Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

CHAPTER 1
INTRODUCTION
Due to the very high customer specific quality and availability requirements regarding machine
tools and to reduce the machine breakdown time caused by wear and tear damage, the use of
condition monitoring (CM) approaches is increased over the last decades. Description, issues and
possible solutions to deal with sensor-based CM and diagnostics applications can be widely
found in literature. In this context, the metrological characteristics of the systems called to
provide the needed data allow evaluating the quality of the results of the further analysis and to
remove possible causes of errors in the diagnosis process. Several innovative approaches in
manufacturing systems and management raise the interest in considering the contribution the
metrology offers in facing engineering issues, and to handle with its methodologies and basics.
Furthermore, the application of CM methods is a suitable example, where experts are called to
manage with a huge amount of data, justifying the adoption of optimized solutions, which take
into account, among others:
(1) The application of CM and data mining techniques.
(2) The exploitation of data coming from disparate sources, sensors and other equipment.
(3) The most appropriate methods of data processing, to perform meaningful and coherent
analysis.
In this framework, the key-factor lies in the possibility of being able to provide meaningful
indications, by synthetizing the information coming from a big data environment, also for
addressing the issues connected to the industrial reality, which often would rather reduce the
quantity of data needed for the analysis. Nevertheless, recent advances in technologies such as
Internet of Things, Machine to Machine communications and Cloud computing technologies
create the opportunity to transfer models and techniques form the research lab to industry. When
the analysis is devoted to the study of complex and technologically-advanced machining tools,
the failures are caused in many cases by mechanical components, such as the mean spindle and
the feed drives (especially the ball screws).
Normally the breakdown of these mechanical components is induced through wear. In general,
the CM systems for production machines available on the market focus on different parts of the

Department of Mechanical Engineering, SIT, Mangaluru Page No1

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

machine (e.g. feed drives or mean spindle). All these systems pursue the strategy of fault
detection in early stage to avoid machine breakdown caused by component failure. These
systems can be divided into mobile systems, which can be used manually to measure signals (e.g.
vibration) on the components and systems and into upgradable systems, which can be integrated
in already installed machines. Furthermore, there are CM solution integrated in some of the new
production machines. Due to high requirements regarding the diagnostic accuracy, which can be
not granted by using mobile CM systems, because of the uncertainty of the measurement, the
most common CM solutions for machine tools are permanently installed ones. Therefore, in case
of not-embedded sensors, in order to have reliable results, reliable measurements on the one hand
and reliable algorithms, on the other hand have to be guaranteed.
The aspect of interest to guarantee the reliability of the algorithms are resolution and selectivity,
i.e. their ability to classify the defects and to select the most suitable features for the analysis.
Regarding the measurements, their reliability is guaranteed by means of preventive solutions
involving rigorous methods and operating procedures, able to avoid or minimize errors from the
beginning. In this sense, the steps of the procedure for CM applications have to be all optimized,
data acquisition included.
The steps optimized and the most suitable indicators for CM algorithms selected, an effective
and efficient CM procedure is obtained. This latter is based on the contemporary choice of a set
of three indicators evaluated on vibration and AE measurements. The ability to classify the
defects, which are different for location, type and gravity, is related to the suitable choice of the
indicators and to the accuracy of measurement data.
The data collected from the sensors and then the data is denoised from the unwanted signals.
This data is then compared to the raw signal obtained from the gear running in the best condition.
If the signals obtained differ from the original raw signal then it is clear that the gear is faulty or
some other part in the system is faulty.
The main benefit of condition monitoring is the lowering of maintenance costs. Although
adopting a ‘fix as it breaks’ approach can initially lower costs, it is always more expensive in the
long run. Furthermore, broken machinery can lead to a ‘ripple effect,’ where other equipment
down the production line also malfunctions, causing further damages. Most UK manufacturers
report that ‘53% of machine downtime is caused by hidden internal faults. Not only does this dig

Department of Mechanical Engineering, SIT, Mangaluru Page No2

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

into your potential profit margin, but it can also have adverse effects on delivery schedules and
agreements with clients. These faults are virtually impossible to trace without the use of
predictive maintenance techniques, and as such, condition monitoring is a worthwhile investment
for any manufacturing company who values reliability. Good machinery and high-quality parts
are made to last, and condition monitoring can even help to extend the lifespan of components
and the longevity of your machines. With frequent and reliable updates about your machinery,
you have a real-time list of potential maintenance tasks and their priority. This can help you to
liaise with maintenance companies and plan around scheduled projects, saving you time and
money in the process.

Department of Mechanical Engineering, SIT, Mangaluru Page No3

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

CHAPTER 2
LITERATURE REVIEW
Pengxing Yi et al. [1] provides a review about the distinctive modal properties of the wind
turbine gear box using modal frequency analysis. Zhang, and Xue and Li performed a modal
analysis on a laboratory gearbox and identified the first 10 mode shapes and modal frequencies.
Based on the experiment results, they revised the FE model and established a foundation for
structural optimization. Helsen et al. discussed modeling methods for building an appropriate FE
model of the gearbox, investigated the influence of structural change, and concluded that the
higher mode had a higher sensitivity than the lower mode. Park et al studied the influence of the
gearbox using different meshing methods and concluded that different structures required
different meshing methods. The modal frequencies of the system decrease after coupling, thereby
diminishing the vibration.

Abdulrahman S. Saitet al.[2] reviewed the literature on condition monitoring of a gearbox.. The
use of vibration-based analysis damage detection techniques is classified and discussed in
details. The capability of each technique to sense failure and damage in rotary equipments is
addressed. Many types of defects or damage will increase the machinery vibration levels. These
vibration levels are then converted to electrical signals for data measurement by accelerometers.
In principle, the information concerning the health of the monitored machine is contained in this
vibration signature. Hence, the new or current vibration signatures could be compared with
previous signatures to determine whether the component is behaving normally or exhibiting
signs of failure. In practice, such comparisons are not effective. Due to the large variations,
direct comparison of signatures is difficult. Instead, a more useful technique that involves the
extraction of features from the vibrational signature data could be used. Features like raw signal,
time synchronous averaged signal, residual signal, difference signal and band-pass mesh signal.
All this data is then compared to the raw signal and then determined if the vibration signals
obtained are differing from the original raw signal. if it does then it means the gear is faulty.

Giorgio Dalpiaz et al. [3] has provideda review on the literature of condition monitoring of a
planetary gearboxes. The present work concerns the development of a new TSA-based metric,

Department of Mechanical Engineering, SIT, Mangaluru Page No4

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

namely NA4mod, for pitting detection in planetary gearboxes. The new proposed metric is
computed from the TSA of the gear of interest and, in particular, it is based both on the regular
signal and the residual one obtained from the TSA. In the case of planetary gearboxes,
theevaluation of the TSA evaluation is complicated by the movement of the planets, that causes
the continuous variation of the distance between the meshing teeth and the fixed transducer
mounted outside the annulus. The variation of such a relative position produces a periodic
variation in the transfer function between the two components. For this reason, a specific
procedure, able to eliminate the effect of the variation of the transfer functions, is used in order
to extract the TSA of each gear from the overall vibration signal. Firstly, by using a tachometer
signal, theacceleration time signal is resampled in the angle domain according to the rotation of
the carrier of the planetary stage of interest. Then, starting from the angle-based signal referred
to the carrier, the signal of the tooth of the planet that meshes with the tooth of the annulus under
the accelerometer is extracted one by one. In order to achieve this goal, a time window of
appropriate length is used and the operation is repeated once per turn of the carrier, until the
signals relating to all the teeth of the gear of interest, planet or sun, have been extracted. Finally,
the operation is repeated, this allows the application of various signal analysis techniques,
specifically designed for the treatment of the TSA, and of a series of related signal metrics to be
used for fault detection and diagnosis.

Giulio D’Emiliaa et al. [4] provides a review of literature of condition monitoring of ball
screws. To train and evaluate the Condition Monitoring (CM) algorithm for the ball screw, the
data (e.g. vibration) has to be acquired and structured in a way, which allows using these data in
a development tool, such as MATLAB. The accelerometer has to be positioned in such a way
that it should improve the signal to-noise ratio of measurement data. Sensors’ calibration, their
installation, positioning and data processing techniques guarantee the definition of the systems
signature, working in no damage conditions, with a tiny uncertainty range, allowing identifying
the occurrence of tiny variations of the working conditions with respect to the arise of defects.
Then, signal processing, feature extraction and feature selection is done.The artificial damages
created using the Laser powder cladding method are located on the running surface of the ball
screws.To evaluate the implemented application data were collected using the AE and
acceleration sensors. The tests were repeated 15 times under the same operating parameters.

Department of Mechanical Engineering, SIT, Mangaluru Page No5

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

Therefore, a large dataset is set up. The procedure allowed identifying in a very satisfactory
manner the artificial defects, depending on the best-suitable features extraction.

In contrast to previous applications, N Baydar and A Ball [5] presents a review aboutwhether
acoustic signal can be used effectively along vibration signal to detect the various local faults in
gearboxes using the wavelet transform. Two commonly encountered local faults, tooth breakage
and tooth crack, were simulated. The results from acoustic signals were compared with vibration
signals. The results suggest that acoustic signals are very affective for the early detection of
faults and may provide a powerful tool to indicate the various types of progressing faults in
gearboxes.

Emmanuel Resendiz-Ochoaet al. [6] presents a novel methodology based on noninvasive

monitoring for assessing different levels of uniform wear in a gearbox transmission system.
There are four important point that must be highlighted in this new proposed methodology. The
first is the reliability of using infrared images and the statistical features of hot spots, due to the
fact that the temperature increase in the gears depends on the work performance of the uniform
wear condition of the gears. The second is the implementation of dimensionality reduction
strategies for the optimization of the set of statistical features estimated from the infrared
imaging, since it allows the elimination of less discriminate features and the compression of
more significant statistical features. The third is the classification based on an artificial neural
network, which is capable of identifying and classifying the different case studies proposed in
this paper for the study of uniform wear in the gears of the gearbox transmission system.The
fourth is the comparison that is made between infrared imaging and vibration signals to
determine the reliability in the results for the diagnosis of failures.

The aim of this research done by BarshikarRaghavendraRajendra et al. [7]is to advance the
field of condition monitoring and fault diagnosis in gearbox operating under different load
conditions. In order to perform accurate and reliable analysis on gearbox, an experimental set up
is designed that can accurately repeat the measurements of signals and can introduce a particular
fault of the gearbox faults. In the present research work, Lab VIEW environment is used to
diagnose the faults with direct online condition monitoring.This research work investigates the
feasibility of detecting mechanical faults such as bearing failure and wear gear failure, gear tooth

Department of Mechanical Engineering, SIT, Mangaluru Page No6

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

broken failure using the spectrum of current of a motor. Experimental results show that the
characteristic frequencies could not seen in the power spectrum if bearing fault and gear fault are
small in size. As severity of fault increases, the characteristic frequencies become visible. In the
research work, an experiment has also been conducted to detect the gear box fault.

Sayed Y. Akl et al. [8] provides a review on how to enhance a reliability program if it is
accompanied with other condition-based monitoring (CBM) techniques, such as vibration
analysis. Through this study, a subsystem of a critical gearbox was monitored using wear particle
analysis technique. Elemental analysis of the worn particles was implemented using
spectrometer, through which a trend of wear process inside the gearbox was established to
evaluate the health condition of internal components within the gearbox. Ferrography analysis
was also implemented to investigate the concentration of each wear mode that had been
generated within the lubricating system in particle/ml. Ferrography analysis was implemented
using laser Netfines (LNF) equipment. Results of both elemental and ferrography analyses had
justified each other.It was recommended for this company to apply the wear particle analysis to
their critical machines to enhance their maintenance program.

Yolanda Vidalet al. [9] gives us a review on how without using specific tailored devices for
condition monitoring but only increasing the sampling frequency in the already available (in all
commercial wind turbines) sensors of the Supervisory Control and Data Acquisition (SCADA)
system, a data-driven multi-fault detection and classification strategy is developed. A promising
strategy to detect and classify multiple wind turbines faults was presented using only
conventional SCADA data with an additional, but feasible, high-frequency sampling from the
sensors. That is, the fault diagnosis strategy does not involve the supplementary installation of
costly purpose-built data sensing equipment for wind power plants.

Saurabh S. Shahapurkaret al.[10]provides us a review about literature on how to detect

different type of gear tooth faults, aexperimental data is taken from single stage gearbox set up
with help of FFT analyzer. Vibration analysis techniques are used for detection of fault in gear
system, fluctuation in gear load such as a method for detecting the evolution of gear faults based
on time- frequency analysis through Matlab. The various types of defects are created on gear
tooth such as one corner defect, two corner defect, three corner defect, and Missing tooth.The

Department of Mechanical Engineering, SIT, Mangaluru Page No7

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

validation is done successful by taking input signal from FFT analyser to Matlabprogram for
calculating effective statistical parameters in defective condition for time &frequency domain
analysis. Big advantages of this kind of system are that we reach directly at the fault after
opening the whole system.

K. Gopinathet al. [11] provides a review on analyzing the signal behavior between time domain
and frequency domain when the shaft rotating at whirl speed and also indicating the filter
specifications of the signal. In signal, noises are always bepresent and it makes the observation
of these vibration parameters might get inaccurate. It is necessary to remove the noises from the
source. It is to be found that the inverse behavior between time domain and frequency domain
using accelerometer and MATLAB relations. Shaft VRMS Scheme: Whirling is defined as the
rotation of the plane made by the line of centres of the bearings and the shaft. Time Domain: The
most commonly used device for vibration measurement is the piezoelectric accelerometer, which
gives an electrical signal proportional to the vibration acceleration. This signal can readily be
amplified and analyzed.The frequency domain signal or frequency spectrum is a plot of the
amplitude of vibration response versus the frequency and can be derived by using the digital Fast
Fourier analysis of the time waveform.The time domain and frequency domain of various signals
were analyzed using MATLAB. Thesptool is used to analyze and compare the two domains of
signals through FFT. A filter for vibration signals, with magnitude of pass band and stop band
was developed successfully. These analysiswere done through the chosen vibration parameters.
These analyzing methods could be used to prevent the damage of shafts and also an increase the
lifetime period.
2.1 PROBLEM STATEMENT
Gears undergo failure for numerous reasons. Assessing gear damage can be a challenging task,
especially in the industrial equipment’s. Therefore in order to minimize gearbox downtime and
to avoid performance degradation, a practical and robust monitoring system is needed.

To understand this problem, a customized gear test rig is developed. It will analyse the vibration
signals of healthy and unhealthy gear sets which will inturnprovide with early warning of a
malfunctioning gear, and thus it will alert us before a gear undergoes complete failure if
periodically monitored.

Department of Mechanical Engineering, SIT, Mangaluru Page No8

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

2.2 OBJECTIVE

• To fabricatethe customized gear test rig

• To extract the signal from healthy and unhealthy gears using vibration sensor
(piezoelectric sensors)
• To Denoisethe extracted signals using DWT (discrete wavelet transform) and MATLAB
which makes use of DFT (Discrete Fourier Transform) to generate waveform, for
analytical study and to extract features of signals via time synchronous averaging
technique, for analytical study
• To use vibrational data, study of denoising schemes and feature extraction techniques, a
device for condition monitoring of gears is developed to reduce the downtime of the
operational system.

Department of Mechanical Engineering, SIT, Mangaluru Page No9

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

CHAPTER 3
METHODOLOGY
A customized test rig [fig 3.1] is developed compelling with healthy and unhealthy gears of
similar dimensions and study the captured vibration data sets generated between the healthy gear
[Fig 3.2] and unhealthy gears [Fig 3.3] and further use these data sets (vibrations/signals)
generated to compare the behavioral changes of the vibration data set generated from healthy and
unhealthy gears and program the micro-controller accordingly after analysis to predict the health
condition of gears..

Further extract the features Via Time synchronous averaging technique and further use denoising
techniques like DWT (discrete wavelet transform) to understand the signals more precisely. .

FIG 3.1: PROPOSED SYSTEM TO BE FABRICATED

Department of Mechanical Engineering, SIT, Mangaluru Page No10

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

FIG 3.2: HEALTHY GEAR

FIG 3.3: UNHEALTHY GEAR

Thus using this device preventive maintenance activity is conducted periodically, for condition
monitoring of the gear system, which ultimately leads to continuous operation of the machinery
which reduces operational downtime in case of gear fault and hence increasing the productivity
and efficiency of the system.

Department of Mechanical Engineering, SIT, Mangaluru Page No11

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

A CUSTOMIZED GEAR TEST RIG IS

DEVELOPED AND THE SYSTEM IS RUN
WITH HEALTHY GEARS AND UNHEALTHY
GEARS

EXTRACT AND STUDY THE VIBRATIONAL DATA SETS OF

HEALTHY AND UNHEALTHY GEARS OF SIMILAR SYSTEMS

DO ANALYTICAL STUDY OF THE VIBRATIONAL DATA SETS

WITH FEATURE EXTRACTION AND DENOISING TECHNIQUES

DEVELOP SYSTEM FOR CONDITION MONITORING OF GEARS

UNDERSTANDING THE VIBRATIONAL DATA SETS AND
STUDYING THE FEATURE EXTRACTION & DENOISING
TECHNIQUES

USING THE DEVICE COMPARE DIFFERENT GEARS AS

PREDICTIVE MAINTANENCE FOR CONDITION MONITORING

IF NOT PROPER
VIBRATIONAL
SIGNAL / GEARS ARE NOT IN GOOD CONDITION
DATA SET

PROPER

GEARS ARE IN GOOD CONDITION

FIG 3.4: FLOWCHART OF WORKING OF THE PROJECT

Department of Mechanical Engineering, SIT, Mangaluru Page No12

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

CHAPTER 4
PROGRESS ON PROJECT WORK TILL NOW

4.1 FABRICATION OF TEST RIG

The test rig consists of following parts:
1. Stand: It supports the driver gear, driven gear and the stepper motor.

Fig 4.1 STAND SET-UP

2. Healthy gear: A gear is fabricated which has no damage or any other impairment.
3. Unhealthy gear: A gear is fabricated with the same module but has significant damages in
its teeth.

Department of Mechanical Engineering, SIT, Mangaluru Page No13

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

FIG 4.2: HEALTHY GEAR FIG 4.3: UNHEALTHY GEAR

4. Stepper Motor: It is used to run the driver gear which inturn runs the driven gear which is
supposed to be the healthy or unhealthy gear.

FIG 4.4 STEPPER MOTOR

FIG 4.5 TEST RIG DIMENSIONS

Department of Mechanical Engineering, SIT, Mangaluru Page No14

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

FIG 4.5 COMPLETE SET-UP

4.2 GEAR FAULT ANALYSER

• The vibration data extraction of healthy and unhealthy gear is done by ADXL345 as
shown in the figure [Fig 4.5].
• Analysis of the data generated from the vibration data extraction of healthy and
unhealthy gear is carried out.
• The data obtained from the healthy and unhealthy gear is then compared with each
other.
• After analysing the data, we have programmed the microcontoller to predict the
condition of the gear i.e if the gear is healthy or not.
• If the gear is said to be unhealthy then the buzzer in the system beeps indicating that
the gear is unhealthy.

Department of Mechanical Engineering, SIT, Mangaluru Page No15

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

START

DIRPIN HIGH

STEP PIN HIGH

DELAY 5000 MS

STEP PIN LOW

DELAY 5000MS

READ ACCEL(X,Y,Z)

VALUE =Z-Y

YES
VAL>
50

PRINT VIBRATION VALUE

NO

PIN 11 HIGH

CNT=0

CNT++

YES
CNT>
60

NO PIN 11 LOW

CNT=0

END

Department of Mechanical Engineering, SIT, Mangaluru Page No16

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

REFERENCE
[1] Pengxing Yi, Peng Huang, Tielin Shi, (2016), “Numerical analysis and experimental
investigation of modal properties for the gearbox in wind turbine”, Frontiers of Mechanical
Engineering, pp 388–402.

[2] Abdulrahman S. Sait, Yahya I. Sharaf-Eldeen, (2011), “A Review of Gearbox Condition

Monitoring Based on vibration Analysis Techniques Diagnostics and Prognostics”, Rotating
Machinery, Structural Health Monitoring, Shock and Vibration Volume 5, pp 307-324.

[3] Giorgio Dalpiaz and MattiaBattarra, (2020), “Condition monitoring indicators for pitting
detection in planetary gear units”, Diagnostyka Vol. 21, No. 1, pp 3-10.

[4] GiulioD’Emiliaa, AntonellaGasparia, EckhardHohwielerb, AbdelhakimLaghmouchib,

EckartUhlmannb, (2018), “Improvement of defect detectability in machine tools using sensor-
based condition monitoring applications”, Procedia CIRP 67, pp 325 – 331.

[5] N Baydar and A Ball, (2003), “Detection of Gear Failures via Vibration and Acoustic Signals
using Wavelet Transform”, Mechanical Systems and Signal Processing Volume 17, Issue 4, pp
787-804.

[6] Emmanuel Resendiz-Ochoa, Juan J.Saucedo-Dorantes, Juan P. Benitez-Rangel,Roque A.

Osornio-Rios and Luis A. Morales-Hernandez, (2020), “Novel Methodology for Condition
Monitoring of Gear Wear Using Supervised Learning and Infrared Thermography”, Applied
sciences, pp 1-18.

[7] BarshikarRaghavendraRajendra, Santosh.VBhaskar, (2013), “Condition Monitoring of Gear

Box by Using Motor Current Signature Analysis”, International Journal of Scientific and
Research Publications, Volume 3, Issue 8, pp 1-6.

[8] Sayed Y Akl,Hesham A. Abou El Anein, Sherif El-Soudy, (2019), “Condition Based
Monitoring of Gearbox Transmission Using Wear Particle Analysis Technique”, Proceedings of
the International Conference on Industrial Engineering and Operations Management,pp 68-74.

[9] Yolanda Vidal, FrancescPozo and Christian Tutivén, (2018), “Wind Turbine Multi-Fault
Detection and Classification Based on SCADA Data”, Energies, pp 1-18.

[10]Saurabh S. Shahapurkar, Hemant S. Pansare, Prashant P. Dhebe, Chetan S Wagh, Prof.

AmitDesale(2015), “Detection of Fault in Gearbox System Using Vibration Analysis Method”,
International Journal of Engineering and Applied Sciences (IJEAS), volume 2, Issue-5, pp 119-
122.

Department of Mechanical Engineering, SIT, Mangaluru Page No17

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

[11] K. Gopinath , S. Periyasamy, (2016), “Vibration Analysis on Rotating Shaft using

MATLAB”, IJSTE - International Journal of Science Technology & Engineering, volume 3,
Issue-6, pp 248-252

Department of Mechanical Engineering, SIT, Mangaluru Page No18

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

Department of Mechanical Engineering, SIT, Mangaluru Page No19

 Evaluation of Denoising and Feature Extraction Techniques for Gear Condition Monitoring

Department of Mechanical Engineering, SIT, Mangaluru Page No20