Turbomachinery Monitoring and Diagnostics

Aly El-Shafei[0000-0002-3124-2120]
1 Cairo University, Giza 12316 Egypt

and RITEC, Maadi 11431, Cairo, Egypt

[email protected]

Abstract. Turbomachinery condition monitoring and diagnostics has come a

long way in the last few decades. The main tool for monitoring and diagnosis
of turbomachinery is the use of vibration measurement and analysis. Along with
machine operating parameters, machine vibration analysis provides a complete
picture of turbomachinery condition and can be used to perform a complete
analysis and a diagnostic result can be subsequently obtained. Usually diagnos-
tics is performed after an anomaly has been detected through monitoring. This
is the approach taken by ISO 13373 series of diagnostic standards. This chapter
discusses methods of vibration measurement for machinery monitoring and di-
agnostics. The sensors used, their operating principles and installation methods
are described, as well as the required instrumentation and measurement set-up.
The main tool for vibration diagnosis is the spectrum, however, for a complete
and accurate diagnosis other vibration tools may have to be considered, includ-
ing the overall amplitude, time waveform, the phase, the orbit and in some cas-
es an operating deflection shape analysis (ODS) may need to be performed. Fi-
nally recent trends shall be briefly discussed, including the use of wireless sen-
sors, cloud monitoring, expert diagnosis, the use of artificial intelligence in di-
agnosis, as well as the very recent use of vibration motion magnification tech-
nology.

Keywords: Turbomachinery, Condition Monitoring, Machinery Diagnostics,

Operating Parameters, Vibration Measurement, Signal Processing, Diagnostic
Process, Remote Monitoring and Diagnosis.

1 Introduction

This chapter summarizes the state-of-the-art in turbomachinery condition monitoring

and diagnostics. The condition monitoring market has expanded tremendously in the
last few years [1], and many technologies have been promoted for machinery condi-
tion monitoring including infra-red thermography, oil analysis, motor current analysis
and ultrasound. However, vibration condition monitoring is by far the leader in this
market, with more than 65% market share [1]. In fact, this author has spent more than
30 years in this field and has monitored (along with his teams) tens of thousands of
machines and it is the opinion of this author that it is sufficient to monitor vibration
and operating parameters to fully diagnose any rotating machine and many reciprocat-
ing machines. Thus, this chapter is dedicated almost exclusively to the vibration con-
2

dition monitoring and diagnosis of machines. The other condition monitoring tech-
nologies are useful and have specific usage, for example infra-red thermography is
very useful for monitoring of electric panels or furnaces, ultrasound is extremely use-
ful for diagnosis of electric discharge in transformers, while oil analysis is generally
useful in confirming machine condition and the state of the lubricant. Motor current
signature analysis is specific for motors, although motor diagnostics can be done by
vibration analysis through experienced users.
Over the years the vibration measurement and analysis tools have evolved. In
1993 [2], the author summarized the then state-of-the-art of vibration measurement
for machinery monitoring and diagnosis. The basic machinery vibration analysis
tools are described in detail by Eshleman [3] and Peters and Eshleman [4]. A coher-
ent methodology for machinery vibration monitoring is described by El-Shafei and
Rieger [5], while a whole series for vibration machinery diagnostic standards, ISO
13373, were developed by experts for the International Organization for Standardiza-
tion (ISO) [6-12]. The ISO 13373 series is based on the author’s work [13]-[14] in
developing a flow-chart approach for machinery diagnostics. An explanation of the
use of ISO 13373 standards is given in [15] and [16]. Examples of Diagnosis of spe-
cific machines is given in [17] and [18]. It should be noted that the diagnosis of gear-
boxes is actually quite elaborate and may require more advanced analysis tools [18].
This chapter starts by reviewing the vibration measurement tools and the instru-
mentation used in machinery monitoring and diagnostics, followed by a review of the
vibration analysis tools used for machinery diagnostics. This is followed by the de-
scription of the principles of machinery condition monitoring. Next, the process for
machinery diagnostics is described with some examples. Finally, the chapter ends
with recent trends for machinery vibration monitoring and diagnostics.

2 Vibration Measurement and Instrumentation

In order to use vibration as a monitoring and diagnosing tool, vibration needs to be

measured successfully. Vibration needs to be sensed (i.e. measured and transformed
into an electric signal) then processed and converted to a digital signal for analysis.
All of these steps are important and should be done correctly. If any of these steps
fail, then the whole measurement process is jeopardized. It is extremely important to
measure correctly. Guidance on the vibration measurement process for condition
monitoring and machinery diagnosis is given in [2-4] and [7].

2.1 Vibration Sensors

There are many types of vibration sensors. The most common sensor for vibration
analysis is an accelerometer, which, as its name implies, measures absolute (seismic)
acceleration, most commonly through a piezoelectric crystal that converts accelera-
tion into electric charge [2].
Accelerometer mounting is very important. There are many ways for mounting an
accelerometer (depending on the application), but the most common mounting meth-
ods are by a stud, mounting pad or magnet [3]. The mounting method affects heavily
 3

the useful frequency range of an accelerometer, which can be reduced from 6 kHz for
stud-mounting to 2 kHz for magnet mounting, for a typical accelerometer [3]. This is
due to the need to avoid the mounting resonance and to use the accelerometer in the
region of constant sensitivity [2]. However, if the accelerometer is used for envelope
analysis [4], then usually the acceleration data are processed at the mounted resonance
to use the resonance magnification of the measured signal for further processing [4].
General machines are usually monitored in units of velocity. In this case, the ac-
celeration signal sensed by the accelerometer is integrated to velocity (usually elec-
tronically). However, a high-pass filter at a very low corner frequency (around 0.5
Hz) is used before integration to avoid the inevitable magnification near zero frequen-
cy. This is the most common method for measuring vibration velocity. Laser vibra-
tion velocity sensors are used in the market, but are expensive, thus are of limited use
in monitoring. However, they are used in particular diagnosis measurements, espe-
cially for torsional vibration.
As a general practice, acceleration is used to monitor machines in the frequency
range above 1 kHz, and velocity is used in the range from 20 Hz up to 1 kHz, while
vibration displacement is used below 20 Hz [3].
To measure displacement, an accelerometer signal can be double integrated, with
the same signal processing described above. This is for absolute displacement. For
relative displacement a Linear Variable Differential Transformer (LVDT) [2] can be
used to measure vibration (example on turbine casing expansion). However, the
LVDT is a contact measurement and is a relative measurement. The LVDT consists
of two energized coils acting as a transformer. When the intermediate rod is centered,
the differential voltage from the transformer coils is zero. Upon rod displacement, the
differential voltage measured is proportional to the rod displacement. The LVDT has
a maximum frequency of 400 Hz [2].
For a contactless relative measurement, a proximity probe is used. This is most
commonly used on rotors mounted on fluid film bearings. The proximity probe is
essentially a coil of wire inserted in a polymer probe and installed in proximity of the
rotating shaft. The coil in the probe is energized by an oscillator supplied with a car-
rier frequency in the radio frequency range, and when the rotating shaft (made of
magnetic conducive material) approaches the proximity probe, the resulting magnetic
field produces eddy currents on the surface of the rotating shaft, which represent loss-
es in the magnetic field. As the shaft moves closer to the probe, the eddy current
losses increase. Therefore, the proximity probe actually measures the gap between
the probe and the shaft [2]. The measured signal is then processed through a demodu-
lator that removes the radio frequency carrier signal, leaving the measured vibration
displacement.
Other applications may require the use of strain gages for measurement of vibra-
tion displacement. This is particularly important in the diagnosis of machines, as the
measurement is done at the source (not transmitted through machine casing). Strain
gage measurement has the advantage of being able to measure both lateral and tor-
sional vibration. For measurements on a rotating shaft, the measured signal is trans-
mitted through telemetry or slip rings [2].
4

2.2 Instrumentation
Instrumentation are used to process the signal sensed by the vibration transducer. The
most important aspect in the measurement loop is to consider impedance matching of
all instruments in the loop [2], from the sensor to the measurement and display in-
strument.
Many intermediate instruments are used: amplifiers are used to improve the signal
to noise ratio, integrators are used to transform the signal from one domain to the
other, while filters are used to select certain frequency ranges. As discussed, a high
pass filter is used before integration, while a low pass filter with a high corner fre-
quency is used before sampling as an anti-aliasing filter. Band pass and notch filters
are used in particular applications. Tracking filters use band pass filters to measure
transient response of rotating machinery during start-up and shut-down.
The final instrument in a measurement loop is essentially a voltmeter capable of
displaying the signal in the time domain (oscilloscope) or frequency domain (spec-
trum analyzer) [3].
Virtual instruments and virtual instrument loops can be created using specialized
software. However, when virtual instruments are used, care should be taken in the
analog signal processing before sampling to ensure that problems such as aliasing are
avoided [4].

2.3 Digital Data Processing

In our digital world, there are many aspects of digital signal processing that has to be
considered for vibration condition monitoring and diagnostics. The most important
issue is that Shannon’s sampling theorem has to be respected. This theorem states
that the sampling rate has to be twice as fast as the highest frequency of the signal.
This is to avoid frequency folding [3]. Since we do not know the vibration signal
beforehand (that is why we are measuring it), but we know the frequency range that
we are interested in, which is Fmax (from the machine knowledge) [3], we can have a
low pass anti-aliasing filter applied to the vibration signal at this maximum frequency
to eliminate any unwanted noise above F max. Actually, because of the characteristics
of the anti-aliasing filter, the sampling rate is 2.56 Fmax. This translates in the number
of samples actually is also 2.56 times the number of lines in a spectrum [3].
Two very important equations, which are quite useful in setting-up a vibration
condition monitoring and diagnosis system are [3]:

Ts = # of Lines / Fmax (1)

and
Res = 2 WF Fmax/ # of Lines (2)

where Fmax is the frequency range, # of Lines is the number of lines in the spectrum,
Ts is the measurement time, Res is the resolution, and WF is the window factor. The
resolution is the ability to distinguish between two frequencies in a spectrum [3] and
is very important in the diagnosis of electric motors [4], [7], [12]. The window factor
 5

is used when a window is applied to the vibration measurement. The purpose of the
window is to force the beginning and end of the measurement to zero, thus ensuring
periodicity and reducing leakage in a spectrum [2-3]. Leakage is a spectral phenome-
non where spectral energy leaks from one spectral line to neighboring lines because of
lack of periodicity in the vibration signal [2-3]. The most common windows are uni-
form (with a WF of 1.0), Hanning (with a WF of 1.5) and Flat Top (with a WF of
3.8). In almost all vibration condition monitoring and diagnosis applications, a Han-
ning window should be used. However, for certain applications, other windows may
be used. For example, a uniform window is used for bump tests, and flat-top window
is used for accurate and specific amplitude measurement [4].
Due to noise, a single measured spectrum is usually not indicative of the actual
machine vibration. Generally multiple spectra are measured at the same point and are
averaged out to reduce noise. 8 averages are recommended, but this may significantly
increase the measurement time. The minimum acceptable number of averages is 4
averages. For low Fmax or for a high # of lines, the measurement time according to
equation (1) would be high. In this case, overlap averaging can be used [4], where
part of the previous data block is used in the subsequent data block.
Specialized averaging procedures are sometimes used. Peak-hold averaging is
sometimes used in lieu of a tracking filter to track the transient behavior of a machine.
Synchronous time averaging is used to average the time waveform, triggered by a
certain event, to accentuate this particular event [4].
In summary, to set-up a vibration condition monitoring and measurement system, it
is important to do the following:
 The sensor frequency range is suitable for the application. The sensor sensitivi-
ty should be considered in set-up.
 The Fmax should be selected based on the expected frequencies generated by the
machine [2].
 The # of lines should be selected to ensure adequate resolution and suitable
measurement time.
 Hanning window should almost always be selected.
 Averaging should be selected to reduce the noise effect. Overlap processing
may be used to reduce measurement time.
Finally, if a compromise set-up cannot be achieved, then two measurements should
be considered at the same point with two different measurement set-ups to ensure
accurate, useful and meaningful measurement.

3 Tools for Vibration Monitoring and Diagnosis

There are many tools for vibration monitoring and diagnosis. The most important of
which are the time waveform and the spectrum.

3.1 Vibration Time Waveform

The vibration time waveform is a plot of vibration data versus time. This is the most
basic vibration data. The peak overall value is obtained from the time waveform. It
6

is actually a very useful tool for visualizing the machine behavior. Directional forces,
truncation and pulses are clearly seen in the time waveform [4]. Some effects such as
beating and modulation need to be distinguished in a spectrum. Figure 1 shows the
time waveform showing directional forces due to misalignment [18].

Figure 1 Time waveform showing a directional force due to misalignment [18]

3.2 Vibration Spectrum

The vibration spectrum is the FFT of the vibration data shown in the time waveform.
Even though it contains the same data, yet the spectrum display is a very powerful
tool for vibration diagnosis, as it illustrates the frequency content of the signal. The
role of diagnosis is to relate each significant frequency line in the spectrum to a par-
ticular machine fault [4]. Figure 2 shows a spectrum on a cement mill showing beat-
ing between two frequencies [18].

Figure 2 Spectrum showing beating between two frequencies [18]

 7

3.3 Vibration Waterfall

Vibration waterfall plot is simply a three dimensional plot of spectra as a certain
operating parameter is changed, usually speed. Figure 3 shows a waterfall plot show-
ing progressive looseness with time [4]. For rotating machinery, the waterfall plots
may be in the form of a full spectrum, where positive and negative frequencies are
plotted, indicating forward and backward whirls.

Figure 3 Waterfall plot showing progressive looseness with time [4]

3.4 Envelope Spectrum

The envelope spectrum is a special measurement to detect impacts in a signal by ac-

centuating the signal envelope [4]. Usually envelope measurements are used to detect
early rolling element bearing faults. Figure 4 shows an envelope spectrum for a high-
ly loaded gearbox [18].

Figure 4 Envelope spectrum of a highly loaded gearbox [18]

8

3.5 Phase
Phase is defined as the time difference between two signals, at the same frequency,
proportioned to their period (which is a complete cycle of 360 degrees). Absolute
phase of a signal is defined relative to a reference point. For rotating machinery, the
reference point is usually a key way on the shaft, or a reflective tape installed on the
shaft, sometimes termed keyphasor. To measure absolute phase a proximity is used to
monitor the keyway, or a laser probe is used to monitor the reflective tape. A pulse is
generated by the proximity probe whenever the keyway passes in front of it, and simi-
larly a pulse is generated when the reflective tape passes in front of the laser probe.
This pulse is then the reference signal for phase at the running speed. Phase is then
defined by the angle between the generated pulse and the first peak in the measured
vibration signal [4].
Phase is very important in machinery diagnostics. Many faults show vibration at
the synchronous frequency, i.e. a frequency equal to the running speed, usually
termed 1x frequency. This renders the spectrum, which is the main diagnostic tool,
useless. In order to distinguish between these faults, phase is then used [13]. For
example, to differentiate between unbalance and misalignment, phase is measured
across the coupling. For rigid rotors, if the vibration signals across the coupling are
in-phase, then the problem is unbalance, if they are out-of-phase, then the problem is
misalignment. Many more examples are given in the ISO 13373 series [6-12].

3.6 Orbit

The orbit is a plot of two vibration signals from two proximity probes 90 degrees
apart monitoring a shaft (usually mounted on fluid film bearings). Time is eliminated,
and the two signals are plotted against each other, thus providing the actual trajectory
of the rotor in the bearing. The orbit is a very useful tool for visualizing the shaft
vibration. A typical unfiltered orbit for a rotor in a fluid film bearing is shown below
in Figure 5. A filtered orbit at 1x would show as an ellipse.

Figure 5 Typical unfiltered orbit for a rotating machine (µm).

 9

3.7 Shaft-Centerline Plot

The Shaft Centerline Plot (SCP) is also obtained from two proximity probes 90 de-
grees apart monitoring a shaft on fluid film bearings. However, for an SCP plot, the
gap voltage (i.e. the static component) is plotted as the machine speed is changed.
This is in contrast to the orbit plot where the vibration signal (i.e. the dynamic com-
ponent) is plotted. It may be considered that the SCP is the locus of the orbit center
under speed changes.
The SCP is a very important tool in the evaluation of fluid film bearing condition.
Usually the shaft is sitting at the bottom of the bearing at start-up. As the speed in-
creases, the shaft centerline lifts and follows a certain trajectory. For heavily loaded
bearings the shaft centerline remains near the bottom of the bearing, which can lead to
excessive wear or rubbing under certain conditions. If however the bearing is lightly
loaded, and the shaft centerline approaches the center of the bearing, then this ma-
chine would be prone to instability. Figure 6 shows a typical SCP for a rotating ma-
chine during start-up.

Figure 6 Shaft Centerline Plot for a rotating machine on fluid film bearings

3.8 Frequency Response Function

The Frequency Response Function (FRF) is a system characteristic function that rep-
resents the system behavior in the frequency domain. It is the Fourier Transform of
the Impulse Response Function (IRF), and is closely related to the Transfer Function
obtained by the Laplace Transform of the IRF, which is used frequently in control
system analysis.
All the tools used in this section from 3.1 to 3.7 are signal analysis tools, meaning
that a vibration signal is measured and is analyzed. However, the FRF is a system
analysis tool. It does not describe a vibration signal, but it actually represents the
system behavior, in the frequency domain, to particular excitation, often in a con-
trolled test. It is possible from a measured FRF to identify the natural frequencies and
10

their associated damping ratios, and even the mode shapes (through further analysis)
in a certain measured frequency range. Figure 7 shows the magnitude plot of a typical
FRF. Peaks in the magnitude FRF represent natural frequencies.

Figure 7 Typical magnitude FRF for a Girth Gear [18]

This might be straightforward for structures, however for rotating machines with
speed dependent characteristics, this requires special attention and deeper understand-
ing. Consider a rotating machine with speed dependent characteristics, as shown in
Figure 8 [19]. The natural frequencies change with speed. Consider that the machine
is operating at a certain speed. If the machine is excited (for example by a magnetic

Figure 8 FRFs for a rotating machine with speed dependent characteristics [19]
 11

bearing), then the FRF measured would show two resonant peaks at the two natural
frequencies shown on the blue FRF. If however, the machine is operating at a differ-
ent speed, and the same test is repeated, then we would get the green FRF shown in
Figure 8. Notice that the peaks of the green FRF (and thus the natural frequencies)
are different from those of the blue FRF. Similarly the red FRF shown in Figure 8, at
a higher speed, shows different natural frequencies, clearly illustrating the speed de-
pendent characteristics of rotating machinery. Notice that all of these are natural
frequencies, but excited at different speeds.
If however, the machine is just excited by the unbalance, and the measured signal
is filtered at the running speed, where the measured vibration is synchronous with the
running speed (i.e. 1x), then we get a synchronous FRF, which is the black FRF in
Figure 8. Note that the synchronous FRF is just the well-known Bode plot used fre-
quently for rotating machinery. Notice that the peaks of the synchronous FRF (black
FRF in Figure 8) consists of one peak from the green FRF and one peak from the red
FRF, illustrating the change in characteristics and natural frequencies with speed. The
peaks in a synchronous FRF are called critical speeds, and are very important for the
analysis and diagnosis of rotating machinery.

Bode Plot. The FRF is a complex number, which can be presented as magnitude and
phase or real and imaginary. For rotating machinery, the Bode plot is a plot of the
magnitude of the synchronous FRF versus speed and phase of the synchronous FRF
also versus speed. Figure 9 shows the magnitude and phase plots of a Bode plot at the
horizontal and vertical directions of the drive end bearing of a rotor. The upper Fig-
ures 9 (a) and 9 (b) represent the magnitude Bode plot, while the Figures 9 (c) and 9
(d) represent the phase Bode plot. Notice that balancing does not affect the phase
Bode plot.

Figure 9 Bode plot of a rotor before balancing and after balancing

Polar Plot. The Polar plot is another representation of an FRF, where the imaginary
component of an FRF is plotted versus the real component, and speed (or frequency
12

depending on the FRF type) is implicit. Figure 10 shows a Polar plot of a synchro-
nous FRF during a turbine shutdown. Each circle represents a resonant condition.
The critical speeds can be identified by a 90 degree phase shift in each circle.

Figure 10 Polar plot of a synchronous FRF during a turbine shutdown.

Real and Imaginary Plots. The real and imaginary components of an FRF can be
plotted versus frequency (speed in case of synchronous FRFs). It may be advanta-
geous to use this particular representation for proper identification of the system char-
acteristics. Table 1 shows the advantageous use of real and imaginary FRFs. For
example, for a mobility FRF, the real plot of the FRF can be used to obtain the mode
shape, while the imaginary plot can be used to obtain more accurate estimates of
damping. Figure 11 shows an example of the real FRF plot for a mobility measure-
ment on a pump rotor, which was later used to confirm the mode shape.

Table 1. Real and Imaginary FRFs

FRF MEASURES REAL IMAGINARY
Dynamic Compliance Displacement/Force Damping Mode

Mobility Velocity/Force Mode Damping

Accelerance Acceleration/Force Damping Mode
 13

Figure 11 Real plot of a mobility FRF [17]

3.9 Operating Deflection Shape (ODS)

The Operating Deflection Shape (ODS) analysis is a powerful tool for visualization of
machinery and structural vibration. Often used with animation, ODS can be a power-
ful tool for visualizing machine vibration. To measure ODS, the machine or structure
is divided into grid points and measurement of amplitude and phase is then conducted
and plotted, which can be a time consuming process. The result is usually an impres-
sive figure clearly identifying the root cause of the problem. Figure 12 shows the 6x
ODS of a cooling fan base with excessive horizontal deflection at the blade passing
frequency causing shaft failures [14], while Figure 13 shows 3D 1x ODS plots of
steam turbine bearings, where the high pressure rotor was experiencing excessive
directional steam induced vibration. The axes represent the actual physical dimen-
sions of the structure. In both figures, the deformed structure is compared to the un-
deformed skeleton to clearly visualize the deformation at that particular frequency.
ODS is increasingly being replaced by vibration video motion magnification. How-
ever, ODS still has the advantage of being a 3D measurement, while vibration video
motion magnification is not.

Figure 12 ODS of cooling fan base [14]

14

Figure 13 3D ODS of steam turbine bearings

4 Vibration Condition Monitoring

The purpose of vibration condition monitoring is to be able to determine the condition

of a machine and to be able to use this information to make a decision on the func-
tionality of the machine and to plan for any required maintenance of the machine.
This is the basic premise of Condition-Based Maintenance (CBM).

4.1 Condition Based Maintenance System

The general objective of a Condition-Based Maintenance System can be summarized
as the ability to view multiple datasets associated with a specific system or component
set, in an environment that facilitates better operation and maintenance decisions [5].
This means that the CBM System must have the ability to access all data sets in a
timely manner, and the system must be designed for scalability, from relatively small
data sets to substantial databases.
The CBM System must be able to display these data in a useful manner, so that the
operator can determine the state of monitored machinery relative to time and process
demand level. It is important to represent attributes such as condition and rate of
change.
The goal of the CBM System is to aid in forecasting or predicting possible condi-
tions for any selected piece of machinery. In order to successfully achieve this goal,
accurate data correlation is required. The CBM System should allow baseline or
standard operating parameters for machinery to be input from an external data analy-
sis source. In particular, it is desirable that the CBM System can be configured to
select key condition identifiers from all available condition identifiers for display
and/or correlation. Data sets should be possible to be correlated against each other in a
meaningful manner. For example, pressure vs. temperature, and specific vibration
frequency parameters vs. load etc. are examples of simple correlated sets. Any corre-
 15

lation analysis should take into account the asynchronous nature of the data to be
correlated, and provide algorithms that allow accurate cross correlation.
It is possible with such a CBM System to receive meaningful data that can be in-
corporated by plant operators to provide accurate condition assessment of the ma-
chine, as well as possible fault diagnosis. It is therefore important to differentiate
between the requirements of condition monitoring and those of fault diagnosis. A
plant operator should be able to utilize the condition data received from the CBM
System to accurately assess the condition of the machine and provide a diagnosis for
the possible machine fault.

4.2 Condition Monitoring Versus Machinery Diagnostics

At this point, it is important to differentiate between the requirements of condition

monitoring versus those of machinery diagnostics. Consider Figure 14, which is the
famous bathtub curve for probability of failure of machines with time [5]. In the run-
in period, the probability of failure is high, until the machine goes into normal opera-
tion, which (if the maintenance work is done correctly) should be a long period of
time. Upon the inception of a fault, point A on the bathtub curve, the probability of
failure starts to rise again.
We will not concern ourselves with the run-in period, since usually careful human
attention is given to new or repaired equipment until they pass the run-in period.
However, it is important to clearly define the requirements for the jobs of monitoring
and diagnosis. In normal operation, the purpose of monitoring is to detect any change
in machine condition that would lead to a machine fault (point A on Figure 14), while
the purpose of the diagnosis is to find the cause of the fault and time to failure (be-
yond point A on Figure 14).

Figure 14 Bathtub Curve

For plants with large numbers of machines, the task for monitoring machines can
be quite burdensome, particularly that vibration data (the main tool for CBM Sys-
tems) are dynamic in nature, and usually require measurements on a large number of
points on the machine, in different domains, and with different spectral ranges and
resolutions [2]. Moreover, most of the time, condition monitoring data are not provid-
ing new information on the condition of the machine, but actually the condition data
16

should be telling the user most of the time that the machine is in good condition (oth-
erwise that plant has a major problem). Thus, it makes a lot of sense to reduce the
amount of data collected on a machine during monitoring to simple numbers that can
clearly indicate the level of vibration on the machine.
On the other hand, for diagnostics, the CBM system already has determined that
the machine is faulty, and it is required to diagnose the cause of the fault. This usually
requires the full analysis of the dynamic data measured on the machine. This means
that all the data on all points on the machine, on all directions, and with different for-
mats, spectral ranges and resolutions, are to be used in the diagnosis [2]. Moreover,
special testing, such as Operating Deflection Shape (ODS) analysis, may be required
to accurately diagnose the cause of the fault, as well as the time to failure.

4.3 Condition Evaluation Based on Statistical Data

Statistical analysis is usually used as a condition evaluation tool rather than as a diag-
nosis tool. This is because statistical analysis can help reduce the condition evaluation
burden, and accurately determine the condition of the machine from a large amount of
data.
Vibration data usually come in different formats, and contain a wide set of infor-
mation. The vibration data can be presented in spectrum, time waveform, orbit, ampli-
tude, phase, and many other data formats. In order to be able to judge a machine con-
dition, the vibration data are usually condensed to a single number. This single num-
ber is termed the Overall Value, and can be quantified as peak value or root mean
square (rms) value, and either as displacement, velocity or acceleration data [20]. This
Overall Value is then trended and compared to limits.
Using vibration measurement to evaluate machine condition, computerized
maintenance systems can evaluate machine condition based on overall levels of
measures. Overall levels of a measure are typically judged in terms of limits. Two or
three alarm values are typically used in the trending process. An alert limit may result
in initiating collection of a spectrum or a time waveform, or changing the frequency
of periodic measurement to be taken, and usually signifies a deterioration in machine
condition. An alarm limit would signify the need to shut down the machine. Normal,
surveillance and shutdown limits are the most commonly used limits for condition
evaluation [20]. The overall levels should be monitored and recorded for a period, in
order to be compared with each other.
Usually, a trend plot is collected in velocity data, and the monthly data can be plot-
ted as shown in Figure 15. Either the limits chosen can be selected based on ISO or
Vibration Institute standards, or experience gained with the machine. The data accu-
mulated are compared to the limits, and machine condition can be evaluated. If the
Overall Value is below the first limit, then the machine is normal; on the other hand if
the overall value is between the two limits, then the machine should be under close
surveillance, and maintenance action should be planned. However, if the vibration
level is above the second limit then the machine should be shut down and mainte-
nance action performed immediately.
 17

Figure 15 Trend Plot

In the trend plot, some statistical analysis, such as a regression analysis or correla-
tion analysis may be useful. This will aid in understanding and evaluating the trend of
all points on a machine, as well as forecast the future behavior based on its history.
Figure 15 shows actual trend data (Navy lines) for a point on a machine, with three
limits shown (green line for Alert limit, yellow line for Alarm limit and red line for
Danger limit), and a regression line (light blue line). Note that the machine condition
was normal, but a fault developed some time during June 2001. Maintenance action
during August 2001 reduced the Overall value within normal condition.
Basing maintenance action on the Overall Value can be useful, but is incomplete.
The reason is that the Overall Value may not always be sensitive to the change in
machine condition. This is particularly true for rolling element bearings [21]. The
change in bearing condition and initiation of bearing fault rarely affect the wide band
measurement embedded in the Overall Value. Rather small changes in the amplitude
of particular bearing frequencies indicate deterioration in bearing condition. Usually
different techniques such as ultrasonic detection or spectral analysis of vibration data
would be required. However, spectral analysis is an involved process. There are two
simple techniques that utilize spectral data to trend the vibration, that can be used to
determine the condition of machines, and are much more sensitive (than overall val-
ues) to the change of individual defects in a machine.
These two techniques for trending of the vibration data are band alarms and spec-
trum enveloping. In the band alarm technique, limits are defined for preset frequency
bands not for the entire frequency span. This technique is powerful in monitoring
bearings; since the vibration levels of bearing fault frequencies are always lower than
other frequencies appearing in the spectrum. Figure 16 represents a sample of a band
alarm plot.
18

With the band alarm technique, a single value is measured for each band of interest
and these band amplitudes are trended and subjected to limits, same as with the over-
all value. However, with the band amplitudes, these measures of the machine condi-
tion are each sensitive to a particular fault in the machine. Thus it is possible to detect
early faults with band amplitudes, before they affect the overall value of the meas-
urement. Thus, the condition of the machine, and some basic diagnostics, can be eval-
uated through the trending of band amplitudes, and compared to band limits.

Figure 16 Band Alarm Using 6 Bands

The second technique, the spectrum enveloping technique, uses a spectrum mask
for each peak in the spectrum, such that each frequency component has its own alarm
and shutdown levels. Using this technique, it is possible to monitor the changes that
occur to each vibration component and to determine whether this component exceed-
ed the previously set shutdown level, or not. Figure 17 illustrates a sample for the
spectrum enveloping. Basically, the spectrum enveloping technique does not provide

Figure 17 Spectrum Enveloping

 19

any particular amplitude, but rather it provides a spectrum mask. This spectrum enve-
lope is usually based on the current normal spectrum, and both an alert and a danger
envelope are initiated, such that each narrow band in the spectrum has an alert limit of
twice the normal value. If in any subsequent measurement a single line reaches one of
these envelope limits at any frequency, the user can immediately be alerted of the
deterioration of the condition of the machine.

5 Vibration Diagnosis

It has been debated for a long time whether the diagnosis of rotating machinery could
be standardized. In fact the debate always centered on whether the diagnosis process
is an art, where the knowledge of a human being in diagnosing machine faults is a
personal process that relies on the knowledge and experience of an analyst, or wheth-
er it is a science that can be explained and documented.
However, the International Organization for Standardization (ISO), through the
ISO 13373, series has adopted an approach that relies on Fault Tables to document
available knowledge on machine faults, and the use of step-by-step flow charts or
process tables to document the structured procedure that is used by many analysts in
reaching an appropriate diagnosis of a machine. This process, particularly the use of
flow charts for diagnosis, was pioneered by the author in his papers on the diagnosis
of installation faults [13] and the diagnosis of fans [14].
The author of this chapter has participated in developing the ISO 13373 series as
the Convenor of ISO TC108/SC2/WG10 and as the project leader for many of its
standards. This section summarizes the author's opinion and many of the approaches
that are accepted now as international standards.

5.1 Standards Overview

The ISO 13373 series consists of the following parts, under the general title Condition
monitoring and diagnosis of machines — Vibration condition monitoring:
 Part 1: General procedures
 Part 2: Processing, analysis and presentation of vibration data
 Part 3: Guidelines for vibration diagnosis
 Part 4: Diagnostic techniques for gas turbines and steam turbines on fluid
film bearings
 Part 5: Diagnostic techniques for fans and blowers
 Part 7: Diagnostic techniques for machine sets in hydraulic power generating
and pump-storage plants
 Part 9: Diagnostic techniques for electric motors
The following parts are under development:
 Part 6: Diagnostic techniques for gas turbines on rolling element bearings
 Part 8: Diagnostic techniques for industrial pumps
 Part 10: Diagnostic techniques for generators
 Part 11: Diagnostic techniques for gearboxes
20

ISO 13373-3 [8] is the base document for the diagnosis process. It provides guide-
lines for the diagnosis of machines. ISO 13373-1 [6] describes the measurement pro-
cedures on machines for diagnosis purposes, while ISO 13373-2 [7] describes the
processing and analysis of data for diagnosis purposes. The other parts [9-12] are
machine specific standards that describe the unique diagnostic procedures for each
machine type.
ISO 13373-3 [8] is described in the standard as: "has been developed as a set of
guidelines for the general procedures to be considered when carrying out vibration
diagnostics of machines. It is intended to be used by vibration practitioners, engineers
and technicians and it provides them with useful diagnostic tools. These tools include
diagnostic flowcharts, process tables and fault tables. The material contained herein
presents a structured approach of the most basic, logical and intelligent steps to diag-
nose vibration problems associated with machines. However, this does not preclude
the use of other diagnostic techniques." [8]
The diagnosis process is described in ISO 13373-3 [8] as consisting of flowcharts,
process tables and fault tables. The flowcharts and the process tables are essentially a
step-by-step question and answer procedure that guides the user in the diagnosis pro-
cess. The flowcharts are used for an overview of the vibration events and characterize
the features, while the process tables are used for more in-depth analysis. The fault
tables are used to illustrate common machinery events and how they manifest them-
selves [8].

5.2 Initial Analysis

The diagnostic process starts by asking some basic questions, both on the safety of the
humans and machines and to ascertain the history and current status of the machine.
This is included in the normative Annex A in 13373-3 [8], which specifies the sys-
tematic approach to the vibration analysis of machines:
a) Annex A.1 is used to gather background information regarding the machine,
nature and severity of the vibration.
b) Annex A.2 is used to answer a set of questions aimed at arriving at a proba-
ble diagnosis of such common faults as unbalance, misalignment and rubs.
c) Annex A.3 is used to set out certain considerations when recommending ac-
tions following a probable diagnosis.

5.3 Common Analysis

The diagnosis of faults common to a wide range of machines are shown in other an-
nexes in 13373-3 [8]:
— Installation faults and examples are described in Annex B.
— Radial hydrodynamic fluid-film bearing faults and examples are described in
Annex C.
— Rolling element bearing faults and examples are described in Annex D.
 21

The use of ISO 13373-3 [8] is necessary with all machine specific diagnostics
standards [9-12], and should be used in conjunction with these machine specific
standards, as it describes the initial analysis that must be applied to the diagnosis of
all machines [8] as well as common faults to all machines including installation faults,
fluid film bearing faults and rolling element bearing faults.

5.4 Fault Table

A sample fault Table is given below for the diagnosis of fans and blowers [10]. The
systematic approach to vibration analysis of fans and blowers is given by the fault
table in Table 2. The fault table includes mainly installation faults. For faults regard-
ing fan or blower bearings, see ISO 13373-3 [8], Annexes C and D. Several faults can
give similar indications and further investigation would be necessary to distinguish
between them.

Table 2. Fault table for fans and blowers

Fault Vibration characteris- Other descriptors Comment

tics
Shaft misalign- 1x, or 1x and 2x, some- Directional force 180° There are two types
ment/ concen- times 1x and 2x and 3x. phase shift across cou- of misalignments:
tricity pling. Offset misa- parallel and angular,
errors lignment tends to pro- and in most cases
duce phase shift across there would be a
the coupling in the combination of the
radial direction, while two.
angular misalignment
tends to produce the
phase shift in the axial
direction.
Looseness Usually a series of Looseness can be at Looseness can be at
peaks at rotational bearings or skid, or the bearing housing
speed and integer har- anchor bolts. Check for (sometimes due to
monics of rotational difference in amplitude the bearing installa-
speed, generally the and/or phase at the tion), and/or at the
amplitude of these interface to discern pedestal or the skid.
peaks decreasing with position of looseness.
higher harmonic num-
bers.
Excessive bear- 1x. With low amplitude Directional. Can be due to wear,
ing clearance harmonics. in both fluid film
and rolling element
bearings.
22

Piping strain 1x Directional, wave clip- Piping flanges

ping in time waveform. should match with-
out jacking.
Soft foot 1x, plus 2x line fre- Soft foot test. Soft foot is the con-
quency in the electric dition that exists
motor when all feet are not
correctly supporting
the machine. See
also ISO 13373-9.
Shaft rubbing Clipping in time wave- Not commonly ob-
form, with 1x and mul- served on fans.
tiple harmonics in spec-
trum. Light rubbing can
cause rotating vectors
(spiral vibration).
Unbalance 1x Phase shift across cou- Unbalance is often
pling depends on the due to erosion, or
mode. Cylindrical deposits on blades.
modes tend to have 0° Overhung fans may
phase shift across the require a couple
coupling, while conical balance, while cen-
modes tend to have tre-hung fans can
180° phase shift. generally be bal-
Usually, 90° phase anced in a single-
shift between the hori- plane.
zontal and vertical
measurements at the
same bearing location.
Bent shaft 1x similar to unbalance, Can cancel with unbal- Rarely seen on fans.
manifests itself at slow ance at particular rota-
roll speed. tional speeds.
Casing distor- 1x, sometimes 2x. 180° phase shift from Only important
tion end to end. where bearings are
integral with the
casings.
Resonance High vibration at a par- Resonance testing Avoid operating
ticular frequency. indicates natural fre- close to a resonant
quency. frequency e.g. by
changing speed, or
by changing reso-
nant frequency, e.g.
by stiffening ma-
chine or adding
 23

mass. Sometimes
damping may be
needed.
Tilting founda- High 1x vibration levels Rocking motion in 1x ODS study to ana-
tion that cannot be explained ODS. lyse problem in
by unbalance, misa- more depth.
lignment, bent shaft or
eccentricity.
Aerodynamic Blade passing frequen- Can have high noise. Usually caused when
forces cy. fan is operating off
best efficiency point.
Belt faults Belt Passing frequency. Less than 1x. Typically due to belt
wear, misalignment
and/or incorrect
tension.
Belt resonance Belt resonance frequen- Usually less than 1x. Usually due to lack
cy. of belt tension.
Excessive belt 1x Directional. Similar symptoms to
tension misalignment
Belt pulley Usually directional 1x, Sometimes visually
eccentricity sometimes 1x and 2x. observed as wobbly
motion.
NOTE ODS stands for operational deflection shape.

The Fault table (Table 2) above includes the most common faults in fans and blow-
ers including the descriptors of their vibration characteristics, as well as other fault
descriptors. To complement the Fault table, ISO 13373-5 [10] for fan diagnostics
introduces the concept of a Symptom table (Table 3 below). The Symptom table
includes the observable symptoms of typical faults, in the form of elevated vibration
signals (subharmonic, 1x, 2x, ..., etc.) as well as vibration phase and other diagnostic
discriminating factors. The Symptom table distinguishes between symptoms almost
certain to be seen if fault occurs and symptoms that may or may not be seen.

5.5 Symptom Table

An example of Symptom table used by ISO 13373-4 [9] for steam and gas turbine
diagnostics, is given below in Table 3.
24

Table 3. Observable Symptoms of typical faults

Elevated
Vibration Time Cri Barring
Signals tic
Va
al
Su Gr In rie Re
sp Bar s pe
dd ad St ee Au cr
Fault type Tr rin wi at Comments
en ua ea d di- ea
an g th ab
Sub >2 ap l dy ch ble se
1x 2x si not loa le
1x x pe in St an Ru d
en pos d
ar cr at ge bbi sl
t si-
an ea e d ng ow
ble
ce se roll
Shaft unbal- Immediately
• • • • •
ance (generic) evident
Shaft unbal- Most effect at
ance (loss of • ○ • ○ • • ○ bearings of
material) affected rotor
Bearing Ele- Occurs fol-
vation change ○ • ○ ○ • • ○ ○ lowing transi-
ent.
Permanent
• ○ • • • ○ •
Bend
Transient E.g. During
bend – no • • ○ ○ ○ ○ temperature
rubbing changes
Transient During speed
bend - hard • ○ ○ • ○ ○ or load
rubbing change
Oil whirl (or Whip locks
whip) • ○ ○ ○ ○ rotor to 1st
critical speed
Steam in- Load depend-
duced vibra- ent. Modify
tion admission
• ○ ○ •
sequence;
repair dia-
phragms;
install nozzle
blocks
 25

properly
Steam whirl / Load depend-
steam induced ent. Reducing
vibration steam flow
• ○ ○ •
rapidly re-
moves prob-
lem
Rotating stall Return to
in GT correct flow
• • • • immediately
removes
problem
Differential
creep on ro- • • • •
tors
Rotor Crack Critical speed
may reduce
• ○ • • •
and show two
peaks
Looseness in
bearing or • • • ○ • • ○
pedestal
Morton Effect Periodic vari-
/ light rubbing ation in am-
• • ○ plitude ap-
prox. 30 –
120 mins
This table is not exhaustive but contains the most prevalent faults associated with
steam and gas turbines with fluid film bearings
• Indicates symptom almost certain to be seen if fault occurs.
○ Indicates symptom may or may not be seen.
26

5.6 Flow Chart

The use of flow charts in machine diagnosis is quite useful, and presents a step-by-
step approach to machine diagnosis that can be applied in the field. The diagnosis of
Installation Faults in ISO 13373-3 [8] is actually based on reference [13]. The stand-
ard methodology illustrated in Figure 18 starts by visual inspection but considers
spectral analysis as the main component of the testing of the installed machine (M/C).
In addition, resonance testing, time waveform analysis, phase analysis and operational
deflection shape (ODS) analysis are used if and when judged necessary.
Basically, a visual inspection of the machine and the site should be completed be-
fore any testing of installed machinery be performed. In many cases, the presence of
skid looseness and/or piping strain would be evident to the naked eye. Actually, it is
suggested that all skid and anchor bolts be tightened before testing an installed ma-
chine. Also, all piping connections should be checked before testing. All flanged
connections should be checked to make sure that connecting bolts pass through the
flanges without any restriction, thus causing no piping strain.

Figure 18 Flow chart for installation faults [13]

 27

Spectral analysis is the core of the diagnosis of rotating machinery. These spectral
data should be measured on all bearings on the driver and driven machine, in all three
directions, horizontal, vertical and axial. Complete knowledge of the machine should
be available to identify characteristic frequencies. The purpose of the spectral analy-
sis is to identify the frequencies causing the machine to vibrate. If all vibration ampli-
tude levels are within acceptable limits, then the machine would be accepted as nor-
mal. However, if any of the spectral components has high amplitude, then spectral
analysis is used to correlate the frequency of the high amplitude vibration to a ma-
chine frequency.
The result of the spectral analysis of the high amplitude vibration is one of three
cases: a) at a known frequency, b) at an unknown frequency, or c) at the running
speed. By a known (unknown) frequency it is meant that the reason of the presence
of this frequency in the spectrum is known (unknown). The first case is the easiest to
analyze. If the high amplitude vibration is at a known frequency, then the problem is
correlated to that known frequency. For example, if 2x vibration is identified then this
is usually correlated to misalignment. If decreasing harmonics of the running speed
are present in the spectrum, then this spectrum shape is usually correlated with loose-
ness in the bearings or the skid. If however, there were unknown frequencies in the
spectrum, then additional testing would be required to determine the source of the
unknown frequencies. Amongst the additional testing that may be required are: reso-
nance testing (including impact (bump) test and critical speed test), modal testing, and
flow characteristics testing. The purpose of the resonance testing is to correlate the
unknown frequency to natural frequencies (stationary components) or critical speeds
(rotating components) of the machine. Modal testing is a more advanced form of
resonance testing, where all the modal characteristics of the machine are determined,
including natural frequencies, damping ratios, and mode shapes. Modal testing is
rarely used in the field, as it is an elaborate testing method, and is usually time con-
suming and costly. However, when justified, it can be a very powerful tool to obtain
the machine characteristics and both identify clearly the unknown frequency in the
spectrum, and suggest a solution to the problem. As for the flow characteristics test-
ing it is always a good idea to make sure that the rotating machine is operating at or
near the best efficiency point, otherwise high amplitude vibration is to be expected.
This is the case for recirculation and cavitation in pumps, and stall in compressors.
The most difficult case occurs when the spectral analysis reveals high 1x vibration.
There are many faults, related to installation problems, that lead to high 1x vibration.
Amongst these faults are unbalance, misalignment, casing distortion, tilted founda-
tion, skid leveling, piping strain and excessive bearing clearance. In this case, special
vibration measurements have to be conducted on the machine to describe the nature of
this 1x vibration, and to distinguish between the different 1x faults. These measure-
ments include: time waveform measurement, phase measurement, and measurement
of the ODS.
The time-waveform measurement can be used to distinguish between misalign-
ment, piping strain and excessive bearing clearance. For piping strain, it will be quite
clear that the forcing on the machine is directional, usually in the horizontal direction,
28

and this directional force would be acting on the whole machine. Inappropriate bear-
ing clearance also results in directional forces, however this would be localized at the
bearing with the inappropriate clearance. This is particularly true for special geome-
try bearings, such as lemon bore or multi-lobe bearings.
The phase analysis is quite important to diagnose unbalance, misalignment, bent
shaft and casing distortion. In many cases, misalignment (the main installation anom-
aly) manifests itself as vibration at 1x only. One of the best ways to distinguish be-
tween 1x vibration due to unbalance and 1x vibration due to misalignment is to meas-
ure phase across the coupling. If there is a 180° phase shift across the coupling, then
the problem is misalignment. If no phase shift occurs across the coupling, then the
problem is unbalance. Casing distortion can be easily identified by 180° phase shift
across the machine (side-to-side or end-to-end) in the horizontal, vertical and/or axial
directions. A cocked bearing can be identified by measuring phase around the bear-
ing housing and noticing the phase shift due to the wobbly action of the cocked bear-
ing. In many cases, a coupled time-waveform-phase analysis is quite useful in visual-
izing the vibration pattern and identifying the problem.
If the 1x vibration problem is still not solved after the time waveform and phase
analysis, then an Operational Deflection Shape (ODS) should be measured. The ODS
is useful in identifying problems of tilted foundation, skid leveling, skid looseness,
and shaft parallelity. In the ODS measurement, phase-referenced 1x vibration is
measured at grid points on the machine structure or skid. This reveals the actual de-
flection shape of the machine under the operating load, and at the operating speed.
Note that the ODS is not a mode shape of the machine or structure, unless the ma-
chine is in resonance, but it can be considered as a summation of the contribution of
all of the modes of vibration. ODS analysis can be quite useful in identifying installa-
tion problems, as it provides a visualization of the actual vibration pattern of the ma-
chine and /or skid. In particular, if a machine skid exhibits a node in its ODS, then
this is a clear indication of a tilted foundation or a leveling problem in the skid. Ac-
curate measurement of skid and/or foundation levels would then be required to con-
firm the results of the ODS analysis.
Both reference [13] and ISO 13373-3 [8] provide many examples of the application
of this methodology in diagnosing installation faults.

6 Recent Trends in Vibration Monitoring and Diagnosis

The vibration monitoring and diagnosis technologies are at an inflection point. The
advent of low cost Micro-Electro-Mechanical-Systems (MEMS) accelerometers, with
wireless sensors (thus eliminating costly cabling) provides for readily available data at
reasonable cost. This means that eventually handheld data collectors shall be replaced
by a wide range of low cost wireless accelerometers that provide vibration data if and
when needed. These data can then be transferred to a cloud repository for further
analysis. Condition monitoring tools can be used as described in Section 4 for condi-
tion evaluation. Moreover, historian software are being developed to store and ana-
lyze Big Data. This can be further used by Artificial Intelligence (AI) techniques
 29

such as deep learning to provide useful condition evaluation and recommended ac-
tions.
Automatic diagnosis [5] is still further away. Some rule-based expert systems are
readily available in the market, and vary in their effectiveness. AI based diagnosis is
still at its infancy but will probably make its way to the market. However, a very
important tool for diagnosis emerged in the last few years which is vibration video
magnification. This technology is starting to replace ODS as a visualization tool, and
fits well with the current trends as it provides Big Data, that can plausibly be integrat-
ed in future automated diagnostic software.

6.1 MEMS Accelerometers

MEMS accelerometers are revolutionizing vibration measurement. Their low-cost

will eventually allow many more machines to be permanently monitored online.
The principle of operation of MEMS accelerometers is simple [22]. They are typi-
cally based on capacitors with a structure that uses two capacitors formed by a move-
able plate held between two fixed plates. Under zero net force the two capacitors are
equal, but a change in force will cause the moveable plate to shift closer to one of the
fixed plates, increasing its capacitance, and further away from the other fixed plate,
reducing that capacitance. This difference in capacitance is detected and amplified to
produce a voltage proportional to the acceleration. The dimensions of the structure are
of the order of microns [22].
The author has used MEMS accelerometers. The user needs to be aware of their
characteristics. An unannounced 4 kHz sensor resonance had to be mitigated by sig-
nal processing. The technology is still under development, but as it matures it will
certainly replace piezoelectric accelerometers in many applications due to their low
cost.

6.2 Wireless Sensors

Accelerometers with wireless connections send their acceleration signals through Wi-
Fi or Bluetooth technologies to a gateway that is connected through Ethernet with the
Internet. The wireless network should be designed such that it does not overload the
gateway(s).
Wireless sensors tend to be battery powered, some with impressive battery lives of
5 years. However, battery life depends on the use. Frequently using the sensor drains
the battery power. A common set-up is to have the sensor collect data for 10 seconds
every 8 hours. This tends to be a suitable setup to preserve battery life, but has to be
evaluated for providing suitable monitoring data as described in section 2 of this
chapter. Finally, whatever strategy is used for data collection, care needs to be exer-
cised to wake the sensors with a certain time lag in order not to clog the gateways.

6.3 Cloud Monitoring

Cloud vibration monitoring has been available for quite some time now. Not only
online systems send their data to the cloud, but also handheld data collector systems,
30

through their software, can upload their routinely collected data to the cloud. The
cloud monitoring software has evolved and now can provide different levels of data
access and display, allowing the technicians to monitor the technical data, plant man-
agers to monitor plant equipment status and program management, and executives to
do asset management.
Cloud Monitoring allows enterprises an opportunity to better manage their plants.
Opportunities of centralizing the analysis or remote monitoring, or even outsourcing
the condition monitoring services, provide unique strategies that can benefit the whole
organization.

6.4 The Use of Artificial Intelligence in Monitoring and Diagnosis

The presence of Cloud Big Data on condition monitoring and process parameters is a
ripe market for the application of AI technology for condition monitoring. In fact
several Asset Performance Management (APM) software are now available in the
market, using AI, particularly deep learning. These AI-based software require plant
data to be assembled in historian software to be easily accessible by the condition
monitoring software. This seems to be a successful application of AI in condition
monitoring and it appears to be gaining momentum in the market.
However, using AI in vibration diagnostics is still at its infancy. The author ven-
tured in using neural networks some time ago [23], [24] but developing a complete
automated vibration diagnosis is still not available. Major efforts are currently being
exerted in this direction [25], and we shall probably see the success of these technolo-
gies in AI-based vibration diagnostics in the future.

6.5 Vibration Video Motion Magnification

A recent technology has evolved that is a video-processing method that uses every
pixel in the camera and detects vibration motion. The process decodes light to pull out
information that is indicative of motion. The process involves measuring a video of
the machine or structure, identifying points of high vibration, analyzing the spectrum
of points of high vibration, identifying frequencies of interest, then amplifying and
filtering at the selected frequencies. The result is video of motion at the particular
frequency selected, showing the machine or structure behavior at a particular frequen-
cy. The result is very similar to ODS, but much more rapid in measurement and anal-
ysis. The author has used it many times in analyzing the vibration of machine casings
and supporting skids, and the analysis of bridges and condition monitoring of bridges,
updating finite element models based on measured bridge natural frequencies and
mode shapes. Note that vibration video motion magnification (VVMM) is a 2D
measurement, and it may be necessary to take multiple videos in multiple directions
to get a true 3D representation, and it needs to be emphasized that appropriate lighting
may be needed for in-door measurements.
An example is shown below for VVMM. A high speed test rig for testing turbo-
charger bearings was experiencing high unexplained vibration at particular frequen-
cies [26]. Measurement of VVMM videos of the test rig identified problems at 100
Hz and 180 Hz, while the test rig was operating at 42000 rpm. VVMM was used to
 31

filter at 100 Hz frequency, showing that the shaft is exhibiting an axial motion as
shown in Figures 19 and 20, which is obvious in the coupling contraction and expan-
sion, while there is an independent 180 Hz frequency of motor bracket rocking mo-
tion as depicted in both Figures 21 and 22. Of course, these figures are more impres-
sive in video format.

Figure 19. Test rig shot at coupling contraction (100 Hz).

Figure 20. Test rig shot at coupling expansion (100 Hz).

Figure 21. Test rig shot at motor bracket original state (180 Hz).

Figure 22. Test rig shot showing motor bracket is tilted towards the shaft (180 Hz).

The use of VVMM technology was quite useful in troubleshooting this test rig [26]
which allowed for further testing of turbocharger bearings.
32

It is conceivable that VVMM technology shall be used with Big Data and AI to
provide future automatic diagnostic capabilities.

7 Conclusion

This chapter reviewed the vibration measurement tools and the instrumentation used
in machinery monitoring and diagnostics. Also, a review of the vibration analysis
tools used for machinery diagnostics was presented. The description of the principles
of machinery condition monitoring were elaborated, and the process for machinery
diagnostics was described with some examples. Finally, the chapter discussed the
recent trends for machinery vibration monitoring and diagnostics.

References
1. Global Condition Monitoring Services Market, Forecast to 2025, Frost & Sullivan, July
(2018).
2. El-Shafei, A., Measuring Vibration for Machinery Monitoring and Diagnostics, Shock and
Vibration Digest, Vol. 25, No. 1, pp. 3-14, January (1993).
3. Eshleman, R.L., Basic Machinery Vibration, The Vibration Institute, USA (2002).
4. Peters, J. and Eshleman, R.L., Machinery Vibration Analysis, The Vibration Institute,
USA (2020).
5. El-Shafei, A., and N. Rieger, Automated Diagnostics of Rotating Machinery, Proceedings
of ASME Turbo Expo, Atlanta, Georgia, USA, ASME paper GT- 2003-38453, June,
(2003).
6. ISO 13373-1 Condition monitoring and diagnosis of machines — Vibration condition
monitoring: Part 1: General procedures, 2002.
7. ISO 13373-2 Condition monitoring and diagnosis of machines — Vibration condition
monitoring: Part 2: Processing, analysis and presentation of vibration data, 2005.
8. ISO 13373-3 Condition monitoring and diagnosis of machines — Vibration condition
monitoring: Part 3: Guidelines for vibration diagnosis, 2015
9. ISO 13373-4 Condition monitoring and diagnosis of machines — Vibration condition
monitoring: Part 4: Diagnostic techniques for gas and steam turbines with fluid-film bear-
ings, 2021, in press.
10. ISO 13373-5 Condition monitoring and diagnosis of machines — Vibration condition
monitoring: Part 5: Diagnostic techniques for fans and blowers, 2020.
11. ISO 13373-7 Condition monitoring and diagnosis of machines — Vibration condition
monitoring: Part 7: Diagnostic techniques for machine sets in hydraulic power generating
and pump-storage plants, 2017
12. ISO 13373-9 Condition monitoring and diagnosis of machines — Vibration condition
monitoring: Part 9: Diagnostic techniques for electric motors, 2017
13. El-Shafei, A., Diagnosis of Installation Problems of Turbomachinery, Proceedings of
ASME Turbo Expo, Amsterdam, The Netherlands, ASME paper GT-2002-30284, June
(2002)
14. El-Shafei, A., Fan Diagnosis in the Field, I. Mech. E. International Conference on Vibra-
tions in Rotating Machines, Exeter, UK, (2008).
 33

15. El-Shafei, A., Machine Diagnosis as a Standardized Approach, IFToMM International

Conference on Rotor Dynamics, pp. 207-221, Springer, 2018
16. El-Shafei, A., Standardized Fan Diagnostics According To ISO 13373-5, Proceedings of
the Second Vibration Institute Middle East Conference, RITEC, October (2020).
17. El-Shafei, A., Diagnosis of a Recurring Pump Shaft Failure, 37th Annual Meeting of the
Vibration Institute, San Antonio, Texas, June (2014).
18. El-Shafei, A., Diagnosis of Gearbox Faults in the Field, 8th International Conference on
Vibrations in Rotating Machinery, Swansea, UK, September (2004).
19. Younan, A. A. and El-Shafei, A., Model Calibration of Anisotropic Rotordynamic Systems
With Speed Dependent Parameters, ASME Journal of Engineering for Gas Turbine and
Power, Vol. 130, July, pp. 042502-1 to 042502-10 (2008).
20. Eshleman, R.L., Machinery Condition Analysis, Vibrations, Vol. 4, No. 2, pp. 3-11 (1988).
21. Berggren, J.C., Diagnosing Faults in Rolling Element Bearings – Part III Electronic Data
Collector Applications, Vibrations, Vol. 5, No. 2, pp. 8-19, (1989).
22. http://wikid.io.tudelft.nl/WikID/index.php/MEMS-based_accelerometer, accessed on
10.04.2021.
23. Hassan T, El-Shafei A, Zeyada Y, Rieger N., Comparison of neural network architectures
for machinery fault diagnosis. In: ASME Turbo Expo 2003, collocated with the 2003 In-
ternational Joint Power Generation Conference, American Society of Mechanical Engi-
neers Digital Collection, pp. 415–424, (2003).
24. El-Shafei A, Hassan TA, Soliman AK, Zeyada Y, Rieger N., Neural network and fuzzy
logic diagnostics of 1x faults in rotating machinery. ASME J Eng Gas Turbines Power
129(3):703–710, (2007).
25. Nath, A.G., Udmale, S.S. & Singh, S.K. Role of artificial intelligence in rotor fault diagno-
sis: a comprehensive review. Artif Intell Rev 54, 2609–2668 (2021).
26. Ibrahim, M.S., Dimitri, A.S., and El-Shafei, A., Troubleshooting of a High-Speed Turbo-
charger Test Rig, Proceedings of the Second Vibration Institute Middle East Conference,
RITEC, October (2020).