EIGEN VALUE ANALYSIS, EXPERIMENTAL MODAL

ANALYSIS, TIME HISTORY ANALYSIS BY NUMERICAL

METHODS OF 4DOF SHARE BUILDING STRUCTURE
SUBJECTED TO RANDOM VIBRATION

By
Khawaja Ali
(15RB909)
Structural Engineering laboratory

ASSIGNMENT#2
Submitted to: Associate Prof. Mayuko NISHIO
Course Name: Engineering Structural Dynamics (RB00025)

Civil Engineering Department

Yokohama National University Japan

Contents
1.

INTRODUCTION ..................................................................................................................................... 4

2.

SIMPLIFIED MODELS ............................................................................................................................. 4

2.1 About the 4DOF structure................................................................................................................... 4
2.2 Material and properties ...................................................................................................................... 5
2.3 Geometry ............................................................................................................................................ 5
2.3 Experimental setup ............................................................................................................................. 5
2.4 Data Acquisition .................................................................................................................................. 5

Q (1) MODELING PROCEDURE FOR 4DOF STRUCTURE BY LUMPED MASS................................................... 6

Calculations of M & K matrices ................................................................................................................. 7
Calculation of Stiffness.......................................................................................................................... 7
Q (2) CALCULATION OF THE VIBRATION MODES AND RESONANT FREQUENCIES ....................................... 8
Eigenvectors .............................................................................................................................................. 9
Eigen values .............................................................................................................................................. 9
Q (3) STEP BY STEP TIME HISTORY ANALYSIS ALGORITHM, DIFFERENCE BTW BOTH METHODS. ............. 10
Time-integration procedure (Newmark-beta) ........................................................................................ 10
Time-integration procedure (4th-order Runge-Kutta)............................................................................ 11
Comparison of both methods ................................................................................................................. 12
Q (4) FFT SPECTRA OF BOTH INPUT AND OUTPUT ..................................................................................... 13
Assigned Values for FFT spectra: For sensor number 4 .......................................................................... 13
FFT Spectra of Input ................................................................................................................................ 13
FFT Spectra of Output ............................................................................................................................. 14
Comparison of different windows .......................................................................................................... 15
Windows Comparison ............................................................................................................................. 15
By changing sampling frequency and FFT length.................................................................................... 16
Q (5) AUTOCORRELATION FUNCTION AND POWER SPECTRAL DENSITY .................................................... 17
Autocorrelation of input and output ...................................................................................................... 17
Power Spectral Density Based on FFT Approach .................................................................................... 17
Power Spectral Density Based on pwelch Approach .............................................................................. 18
Q (6) FREQUENCY RESPONSE FUNCTION (FRF)........................................................................................... 19
Q (7) MODAL PROPERTIES OF 4DOF STRUCTURE FROM FRF ..................................................................... 21
Half Power Method ................................................................................................................................. 21
2

Comparison of modal properties from FRF with numerical modal analysis........................................... 24

Q (8) RELATIONSHIP BETWEEN STRUCTURAL PROPERTIES CHANGES AND DYNAMIC RESPONSE CHANGES
.................................................................................................................................................................... 25
Q (9) SENSOR, ITS PRINCIPLE AND SPECIFICATION ..................................................................................... 26
Principle of Piezoelectric sensor ............................................................................................................. 27
CONCLUSION............................................................................................................................................... 28
REFERENCES ................................................................................................................................................ 29

1. INTRODUCTION
The vibration analysis is required in many engineering applications such as calculation and
design of structures and mechanical devices, mechanical maintenance or predictions of failure.
One of the fundamental aspects of vibrations analysis is to obtain the natural frequencies of the
system studied. This analysis becomes relevant not only for the prediction of displacements, and
the related deformations and tensions, but also for their control. These aspects influence the
resistive calculation of components, as well as affect the use of them. Hence, the displacements
caused by dynamic loads need to be reduced. There are a great number of situations in which it is
possible to reduce, but not to eliminate the dynamic forces that excite a system inducing a
vibratory behavior on it. Loads produced by an earthquake or by wind are unpredictable
variables that subject structures to variable dynamic loads. The resulting displacements damage
the structure and, in many cases it could not be restored. These reasons make the study of
vibrations in structures and the control of their behavior under dynamic loads an interesting
study.

2. SIMPLIFIED MODELS
2.1 About the 4DOF structure
The target structure in this work is a four-story share building structure (1 base floor+1st
floor+2nd floor+3rd floor) as shown in figure1. It consists of four floor plates and twelve columns
assembled by bolted joints. In each joint, a small block is placed to reduce the stress
concentration. In addition, a suspended column and a bumper are attached on the 3rd and 2nd floor
plates, respectively; however, this impact mechanism do not work in this study; i.e. just treated
as additional mass.

Figure 1 4DOF share building structure

2.2 Material and properties

This physical model could represent a prototype building made of the same material, i.e. the
density () and Young modulus (E) do not vary. In the present analysis, the properties for the
structural aluminum are taken as: = 2700 kg/m3 and E = 70 GPa.

2.3 Geometry
Members
Plate for a floor
Columns
Screw block at each joint
Bumper on 2nd floor
Suspended column on 3rd floor

Geometry (cm)
30.5x30.5x2.5
17.7x2.5x0.6
-------

Mass (kg)

0.04
0.2
0.26

Basic dimensions of the 4DOF structure are shown in Figure2.

2.3 Experimental setup

The structure slides on rails that allow movement in one direction (the x-direction only). The
slide mechanism is completely smooth. An electromagnetic shaker provided a lateral excitation
to the base floor along the centerline of the structure. The structure and shaker were mounted
together on a common base plate and the entire system rests on rigid foam. The foam was
intended to minimize extraneous sources of unmeasured excitation from being introduced
through the base of the system. A force transducer (Channel #1 in Figure2) with a nominal
sensitivity of 2.2mV/N was attached at the end of a stinger to measure the input force from the
shaker to the structure.
To measure the structures response, four accelerometers (from Channel#2 to Channel#5 in
Figure2) with nominal sensitivities of 1000mV/g were attached at the centerline of each floor on
the opposite side from the excitation location. A Dactron Spectra book dynamic data acquisition
system was used to collect data.

2.4 Data Acquisition

The input force was the band-limited (20-200Hz) random signal applied to the base floor. The
outputs are accelerations in all floors. The data are with the sampling frequency of 640Hz, and
the length of data was 4096 points.

Figure 2 Basic dimensions of the 4DOF structure (cm)

Q (1) MODELING PROCEDURE FOR 4DOF STRUCTURE BY LUMPED

MASS
The lumped mass modeling strategy was used in this analysis for 4DOF share building
structure.
In this lumped mass strategy, the inertial force is centralized on mass m; it is thus suitable to
use this modeling when the structure has characteristic in its mass distribution; e.g., share
building, tower structures, wind turbine etc.
Each mass is calculated by using density , cross sectional area A and length of each element
L.

Figure 3 Shear building model of the test bed structure

Calculations of M & K matrices

m1 (Base Floor) = 6.73 kg
m2 (1st Floor) = 6.73 kg
m3 (2nd Floor) = 6.93 kg
m4 (3rd Floor) = 6.84 kg
Calculation of Stiffness:

Consider a column of effective length with boundary

conditions,
Applying Beam flexural theory,
Figure 4 calculation of
Stiffness

2
=
2
+ = 0
2

+
=0
2

2
Integrating twice and we get,
2 3
=

+ 1 + 2
4
6

Applying initial boundary conditions, we get the general deflection formula.

=

2
3 2
12

When x=L then y will be maximum

3
=
12
So the Stiffness (influence coefficient) of the column can be written as
=
, =

12
3

3 0.025 0.006
=
12
12

= 0.177 0.025 = 0.152

7

= 4.5 1010 4

12 70 109 4.5 1010

=
=
107636.8
0.1523

1 = 10 , 2 , 3 , 4 = 4 = 430547.2
The Equation of motion of 4DOF system by using Lagranges Equations is given
by
1
0
0
0

0
2
0
0

0 0
0 0
3 0
0 4

1 + 2
1
2
2
+
3
0
4
0

2
2 + 3
3
0

0
0
3 0
3 + 4 4
4 4

1
1
2
2
3 = 3
4
4

Here,
6.73
0
6.73
= 0
0
0
0
0

Figure 5 Simplified
model

0
0
0
0 kg
6.93 0
0 6.84

4.3056 4.3055 0
0
0 N/m
= 1.0 + 005 4.3055 8.6109 4.3055
0
4.3055 8.6109 4.3055
0
0
4.3055 4.3055
95.1321
6.6530 41.0988 47.2958
61.8723 14.4626 40.6744
, = 6.6530
41.0988 14.4626
63.9620 8.3179
47.2958 40.6744
8.3179 96.3686

Q (2) CALCULATION OF THE VIBRATION MODES AND RESONANT

FREQUENCIES
The analytic model used to estimate vibration modes of the proposed structure is a simplification
of the continuous three-dimensional model. The structure is reduced to a plane mass-spring
model whose components are detailed in Figure 5.
The frequencies (fi = , being wi the eigenvalues) and vibration modes (their related
eigenvectors, = { i}) of the problem are computed assuming free vibrations (i.e., without
external load Q) as follow:
= 0

3.5

3.5

3.5

3.5

2.5

2.5

2.5

Floor

Floor

Floor

Floor

Table 1 summarizes the analytic results obtained for frequency and Eigen modes (which are
schematically represented in figure 6.

2.5

1.5

1.5

1.5

1.5

1
-0.5 0 0.5
1st mode

1
-0.5 0 0.5
2nd mode

1
-0.5 0 0.5
3rd mode

1
-0.5 0 0.5
4th mode

Figure 6 Eigen Mode shapes

Eigenvectors = { i} are as follows (from MATLAB)

0.1916 0.2523 0.1910 0.1083
= 0.1916 0.1058 0.1865 0.2566
0.1916 0.1021 0.1953 0.2429
0.1916 0.2490
0.1935 0.1002

Eigen values are as follows

12

=0

22

= 37140

, 32

= 126470

, 42

= 215520

Table 1 Results of Eigen Problem Analysis

Modes Resonant Frequencies (Hz) Initially damping ratios taken (%)

f1
0.0964
1
f2
30.6716
6
f3
56.5995
2
f4
73.8858
1
9

Q (3) STEP BY STEP TIME HISTORY ANALYSIS ALGORITHM,

DIFFERENCE BTW BOTH METHODS.
Time-integration procedure (Newmark-beta)
gamma=0.5;
beta=0.25;
num_Acc=zeros(length(ntime),4);
%%% This is just for mamory
%%% Coefficients
a0=1/(beta*(dt^2))7;
a1=gamma/(beta*dt);
a2=1/(beta*dt);
a3=(1/(2*beta))-1;
a4=(gamma/beta)-1;
a5=(gamma/(2*beta))-1;
a6=gamma*dt;
a7=(1-gamma)*dt;
%%% Effective stiffness for linear system (=constant)
%%% Notice that if in the nonlinear system (= not constant K, M, C), this
procedure should be
%%% included in the iteration
Keff=a0*M+a1*C+K;
iKeff=inv(Keff);
%%% Start of ineration
for t=0:dt:t_end
i=i+1;
%%% step
ftdt=[nForce(i) 0 0 0]';
num_Acc(i,:)=ya';
yt=y;
yvt=yv;
yat=ya;
Reff=ftdt+M*(a0*yt+a2*yvt+a3*yat)+C*(a1*yt+a4*yvt+a5*yat);
y=iKeff*Reff;
ya=a0*(y-yt)-a2*yvt-a3*yat;
yv=yvt+a6*yat+a7*ya;
%%%%% convergence check based on displacement
for s=1:100 %%% Max iteration
ys=inv(K)*(-C*yv-M*ya+ftdt);
es=abs(ys-y);
if es <= 0.000000001 %%% convergence condition
break
elseif es > 0.000000001
y=ys;
ya=a0*(y-yt)-a2*yvt-a3*yat;
yv=yvt+a6*yat+a7*ya;
end
end

10

Force(N)
Acc.(m/s 2) Acc.(m/s 2) Acc.(m/s 2) Acc.(m/s 2)

Time-integration procedure (Newmark-beta)

100
0
-100

1st floor
20
0
-20

3
2nd floor

20
0
-20

3
3rd floor

20
0
-20

3
4th floor

20
0
-20

3
time(s)

Figure 7 Time History Response (Newmark-beta method)- Overlay experimental data (black)
Circles indicate the maximum responses of random events

Time-integration procedure (4th-order Runge-Kutta)

%%%%% State-space discription of equation of motion
I=eye(n,n);
O=zeros(n,n);
A=[O I;-inv(M)*K -inv(M)*C];
A=sparse(A);
%%% this is just for saving memory
B=[O O;O inv(M)]
dPP=zeros(length(ntime),8);
%%% This is just for save mamory
for t=0:dt:t_end
i=i+1;
ft=[nForce(i) 0 0 0]';
Bu=[0; 0; 0; 0; inv(M)*ft];
dPP(i,:)=(A*p+Bu)';
pt=p;
d1=dt.*(A*pt+Bu);
p2=pt+d1/2;
d2=dt.*(A*p2+Bu);
p3=pt+d2/2;
d3=dt.*(A*p3+Bu);
p4=pt+d3;
d4=dt.*(A*p4+Bu);

11

Acc.(m/s 2) Acc.(m/s 2) Acc.(m/s 2) Acc.(m/s 2)

Force(N)

p=pt+(d1+d2*2+d3*2+d4)/6;
end
%%%%% Pick up acceleration
num_Acc=dPP(:,5:8);
%%% p=[y;y'] dp=[y';y'']

Time-integration procedure (4th-order Runge-Kutta)

100
0
-100

1st floor
20
0
-20

3
2nd floor

20
0
-20

3
3rd floor

20
0
-20

3
4th floor

20
0
-20

3
time(s)

Figure 8 Time History Response (Runga-kutta method)-Overlay experimental data (black)

Circles indicate the maximum responses of random events

Comparison of both methods

It is concluded that the Newmark method with an adequate check on the solution
accuracy could give qualitatively the same results as the Runge-Kutta method. Both the methods
apply the error correction on the displacements keeping the time-step constant. The Newmark
method produces results which are comparable and almost same qualitatively with the results
obtained by Runge Kutta simulation. Moreover, Newmark method is unconditionally stable
method.

12

Q (4) FFT SPECTRA OF BOTH INPUT AND OUTPUT

A fast Fourier transform (FFT) is an algorithm that computes the discrete Fourier transform
(DFT) of a sequence, or its inverse. Fourier analysis converts a signal from its original domain
(often time or space) to a representation in the frequency domain and vice versa.

Assigned Values for FFT spectra: For sensor number 4

FFT length = 4096
Sampling Frequency = 640 Hz, Nyquist Frequency=320Hz
Windowing used= Hanning
Data= Data_random

FFT Spectra of Input

Figure 9 shows the effect of windowing (hanning) on FFT input spectra. Magnitudes of FFT
almost become half after applying windowing.
FFT Spectrum of input force before windowing

FFT force (N)

6000

4000

2000

50

50

100

150
200
250
300
Frequency (Hz)
Windowed (Hanning) FFT Spectrum of Input Force

350

FFT force (N)

3000

2000

1000

100

150
200
Frequency (Hz)

250

300

Figure 9 FFT Spectra of Input (before and after windowing)

13

350

FFT Spectra of Output

Figure 10 shows the effect of windowing (hanning) on FFT output spectra. Magnitudes of FFT
reduce after applying windowing and peaks become more sharp and clear.

FFT Response Spectrum before windowing

Fourier Fesponse

4000
3000
2000
1000
0

50

50

100

150
200
250
300
Frequency (Hz)
Windowed (Hanning) FFT Response Spectrum

350

Fourier Response

3000

2000

1000

100

150
200
Frequency (Hz)

250

300

Figure 10 FFT Spectra of output before and after windowing

14

350

Comparison of different windows

Among different windows i.e. hanning, hamming, blackman-harris and Kaiser, blackman-harris window
give minimum Fourier response and sharper peaks and small side lobes with less width of main lobe.

Fourier Response

Fourier Response

Windowed (Hanning) FFT Response Spectrum

Windowed (Hamming) FFT Response Spectrum
3000
3000

2000

1000

1000

0
100
200
300
400
0
100
200
300
400
Frequency (Hz)
Frequency (Hz)
Windowed (Blackman-Harris) FFT Response Spectrum
Windowed (Kaiser) FFT Response Spectrum
3000
4000
0

Fourier Response

Fourier Response

2000

2000

1000

100
200
300
Frequency (Hz)

3000
2000
1000
0

400

100
200
300
Frequency (Hz)

400

Figure 11 FFT response spectra of output for different windows

Windows Comparison
Windows Comparison
1
Hanning
Hamming
Blackman-Harris
Kaiser

0.9
0.8

Amplitude

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

3
4
Time (sec)

Figure 12 Comparisons of Windows

15

By changing sampling frequency and FFT length

When Sampling Frequency is changed from 640Hz to 160Hz then data becomes less accurate
(more disturbed). Chances of Aliasing effects are increased which can affect results as shown in
figure 13. And when max frequency is lesser than Nyquist frequency then aliasing effects are
seen in figure 14. Aliasing occurs because signal frequencies can overlap if the sampling
frequency is too low. This is one of the important concepts of sampling theory.
Windowed (Hanning) FFT Spectrum of Input Force

FFT force (N)

3000

2000

1000

50

50

100

150
200
250
Frequency (Hz)
Windowed (Hanning) FFT Response Spectrum

300

350

300

350

Fourier Response

1500

1000

500

100

150
200
Frequency (Hz)

250

Figure 13 FFT response for Sampling Freq = 160Hz

FFT Spectrum of Input Force

FFT force (N)

800
600
400
200
0

50

100

50

100

150
200
250
Frequency (Hz)
FFT Response Spectrum

300

350

300

350

Fourier Response

300

200

100

150
200
Frequency (Hz)

250

Figure 14 Aliasing Effect when max freq is less than Nyquist Freq

16

Q (5) AUTOCORRELATION FUNCTION AND POWER SPECTRAL

DENSITY
Autocorrelation of input and output
Autocorrelation can assist in detecting periodicities. Above graph in figure 15 shows the
autocorrelation of input signals and below one show autocorrelation of output signals
(correlation is strong at lower length of ACF).
5

Autocorrelation

Auto-Correlation Function for Input

x 10

15
10
5
0
-5

500

1000

500

1000

x 10

Autocorrelation

1500

2000
2500
3000
3500
Length of ACFx
Auto-Correlation Function for Output

4000

4500

4000

4500

0.5
0
-0.5
-1

1500

2000
2500
Length of ACFy

3000

3500

Figure 5 Autocorrelation of input and output

Power Spectral Density Based on FFT Approach

Figure 16 shows power spectral density graphs for input and output data based on FFT for
random vibrations. Output PSD shows two peaks which may represent the dominating modes of
vibration.
Spectral Density (db/Hz -1)
Spectral Density (db/Hz -1)

FFT Approach:Power Spectral Density for Input

x 10

50
6

x 10

100

150
200
250
300
Frequency (Hz)
FFT Approach:Power Spectral Density for Output

350

50

100

150
200
Frequency (Hz)

250

300

Figure 6 PSD based on FFT approach

17

350

Power Spectral Density Based on pwelch Approach

Figure 16 shows power spectral density graphs for input and output data based on pwelch for random
vibrations. Output PSD shows two peaks which may represent the dominating modes of vibration.

Spectral Density (Hz -1)

Spectral Density (Hz -1)

pwelch Approach:Power Spectral Density for Input

80
60
40
20
0

50

100

150
200
250
300
Frequency (Hz)
pwelch Approach:Power Spectral Density for Output

50

100

350

40
30
20
10
0

150
200
Frequency (Hz)

250

300

350

Figure 17 PSD based on pwelch approach

The power spectral density and autocorrelation functions are related by Fourier Transform pair
as shown in above graphs
= ( )
This pair is written as,

And

These two relations are called as Wiener-Khintchine relations. These relations are very useful in
random signal processing.

18

Q (6) FREQUENCY RESPONSE FUNCTION (FRF)

When acceleration input data measured by sensor#4 is used to find the frequency response
function then predominant frequency was found to be 71.2Hz.
Frequency Response Function

Amplitude dB

2
1.5

Predominant frequency
fn=71.2Hz

1
0.5
0

50

100

150
200
250
Frequency (Hz)
Frequency Response Function

50

100

300

350

300

350

Phase (degree)

4
2
0
-2
-4

150

200

250

Figure 18 FRF of data measuredFrequency

with Accelerometer
(Channel 5) Sensor#4
(Hz)

When acceleration input data measured by sensor#3 is used to find the frequency response
function then predominant frequency was found to be 54.4Hz.
Frequency Response Function

Predominant frequency
fn=54.2Hz

Predominant frequency
fn=71.1Hz.

50

100

150
200
250
Frequency (Hz)
Frequency Response Function

50

100

300

350

300

350

4
Phase (degree)

Figure 19 FRF of data

measured with
Accelerometer
(Channel 4) Sensor#3

Amplitude dB

2
0
-2
-4

19

150
200
Frequency (Hz)

250

When acceleration input data measured by sensor#3 is used to find the frequency response
function then predominant frequency was found to be 30.5Hz. The magnitude plot clearly
exhibits the resonances. The phase plot reflects the existence of a vibration mode from its visible
phase change.
Frequency Response Function

Amplitude dB

Predominant frequency
fn=30.5Hz

50

100

150
200
250
Frequency (Hz)
Frequency Response Function

50

100

300

350

300

350

Phase (degree)

4
2
0
-2
-4

150
200
Frequency (Hz)

250

Figure 20 FRF of data measured with Accelerometer (Channel 3) Sensor#2

20

Q (7) MODAL PROPERTIES OF 4DOF STRUCTURE FROM FRF

Half Power Method
The modal damping ratios are evaluated employing Half Power Method adapted to the Fourier
spectrum. Figure 21 shows the evaluation method of damping ratio using Half Power Method on
Fourier spectrum of data. Using fn as the n-th modal natural frequency identified by above
method, the n-th modal damping ratio n is given by f /(2fn) where f is the band width in
frequency domain as half power. Rmax is the amplitude of the n-th modal natural frequency fn on
Fourier spectrum.
The peak of the frequency response curve occurs at 71.2Hz.
Frequency Response Function

Rmax=1.93

Amplitude dB

1.5

Rmax/(2)=1.36
1

0.5

0
66

68

70 fa fn fb72
Frequency (Hz)

74

76

78

Figure 21 Half Power Method for Sensor#4 data

Assume that damping is small, the natural frequency f4=71.2Hz.

The peak value of the FRF response is 1.93. Draw a horizontal line at 9.3/ 2 as shown in
above graph. This line intersects the frequency response curve at fb=71.8Hz and fa=70.8.
Therefore,
4 = ( )/24
21

4 = 71.2
4 =

71.8 70.8
= 0.007 = 0.7%
2 71.2

Similarly, half power method can be used to find the damping ratios for Sensor #3 and
Sensor#2 data.

Frequency Response Function

2.5

Rmax=2.1

Amplitude dB

``

1.5

Rmax/(2)=1.48
1

0.5

50

52

fa54 fn fb

56
Frequency (Hz)

58

Figure 22 Half Power Method for Sensor#3 data

3 = ( )/23
3 = 54.2
3 =

55 53.8
= 0.0111 = 1.11%
2 54.2

22

60

Frequency Response Function

1
0.9
0.8

Rmax=0.76
Amplitude dB

0.7
0.6
0.5 Rmax/(2)=0.54
0.4
0.3
0.2
0.1
0
20

fa30fnfb

25

35
Frequency (Hz)

40

Figure 23 Half Power Method for Sensor#2 data

2 = ( )/22
2 = 30.5
2 =

31 29.5
= 0.0246 = 2.46%
2 30.5

23

45

Comparison of modal properties from FRF with numerical modal analysis

Table 2 Comparison of modal properties of 4DOF structure from FRF and numerical analysis in Q (2)
Natural frequency (Hz)

Modal damping ratios

Modes

fn (Hz) using
(Eigen Value
Analysis)

fn (Hz) using Half

Power
Method/FRF

Error
(%)

Initially
Assumed
Damping Ratio
(%)

Damping Ratio (%)

using Half Power
Method/FRF

Error
(%)

f1
f2
f3
f4

0.0964
30.6716
56.5995
73.8858

0
30.5
54.2
71.2

0
0.559
4.2
3.63

1
6
2
1

0
2.46
1.11
0.7

0
59
44.5
30

Table 2 shows the natural frequencies and modal damping ratios measure by using numerical
analysis (Eigen value analysis) and frequency response function FRF. It also clearly shows the
errors (percentage difference) in values. Natural frequencies are almost same by both methods
but as far as damping ratios are concerned; there is a large error value between initially taken
damping ratios in numerical analysis and calculated damping ratios by FRF curve using Half
Power Method. This difference may be due to invisible noise signals.
Only FRF curves have been used to find modal properties by Half Power Method in this study
but not phase curves because with the presence of damping, the change of phase is less than clear
cut. In this case, the phase plot is not a useful source of information for identifying vibration
modes. Thats why, FRF response curves have been used for finding the modal properties.
Moreover, FRF is fft(output)/fft(input), so frequency resolution is also an important matter of
concern.

24

Q (8) RELATIONSHIP BETWEEN STRUCTURAL PROPERTIES

CHANGES AND DYNAMIC RESPONSE CHANGES
Mode shapes of a structure are very important to MDOF structural dynamics because they are
essential to perform modal analysis. Modal participation factors and effective modal mass are
calculated from the mode shapes of the structures.
The dynamic response of a structure is dependent upon the two very important parameters which
are mass and stiffness of the structure. If these two parameters change, the whole dynamic
response of structure can change accordingly. For example, if there is a crack in a beam element
in a structure, due to presence of that crack, its local effective stiffness will be reduced with
negligible loss of mass and deflection will be larger than before. This leads to reduction of
resonant frequencies.
Modes shapes and natural frequencies are functions of the structural properties and boundary
conditions. If structural properties change, natural frequencies change accordingly but mode
shapes may not necessarily change. But if boundary conditions of the structural system change,
then natural frequencies and mode shapes both change.
When excitation is very simple then response can easily be determined but when excitation is a
random vibration then it is impossible to find the response by solving linear equations of motion.
In that case, modal analysis is compulsory option but if the displacement is beyond the elastic
range then the response of the structure behaves nonlinearly. Therefore, the only method to
obtain the accurate solution is a numerical method. The advantage of using the average
acceleration method is its unconditional stability. In the numerical analysis, the accuracy of the
solution is affected by the time steps in the analysis.
Figure 24 shows the time history response of 4DOF structure (overlap of sensor data and by
numerical method-Newmark beta method) by using newly calculated damping ratios as shown in
table 2. This figure 24 can be compared with figure 7. By looking at both figures, we come to
know that by using the newly found damping ratios (0%, 2.46%, 1.11% and 0.7%), response
calculated by numerical method is increased and it makes sense that lesser the damping, larger
the response. So, dynamic response of structure is dependent upon the structural properties.

25

Acc.(m/s 2)
Acc.(m/s 2)

20
0
-20

Acc.(m/s 2)

1st floor
20
0
-20

20
0
-20

Acc.(m/s 2)

Force(N)

Time-integration procedure (Newmark-beta)

100
0
-100

20
0
-20

3
2nd floor

3
3rd floor

3
4th floor

3
time(s)

Figure 24 Response of random loading for calculated damping ratios (determined from FRF)

Q (9) SENSOR, ITS PRINCIPLE AND SPECIFICATION

Accelerometers are sensing transducers that provide an output proportional to acceleration,
vibration, and shock. The piezoelectric accelerometer is the most popular class of these devices;
other types are based on piezoresistive, capacitive, and servo technologies. These self-generating
devices are characterized by an extended flat-frequency response range, a large linear amplitude
range, and excellent durability.
High stability similar to quartz accelerometers is achieved by means of an artificial aging process
of the piezoceramic sensing element. The sensitivity of ceramics compared to quartz materials is
about 100 times higher. Therefore, piezoceramic accelerometers are the better choice at low
frequencies and low acceleration. Piezoelectric accelerometers are widely accepted as the best
choice for measuring absolute vibration. Compared to the other types of sensors, piezoelectric
accelerometers have important advantages:

Extremely wide dynamic range, almost free of noise - suitable for shock measurement as well
as for almost imperceptible vibration
Excellent linearity over their dynamic range
26

Wide frequency range, high frequencies can be measured

Compact yet highly sensitive
No moving parts - no wear
Self-generating - no external power required
Great variety of models available for nearly any purpose
Integration of the output signal provides velocity and displacement

Principle of Piezoelectric sensor

The sensing element of a piezoelectric
accelerometer consists of two major parts:

Piezoceramic material

Seismic mass

One side of the piezoelectric material is connected to a rigid post at the sensor base. The so -called
seismic mass is attached to the other side. When the accelerometer is subjected to vibration, a
force is generated which acts on the piezoelectric element. According to Newtons Law this force
is equal to the product of the acceleration and the seismic mass. By the piezoelectric effect a
charge output proportional to the applied force is generated. Since the seismic mass is constant the
charge output signal is proportional to the acceleration of the mass
A piezoelectric accelerometer can be regarded as a
mechanical low-pass with resonance peak. The seismic
mass and the piezoceramics (plus other "flexible"
components) form a spring mass system. It shows the
typical resonance behavior and defines the upper
frequency limit of an accelerometer. In order to achieve
a wider operating frequency range the resonance
frequency should be increased. This is usually done by Figure 25 shows a typical frequency response
reducing the seismic mass. However, the lower the curve of an accelerometer when it is excited by
a constant acceleration
seismic mass, the lower the sensitivity. Therefore, an
accelerometer with high resonance frequency, for
example a shock accelerometer, will be less sensitive whereas a seismic accelerometer with high
sensitivity has a low resonance frequency. Specification of Model: 393B04 is given below.

27

CONCLUSION
Natural frequencies and damping ratios were evaluated by modal analysis of 4DOF system. In
first mode of vibration, there is no deformation of structure but structure as a whole is shifting
from its base to some other location so its natural frequency and damping ratios are nearly zero.
The amplitude vs frequency plot of an FRF is sometimes perceived to be the crucial plot for
identifying vibration modes as it manifests resonance peaks. This fallacy does not accord with
two scenarios: (1) not every mode will appear on every measured FRF; and (2) not every peak of
an FRF is an authentic mode.
The quality of modal analysis relies critically on the quality of the measured FRF Data. The
modal properties derived from erroneous FRF data are susceptible to unacceptable errors. The
assessment of measured FRF data is basically to ascertain two things: (1) the structure satisfies
the assumptions modal analysis requires; and (2) human and system errors are minimized or
eliminated.
In general, to preserve the full information in the signal, it is necessary to sample at twice the
maximum frequency of the signal. This is known as the Nyquist rate.
Sometimes the highest frequencies components of a signal are simply noise, or do not contain
useful information. To prevent aliasing of these frequencies, we can filter out these components
before sampling the signal. Because we are filtering out high frequency components and letting
lower frequency components through, this is known as low-pass filtering.
The effect of physical parameter changes of a structural system on its dynamic properties which
are in the forms of natural frequencies and mode shapes are very important to carry out structural
modification.

28

REFERENCES
1. https://www1.udel.edu/biology/rosewc/kaap686/notes/windowing.html

2. Communication Systems I By J.S.Chitode

3. Modal Analysis by Jimin He and Zhi-Fang Fu
4. Digital Signal Processing, Part 3, Discrete-Time Signals & Systems Case Studies by S R
Taghizadeh
5. Experimental vibration analysis for a 3D scaled model of a three-floor steel structure

Ernesto F. Castillo; Marcela A. Cruchaga

29