearity. Due to the above stated factors, random excitation is very
commonly used in actual testing conditions. Hence, test engineers need a nonlinear
detection method that is compatible with normal modal analysis methods employing
random excitation. Therefore, in this thesis, random excitation is used in detecting the
nonlinear material properties. For practical purposes, in many cases, they are regarded as
linear structures because the degree of nonlinearity is small and therefore, insignificant in
the response range of interest. Most theories, upon which structural dynamic analysis is
9
founded, rely heavily on this assumption of linearity (superposition principle). The
superposition principle states that, the deflection due to two or more simultaneously
applied loads is equal to the sum of the deflections caused, when the loads are applied
individually. But for some cases, the effect of nonlinearity may become so significant
that it has to be taken into account in the analysis of dynamic characteristics of the
structure. The present thesis focuses on the location of nonlinearity based on the
Nonlinear structures are often divided into three main types: zero memory, finite memory
and infinite memory systems. The zero memory type of system is the most simple of the
three types, as it only applies the nonlinear operator at system input, whereas the infinite
memory type of system applies nonlinearity to the system response as well. A typical
infinite memory type of system for a MDOF system can be written as [12] [3]
[M] x(t) +[C] x(t) +[K] x(t) +[Kn] x3(t) = f (t) (2.7)
applying it to the MDOF system. The mathematical model of a cubic stiffness element
can be expressed as
10
where the coefficient k represents spring stiffness, and the coefficient ε represents the
degree of nonlinearity. The Figure 2.3 below represents both the linear and the nonlinear
behavior of a cubic stiffness element. It can be seen that the overall stiffness changes with
Cubic stiffness is applied to the simulation model used in this study to observe its
nonlinear characteristics by exciting the system at five forcing levels. The FRF and
coherences of a nonlinear system can be seen in the Figure 2.4. It can be seen from the
FRF and COH function plots that the anti-resonances and resonances are changed as the
excitation force level changes and thus it can be assumed that the system is non-linear.
11
Figure 2-4: FRF and Coherence of nonlinear system
developed. The same is not true for the case of a nonlinear system. In most of the
situations, it is necessary to first detect the presence of nonlinearity. A lot of work is done
in this direction and quite a number of procedures are suggested. A brief review of some
M. Simon and G. R. Tomlinson [4] proposed a Hilbert transform technique to detect and
quantify structural nonlinearities. The basis that the Hilbert transform technique can be
used to identify nonlinearity is due to the fact that for a linear structure, the real and
imaginary parts of a measured FRF constitute a Hilbert transform pair, whereas for the
calculating the Hilbert transform of the real part (or the imaginary part) of a measured
12
FRF and comparing it with the corresponding imaginary part (or real part), the existence
which is based on the mapping of different estimates of stiffness and damping for each
respectively. The equivalent stiffness and damping of a linear system are constant while
for a nonlinear system stiffness and/or damping vary. This method gives an idea of
He J. and D.J. Ewins [7] proposed Inverse Receptance method in which nonlinearity is
detected as whether it exists in the stiffness or damping, by displaying the FRF data in
inverse form. For a linear system a plot of real part of inverse FRF against ω2 and the
imaginary part against ω yields a straight lines while for non-linear systems the plots are
not straight lines. The nonlinearities associated with stiffness show up in the real part
while in the imaginary part the nonlinearities due to damping show up.
method to detect nonlinearities. In this method, the system is excited at only a few chosen
set of frequency lines. It is shown that by exciting the system only at a selected set of
frequency lines, the even nonlinear disturbances can be determined at the even frequency
lines while at unexcited odd frequency lines the odd nonlinear distortions can be
determined.
13
Kim W-J and Y-S Park [10] proposed non-causal power ratio (NPR) method. It is a
causality check method that quantifies the non-linearity. The NPR value grows with the
the non-linearity and also the type of nonlinearity by examining the variation of the NPR
14
3. Non-linear Detection Method (Combined Coherence
Function)
In this chapter the theory of the combined coherence is discussed and mathematical
equations for both the ordinary and multiple combined coherence (MCCOH) are derived.
In general structures are represented by assuming lumped masses as node elements with
mass and no stiffness, and are connected by stiffness and damping terms. The distribution
of mass is important in dynamic analysis. The general representation of the structure and
At any node point if Newton’s law is applied and an equation of motion is developed,
then the acceleration is the sum of both the internal force terms caused by stiffness and
15
damping terms, and the external force terms. Considering a 2 DOF model shown in the
.. .. . .
(m1 + j 2 / r22 ) x1 − ( j 2 / r22 ) x 2 = −(c01 + c12 + c 20 ) x 1 + c12 x 2 − (k 01 + k12 + k 20 ) x1 + k12 x 2 + f1
(3.1)
.. .. . .
(m2 + j 2 / r22 ) x 2 − ( j 2 / r22 ) x1 = −(c12 + c 20 ) x 2 + c12 x 1 − (k12 + k 20 ) x 2 + k12 x1 + f 2 (3.2)
When motions of two degrees of freedom are combined under the condition of equal
mass i.e., (m1 = m2), the contribution of motions due to internal forces between degrees
of freedom will disappear. The equation obtained by combining the motions of DOF is
.. .. . .
x1 + x 2 = 1 / m[−c 20 x 2 − c01 x 1 ] + 1 / m[−k 20 x 2 − k 01 x1 ] + [ f 1 + f 2 ] / m (3.3)
motions between these DOF’s, the drops in coherence due to non-linearity would go
away but the low coherence values due to digital signal processing errors would not
improve. The critical condition for this method is the equality of masses between the
degrees of freedom between which the motions are combined. If the masses are not equal,
16
the detection method can still be applied, if the motions are scaled according to the mass
ratio. In this thesis, a test case is run to see if scaling the masses would improve the
Since the CCOH function is based on the sum of the motion between two DOF’s.
( X r + X p ) Fq* ( X r + X p ) * Fq
CCOH ( pr ) q = (3.5)
Fq Fq* ( X r + X p )( X r + X p ) *
( X r Fq* + X p Fq* )( X r Fq + X p Fq ) *
CCOH ( pr ) q = (3.6)
Fq Fq* ( X r X r* + X r X *p + X p X r* + X p X *p )
| GXFpq + GXFrq | 2
CCOH ( pr ) q = (3.7)
GFFqq (GXX rr + GXX rp + GXX pr + GXX pp )
Ni Ni
H pq (ω )GFFqt (ω ) H *pt (ω )
MCOH p (ω ) = ∑∑ (3.8)
q =1 t =1 GXX pp (ω )
after following the similar steps as for CCOH, MCCOH can be derived as
Ni Ni
( H ps (ω ) + H rs (ω ))GFFqt (ω )( H pt (ω ) + H rt (ω )) *
MCCOH p + r (ω ) = ∑∑ (3.9)
s =1 t =1 GXX pp (ω ) + GXX pr (ω ) + GXX rp (ω ) + GXX rr (ω )
17
3.3 Applying CCOH formulation to the Roscher Theoretical Model
Roscher applied the combined coherence function to the data generated from the
theoretical lumped parameter (M, K, C) model with static coupling. The model used by
the Roscher is shown in the Figure 3.3. Roscher had applied the combined coherence
formulation for various testing conditions for different types of displacement and velocity
related nonlinearities. There was complete improvement in the combined coherence for
some of the cases and in some cases, for some frequency ranges, the combined coherence
did not show improvement. Only a few cases were tested for different kinds of
coherence in determining the nonlinearities, was not extensively studied. In this thesis, a
study is done on how the mass distribution affects the combined coherence by simulating
18
Figure 3-3: Roscher Theoretical Model
A few cases simulated by Roscher are shown in the Table 1-1. As it can be seen from the
CCOH in the range of 16 to 18 Hz and this can be due to the nonlinear motion entering
through other paths. This drop in CCOH still needs to be studied, before CCOH can be
Table 1-1: Sample test cases of combined coherence applied to Roscher model
19
Figure 3-4: FRF and Coherence for Case 1
20
Figure 3-5: Comparison of Coherence and CCOH for Case 1
21
Figure 3-7: Comparison of Coherence and MCCOH for Case 2
Doug Coombs applied combined coherence to a real world structure. The system
consisted of an H-frame with (2x6x0.25) with another square frame (2x2x0.125”) steel
tubing. These two frames were connected at 4 discrete points giving various options for
of 450 in order to get energy in all three directions. The following testing scenarios were
22
Figure 3-8: Line diagram of Doug Coombs model
on combined coherence were studied. For a nominal linear connection between the
connections, the improvement in combined coherence was near the resonances instead at
the anti-resonances raising a question if leakage is affecting the combined coherence. For
many of the testing cases the improvement in the combined coherence was small when
compared to multiple coherence. In one case, when the combined coherence is examined
23
by changing the location of input to the square frame, the previous large improvements in
the combined coherence away from the input locations were not seen.
In this thesis, a study is done on a theoretical model with dynamic coupling similar to the
real world system used by Coombs to study the behavior of combined coherence for
A 4 DOF model with rotary inertia is used to study combined coherence. Figure 3.9
shows a near real time 4 DOF model, similar to that used by Doug Coombs, which is
dynamically coupled. The mi, cij, and kij variables denote the mass, linear viscous
damping, and linear stiffness parameters; the fi variables denote the applied external
forces. The independent coordinates, xi, are defined with respect to an absolute
coordinate system. The idea of this type of model is to study the effect on combined
coherence when the path of energy is across the boundary and to get dynamic coupling
between the degrees of freedom. As can be seen from the equations of motion, degrees of
24
Figure 3-9: Theoretical 4 DOF lumped model
The equations of motion of the model are expressed in terms of a set of coordinates that
are defined with respect to the unique static equilibrium point of the linear system:
25
m1 + j2 / r22 − j2 / r22 0 0 &x&1 (t)
&&
− j2 / r2 m2 + j2 / r2
2 2
0 0 x2 (t)
0 0 m3 + j4 / r44 − j4 / r44 &x&3 (t)
0 0 − j4 / r44 m4 + j4 / r44 &x&4 (t)
c01 + c12 + c13 + c14 − c12 − c13 − c14 x&1 (t)
− c12 c12 + c23 + c24 − c23 − c24 x& (t)
+ 2 (3.10)
− c13 − c23 c13 + c23 + c34 − c34 x&3 (t)
− c14 − c24 − c34 c14 + c24c34 + c04 x&4 (t)
k01 + k12 + k13 + k14 − k12 − k13 − k14 x1 (t) f1 (t)
− k12 k12 + k23 + k24 − k23 − k24 x (t) f (t)
+ 2 = 2
− k13 − k23 k13 + k23 + k34 − k34 x3 (t) f 3(t)
− k14 − k24 − k34 k14 + k24k34 + k04 x4 (t) f 4 (t)
Frequency response function and coherence are evaluated using the dynamic stiffness
method, where the FRF matrix was computed by inverting the system impedance matrix
at each frequency of interest. This provided a means of checking the simulink model,
which used time domain integration to obtain the responses. It can be seen from the FRF
plots below that the dynamic stiffness method results matched the results obtained
through simulink model perfectly. Figure 3-10 below shows the FRF’s of all 4 DOF’s for
26
27
Figure 3-10: Comparison of Analytical and Simulation Results
28
4. Application of Combined Coherence to Analytical model
In this chapter, the simulation results obtained for various cases from a 4 DOF system
MATLAB Simulink model are presented. The 5th order fixed-step Dormand-Prince ODE
method is used. The sample time, ∆t, is set at 0.005 seconds and 216 time steps are
computed, resulting in 327.68 seconds of signal for each simulation. Data is processed in
the Fourier frequency domain and FRF’s are determined for each simulation using the H1
FRF calculation [11] with Fjk(ω) as the input and Xik(ω) as the output. The H1 calculation
In this section, simulations are done to verify whether the system is linear or non-linear,
by exciting the system with five different force-exciting levels. Further, the effect on the
levels is studied. The following MIMO cases shown in Table 4-1 are simulated.
29
Case Location of Force (Increased in steps M1, M2, M3 & ε
Non-Linearity of 10 N) M4 (Kg)
F3 = 20 to 60 N
F3 = 20 to 60 N
It can be seen from the FRF and coherence function plots (Figures 4-1 and 4-2), that the
anti-resonances and resonances are changed as the excitation force level changes. Thus, it
can be assumed that the system is non-linear. The drops in coherence can be attributed to
digital signal processing errors as well as to the non-linearity. For example, from the
coherence plot (coherence 1) of case 4.1.1, it can seen that the drop in coherence value at
It can be seen from the MCCOH of case 4.1.1 (Figure 4-1), at lower forcing levels the
MCCOH showed improvement while at higher forcing levels it still showed improvement
but with more distortion. The distortion at a forcing level of 70 N is more when compared
to a forcing level of 30 N. Hence, it can be concluded that the nature of the improvement
in the MCCOH is inversely proportional to the forcing level, i.e., at lower force levels the
30
ability of the MCCOH to detect non-linearities is greater when compared to higher force
levels.
31
Figure 4-2: FRF’s, Coherences and MCCOH for Case 4.1.2
32
4.2 SIMO Situations for a system with Dynamic Coupling
In this section, the following cases shown in Table 4-2 are simulated for a system close to
the real world testing conditions where the mass distribution between the DOF is uneven
Non-Linearity
It can be seen from the FRF plots that there are distortions at both resonances and anti-
resonances. From the coherence function plots, it can be seen that the drops in the
coherence value can be attributed to digital signal processing errors as well as to the non-
linearity. In the coherence function plot of Case 4.2.1, one could see the drops between 5
but are due to non-linearity. Also, one could see the drops at 3.5, 6 and 8 Hz that are at
resonances or anti-resonances and the drops in the higher frequency range (above 20 Hz).
33
As a next step, the CCOH for responses 1 and 2 is compared with ordinary coherence. It
can be seen that the drops in the higher frequency range and the drops associated with
non-linearity (5-6 Hz) are completely eliminated and one could only see the drops at
completely eliminated but has shown improvement over the ordinary coherence. The
complete clear up of the CCOH is not seen because of one or a combination of three
factors:
1. Dynamic coupling
2. Due to the non-linear motion entering the system from other paths
3. Mass difference between the DOF’s between which the CCOH has been
computed
Similar results have been observed for all other combination of cases i.e., when the non-
linearity is located between DOF’s 1 and 3, 1 and 4, 2 and 3, 2 and 4, and 3 and 4 that
there are drops in coherences due to non-linearity, leakage and at higher frequencies.
CCOH has shown improvement at anti-resonances but not at resonances and complete
34
Figure 4-3: FRF’s and Coherences of Case 4.2.1
35
Figure 4-4: Coherence and CCOH of Case 4.2.1
36
Figure 4-5: FRF’s, Coherence and CCOH for Case 4.2.2
37
Figure 4-6: FRF’s, Coherences and MCCOH for Case 4.2.3
38
Figure 4-7: FRF’s, Coherences and CCOH for Case 4.2.4
39
Figure 4-8: FRF’s, Coherences and CCOH for Case 4.2.5
40
Figure 4-9: FRF’s, Coherences and CCOH for Case 4.2.6
41
4.3 MIMO Situations for a system with Dynamic Coupling
In this section, the MIMO situations shown in the Table 4-3 below are simulated. The
testing conditions, severity of non-linearity, locations of non-linearity, mass and all other
conditions are similar to that of the previous case (4.2) except for the input given at two
DOF’s between which the non-linearity is located. Multiple inputs determine if the
structure responds in a non-linear regime. More often, most modal analysis procedures
involve the application of multiple inputs in order to get more uniform energy
distribution. Whereas, the SIMO situation induces non-linear behavior in the vicinity of
the input location and structure might not be excited well at remote points, therefore,
42
Case Location of Force M1 M2 M3 M4 ε
Non-Linearity
F2 = 40 N
F3 = 40N
F4 = 40 N
F3 = 40 N
F4 = 40 N
40 N
The FRF estimation, the MCOH and the MCCOH obtained are as shown in the Figures
(4-10 to 4-16) below. As seen from the plots below, the results obtained in this case are
similar to that of the previous case. There are frequency shifts in the FRF’s at resonances
and anti-resonances. Also, there are low coherence values due to non-linearity and digital
signal processing errors, like leakage at resonances and anti-resonances. One can see the
43
resonances. At some frequencies, the MCCOH has not registered complete improvement.
In the coherence plot of Case 4.3.1, one can see the drop in coherence over the frequency
range of 5 – 7 Hz, which is at anti-resonance, is completely improved. The drop over the
frequency range of 10 –14 Hz, which is near the resonance, is not improved. It can be
Similar results have been observed for all other combination of cases i.e., when the non-
linearity is located between DOFs 1 and 3, 1 and 4, 2 and 3, 2 and 4, and 3 and 4 that
there are drops in coherences due to non-linearity, leakage and at higher frequencies.
MCCOH has shown improvement at anti-resonances, but not at resonances, and complete
It can be seen from the figures that the MCCOH has shown improvement over the
MCOH in all of the above situations but for some frequency ranges the complete
1. Dynamic coupling.
2. Due to the non-linear motion entering the system from other paths.
3. Mass difference between the DOF’s between which the MCCOH has been
computed.
But, it is not clear from this case whether the incomplete improvement in the MCCOH is
due to either dynamic coupling or due to the mass difference or because of the non-linear
44
Figure 4-10: FRF’s, Coherences and MCCOH of Case 4.3.1
45
Figure 4-11: FRF’s and Coherences of Case 4.3.2
46
Figure 4-13: FRF’s, Coherences and MCCOH of Case 4.3.3
47
Figure 4-14: FRF’s, Coherences and MCCOH of Case 4.3.4
48
Figure 4-15: FRF’s, Coherences and MCCOH of Case 4.3.5
49
Figure 4-16: FRF’s, Coherences and MCCOH of Case 4.3.6
50
4.4 Effect of Dynamic Coupling on Combined Coherence
In the previous cases, it is seen that the complete improvement of the combined
coherence is not observed and reasons for it are attributed to dynamic coupling, mass
difference and/or path of energy. In this section, the following cases are simulated to
study the effect of dynamic coupling on the MCCOH. The rotary inertia term has been
reduced by 100 times (i.e., making the dynamic coupling between DOF’s negligible.)
Non-Linearity & M4
F2 = 40 N J4=M*R42/200
F3 = 40N J4=M*R42/200
F4 = 40 N J4=M*R42/200
F3 = 40 N J4=M*R42/200
F4 = 40 N J4=M*R42/200
F4 = 40 N J4=M*R42/200
51
The FRF estimation, the MCOH and the MCCOH obtained are shown in Figures (4-17 to
4-22) below. It is concluded from last case that dynamic coupling is one of the reasons
why the MCCOH has not shown complete improvement. Therefore, it is expected from
this case, that the MCCOH will show improvement, as the dynamic coupling is made
negligible. It can be seen from the MCCOH plot of Case 4.4.1 the improvement in
MCCOH is complete whereas in all other cases (4.4.2 to 4.4.6) there is not complete
improvement in MCCOH. In the MCCOH plot of Case 4.4.2, it can be seen that over the
frequency range of 15 – 18 Hz, the MCCOH has not shown improvement. Though, the
effect of dynamic coupling is made negligible, the MCCOH has not shown complete
improvement in all the cases. Therefore, it can be concluded from this case that the
dynamic coupling has no effect on the MCCOH. So, the incomplete improvement of
MCCOH might be due to either the mass difference or the non-linear motion entering
52
Figure 4-17: FRF’s, Coherences and MCCOH of Case 4.4.1
53
Figure 4-18: FRF’s, Coherences and MCCOH of Case 4.4.2
54
Figure 4-19: FRF’s, Coherences and MCCOH of Case 4.4.3
55
Figure 4-20: FRF’s, Coherences and MCCOH of Case 4.4.4
56
Figure 4-21: FRF’s, Coherences and MCCOH of Case 4.4.5
57
Figure 4-22: FRF’s, Coherences and MCCOH of Case 4.4.6
58
4.5 Effect of Location of Input and Path of Energy on Combined Coherence
This case is simulated for the MIMO situations as above but the forcing function is not
placed directly on the DOF that is associated with the non-linearity. These cases are
simulated to study the ability of the combined coherence to detect the non-linearity when
the energy comes from the linear path and also when input is placed some distance away
Linearity M4
F4 = 40 N
F4 = 40N
F3 = 40 N
F4 = 40 N
F3 = 40 N
F2 = 40 N
59
The FRF estimation, the MCOH, and the MCCOH obtained are shown in Figures (4-23
to 4-28) below. It can be seen that for Cases 4.5.1 and 4.5.6 where the forcing function is
away from the DOF’s between which the non-linearity is located, the MCCOH has
registered a drastic improvement. For these cases, when compared with Cases 4.3.1 and
4.3.6 respectively for the same level of excitation, the FRF’s and coherence functions are
not distorted as much as when the forcing function is directly placed at the DOF’s where
the non-linearity is located. Whereas in Cases 4.5.2 to 4.5.5 the result is reversed, the
FRF and MCCOH are distorted more than Cases 4.5.1 and 4.5.6 and also there is not
complete clear up of the MCCOH function. The improvement in the MCCOH in Cases
4.5.1 and 4.5.2 can be because the forcing function is away from the DOF where the non-
linearity is being located and the energy is entering through a more linear path. However,
it has been concluded in the previous case that the dynamic coupling has no affect on the
MCCOH, so from this case it can be concluded that the location of inputs and energy
path are critical in determining the ability of the MCCOH in detecting the non-linearities.
60
Figure 4-23: FRF’s, Coherences and MCCOH of Case 4.5.1
61
Figure 4-24: FRF’s, Coherences and MCCOH of Case 4.5.2
62
Figure 4-25: FRF’s, Coherences and MCCOH of Case 4.5.3
63
Figure 4-26: FRF’s, Coherences and MCCOH of Case 4.5.4
64
Figure 4-27: FRF’s, Coherences and MCCOH of Case 4.5.6
65
4.6 Effect of Mass Distribution on Combined Coherence
In this section, MIMO situations are simulated by considering equal mass at all DOFs.
This is to see how the mass difference of the DOF’s between which the non-linearity is
associated, affects the MCCOH. The masses at all the DOF’s are made equal. The
Non-Linearity M4
F2 = 40 N
F3 = 40N
F4 = 40 N
F3 = 40 N
F4 = 40 N
F4 = 40 N
66
It can be seen from the plots of the MCCOH, that it has shown complete improvement in
Case 4.6.6 and in all other cases from 4.6.1 to 4.6.5, the MCCOH is improved but still
exhibits drops over some frequency ranges. For example, from the MCCOH plot of Case
4.6.2, it can be seen that over the frequency ranges of 7 – 8 Hz and 13 – 18 Hz there is no
complete improvement in the MCCOH. By comparing the MCCOH of Case 4.6.6 and
Case 4.3.6 it can be concluded that the mass inequality between the DOF’s can be a
possibility for the MCCOH to detect non-linearities. Even though the mass difference
between the degrees of freedom 3 and 4 in Case 4.3.6 is small (6 kg, this difference is
mass difference, the MCCOH has shown great improvement. In all other cases, the
MCCOH has shown improvement when compared to Cases 4.3.1 to 4.3.6 but of much
smaller values. It is expected that when the mass inequality between the degrees of
freedom is eliminated, the combined coherence should show greater improvement. But
from these cases, it can be concluded that besides the mass inequality, the path of energy
is also critical in detecting the non-linearities. This can be seen from Cases 4.6.1 to 4.6.5,
67
Figure 4-28: FRF’s, Coherences and MCCOH of Case 4.6.1
68
Figure 4-29: FRF’s, Coherences and MCCOH of Case 4.6.2
69
Figure 4-29: FRF’s, Coherences and MCCOH of Case 4.6.3
70
Figure 4-30: FRF’s, Coherences and MCCOH of Case 4.6.4
71
Figure 4-31: FRF’s, Coherences and MCCOH of Case 4.6.5
72
Figure 4-32: FRF’s, Coherences and MCCOH of Case 4.6.6
73
4.7 Effect of Spatial Density of Masses on Combined Coherence
In this section, MIMO situations similar to the real world testing situations, where the
system consists of more than one component, with a difference in mass densities are
simulated. As an example, it can be seen from the Doug Coombs model there are two
Non-Linearity M4
F2 = 40 N
F3 = 40N
F4 = 40 N
F3 = 40 N
F4 = 40 N
F4 = 40 N
74
It can be seen from the MCCOH plots of Cases 4.7.1 and 4.7.6 that the MCCOH has
shown improvement where the mass difference between the DOF’s for which the
MCCOH is computed is negligible. In other cases, the MCCOH has not shown any
improvement at all due to the huge mass difference between the DOF’s. In Cases 4.7.1 to
4.7.6, the improvement in the MCCOH is not complete due to the relative motion
75
Figure 4-33: FRF’s, Coherences and MCCOH of Case 4.7.1
76
Figure 4-34: FRF’s, Coherences and MCCOH of Case 4.7.2
77
Figure 4-35: FRF’s, Coherences and MCCOH of Case 4.7.3
78
Figure 4-36: FRF’s, Coherences and MCCOH of Case 4.7.4
79
Figure 4-37: FRF’s, Coherences and MCCOH of Case 4.7.5
80
Figure 4-38: FRF’s, Coherences and MCCOH of Case 4.7.6
81
Cases 4.8: Effect of Scaling of Motions of DOF on Combined Coherence
In this section, MIMO situations are simulated by scaling the motions at the DOF’s at
which the non-linearity is located. The crucial condition for the combined coherence
technique in determining the location of the non-linearity is the mass equality; however
in real world testing conditions, mass equality between the DOF’s is not achieved. This
case is tested for MIMO situations to study if scaling the motions would improve the
Non-Linearity M4
F3 = 40 N
F4 = 40N
F3 = 40 N
F4 = 40 N
F4 = 40 N
82
It can be seen from the MCCOH plots that the MCCOH has shown improvement over the
ordinary coherence, but not improved completely. The complete improvement in the
MCCOH is not seen as it can be affected by non-linear motion entering from other paths.
Case 4.8.5 is simulated similar to Case 4.3.6 and it can be seen from the MCCOH of case
4.8.5 that there is not much improvement when compared to Case 4.3.6. From this
comparison, it can be concluded that non-linear motion entering from other paths has an
83
Figure 4-39: FRF’s, Coherences and MCCOH of Case 4.8.1
84
Figure 4-40: FRF’s, Coherences and MCCOH of Case 4.8.2
85
Figure 4-41: FRF’s, Coherences and MCCOH of Case 4.8.3
86
Figure 4-42: FRF’s, Coherences and MCCOH of Case 4.8.4
87
Figure 4-43: FRF’s, Coherences and MCCOH of Case 4.8.5
88
5. Conclusions
It is observed from all of the situations simulated above that, the combined coherence is
able to separate the nonlinearities from digital signal processing errors but that there is no
complete improvement in the combined coherence over some frequency ranges. The first
case is simulated for MIMO situations to study how the increase in the force affects the
combined coherence. It is concluded from this case that the nature of the improvement in
the MCCOH is inversely proportional to the force level, i.e., at lower force levels the
ability of the MCCOH to detect non-linearities is more when compared to that of higher
force levels.
The second and third cases are simulated for SIMO and MIMO situations respectively to
study the nature of the improvement of the combined coherence. It is concluded from
these cases that it is unclear whether the incomplete improvement of the combined
coherence is due to dynamic coupling, mass inequality and/or path of energy. The fourth
case is simulated to study the effect of dynamic coupling on the combined coherence by
making it negligible. This case has showed that the dynamic coupling has no effect on the
combined coherence in detecting nonlinearity. The fifth case is simulated for MIMO
situations to study the path of energy. The combined coherence has shown complete
improvement when the location of input is away from the DOF’s between which
nonlinearity is located. It is concluded from this case that the path of energy is critical in
detecting the nonlinearities. The sixth case is simulated for the MIMO situations with
equal masses. It was expected from this case that the combined coherence would improve
89
significantly over the Case 3, which is simulated with the same testing conditions except
for the difference in masses at DOF’s. However, the results obtained have not shown the
complete improvement over Case 3. It is concluded that besides the mass inequality, the
The seventh case is simulated for MIMO situations similar to real world testing
conditions with a difference in the mass distribution, i.e., with one of the masses lighter
than the other. The combined coherence has not shown improvement when the
nonlinearity is located between the DOF’s with large mass differences. The eighth case
is studied for MIMO situations by scaling the motions of the DOF’s that are associated
with nonlinearity. Scaling the motions of the DOF’s that are associated with nonlinearity
has shown improvement over the same cases where no scaling of motion is done. Scaling
the motions did not completely improve the combined coherence and that can be due to
It can be concluded from the above discussion that the dynamic coupling has no effect on
crucial condition, and the path of energy are the primary elements affecting the ability of
the combined coherence in detecting the non-linearity. Scaling the masses can improve
the combined coherence, but even when the scaling is done, because of the nonlinear
motion entering from other paths, combined coherence could not improve completely.
Because the equality of mass between degrees of freedom cannot be achieved in the real
90
world systems and also energy can enter through non-linear paths, it is difficult to detect
91
6. Future Work
Though the combined coherence, in all the cases studied, is able to separate the
nonlinearity from the digital signal processing errors, drops are observed over some
frequency ranges. Work still need to be done to study the ability of combined coherence
1. It was observed that, in most of the situations high frequency distortions in the
2. The system with single location of nonlinearity was studied, whereas in real
structures nonlinearities are located at more than one location. Therefore, this
the system.
3. Location of input and path of energy are critical in detecting the structural
coherence.
4. Other area can be the study on the ability of combined coherence in detecting
92
7. References
[1] “Detection of Structural Non-Linearities using the Frequency Response and
Coherence Functions”, Master’s Thesis,
T. Roscher, University of Cincinnati, 2000
[4] “Use of Hilbert transform in modal analysis of linear and non-linear structures”
M. Simon and G. R. Tomlinson
Journal of Sound and Vibration, volume 96, Issue 4, 22 October 1984, Pages 421-
436
[6] “The complex stiffness method to detect and identify non-linear dynamic
behavior of SDOF systems”
Mertens M., H. Van Der Auweraer, P. Vanherck and R.Snoeys
Mechanical Systems and Signal Processing 3 (1), pp37-54
[7] “A simple method of interpretation for the modal analysis of nonlinear systems”
He J. and D.J. Ewins: 1987,
Proceedings of the 5th International Modal Analysis Conference, London
(England), pp 626-634
[8] “An explanation of the cause of the distribution in the transfer function of a
duffing oscillator subject to sine excitation”
Storer D.M. and G.R. Tomlinson: 1991,
Proceedings of the 9th International Modal Analysis Conference, pp.1197-1205
93
[10] “Non-linearity identification and quantification using an inverse Fourier
Transform”
Kim W-J and Y-S Park: 1993,
Fourier Transform, Mechanical Systems and Signal Processing 3, pp 239-255
94
8. Appendix
95
8.2 Simulink Model when the non-linearity is between DOFs 1 and 3
96
8.3 Simulink Model when the non-linearity is between DOFs 1 and 4
97
8.4 Simulink Model when the non-linearity is between DOFs 2 and 3
98
8.5 Simulink Model when the non-linearity is between DOFs 2 and 4
99
8.6 Simulink Model when the non-linearity is between DOFs 3 and 4
100