Vibrations and accoustics lab re-

port (MECA-H-411)
Jawad Bessedjerari, Ali Hussnain, Mikaïl Köroglu,
Gaetan Moussiaux, Joseph Rosenzweig

Data submitted
15/12/2015
 Abstract
The modal parameters are a very important part when studying a
structure. They give information about the dynamic response of the con-
struction during excitation. Dangerous situations can be avoided with a
good knowledge of these parameters.
The present report covers a modal study of a structure which consists of
4 beams and 3 plates. The roving hammer test is used to find the fre-
quency response function, based on the principle of reciprocity. Because
only ten modes are studied, the hammer was chosen to excite only a low
frequency range. The resonance frequencies for the different modes are
presented in the report. It is found that they are in a frequency domain
ranging from 7 to 15 Hz and that their damping ratios vary between 0.127
% and 3.2727 %. Furthermore, the first ten modes shapes are discussed.

Keywords : modal analysis - modes - hammer - resonance frequency

- damping - reciprocity - frequency response function

1
Contents

1 Introduction 3

2 Results and discussion 4

2.1 Task 0: Choice of the hammer tip and the sensor locations . . . . 4
2.2 Task 1: Data acquisition . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Task 2: Calculation of the Frequency Response Functions (FRFs) 8
2.3.1 Frequency Response Function . . . . . . . . . . . . . . . 9
2.4 Task 3: Modal parameter estimation and validation . . . . . . . 12
2.5 Task 4: Modal parameter validation . . . . . . . . . . . . . . . . 14

3 Conclusion 16

Appendices 18
.1 Matlab code . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
.1.1 Script to process measurements . . . . . . . . . . . . . . 19
.1.2 Calculating the H1 estimater for the frequency response . 20
.1.3 Calculate and plotting the Modal Assurance Criterion . . 22
.1.4 Plotting the different modes . . . . . . . . . . . . . . . . 22
.2 Mode shapes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2
Chapter 1

Introduction

In this paper, the natural characteristics of a structure are studied, which are the
frequency, damping and mode shapes. The process which describes a structure
in terms of these characteristics is called modal analysis [1].
To do this, a roving hammer test will be applied to the system. It consists of
hitting 58 points on the beams and plates of the construction and measuring the
acceleration with 3 accelerometers placed in a certain manner. This strategy is
sufficient for obtaining the modal parameters due to the principle of reciprocity.
Once all the measurements are done, the frequency response function of the
system can be calculated. Ultimately, the searched modal parameters will be
calculated based on this function.

The main objectives of this lab are

• Getting familiar with data acquisition and vibration measurements

• Identification of modal characteristics of a vibrating structure

• Validation the acquired data by using mathematical validation tools

Chapter 2 explains step by step how this is done. A conclusion is presented in

chapter 3 and finally the used matlab code and some extra figures can be found
in the appendix.

3
Chapter 2

Results and discussion

2.1 Task 0: Choice of the hammer tip and the

sensor locations
In the setup preparation, one of the choices that had to be made was the hammer
tip. We know intuitively that the hammer will be making longer contact with
the structure when the tip is soft. Hence, the softer the tip, the less the impact
will look like a Dirac pulse. We know from the theory that a Dirac pulse excites
all the frequencies of the system. When we deviate from a Dirac pulse because
of the tip being softer, then the excited frequency range will be reduced. This is
illustrated in the figures below.

Figure 2.1: Dirac Pulse and Fourier Transform

Figure 2.2: Deviated Dirac Pulse and its Fourier Transform

4
 The question that arises is thus: what is the range of frequencies that we
want to excite? In this lab it is asked to identify the 10 first modes of the system,
so the tip that we will choose will be according to a frequency range that contains
the frequencies corresponding to the 10 first modes. According to a finite element
model we know that the first 10 modes will surely be included within a range of
150Hz. Hence we will be using the rubber tip because the frequency range that it
will excite is wide enough so that it contains this frequency. As a matter of fact,
the Matlab plot below in figure 2.3 shows that at 150Hz the according amplitude
is high enough so that it can be considered as a response of the structure due to
the hammer impact.

Figure 2.3: Spectral structure response when hit at point 14

Indeed, the amplitude that corresponds with 150Hz is higher than -40dB,
which is the minimal amplitude that can be considered as reliable. Below this
amplitude we cannot distinguish between the response due to noise and due to
the impact. Finally we can see that the response is like expected, in contrast with
some other points where we can see that the frequency response is fluctuating
which points to the fact that we double hit the structure with the hammer. An
example of such a frequency response can be seen in figure 2.4, which is the
frequency response of the structure when we hit at the second point.
The second choice that had to be made was the location of the accelerom-
eters. These have to be put at points where the deflection is the highest, that
is where the acceleration is the highest. Intuitively we know that those points
will be located somewhere on top of the structure since there we are located the
furthest away from the support. We thus chose 3 points, 1 for each direction,

5
 Figure 2.4: Spectral structure response when hit at point 2

the furthest possible away from the support and put the accelerometers there
oriented in the 3 different directions. More specifically we choose to place the
accelerometers at points 45, 54 and 34.

2.2 Task 1: Data acquisition

The purpose of this experiment is to compute the Frequency Response Func-
tions(FRFs). The measurement technique that will be used to this aim is called
the roving hammer test. Firstly, 58 points of the structure are chosen in a specific
way1 , which defines the degrees of freedom of our system.
The degrees of freedom of our system is relevant here because the more
degrees of freedom our system(which has in reality an infinite number of degrees
of freedom) has, the better we can represent its modeshapes afterwards. Indeed,
at higher frequencies, the modeshapes become more complex and hence require
many measurement points to represent them with the FRFs. The structure is
then hit at those points and the structure response is measured at 3 of its degrees
of freedom using accelerometers.
One could think that 3 measured points is way too few to represent the
modeshapes of the system afterwards. Placing 58 accelerometers would be not
practical and would even change the structure through it’s mass for example,
which would raise considerably. A trick to get around this problem is the use of
1
among other things to make sure these points do not correspond with nodal points

6
Maxwell’s Reciprocity Theorem, which states that the response of a linear system
at a location x when excited at a location y is exactly equal to the response at
location y when excited at location x. Hence, exciting the system at 58 locations
and measuring at 3 locations is equivalent with exciting the system at 3 locations
and measuring at 58 locations.

Once task 0 is performed we can start the measurement procedure. The type
of sensors used are IEPE(Integrated Electronic Piezzo Electric) accelerometers
which measure the vibrations in this solid structure using the Piezzo electric ef-
fect whereby a force acts on a Piezzo electric element after the acceleration of
the accelerometer, which in turn generates a voltage in response. The quantity
that is measured with the accelerometers is thus the accelerations at some points
of the structure of which the choice is made in a manner described above. Note
that the impact force isn’t measured with the accelerometers but with a force
impedance in the hammer that is connected to the computer.

The setting of the measurement parameters are as follows:

• T = Total measure time [s] = 10

• FS = Sampling frequency [Hz] = 2048

• Nb = Number of averages = 3

This measuring time is necessary as it prevents leakage when the response

signal is transformed with FFT. Indeed, after 10 seconds the time response of
the structure is pretty much equal to 0. We shall elaborate more on this in task
2 (see windowing).
The sampling frequency has to be twice the highest frequency where we are
interested in according to the Shannon Theorem. The highest frequency like
stated above corresponds roughly with a frequency of 150Hz so we would need
a minimal sampling frequency of 300Hz. We choose the sampling frequency to
be equal to 2048Hz, which is way above the minimal sampling frequency. This
is good for getting a more accurate reproduction of the modeshapes afterwards
since the amount of information acquired is more than sufficient.
Finally, we chose the number of averages to be equal to 3, i.e. we chose to
measure 3 times at the same point to ensure a minimal reduction of the noise.
More than 3 measurements per point would obviously give more accurate results
but it would then also take a lot more time to perform the experiment, which is
limited in this project.

7
2.3 Task 2: Calculation of the Frequency Re-
sponse Functions (FRFs)
In order to calculate FRFs, we need to transform the signal to the frequency
domain. In practice, signals are sampled and a discrete form of the Fourier
transform is used to transform this sampled signal, which is called the Fast
Fourier Transform or FFT. However this technique has its limitations. We know
from calculus that a function can be transformed in the Fourier domain provided
this function is periodic. Hence, when applying FFT on a sampled signal, a
piece of this sampled signal will be duplicated in order to make it periodic. The
problem is that if the number of periods in the acquisition of this piece of signal
is not an integer, the begin point and the endpoint of this piece of signal won’t
match and we will be having discontinuities like can be seen in figure 2.5.

Figure 2.5: Spectral leakage

These discontinuities show up as high-frequency components that are not

present in the original signal. The spectrum you get by using a FFT, therefore, is
not the actual spectrum of the original signal, but a smeared version. It appears
as if energy at one frequency leaks into other frequencies. This phenomenon is
known as spectral leakage, which causes the fine spectral lines to spread into
wider signals. The point of windowing is to reduce the amplitude of the discon-
tinuities. Windowing consists of multiplying the time record by a finite-length
window with an amplitude that varies smoothly and gradually toward zero at the
edges. This makes the endpoints of the waveform meet and, therefore, results in

8
a continuous waveform without sharp transitions [2]. In our experiment window-
ing is not necessary since the signals on which we apply FFT have more or less
endpoints that are equal to 0. For the impact force it is quite obvious and for the
response measured by the accelerometers we chose a sufficient large measuring
time in order for the response to decay to 0 as can be seen in the next paragraph.

2.3.1 Frequency Response Function

When the data is acquired, the FRF can be calculated for each point. After
this, a global FRF (for the whole system) can be calculated if the quality of the
singular FRF’s is good. The starting point to calculate them are the input and
output of the system in the time domain, which are shown below for a few points.

Figure 2.6: The input (force) in blue and the output (acceleration) in orange
when hitting point 5

As can be seen on these figures, the amplitude of the oscillations goes to

zero after some seconds. This is the reason why windowing is not necessary.

(a) Point 40 (b) Point 52

9
 After doing some calculations(appendix .1), the FRF of these points can be
found. Their amplitude and phase are shown in the figures below.

(a) Point 5 (b) Point 5

(c) Point 40 (d) Point 40

(e) Point 52 (f) Point 52

Figure 2.7: FRF amplitude (left) and phase (right) in function of the frequency

10
 These figures show that the measurement becomes noisier with higher fre-
quencies. The peaks in the amplitude figures show the resonance frequencies
for the system. The positive peaks (resonance) are all separated by the negative
peaks (anti-resonance), in accordance to the theory [1]. The resonance frequen-
cies can also be detected in the phase graphs. Each time a resonance frequency
is passed, the phase loses 180 degrees. When an anti-resonance is passed, the
phase gains 180 degrees [1].
The quality of the results can be reviewed by calculating the coherence func-
tion (equation 2.1).

|Gˆuf |2
γ2 = (2.1)
Gˆuu · Gˆf f
The quality of the results is good if the value of gamma is equal to 1. The
more it differs from 1, the worse the quality. The coherence function is plotted
in function of the frequency in figure 2.8. It is again remarked that the quality of
the results is good for low frequencies, but it deteriorates around 200 Hz. This
is because the rover hammer test mostly excites the lower frequencies (as can
be seen in figures 2.3 and 2.4), the energy in the higher frequencies is low and
therefore more vulnerable to noise.

Figure 2.8: The coherence function for point 21 in function of the frequency

11
2.4 Task 3: Modal parameter estimation and
validation
Once the data is acquired and processed, the most important part of the project
can start: to determine the modal parameters of the system and mode shapes.
The modal parameters are:

• Poles with damped natural frequencies and damping ratios

• Mode shapes

To do these calculations, a function needs to be found that fits the FRF graph
as good as possible. Finding a polynomial function which fits the FRF of the
whole system, can be done by using the function Polymax. Polymax generates
a stabilization chart (figure 2.9) on which the poles are selected through which
the polynomial function needs to go. The selected poles were the stable ones
corresponding to a peak in the graph.

Figure 2.9: Stabilization chart

Once the polynomial function is found, the resonance frequencies and damp-
ing ratios are calculated by the program. They are displayed in table 2.1. All
the frequencies displayed in the table result in a resonance of the structure. It is
remarked that the frequencies of 112 Hz, 58 Hz and 116 Hz are the ones which
results in the largest oscillations, since they have the lowest damping ratio. All
the resonance frequencies are also in the range for which the coherence function
was approximately equal to 1, ensuring a good quality of the data.

12
 Table 2.1: Resonance frequencies and damping ratios for the exited modes

Mode number Resonance frequencies [Hz] Damping ratios [%]

1 7.68 0.573
2 11.03 3.020
3 20.09 0.495
4 21.46 0.129
5 58.15 1.205
6 82.86 0.175
7 90.38 0.365
8 92.41 3.272
9 116.10 0.158
10 112.82 0.127

The shape of two modes are given in figures 2.10 and 2.11. Each colour
represents a different beam of the system, and there are two plots for each
mode. One in the x-direction and one in the y-direction. The rest of the mode
shapes can be found in appendix .2.

(a) mode 3-x (b) mode 3-y

Figure 2.10: The third mode shape of the structure for the x direction and y
direction. Each colour on the plot represents a different beam.

13
 (a) mode 6-x (b) mode 6-y

Figure 2.11: The sixth mode shape of the structure for the x direction and y
direction. Each colour on the plot represents a different beam.

2.5 Task 4: Modal parameter validation

In this task, we verify results of modal parameter estimation phase by using
Intuitive validation tools and Mathematical validation tools.

Intuitive Validation Tools This tool is observant and perceptive. It needs

visual inspection of mode shapes. We compare the already measured frequency
Response Function with estimated modal parameters.

Mathematical Validation Tools Modal assurance criterion (MAC) is an im-

portant mathematical tool to check quality of estimated mode shapes. It com-
pares different sets of estimated mode shapes or investigate within one set. To
see if the orthogonality condition is respected. The MAC is expressed by the
Eq.(2.2).

|{ψ}H
r {ψ}s |
2
M AC({ψ}r {ψ}s ) = (2.2)
({ψ}H H
r {ψ}r )({ψ}s {ψ}s )
This number should be equal to 1 when we compare the 2 same modes and
when we have 2 different modes then it should be equal to 0 in theory (the
orthogonality condition). To consider all the modes at the same time and to
have a better representation we can use the bar3 function in matlab to plot a
3D graph. This is done in Figure 2.12 with the MAC in function of the different
mode shapes.

14
 Figure 2.12: Auto MAC

Figure 2.12 tells us that the modes that we have chosen with Polymax are
quite correct because the values on the diagonal are all equal to 1 and the others
are not all equal to 0 but really close to it. The chosen modes are consequently
acceptable.

15
Chapter 3

Conclusion

After this project we can have a good understanding of the meaning of the modal
analysis in a practical example with the structure used. We learned a method
to obtain this modal analysis which is called the hammer test. A mathematical
procedure was also done to convert our given signals in the time domain into
frequency response functions. These FRFs were then used to find the different
modal parameters thanks to the Polymax function in Matlab. With Polymax, we
were able to choose the stable poles by selecting the peaks on the stabilization
chart. Once the polynomial function was found, we obtained different modal pa-
rameters such as the mode numbers,the resonance frequencies and the damping
ratios. A representation of the modes was then possible. To see if the poles that
we had chosen were good we made a modal parameter estimation using a 3D
representation of the MAC function.
At the end, the goal of modal analysis is to see if our structure is sensitive to
vibrations and to see at which frequencies the vibrations are the most critical.

16
Bibliography

[1] P. Avitabile, Experimental Modal Analysis: A Simple Non-Mathematical Pre-

sentation.

[2] N. Instruments, Understanding ffts and windowing [cited 2015-12-10].

URL http://www.ni.com/white-paper/4844/en/

17
Appendices

18
.1 Matlab code
.1.1 Script to process measurements

clear all
close all
clc

load('TDD_loc_1.mat');
matrix = TDD.a_data;

[Nt x y] = size(matrix);
Nb = 3; % Number of averages
u1 = zeros(Nt,1,1);

Fm = 13.1072e6; % Hz: internal time base NI-9234 (Fix value!)

K = 25; % Can be modified for changing the sampling rate

Fs = Fm/(256*K); % sampling frequency of NI DAQ frequency Hz

Nf = (Nt/2+1); % Number of points in the frequency domain

freq = Fs/2*linspace(0,1,Nf); % creating the x-axis (see help fft)

for i=1:58

fileName = sprintf('TDD_loc_%d.mat',i);
load(fileName);
matrix = TDD.a_data;

% Creating new matrix with changed signs for the force on some of the
% beams

% Changing the sign of the force for the right points

if (25<= i && i <= 36)
matrix(:,1,:) = - matrix(:,1,:);
end

if (37<=i && i<=47 && mod(i,2)==1)

matrix(:,1,:) = - matrix(:,1,:);
end

[estimater gamma] = H1(matrix);

FDD.a_frf(:,i,:) = estimater;

% The following code was used to create the required plots

19
% figure
% plot(freq,db(estimater(:,1)))
% xlabel('frequency [Hz]','FontSize',16)
% ylabel('amplitude FRF [dB]','FontSize',16)
% figure
% plot(freq,angle(estimater(:,1)))
% xlabel('frequency [Hz]','FontSize',16)
% ylabel('phase [radian]','FontSize',16)
% figure
% plot(freq,gamma(:,1))
% xlabel('frequency [Hz]','FontSize',16)
% ylabel('coherence function','FontSize',16)

% input = matrix(:,1,1);
% output = matrix(:,2,1);
%
% figure
% plot(1:Nt,input,1:Nt,output)
% xlabel('time [s]','FontSize',16)
% ylabel('input and output [V]','FontSize',16)

end

FDD.a_fre = freq;

EST_Polymax = Polymax_LSFD(FDD)

% plot(freq,gamma(:,1))
% figure
% plot(freq,db(estimater(:,1)))
%

.1.2 Calculating the H1 estimater for the frequency re-

sponse

function [ H1 , gamma ] = H1( matrix )

% Function that returns the H1 estimater for the FRF and also the gamma
% value to check its realibility

[Nt x y] = size(matrix);
Nb = 3; % Number of averages
u1 = zeros(Nt,1,1);

Fm = 13.1072e6; % Hz: internal time base NI-9234 (Fix value!)

K = 25; % Can be modified for changing the sampling rate

20
Fs = Fm/(256*K); % Sampling frequency of NI DAQ frequency Hz

Nf = (Nt/2+1); % Number of points in the frequency domain

freq = Fs/2*linspace(0,1,Nf); % creating the x-axis (see help fft)

for i=1:Nb
u1(:,i) = matrix(:,2,i); % Accelerometer 1, three measurements
u2(:,i) = matrix(:,3,i); % Accelerometer 2, three measurements
u3(:,i) = matrix(:,4,i); % Accelerometer 3, three measurements
f(:,i) = matrix(:,1,i); % Force, three measurements

% Taking fft of force and taking only half the domain (as explained on
% paper)
F(:,i) = fft(f(:,i));
Fshort(:,i) = F(1:Nf,i);

% The same for the acceloremeters

U1(:,i) = fft(u1(:,i));
U2(:,i) = fft(u2(:,i));
U3(:,i) = fft(u3(:,i));
U1short(:,i) = U1(1:Nf,i);
U2short(:,i) = U2(1:Nf,i);
U3short(:,i) = U3(1:Nf,i);

% Auto/cross-power spectra as in page 6 of the guide

Gff(:,i) = (Fshort(:,i).*conj(Fshort(:,i)));
Gu1f(:,i) = (U1short(:,i).*conj(Fshort(:,i)));
Gu2f(:,i) = (U2short(:,i).*conj(Fshort(:,i)));
Gu3f(:,i) = (U3short(:,i).*conj(Fshort(:,i)));
Guu1(:,i) = (U1short(:,i).*conj(U1short(:,i)));
Guu2(:,i) = (U2short(:,i).*conj(U2short(:,i)));
Guu3(:,i) = (U3short(:,i).*conj(U3short(:,i)));
end

%% Averages of the 3 measurements ( 3 hits of the hammer)

Gff_avg = mean(Gff,2);
Gu1f_avg = mean(Gu1f,2);
Gu2f_avg = mean(Gu2f,2);
Gu3f_avg = mean(Gu3f,2);
Guu1_avg = mean(Guu1,2);
Guu2_avg = mean(Guu2,2);
Guu3_avg = mean(Guu3,2);

gamma_col1 = sqrt(abs(Gu1f_avg).^2./(Guu1_avg.*Gff_avg));
gamma_col2 = sqrt(abs(Gu2f_avg).^2./(Guu2_avg.*Gff_avg));
gamma_col3 = sqrt(abs(Gu3f_avg).^2./(Guu3_avg.*Gff_avg));

21
gamma = [gamma_col1 gamma_col2 gamma_col3];

%% H1 estimater
H1 = [Gu1f_avg./Gff_avg Gu2f_avg./Gff_avg Gu3f_avg./Gff_avg];

end

.1.3 Calculate and plotting the Modal Assurance Criterion

function [ mac ] = MAC( modes )

% modes 58x16

[x Nmodes] = size(modes);

for j=1:Nmodes
ms = modes(:,j); % load mode s
for l=1:Nmodes
mr = modes(:,l); % load mode r
mac(j,l) = ((abs(mr'*ms)^2))/((mr'*mr)*(ms'*ms));
% Comparing the two by using formula (3) on page 15
end
end

% Creating a bar plot with appropriate colours

b = bar3(mac);
colorbar
for k = 1:length(b)
zdata = get(b(k), 'Zdata');
set(b(k),'CData',zdata);
set(b,'EdgeColor','k');
end

.1.4 Plotting the different modes

function [ output_args ] = plotmode( m, m_nr )

% Function that plots the displacement of a certain mode
% Input : m = modes; m_nr is the number of the mode to plot
% m is a matrix 58 x 16
% First beam: points 1 to 12

for i = 1:6
% The displacement can not be a complex
% number. We have to take the amplitude of these numbers and use the
% sign of the real part. d stands for displacement

dx1(i) = sign(real(m(2*i,m_nr)))*abs(m(2*i,m_nr));

22
 dx2(i) = sign(real(m(2*(i+6),m_nr)))*abs(m(2*(i+6),m_nr));
dy1(i) = sign(real(m(2*i-1,m_nr)))*abs(m(2*i-1,m_nr));
dy2(i) = sign(real(m(2*(i+6)-1,m_nr)))*abs(m(2*(i+6)-1,m_nr));
dx3(i) = sign(real(m(2*(i+12),m_nr)))*abs(m(2*(i+12),m_nr));
dx4(i) = sign(real(m(2*(i+18),m_nr)))*abs(m(2*(i+18),m_nr));
dy3(i) = sign(real(m(2*(i+12)-1,m_nr)))*abs(m(2*(i+12)-1,m_nr));
dy4(i) = sign(real(m(2*(i+18)-1,m_nr)))*abs(m(2*(i+18)-1,m_nr));
end

%% Plotting the displacements for each beam

figure
plot(dx1,(1:6))
xlabel('Displacement in the x-direction','fontsize',14)
ylabel('Point number','fontsize',14)
hold on
plot(dx2,(1:6),'r')
plot(dx3,(1:6),'y')
plot(dx4,(1:6),'g')
hold off
figure
plot(dy1,(1:6))
xlabel('Displacement in the y-direction','fontsize',14)
ylabel('Point number','fontsize',14)
hold on
plot(dy2,(1:6),'r')
plot(dy3,(1:6),'y')
plot(dy4,(1:6),'g')
hold off

end

.2 Mode shapes

23
 (a) mode 1-x (b) mode 1-y

(c) mode 2-x (d) mode 2-y

(e) mode 4-x (f) mode 4-y

Figure 1: Mode shapes 1,2 and 4 of the structure for the x direction and y
direction. Each colour on the plot represents a different beam.

24
 (a) mode 5-x (b) mode 5-y

(c) mode 7-x (d) mode 7-y

(e) mode 8-x (f) mode 8-y

Figure 2: Mode shapes 5,7 and 8 of the structure for the x direction and y
direction. Each colour on the plot represents a different beam.

25
 (a) mode 9-x (b) mode 9-y

(c) mode 10-x (d) mode 10-y

Figure 3: Mode shapes 9 and 10 of the structure for the x direction and y
direction. Each colour on the plot represents a different beam.

26