ISO Standards

13373 Vibration condition monitoring general procedure

17359 General guidelines for condition monitoring

13379 Data interpretation and diagnostic techniques

13381 Condition monitoring and diagnostics of machines - prognostics

Successful Program
Instead of tracking cost that you saved per year, track uptime, MTBF, OEE
etc. Because cost will keep on reducing because of successful program
and less downtimes, it would look like that you are not adding value. You
can also track vibration levels before and after program.

Percentage of work on machines based on condition should be greater

than time based or breakdown based.

Make case studies to get others know what you have done.

Don’t always publish bad news that bearings failed. Also share good news
that someone did great.

Always keep bearings for study.

Instead of publicizing someone has done something wrong, use the

opportunity to train him.

The Analysis

Harmonics occur due to non-linear vibrations.

Sidebands occur when there is amplitude modulation.

The peak sidebands are grouped around is called center frequency.

The first peak of a harmonic series is called fundamental frequency.

Humps are due to cavitation, noises.

Fault verification is necessary to confirm whether issue was resolved or

not.
Condition Monitoring Technologies

 Oil analysis
 Vibration analysis
 Wear particle analysis
 Ultrasound/acoustic emission
 Thermography
 Motor circuit and signature analysis

Condition monitoring =/= Predictive maintenance

Condition monitoring is just monitoring, predictive maintenance is to act

in accordance with condition.

Do the right test, in a right way and in the right time.

Condition monitoring = Health monitoring

If we take care of the machines, they will be more reliable.

Predictive maintenance will reduce downtime, labor cost, parts inventory,

increase profits.

Each technology requires careful test frequency.

Make enough measurements in a balanced manner to not miss any fault
event

If program is running properly, each stakeholders take interest in how to

make things better.

Different technology results can be analysed together for better

understanding.

Monitor vibrations and find changes over time.

Vibration Analysis

It tells us nature and severity of the problems.

Tests are performed usually every 30-90 days.

You can start from 30 days but if you are confident machines are okay,
you can go till 90 days.

In hazardous locations, we mount permanent sensors and junction boxes

outside hazardous areas.

In remote areas, permanent sensors can be mounted on critical machines

to save time in mobility.
Acoustic Emissions and Ultrasound

Used for very high frequency sounds such as;

 Air leaks (40kHz range)

 Steam leaks
 Lubrication problems (bearing impacts due to lack of lubrication at
30kHz range)
 Electrical faults (Arcing, nuisance corona, destructive corona)
 Flow related (limited)

Range above 20,000 Hz is called ultrasonic range.

Human ear can only detect sound in the sonic range (20 to 20,000Hz).

Sounds are directional and volume depends upon distance from the
source.

Sounds can travel though air/gas, liquid and solid objects but not
vacuums.

Heterodyne technique (demodulation) is used to make the sound/signal

audible to human ears.

Two modes to detect sounds, airborne or direct contact.

Sophisticated tools can even detect bearing faults of wheels by just taking
airborne data of moving train.

Two ways to use ultrasound data, either listen for faults, or view time
waveform and spectrum.

Can detect bearing faults? Debateable.

Useful in high noise environments where abnormal noise is not heard

easily.

Thermography

It is the study of radiated heat energy using infrared imaging system.

It is NOT measuring temperature. It is measuring amount of radiated

energy.

Celsius shown doesn’t actual mean temperature is being measured.

Actually, radiated energy is proportional to heat which can indicate faults.

Spot radiometers measure temperature.

Temperature is measured as the average of the area within the laser’s

field of view, not exactly at the laser point.
It has certain other applications in military, rescue and animal husbandry,
mountains/fires etc.

Mechanical Applications: Used for monitoring pipes, bearings, belts, steam

systems, and refractory plants to detect overheating, friction, or insulation
loss.

Electrical Applications: Used for inspecting transformers, motors, control

panels, loose connections, unbalanced loads and overloads to identify
abnormal heating or electrical resistance.

Vibration monitoring picks problems earlier than thermography.

Thermal images should be used along with visual images.

You may not know exact temperature but relative temperature.

Camera detects radiated energy and displays the temperature gradients

(Useful in knowing hottest or coldest locations).

Surface type, wind conditions and other factors affect accuracy.

Camera looks at radiated electromagnetic energy from first 1/1000” of the

surface.

Things may be hot from inside which camera is not picking, so need to
open cabinets or look at windows to detect faults.

Camera detects three forms of radiation, emitted heat, reflected heat and
transmitted heat. That’s why it is important to know object’s emissivity.

Following are emissivity of some objects.

 Black Body = 1.0

 High (0.9 and above): water, snow, ice, soil, minerals, glass.
 Low (0.2 and below): polished metals

Colour and paint can also affect emissivity.

If there is wind, even relative temperatures/gradient are not reliable.

Sun light will also affect temperatures reported.

Electric Motors

Some tests include;

 Vibration (stator, rotor, air gap, soft foot, bearings)

 Current and flux (rotor)
 High potential (insulation)
 Surge (stator)
Offline Tests include;

Winding resistance, meg-ohm, polarization index, high potential, surge

testing, motor circuit analysis.

Motor Circuit Analysis (can detect defects at early stage):

 Winding shorts
 Winding contamination
 Insulation to ground faults
 Air gap faults
 Rotor faults (voids and broken bars)

Motor Signature Analysis (running condition):

We check flux (like magnetic fields).

View current and voltage signature (treated like vibration spectrum).

Can detect broken rotor bars.

We get LF but no sidebands if there are no broken rotor bars. db difference

between amplitude of sideband and LF tells severity.

Signal Processing

Overview:

Data collector takes analog signal from sensor.

Analog signal is converted into digital signal (because we need numbers).

How quickly we digitize it, how long does it take to digitize, how many
samples to keep etc.

Process the signal and convert into into rms overall readings, envelop
spectra etc.

For that we have to filtering, integration, sampling windowing and

averaging.

Filtering

They are used to remove effects of integration, aliasing

Used to remove low frequency vibrations to focus on high frequency

vibrations.

Low pass filters: Allows (pass) only low frequencies and reject the rest.

Band pass filters: Allows only certain range of frequencies and reject all
above and below this range. (best for bearing defects)
High pass filters: Allow only high frequencies and reject the lower ones.

Band stop filters: Block certain range of frequencies and allow all the rest.

Ideal filter design is such that it will either allow 100% to pass through or
100% to block from certain frequency.

But in actual, filters are not perfect, and signal is only partially blocked
from certain band/frequency. Means amplitudes above the allowable band
may reduce eventually becomes 0 at higher region.

In past there were analog filters with capacitors and resistances etc now
we mostly use digital filters, have much sharper cut off.

Integration is a form of filter.

Allow signal to settle first. For example when changing settings. Otherwise
we will see bad spectra, time waveforms and can see ski-slope.

Readings change with change in filter settings.

Hitting sensors hard would generate ski-slope.

Digital filters perform better, use mathematical calculations.

When to use filters?

Anti-aliasing filters, integration, tracking, demodulation, peakview,

acceleration enveloping.

There are tracking filters too, who track speed of machine at run up coast
down test, and changes setting of filters automatically when machine
speed is changed.

In Peakvue, acceleration enveloping, we reject low frequency vibrations to

focus on high frequencies for example when studying for bearing damage.

Integration

The analog signal from accelerometer is naturally proportional to

acceleration.

To monitor velocity, we have to convert signal from acceleration to

velocity using integration.

Integration and differentiation processes are done using filters.

Integration reduces high frequency vibrations and amplifies low frequency

vibrations and introduces a 90 degrees phase change.

Lower frequency vibrations are more readable to displacement and higher

frequency vibrations are more readable to acceleration.
When balancing, focus on displacement, when checking gearbox, focus on
acceleration.

Relationship between acceleration and velocity is 1/f.

A = v/t  axt = v  v=a/f

If f is very small, velocity will be very large and if f ix 0, velocity is infinite.

Converting acceleration to displacement means very high amplitude, low

frequency signal is gonna come through.

Ski-slope means lower frequencies are amplified by integration process.

That’s why there are high pass filters to block frequencies below like 10hz
which will corrupt data.

If analyzer or we don’t remove, low frequency vibrations then spectrum

will have so much noise.

Reason why people may not get low frequency readings when they are
inspecting low frequency vibrations/low speed motors, is because the
software is cutting it out due to filtering.

When we integrate, we push down higher frequencies and lift up lower

frequencies. When we integrate again, we do this again.

High pass filter at 5 Hz in velocity is used to push down lower frequencies

where as at 10 Hz in displacement.

We need high signal to noise ratio.

Noise can come from electronics, transducers, flow, process or external
machines.

Noise can come if displacement probes are used at high frequency like
1kHZ, accelerometer is used at low frequency or quiet machine.

If we want to avoid noise, avoid integration and do the following.

View data from displacement probes in mils or um.

View data from accelerometers in gs or mm/s2.

Integrations can be done through analog circuitry or digitally.

Accelerometer>integration at analog circuit inside data collector>analog

signal>digital converter (time waveform and spectrum will be in velocity).

Different analyzers have different methods of integrations.

When machines are run at low RPM, then integration followed by high
pass filter attenuates the low frequency vibrations, giving almost no
meaningful signal. Or if simply integration, then only noise will be
produced.

High sensitivity sensors (1V/g) are best for low speed machines, which
give stronger signal.

In advanced analyzers, the builtin features are enough that you may
never have the need to change settings to remove noise.

Sampling/Digitization:

Analog signal coming from transducers have all the information of

vibrations.

We have to convert this analog signal into numbers to make it useful.

Data collector converts analog signal into digital signal. The process is
called digitization.

We need to capture the signal for a period of time to make it more

understandable, that is done by Fmax and lines of resolutions.

LOR dictates how many of these numbers will be acquired.

Velometer, accelerometer or proximity probe generates analog signal.

Data collector acts like volt meter. It reads/acquires data at fraction of

seconds.
The rate at which we acquire data is called sampling rate or sampling
frequency (samples per second).

Sampling rate/frequency is how many times per second we acquire value

of voltage from transducer.

The analyzer will choose the best setting according on Fmax.

Fmax decides how much time there is between each samples.

All voltage values at particular interval of time joined together to form a

series of numbers, called timewave form.

If sampling rate is low, then if a spike occurs between two samples, then
we will never know what happened.

We select sample rate in such a way that we could record all kinds of
frequencies we are interested in.

If we are interested in very high frequency event, that can be captured

within two samples, then we have to choose very high sampling rate,
which will make us choose higher Fmax.

In low speed machines higher Fmax is not needed, but in high frequency
events like gear mesh or bearing defects, we need higher Fmax.

Fmax = Fs / 2

To correctly capture a signal, the sampling rate must be at least twice the
highest frequency component present in that signal.

If your signal has vibration energy above Fmax (e.g., 6 kHz in this case),
the analyzer will misinterpret it as a lower frequency — that’s aliasing.

Triggering:

The analyzer waits ‘some time’ before it starts taking data.

Examples

 Impact test (sensor is placed but data collection only starts when
first bump/strike occurs)
 Order tracking
 Run up, coast down
 Phase measurements
Data collector can be triggered based on vibration levels. Means if
vibration reaches certain amplitude, only then data is captured.

Normally in this case, vibration is continuously being monitored, we set a

% pre-trigger, for example, 10% pre-trigger, then whenever first time
amplitude reaches a trigger limit, collector will show 10% of the data
before the impact and 90% after like in the figure below.

Set the input gain (amplitude limit) first before doing triggering event.
Because the first huge impact may be missed by data collector. It will then
set its gain. #

In triggering, we cannot afford delay so we have to set gain ahead of time.

It is used in the background quite often.

Fast Fourier Transform:

Fourier transform was used to transform infinite time series into an infinite
frequency spectrum.

Where, a=amplitude, f=frequency and 0=phase.

Fourier transform sets mathematical foundation but it is too time

consuming.

Discrete fourier transform is the broader form of creating spectrum from

finite time series, but it very time consuming.
Fast fourier transform (FFT) is a form of DFT but with a faster algorithm.

To achieve speed, the number of samples in time series must be a power

of 2. 256, 512, 1024, 2048 etc.

Instead of infinite, we use discrete time waveforms.

FFT works best on waveforms that include periodic signals.

It will build sine waves out of every peaks.

All FFT conversions have phase components too. Spectrum shows

amplitude and frequency but not phase.

To make phase meaningful, we need a reference.

Unless we perform cross channel measurements, phase readings are

discarded.

Time waveforms/records will have a defined number of samples, N =

1024, 2048, 4096 etc.

From 2048 samples in time waveform, we end up with 1024 lines, that
represent 1024 amplitude and 1024 phase.

When analyzer performs FFT calculations, it takes out real part and
imaginary part.

We can combine into to form amplitude and phase.

Real and imaginary part makes right angle.

Using Pythagoras theorem, we will find amplitude.

Used square of 2 for ease of calculations right.

Unless it’s the case of ODS or cross channel phase measurements, we

throw these 1024 phase values away.

FFT have limitations.

 Aliasing

We need atleast two samples per cycle if we wish to resolve that

frequency.

 Leakage
 Resolution (picket fence effect)

We only have data at specific frequencies.

Aliasing:

Nyquist Theorem

If samples are low, we won’t be able to record any frequency as after

joining these points, it will be a flat line.

We therefore need to sample faster.

If time period of input signal is larger than the time record (time between
two samples), then we won’t be able to capture the input frequency /
spectrum.

Sampling rate must be greater than twice the highest frequency of

interest.

Nyquist theorem: Fs > 2 x Fmax (Fx = 1/Ts)

Fs = Sampling Time, Ts = Sampling Time.

For example, if I have to inspect sidebands of gearmesh, then that will be

our Fmax.
Not enough sample points in above picture, this will create a signal that
never existed.

Fs < than Fmax here in the above pic.

We will be confused what is generating this lower frequency vibration

(dotted). In actual it is not real.

Two signals are said to be alias if the difference between their frequencies
falls in the frequency range of interest.

This always happened in the process of sampling.

For example, if sampling frequency is 37.5hz (less than twice of Fmax),
and actual frequency is 30hz. Then will see frequency of 7.5hz in the
spectrum.

Rule1: Fmax should cover all frequencies under consideration.

Rule2: Whatever Fmax we choose, there will always be higher frequency

than that (harmonics of gearmesh etc).
If sampling frequency is exactly twice that of input frequency, then the
result will be 0 and no peaks will be observed. In the picture, Fs=60, and
Fmax=30

If Fs>2Fmax, the resultant frequency in spectrum will be accurate.

If Fs is too high, then resultant frequency will be very perfect, however
data collector will take a lot of time since it has to deal with a lot of
numbers.

When we choose our Fmax, the data collector chooses how quickly it
needs to sample the data in order to give you a good valid spectrum.

In case of multiple frequencies (harmonics etc), use Fmax as highest input

frequency like 6X (180Hz) here in the pic. Then sampling frequency should
be atleast slightly greater than twice of 180Hz, in the pic it is 363.8Hz.
This will read the actual frequency and all the lower frequencies as well.

Coping with Aliasing

Look at the above picture. Lets say, we are interested only in frequencies
below 25Hz, then we will set Fmax=25Hz, and sample rate > 2x25Hz =
52Hz. But even then, there could be another frequency at 30Hz, but due
to aliasing problem, it will appear at 52-30 =22Hz. So even if you tried to
stay out of aliasing issue from a 20Hz input frequency, you got the
problem because of 30Hz frequency. You may mistakenly take it as real
frequency.

This means all vibrations above Fmax will have alias problem and will be
shown in region we are interested in. That’s bad.

Because of this issue, why cant we use low pass filter to allow only
frequencies upto Fmax and ignore the rest.

But we know in reality, its actually like below graph.

This means still some frequencies above Fmax or the frequency above
which we need to filter, will be able to pass. So they will once again
appear on spectrum which is a problem.

So whats the solution?

Fs was slightly above 2Fmax right? This has created the problem.

Instead we will use Fs=2.56Fmax

If input frequency appears above Fmax but within green area, the alias
peak will also appear in that area, but Fmax wont allow it as the alias
frequency is above Fmax.

For example,

If Fmax=25Hz, and Fs=2.56Fmax=64Hz, and Fin=30Hz, Then alias peak

will be at Fp= Fs-Fin=64-30=34Hz.

Moreover, Filter is working well in blue area, so it will filter out any
frequency in blue area.
This way, all peaks slightly above Fmax in green area, alias peaks will be
in green area but above Fmax, so they will be discarded and all peaks in
blue area will be automatically filtered using low pass filter.

So sampling frequency should not be 2Fmax but 2.56Fmax.

If Fs=2.56Fmax, it may mean N number of samples. Lets say 2048

samples.
Then we will get N/2 = 1024 lines, but we have to get rid of lines above
Fmax in green region. So we will only get N/2.56 = 800 lines.

Why 2.56? We could have used smaller number. If we use N=2048

samples, we end up at nice number of 800 lines. If we use 4096 samples,
we end up at nice 1600 lines.
This is because if our filter was ideal, then we would be getting 1024 lines
out of 2048 samples not 800. This margin is given because of non-ideal
filters.

Modern filters are digital, which may have better filtration. So we may get
900 lines or 950 lines.
Some analyzers have good filters so they will have better lines and Fmax.

So, to avoid aliasing, Fs=2.56Fmax

Low pass filter at Fmax

In old analyzers, with each Fmax we choose, one filter is used. So when
we change

Fmax, analyzer takes settling time, to change filter and then to allow
taking measurements.

New analyzers have Delta-Sigma, where we do things differently. They are

able to take thousands of samples at very high rate. Then when we select
any Fmax, it will give the data and discard the rest.

Rather than providing one filter for every Fmax, it will only have 1-2
filters.

Sampling and Averaging:

We can control two parameters when taking measurements, sample rate

(samples/sec) and total samples.

If we want 1000 samples/sec, and total 3000 samples it will take 3 secs.

We need to know the effects of samples vs LOR and Fmax on spectrum.

Increasing Fmax requires increasing Fs=Sampling rate by 2.56 times.

(Means if we need to resolve higher frequency, we have to use higher
sampling rate).
This means whatever maximum frequency we need to analyze, analyzer
will select Fs=2.56Fmax.

Rule2: Number of samples N is proportional to resolution or lines.

For example, if Fmax = 1000 Hz, then Fs will be 2560 Hz means, it will
take 2560 samples per second.
Now, we have to choose total samples or LOR. If total samples are 2048,
we will take readings in less than a second (2048/2560 = 0.8sec).

But we normally do average so it takes longer time than that.

Despite knowing Fmax and Fs, is our job done? No. We need to see lines of
resolution, or total samples. A good LOR will improve resolution or
spectrum details.

You may need different settings for TWF and Spectrum.

The greater the resolution, the greater the time taken and memory used.

The faster we sample, the more data we will collect.

Lower resolutions or record length or total samples, means broader peak.

Greater resolutions or record length or total samples, means sharper peak

.
In the above chart, 800 resolution/lines with 800 Fmax means, each
frequency is represented by 1 line. In this case, only vibration appears at
15 Hz, which is our input frequency. One thing to note is that if number of
lines are increased to twice, then peak will be much sharper.

Measuring time = Time it takes to measure 1 time record.

Frequency (spectrum) resolution = Fmax/LOR

We don’t know how much vibrations or peaks in between these lines. For
example, no peaks or vibrations will be produced between 104 and 104.5.

Table below shows relations among each parameters of formula above.

If we double the Fmax, keeping LOR constant, then time record will be
halved because sample rate will also be doubled.
If we double LOR, then sample time will be doubled as analyzer will take
longer time to inspect spectrum.

Low Fmax but very high LOR, not very good idea.

Zoom Analysis:

To see pole pass or gear mesh frequencies, we need to see high resolution
data.

One option we have is to increase line of resolution, like 3200, 6400 lines
etc.
Or we can choose zoom measurements.
If we select Fmax=1600Hz and 1600 lines, then we will have 1Hz between
each lines.
But in reality, in spectrum we dont see each lines.
Even on high resolution it is difficult to see details.

Analyzers will have Pseudo-Zoom (means specific setting). Options will

have specific Fmax, Fmin and LOR.

If we need data between 600 to 800 Hz, the analyzer wont do anything
clever, instead discard frequencies that are not within this selected range
and give the spectrum of selected range.
We are not improving resolution, we are just zooming.

We need to check analyzer what setting it provides and what settings are
good for our measurements.

There is another thing called True-Zoom or ‘band selectable analysis’. It

performs heterodyning. It shifts the range of frequencies we are interested
in to lower region towards left of spectrum, a technique similar to
ultrasound where sounds louder than audible range are brought in lower
range with correct labels.

Dynamic Range:
So far we have seen frequency resolution, its amplitude resolution.

Number of bits are used in analyzer to convert analog signal to digital.

Dynamic range is the ability of analyzer to show low amplitude peaks in

the presence of high amplitude peaks.

1980s, dynamic range was a big issue. Low amplitudes were lost.

There is voltage range in analyzer, if there is a voltage, that is above the

scale, the analog to digital converter will get overload. If its too low, then
data will not be very perfect.
The above is the example of poor dynamic range, with time waveform
chunky.

Dynamic range is quoted as dB value.

dB is the ratio of smallest signal to biggest signal.

A very ‘wide’ or good dynamic range means we can resolve very small
signals in the presence of very large ones.

Old analyzers had 12 bits, which means 72 dB (theoretical) according to

following formula.

The lowest amplitudes get hidden under raised noise floor.

The lowest amplitude should be higher than the noise.

New analyzers have 16 bit A/D converters that provide dynamic range of
96 dB.

If input signal was 1 volt, we can still detect changes as small as 0.015
mV.
If an analyzer has 5 bit=32 dB A/D convertor , then lowest voltage it can
measure on the presence of a 1V signal, is 0.03125V or 31.25mV.

Means any voltage signal below 31.25mV will not be shown properly.

What if signal is only 0.05 volts or 50mV?

Dynamic range will also be reduced greatly. 20 log (31.25/50) = 4dB

(which is very terrible measurement).

Analyzers took time to set their gain, means it waited some time before it
decided which setting/scale to choose.

If the voltage is too high, it may start resetting because voltage received
will be much higher than selected scale. It will then choose new scale.

If analyzer keeps waiting, it may be checking gain setting.

Some analyzers save the gain settings they use for future.

Some analyzers allow you to pre-set input range to save time required for
‘auto-range’ function.

Windowing and Leakage:

Ideally, time records should be infinite, but in actual they are not.

What comes before and after record should be same, but in actual it is
not.

The FFT result will be same if time records are infinite and before and after
record, waves can be joined perfectly. But that rarely happens.

In above case, spectrum will be affected, this issue is called leakage.

This is the real signal, FFT calculation is assuming this time record is
identical and infinite and this block can be fit before and after to get same
signal. In actual it doesn’t happen.

Imagine fourier transform will have trouble if this doesn’t happen?

If we are unable to solve this issue, then base will be very broad.

Solution to this problem is Windowing. Means force all frequencies to start

and end at 0.

Windowing changes the shape of the time waveform and squeezes it from
both ends so it begins and ends at 0.
This way, we will be able to get repeatable measurements.
This is what windowing is all about.
This means if we get FFT of first, and the second, we will get same result.

Windowing is necessary and you cannot get away from it.

But does windowing change resolution or produce modulation effects?

Lets find out.

Most common type of windowing is Hanning.

If we choose 800 lines or 2048 samples, we will get blue signal. And then
we apply windowing. The resultant wave will have similar resolution.

Previously we have read, we have 1600 Hz measurements and 1600 lines

spectrum, we get 1Hz resolution. The windowing effect reduces this
accuracy of spectrum.
1Hz resolution, 15Hz frequency. If frequency is changed to 15.2Hz, we will
still see peak at 15 Hz as my lines are 1Hz apart.

If machine speed is increased to 15.2 Hz, then we make get peak like this.
Neither 15, nor 16, amplitude drops.
If its important to us to know more accurate peaks, then we have to
change resolution.

We improved resolution to 0.5Hz, this way, all lines are 0.5Hz apart and
we will get more sharper peak.

Now, we have high resolution data, no leakage issue.

But, what if machine is moving at 15.6 Hz?
The amplitude drops because energy is leaking from key frequency.

We improve this issue by windowing, although peaks are still different, but
sharper than before.

The amplitudes of either sides of key frequencies are equal. If we increase

frequency, the ratio changes in very predictable way.

This is Hanning windowing.

But the good thing is that, the software would know analyst used hanning
window so it will reverse engineer the spectrum. This means, it will
identify actual frequency with actual amplitude.

With functions, named such as ‘Locate’, we will be able to identify actual

frequency and peak.
But it can fail in one situation. If two peaks are too close to each other. It
will pick one peak. We need better resolution for that.

Look at the above graph higher resolution graph. If we turn off, hanning,
we will still get two peaks, because we have higher LOR as shown below.

But still, spectrum resolution will be affected and broader base will be
formed.

Peaks will always appear at bin/line. If we have a signal of 48.25Hz, and

our bins / resolution is 0.5 Hz apart, then two bins will be at 48Hz and
8.5Hz. Not 48.25Hz.
The peak’s amplitude will be dropped to 0.85 if it appears at Hz.

Our calculation of resolution must be altered to include the window factor.

Any time when two peaks are closed to each other, Focus on bandwidth,
not just resolution when analyzing data.

Because resolution calculation will tell us we will be able to see two peaks.

But bandwidth calculation will say, no these two peaks will be merged
together.

Example, 1600Hz frequency, and 800 lines, spacing will be 2Hz.

Hanning window: window factor 1.5
Bandwidth = 2x1.5=3Hz.

Hanning most commonly used.

Sometimes people get confused with Hamming, its not used much these
days.

Amplitude accuracy is poort, 15% or 1.5 dB error.

Most software can compensate.

The further we are away from bins, the greater the affect of amplitude is.

‘Flat Top’ window has great accuracy and have error of only 1% or 0.1 dB.
But it has poor resolution with broader peaks.
Flat top windowing are not used much.

Rectangular / Uniform / No window option.

Windowing has very bad affect on time waveform. It flattens start and end
of the wave.
What if wave has very important info at start and end?

If any useful information is there at the beginning or end, it will be

squeezed out.

Like bearing impact or gear tooth impact.

Exponential Window is useful in modal and impact test to know the

response of structure after impact.

Hanning window will squeeze the impact at the beginning. So cant use
that.

So whenever taking data, focus on choosing Fmax rather than LOR,

because if Fmax is not high enough we will loose faults.

Averaging:

Vibration changes from moment to momentum due to noise, and as

rotating elements go through cycles.

To reduce impact of noise and variation and to maintain repeatability we

do averaging.
Two measurements taken 5 mins apart should be same.

Linear Average = (S1+S2)/2

All four spectrum will be added and divided by. Every single lines.

RMS Averaging

Averaging doesn’t remove noise it averages it.

If peak is fluctuating, we will get average of it.

4 averages are done by many people but more can be done.

Shaft must have rotated atleast a few times.

If there are shock loads, impacts, cavitation etc, multiple such events
should have been collected.
To see everything clearly, there should be more samples, high resolution
along with averaging.
Low resolution data with high averaging wont give full details.

When we averaged, higher frequencies have more variations and can be

seen in the graph.

When we get averaged peak, we would not know whether it was a

constant peak or a peak with very high variation.

We get repeatability after 6-7 averages.

Averaging starts right after pressing start button. In the above case,
spectrum will be averaged and low amplitudes will be given, but much
more impacts are about to happen which wont affect the averaged
amplitude much.

So we need to listen to machine, if the impacts are late, we may have to

increase resolution so it takes more time to collect data, so averaging will
also take more time, hence capturing better results.
How averaging works? The analyzer captures a time record, windows the
data, creates the FFT and then computes the average.

Overlap Averaging:

We lose some data because windowing squeezes out data at start and
end. Important information/impact at start or end may get lost.

We don’t want that information to get wasted. So we can overlap the time
records, to save time and ensure we don’t lose important data.

This one is % overlap. But we will leave after 4 averaging for instance, and
don’t have to finish entire to capture entire waveform.

Exponential Averaging:
Linear averaging uses all history of data equally, exponential averaging
give more weightage to recent data than older data.

We don’t use method in our daily routine.

Subtraction/Negative Averaging:
It’s a great function but not available in many analyzers.

During bump test, we will start averaging as we start to bump.

First step, bump+machine own vibration.

Second step, no bump but just machine vibration.

It will subtract all latest vibrations from first one. To get only bump
vibrations.

Best for impact testing and finding vibration from resonance.

The only problem if the machine is generating important vibrations , then

it will be subtracted.

Peak Hold Averaging:

We use it for special test, for daily routine we use linear averaging.
A very good technique, we take maximum of all instances, to get a peak
spectrum.

Incase of linear averaging, impacting or when bump testing, all impacts

will be averaged, resulting in lower amplitudes.

While in Peak Hold, it will give maximum of all.

It can also be used in runup coast down, to get maximum value when
changing speed.

Time Synchronous Averaging:

It’s a special type of averaging useful for gearbox analysis.

Data collector starts collecting at arbitrary time as soon as we press go

button. At that point, machine could be doing anything.

It can reveal low amplitude peaks.

Averages in time domain, instead of frequency domain.

Then perform FFT on averaged time waveform.

When we average in frequency domain, we don’t remove noise, we

average it.

Where as when we average in time domain (using time synchronous

averaging), we not only remove noise, but also unwanted vibrations
coming due to any reason.

It achieves the results by synchronizing its measurements.

Time records must be synchronized.

We must use a trigger for each record.
Built-in tachs or tachometers are often used.
The above time records are not synchronized. It will result in cancelling of
waves as averaging of these will be 0. Therefore it wont give any peaks.

These are all synchronized using tach at 1X of shaft speed (motor,

gearbox etc).

After 1st time record is finished, it waits until new tach pulse is generated
at same point. Then next time record is started.

We may need 100s of averaging, and after sometime all noise, and non
synchronous vibrations will be smoothed out.

If there is other vibration such as bearing frequency, and since due to non-
synchronous nature, all time records will have bearing impacts at different
times, so they will be averaged out.

Ideal technique for complex gearbox. Because other non-synchronous

gears will be averaged out.
Some analyzers have tachs with multipliers, useful to multiply and divide
other sources of synchronous vibrations.

Problems with TSA:

Eliminates all non synchronous peaks even bearing frequencies.
Difficult to set up/

May need synchronization with shaft we don’t have tach for.

Order Tracking:

Not very common, but very useful in variable speed machines.

Some applications include run up, coast down, and VFD machines such as
in mining industry (draglines), elevators, vehicles, cranes, hoists.

Unless precautions, are taken, spectra will be ‘smeared’.

In previous lectures, we assumed machine speed is constant, when

averaging, sampling and all. That’s why everything works well.

If speed changes during a time record, after all the processes and later
averaging, all peaks of variable frequencies will be averaged out, creating
a broad smeared peak (distributed over several frequencies).

To overcome this, first we need to know how fast speed is changing.

Amplitudes will drop as averaging will spread the peak over a larger
portion. It may develop two different peaks closed to each other.

Larger the number of averages, larger the amplitude drop due to more
variation in speed.
Higher frequency will have smaller number of samples.

Lower frequency will take longer time to complete the cycle, will have
more samples.

1X running speed may not be constant, it actually gets shorter and shorter
as machine gets faster and faster.

Relationship between number of samples per cycle and time between

each sample is not consistent with actual cycle that we are measuring.

When order tracking is off, and machine run up testing is done. Look at
analog signal, digital signal will also be changed accordingly.
If order tracking is turned on, we will have different time waveform. Means
digital time waveform will be uniform. Actually we are telling the tool
instead of sampling every Xsecs, look at input signal and change sample
accordingly.

In mining industry, speed changes dramatically. So order tracking is very

important application here.

One solution (avoid order tracking) is to collect 1 or 2 average when

machine is at constant load or speed, then stop the process and when
load and speed is changed.

Repeat the process to collect additional averages.

It may require several tests before we have sufficient data. There could be
several minutes of delay between each test due to operating conditions

Solution 2 is to measure speed as each spectrum is collected, adjust each

spectrum according to difference in speeds. Then we can average
together. Smearing will still happen.

Solution 3 is the vary sample rate according to fluctuations in speed. Use

tachometer to record instantaneous machine speed

We actually are trying to change sampling frequency in relation with

varying machine speed.

Tracking ratio synthesizer:

Tracking ratio synthesizer will take tach as input and produce pulse stream
at a much higher frequency, like 100 pulses per cycle.

The analyzer has to do much faster sample, it needs something to tell it

how faster it has to sample.

So, order analysis or order tracking is when we adjust the sample rate
accordingly.
No Hz and CPM are gone, we are dealing with Orders of shaft rate
(whatever the shaft frequency is).

The tracking ratio synthesizer is an expensive device.

TRS can also produce output that is at a different ratio of running speed.

Problem occurs when speed changes too quickly.

Shaft encode is like a tach, but producing a large number of pulse.

Like 1 pulse per degree of rotation.

There is a digital way of doing that too. Capture all the data first, digitize
it, and also digitize the tach signal too.

Analyzing the tach signal will inform us how speed of machine varies.

Modern systems can do in real time.

Fixed frequencies will move across spectrum.

Like 2xLF, Resonance, external vibrations.

Time Waveform Analysis

Useful to diagnose problems when they are difficult to analyse through

spectrum.
Best for gearboxes and bearings.

Anytime we see harmonics and sidebands, noise floor, time waveforms

can be used.

Time waveforms can detect severity of the faults.

Bearing frequencies not showing much in velocity.

Checking this, will make us think about inspecting time waveform.

Shows some impacting.

Time waveform is a digitized version of analog signal coming from sensor.

We have to select settings, how quickly it is digitizing and how long it is

digitizing using Fmax, LOR, Samples.

Duration of waveform depends upon resolution settings.

Number of samples or sample rate depends upon Fmax.

X axis defines time in seconds. You need to know how much time to
include, keeping in mind gear mesh, bearing balls, number of teeth etc.
The data should be enough to include the impact.

If machine is telling you story, you need to listen to entire story, for that
you have to spend more time on machine.

Settings will determine how good time waveform is.

Above picture has different LOR and Fmax.

You generally get two choices, you will decide on Fmax and LOR to get
best possible spectrum and waveform.

LOR is same for both. But Fmax is different, so sample rate is different.
(3413 vs 213 samples per sec) / 16 times lesser in second one.

Memory is not the issue in analyzer anymore.

In both waves, we are standing at machine for same time. Setting is

important.
Sometimes, even using same sample rate, resolution and timing,
waveform can be different. Because measurement started at different
point of cycle of the machine.

If timing is more, then we will be able to see full waveform which could be
identical.

You can either set sample rate directly or use Fmax to set.
You can also specify number of samples to acquire 1024, 2048. You need
to also tell how long do you want to take data, for that, use LOR.

Why settings are important?

Because machine wants to tell you story, either long or short. We need to
capture enough cycles (not too low, not too high) to be able to inspect
bearing faults, cavitation, rubs, beats, modulation or gear teeth/mesh.

The solution is to get as many samples per second as possible along with
how long can you wait patiently.

1987, Catlin. If T is period of rotation.

In earlier times, memory was scare.

Normally, 6-10 rotations of the shaft should be captured.

If too many rotations are captured, it will harder to see waves.

Solution 1: Zoom in and look more closely, then zoom out for broader
picture like modulation etc.

Solution 2: Take two time waveforms. 1, for 5-10 seconds (acc g), and
second for quarter of a second for example (vel mm/s).
Long time period will allow us to see beating phenomenon. This would not
normally be collected.
We have to listen to machine in order to capture this long data. Without
knowing, no one would be able to capture this long waveform.

In live spectrum, we will see peaks up and down. But in spectrum, it will
be averaged.

It becomes important if somebody else collects such data you then should
be alert why he or she collected this. Go and inspect because otherwise
information would be lost.

Choose a time window so 6-10 cycles can be seen.

Too many cycles of rotation hampers the analysis (unless you collect a
large number of samples).

What unit to measure?

Incase of bent shaft, misalignment, imbalance, use velocity. Incase of

bearing defect, cavitation, gearmesh, use acceleration.

Some analyzers ask first. Some convert in software.

Extracting Information:

Time between events should tell nature of fault. It can determine

frequency using T=1/f

Software may be able to detect orders.

What is causing these spikes? Take assistance from software to determine
times of two consecutive impacts, subtract to get time period of that
impacting cycles. Use T=1/f to find frequency. 0.0448-0.0339 = 0.0109,
1/0.0109 = 91.74 Hz=5504 CPM.

What 91.74 Hz ( 3.09X)? Is it just 3 with an error or is it 3.09 and

harmonics telling bearing frequencies?

If each ball is making two impacts (bing bing… bing bing.. ) it will also
create same frequency for each in spectrum. Waveform will tell better.
Harmonic?

Looking at time difference, we can tell its belt damage.

Time waveform is instantaneous picture of whats happening.

If two peaks are closed to each other, spectrum will average it and we
may lose one peak. But time waves will generate both.

We can find out beating symptoms.

Amplitude modulation is common in bearing and gearbox faults.
Impacts are greater in load zone.

Anytime we see vibrations rise and fall we do get sidebands.

Sidebands can come from many reasons, there are many reasons of
vibrations rise and fall wrt to 1X.

Time waveform will reveal why? Possible bearing fault.

If spectrum doesn’t reveal anything, time waveform will make it clear.

If non linear vibrations are clipped, spectrum will show harmonics.

Time waveforms will reveal further information.

Impacts can be easily seen in waveform.

Many reasons for ski-slope, but if we go to time waveform, it will give

exact reason.

Rotating looseness generates impacts per revolution.

If you see a lot of harmonics, jump to the waveform.

Classic Looseness

In looseness, sub harmonics are also present (1/3X and harmonics.

Cavitation

Noise floor raised up.

Time between impacts will not be periodic. Because of random burst of

energy.

People argue that how much data is enough data? You don’t have to
analyze all, just get all information you can and analyze the unusual.

Different orientations will give different spectrum. Like vertical in above

graph showing best impacts.
Impact at every 1X revolution. May be due to gear problem.

Good time waveform, no abnormal impact.

If you want to know, or tell people , clients about gearbox, check time
waveform.

Resolution Vs Fmax

With LOR and Fmax setting, the analyzer has to collect certain
amount/time of data, in order to give us what we have asked for.

Lets increase 2X LOR, the amount of time needed to give 1600 LOR will be
doubled.

Higher resolution will take more time.

If we have to control our time waveform using Fmax then we have to use
a higher Fmax. Because to be able to read higher Fmax, analyzer has to
collect more samples and at faster rate. But this will shorten our time
waveform.
Higher Fmax at constant LOR will reduce time span of wave because it will
have higher sample rate, so recording will be finished.
We wont be able to detect any beating and all.

So what can we do?

Use higher Fmax to get detail, but also higher LOR to record for longer
time to detect impacting, beating etc
Hotwell Bursts of Energy

Something is happening in machine every now and again. Its not periodic.

Unless you don’t listen, you wont understand what is happening just by
looking at spectrum.

These impacts are not occurring at regular interval of time. So if we

reduce Fmax (to reduce sample rate to increase total time taken), we
would still not be able to record impacts / bursts.
Windowing will eliminate impact if occurred at start. And averaging will
further reduce it.

It is very easy to not able to capture anything via waveform when

machine is telling you something.

Key thing is to capture for as long as possible.

LISTENING is very important.

5-6 secs will be atleast better to capture many meaningful impacts.

Cavitation
Spectrum is continuously changing.

If we take data for 5-6 secs, we will hear and see bursts of energy,
randomly with no period. Random events.

In spectrum we may see vane pass and noise.

Looking at spectrum (accel. purple), we may see noise floor, not rising and
falling much due to averaging. Not a very good tool to analyze something.
Best info is in time waveform with best settings.

Increase resolution to get more information and indication of what is

happening.

Lot of information can be lost in spectrum so never lie in spectrum alone,

always take waveform.

Ski Slope:
Large ski slope. Why?
Was there very high frequency vibration that is causing this issue? Any
clipping?

That could be due to measurement/connection problem.

Something physical/electronic, not due to machine.

The whole waveform will be offset.

You may want to send this report to OEM that makes accelerometer as
evidence of what is happening.

Phase Analysis

Very important topic in diagnosing.

Common fault conditions can be accurately diagnosed using phase

analysis.

Should have two channels in analyzer.

It is used to visualize motion of the machine.

It is used to detect unbalance, misalignment, bent shaft, eccentricity,

cocked bearing, resonance, looseness.

It is important to record motion of the machine relative to two points. How

one point is moving relative to second point.

Phase readings don’t have to be taken at all times. Only when vibration
changes. Then confirm fault using phase analysis.

Phase measurements also help us in determining where weights need to

be added.
Understand forces involved with fault condition. Relate with expected
movement of the machine.

Timing of differences of movement of the machine is important.

All different types of motion represent different fault conditions. Circular

motion, rocking motion, up and down motion, axial movement etc.

We can look at entire structures, equipment and understand how one

point moves in relation to another, using ODS.

It is difficult to understand motion with a few readings, so we do a lot of

readings to plot entire structure.

Phase is all about relative timing.

Phase angle:

Talking about 1X of the running speed.

1X = sine wave.

Two sine waves from two different sensors.

Find time period (crest to crest or trough to trough) of 1 wave. It will be

360 degrees.

Now find time difference from crest to crest between two waves. You will
find lag positive or negative (such as 120 degrees).
A single vibration, 1 wave wont give us phase. We need to compare it with
any reference.

Reference from either tach, or vibration from other source.

During balancing, the reference tells us where to place weight.

We can use once-per-rev reference using a tachometer signal (photocell

and reflective tape).

Time between each each blip is 360 degrees or 1X shaft as we paste

reflective tape.

The time between blip and crest of wave will represent time difference or
phase.
In above picture it is 90 degrees.

Tach signal is yellow. Green is vibration. In this case no phase angle both
signals are inline with each other.

But if I paste my reflective tape right at heavy spot, then it wont give any
phase as well.

Normally, when we put reflective tape, we have no idea where it is relative

to the balance / heavy spot.

Leading and Lagging Phase:

Phase angles are actually measure of amount of phase lag.

It can be argued that sine wave is leading compared with tach signal.

But actually its lagging.

If we look carefully, sometime AFTER tach signal we see a crest. So its

lagging compared with tach signal.

We measure phase lag. If there is lag, sign will be positive.

If we change speed, phase angle wont change. Vibration amplitude may

change but angle wont change unless it hits natural frequency and
resonance occurs.

Transducer Type

Displacement Sensor:

We were using displacement sensor in line with tachometer reference.

What if we move sensor by 90 degree ? Tachometer and displacement

probes are 90 degrees apart.
Our readings will be shifted by 90 degrees. (-90 or +90 same. Direction
different.)

All these readings are relative.

We need to know how phase changes with respect to transducer location.

Its easier to keep sensor and tach in line. But if you cant, always
remember the difference in mind.

What if we use velometer or accelerometer? Would I get different phase

angle? ABSOLUTELY.
It will have to double integrate to reach displacement. But phase shift will
be accounted for and if units are changed to displacement then same
phase readings will be seen compared with using displacement sensor.

90 degrees phase difference between max velocity and max

displacement.
90 degrees phase difference between max acceleration/min displacement
vs max displacement.

Max velocity is when displacement is 0. Max acceleration is when

displacement is minimum.
Using

Velocity always leads displacement by 90 degrees (-90 phase).

Accelerometer:

Acceleration always leads displacement by 180 degrees (-180 phase)

180 degrees phase makes it look as if heavy spot is on the opposite side
of the shaft.

During balancing my first reading wont matter, what matters is how my

second readings
If vibration is already too high and you don’t want to risk adding trial
weight during balancing, then you must figure out where you heavy spot
is during phase measurement. The reflective tape should be where my
heavy spot is. So trial weight should be opposite to counter it. It may not
balance but vibration will be reduced for sure.

Be careful about units. Because unit selected is of displacement and if I

put my trial weight opposite to heavy spot, then vibration will become
severe. Use acceleration in case of accelerometer.

If machine is small, not critical, not much vibrations then you may put trial
weight anywhere.

If we always use same sensor for all measurements, then it doesn’t

matter.
It matters if we change sensor type. Only important in balancing.

Heavy Spot vs High Spot:

If we are not sure about flexible or rigid rotor, this will be easy section else
we will read in dynamic section.

High Spot

A high spot is the location on a rotating machine where the vibration is

highest. It shows where the shaft or rotor is bending outward the most
during rotation.

Heavy Spot

A heavy spot is the location on the rotor where extra mass (weight) is
concentrated. It is the point that causes imbalance because it is heavier
than the rest.

This means unbalance can be represented by single amount of mass at

single angle.

Heavy spot due to imbalance may be due to any issue, like dirt, deposit,
erosion build up etc.

The net effect of all these effects will be represented by single point called
heavy spot.

Does heavy spot not same as high spot? No because of dynamics of the
shaft, rotor, because of natural frequency, bending mode, phase shifts etc.

Due to effect of all those dynamic effects there will be difference in heavy
spot and high spot.

Unbalance creates heavy spot which results in centrifugal forces resulting

in displacement values.
Crest of the wave is not heavy spot, its high spot.

Displacement sensor is sensitive to high spot.

Because of change of speed, we have some revolve going on.

High spot is the closest to the sensor. Heavy spot is the heaviest.
Increase in speed, shifts high spot, creating delay between two spots.

High spot lags the heavy spot. Means heavy spot goes first.

Rigid vs Flexible Rotor:

At low speed, (speed below 50% of first natural frequency), high and
heavy spots will be closer.

If we goes towards critical speed, phase shift starts.

At resonant frequency the phase shift is 90 degrees and amplitude
becomes maximum as shown in pic below.

As speed increases, the phase lags increase to 180 degrees, amplitude

drops.
Turbines run above first, even second critical speeds to be able to operate
smoothly away from natural frequencies.

Most machines we diagnose are rigid rotors running upto 50-60Hz, like
fans pumps compressors motors.

Big machines have mechanical lag as vibration is transmitted from shaft

through bearings to housing there is delay. A lot is going on in those
machines.

We need to keep everything constant when taking readings.

Even in vibration sensor there is sensor lag (crystal in accelerometer).

Electronics also cause phase lags. That lags may be documented by

suppliers, OEMs.

So phase lags are important, most important during balancing.

When we perform relative measurements, phase lag doesn’t matter as

long as .

We don’t change transducers or machine speeds.

Representing Phase:

How to represent phases on graphs, etc.

1. Vector Diagram (helps in balancing)

Each rings represent mils of vibrations.

For example 5 mils of vibrations at 60 degrees. (right pic)
Example.

What is phase convention?

True Phase: The phase angle increases opposite to the direction of

rotation. (Rotating protractor convention)

The phase angles increase in the direction of rotation. (fixed protractor

convention).

Phase can be represented by bubble (vector) diagrams as shown below.

Different types of representations.

If both angles are written as 45 degrees for example, it doesn’t mean they
are 45 degrees, it just means they have same angle.
Bubble diagram on machine.

These readings help us differentiate between cocked bearings on shaft,

housing, bent shaft, how much it is bent, rotor eccentricity, imbalance and
misalignment using combination of measurements.

We can also focus on single component.

Measuring Phase:

There are three ways to measure phase.

1. Light reflecting tape (or paint spot or anything contrasting to pick
difference between light and dark. There are now systems in which you
don’t have to put reflecting tapes.

Light is deflected back and produces voltage difference that is used by

analyzer.

2. Photo tachometer and Laser tachometer

3. Keyway or set screw

Look at analyzer and check if it has two channels or what method you can
use.

The tach pulse is generating once per rev signal and accelerometer is
picking another signal. The data collector derives the phase at running
speed or 1X.

Only 1X is considered while measuring phase.

Moving the accelerometer will change phase wrt phase.

Its all about relative position of tach and sensor.

Phase is not normally perfect, values fluctuate slightly 4-5 degrees ex.

If fluctuations are too high 90 degrees to 60 degrees, means very random

phase, means check looseness, cavitation etc.

When you place sensor, readings will be any value, 0, 300, 170. You
should not be worried about that.
The only thing we need to focus on is how readings change when sensor
location is changed, from above the shaft or below the shaft. From side to
other side, from face of motor to face of pump etc.

Two channels:

If you don’t have two channels, use tachometer as other reference

otherwise best is to use two sensors using two channel analyzer.

Phase can be calculated using several methods, remember spectrum has

all the phase details when computing FFT. Discarded due to no reference.

It would be useful if we wished to compare 1X and 2X.

You may be keeping second sensor in the bag with you to measure phase
when needed.

When using two phase channels, the phase readings may be in -ve. Its
because, one position with respect to other, simply swap the position.
Signs doesn’t matter.

Can keep on sensor fixed and other movable to identify phase change.
When we compute FFT, phase is calculated at every frequency. During
time waveform we don’t control time of first measurement so phase is
random doesn’t have any value.

Its very important to keep sensors aligned in same direction. Because

sensors locations are important to diagnose what kind of relative motion
machine is undergoing.

Always remember, location. For example you are taking motor axial and
pump axial both are in different directions, so adjust accordingly (180
degrees adjustment needed).

Unbalance:

Phase readings give idea of dynamic forces exerted due to problems in

machines.

Multiple phase readings help us understand motion of the machine in all

directions.

Some sophisticated machines are there, just enter phase readings and
know animation.

ODS is a great example in which we study entire structure.

Amplitudes are also important which helps us distinguish between

different faults.

Axial (phase) readings are very important. You have to take it if you
suspect any faults.

Both sensors should be inline. If not, at 180 degrees direction apart, then
adjust for the readings by 180 degrees.
If you don’t do it, you will be confused.

We need to beware of phase references. Strobe is a great tool to use.

If first reading is 23 degrees and second regain is 28 degrees which is

opposite direction to first, then adjust it 180+23 = 203

Why we use phase analysis, because many faults give peaks at 1X so we

need to differentiate between them.

People can get confused unbalance and consider it as other faults.

Static Unbalance:

Why we consider different types of unbalance? Because it helps us in

knowing what technique should we used, single plane or two plane
balancing etc.
Pure static unbalance means it is out of balance in single plane. (common
if blades/vanes on the rotor are narrow compared with their diameter.

If machine has couple unbalance but we performed static balancing, then

job is not done.

Majority vibrations in radial, amplitude varies with speed.

Both ends of machine are in phase. 90 degrees between vertical and

horizontal.

Horizontal velocity normally higher than vertical. (because machine may

allow horizontal movement more compared with vertical).

We get circular motion. Both inboard and outboard are moving in same
direction.

The greater the out of balance weight, greater the center of mass moves
away from physical center.

In above pic, machine is forced to rotate around the center, which is

center of out of balance weight.
Whenever we talk about in phase or out of phase or certain degrees, we
are talking about possible +/-30 degrees variation. Its normal.

Between Horizontal and Horizontal at two ends of machine, phase

difference will be 0 (+30/-30). In phase.

90 degrees phase difference between Horizontal and vertical at same

bearing.

In theory, horizontal vibrations should be equal to vertical vibrations due

to imbalance phenomenon, but to stiffness at vertical direction, vertical
vibrations are lesser almost half of horizontal.

If horizontal too high, go for structural / resonance issue check as well. It

may have imbalance and structural looseness too.

First perform other rectification then do balancing.

The amplitudes will be different, but ratios of horizontal to vertical

vibrations will be same and phase difference will be same between two
bearing ends.

Axial readings will be low may be 1/3 of horizontal readings.

Axial readings will be inphase from one end to other end. But don’t rely on
this axial phase readings. Focus on horizontal mentioned above.

Comparing all.

Couple Unbalance:

Pure couple unbalance is rare.

Can happen when a machine is statically balanced and couple is not

corrected.

Equal mass is placed at opposite locations of the rotor.

1X horizontal, greater than vertical.

Both ends of the machine are 180degrees.
Horizontal vertical are 90 degrees.

May excite resonance.

Because of stiffness and other reasons, we get non linear movement. Not
nice smooth movement.

Because of some harmonics or other frequency waves.

We may wonder are these harmonics real? Are these due to

misalignment?

Not pure circular orbit.

Wave is not proper sinusoidal.

Machine is not vibrating linearly due to out of balance forces.

Phase readings will prove it for sure which fault is there.

If looseness, too many harmonics, other wise we can distinguish it by
using phase analysis.

The presence of harmonics will confuse us with misalignment or

looseness.

This type of imbalance may generate some axial vibrations too. But real
forces are radial.

Rocking but circular motion too.

Some people confuse vibration of couple imbalance with 2X vibrations,
but its 1X.

IF red masses increase in size, the center of mass wont change because
both masses are in opposite direction. Both forces will counter act it. If
one of the forces will be greater, then center will be changed and it will
become dynamic unbalance.
A round about 180 degrees difference between horizontal of both ends.

Horizontal may be higher than vertical about 2X.

Horizontal and vertical will have 90 degrees phase.

Axial readings will be much lower BUT they will be out of phase, one side
to the other due to rocking motion.

If amplitudes are very low, phase readings will have more chances of
being unreliable and have more variations.

Any presence of misalignment bent shaft will affect axial readings.

Dynamic Unbalance.

It’s the combination of couple and static unbalance. It looks like 2X, NO.

Combination of static and dynamic unbalance movement.

1X high.

Amplitude proportional to speed.

90 Degrees between vertical and horizontal.

Horizontal often higher than vertical.

May excite resonance or resonance.

Phase difference between end to end will be either closer to 0 (if static is
dominated) and 180 (if couple is dominated)

With 0-180 +/-30 degrees possibility, how can u say what balance is
there, you don’t care.

If closer to 0 then do static balancing in single plane.

If closer to 180 then do two plane balancing.
We have to remove couple and static part. Center of mass may have
offset.

Phase not well defined.

Same horizontal to vertical ratio.
Amplitude varies.

Axial phase readings will be either of static or couple or combination of

both.

Vertical Machines:

Unbalance force will create radial vibrations.

Vibrations will be low in axial direction.
Higher we go up, higher will be the vibrations.

Large 1X radial.

If you don’t see pure 90 degrees or 1X, it means there are other forces
too. Don’t worry.

Radial and Radial at same direction will be in phase.

Overhung:

Important differences with overhung machines.

Very important to know if machine is overhung.

Forces are being generated in radial and axial.

1X high vibration.

High in axial and radial directions.

Can confuse with angular misalignment due to presence of axial

vibrations.

Readings should be taken from bearing closest to overhung impeller or fan

blades.

Horizontal and vertical at 90 degrees.

Bending moment causes high axial readings.

Readings will be in phase axially.

Readings will be between 0-180 degrees in horizontal to horizontal and

vertical to vertical.

Phase difference between horizontal to horizontal and vertical to vertical

will be same.

Misalignment:

Misalignment is very tricky to diagnose.

We need to know our alignment practices. Do we do precise laser
alignment or do not align properly at all.

Forces exist based on misalignment.

Its not very simple relation. A lot of time, machine has misalignment and
there is no change in spectrum.

Depending upon types of machines coupled, the forces will be different.

Depending upon presence of how much angular vs how much offset

misalignment, the forces will be different.

Relying on spectrum alone will be confusing.

It can generate only 1X vibration , similar to unbalance, and several

harmonics, similar to looseness.

Some rules exist are, if there is 1X high axial, there could be angular
misalignment, if 2X high in radial there could be offset misalignment.

3X, 4X may be present, evidence of misalignment.

Phase readings will be different depending upon above said issues. You
need to figure out what forces are exerting to come to a decision.

In misalignment, you are generating forces and also causing the machines
not to rotate smoothly. Because of angles introduced, its fighting to rotate.
Creating non linear vibrations. Wobbling, spinning etc. Creating
harmonics.

But we can distinguish unbalance from misalignment.

Rules of unbalance is clear. Phase angles, vibrations make sense.

Little bit harder to understand how phase angles make sense in

misalignment.

Eccentricity, cocked bearings etc generate these forces.

If misaligned shaft centers are parallel but not coinciding, it is parallel or

offset misalignment.

If shafts meet at a point but not parallel, making some angle, is called
angular misalignment.

Almost all times, we have combination of two.

A little shim at the feet of the machine changes vibration forces at each
rotation of the shaft, thus alignment and phase.

Coupling might tolerate some offset misalignment. Different couplings will

generate different forces.
Coupling these two shafts mean forcing coupling to join together, creating
forces. Shaft may bend little.

This scenario is also possible that machines with misalignment running for
long time, but as laser alignment is done first time, it fails.

Take whole situation into account.

High 1X axial due to angular vibration

High 2X radial due to offset

Check for thermal growth, it can create angular misalignment.

It be have several peaks upto 8X, but it wont have noise floor. Harmonics
wont be very strong. This will help in distinguishing between looseness
and alignment.

Coupling plays a major role in vibration amplitudes in misalignment.

Phase readings at ends of the machine are not good indicators, they vary.

We wont see phase readings of 90 or 270 degrees between vertical and

horizontal at same bearings. Instead they will be closer to 0 or 180
degrees.

If strange phase angles difference between horizontal vs vertical readings

at each bearings, then it may not be unbalance, may be misalignment,
supported by misalignment spectrum.

Amplitude varies and follow different rules. Generally vertical readings are
high.

Compare ratio of horizontal and vertical readings. Different ratios. Vertical

may be higher in one lower in other.

Axial readings are interesting one. If substantial angular misalignment

then axial readings will be higher.

If offset is dominated, then radial readings will dominate.

In axial misalignment, phase difference will be 180 degrees across
coupling.

The readings across the coupling may be out of phase or in phase. Though
it didn’t tell much, this situation will help us distinguish between
misalignment and cocked bearings.

Eccentricity:

Center of rotating is offset from geometrical center of gear, bearing or

rotor.

We get 1X peak.

Center of mass is offset so it will mimic strong imbalance but phase

readings will be much different.

Its wobbling.

Forces will damage belts, bearings, gears etc.

Highest vibrations in direction of orange line.

Vibration readings in line with the belt. Horizontal and vertical.

They will be in phase or out of phase, 0-180 degrees between horizontal

to vertical.

Very difficult to visualize.

But unbalance will be 90 degrees H to V. Here its 0-180 H-V. That’s the
difference.
Bent Shaft:

High 1X axial vibrations.

The mass of shaft is not balanced around center lines, wobbling, so will
create unbalance type symptoms so vibration will be high in radial too.

If bent is near shaft center, it will like imbalance so we will get strong 1X
peak.
We will get 2X peak if bend is closer to coupling.

We have to figure out is it bent shaft or misalignment?

Need to ask why it could be bent? Thermal growth?

We will get 180 degrees phase shift in axial.

Horizontal and vertical will also have 1X and 2X, but need to focus on axial
readings.

Out of phase in axial (180 degrees).

Tricky to check. But we will see mostly in phase readings across 3,6,9,12 o
clock readings around coupling.

All in phase.

Cocked Bearing:

Sort of misalignment. Depends upon how it is installed.

Thermal fitting?

Either inner race is titled and not installed correctly on shaft or outer race
is titled and not corrected installed on housing.
Because of angle in bearing, we will see 1X axial.
High 2X and 3X will confirm cocked bearing.

Phase angles will change continuously. 0,90,180,270 as bearings wobble if

its cocked on the shaft.

If its mounted correctly, it wont general axial readings, its just rotating.

If outer race cocked on housing, we will see high axial but not changing
phases.

If outer race is cocked, axial phase will be 180 side to side. (left to right or
up and down of same bearing)

If inner race is cocked, then 90 degrees from 3 to 6 or clock, 6-9 and 9-12
at same bearing.
Phase Comparison Table

Fault Type V-V Phase H vs V A-A Phase Notes / Key

Phase Behavior
(same
bearing)

Static 0° 90° Unreliable - High 1X in H

Unbalance &V
- H ≈ 2× V
- A low
- Ratio of H:V
same
- Vibrations
high at other
unbalanced
end

Couple 180° 90° Unreliable - High 1X in H

Unbalance &V
- H ≈ 2× V
- A low
- Ratio of H:V
same
- Vibrations
high at other
unbalanced
end

Dynamic 0°–180° 90° Unreliable - High 1X in H

Unbalance &V
- H ≈ 2× V
- A low
- Ratio of H:V
same
- Vibrations
high at other
unbalanced
end

Overhung 0°–180° 90° Unreliable - Highest 1X

Rotor in A
- High 1X in H
&V

Vertical In-phase – – - High 1X in H

Rotor radial-to- &V
radial; 90° - Vibrations
to opposite increase with
direction height

Offset 180° 0° or 180° - High 2X and

Misalignmen 180° moderate 1X
t in H & V
- Moderate
1X & 2X in A
- Ratio of H:V
not same

Angular 0° 0° or 180° - High 1X &

Misalignmen 180° moderate 2X
t in A
- Moderate
1X & 2X in H
&V
- Ratio of H:V
not same

Eccentricity – 0°–180° 0° or 180° - High 1X in H

&V

Bent Shaft – – 0° at 12/3/6/9 - High 1X &

on same 2X in A
bearing - Bend near
180° between center → 1X
bearings high
- Bend near
coupling →
2X high

Cocked – – A-A shifts 90° - High 1X, 2X,

Bearing per 3X in A
(Inner Race) 12→3→6→9→1
2

Cocked – – 180° - High 1X, 2X,

Bearing top↔bottom 3X in A
(Outer Race) or left↔right
System Dynamics
Introduction
Dynamics play a very important role. It affects us in health of the
machines, diagnose problems of the machines.
Need to understand natural frequency, resonance, mass, stiffness,
damping.

The dynamics of the machine affects the amplitude of vibrations, some

are attenuated, some are amplified because of it.

Because of dynamics of sensor and how its mounted, what we get in data
collector is affected.

Filters, integrators also have dynamics.

Sensors have single degree of freedom systems, having natural

frequencies around 30kHz.

We should be suspicious of high vibration as it may be amplified due to

natural frequency.

Every component can be broken down into three qualities: stiffness, mass
and damping.

We can model every machine and component like this.

Depending upon stiffness, mass, viscosity of damper, we may get different
vibration. It can be high or low frequency vibration.

How does this system affect vibrations?

Mass Effect
Increasing mass will tend to lower natural frequency as system will
bounce more slowly.

Decreasing mass will tend to increase natural frequency.

Stiffness Effect
Increasing stiffness will tend to increase natural frequency and vice versa.

Damping Effect
If we reduce damping, the wave/vibration will take a lot of time to die out.
If we increase damping, the wave/vibration will die out soon.

This was an example of single degree of freedom system.

Types of Damping
Damping results damping.

Example is bicycle pump, if I push and pull, in-out faster, it will resist
more.

1. Frictional damping
 Part rubbing against each other, Lubrication is used to reduce
frictional damping.

2. Hysteresis damping

Internal molecules provide source of damping by friction.

3. Viscous damping

Resistance to fluid flow. It is the most common type.

Lubrication provides viscous damping.

Fan pushing fluid, impeller pushing fluid, rotor turning in electric

field.

Damping
Undamped Natural Frequency
Undamped natural frequency means with 0 damping.

If there is no damping, nothing will stop body from keep oscillating forever.

But in reality, damping is always there.

Damped Natural Frequency

Damping causes loss of energy.

The oscillation will always stop or will never start.

Zeta is the ratio of viscous damping and critical damping.

If we keep increasing damping then at one point, body will stop moving.
Any increase will start oscillation, and any further decrease will stop body
from moving ever and it will be overdamped.

We just need enough damping so vibrations produced get immediately die

out.

The ideal situation is when wave comes as quickly as possible but don’t
oscillate at all. If we overdamp, it will come too slow. If we under damp, it
will oscillate a lot. So We need perfect ratio.

This magic ratio is known as zeta.

If actual damping is 0, then cv=0, and

 wd=wn,
 fd=fn

Critically Damped Systems

When it is critically damped, then cv=cc, and
No oscillation at all, just a simple decay.

Overdamped Systems
There is no oscillation at all.

Takes longer to decay compared to critically damped.

Damped natural frequency will be 0.

Underdamped Systems
Underdamped systems will oscillate at its natural frequency.

Consider damping ratio is 0.

Damping and Amplitude Ratio
Below graph shows relationship between damping ratio, amplitude ratio
and frequency ratio.

When damping ratio is 0, system will vibrate at natural frequency and will
have maximum amplitude.
When damping ratio is 1 or critically damped, system will have lowest
amplitude.

Check amount of amplitude with varying damping values.

It is the ratio of excitation compared to response, or vibration we are

putting in vs vibration we observe out.

When damping is 1, it means if we are putting 1mil of vibration, we are

getting 1mil of vibration response. Compared with when damping is 0.05,
when we add 1mil of vibration, then it will amplify to 100 mils.

We are exciting natural frequency using a driving force. The driving force
may be running speed.

Our main interest in dynamics is situation where system is underdamped

and causes amplification of vibration.
When we do our tests, such as run-up and coast-down, bump test, ODS
test, modal test, we are trying to find out what our natural frequency is
and get a sense of amount of damping we have got (or how much
amplification we have got).

Knowing that, we will try to either;

 Stop putting in vibrations at that frequency (correcting unbalance

etc), or
 We have to change the speed of the machine so running speed
doesn’t correspond to natural frequency (pump vanes doesn’t come
close to natural frequency etc), or
 We have to change structure.
o Either add damping (to shift amplitude down)
o Or change stiffness/mass (to shift natural frequency)

First challenge is to know causes of high vibrations, then find natural

frequency and amplification.

Phase relationships also change with change in damping ratio.

Same damping ratio will have phase on higher frequencies.

0 degree phase at low frequency and 180 degree phase at high frequency.

Rate of change of phase angle varies. Means phase doesn’t change with
same trend/ratio etc.

If a system is lightly damped, then phase readings will not change quickly
and stays near 0 until it changes abruptly near its natural frequency,
passing from 90 all the way to 180 degrees.

If a system is heavily damped, then phase changes relatively quickly.

Critically damped system will have almost linear curve.

We might be looking at system with moderate damping. Because if
critically damped, then no issues.

We are interested in how phase changes when we increase the speed of

the machine. We are basically exciting the resonance or the mode shape.

When phase is 90 degrees, it means that speed is the natural frequency.

Increasing speed increases phase lags.

SDOF (Single Degree of Freedom) System

Modelled as single mass, stiffness and a damper.

Freedom to move in one axis at one frequency.

But real systems are multi degree of freedom systems (MDOF).

They can move in multiple axis at multiple frequencies.

When we look at modal analysis or ODS analysis, we are more interested

in multiple degree of freedom systems (wobbling, twisting,…), means we
have to explore multiple natural frequencies.

We can break it down into multiple single degree of freedom systems.

Single Degree of Freedom in Detail

The dynamic force is combination of spring displacement or stiffness,
damping of velocity, force due to accelerated mass.

Differentiate displacement to get velocity.

Integrate velocity to get displacement.

Stiffness: Force required to compress or stretch spring.

Damping: Resisting force due to damping/friction/viscosity.

If pump handle is moved slowly, no issue,

If its moved quickly, movement becomes difficult.

This is because quick movement increases kinetic energy, gas
temperature. This will increase viscosity or the resisting force.

In case of Velocimeter, we got coil with magnet moving inside which

creates magnetic field creating damping effect and resists movement,
producing output in terms of velocity.

Mass: Mass at rest needs to be sped up (accelerated) to get in motion.

So the total force should be equal or greater than sum of stiffness,

damping and mass factors.

In case of accelerometer, we have crystal with mass on top and when

accelerometer tries to accelerate, the mass tries to stay still, we get
compression and expansion of crystal, F=ma, so output of sensor is
acceleration.

At low frequency, acceleration is low but displacement is high, stiffness

dominates.

At high frequency, acceleration is high and displacement is low, mass

dominates.
At all frequencies, velocity remains a good indicator.

Excited Natural Frequencies

Single degree of freedom has single natural frequency.

Multi degree of freedom has multiple natural frequencies.

Natural frequencies lie dormant. Consider example of tuning fork.

Tuning fork is example, it only produces sound when excited about natural
frequency.

We can excite natural frequency either by hitting it or by some driving

force.

If we pull the spring mass system, it will resonate at its natural frequency.
Damping will cause vibration to stop.

If we excite with constant low forcing frequency? Not by pulling mass but
by lifting up and down from red point starting from slow, going faster and
faster.
What will be the green wave’s behaviour? Will it always vibrate at its
natural frequency or some other frequency? How much amplitude would it
be?

Even if we are not seeing any physical spring mass system in machine, we
can represent it with it.

Consider the simulator below.

Spring is reducing amplitude of green signal.

If we increase mass of the system, it will still vibrate at same frequency

with increased amplitude.

If mass is increased, amplitude will increase, if mass is decreased,

amplitude will decrease.
What is affecting amplitude of green signal? Red signal is excitation and
green one is response to that excitation.

Lets change the forcing frequency.

If we increase the red frequency, we will see larger green signal.

Look at damper, its causing a delay.

Green signal is experiencing a lag.

Increasing frequency of red will further increase amplitude of green curve.

The spring is really stretching. Like bouncing!

Delay or lag is also increased.

They are no longer in phase.

Look at the critical frequency and phase curves and resonance points.

Phase angle (lag) has been increased to 90 degrees or 1 quarter of the

cycle.
As we pass the natural frequency, the mass amplitudes get lower and
lower.

Its not like green mass is always vibrating at natural frequency but as
response to exciting frequency.

If we increase frequency of the force further, phase lag becomes 180

degrees.
It will mean as if two mass is moving opposite to the forcing frequency.
Its just because of time delay.

The amplitude will become very low.

If we cross natural frequency, then we will get very low amplitude of

response compared with what we are putting in (exciting force).

Now lets change mass and stiffness of the system at fixed frequency.

As we increase mass, we decrease natural frequency.

If we increase stiffness, natural frequency moves farther.

If we increase damping, amplification will be decreased. Phase will be
increased.

90 degrees phase shift means we are at natural frequency but amplitude

is much lower.
In case of critical damping, almost smooth vibrations.

In case of lightly damped, we will get enormous vibration amplitudes.

A little excitation will give huge amplification.

We will give a little bit of vibrations and we will get a lot.

If is system is very lightly damped and are we doing run up test and phase
doesn’t change, but changes suddenly, you should know why.

Increase Frequency to Natural Frequency

Increase frequency until:

Lag between ex increases until it is actually 90 degrees.

Amplitude also increases to maximum.

Well Beyond Natural Frequency
Amplitude becomes minimum.

Lag becomes 180 degrees.

If forcing frequency is too low, the body acts as rigid body, and it appears
the stiffness is very high.

\\

We call this region, ‘Stiffness Controlled Region’.

Changing damping is not affecting anything much in this region and

system appears to be very stiff.
When we move slightly above natural frequency, the region will be
dictated by amount of damping. This region is called Damping Controlled
Region.

If there is less damping, vibration is more and vice versa.

This time, system looks to have weak stiffness.

If we move further up, stiffness appears to be very low.

This would also mean we are hardly moving the mass.

We are pulling this spring, but mass is moving little.

If we change stiffness or damping, nothing really happens.

This region is called ‘Mass Controlled Region’.

Bode Plots
Relationship between frequency vs amplitude and phase vs amplitude.

Lightly Damped

Heavily Damped

Bode plots are very useful specially in run up coast down tests, modal
analysis etc.
Based on shape of bode plot, we can assess damping and confirm whether
resonance exists or not.

Dynamic Response of Rotor

Convert spring mass system to a shaft system.