Paper Title (use style: paper title)

Paper Title (use style: paper title)
Measurement and Preliminary Analysis of Powertrain Vibrations in Vehicle
Production Environment
Andreas Manolatos, Ioannis Kagalidis, Branislav Vuksanovic
School of Engineering
University of Portsmouth
Portsmouth, UK
Abstract—In a vehicle production environment, obtaining
information on the condition of the assembled vehicle, allows
for increased quality while minimizing rework minutes at the
same time. The cost of this information is associated with
additional time in production, increasing the overall cost of the
vehicle. The focus of this paper is development of a system for
contactless measurement and investigation of vibrations
present in the vehicle and the Rolling Road vehicle testing
facility as a mean of identifying failures or degradation in the
vehicle or testing facility. Measurements are performed using a
contactless Laser Vibrometer based system which facilitates
capturing the vibrations without interfering with the vehicle
and other standard testing procedures. Signal conditioning and
analysis methods are applied to measured vibration signals in
an attempt to detect irregularities and determine their
transient characteristics from the captured signals. Data
analysis techniques employed in this initial trial of the system
include Short Time Fourier and Wavelet Transforms.
Proposed testing method including experimental setup,
captured signals and some preliminary analysis results are
discussed in this paper.
Keywords-signal processing; vibration analysis,
stationary signals; laser vibrometry; wavelet analysis
In a vehicle production environment, the need for faster,
more accurate and more efficient means of testing is of
paramount importance. While the vehicle is expected to be
tested thoroughly in each of the individual stages of
assembly, there are still ways of obtaining additional
information by detecting, monitoring and subsequently
analysing the auditory and vibration information provided by
the vehicle itself when executing the first full engine test-run
during the production stage and before it is eventually
released to the end customer. Detailed analysis of these
signals can result in a significantly better overview of the
vehicle and help in identifying potential problems related to
the vehicle. Valuable additional information about the
vehicle as a whole or individual sub-systems and modules
can be gained. This includes the engine, the exhaust system,
the horn, the vehicle sound generator in hybrid variants as
well as the infrastructure and the Rolling Road facility.
Vibration analysis is becoming a widely used tool in
diagnosis of various machine faults, including bearings,
gears and similar equipment. Gearboxes in wind turbines,
automobiles and helicopters often run at time varying speeds
and/or loads. This operation gives rise to a non-stationary
vibration signals which can be monitored and further
analysed to provide more information about the condition of
the gearbox. Bafroui et al. [1] have investigated and reported
the use of wavelet energy and Shannon entropy measures for
feature extraction and detection of faults in gearboxes. Chen
et al. [2] carried out fault feature extraction for gearboxes
under non-stationary conditions during the run-up or rundown process. During the run-up or run-down periods gear
drive stimulation forces change both in amplitude and in
frequency allowing some events and phenomena to become
more obvious under the varying speed conditions, usually
not evident at constant speed operation.
Vibration signals have been extensively utilized since
they provide the most intrinsic information about mechanical
faults [3]. Common vibration sensors include piezoelectric
and MEMS accelerometers, displacement as well as velocity
sensors which are all widely used in bearing fault detection
of the gearbox in various industrial applications [4], [5].
In many practical applications attaching the sensory
element (acoustic sensor or accelerometer) directly on the
rotating parts is impractical, can affect the operating
conditions of the device under test and is not cost-effective.
This can also be the case in industry or harsh working
environments which presents a challenge in distinguishing
and extracting signal characteristics [6]-[8].
The focus of this work is detection, identification and
prediction of a vehicle and to a certain extent, testing
infrastructure acoustic emission and vibration during the
production stages. Very limited amount of research has been
attempted to address these challenges and topics directly in
the production environment.
This paper details design and testing process of the
vibration measurement and recording system and discusses
the signal processing techniques used to condition the
measured signal and extract the useful information from it.
The measured signal contains the information about
vibrations originating from a moving components and is of
the analogue nature, thus the first step includes conversion of
the captured signal to its discrete version. According to their
statistical properties, signals emanating from the measuring
equipment and the vehicle itself and any faults or deviations
from the expected performance, can be further considered to
be of non-stationary nature, i.e. statistical properties of those
signals are expected to have time-variable statistical
properties [9]. This property would make direct application
of the most common signal analysis tool - Fourier Transform
(FT) unreliable. FT would in most cases not be able to
determine the transient signal characteristics as the results
will reflect the spectral content of the measured signal
averaged through the duration of the signal [10].
Time-frequency analysis techniques, such are Short Time
Fourier (STFT) and Discrete Wavelet Transform (DWT)
have therefore been used in this work instead of FT. Those
techniques allow for better signal representation in both the
frequency and time domains which makes them more
suitable for the analysis of non-stationary type signals.
Furthermore, these methods provide more information of the
energy distribution over the frequency bands [11].
The rest of this paper is organized in the following
manner. Section II describes the experimental setup
developed in order to measure and record vehicle vibration
signals. Main elements of the system - sensors, conditioning
circuitry and recording facility are discussed and the reasons
for this choice justified. Section III contains some
preliminary results obtained during the system testing and
validating process. Signals captured during the vehicle tests
and processing of those signals are explained and results
illustrated indicating the potential use of developed system.
Conclusions and further steps in this project are outlined in
the final, Section IV of the paper.
A. Sensor and Associated Circuitry
In order to measure the vibration from the product under
the test, a Laser Vibrometer system was designed
specifically for this application. Designed system is
recording the vibration information carried by a visible light
laser beam reflected off the target surface. Schematic of the
circuit and a physical 3D image of the measurement device
developed for this experiment are shown in Figure 1.
The device's main components are an active
photosensitive element and a small amplifier to increase the
combined signal fed to the audio input of a computer system
for further processing. As it is aimed as a stand-alone device,
a 9V battery is used to power the circuit and a variable gain
is added for fine tuning and calibration. Due to the external
and environmental operating conditions, an additional lownoise pre-amplification stage was used between the active
photosensitive element and the final stage amplifier, to
increase the driving potential of the received signal as well.
To further secure the device from unwanted noise sources,
high-precision components, short track lengths and an
insulated casing was used while the remaining noise
components are filtered out using digital filters implemented
in software.
The photosensitive element of the device is a visible light
phototransistor that offers high reaction speeds, very low
noise characteristics and low-effort alignment. Gain and
sensitivity settings can be independently configured for
optimized performance of the device and increased operating
Figure 1. Laser vibrometer circuit and 3D image
The interface of the device is a standard 3,5mm audio
interface, which allows easy connectivity with computer
terminal and loudspeaker isolation when the audio interface
was connected. Finally, a red-tinted focusing lens was used
to improve focusing of the incoming laser beam.
The signals obtained through the measurement procedure
contain features and components that can be attributed to the
vehicle itself, the testing equipment used in the measurement
process and infrastructure. In addition to this, signals can
also be affected by the noise from the surrounding
environment and parasitic interference from the enclosure of
the testing facilities. The desired information or the
components related to condition of the vehicle under test as
well as the testing facility itself is polluted by the
environmental and the valuable information will have to be
extracted from the measured signal through further
processing of the captured signal in discrete-time domain.
B. Experimental Setup
An initial trial of the system involved measuring the
vibration and internal noises of 11 vehicles during actual
testing in the Rolling Road facility. A low power (1mW)
Class 2 650nm (red) laser was used as the targeting beam,
Figure 2. Experimental setup
It is clear that the vibrations increase with the increase in
the vehicle acceleration and differences in those patterns
should be further analysed to give more information about
the condition of the facility or the vehicle under test.
Another trial to compare the performance of the system at a
longer distance (3m) and different angles (30 – 60 degrees)
was performed with similar results.
Time [s]
Frequency [Hz]
which is the maximum commercially allowed laser device. A
reflective surface was used to reflect the laser beam with the
vibrometer positioned 1.5m from the target point, at a 45degree angle as indicated in Figure 2.
The vibrometer had the lens, the gain and sensitivity
adjusted to compensate for the ambient and environmental
noise. It was placed on different surface materials and on flex
bases to dampen any other vibrations that could pollute the
results of the trial using telescopic tripods to mount both the
vibrometer and the laser device. Each test signal is
approximately 2 minutes long in duration, depending on the
vehicle options and the driver’s speed in carrying out the test
in the Rolling Road facility.
The system was tested using an omnidirectional
microphone as a reference sensor in order to compare
acoustic data captured with the microphone to vibration data
taken by the vibrometer. Sample results - spectrograms of the
signals captured by two sensors are shown in Figure 3.
Spectrogram represents a spectrum of frequencies present in
the measured signal as they vary within the measured time
frame and is a suitable tool for the visual inspection of this
type of measured data.
Comparing the two spectrograms recorded from both the
microphone and the vibrometer, it is apparent that there are
no audio or vibration data from the microphone signal in the
0 - 900 Hz range. The microphone signal seems to contain
useful information from 1000 Hz and above with
significantly higher content of information.
The vehicle RPM acceleration curve can be seen in both
signals with the microphone performing better on the higher
frequencies and the vibrometer in the lower frequencies as
expected with a clear indication in the 100 – 3000 Hz range
that the vibration information can be detected. Low
frequency range between 0 - 1 kHz is the spectral content of
vehicle vibrations not detected by the microphone. The aim
of the further work on this project is to try and use this
portion of spectral information in order to try and identify
and predict upcoming problems with the vehicle or the
Figure 3. Microphone and vibrometer data comparison
C. Measurement Process and Data Collection
The data were gathered in Rolling Road facility C in the
BMW Plant Oxford using the experimental setup described
in the previous section thus allowing access to real data in
everyday working conditions. The possible contaminating
sources have been identified as the adjacent Rolling Road
facilities and the ambient or environmental sounds from the
production area. The rest of the infrastructure and equipment,
in regards to the status and faults of the vehicle after testing
and the facility itself, was already configured and calibrated.
The data capturing equipment was installed outside the
facility to avoid any common mode errors from the
vibrations or industrial equipment interference using
telescopic tripods to mount and stabilise both the laser beam
and the data capturing circuit. The beam was aimed through
the double paned window of the facility and focused on
providing the optimum reflection pattern. A laptop was used
to record the captured in the real-time and a microphone was
placed inside the facility to record the audio signal of the
facility and the vehicle under test as additional information.
The driver was bringing the car into the facility and was
performing the standard set of tests and operations - scan the
vehicle identification number, activate horn, start the test,
follow on screen instruction – including pre-warming drive
curve if the engine temperature is low, speed ramp-up with
progressive gear changes, turbocharger adaption and repeat
on failure, variable valve flushing etc. depending on the
vehicle series, options and engine type – and drive off. Each
test took a different time depending on the content of the test
and vehicle itself.
The vibrations from the vehicle were first detected when
a vehicle was in the facility and the current speed was 26
km/h. The best performance was measured when the speed
was 80+ km/h. During these recording sessions, not all
vehicles were successfully captured due to misalignments of
the beam or the focusing lens on the vibrometer.
In turn, each result was further processed, correlated to
the microphone signal and stored in an attempt to determine
the optimum setup and data gathering process.
The aim of the preliminary data analysis is to establish
whether the additional data about the condition of the vehicle
and testing facility can be detected while the vehicle is under
test by using the Laser Vibrometer based contactless method
for data gathering. The vibration data gathered from the
system were classified in two groups discussed in the rest of
this section for further analysis.
A. Horn Detection Results
The horn of the vehicle is located in the engine
compartment and is powerful enough to generate vibrations
that can be detected with the existing setup. Cars under the
test can have one of two different types of horns - the high,
500 Hz and low 400 Hz frequency variant including +/- 25
Hz tolerance for both types.
The aim of this test is to merely establish whether the
measurement system and captured vibration signals can be
used to detect and differentiate between the two different
horn types. Due to the noise contamination of the
measurement area from adjacent lines and vehicles under the
test activating their horns at the same time, the direct
measurement of the horn vibration from the vehicle can
ensure that the target vehicle is monitored and not an
adjacent one.
The sample horn test data shown in Figure 4 include a
time-synchronised comparison of the recorded data from the
measurement system targeting a production vehicle during
the Rolling Road test. The measurement lasts between 1 – 2
seconds and the test requires the activation of the horn four
times. This test is performed by the driver in order to validate
the functionality of the horn.
Time [ms]
Figure 4. Horn detection signal analysis
The first plot shows the recorded vibration data after
normalization, recorded live from the vehicle and contains
four horn activations. Horn activation instants are hard to
detect by visual inspection of this signal.
The second plot on this figure contains the STFT
spectrogram of the original normalized waveform. This
analysis shows more information on the frequency content of
the recorded signal with the four horn activations centered at
495 Hz but it is not clear where each activation starts and
Furthermore, the fourth activation can be lost in the
background noise level as the recorded level was lower than
the rest of the activations.
The graphs below the spectrogram show the original
signal decomposition at 4 levels using a db4 wavelet after
performing a DWT and identifying this as the optimum level
of decomposition. The S graph is the original signal followed
by the relevant approximation (a4) on the 4th level, resulting
in an almost exact reconstruction of the original signal. The
next graph contains the 4th level decomposition detail (d4)
which show very clearly the point of each horn activation
(119ms / 432ms / 824ms / 1337ms) and the relative time
duration of each one, effectively detecting these activations
as discontinuities and irregularities of the original signal
regardless of the noise level present in the original
normalized recording. The 3rd level decomposition detail
contains the same information but there is an irregularity
detected in the beginning of the signal changing the scale of
the detail result.
The last two detail levels contain vibration and noise
information but no usable information can be extracted from
B. Drive Curve Comparison and Gear Shift Detection
DWT based approach described in the previous section
has been used for the second set of tests aimed at evaluating
the measurement procedure and recorded data. The aim is to
use vibration signals to try and identify when and how the
gear shift changes occur while the vehicle is driven in the
Rolling Road for the first time. During this test the vehicle
does the first full driving cycle after it has been assembled,
thus this is the first time where the vehicle is tested in driving
conditions. The testing process involves the majority of the
powertrain checks by doing a warm-up drive cycle if the
engine temperature is low and then follows a drive up curve
that gradually increases the speed of the vehicle while
sequentially changing gears, the turbocharger adaption and a
repeat in the drive curve if the adaption has failed. Variable
valve timing check and flushing, braking and drive-out curve
are performed at the end of each test.
The driving curve has a variable testing time which
depends on how closely each driver follows the on-screen
instructions, what is the vehicle type and what are the
installed options. On average, this test is expected to last
approximately 90 seconds assuming the adherence to the
instructions by the driver, no warm-up cycle and no repeated
test steps. During this drive cycle, the measured vibrations
include the ones coming from the vehicle and its vibrating
components as well as the vibrations coming from the testing
facility itself, even though the target of the vibrometer is the
vehicle under test. Direct separation of those vibrations using
a single vibrometer is not possible as both the vehicle and the
facility have to be monitored to identify the possible source
of the vibrations. This fact should be kept in mind while
using the recorded signal, the sum of those vibrations, to
identify problems that appear during the testing process.
Given the fact that there is a reference drive curve along with
the expected steps to be followed, the measured vibrations
can be correlated to identify any discrepancies that highlight
a possible problem either on the vehicle or on the facilities
side. Problems of this type are characterized by abrupt
changes or degradation in the transient features of the
expected signal.
The mechanical stress on the powertrain happens during
gear shift changes as most of the powertrain components are
engaged at the same time to maintain an optimum
performance. The same stress can be propagated to the
testing facility and serve as a vibration trigger on the
facility's side and vice versa.
Sample signal recorded during this drive test and the
results of the processing are shown in Figure 5. Drive curve
identifying various events during the test is superimposed at
the top of this figure to provide easier identification of
various events during this test. The first plot below the drive
curve shows the recorded vibration data after normalization
as it was recorded live from the vehicle and includes the start
of the drive test until the vehicle comes to a standstill just
prior to the drive off. Again, no information in regards to the
condition of either the vehicle or the facility can be extracted
by this waveform as it is expected.
The second plot from Figure 5 contains the STFT
spectrogram of the original normalized waveform but this
time the frequency information can be seen and compared to
the actual drive curve of the vehicle. The spectrogram
deliberately shows information up to 1500 Hz to include and
focus on the low frequency vibrations and any harmonics
present. Visual inspection of the spectrogram outlines the
changes in the vehicle speed during the drive curve, starting
from 0 km/h, gradually increasing until 150km/h and
remaining there for a small period of time and gradually
reducing speed until the vehicle comes to a standstill again
prior to driving out. This information can be used when
compared with the reference drive curve to detect any
anomalies on its own, to identify deviations from the
reference drive curve and to indicate possible problems with
either the vehicle or the testing facility.
Time [ms]
Figure 5. Gear-shift detection results
Next plot on this figure contains the results of the wavelet
analysis of the same signal. The signal was decomposed at 4
levels using a db4 wavelet after performing a DWT and
identifying this as the optimum level of decomposition. The
S graph is the original signal followed by the relevant
approximation (a4) on the 4th level, resulting in an almost
exact reconstruction of the original signal.
The level 4 detail (d4) holds the most interesting
information, where individual irregularities are highlighted
by the wavelet analysis. Each highlight covers a band that
happens during a gear shift change when compared to the
reference drive curve. Each band now, in turn, indicate the
transient characteristics of the vibrations governing each gear
change that can be used further to identify any deviations
that can be translated in problems either in the vehicle or in
the testing facility. It is also worth noting that another
interesting area of analysis is the steady speed point, where
the vibrations are stronger due to the increased speed and the
slowdown roll-off curve where breaking can trigger different
vibrations from other vehicle components such as brake
calipers and suspension mechanisms which can also be used
to monitor any upcoming problems.
The rest of the details (d1 – d3) contain vibration and
noise information but no usable information can be extracted
from them.
The initial steps in setting up and testing a novel system
for the vibration analysis in a vehicle commissioning and
testing environment have been described in this paper. The
necessary hardware including the measurement sensors,
signal capturing and conditioning circuitry as well as signal
recording facility have been designed and trialed. A number
of vehicles assembled at the BMW plant in Oxford have
been tested using designed system and their vibration
signatures recorded and analysed. The preliminary analysis
of the captured signals focused on establishing the clear
correlation between the recorded signal and the vehicle
vibrations in the vehicle testing and commissioning
environment. Recorded raw signals have been processed
using various digital signal processing (DSP) techniques in
order to reduce the level of noise and extract the most
important signal components which can be reliably
recognized and interpreted.
The use and effectiveness of wavelet analysis for this
task has been highlighted by comparing the results of the
DWT and STFT analysis methods with the raw signals. The
wavelet properties allow for an accurate recognition of
timing information in the signal as well as extraction of
irregularities and transient characteristics which can then be
used to identify existing problems or even pinpoint
degradation in these elements that can prompt for preventive
maintenance if needed.
Concerning the measurement system, the drawback of a
purely acoustical measurements, i.e. microphone-only
approach have been exposed as the additional and unwanted
information from the adjacent facilities where sounds from
other vehicles running the same test at the same time
contaminate the measured signal. This was demonstrated by
the second horn signal being detected in the initial horn test
measurements on the vehicle. The vibrometer based system,
on the other hand, while recording only the noise emanating
from the vehicle under the test and test facility has a potential
problem of not being able to detect higher frequency content
present in the signal. One of the immediate tasks in the
continuation of this work is to establish how important is the
higher frequency information in providing additional
information on the condition of the vehicle and the facility.
The ideal approach might therefore be to combine the data
captured by both sensors in order to achieve a more holistic
approach. The need to adopt this approach is highlighted by
the fact that detection reliability of the current system is
below 99%.
On a broader scale, this work should aim for the system
that can be used on various sites and plants of vehicle
production by adapting to the individual characteristics of
each environment. In addition to this the possibility of an
onboard solution that can supplement the existing on-board
diagnostics of a vehicle by analyzing the vibrations present
on the vehicle.
With those, broader aims and targets in mind the future
work will focus on investigating the advanced techniques for
signal processing and analysis. Possible algorithms and
methods to investigate will include the analysis of the
acquired vibration signal using ensemble empirical mode
decomposition (EEMD) and singular spectrum analysis
(SSA) methods for signal decomposition, denoising and
feature extraction. Adaptive Noise Cancellation (ANC) can
also be applied to try and reduce the unwanted noise present
in the measurements.
Further study would then include the use of data mining
techniques as well as pattern recognition and prediction
algorithms such as Artificial Neural Networks (ANN) and
Independent Component Analysis (ICA) in the context of
vehicle testing in production environments which might also
be applicable to an area such as the automotive industry.
Hojat Heidari Bafroui, Abdolreza Ohadi, Application of wavelet
energy and Shannon entropy for feature extraction in gearbox fault
detection under varying speed conditions, Neurocomputing, Volume
133, 10 June 2014, Pages 437-445, ISSN 0925-2312
[2] X. Chen, Z. Feng and M. Liang, "Fault feature extraction of planetary
gearboxes under nonstationary conditions based on reassigned
wavelet scalogram," 2015 IEEE International Instrumentation and
Measurement Technology Conference (I2MTC) Proceedings, Pisa,
2015, pp. 294-299.
[3] Y. Lei, F. Jia, J. Lin, S. Xing and S. X. Ding, "An Intelligent Fault
Diagnosis Method Using Unsupervised Feature Learning Towards
Mechanical Big Data," in IEEE Transactions on Industrial
Electronics, vol. 63, no. 5, pp. 3137-3147, May 2016.
[4] H. Zhang, Q. Yue, H. Zhou, J. Si and X. Shi, "Automatic selection of
frequency-band based on wavelet for gear fault diagnosis," Computer
Science & Education (ICCSE), 2015 10th International Conference
on, Cambridge, 2015, pp. 524-527.
[5] Y. Qu, J. Zhu, D. He, B. Qiu and E. Bechhoefer, "Development of a
new acoustic emission based fault diagnosis tool for gearbox,"
Prognostics and Health Management (PHM), 2013 IEEE Conference
on, Gaithersburg, MD, 2013, pp. 1-9.
[6] V. B. Pagi, R. S. Wadawadagi and B. S. Anami, "An acoustic
approach for multiple fault diagnosis in motorcycles," SoftComputing and Networks Security (ICSNS), 2015 International
Conference on, Coimbatore, 2015, pp. 1-7.
[7] H. V. Khang, H. R. Karimi and K. G. Robbersmyr, "Bearing fault
detection based on time-frequency representations of vibration
signals," Electrical Machines and Systems (ICEMS), 2015 18th
International Conference on, Pattaya, 2015, pp. 1970-1975.
[8] S. H. Kia, H. Henao and G. A. Capolino, "A real-time platform
dedicated to on-line gear tooth surface damage fault detection in
induction machines," Electrical Machines (ICEM), 2014 International
Conference on, Berlin, 2014, pp. 1478-1484.
[9] M. R. Wilkinson, "Condition monitoring for offshore wind turbines
"Eng. D., University of Newcastle upon Tyne, UK, 2008
[10] T. Sawicki, A. K. Sen, and G. Litak, "Multiresolution wavelet
analysis of the dynamics of a cracked rotor," International Journal of
Rotating Machinery, vol. 2009
[11] A. A. Jaber and R. Bicker, "A Simulation of Non-Stationary Signal
Analysis Using Wavelet Transform Based on LabVIEW and Matlab,"
Modelling Symposium (EMS), 2014 European Pisa, pp. 138-144
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF