Vibration Analysis For Predictive Maintenance of A Rotary Pump
Vibration Analysis For Predictive Maintenance of A Rotary Pump
Vibration Analysis For Predictive Maintenance of A Rotary Pump
GUSTAV OSSWALD
GUSTAV OSSWALD
Gustav Osswald
Abstract
Predictive maintenance based on condition monitoring uses sensor and system data to prevent
damage, in advance to a failure occurring, allowing for a service to be performed at an optimal
position in time. Condition based predictive maintenance estimates time of system failure based
on a priori information, which has shown to be much more cost effective than traditional
maintenance methods. Typically, there are applications where either of the maintenance
methods, reactive, preventative or predictive maintenance, prove most sufficient. The expensive
downtime in industrial processes and systems has come to focus development of predictive
maintenance which often is found to be the optimal solution in these settings. In order for a
predictive maintenance algorithm to be developed, there has to be in depth knowledge about the
system and big data to base the algorithm on.
This project is aimed at analysis and condition monitoring of the AkzoNobel -
intelliCURE separate spreader used in the lamella and beam industry. Specifically targeted at
the detection of cavitation in the transportation pumps, which is a common destructive
phenomenon occurring in pumps. Cavitation which is the formation and implosion of cavities in
the liquid, produces excessive shock waves resulting in vibrations. Depending on severity,
cavitation can, in time, lead to internal damage and cause leakage. The type and amount of
cavitation in a system is dependent on the rotational operating speed, which results in lower
amount of vibrations for slow speeds. Where low amount of vibrations presents challenges of
measurability.
In order to detect the low amount of vibrations before severe cavitation development the
optimal solution of measurability must be applied. Therefore, the research investigates the
optimal solution for cavitation detection in terms of accelerometer configuration based on sensor
position and sensor type in relation to reliability.
The results of the study found that the fault mode vibrations caused by cavitation were
detectable at lower rotational speeds than what industry recommended as the limit for an
implementation. Additionally, the position and sensor type results in different performance to
detect cavitation at slow pump speeds. An optimal configuration was found for the specific use
case of the rotary bi-winged positive displacement pump.
Examensarbete MMK TRITA-ITM-EX 2019:435
Gustav Osswald
Sammanfattning
Prediktivt underhåll baserat på tillståndsövervakning använder sensorer och systemdata för att
förhindra skador, innan fel uppstår, vilket möjliggör att schemalägga en service vid en optimal
tidpunkt. Tillståndsbaserat prediktivt underhåll estimerar tidpunkten för när fel uppstår i system
baserat på a-priori information, som har visats vara mycket mer kostnadseffektiv än traditionella
underhållningsmetoder. Typiskt finns tillämpningar där vardera av underhållsmetoderna,
reaktivt-, förebyggande- eller prediktivt underhåll kan vara bäst lämpade. De dyra driftstoppen
inom industri har lett till fokus på utveckling av prediktiva underhållsmetoder som ofta är ansett
som den optimala lösningen i dessa förhållanden. För att en prediktiv algoritm ska kunna
utvecklas krävs djup kunskap om systemet och stor mängd data att basera en algoritm på.
Projektet fokuserar på vibrationsanalys och tillståndsövervakning av maskinen
AkzoNobel - intelliCURE strängspridare som används i lamell- och balkindustrin. Forskningen
är specifikt inriktad mot att upptäcka kavitation i transportpumparna som används i
applikationen, där kavitation är ett relativt vanligt förekommande destruktivt fenomen som
uppstår i pumpar. Kavitation, som är formandet och implosion av håligheter i vätska, leder till
chockvågor som resulterar i vibrationer. Beroende på intensitet av förekommande kavitation, så
kan det i tid leda till interna skador och resultera i läckage. Typen och intensiteten av
förekommande kavitation i systemet är beroende av rotationshastigheten, som leder till lägre
vibrationsgrad för lägre hastigheter. Där låg vibrationsintensitet presenterar utmaningar i
mätbarhet.
För att detektera den låga graden av vibrationer innan allvarlig kavitation utvecklas måste
en optimal lösning för mätbarhet användas. Därav undersöker denna forskning en optimal
lösning för att upptäcka kavitation med hjälp av accelerometerkonfiguration baserat på
sensorplacering och sensortyp samt hur det relaterar till pålitlighet.
Resultatet från studien visar att mätbarheten av vibrationerna som uppstår vid lägre
hastigheter är bättre än vad som rekommenderas som gräns för användandet av vibrationsanalys
på industriella applikationer. Samt att positionen och typen av sensor resulterar i olika
möjligheter att detektera kavitation vid låga pumphastigheter. En optimal lösning hittades för det
specifika användningsområdet av en dubbelvingad vingrotorpump som studien applicerades på.
Acknowledgement
The project resulted in great learning and provided interesting opportunities regarding
condition monitoring, vibration analysis and predictive maintenance. I would like to
express my special appreciation and gratitude to the following for their contributions to
my research and this thesis:
Associate Professor Hans Johansson for his discussions and acting as examiner of this
thesis.
Ph.D. candidate Xin Tao for her research insights and great supervision, resulting in a
deeper study and additional learning.
Researcher Fredrik Asplund for his guidance regarding research methodology and re-
search methods.
Researcher & Lab Chief Ulf Carlsson for his assistance of data interpretation and guid-
ance regarding vibration analysis.
Akzo Nobel Adhesives AB and the machine department for providing the opportunity to
carry out this thesis at an industrial cooperation and granting the resources and support
regarding the application and pump malfunctions.
Andreas Ritola for his supervision, close collaboration and guidance of industrial pro-
cesses and equipment.
Donald O’Boyle III for his great supervision, assistance and guidance of software and
applications.
PCB & Omniray AB for their guidance of accelerometers and borrowing of a lab-sensor.
ifm electronic ab and Ingemar Sjöberg for their interest and guidance of both hardware,
software and the sponsoring of a signal conditioner and sensors.
Contents
Contents
1 Introduction 3
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Maintenance methods . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Vibration analysis of pumps . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Predictive maintenance . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.2 Cavitation & vibration analysis of pumps . . . . . . . . . . . . . . 6
1.2.3 Industrial application . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Research Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Delimitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5 Research Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6 Ethical & sustainable considerations . . . . . . . . . . . . . . . . . . . . . 13
1.7 Report outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2 Frame of Reference 15
2.1 Industrial maintenance approaches . . . . . . . . . . . . . . . . . . . . . . 15
2.1.1 Reactive maintenance . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.2 Preventative maintenance . . . . . . . . . . . . . . . . . . . . . . . 16
2.1.3 Predictive maintenance . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.4 Condition monitoring techniques . . . . . . . . . . . . . . . . . . . 18
2.1.5 Reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Pump characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.1 Rotary positive displacement pumps . . . . . . . . . . . . . . . . . 21
2.2.2 Failure modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.3 Cavitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3 Accelerometers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.1 Accelerometer position . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.2 Accelerometer type . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4 Vibration Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.1 Time domain analysis . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.2 Frequency domain analysis . . . . . . . . . . . . . . . . . . . . . . 32
2.4.3 FFT and normalized FFT . . . . . . . . . . . . . . . . . . . . . . . 33
3 Implementation 35
3.1 Laboratory test specification . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.1.1 Test rig . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1.2 Sensor data acquisition . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1.3 Tested accelerometers . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.1.4 Signal processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Preliminary tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3 Case study and test specification . . . . . . . . . . . . . . . . . . . . . . . 40
3.4 Evaluation of tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4 Results 47
4.1 Preliminary test results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2 Case study results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2.1 Results 30 rpm study . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.2.2 Results 50 rpm study . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.3 Optimal accelerometer configuration . . . . . . . . . . . . . . . . . . . . . 59
6 Future recommendations 69
6.1 Predictive maintenance development . . . . . . . . . . . . . . . . . . . . . 69
6.2 Future research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Bibliography 72
Abbreviation Description
ML Machine Learning
1
NPSHa Net Positive Suction Head available
2
Chapter 1
Introduction
An introduction is given to the study, the subject and the over all application the study is
based upon. Initially the Background section 1.1 gives an introduction to how vibration
monitoring is useful, industrial maintenance methods, condition monitoring and vibra-
tion analysis for pumps. Followed by the Purpose section which describes an application
area for the research and further specifies the purpose behind the research. Thereafter,
a statement and formulation of the research question is presented together with the de-
limitations and the methodology that was chosen for the research. The chapter ends
with a brief description to the other following chapters in the report.
1.1 Background
In this section a brief description is given on the background of the subject of maintenance
methods such as predictive maintenance and vibration analysis of pumps with causes of
failure such as cavitation. Further detailed information of the subjects are presented in
chapter 2.
3
[3]. Predictive maintenance algorithms are based on statistics, Bayesian models, machine
or deep learning methods. In turn, predictive maintenance can be based on either logs
and alarms or on sensor data where the sensors present the condition of operation, named
condition monitoring. These methods and corresponding assemblies use the sensor data
to develop a span of what data is to be considered as normal operation, which provides
the opportunity to recognize anomalies in the system. These predictive maintenance
algorithms usually find the signs of mechanical wear and other types of approaching
failures before they occur [3][22]. In order to perform any predictive maintenance or
fault detection in a system, a large amount of historical data is needed to implement the
statistical or machine learning predictions. Although it might not always be possible
to gain information from a predictive maintenance model due to numerous reasons, one
can still gain insight into the workings on a piece of equipment by studying its logs,
alarms and sensor data [40]. The use of events, alarms and condition monitoring based
on logged sensor data and process parameters pose many challenges and have not yet
been fully explored [40]. Numerous sensors and their corresponding data have shown
to give relevant information, which can be used to estimate the health or lifetime of
components. Specifically, it has been shown, that condition monitoring with different
sensor data can give insight regarding the health of a pump. Condition monitoring
based on vibration measurements with the use of accelerometers has been shown to be
one of the most prominent of methods, as vibrations occur in a system as a result of the
development of a failure. A further presentation of the different maintenance methods
are given in chapter 2 section 2.1.
In order to monitor the condition of a system, the failure modes must be profiled
[8]. For rotary pumps, there are many failure causes such as cavitation, poor lubrication
between seal faces and bearing failure vibrations. Typically these are the main causes
of leakage at seal surfaces [8]. Leakage at the axial seals are heavily affected by the
vibrations in the system.
Cavitation is a phenomenon where formation of cavities (bubbles) develop within
the fluid. These cavities later implode causing a shock wave transmitting and causing
vibrations in proximity to the implosion. Depending on the extent of the present cavi-
4
tation it can cause serious damage to pumps [30]. If the implosion is in close proximity
to a mechanical surface, cavitation erosion phenomenon may easily occur, resulting in a
lower component lifetime [50][8]. Such damage can greatly reduce the lifetime of seals
and other internal components [8].
Figure 1.1: Example of blistering on a bearing to the left and erosion on a gasket to
the right, possibly caused by cavitation.
1.2 Purpose
In this section a presentation is given to the purpose of the project where an industrial
application lay the foundation of the interest in the corresponding accelerometer con-
5
figurations involved in the research. A motivation is thereby given to the interest in
accelerometer type and position.
6
their unique frequency responses and resonance, a multiplicity of acceleromter must be
tested where the type of is of relevance to the measurability.
The vibrations caused by failure modes are unique in amplitude and frequency de-
pending on what caused them. The defect vibration could, for example, occur periodi-
cally, dependent on the rotational speed, be random in nature or have high/low frequency
content. The types of signal processing are dependent on these vibration characteris-
tics. Where cavitation both occurs periodically and is random in nature, the noise of
the vibration signal lay within the high frequency domain approximately 4-25kHz. Vi-
brations from other failure modes such as poor lubrication or bearing failures are of
periodic characteristic and are dependent on the rotational speed, these usually appear
within the low frequency domain for slow speed applications, which are hard to detect
with the normally used piezoelectric accelerometer for slow speed applications since the
IEPE sensors does not provide the option to measure frequencies down to 0Hz due to
accelerometer type functionality. Since the cavitation occurs in different parts of the
pump, the position of the accelerometer is of relevance as the location of the sensor
should be in close proximity to the vibration source, in order to get the best signal for
cavitation detection the position is of interest.
7
Figure 1.2: The industrial application, AkzoNobel intelliCURE seperate spreader.
The correlation of how specific sensor data could prove useful for fault detection in
pumps, specifically the rotary bi-wing positive displacement pump is an area of interest.
This type of pump is used in the separate spreader, where two identical copies trans-
port glue and hardener from their respective tanks to two separate application nozzles
spreading the glue or hardener on lamellas. One of the most common failures in this
kind of pump is leakage at the axial seal [8], which is verified by experts at AkzoNo-
bel. Normally these seals have a rather short lifetime considering the pumps being used
within production systems [8]. The seal in this particular application has a range of
lifetime from 1-18 months heavily dependant of the operating conditions.
One of AkzoNobel’s problems is the lack of reliable automated monitoring of a faulty
axial seal in the pump for their application. A damaged seal can lead to potential low
volumetric leakage of glue or hardener along the axis connected to the motor and gearbox.
Where as little as one droplet per day, in turn could lead to production downtime, if
this fault isn’t manually detected. A collection of adhesive inside the gearbox could lead
to the destruction of the gearbox or motor. Although the broken component is a loss in
itself the larger economical impact is caused by production downtime.
In order to predict when a fault occurs one has to indirectly examine failure causes,
also denoted as failure modes, that can lead to leakage, such as cavitation, poor lubrica-
tion and bearing vibrations. If the failure causes can be examined with sensor data, then
a predictive algorithm can be developed with this data to find patterns of the condition
monitoring. An algorithm like this could in turn predict how the lifetime of the seal
8
changes when the different failure causes are active for a certain amount of time in the
system. Real time analysis of vibrations and spectral frequency analysis in a system
has been successful to detect and examine the most common of these failure causes,
such as cavitation [32][44][42]. Where the Fast Fourier Transform (FFT) is a useful tool
for transferring real time data to frequency domain. Earlier research has stated that
vibration analysis and signal processing is successful for cavitation detection. The most
relevant damaging failure mode to monitor is cavitation since the phenomenon and the
occurring vibrations have a big impact on the lifetime of the axial seal. Some other error
modes that are a negative factor to the lifetime of the seal are poor seal lubrication,
bearing vibrations, misalignment of bearings or shaft, imbalance of shaft or impeller.
The position of the accelerometer has been shown to impact the readings and thereby
also the reliability and maintainability [42]. Since it is known that vibrations propagates
through structures, another interesting area is how machines mounted on the same base
cause a negative effect on component lifetime. As in study [4] where they are exam-
ining two pumps mounted 15 cm apart and how the vibration from one pump can be
detected by the accelerometer mounted on the other pump [4]. Even though the research
in [4] was not within the focus of this research scope it was still an interesting area for
the application of the tested pump which was similarly tested for the application under
study. A major concern for cavitation detection in a slow speed application is using an
optimal accelerometer configuration, as the slow speed results in low intensity of vibra-
tions making it harder to detect development of failure modes than for stronger signals.
Therefore the research examines the possibilities of detecting developing failure modes
at low operational speeds of a rotary bi-winged positive displacement pump.
The research focuses on examining vibrations with one or multiple high-bandwidth
accelerometers, FFT and vibration analysis to detect the common failure mode cavitation
that in turn causes leakage at the axial seal for a rotary bi-winged positive displacement
pump. These main causes have been shown to be, cavitation, poor lubrication and bear-
ing vibrations [8][44]. The following research questions were derived from the problem
formulation.
Given the industrial challenges and recent research advances the following main re-
search question was defined.
9
Hence an explanation for the criteria for how an optimal accelerometer configuration
solution was evaluated is given. Since the work involves preparation for predictive main-
tenance, the result from the study should find the best position to classify a signal as
normal or cavitation during current operating conditions. Therefore the optimal solution
is determined by the difference between a signal from normal operating conditions and
a signal from faulty operating conditions (cavitation in the system). The evaluation was
done both with manual spectral analysis and vibration anaylsis. Values of interest in
the evaluation was comparing the peak value of the cavitation noise from time domain
and the energy for an interval where the high frequency cavitation noise was present.
Further description of the implementation is given in chapter 3 where the method of
evaluation is further presented in section 3.4.
The second research question was defined in order to validate the use of earlier applied
methods to a different type of pump at slower speeds.
• What similarities of cavitation detection using vibration spectral analysis exist be-
tween the accelerometer data from a gerotor pump and centrifugal pump compared
with a rotary pump at slower speeds?
The study was focused on the structural vibrations generated in the pump from
normal running conditions in comparison to the occurrence of cavitation. The optimal
accelerometer configuration in terms of type and placement are evaluated based on the
difference between the signals from normal versus cavitation operating condition. Where
the comparison of cavitation detection between a slow speed application of a rotary pump
was tested and compared to the results of the earlier research in [8][9].
1.4 Delimitations
To narrow the scope of the project, some limitations were determined. The main entries
can be seen in the list below with further description presented in the following para-
graphs.
5. Maximum amount of data analysis was set to one packet per accelerometer con-
figuration.
10
7. The sample rate was set to 50kHz resulting in a frequency spectrum of 0-25kHz
after transformation.
8. Only FFT, normalized FFT and simple filters such as High-Pass Filter (HPF),
Band-Pass Filter (BPF) were examined, no other transformers were evaluated.
9. There was no tests to validate that the vibrations were damaging to the pump or
its components.
10. No confirmation was given to the type or amount of cavitation formation in the
pump.
Firstly, pumps are affected by more failure modes than what has been explicitly dis-
cussed previously (cavitation, poor lubrication, etc). A study of all the failure modes
would be too large for a thesis project to carry out due to the extensive testing re-
quired for each failure mode. Some examples of failure modes that was excluded from
the research are poor lubrication, bearing vibration, misalignment, imbalance of axis,
imbalance of impeller and mechanical looseness.
The tests were only carried out on one specific type of pump (rotary bi-winged
positive displacement pump), there was no tests on other pump variations. The study
focuses on the optimal accelerometer configuration for the specific use case of the pump
in the system application and specifically for cavitation detection. Where only two
types of accelerometers were tested, these were the Integrated Electronics Piezo-Eleztric
(IEPE) and the Micro Electro Mechanical Systems (MEMS) accelerometer.
Only a smaller amount manual analysis was performed, therefore it was limited to
only analyze a maximum of one packet (graphs) of data per accelerometer position. Only
one size of data packets of 20 seconds worth of data was evaluated with a fixed sample
rate. The sample rate was set to Ts = 50kHz, as it is important to have a high sampling
frequency since the signal is prominent in high frequency.
The transform of continuous data to discrete domain was performed mainly with
FFT. Performance of the softwares were not evaluated, where the softwares used in the
study were ifm VES004, MATLAB and its application Signal Analyzer, no comparisons
between the options of software was done. The software choice was dependant on the
options for filtering, transformers and availability.
For examination of how vibrations propagate from one pump to another the study
only included tests on the existing separate spreader at the AkzoNobel lab. Where the
results from these tests are not included in the report. Adapting the solution from the
study to an optimal solution to detect more failure modes such as bearing vibrations,
miss-alignment, imbalance of axis, imbalance of impellers or mechanical looseness is of
interest to the industrial market but was not elaborated or evaluated.
11
1.5 Research Methodology
Even though a quantitative method would be better suited to prove the earlier research
and to make the results from this project more robust and reliable, the scope demanded
a qualitative approach. This means fewer data packets were analyzed, where the tests
provided sufficient data to draw conclusion and discuss the result. As stated in section
1.6, manual data analysis and a qualitative approach results in a risk of letting bias
affect the results, discussion and conclusion. In order to reduce the risk and impact of
possible bias, a second opinion was given to the results.
The structure for the research methodology of the study initialized with a literature
study in order to establish what methods to be used and give a clear connection to
earlier research. Practical measurements were taken as reference in order to design a
case study. The way of evaluation, analysis and conclusions of collected measurements
and the corresponding performance was performed manually. A short and representative
presentation of the methodology in form of an illustration of the steps of methodology
over time can be seen in figure 1.3.
Figure 1.3: An illustration of time series of applied methodology for the study.
From the preliminary literature study a number of methods were decided and as-
signed for the practical measurements. Since vibration analysis has been an accepted
field for such a long time there are many established methods regarding data evaluation.
Some methods for signal processing that has shown practical for cavitation detection
in past research are frequency spectral analysis using FFT, packet analysis using Dis-
crete Wavelet Transform (DWT), and time-domain analysis using NLAR [44][16][35][33].
Where the most fundamental of these methods was chosen, namely vibration frequency
spectral analysis using FFT, in order to focus the research on the type of accelerometer
and position in terms of reliability.
As seen in figure 1.3 the preliminary study resulted in practical measurements and
preliminary tests of the available accelerometers. Thereby verifying the chosen methods.
The experimental preliminary tests formed the foundation to develop a case study upon.
The following case study had its position tested for multiple positions in relation to
the two positions used in the preliminary experimental test. The preliminary practical
tests were repeated for different rotational speeds, starting at 200 rpm, successively
lowering the rpm with decrements of 50 rpm down to the final tested speed of 50 rpm.
The positions chosen are based on recent research, industrial guidance and additionally
strategic positions based on cavitation occurrence. A further presentation of the case
12
study is given in chapter 3 in subsection 3.3.
13
sition. The last section in the implementation chapter, is a presentation of how the case
study was performed to provide a better understanding of the following two chapters
regarding results and the corresponding discussion and conclusion.
The report ends with recommendations and suggestions for future research and work.
This was kept separate to the recommendations for the specific application the study is
performed on.
14
Chapter 2
Frame of Reference
In order to grasp the subject, its challenges and results of the study a frame of reference
is given regarding the subject of accelerometer configuration and vibration analysis of
rotary pumps. An explanation is given to how the study correlates to predictive mainte-
nance and condition monitoring. Comparisons are given to earlier studies and explains
some of the industrial challenges within the area of condition monitoring and cavitation
detection. Where the chapter presents the most relevant information gathered from the
preliminary literature study.
15
2. The time for repair is short and the cost is small.
A pencil used in an office is a good example on a component for when a reactive main-
tenance is the optimal strategy, it is used until the ink runs out and is then repaired or
replaced.
Reactive maintenance is not an optimal solution for the pump application in this
study since the time for repair is rather long and the cost large. Additionally the pro-
duction disturbance for a failure might result in complete production downtime and is
therefore very costly.
16
2.1.3 Predictive maintenance
The third option to strategy of maintenance is predictive maintenance where a large
amount of historical data is used to estimate time of component failure. A predictive
algorithm could be based either on logs of system operation, alarms of failure or condition
monitoring. The condition based predictive maintenance strategy uses sensor values
showing the operational condition of the machine where the changes can be tracked to
detect anomalies in a system and thereby give a prediction of how data fluctuates. These
methods usually find an upcoming anomaly in far advance to the occurrence [5][41].
Predictive maintenance is considered the optimum choice dependant on the following
five preconditions [7].
2. Repair parts are expensive and not immediately available from storage or supplier.
Additionally there is one more condition that should be added to these factors that is not
included from the sources used to describe these methods. The additional precondition
that should be included therefore is stated by this study to be;
• When the lifetime is hard to estimate and not closely related to operational time
but rather operational condition.
All of these maintenance strategies has been shown to provide an optimal strategy
given different preconditions and circumstances of machine application [7]. The optimal
solution for the application covered in this study would deem either preventative main-
tenance or predictive maintenance depending on the measurability and how early the
development of failure modes could be discovered. If condition monitoring allows the
development of a predictive maintenance strategy this method would be best in terms
of efficiency, maintainability, reliability, sustainability and economical expenses for the
tested application. This is due to the application fulfilling the following preconditions; 1,
2, 3, 5 as well as the precondition suggested to be included by the author of this study.
Therefore the research studies the possibility of condition monitoring on the specific
process pump used in the application.
If an unexpected production downtime due to component failure occurs before the
scheduled maintenance the costs related to the downtime are large. Additionally with
uncertain conditions the lifetime of the axial seal is hard to estimate [32], in the pump
17
application, lifetime could vary from 1 to 18 months according to the AkzoNobel ex-
perts. In some cases customers ignore the recommendations of maintenance therefore
resulting in additional damages and causing a longer production downtime than the
original maintenance would result in. It is also hard to determine that the pumps in the
application does not operate outside the pumps Best Efficiency Point (BEP) at customer
locations. Operating outside the BEP may cause conditions developing cavitation and
greatly reducing the lifetime of components. Predictive maintenance could be based on
monitoring the conditions in a system following is a presentation of condition monitoring
techniques.
Figure 2.1: An illustrative example of a typical machine failure and the corresponding
warning signs according to National Instruments [20].
As seen in figure 2.1, the vibrations and harmonic signatures often carry the first
warning signs that the machine is prone to failure [20]. It has been shown that vibrations
18
in some cases provide up to three months notice prior to the actual failure date providing
a lead time for prevention of failure [20][42]. Monitoring the conditions and its data with
vibration analysis hardware and software helps to early predict failure and schedule
proper maintenance [2].
Even though predictive maintenance has proved valuable many industries are still
to implement both online condition monitoring and predictive maintenance methods. It
is still fairly common for route based condition monitoring where technicians manually
evaluate the conditions, either by human senses or with the help of instrumentation
[20][26][19].
The best and most comprehensive condition based maintenance programs utilize
sensor fusion or a variety of sensing technologies, where vibration monitoring generally
is a key component [15][49]. Vibration based condition monitoring has been estimated
to be the most widely used technique with 80% of the parameters measured likely to be
vibration based [37]. Although, vibration based condition monitoring does not provide
information to all sources of failures and is limited to monitor the mechanical conditions.
Therefore in order to utilize a more reliable and efficient maintenance program it must
include additional diagnostic techniques. Some of the known available techniques are:
• Vibration based
• Acoustic based
• Current based
• Thermography
• Tribology
• Visual based
Where vibration analysis and monitoring of vibrations is the most widely used and
reliable technique for condition based maintenance [31][37].
2.1.5 Reliability
Since reliability is a central part in the thesis as a part of the main research question,
a presentation of its definition in relation to engineering is given. The term reliability
can be defined in multiple ways, one way to describe it is as the ability of a system
or component to perform its required functions under stated conditions for a specified
period of time. Where reliability engineering focuses maximizing the component lifetime
and decrease the risk for early failures [26].
This is closely related to the component lifetime and certainty of system performance
and health, which correlates to condition monitoring and predictive maintenenace.
19
Figure 2.2: The reliability bathtub curve showing the change of failure rate over time.
A commonly used tool in the concept and evaluation of reliability is the bathtub curve
hazard function which illustratively explains the observed failure rates due to infant
mortality, random failures or wear out failures, this illustration can be seen in figure
2.2. As the study focuses on detection of cavitation using an optimal accelerometer
configuration for the stated application, it results in an increase of reliability for the
machine as it can help to reduce the infant mortality rate as well as the wear out failures.
If cavitation can be detected in an early stage before the level of cavitation causes severe
damage to components the reliability of the pump and there by the entire system can
be increased. When an optimal accelerometer configuration is applied it increases the
reliability of the system as both the lifetime can be increased and a reduction to the
infant mortality rate of components. For slow speed applications the subtle early changes
of system operating conditions are hard to detect where the optimal configuration is a
need to increase the reliability. The reliability of a pump could in terms be increased if
the failure modes were monitored allowing for development of predictive maintenance,
following is a presentation to pumps and their corresponding characteristics.
20
Figure 2.3: The functionality of a centrifugal pump, retrieved from [47].
The application for the pump is of high relevance to the design decision of machinery
and what pump to include in the system [32][38]. The type of pump best suited for an
application is dependant on many factors, example of important factors are the type
of liquid to transport, the distance it should be transported and the volume needed
over a specific time frame. Since the pumps operate at different ways it results in
different types of failure modes occurring at different parts of the pump and different
vibrations in the system, even though some similarities are shared [30]. Earlier studies
of vibration analysis and cavitation detection are typically performed on centrifugal
pumps, where the pumps uses the centrifugal force to push the liquid from the inlet
to the outlet. Centrifugal pumps are the most commonly used pump type by today
[38]. These pumps usually operate at high to very high operational speeds and is mostly
suited for low viscosity liquids and large volume transportation applications. Rotary
positive displacement pumps on the other hand which often operate at slow operational
speeds and are heavily used in the food industry with high viscosity liquids, which are
not equally popular to study.
21
the name, bi-winged positive displacement pump. The secondary gearbox often has no
change in gear ratio but has a main purpose to translate the single axis rotation to a well
timed dual axis output from the gearbox. It is of high importance to keep the timing
of the impellers at a fixed setting, since a change of just one gear tooth might cause
a collision of the impellers. Followed by the secondary gearbox is a number of seals in
order to prevent liquid from escaping the pump housing into the gearbox. On each of
the two axis entering the pump house an impeller is mounted to move the liquid from
the inlet to the outlet. The outer part of the pump is covered by a pump housing cover
which closes the build of the pump. An exploded view of a rotary bi-winged positive
displacement pump can be seen in figure 2.4.
The rotary positive displacement pump operates by rotating its impellers in a pre-
cisely timed manner, creating lower pressure at the inlet than at the outlet. By the
rotation of the impellers the liquid is forced into the space between the impeller and
the pump housing. As the impeller turns the liquid is forced out of the pump house as
the impellers connect in the middle of the pump. An illustration of how rotary positive
displacement pumps operate can be seen in figure 2.5.
22
Figure 2.5: An illustration of the functionality of the rotary bi-winged positive
displacement pump, retrieved from [11].
As described in the earlier section 2.2 each pump type is affected by different failures
even though they share some common failure modes. A fairly common failure for the
rotary positive displacement pump is damage to the axial seals due to various failure
modes. The seals can be seen in figure 2.4 between the pump housing and the impellers.
As a result when these seals are damaged, leakage could occur either leaving the pump
housing or entering the gearbox.
As mentioned in section 2.2, the majority of research has been focused at the more
commonly used centrifugal pump which often operate at higher speeds than the rotary
positive displacement pump [38][44][8]. Since earlier research has not fully explored slow
speed applications, the industry is still not convinced vibration analysis is a viable option
even for slower speeds [45]. Where article [45] is examining a slow speed application for
the earlier mentioned centrifugal pump. The argument is given that the sensors on the
market are not ideal and can’t measure frequencies down to 0 Hz and that the vibrations
occurring at slow speeds are too weak to capture or recognice the anomalies from the
failure modes. It was recommended not to consider vibration analysis for slower speeds
than around 100 rpm, since the validity for slow speed applications is considered poor
for many of the failure modes. Following is a presentation to the different types of failure
modes in pumps.
23
which results in additional multiple failure modes. When cavitation occurs in a system it
results in uneven flow due to fluid pulsations as a result of being starved of liquid. When
the cavities implode it results in excessive vibrations, often propagating through the en-
tire structure. The vibrations result in uneven loads, resulting in additional mechanical
damage to components such as impellers, seals and bearings [24][27].
2.2.3 Cavitation
Cavitation as mentioned both in chapter 1 section 1.1 as well as in previous subsection
2.2.2, is an unwanted phenomenon also referred to as a failure mode which occurs in the
transportation of liquids. The phenomenon is the development of cavities or bubbles,
which for pumps occurs at the suction port. These cavities are then transported to
the delivery side of the pump where a higher pressure forces the cavities to implode
resulting in a shock wave from each of the cavities and vibration when the shock wave
hits structural components [44][24][27]. A visual example of cavitation can be seen in
figure 2.6.
Figure 2.6: Cavitation in two different type of rotary positive displacement pumps.
It occurs in pumps when the inlet pressure drops below the vapour pressure for
the media transported by the pump. The condition required for cavitation to develop is
defined as the Net Positive Suction Head (NPSH) available is lower than the Net Positive
Suction Head required which is the opposite of the equation used for designing pump
systems stated by equation 2.1.
24
System and pump developers go to great length when designing pump systems to
ensure that equation 2.1 is fulfilled even when operated outside its best efficiency point
(BEP) to prevent cavitation bubble collapse inside the pump. Where centrifugal pumps
are usually operated smoothly within the range of 85% to 110% of its respective BEP
[24].
NPSH takes multiple factors into consideration such as the suction piping and con-
nections to the inlet, the fluid velocity and the absolute pressure of the transported
medium [8][24]. In essence, cavitation occurs as a result of reduction in suction pres-
sure, which can be the result of a clogged inlet, an increase in suction temperature or a
increase of operational pump speed or flow rate above the levels the pump has been de-
signed [24]. If the NPSH margin from available and required is small it heavily increases
the occurrence of cavitation more frequently, this could be due to design constraints or if
the system develops a malfunction such as clogging of the suction port of the pump. The
NPSHa is calculated using characteristics from the pump’s inlet nozzle and is thereby
independent of the pump or the pump characteristics which assists system designers to
avoid a design which may cause cavitation even though the pump is operated within
BEP. The NPSHr is defined as the calculated amount of NPSH that is required to avoid
cavitation in the pump and is independent of the system characteristics [24].
Since the research focuses on cavitation detection it is important to understand the
phenomenon to an extent that the different damaging factors occur in different areas
thereby being important for the position of a sensor.
The failure mode of cavitation in a system or a pump leads to multiple damaging
factors to a system or pump, as defined in the article [24] the four symptoms are as
follows.
(a) Erosion: Initially when the cavities are formed at the low pressure inlet of the pump
it does not introduce any erosive capabilities. However when these cavities are
transported to a high pressure at the outlet side of the pump they instantly implode
exerting enormous local stress on surfaces in close proximity to the collapse, causing
damage to internal components such as the impellers and surfaces such as the pump
house casing. Signs of erosion due to cavitation in pumps will appear as pitting due
to the water hammering action caused by the collapse of the cavities. When the
bubbles implode they result in a shock wave transmitting and hitting surfaces at
the local speed of sound just like a jet beam of liquid, which can results in surface
damage if the pressure is higher than the ultimate strength of the material. Damage
from cavitation in form of erosion on a pump housing cover can be seen in figure 2.7
25
Figure 2.7: Erosion damage on the pump housing cover, caused by cavitation in a
rotary positive displacement pump.
It has been shown that the damage due to cavitation increase rapidly corresponding
to the volume of the fluid. The rates of present erosion increase by a factor four when
the capacity of the shock less flow is raised from 100% to 120%, which is generally
1.1 to 1.3 times the capacity of the BEP flow.
(b) Vibration: The vibrations that propagate due to the cavitation are characteristically
random in nature and given as pulsations of high amplitude from the delivery of fluid
to the outlet. However it is differently argued in research if these vibration are found
in low frequency range such as 0 to 10 Hz or from the noise propagating as vibration
in high frequency domain with a range from 4 kHz upwards depending on the pump
type, design and amount of cavitation [24][44][8].
(c) Noise: Usually cavitation results in noise in terms of a sharp crackling sound as
the cavities are collapsing under the high pressure. This noise has a correlation
which can measure the severity of the cavitation. The noise usually occurs in close
proximity to the inlet of the pump. If the crackling noise is random in nature and is
present with intensity knocks, it indicates cavitation in present in the suction side
as recirculation. This type of cavitation does not lower the efficiency if the NPSH
requirement is met but still generates conditions which can cause damage to the
pump. As described there are multiple types of cavitation, these types are classified
based on the location of the cavities inception and location for the implosion of the
cavities, additionally each type is accompanied with its own range of acoustic noise
emission. Sheet cavitation is the first type, which forms cavities across the vane
surface when the pump is operating close to its designed flow with low pressure
at the pump inlet. This type of cavitation results in a broad band noise, with
low amplitude, it typically appears from 2 kHz to 40 kHz depending on the pump
26
characteristics [24][36]. The second type is called cloud cavitation, where cavities
form downstream of the cavity sheet and occurs if the pump is operated outside of
its designed flow with low pressure at the suction side of the pump. Cloud cavitation
is the loudest type of cavitation which is said to sound like "pumping gravel". This
type of cavitation generally appears in high frequency domain, such as 20 kHz to 40
kHz. Vortex cavitation is the last type of the three types of cavitation, it usually
occurs when the pump operate at very low flows and is a highly unstable form of
cavitation. It usually develops in the inlet back flow regime. Even though cavities
implode similarly to the other types of cavitation, it is considered as less damaging
since the collapse of the cavities usually occur away from solid surfaces. This kind
of cavitation generates a low frequency pulsating beat in the range of 1 Hz to 4 Hz,
this phenomenon is known as cavitating surge.
(d) Efficiency reduction: Since bubbles are created in the passage of liquid from the
inlet to the outlet, the volume of the transported liquid is reduced from one side
of the impeller to the other, thus resulting in a reduction of delivered liquid. The
lowered efficiency could be an indication of cavitation occurring in the pump or
system in advance to fully developed cavitation occurrence. The efficiency of the
pump can drop drastically over time to the point where the efficiency is considered
as poor [27]. In some occasions and research attempts it has been found that the
efficiency slightly increased moments before initial cavitation occurs. A cause of this
is reasoned to be the reduction of friction at the initialisation of cavitation due to
the separation in the flow, right in advance to the implosion of the caveties [27].
2.3 Accelerometers
An accelerometer is a type of sensor that can measure the displacement, velocity or
acceleration at a given point [26]. There are different types of accelerometers that op-
erate with different functionalities at different ways [43]. For instance, an accelerometer
can have one or multiple axis giving the opportunity to take multiple orientations of
measurements for a certain position.
When selecting an accelerometer for a specific application it is important to choose
one appropriate to its purpose and to the environment it will be exposed to. There are
27
a number of factors effecting what technical requirements the application needs, where
some of the important factors are accelerometer type, sensitivity, frequency bandwidth,
self resonance frequency and range. In addition to these factors there are a lot of con-
siderations that should be taken into account before deciding on a sensor for a specific
application.
Some aspects concerning the condition monitoring performance are the position with
good mounting and type of the sensor where both of these aspects results in different
capability of readings.
28
to the bearing in the respective direction as close to the source of vibration as possible
[43].
Another factor that is important to consider is the way of mounting, as all of the
possible mountings allow for small slack and thereby acts as a filter. There are many op-
tions for accelerometer mounting, such as stud, screw, adhesive and magnetic mounting.
There is a fairly unanimous agreement of what method gives the best response, however
there still are discussions on when the different methods are possible to use for high
frequency domain. According to different suppliers the different mounting methods are
viable for different measurement ranges. According to accelerometer producer CTC, the
mounting methods ranges for stud mounts and adhesive mounts in theory are around
0-15kHz, they also specifically state that real environments often lower the possibility
of range, resulting in a stud mount range of 0-10kHz and an adhesive mount around
0-3kHz [34]. Another accelerometer producer PCB Piezotronics states the performance
of accelerometer accuracy decreases with a logarithmic scale at higher frequencies where
the noise increases for high frequency where different mounting options have different
performance of noise reduction [14].
According to the company Hansford Sensors, bad mounting of accelerometers in vi-
bration monitoring of industrial applications is surprisingly common in practice [43]. The
unanimous known best performance of mounting techniques is said to be stud mounting
where the more stiff options give the better frequency response. For adhesive options
the same conclusion applies regarding the stiffer connection the higher bandwidth.
29
of operation and functionality. The IEPE versus the MEMS accelerometers are of high
relevance where the IEPE sensors have been used for a long period of time and are
still the most widely used. On the other hand the MEMS sensors are rather new to
the industrial market and functionalities allow them to measure for slower speeds due to
being able to measure lower frequencies. According to the research paper [1] the demand
for wireless sensing nodes is increasing where the MEMS sensor is a viable option. As
IIoT, industry 4.0, predictive maintenance and condition monitoring are growing fields
and the performance of the MEMS sensors are increasing while its prices are decreasing,
the industrial MEMS accelerometer are getting closer to the performance of the more
traditional IEPE or ICP accelerometers. The article found that two of the three tested
MEMS sensors achieved good test results with similar performance to the IEPE type,
where the big difference was the phase shift of 1-5%, it also states the following three
sentences; "MEMS sensors could be a good alternative to standard sensors mainly for
wireless implementation as there is no need to carry heavy charge amplifiers, but the
choice has to be made according to specifications and through validation tests. MEMS
sensors have also to resist harsh environments using an appropriate packaging. More
investigations with various MEMS accelerometers to understand the future direction for
improvements are being carried out.". This lies as a foundation for this research where
the performance of a MEMS accelerometer is compared to a IEPE accelerometer for
cavitation detection at slow speed applications of a positive displacement pump.
30
2.4.1 Time domain analysis
Time domain is the most fundamental of methods, where vibrations are measured and
plotted over time with displacement, velocity or acceleration as amplitude. This gives
a clear view of how the vibration change over time in a system. For example if the
vibration amplitude increase over time for a set system state of operation, this could
prove that a fault is developing [42][44].
Time domain is useful for the over all evaluation of the system state. It helps to
study subtle changes to the vibrations in the operation. The downside of time domain
analysis is that it could be hard to detect the distinct patterns or differ fault modes from
one an other. In other words, it is very hard to distinguish the source of vibrations from
the time domain signal [7][13]. As different sources of vibration results in a superposition
of vibration, creating a more complex signal containing information from all vibration
sources, this results in limits to what can be recognized in a complex signal. Time
domain analysis is typically devoted to peak, average, Root Mean Square (RMS) and
envelope values of the vibration force amplitude.
An example of the vibrations in time domain can be seen in figure 2.8 where the
plot shows a complex signal hard to distinguish the multiple sources of vibration from a
running condition.
Figure 2.8: The raw time domain data where the plot shows the complex vibrations
from the test rigs normal operating conditions recorded in ifm VES004 software.
Typically, time domain analysis is devoted to detect the integral performance of the
tested part. For example if the intention is just to measure a peak or an average level
of vibrations no other domain than time domain has to be considered. Typically for
31
condition monitoring, the time domain data is consistently compared with historical
data. However, since system vibrations often change over time and other sources of
vibration can propagate to the measured point, it complicates the whole method as
it is challenging to compare the acquired data with historical data and draw accurate
conclusions in such a format [31]. Time domain analysis is a viable option for real
time applications where specific signals carry their features and are relatively easy to
distinguish if the environment is controlled [44]. As most data is recorded in time domain
a quick analysis of the data can be given in advance of choosing another analysis method.
32
Figure 2.9: At the left, four different signals where the bottom one is the wave
superposition, to the right their corresponding frequency spectra.
The frequency domain and the corresponding frequency spectrum is widely used
in vibration analysis where components in a system have different sources of vibration
resulting in different fundamental frequencies [27][7].
Since the frequency domain analysis has been proved viable for similar research
[44][8], the decision was taken to analyse the vibrations mainly in frequency domain.
fs ≥ 2fc (2.2)
FFT is an optimized implementation algorithm for the DFT computation [21]. It is a
commonly used tool or method for vibration analysis, since differences to the vibration
signals from continuous time are hard to distinguish from one and other. The vibrations
measured in continuous time signal consist of multiple vibration sources resulting in
different frequency of the vibrations for different type of vibrations and failure modes.
The literature study proved FFT to be a viable transformer for data in similar vibration
analysis studies [44][8]. A normalization can be done to the abscissa of the frequency
domain where a change of speed does not affect the plot. Resulting in the vibrations
33
from a source, occurring at its fundamental frequency still being found at the same
position for different speeds after the normalization, although this might cause a change
of amplitude [31]. A version of FFT known as H-FFT was initially tested but deemed
a poor option for cavitation detection as it resulted in dampening of the high frequency
noise.
34
Chapter 3
Implementation
This chapter presents the practical work and how the study was constructed. A presen-
tation is given to the test rig, the laboratory equipment, the preliminary tests and the
case study. As described in section 1.5 the preliminary tests were performed as prac-
tical measurements to assure the capability of measurement of the chosen method and
equipment. This assisted in the design of the case study which consisted of a number of
combinations further explained in this chapter.
35
The tests were as mentioned carried out on the AkzoNobel separate spreader where
two rotary positive displacement pumps can be individually run. In order to force
cavitation to occur in the system, the liquid flow was reduced at the suction side of the
pump by placing different orifice plugs at the connection of the tank outlet connected to
the pump suction side, resulting in a pressure drop, similarly as in study [9]. The orifice
plug was designed with CAD and 3d-printed in thermoplastic polyurethane TPU 95A,
using Solid Edge ST10, Cura and the Ultimaker 3 Extended. The plug consists of two
parts, the main plug and the interchangeable orifice. The part was constructed in two
parts to allow for a quick change of orifice and save time and material from the printing.
The plug and the different orifice of 10mm, 5mm and 3mm can be seen in figure 3.1.
Figure 3.1: The four 3d-printed parts, one plug and three interchangeable orifice parts
to control the suction flow and pressure.
During the collection of data the system ran collecting data according to the test
specification at a set sampling rate of 50kHz which is in accordance of the Nyquist
sampling theorem presented by equation 2.2. Each accelerometer type and position of
the sensor was collected for 20 seconds of raw data resulting in 1.000.000 data points for
each test.
Since the laboratory copy was not to be damaged in any way for the tests, stud mount-
ing of the accelerometers was removed as an option. Therefore the second best option
for high bandwidth would be adhesive mounting according to the industrial guidelines.
Since the sensors were planned to be moved between multiple positions an alternative
to the adhesive mounting was used. Flat surface nuts were glued to the pumps im-
peller housing with a stiff superglue. Allowing fast connection and disconnection of the
accelerometers with a bolt to the adhesive mounted nuts.
36
3.1.1 Test rig
The tests were performed in a traditional test environment, in an application setting for
the pumps rather than on a test bench. The test rig consists of a laboratory copy of the
AkzoNobel intelliCURE separate spreader. The machine structure is fairly similar to the
bench test rigs that has been used in earlier research [29][8]. The machine is constructed
with two parallel and equal transportation loops for glue and hardener. These loops
with equal design are constructed of a tank closely connected to the suction port to the
pump. The pump discharge side is followed by a pressure sensor and flow meter leading
by tubing up to a discharge nozzle which transport the liquid back to its tank. In order to
remove noise from water impacting the surface in the tank a tube was connected instead
of the traditional sparge. This does not affect the usability in a real environment since
the viscosity of the glue and hardener helps to protect against noise from the surface
tension in the tank. An illustration of the test rig can be seen in figure 3.2.
Figure 3.2: An illustration of one of the equally designed flow loops in the test rig.
The machine’s main purpose is to apply glue and hardener to lemallas with high
precision. Even though the machine transports high viscosity liquids the testing was
performed with water as the transported liquid. This results in risk of re-circulation at
both the suction and discharge side on the pump. Additionally since the application
uses variable speed for precise liquid delivery if cavitation develops it reduces efficiency
and the speed will therefore increase, causing further cavitation development resulting
in worse operating conditions.
37
of data. The packets recorded for each of the tested combinations were 20 seconds worth
of data. In order to fulfill the Nyquist sampling theorem the data acquisition recorded
with a sampling frequency of 50kHz since the cavitation signal was estimated to appear
within 4-20kHz. It was also the lower of the two alternatives of sampling frequency for
the DAQ. The technical specifications for the DAQ can be seen in table 3.2.
Specification Metrics
Connecting voltage: 24V
Current consumption: 200mA
Amount of connections: 8 (configurable)
Analog resolution: 12-bit
Sampling frequency fs : 50kHz or 100kHz
Measure bandwidth: 0-12.000Hz
Communication interface: Ethernet
Protocol: TCP/IP
The IFM VES004 software exports the recorded raw data with 9 informational
columns resulting in a size of approximately 100MB for each csv file. In total the
recorded data volume for all tests, preliminary, exclusive tests for AkzoNobel and the
test case resulted in approximately 20GB of stored data. The 12-bit resolution and the
20 seconds of recording with 50kHz sampling frequency, giving a total of 1.000.000 sam-
ples, deemed viable from the preliminary tests and were therefore used in the case study
aswell.
The VSE100 allows connection by Ethernet to a PC allowing storage of real time
data recording. This dataset of time domain data was exported for each accelerome-
ter configuration to a csv file format. The files could then be imported to MATLAB
where altering of the table could be made in order to prepare the data for mathematical
transformation in form of FFT or importation to the application Signal Analyzer.
38
Table 3.3: Table of the two types of accelerometers used in the research.
39
recommendation of 100 rpm up to a speed of 200 rpm. The lower limit was set to 50 rpm
to validate that the cavitation noise in high frequency domain would disappear. The
range was thereby set as 50-200 rpm. Since earlier studies have acknowledged vertical
and horizontal positions on the pump housing and the industrial guidelines equally
recommended these positions, these were the two positions used in the preliminary tests.
The orifice plug used for the preliminary tests was the 5mm orifice plug.
Pump Speed [rpm] Sensor type Sensor position Orientation Running condition
50 IEPE Impeller casing Horizontal/Radial Normal
50 MEMS Inlet impeller casing Vertical Normal
50 IEPE Impeller casing Horizontal/Radial Cavitation
50 MEMS Inlet impeller casing Vertical Cavitation
100 IEPE Impeller casing Horizontal/Radial Normal
100 MEMS Inlet impeller casing Vertical Normal
100 IEPE Impeller casing Horizontal/Radial Cavitation
100 MEMS Inlet impeller casing Vertical Cavitation
150 IEPE Impeller casing Horizontal/Radial Normal
150 MEMS Inlet impeller casing Vertical Normal
150 IEPE Impeller casing Horizontal/Radial Cavitation
150 MEMS Inlet impeller casing Vertical Cavitation
200 IEPE Impeller casing Horizontal/Radial Normal
200 MEMS Inlet impeller casing Vertical Normal
200 IEPE Impeller casing Horizontal/Radial Cavitation
200 MEMS Inlet impeller casing Vertical Cavitation
40
axial and radial orientation in reference to the pump. Therefore the majority of the
positions used in earlier research was tested in relation to the pump used in this study.
In agreement to the research, industrial applications refer to positions on the impeller
housing in the same orientations. However it is stated in both earlier research and in-
dustrial guides that the sensor should be placed as close as possible to the source of
vibration. Thereof some additional positions were introduced dependant on the cavita-
tion phenomenon presented in section 2.2.3. One of these positions which was decided
to be tested out of curiosity was the position right after the impellers at the discharge
side in an axial direction. This is since it is the closest position to the implosion of severe
cavitation which can cause surface erosion, as seen in figure 2.7.
Following figures present the tested positions with their respective labeling. These
positions are denoted as; T vertical top, H horizontal, A axial, I horizontal inlet, B
vertical bottom, R radial (mixed horizontal and vertical) and O horizontal outlet. The
location for comparison of the MEMS and IEPE performance are in close proximity to its
respective position. However a slight difference to the position of the compared position
did exist since the mounting technique was decided to be adhesive mounting with nut
and bolt. The pump application with some of the earlier labeling of the denotations can
be seen in figure 3.3.
Figure 3.3: Four accelerometers with adhesive mounting using nuts and bolts, at some
of the tested positions of the sensors.
41
positions have been tested for cavitation detection in recent research except positions
A5 and A6 which was added as interesting positions. These positions were tested for
speeds of 30 rpm and 50 rpm which was shown to be operational speeds carrying the
cavitation noise signal.
The evaluation for these tests was done both by calculating the energy difference be-
tween the normal and cavitation condition for the range of the noise and maximum peak
comparison in time domain, which was presented visually. The best visual presentation
of the position locations are presented in figures 3.4 to 3.6 where different views of the
pump present sight of the different positions.
42
Figure 3.4: Top left side view of the tested pump with the denotations at the tested
positions.
Where the positions T1, T2, R1, R2, H1, H2 I1 and I2 can be seen in figure 3.4 at
their corresponding positions from a top left side view.
Figure 3.5: Bottom left side view of the tested pump with the denotations at the tested
positions.
Similarly for the positions B1, B2, O1 and O2 which can be seen in figure 3.5 at their
corresponding positions in a bottom left side view.
43
Figure 3.6: Axial front view of the tested pump with the denotations at the tested
positions.
Lastly is a presentation of the positions A1, A2, A3, A4, A5 and A6 which can be
seen in figure 3.6 at their corresponding positions from a front axial view.
These were the positions of interest from the preliminary literature study, recom-
mendations from industry and proposed by the author, a description is given of how
the data evaluation was done for the signals from the type of accelerometer and tested
positions.
44
the position. This value shows the over all increase of vibration when cavitation occurs,
however it is not a good measurement in its own as the source of vibration is not iden-
tified in this way. But since the difference of maximum peak value is combined with the
difference in energy for the noise signal, the identification includes both level of vibration
force and the certainty that cavitation is present.
The maximum peak value was measured in time domain after applying the HPF. This
results in a value showing the vibration intensity of the noise. A peak value from the raw
measurements before applying the filter could be valuable for a predictive maintenance
algorithm but was not considered for evaluating the type of sensor and position since
the peak value in low frequency domain is not a sure indicator of the cavitation noise.
Power spectral density (PSD) shows the energy from signal strength as a function
of frequency. It is a measurement showing what frequencies correlates to strong signals.
It can be calculated as the integral from the start of the differing signal to its end.
Comparing two energies from two signals could be calculated by cross spectral density.
The PSD is presented by equation 3.1 where the energy is calculated as the integral of
the time domain signal x(t). According to Parseval’s theorem the energy is equal to the
area under the square of the magnitude of the frequency domain signal seen in equation
3.2. Since the FFT shows the PSD of the signal, the energy for a frequency spectrum
could be summed. The cross power spectral density is thereof the distribution of power
per unit frequency.
Z ∞
E= |x(t)|2 dt (3.1)
−∞
+f
X end
E = Es = X(f ) (3.2)
n=fstart
The frequency interval was determined from the presence of the vibration noise from
the cavitation, this was determined by the preliminary tests to be from 4.5kHz up to
15kHz. The energy spectral density is then compared for the two signals in accordance
to equation 3.3 where Ecavitation represents the energy of the signal from cavitation and
Enormal represents the energy of the signal from normal operating conditions.
45
The energy used for the calculation of the percentage in difference between the signals
was used either over a spectra of 4500Hz to 15.000Hz in order to see the general difference.
For the signals that showed a clear peak difference the percentage was taken over the
spectrum where the difference of the peak showed present.
An evaluation of the signal difference in percentage is valuable to easily understand
how big the difference is between the signals. However the negative effect of evaluating
the signal difference with a percentage difference is that the general signal strength is
lost. Therefore the energy from a PSD calculation is better suited for the evaluation of
an optimal solution to show the difference between the signals. The percentage difference
could be an addition of metric to include for a ML application where multiple dimensions
of metrics could be evaluated. It also shows what positions to be considered as valuable
at all. The percentage was given in order to better understand the magnitude of the
metrics, even though not considered as a more valuable evaluation than the energy
difference.
46
Chapter 4
Results
In this chapter the results from the empirical study is presented, analyzed and compared
to results from earlier studies and theory in the field. The field and frame of reference
of the study has been presented with the implementation. Following in this chapter, the
results from the preliminary study and the two case studies are presented.
47
Figure 4.1: Raw data from cavitation and normal operating state over 20 seconds of
recorded samples at four different speeds using the IEPE accelerometer at a horizontal
side position.
The cavitation phenomenon for the preliminary tests were done with a 5mm orifice
in accordance to subsection 3.1.
Figure 4.2: Raw data from cavitation and normal operating state over 20 seconds of
recorded samples at four different speeds using the MEMS accelerometer at a vertical
top position.
48
When the raw signals are transformed with the MATLAB FFT the signal shows
a clear difference for the cavitation signal in comparison with the normal operating
conditions. For the IEPE accelerometer mounted at a horizontal side position, the low
frequency vibrations are stronger than the noise developed by the cavitation in high
frequency domain. The high frequency vibration noise ranges from 5000Hz to 8000Hz
and has more energy than the normal state signal. This shows a difference of the noise
in high frequency domain between normal operating conditions and the failure mode
cavitation, this can be seen in figure 4.3. Even though the signal was somewhat different
between normal operating condition and cavitation in the low frequency domain, the
difference between the signals is easier recognizable for the high frequency domain noise
signal. The peak value in the high frequency domain also shows a clear difference between
the normal and faulty operating conditions.
Figure 4.3: The frequency spectra for cavitation and normal operating state over 20
seconds of recorded samples at four different speeds using the IEPE sensor at a
horizontal side position.
49
Figure 4.4: The frequency spectra for cavitation and normal operating state over 20
seconds of recorded samples at four different speeds using the MEMS sensor at a
vertical top position.
Since the clear difference was verified to be in the noise signal located in high fre-
quency domain, similarly as in the study [8] and [9], therefore a high-pass filter was
decided to be applied at 4000Hz for the future analysis of the noise signal.
Measurement of the vertical top position showed that both the MEMS and the IEPE
accelerometer were capable of showing a minimal difference to the measured vibration for
speeds of 30 rpm. The data was hard to distinguish for speeds of 30 rpm with the FFT,
therefore it was analyzed with a normalized FFT plot. The MEMS accelerometer had a
small spike at 0,45 rad/sample while the IEPE sensor could only show small differences
of data at 0,20 to 0,25 rad/sample for the same vertical top position. This practical
measurement also showed that signal strength for the MEMS sensor was stronger than
or the IEPE sensor. This was found using a normalized frequency plot and normalized
FFT as seen in figure 4.5. This was later verified and can be seen in the plots in appendix
A where a slight differences to the normal and cavitation condition could be noticed in
their corresponding FFT plot.
50
Figure 4.5: Raw data and normalized frequency spectra for cavitation and normal
operating state over 20 seconds of recorded samples at 30rpm for both MEMS and
IEPE at a vertical top position.
Thereby it was shown that there are differences of the type and position of the ac-
celerometer and that the force of vibrations are related to rotational speed for cavitation,
where 30 rpm was set as the limit. With these results the case study was designed for
speeds down to 30 rpm as the limit. A secondary case was designed for a speed of 50 rpm
which had clearer difference between the two signals cavitation and normal operating
conditions.
51
4.2.1 Results 30 rpm study
The study started with tests for a rotational speed of 30 rpm in order to prove the
limits of the measurability for different positions, where a higher speed study could act
as validity to the tests. The case study was performed according to the specifications in
section 3.3, where the denotation of the positions used in the tables can be seen on the
pump in figures 3.4 to 3.6. Following is a presentation of the evaluated test data from
the 30 rpm case study. Where the peak and energy difference between the two condition
signals can be seen in appendices A and C.
Table 4.1: The results from the tested positions in the 30 rpm case study.
The biggest difference for the signal energy in the spectra was found to be the hori-
zontal inlet position I1 and the horizontal axial position A1 for the IEPE sensor and the
horizontal axial position A2 and A6 for the MEMS sensor when the pump operates at
30 rpm. However the T1 and I1 positions had very close similarity of signal energy.
The peak value difference for the tested positions showed that the best position for
the IEPE sensor was either vertical top position T1 or the horizontal axial position A1.
For the MEMS sensor the peak difference showed that the two horizontal axial positions
A2 and A6 had the best performance of detecting cavitation at slow speeds.
When looking at the percentage difference over the peak spectra it is noticeable that
only some of the combinations shows a valuable difference between normal operating
conditions and developing cavitation. The configurations relative difference (percent-
52
age) shows that T1, I1, R1, A1 and A6 are all positions viable for the detection of
cavitation development. If only the relative difference would be used to evaluate an
optimal solution, then the radial position with the IEPE sensor would be the best con-
figuration for the application. Still, the peak and energy difference act as better suited
variables for evaluation as in the case of having a predictive algorithm built, the absolute
values give more information valuable for a remaining useful life calculation. If a mix of
the three evaluation variables would be used the configuration with the MEMS sensor
at position A6 shows the biggest over all difference of the three metrics.
The best performance of the combination of the two variables peak difference and
energy difference, would be as far from zero as possible, therefore in order to evaluate
the results a scatter plot was done to visually see the best performance configuration,
as seen in figure 4.6, where the numbers in the figure correspond to the number of the
tested position and accelerometer type.
Figure 4.6: A scatter plot of the 30 rpm tests, showing the number of the test in
accordance to its variable value of energy and peak difference for the noise signal.
In figure 4.6 it is possible to determine the best configuration. Out of all configura-
tions the result showed the MEMS sensor at a horizontal axial position at the delivery
side of the pump A6 was the optimal configuration. Which was also a position less com-
mon to use, suggested by this research to be studied. Viable position for the IEPE sensor
would be T1, R1, I1 and A1 has these have a far distance to the origin and they have
similar values. It is argued that the A1 position might not measure the cavitation signal
53
even if the signal, when present is stronger, if a window of 10 seconds worth of data was
used instead of 20 seconds, as can be seen in figure 4.8. While T1 has the biggest abso-
lute difference R1 and I1 still shows viable with having a more peak like characteristic,
which is easier to determine as a property of the noise of cavitation development.
The case study shows a clear trend that the MEMS sensor and IEPE sensor gener-
ally have different performance at different positions. The best position was found to
be in relation to the type for a specific speed application of a rotary bi-winged positive
displacement pump tested, as can be seen in figure 4.6 and appendix A.
The result of the MEMS sensor at the position A6 to be the optimal configuration
was most likely due to the position being in the closest position of the vibration source of
the cavity implosion due to the pressure change after the impeller. The risk for internal
back leakage between the impellers could also be a contributing reason for A6 to be an
optimal position, as the risk for back leakage increases as the pressure difference rises
before cavitation develops. This was the reason behind the suggestion of the position to
be studied, where the damage at the position from erosion could be seen in figure 2.7.
The negative numbers in table 4.1 indicates that the value was larger for the normal
operating conditions rather than cavitation thereby not being able to differ the two
signals for a speed of 30 rpm. Even this can be seen in the figure plots from the case
study of the positions at the operational pump speed of 30 rpm presented in appendix
A. The frequency domain plots for the configurations capable of detecting the difference
in operating conditions for both the IEPE and the MEMS sensor can be seen in figure
4.7 and their corresponding time domain signal can be seen in figure 4.8.
54
Figure 4.7: Plots presenting the frequency spectra for cavitation and normal operating
state over 20 seconds of recorded samples at 30 rpm of the viable positions for both
MEMS and IEPE sensors.
The reason for the close similarity between cavitation development and normal op-
erating conditions in the frequency domain, is since the vibration intensity from high
frequency noise for these slow speeds are similar to the background noise. This visu-
ally presents the challenge of detecting cavitation during its development stage and the
importance of an optimal sensor configuration since the vibration intensity is so low.
55
Figure 4.8: Plots presenting the raw accelerometer values for cavitation and normal
operating state over 20 seconds of recorded samples at 30 rpm of the viable positions of
both MEMS and IEPE sensors.
The result for the IEPE sensor was in accordance to the positions stated by earlier
research where a top, radial, inlet and axial position on the pump housing was considered
as good positions for the IEPE sensor. The results from the 30 rpm case study showed
that an optimal accelerometer configuration is dependant on both position and by the
accelerometer type. The results of the data can be seen as plots in appendices A and
C where the data from cavitation and normal operating conditions were compared for
the different positions and the two accelerometer types. It is clear that the type and
position are important factors to consider when detecting cavitation phenomenon for a
slow speed application as the signal is only possible to detect with some of the tested
configurations.
56
Table 4.2: The results from the tested positions in the 50 rpm case study.
57
However it can be concluded that the signal strength varies depending on the severity of
the cavitation. Which is dependant on both the operational speed of the pump aswell
as the inlet orifice. The speed and orifice both changes the NPSHa resulting in different
types of cavitation and intensity of vibrations.
The only position for the IEPE sensor which performed closely to the MEMS sensor
at the same position was the horizontal position H1 which had similar performance to
the MEMS sensor at position H2. Which was also the best position for the IEPE sensor
of the tested positions, as can be seen in figure 4.9. However, the configuration of IEPE
sensor at the horizontal position H1 is a good example of the configurations from the
50 rpm case not acting as optimal in slow speed applications. As the IEPE sensor
at the horizontal side position H1 was not capable of detecting the failure mode for
slower speeds, where other configurations still recognized the signal difference. The H1
position was concluded as the worst position in the 30 rpm case study, which shows that
a configuration needs to be evaluated at the slowest limit where the vibration intensity
still recognizes the difference between the operating conditions.
Figure 4.9: A scatter plot of the 50 rpm tests, showing the number of the test in
accordance to its variable value of energy and peak difference for the noise signal.
Equally as for the 30 rpm case it was found that the MEMS sensor generated a
larger difference between the two signals from faulty and normal operating conditions.
A reasoning to why the MEMS sensor resulted in better detection of the cavitation signal
is presented in the next chapter 5.
58
4.3 Optimal accelerometer configuration
The research focuses on an optimal accelerometer configuration in terms of type and
position of the sensor for cavitation detection of the tested pump application. The
research questions were derived from the problem formulation and the preliminary lit-
erature study. Where the main research questions results covering its topic has been
presented.
Hence the best configuration in terms of accelerometer type and position in regard
to reliability evaluated based on energy difference and difference of peak value of the
high frequency noise signal for the two compared signals is presented. The optimal
configuration was found to be the MEMS sensor located at the axial horizontal position
A6 at the outlet of the pump as it showed the largest difference between the compared
signals. Where the reliability of the MEMS sensor at position A6 was found more reliable
than the other configurations since the optimal configuration for a higher speed would
not be reliable for a slow speed application.
A further discussions and conclusions to the research and the research questions are
presented in the following chapter 5.
59
Chapter 5
This chapter discusses the results of the research and forms a conclusion to both the
research questions. A presentation is given to the overall significance of the covered
research and states the most important points of discoveries and also highlights the
central issues of the research.
5.1 Discussion
The study shows that the vibrations at slow speed applications poses challenges of de-
tecting failure modes and thereby it is a challenge to implement a reliable condition
monitoring method for slow speed applications. This proves the necessity of the re-
search covered by this study and increases the significance of the results. However since
slow speed applications of these pumps are less common to research the interest in the
research field is mild.
The significance of the study is confirmed since the industrial recommendations are
to not consider vibration analysis for slow speed applications under 100 rpm, where the
method has proved useful for slower speeds. It has been shown that the damaging failure
mode cavitation can be detected during its development phase for slow speed applica-
tions down to speeds of 30 rpm. Additionally the study shows the importance of choosing
the correct accelerometer type and position since there were only some accelerometer
configurations viable for this application. As the study was only proved effective in a
lab environment on a specific pump with water, the study needs to be verified in a more
realistic scenario with more viscous fluids.
It could be argued that the energy is not a variable free from bias as it includes the
signal strength which is in favour for the MEMS accelerometer in the study, due to the
sensor self resonance. Another way to compare the signals would be using the coherency
function which shows the resemblance of two signals compared to each other, presenting
a value from zero to one. Another option is to simply calculate the difference between
two signals in terms of percentage, showing how much stronger one signal is compared
to the other. This could be done for the energy over the frequency spectrum where the
61
difference shows present. However since the signal strength also is of interest the energy
was deemed a better way to evaluate the performance.
Another argument could be that the peak value does not show the consistency of the
signal or the source of vibration. Where a RMS or average value would be better suited
in order to evaluate the consistency of the vibration. However other options than the
peak value does not exhibit the most severe vibrations where a peak value over a short
time can give a better understanding of the severity of the vibrations from the cavitation
in comparison to a RMS or average value over a time span. A better option would be
to compare both RMS value and peak value for the time domain signal to better cope
with the characteristic vibrations from cavitation.
The difference of peak characteristics and peak location in frequency spectra between
the two accelerometers was most likely due to the different functionality of the sensors
but also since the MEMS accelerometer has a self resonance close to the frequency where
the cavitation noise is present.
62
relation to the performance is discussed in terms of reliability.
The accelerometer configuration has shown to be important in all of the aspects
included in the stated research question. The type of accelerometer in an implementation
clearly results in different opportunities in terms of measurability. Likewise does the
position, where different positions results in varying signal strength. Where the best
position is as close relation to the vibration source as possible. For cavitation in a rotary
bi-winged positive displacement pumps, these positions are varying dependant on the
type of cavitation occurring in the pump. A reason for the delivery side to be the optimal
position for detection of cavitation development for slow speed applications could be due
to the internal back leakage in the pump, due to the pressure differences at the inlet
and outlet of the pump. Thereby detecting the development of cavitation before the
phenomenon is fully developed.
This will likely give the most accurate measurements in order to as early as possible
detect the development of cavitation and estimate its effect to the seal lifetime.
The amount of accelerometer recommended are therefore discussed to be two for a
rotary bi-winged positive displacement pump. The first, in order to early detect the de-
velopment of vortex cavitation at the suction side, to enable additional time of detection
in advance to a failure. The second, placed after the impellers on the delivery side, to be
used for the identification of type of cavtation and the level of force. This configuration
provides the measurements to more accurately estimate the lifetime of a seal with pre-
dictive maintenance. If more failure modes are to be monitored a combination of amount
and position in regard to monitoring as many of the failure modes as possible needs to
be tested, where the optimal accelerometer configuration for cavitation detection might
not even be a viable position for any other failure mode in slow speed applications. This
also allows to include one of each of the two sensor types in the evaluation increasing
the reliability of detecting the distinct character of the noise peak in frequency domain.
63
displacement pump as well.
Even if these initial cavitation conditions are not directly damaging to the pump,
it will lead to increased speeds for the application where it is intended to deliver a
precise amount of liquid over time. This will result in increased loads, increased amount
of cavities formed and the development of stronger cavitation or even conversion of
cavitation type to a more directly damaging one. Thereby, also resulting in a lower
lifetime of not only the seals in the pump but also other internal components, which in
turn relates to lowered machine reliability.
5.2 Conclusion
Regarding the first research question of finding the optimal accelerometer configuration
for the tested use case a conclusion can be stated based on the results from the case
study.
• What is the optimal configuration of accelerometers in terms of type and position
in regard to reliability when detecting cavitation in a slow speed application of a
rotary bi-winged positive displacement pump?[8][30][42]
The optimal configuration in the tested pump application out of the tested configu-
rations was found to be the MEMS accelerometer in the horizontal axial position A6 on
the delivery side, located directly after the impellers on the pump house cover. This was
found to be the optimal solution since the configuration gave the most clear distinction
between normal operating conditions and development of the failure mode cavitation.
This is helpful for condition monitoring since it helps to detect the type of failure mode
and thereby provides the means for a Machine Learning (ML) classifier to be trained on
future collected historical data. Since the measurability proved effective even for slower
speeds, the severity of cavitation to component lifetime can be estimated using the same
historical data. Thereby showing a trend of vibration development where a ML regres-
sion algorithm can help to predict the end of component lifetime. If cavitation can be
detected in an early stage and be either avoided or lowered it can help to increase the
component lifetime and thereby also the reliability of the machine.
The configuration can however not be directly installed in different applications or
scenarios, as it still has to be validated or confirmed to each application. Since the
pumps are constructed in different ways it results in different resonance frequencies for
the propagation of noise, thereby possibly changing the optimal solution to each appli-
cation or version of the pump type.
The second research question has been answered and it has been shown that there
are clear similarities between the vibration noise that occurs from the cavitation for the
two different pumps.
• What similarities of cavitation detection using vibration spectral analysis exist be-
tween the accelerometer data from a gerotor pump and centrifugal pump compared
with a rotary pump at slower speeds?
64
Even though the signals are somewhat different in terms of both amplitude and fre-
quency spectra, vibration analysis and FFT showed to be a good method for detecting
cavitation for slow speed applications for the bi-winged positive displacement pump, in
addition to earlier research done to the gerotor and centrifugal pump. Following is a
presentation of the results from two studies of comparison. The first study used for com-
parison was "Diagnostic Process by Using Vibration Sensors for Monitoring Cavitation
Phenomena in a Gerotor Pump Used for Automotive Applications" [9]. The researchers
of the study invoked cavitation with a 5mm and a 3mm orifice plate similarly as in this
thesis. The pump was operated at different speeds with different inlet pressure were the
slowest speeds of 2000 rpm is used as a comparison to this study. The sensor in this
study was placed "near the cavitation zone" on the pump housing [9]. The results from
operating the pump at 2000 rpm with the 3mm orifice plug can be seen in figure 5.1.
Figure 5.1: Three frequency spectra from first study of comparison, with a 3mm orifice,
referenced as figure 8 in the results chapter in the compared study, collected from [9].
If compared with the results from the 50 rpm study in frequency spectra it, seen
in appendix B it is noticeable that the peaks due to cavitation compared with normal
operating conditions show a clear indicator of the phenomenon. Even though the noise
characteristics differ between the studies it shows that the noise are characteristic to
sensor type, position and pump type. The lower the inlet pressure is and thereof also
65
the bigger the difference between pressure at inlet and outlet the more intensive the
cavitation develops, which can be seen by the energy in the graphs (as the area under the
curve). When comparing low cavitation or conditions close to no cavitation development
it can be seen that the energy differs between the studies, in this thesis the cavitation
develops as a small difference for a smaller spectrum, as can be seen in appendix A. In
the compared study the cavitation noise can be seen in a broader spectrum showing the
importance of use case specific implementation of the sensors are needed, as can be seen
in figure 5.2, where a 5mm orifice was used at a speed of 3000rpm.
Figure 5.2: Frequency spectra from first study of comparison, with a 5mm orifice,
referenced as figure 5 in the results chapter in the compared study, collected from [9].
The second study used for comparison was "Monitoring of Mechanical Seals in Process
Pumps", where the researcher examines multiple failure modes on a centrifugal pump
[8]. The test bench included eletronically controlled valves allowing the pressure at the
inlet to be changed, invoking cavitation in the system. There was numerous tests to
cavitation detection where it is stated that low pump speed does not fully develop the
cavitation phenomenon and that the process is prolonged and takes significantly longer
to fully develop caviataion.
Thereby it is harder to detect cavitation while in its development state. Following
is a presentation to test 33, where the centrifugal pump was operated at a speed of
1500rpm and a negative inlet pressure [8].
66
Figure 5.3: Frequency spectra from second study of comparison, for a speed of 1500
rpm, referenced as figure 18 in the results chapter in the compared study, collected
from [8].
The peak characteristics vary between the thesis and the second study used for
comparison. The energy level shows the noise of cavitation present in high frequency
spectra and a peak at 10kHz can be seen in figure 5.3. This could be mistaken to
be due to self resonance, which is not the case since the sensor used in the compared
study was PCB - M352C67 which has a self resonance around 35kHz. The indicators for
cavitation is present where it shows that a cavitation has not been fully developed since
the over all vibration levels in this case was lower than for the BEP. However due to the
energy present in high frequency spectra it shows that the development of cavitation is
occurring. When comparing the noise for a higher speed of 3600rpm it is noticeable that
the cavitation noise changes with the speed, as can be seen in figure 5.4, similarly as in
this thesis.
Figure 5.4: Frequency spectra from second study of comparison, for a speed of 3600
rpm, referenced as figure 19 in the results chapter in the compared study, collected
from [8].
67
It has been shown that vibration analysis with time domain and frequency domain
using simple signal processing techniques such as digital filtering and FFT proves to give
valuable information for condition monitoring of pumps, even for slower speeds than what
is generally recommended in the condition monitoring industry. The accelerometer posi-
tion and type with its technical specifications of the sensor are important factors which
results in an optimal accelerometer configuration, which is unique to the application and
pump type.
68
Chapter 6
Future recommendations
In this chapter the recommendations for further studies within the area is discussed.
Additionally recommendations are given to the future work that should be implemented
for the project to be usable in the application.
69
ued raw data for 1-12 times per day. Thereby creating the opportunity to recognize the
development of cavitation in the pumps, either over month, day or hour of operation.
This allows for the development of a regression algorithm predicting lifetime of the seal
dependant on the machines operational condition.
The research areas could be broken down to the following fields of research:
The field of condition monitoring of pumps still has areas where new techniques and
methods should be investigated. One of the most interesting areas is sensor fusion for
condition monitoring of pumps. Allowing the combination of different sensor data from
disparate sources, to reduce the uncertainty and result in a more accurate condition
monitoring technique for pumps. This type of research is just briefly touched at this
time and needs to be investigated for different applications and different types of pumps.
The combination of microphones and accelerometers to detect cavitation is an interest-
ing combination which could prove more beneficial.
The field of vibration analysis of pumps has been around for a long time, but has
still not resulted in optimal methods of analysis for different applications. Different or-
ganizations use different directives, different methods and therefore should be further re-
searched in order for find the best methods for the different failure modes. Since different
opinions are stated about the limits of vibration analysis for pumps it shows that the lim-
its of measurability and vibration analysis for pumps has not been settled. Furthermore
the sensitivity of the accelerometer has a clear connection to the performance and mea-
surability thereby posing possibility for better performance of measurablity with higher
sensitivity. It is also of interest to compare an IEPE and a MEMS accelerometer with
the same resonance frequency as it was found to be a contributing factor for cavitation
detection. Many recent research articles state high frequency bandwidth accelerometers
to be the best suited sensor for cavitation detection since the noise of the implosions are
70
visible in high frequency domain. However since low bandwidth accelerometers intended
for low speed applications often have a self resonance within the domain of cavitation
noise, these accelerometers might result in better performance for cavitation detection
as the self resonance might act as a signal amplifier. Additionally these low bandwidth
sensors often provide higher sensitivity, resulting in an area of interest for future research.
Signal processing of vibration data is another field in itself, posing many challenges
to future research relating to conditions of pumps. Studies have been done with dif-
ferent signal processing techniques. However these methods still needs to be evaluated
for different application areas, where the slow speed applications poses challenges to
measurability. FFT and vibration spectral analysis has been a standard for a long time,
but with newer more advance signal processing methods it might not be the optimal
solution for condition monitoring. For example studies have been done using discrete
wavelet transforms in order to monitor the condition in pumps. Other areas within sig-
nal processing relevant to condition monitoring are optimal filtering techniques, signal
boosting techniques, windowing, smoothing functions and threshold value calculations.
These are all relevant for condition monitoring of slow speed application pumps.
The last area somewhat involved in this thesis which is suggested for further re-
search is predictive maintenance and industrial optimization. There are many available
options for predictive learners in terms of machine learning, deep learning or statistical
approaches where different algorithms prove better for different use cases. Therefore it
could be of high interest to research the best predictive solution based on big data for
cavitation detection.
A study within any of these areas could be done in connection to one or multiple of
the other fields to connect the subjects more. This is heavily recommended since it is
the use case of condition monitoring or predictive maintenance which is the goal of the
vibration analysis or signal processing of the data from the process pumps.
71
Bibliography
[1] Alhussein Albarbar et al. “Suitability of MEMS Accelerometers for Condition Mon-
itoring: An experimental study”. In: Sensors 8.2 (2008), pp. 784-799. issn: 1424-
8220. doi: 10.3390/s8020784.
[2] Juan Pablo Amezquita-Sanchez and Hojjat Adeli. “Signal Processing Techniques
for Vibration-Based Health Monitoring of Smart Structures”. In: Arch Computat
Methods Eng 23 (2016), pp. 1-15. doi: 10.1007/s11831-014-9135-7.
[3] Vladimir V. Echeistov Andrey I. Vlasov and more. “An Information System of
predictive maintenance and analytical support of industrial equipment”. In: IIPP
Journal of Applied Engineering Science 16 (2018), pp. 515-522. doi: 10 . 5937 /
jaes16-18405.
[4] Mifta Nur Farid Anisatul Fauziyah Try Rahadi Sulistomo and Dhany Arifianto.
“Pump Condition Monitoring Through Vibration Analysis.” In: IOP Publishing
Ltd (2018). doi: 10.1088/1742-6596/1075/1/012011.
[5] M.Salmina A.Vlasov A.Yudin and more. “Design Methods of Teaching the Devel-
opment of Internet of Things Components with Considering Predictive Mainte-
nance on the Basis of Mechatronic Devices”. In: International Journal of Applied
Engineering Research 12 (2017), pp. 9390-9396. issn: 0973-4562.
[6] Michiel A. Beijen et al. “Experimental estimation of transmissibility matrices for
industrial multi-axis vibration isolation systems”. In: Mechanical Systems and Sig-
nal Processing 107 (2018), pp. 469-483. issn: 0888-3270. doi: 10.1016/j.ymssp.
2018.01.013.
[7] Ulf Carlsson. CONDITION MONITORING. MWL, Aeronautical and Vehcle En-
gineering, KTH,
[8] Siddharth M. Chittora. “Monitoring of mechanical seals in process pumps”. In:
KTH Industrial Engineering and Management (2018). doi: TRITA-ITM-EX2018:
686.
[9] Emma Frosina Daniela Siano and Adolfo Senatore. “Diagnostic Process by Using
Vibrational Sensors for Monitoring Cavitation Phenomena in a Getoror Pump
Used for Automotive Applications”. In: Energy Procedia 126 (2017), pp. 1115-
1122. doi: 10.1016/j.egypro.2017.08.269.
72
[10] Xavier Escaler et al. “VIBRATION CAVITATION DETECTION USING ON-
BOARD MEASUREMENTS”. In: Fifth International Symposium on Cavitation 5
(2003).
[11] “FACT SHEET - TYPES OF SLUDGE PUMPS”. In: Australian Pork (), [Ac-
cessed Online: 2019-05-10]. url: https://australianpork.com.au/wp-content/
uploads/2013/09/FACT-SHEET-Types-of-Sludge-Pumps.pdf.
[12] Matteo Frigo and Steven G. Johnson. “The Design and Implementation of FFTW3”.
In: Proceedings of the IEEE 93.2 (2005). Special issue on “Program Generation,
Optimization, and Platform Adaptation”, pp. 216-231.
[13] Deepam Goyal and BS Pabla. “Condition based maintenance of machine tools A
review”. In: CIRP Journal of Manufacturing Science and Technology 10 (2015).
issn: 1755-5817. doi: 10.1016/j.cirpj.2015.05.004.
[14] “Guidelines for Mounting Test Accelerometers”. In: PCB Piezotronics, Inc. (), [Ac-
cessed Online: 2019-05-10]. url: https://www.pcb.com/resources/technical-
information/mounting.
[15] Amin Al-Habaibeh, Guoping Liu, and Nabil Gindy. “SENSOR FUSION FOR
AN INTEGRATED PROCESS AND MACHINE CONDITION MONITORING
SYSTEM”. In: IFAC Proceedings Volumes 35 (2002), pp. 25-30. issn: 1474-6670.
doi: 10.3182/20020721-6-ES-1901.01557.
[16] S R. Haqshenas and N Saffari. “Multi-resolution analysis of passive cavitation
detector signals”. In: Journal of Physics: Conference Series 581 (2015). doi: 10.
1088/1742-6596/581/1/012004.
[17] H. M. Hashemian and Wendell C. Bean. “State-of-the-Art Predictive Maintenance
Techniques*”. In: IEEE TRANSACTIONS ON INSTRUMENTATION AND MEA-
SUREMENT 60 (2011), pp. 226-236. doi: 10.1109/TIM.2009.2036347.
[18] Salem Al-Hashmi, Mohamed Alhsnawi, and A Ball. “Processing Vibration signals
for cavitation detection”. In: Journal of Sebha University-(Pure and Applied Sci-
ences) 6 (Oct. 2007), pp. 50-63.
[19] Implementation Strategies and Tools for Condition Based Maintenance at Nuclear
Power Plants. TECDOC Series 1551. Vienna: INTERNATIONAL ATOMIC EN-
ERGY AGENCY, 2007. isbn: 92-0-103907-7.
[20] National Instruments. “Machine Condition Monitoring Technical Library”. In:
(July 2017), [Accessed Online: 2019-04-23]. url: http://www.ni.com/example/
6511/en/#toc2.
[21] J. d. J. Rangel-Magdaleno et al. “FPGA-Based Vibration Analyzer for Continuous
CNC Machinery Monitoring With Fused FFT-DWT Signal Processing”. In: IEEE
Transactions on Instrumentation and Measurement 59.12 (Dec. 2010), pp. 3184-
3194. issn: 0018-9456. doi: 10.1109/TIM.2010.2047130.
73
[22] Hung-An Kao Jay Lee and Shanhu Yang. “Service innovation and smart analytics
for industry 4.0 and big data environment.” In: Procedia CIRP 16 (2014), pp. 3-8.
doi: 10.1016/j.procir.2014.02.001.
[23] A. Kotb and A. Abdulaziz. “Cavitation Detection in Variable Speed Pump by
Analyzing the Acoustic and Vibration Spectrums.” In: Engineering 07 (2015), pp.
706-716. doi: 10.4236/eng.2015.710062..
[24] IlyasMazhar KristofferMcKee GarethForbes and RodneyEntwistle. “A review of
major centrifugal pump failure modes with application to the water supply and
sewerage industries”. In: Curtin Unicersity - ICOMS Asset Management Confer-
ence Proceedings (2011). issn: 1329-7198.
[25] Pietro Larizza. “Chapter 29 Measurement, Testing, and Diagnosis for Micro man-
ufacturing Systems”. In: Micro and Nano Technologies (2015), pp. 675-704. doi:
10.1016/B978-0-323-31149-6.00029-3.
[26] SYED TAFAZZUL MAHMOOD. “USE OF VIBRATIONS ANALYSIS TECH-
NIQUE IN CONDITION BASED MAINTENANCE”. In: KTH Diva (2011). doi:
urn:nbn:se:kth:diva-41319.
[27] Jason Mais. “All you need to know... Alfa Laval Pump Handbook”. In: Gillain Co
NV 2 (2002), [Accessed Online: 2019-04-24]. issn: PM66050GB2 2002.
[28] Jason Mais. “Spectrum Analysis - The key features of analyzing spectra”. In: SKF
(2002), [Accessed Online: 2019-04-24]. issn: CM5118 EN. url: www.skf.com/cm.
[29] Emiliano Mucchi Mattia Battarra. “Incipient cavitation detection in external gear
pumps by means of vibro-acoustic measurements”. In: Measurement 129 (2018),
pp. 51-61. doi: 10.1016/j.measurement.2018.07.013.
[30] P.E Michael Volk. Pump Characteristics and Applications, Second Edition. Taylor
and Francis Group, 2014. isbn: 978-1-4200-2835-5.
[31] R. Keith Mobley. “An Introduction to Predictive Maintenance”. In: Elsevier -
Plant Engineering (Sept. 2002). issn: 9780080478692.
[32] Alisa Filofteia Oprea, Nicolae Popa, and Constantin Onescu. “Considerations Re-
garding the Pumps Vibration Monitoring Using Proactive and Predictive Main-
tenance Programs”. In: Applied Mechanics and Materials 630 (Nov. 2014), pp.
383-387. doi: 10.4028/www.scientific.net/AMM.630.383.
[33] James A. Venning Paul A. Brandner and Bryce W.Pearce. “Wavelet analysis tech-
niques in cavitating flows”. In: The Royal Society Publishing (2018). doi: 10.1098/
rsta.2017.0242.
[34] Jack Peters. “Beginning Vibration Analysis with Basic Fundamentals”. In: Con-
nection Technology Center, Inc. (), [Accessed Online: 2019-05-10]. url: https://
www.hollandindustrial.com/wp-content/uploads/sites/197/2014/12/01-
Beginning-Vibration-Analysis.pdf.
74
[35] M.A.Ferrer P.Henriquez JB.Alonso and G.Gomez. “Fault Diagnosis using Audio
and Vibration Signals in a Circulating Pump”. In: Journal of Physics: Conference
Series 364 (2012). doi: 10.1088/1742-6596/364/1/012135.
[36] A. Kaul P.K. Gupta P.A. Kumar and more. “Neural Network Based Methodol-
ogy for Cavitation Detection in Pressure Dropping Devices of PFBR”. In: Indian
Society for Non-Destructive Testing Hyderabad Chapter (2006).
[37] Mahesh Pophaley and Ram Krishna Vyas. “Plant maintenance management prac-
tices in automobile industries: a retrospective and literature review”. In: Journal
of Industrial Engineering and Management 3 (Dec. 2010). doi: 10.3926/jiem.
.v3n3.p512-541.
[38] Global Pumps. What Are the Differences Between Pump Types?, [Accesed Online:2019-
03-28]. url: https : / / www . globalpumps . com . au / blog / what - are - the -
differences-between-pump-types.
[39] “ROTARY OR POSITIVE DISPLACEMENT PUMPS”. In: The Mc Nally Insti-
tute CD (), [Accessed Online: 2019-04-24]. url: http://www.mcnallyinstitute.
com/CDweb/cdpage.htm.
[40] F.Moerchen R.Sipos D.Fradkin and Z.Wang. “Log Based predictive maintenance”.
In: KDD ’14 Knowledge discovery and data mining (2014), pp. 1867-1876. issn:
1743-3533. doi: doi:10.1145/2623330.2623340..
[41] Nazmus Sakib and Thorsten Wuest. “Challenges and Opportunities of Condition-
based Predictive Maintenace: A Review”. In: Elsevier B.V. (2018). doi: 10.1016/
j.procir.2018-08.318.
[42] Cornelius Scheffer. “Pump Condition Monitoring Through Vibration Analysis”.
In: PUMPS: Maintenance, Design, and Reliability Conference 2008 – IDC Tech-
nologies 5 (2008). issn: 1755-5817.
[43] Hansford Sensors. “How to mount a vibration sensor”. In: (2019), [Accessed Online:
2019-04-23]. url: https : / / www . hansfordsensors . com / how - to - mount - a -
vibration-sensor/.
[44] D. Siano and M.A. Panza. “Diagnostic method by using vibration analysis for
pump fault detection”. In: Energy Procedia 148 (2018). ATI 2018 - 73rd Conference
of the Italian Thermal Machines Engineering Association, pp. 10-17. doi: 10.1016/
j.egypro.2018.08.013.
[45] Christopher Stephen and S. Kumaraswamy. “Experimental Determination of Cav-
itation Characteristics of Low Specific Speed Pump using Noise and Vibration”.
In: Journal of The Institution of Engineers (India): Series C 100.1 (Feb. 2019),
pp. 65-74. issn: 2250-0553. doi: 10.1007/s40032-017-0431-5.
[46] “The AL Series Rotary Lobe Pump is the Low Cost, Big Technology Leader”. In:
Ampco Pumps Company (), [Accessed Online: 2019-05-10]. url: https://www.
ampcopumps.com/al-series-pumps/.
75
[47] “VFD Variable Frequency Drive and Centrifugal Pump”. In: Automation Controls
• OEM Panels (), [Accessed Online: 2019-05-10]. url: http://www.oempanels.
com/vfd-variable-frequency-drive-and-centrifugal-pump.
[48] K. Wang. “Intelligent Predictive Maintenance (IPdM) system, Industry 4.0 sce-
nario”. In: WIT Transactions on Engineering Sciences 113 (2016). issn: 1743-3533.
doi: 10.2495/IWAMA150301.
[49] Xin Xue, V Sundararajan, and Luis Gonzalez-Argueta. “Sensor fusion for machine
condition monitoring”. In: Proceedings of SPIE - The International Society for
Optical Engineering 6932 (Mar. 2008). doi: 10.1117/12.776559.
[50] L.Zhong G.Bo Y.Minguan Z.Ning. “Vibration Characteristics Induced by Cavita-
tion in a Centrifugal Pump with Slope Volute”. In: (2015). doi: 10.1155/2015/
294980.
76
77
Appendix A
78
79
Appendix B
80
81
Appendix C
82
83
Appendix D
84
TRITA ITM-EX 2019:435
www.kth.se