Summary

x1
1 annotation on 1 page by Mark Lynskey

Errors when averaging

• Averaging is used for standard Condition monitoring as it

allows the data to become more statistically correct.
#1
• However the process of averaging when applied to gE readings p. 34
actually damages the data – Why?
• Bearing defect frequencies can be looked at as random signals
SKF Acceleration Enveloping
and other Bearing Defect
Detection Methods

Presented to name/s
Presented by name of presenter
dd Month, yyyy
 Agenda

1. Introduction – Bearing Defect Detection

2. The ‘Sensor Resonant’ Technique

- The method & common pitfalls

3. The ‘Demodulation’ Technique

- The method & common pitfalls

4. SKF Acceleration Enveloping configuration

- The correct set-up in detail

5. Supporting Material

© SKF Group Slide 2 6 August, 2014

1
Introduction – Bearing Defect
Detection
 Introduction – Bearing Defect Detection
Facts:

• All Condition Monitoring Instrument vendors offer "special" measurements for

better early detection of defects in bearings and gears. There is no independent or
common standard such as the ISO standards for RMS velocity.

• Most (but not all) of these "special" measurements boil down to one of two
techniques:

- Sensor resonant technique

- Demodulation technique

• Because of too much prior art, neither of the two techniques are patented, and are
free for any vendor to adopt. Each vendor adds its own ‘spin’ (and name) to these
techniques.

• The following slides discuss SKF’s differentiation:

© SKF Group Slide 4 6 August, 2014

Differentiation #1: SKF’s credibility in bearings

• Trying to technically argue that "our technique" is superior to

"their technique” is like trying to argue that "our religion” is
better than "their religion”.
• There are ardent believers and preachers of both techniques across
the industry, and such argument is best avoided unless you want to
start some flag-burning.
• SKF’s view is that all these measurements work very well in
many applications, and not at all in some applications.
• SKF is the largest manufacturer of bearings in the world, and
so the applications we are asked to address are many and
varied. So to keep our reputation as a bearing manufacturer,
we must use a technique that reliably covers as many
applications as possible.

© SKF Group Slide 5 6 August, 2014

Differentiation #1: SKF’s credibility in bearings

• SKF is a mechanical engineering company, not an

electronics company. SKF understand how bearings behave
and therefore how they fail and their failure characteristics.
• Development of SKF’s implementation of Enveloping
(demodulation) was carried out in a practical research
environment. No other condition monitoring company has
real bearing test facilities such as those of SKF in The
Netherlands, India & China.
• SKF uses the technology of enveloping in their own factories.
It was first used in the 1960’s as a quality tool.
• SKF can and does use the data to help customers develop
the right bearing/machine solution.

© SKF Group Slide 6 6 August, 2014

Differentiation #2: Standardization in Band Filtering

• Within the enveloping technique the two main variables are

- Band pass filtering. What values of High Pass Filter and Low Pass Filter do
you use for the band pass stage?
- Detection type. Do you process the Peak or RMS signal?

• These choices are key to the effective implementation of the

technique in any one application.

© SKF Group Slide 7 6 August, 2014

Differentiation #2: Standardization in Band Filtering

• SKF's differentiation is the Band Pass Filter bands that we have

standardized on and successfully applied over thousands of
machines types and applications across the world for over 20
years.
• It matters not what filter values others vendors do - or do not -
use, the key is standardization of the BPF's, and their selection
guideline, across a wide range of applications a bearing
manufacturer must cover.

NOTE:
SKF were also able to patent this choice of filters when applied in a paper machine

© SKF Group Slide 8 6 August, 2014

2
The ‘Sensor Resonant’ Technique
Sensor resonant technique

• The natural resonance of the accelerometer is used to amplify

the tiny signals generated by the defect.

• These produce a scalar value only, which may ONLY be trended.

• Examples of this technique would be

- "Shock Pulse Measurement" from SPM (requires a special transducer)
- "G Spike Energy (GSE)" from IRD/Rockwell
- "Bearing Damage Measurement" from Pruftechnik
- "High Frequency Detection (HFD)" in the CMVA series Micrologs

© SKF Group Slide 10 6 August, 2014

Sensor resonant technique

At the early onset of a bearing fault

the defect will have sharp edges. As
the rolling elements pass over it they
give off high “spikes” of energy.

© SKF Group Slide 11 6 August, 2014

 Sensor resonant technique

Non-linear response
Amplification by the
sensor
Spikes from ‘overolling’ of a
defect are in this range

Reference

Linear response
Near zero amplification by
the sensor Bearing defect frequencies
are in this range
© SKF Group Slide 12 6 August, 2014
Sensor resonant technique : common pitfalls

• Susceptible to background noise influences

- Any random noise that occurs in the area of the sensor natural
frequency will enter the ‘amplification effect’.

• Poor on slow rotating machinery

- As speed and energy reduces, the frequency of the ‘overolling’ spike
tends to move outside the area of the sensor natural frequency, and
the ‘amplification effect’ is lost.

• If used as a one-off reading for analysis purposes, it can provide

wrong information (see next slide)

© SKF Group Slide 13 6 August, 2014

Common Pitfall: You can ONLY trend the value.

At the early onset of a bearing fault

the defect will have sharp edges. As
the rolling elements pass over it they
give off high “spikes” of energy.

At the defect grows the sharp edges will

be smoothed away and when the rolling
elements pass over it they now give off
lower “spikes” of energy.

© SKF Group Slide 14 6 August, 2014

Common Pitfall: You can ONLY trend the value.

• The scalar values for bearing defects can only be used for trending
purposes. They cannot be used for spot measurements as you have
nothing to compare with.
• The low figure could be due to no defects or a large defect!

Point of failure

Time
© SKF Group Slide 15 6 August, 2014
Common Pitfall: Sensor Specific.

• Two identical sensors from the same

manufacturer will have slightly different
natural or resonant frequencies fn. The
frequencies generated by the bearing
remain the same, so the “amplification
effect” will differ from sensor to sensor. So
fn the same sensor needs to be used each
time to avoid ‘false’ changes in the trend.
Frequency
• A larger discrepancy will be seen if two
sensors from different manufacturers are
used. They may have an identical 100
mV/g sensitivity and linear response, but
they will have different fn points. Hence
different values, so they cannot be
compared.
• SPM minimize this effect by tightly
fn controlling the bandwidth of fn around 30
kHz and matching their instrument
Frequency response accordingly. The drawback of this
approach is a ‘special’ sensor. Use a
‘regular’ sensor and the SPM technique still
works, but the values change.

© SKF Group Slide 16 6 August, 2014

3
The ‘Demodulation’ Technique
 Demodulation Technique

• The Demodulation technique involves using band pass filters and

"envelope" signal processing.

• It is used to EXTRACT tiny signals generated by the defect from the

background vibration.

• These produce a scalar value and an FFT which may be trended or

used for a one off analysis.

• Examples of this technique would be;

- "gE enveloping" from SKF
- "PeakVue" from Emerson CSI
- "Enveloping" from GE/Commtest
- “ESP” from IRD/Rockwell

© SKF Group Slide 18 6 August, 2014

 Demodulation Technique

Raw Signal

Bandpass
filter

Time Signal FFT

Rectifier

Lowpass
filter

© SKF Group Slide 19 6 August, 2014

 Demodulation Technique

• A key to SKF’s implementation of enveloping (gE) was standardizing on

4 band pass filter values across all its instruments:

- ENV1: 5 Hz - 100 Hz
- ENV2: 50 Hz - 1 kHz
- ENV3: 500 Hz - 10 kHz
- ENV4: 5 kHz - 40 kHz

• The SKF enveloping technique has proven to be very successful for

SKF over the past 20 years. There are numerous case examples
documented.

• It has been effective in many applications. When used in a multi-

parameter approach with velocity, acceleration and even acoustic
emission, there are few applications it cannot address.

© SKF Group Slide 20 6 August, 2014

 Demodulation Technique: common pitfalls

• Over the years the gE message has been misused and misinterpreted,
and some still say (even in SKF) that the technique does not work. This is
untrue and the reason for such doubt almost certainly lies in the
configuration of the gE measurement.

• The common configuration pitfalls of enveloping lie with incorrect

- Band Pass Filter selection
- FFT maximum frequency and line resolution
- Detection type and averaging
- Use of low frequency accelerometers

• The rest of this presentation will address in detail the correct

configuration – and why.

© SKF Group Slide 21 6 August, 2014

4
SKF Acceleration Enveloping
configuration
Demodulation Technique

The following slides take us through the following configuration

considerations:

• Band Pass Filter Selection

• Calculation of time block required (based on shaft speed!)
• Calculation of Fmax (based on running speed!)
• A quick way of working out Lines of resolution once Fmax is known
(to obtain time block length)
• Correct sensor selection
• Errors when averaging
• Selection of amplitude descriptor (detection type)
• Alarm limit calculation

© SKF Group Slide 23 6 August, 2014

gE Enveloping Filter Selection – Original slide that
explained filter band selection

The acceleration enveloping process groups energy related to

impulsive defects and excludes all others.
Bandpass filter

5 Hz - 100 Hz
Enveloping
Accelerometer 50 Hz - 1 kHz Detector

500 Hz - 10 kHz

5 kHz - 40 kHz FFT Spectrum

Felt Rolls, Bearings Bearings Gears

Filter #1 Filter #2 Filter #3 Filter #4

Frequency - Hz

© SKF Group Slide 24 6 August, 2014

gE Enveloping Filter Selection – Updated slide that
explains filter band selection

The acceleration enveloping process groups energy related to

impulsive defects and excludes all others.
Bandpass filter

5 Hz - 100 Hz
Enveloping
Accelerometer 50 Hz - 1 kHz Detector

500 Hz - 10 kHz

5 kHz - 40 kHz FFT Spectrum

Felt Rolls, Bearings Gears

Filter #1 Filter #2 Filter #3 Filter #4

Frequency - Hz

© SKF Group Slide 25 6 August, 2014

Band Pass Filter Selection

• A common demodulation guideline is that the HPF value of the

selected BPF needs to be at least 10X shaft speed.

• From the previous slide it can be seen that NO reference is

made to shaft speed so filter selection is independent of, and
NOT dependant upon, shaft rpm.

• Filter 3 is the filter used for bearing fault detection

- Bearing faults will be detected in the other filter bands (Filter 4 at the very
early stages of a fault occurring and filter 2 and 1 when the fault is very
well established or almost at the point of failure)

• Shaft RPM is however VERY important in setting up the

envelope collection correctly!

© SKF Group Slide 26 6 August, 2014

Why is Shaft speed important?

• Enveloping requires adequateTIME-LENGTH

• This is defined by the selected Bandwidth versus the chosen
RESOLUTION (LINES).
• Time-length = Lines / Bandwidth.
• Frequency (f) = 1/t (where t = time) so t=1/f. Knowing this we
can calculate one period of the lowest frequency of interest.
• Optimal time-length is 15X the lowest frequency reciprocal that
you are interested in. If you do not know the bearing details use
40% of 1X.
• The lowest frequency of interest will change with SPEED so the
shaft speed needs to be taken into account when calculating the
lowest frequency of interest

© SKF Group Slide 27 6 August, 2014

 Why is shaft speed important?
A Typical DGGB running at 1800 rpm

The lowest frequency generated by the bearing is the cage defect frequency at 11.39Hz
To calculate the time period for one cycle at this frequency we use T=1/F = 1/11.39
This works out at 0.087 seconds.

© SKF Group Slide 28 6 August, 2014

Why is shaft speed important?
The same bearing running at 50 rpm

The cage defect frequency is now 0.32Hz

To calculate the time period for one cycle at this frequency we use T=1/F = 1/0.32
This works out at 3.125 seconds. 35 times longer than at 1800rpm

© SKF Group Slide 29 6 August, 2014

Why is shaft speed important?

• To obtain the best quality data we should allow the lowest

frequency event occur 15 times during our collection period.
• In the previous example at 1800 rpm we need a minimum of
1.305 seconds of data (15 * 0.087)
• At 50 rpm we need a minimum of 46.87 seconds of data (15 *
3.125)

© SKF Group Slide 30 6 August, 2014

Calculating Fmax

• The calculation of alert and danger limit levels is based on

one of 4 Fmax settings;

1. 40 x N (where N is the running speed)

2. 20 X N (where N is the running speed)
3. Fixed range of 500Hz
4. Fixed range of 1000Hz

• Use your engineering judgement as to which one to choose!

© SKF Group Slide 31 6 August, 2014

 Calculating time block length

Rather than having to breakout pen, paper and calculator there is a

simple way to obtain the info required on lines of resolution (data
block length) – use the Microlog’s ANALYSER module

In the previous example at 1800 rpm we need a minimum

of 1.305 seconds of data. Having chosen an Fmax of
500Hz I can now adjust the Lines of resolution until I see
the required (or greater) acquisition time in the info window
(400 Lines only gives you 0.8 seconds – not enough!)

For the same bearing at 50 rpm we need a minimum of

46.87 seconds of data so for the same Fmax we now
need 25600 Lines of resolution

N.B If the required acquisition time cannot be met by

500Hz and 25600 Lines then the Fmax should be reduced
to either 40 or 20X running speed

© SKF Group Slide 32 6 August, 2014

Errors when averaging

• Averaging is used for standard Condition monitoring as it

allows the data to become more statistically correct.
• However the process of averaging when applied to gE readings
actually damages the data – Why?
• Bearing defect frequencies can be looked at as random signals
in our data block as they do not happen synchronously with
other rotational characteristics.
• Averaging exponentially reduces random frequencies with each
average so we are exponentially removing the bearing fault
frequencies with every average applied.
• In addition to this, the averaging process uses overlapping
which also has the potential to further “Randomise” the fault
frequencies.

© SKF Group Slide 33 6 August, 2014

Errors when averaging - overlap

• Overlaping allows us to reuse the last 50% of the signal block

as the first 50% of the signal in the second data block speeding
up data acquisition.
• This reuse of the signal can make the bearing signals appear
even more random and removed in the average process.

Data Block 1 Data Block 2 Data Block 3

© SKF Group Slide 34 6 August, 2014

 Selection of amplitude descriptor (detection type)

• If we look at a typical data block in the time domain, we can see some tall peaks due to short
duration impacts - potentially bearing defects?

• Using an RMS amplitude descriptor, then we sum the individual values and reduce the
measured level to a much lower calculated value. Individual peaks would have to grow
massively before they have an impact on the RMS value.

• Using a Peak or Peak-to-Peak then we are still not capturing the data correctly, as these two
descriptors are calculated from the RMS value (Pk =1.414* RMS) (Pk-Pk = 2 *Peak)

• The correct way to look at this data is with a TRUE Peak detector as used in the Microlog

© SKF Group Slide 35 6 August, 2014

 Alarm Limit calculation

Where can you find settings for alarm and danger when measuring gE?

The Atlas software has

an option to look at the
alert and danger levels
for any of the bearings
in the database

© SKF Group Slide 36 6 August, 2014

Limit calculation Spreadsheet

• CMC’s have produced a spreadsheet based on the original gE

calculations to allow you to generate alert and danger levels
• Simply enter the machine speed and bore diameter (metric or imperial)
and out pop your limits !
© SKF Group Slide 37 6 August, 2014
Low Frequency cut off

• Low frequency cut off is used in

“normal vibration” to remove potential
issues at low frequency caused by
integration errors, anti aliasing, poor
cabling etc. This is the reason why all
ISO standards start at 10 Hz.
• With Enveloping we have already
applied a high pass filter to 500Hz in
the enveloper, so there is no need to
set a low cut off in the enveloped
spectra.
• Beware: the XA Microlog series
ignores this setting from @Analyst on
an enveloping point. However, the
CMVA series do not - if you use both
Microlog series you may see
inconsistent readings if the LF cut off is
SET the LF cut off to 0 not set to zero.
© SKF Group Slide 38 6 August, 2014
Low frequency accelerometers

• It is widely accepted that when a shaft is rotating at a low

number of revs per minute, then a low frequency accelerometer
should be used.

FOR gE MEASUREMENTS THIS IS WRONG!

© SKF Group Slide 39 6 August, 2014

 Low frequency accelerometers

SKF supply 3 sensors designated as low frequency accelerometers

CMSS 793L / CMSS 793L-FM CMSS 797L / CMSS 797L-FM

CMSS 799LF

Frequency range: Frequency range:

Frequency range:
±5%: 0,6 to 700 Hz ±5%: 0,3 to 1 200 Hz
±5%: 0,6 to 850 Hz
±10%: 0,4 to 1 000 Hz ±10%: 0,2 to 1 600 Hz
±10%: 0,4 to 1 500 Hz
±3 dB: 0,2 to 2 300 Hz ±3 dB: 0,1 to 2 500 Hz
±3 dB: 0,2 to 3 700 Hz

As you can see the maximum useable range at the 3dB point is
well below the 10kHz range of band 3 so the majority of the
energy generated by a bearing defect will not be seen by these
accelerometers

© SKF Group Slide 40 6 August, 2014

Low frequency accelerometers

Band Data
gE Band 3 High pass filter capture range
no signals from within this range are taken into
the gE calculation

Useful range of sensor

© SKF Group Slide 41 6 August, 2014

 Standard sensors

Band Data
capture range

CMSS 2100
gE Band 3 High pass filter

Frequency range:
±5%: 3,0 to 5 000 Hz For Bearing Analysis
±10%: 1,0 to 9 000 Hz
±3 dB: 0,5 to 14 000 Hz Useful range of sensor
on machines with a
low rotational speed
Band Data
and gE band 3 set
capture range you must still use the
standard sensors!
CMSS 2200 / CMSS 2200-M8 gE Band 3 High pass filter

Frequency range:
±10%: 1,0 to 5 000 Hz
±3 dB: 0,7 to 10 000 Hz
Useful range of sensor

© SKF Group Slide 42 6 August, 2014

5
Supporting Material
Supporting Material

Spreadsheet for calculating warning and alert

levels

Atlas bearing database installer

© SKF Group Slide 44 6 August, 2014