frequency response functions and shapes, then an Matlab

PROCESSING OF OPERATING application for shape visualisation is introduced and finally,

some results of the experiment are presented.
DEFLECTION SHAPES
PAVEL SURANEK, JIRI CZEBE, VACLAV PAVELKA, RENATA 2 FREQUENCY RESPONSE FUNCTION
WAGNEROVA Frequency response function (FRF) represents the relation of
Dep. of Control Systems ans Instrumentation, Faculty of the output and the input signal in a frequency domain. FRFs
Mechanical Engineering, VSB-Technical University of between measured points pairs are necessary for the ODS
Ostrava, Ostrava, Czech Republic visualization [Herfert 2015], [Yonghui 2017], [Ganeriwala 2008].
DOI : 10.17973/MMSJ.2019_03_201899 It is accomplished in several steps. Two discretized signals are
measured at the beginning. Let’s call the output channel of the
[email protected] system simply as a “signal” and the input as a “reference
The article deals with measurement and visualization of the signal”. The signal is marked as y(kT) and the reference signal as
operating deflection shapes (ODS) on laboratory task. The aim x(kT). Both signals are then divided into data blocks with 2/3
is to measure Frequency Response Function (FRF) of selected overlay and M realizations of signals y1(kT) ... yM(kT), x1(kT) ...
points of the template. Together with visualization of their xM(kT) are extracted. All the realizations have N number of
magnitude, real part, imaginary part or phase and show samples. In the next step, each of the blocks are weighted by
template‘s ODS. The measuring chain consists of an Hanning window and obtained results are transformed using
accelerometer, laser and shaker. Data acquisition is performed the Fast Fourier transform (FFT) into the frequency domain
via PULSE LabShop application. Visualization of processed data representations Y1(f) ... YM(f), X1(f) ... XM(f). Complex conjugates
is accomplished in Matlab GUI application called “ODS app“. are calculated by changing the sign of the imaginary part of
The article also contains the theoretical basis of the complex numbers and these representations are marked with
measurement principle along with setup of PULSE LabShop superscript "*": Y1*(f) ... YM* (f), X1*(f) ... XM*(f). Now, it is
project and general description of Matlab GUI application possible to calculate auto-spectrum Syy of the signal, auto-
components. spectrum Sxx of a reference signal. And a cross spectrum Sxy.
KEYWORDS
Auto-spectrum, cross-spectrum, deflection shapes, frequency
transfer function, Matlab, transfer function

1 INTRODUCTION
Operating deflection shapes (ODS) analysis is a method for
visualization the object dynamic under self-exciting forces. This
method enables to identify, where the machine is moving with
the maximum displacement, velocity or acceleration and which
frequencies of excitation cause these problems. It is useful tool
for the diagnostics of faults and for the development of new
components or whole machines.
Method is based on experimental measuring of frequency
response functions and interpreting them as an animation of a
simplified geometric model representing the examined sample.
The principle of this method is divided into two phases. The
first phase represents measuring of FRF, i.e. determination of
vibration magnitudes at single characteristic points and mutual
phases (usually to one reference signal). The next step is the
creation of an animation, which illustrates the shape of
vibration during the excitation by selected frequency or by the Figure 1. Dividing the measured signal into M blocks and weighting by
combination of more frequencies. The movement of structure Hanning window
is usually amplified and slowed-down in the visualisation to
The auto-spectrum (also known as power spectrum) of the
make it suitable for visual inspection.
signal weighted block is computed as a product of signal’s Fast
We can rank the most important pros: simplicity, great
Fourier transform and corresponding complex conjugate. The
information value and wide range of implementations. The
resulting auto-spectrum is computed as an average over M
method allows e.g. identification of dynamic properties
measurement realizations.
(resonance frequencies, identification of largest displacement
location on part and design of flexible fitting, misery or
1 M
unbalance). S yy ( f ) Yi ( f )  Yi* ( f ) (1)
This method grew up on its importance at the end of 80’s in the M i1
20th century due to the increasing computing power and the The averaging smooths the resulting spectrum but also
implementation of two-channel (and later multi-channel) signal consumes measurement time, because more samples are
analysers into the diagnostics. needed.
The operating deflection shapes analysis has become The auto-spectrum of the reference signal is calculated in the
unnecessary part of machine diagnostics knowledge and this same way:
paper presents a laboratory task which is used for the
education of mechanical engineers. The next chapter deals with 1 M
a measurement of frequency response functions. It continues
S xx ( f )   X i ( f )  X i* ( f ) (2)
M i 1
with the part which presents the relationship between

MM SCIENCE JOURNAL I 2019 I MARCH

2861
The cross-spectrum is computed as a multiplication between Because the signal from the laser is delayed by 1,28 ms, this
the Fourier transform of the reference signal and the time must be added also to the accelerometer channel (good
corresponding complex conjugate of signal’s Fourier transform. synchronization is a key preposition of this kind of
measurement). The analyser also generates a white noise signal
1 M for the structure’s excitation. This signal is sent to the amplifier
S xy ( f )   X i ( f )  Yi* ( f ) (3) and the shaker’s coil is powered.
M i1
It is also possible to calculate the "backwards" cross-spectrum: White noise is a commonly used measurement signal, which
has theoretically infinite frequency range. Practically it must be
reduced to some finite range, in the measurement presented in
1 M
S yx ( f )  Yi ( f )  X i* ( f ) (4) this paper, the frequency span was 12.8 kHz. Another testing
M i1 signals can also be used to identify various systems. The
Frequency response function is simply the cross-spectrum simplest one is sine testing signal, another are: swept sine,
divided by the auto-spectrum of the reference signal: impulse, step, spread spectrum signal [Vala 2016], [Proto
2016].
S xy ( f )
H 1( f )  (5)
S xx ( f )

Alternatively, it is possible to calculate the FRF as the signal

auto-spectrum divided by the cross-spectrum:

S yy ( f )
H 2( f )  (6)
S yx ( f )
If the frequency response function is calculated as H1
approach, the processing can suppress the noise of output
signal y and if the function is calculated in a H2 way, the noise
in the input signal is supressed. In the case, when it is supposed
that both signals are affected by the noise on a comparable
level, the "H3" is computed as a geometrical average of H1 and
H2.

H 3( f )  H1( f )  H 2( f ) (7)

Figure 3. Measurement chain

Frequency response functions were measured in the range of

6.4 kHz. Sampling frequency is 2.56 times greater, which is
16 384 samples per second. This satisfies the Nyquist–Shannon
sampling theorem with a suitable reserve. The power of two
number is eligible for Fast Fourier transform used later. The
chart consisted of 1600 bars, which means that the block were
4096 samples long (2.56 times more); it corresponds to 250 ms.
FRFs were computed as an average from 100 blocks.
In next figures, there are examples of spectra and FRF, which
were obtained during the measurement of point number 88.
Auto-spectra (Fig. 4) express which frequencies are present in
the signal. Several frequencies in laser signal are dominant; it
can be said, that these are the frequencies at which the sample
tends to vibrate. Auto-spectra are real numbers, while the
cross-spectrum is a sort of complex numbers and brings the
information of the phase (Fig. 5) FRF includes two information
because these functions are complex. Several approaches how
to graphically demonstrate this FRF exists, but probably most
Figure 2. Sample's mesh with attached accelerometer and laser located
at the selected measurement point.
often the magnitude and the phase both depending on the
frequency are depicted. In this case the magnitude part shows
The first step is to draw a mesh on a studied object. In this how many times the output’s amplitude is greater than the
demonstration task, the number of 164 measured points is input’s amplitude at selected frequency. However, for the
defined on the sample, the positions of the points reflect the purpose od ODS, the imaginary part (Fig. 7) is the most
sample’s geometry. Using 6 bolts, the sample was fastened to important one and is used for the illustration of the shape.
the shaker and an accelerometer was glued onto the FRFs are presented for the purpose of ODS visualization in this
appropriate point in the center of this attachment. This paper, but these can be found in many more applications such
accelerometer measures the reference signal and remains in are modal analysis [Maia 1997], [Agneni 2004], [Agneni 2006]
the same position during all of 164 frequency response or is a part in the design in popular passive and active vibration
functions measurements. The output signal was measured control [Wrona 2016A], [Wrona 2016B], [Wrona 2016C].
using the laser vibrometer. The both signals are processed in
the signal analyser BK 3560C and Pulse LabShop software.

MM SCIENCE JOURNAL I 2019 I MARCH

2862
Figure 4. Auto-spectra of measured signals
Figure 6. Frequency response function – magnitude and phase

Figure 5. Cross-spectrum between measured signals

Figure 7. Frequency response function – real and imaginary part

MM SCIENCE JOURNAL I 2019 I MARCH

2863
3 DEFLECTION SHAPES seen, that the left corner, right edge and the area on the right
side have the highest relative amplitudes.
In application presented in this paper, operating deflection
shapes will be extracted from imaginary parts of measured
response functions. The relationship between FRFs measured
on a laboratory sample and final deflection shape is best to
show on an example. Let’s say, we want to study the deflection
shape of the sample during the frequency of 1212 Hz. We will
focus on the measurement point number 88, whose FRF was
captured and described in the previous chapter. The value of
imaginary part at the frequency of 1212 Hz is approximately -
12. This number is placed into a mesh grid as a third value in a
3D surface plot. While this process is repeated for all measured
points, the final shape for 1212 Hz is acquired.

Figure 10. Shape at 684 Hz

Figure 8. The relationship between measured FRF and the operating

deflection shape

4 ODS APPLICATION
Graphical user interface was designed to visualize the
measured data. This application enables to import measured
FRFs saved as text files in ASCII format. These functions can be
displayed as magnitude, phase, real and imaginary part. The Figure 11. Shape at 952 Hz
selection of displayed functions is also implemented and is
accessible through context menu. The main part of GUI is the
visualization of the deflection shape according to the selected
frequency. This shape can be animated, and the video can be
saved in an avi file.

Figure 12. Shape at 1220 Hz

Figure 9. ODS application GUI preview

5 EXPERIMENTAL RESULTS
Results acquired on a laboratory plant are concluded in this
chapter in a graphical form. Figures depict ODSs at a given
frequency. These illustrations enable to understand which
places on the laboratory samples have vibrationally risky
behaviour and which frequencies are connected to these
problems. For example, if we look at the Figure 12, which Figure 13. Shape at 1704 Hz
illustrates the shape at the frequency of 1220 Hz, there can be

MM SCIENCE JOURNAL I 2019 I MARCH

2864
 2004 International Conference on Noise and
Vibration Engineering, Leuven, September 20 -22,
2004. Leuven: Katholieke Universiteit Leuven, 2239-
2248. ISBN 90-73802-82-2.
[Agneni 2006] Agneni, A. and Coppotelli, G. Load Spectrum
Identification from Operative Responses. In:
Proceedings of ISMA2006: International Conference
on Noise and Vibration Engineering, Leuven, 18-20
September 2006, Leuven: Katholieke Universiteit
Leuven, 3061-3069, 2016. ISBN 90-73802-83-0.
[Ganeriwala 2008] Ganeriwala, S. N. et al. Using Operating
Deflection Shapes to Detect Misalignment in
Figure 14. Shape at 2576 Hz Rotating Equipment. In: Conference Proceedings of
the Society for Experimental Mechanics Serie,
Orlando February 4-7, 2008, 1-5.
[Herfert 2015] Herfert, D. and Gollnick, M. "Good Vibrations" -
A complete software solution for experimental,
operational modal analysis and operating deflection
shapes. In Proceedings of the 16th International
Conference on Computer Systems and
Technologies, Dublin June 2015, New York: ACM,
pp. 46-53. ISBN 978-1-4503-3357-3
[Maia 1997] Maia, N. M. M., et al. Theoretical and
Experimental Modal Analysis. Baldock,
Hertfordshire: Research Studies Press, 1997
[Proto 2016] Proto, A., et al Measurements of generated
energy/electrical quantities from locomotion
activities using piezoelectric wearable sensors for
Figure 15. Shape at 2740 Hz body motion energy harvesting. Sensors, April 2016,
vol. 16, no. 4, pp. 524. ISSN 1424-8220
6 CONCLUSIONS [Vala 2016] Vala, D. Advanced AMR sensor using spread
spectrum technology. IFAC-PapersOnLine, 2016, vol.
This paper presents a laboratory task designed for operating 49, no. 25, pp. 511-516. ISSN 2405-8963
deflection shape processing. The mathematical procedure used [Windalage 2009] Windalage, D. W., et al. Effects of
for measuring frequency response functions was introduced Overlapping and Windowing on Frequency
together with a measurement chain for the diagnostics of these Response Function Estimates of Systems with
functions. Laboratory equipment included a vibration shaker Random Inputs. IEEE Transactions on
with its signal amplifier, signal analyser and signal processing Instrumentation and Measurement, 2009, vol. 58,
software, piezoelectric accelerometer and Doppler laser no. 1, pp. 214-220.
vibrometer. The relation between frequency response [Wrona 2016A] Wrona, S. and Pawelcyzk, M. Shaping
functions and operating deflection shape was explained. A frequency response of a vibrating plate for passive
special application designed for the visualization of measured and active control applications by simultaneous
shapes was also presented and finally the experimental results optimization of arrangement of additional masses
obtained on a laboratory sample are shown. and ribs. Part I: Modeling. Mechanical Systems and
The main goal of this paper is to simply demonstrate the Signal Processing, 2016, vol. 70-71, pp. 682-698.
acquiring operating deflection shapes in several simple steps: ISSN 0888-3270
measurement FRFs on a laboratory sample, transferring [Wrona 2016B] Wrona, S. and Pawelcyzk, M. Shaping frequency
imaginary parts onto a mesh and visualize results. response of a vibrating plate for passive and active
control applications by simultaneous optimization
of arrangement of additional masses and ribs. Part
ACKNOWLEDGMENTS II: Optimization. Mechanical Systems and Signal
This work was supported by the European Regional Processing, 2016, vol. 70-71, pp. 699-713. ISSN
Development Fund in the Research Centre of Advanced 0888-3270
Mechatronic Systems project, project number [Wrona 2016C] Wrona, S. and Pawelcyzk, M. Feedforward
CZ.02.1.01/0.0/0.0/16_019/0000867 within the Operational Control of a Light-Weight Device Casing for Active
Programme Research, Development and Education; and the Noise Reduction. Archives of Acoustic, 2016, vol. 41,
project No: SP2019/51 „Applied Research in the Area of no. 3, pp. 499-505.
Machine and Process Control “ supported by the Ministry of [Yonghui 2017] Yonghui, Y. Y., et al. Online Continuous
Education, Youth and Sports. Measurement of the Operating Deflection Shape of
Power Transmission Belts Through Electrostatic
Charge Sensing. IEEE Transactions on
REFERENCES Instrumentation and Measurement, March 2017,
vol. 66, no. 3, pp. 492-501.
[Agneni 2004] Agneni, A. et al. On modal parameter estimates
from ambient vibration tests. In: Proceedings of the

MM SCIENCE JOURNAL I 2019 I MARCH

2865
CONTACTS:
Ing. Pavel Suranek, Ph.D.
VŠB - Technical University of Ostrava, Faculty of Mechanical Engineering, The Department of Control Systems and Instrumentation
17. listopadu 2172/15, Ostrava, 70800, Czech republic
+420 597 324 113, [email protected], http://www.352.vsb.cz/

MM SCIENCE JOURNAL I 2019 I MARCH

2866