You are on page 1of 14

Abstract and Biography

Technical Association of the


Pulp and Paper Industry
Optimized Application of
Feature Extraction Techniques
by Andre J. Smulders SKF Condition Monitoring

Abstract

Biography

Recent advancements in envelope enhancement


techniques as applied to acceleration and acoustics
emissions signals have led to new measurement
solutions for many vibration problems. This paper
discusses the theory of enveloping and how it is
implemented in practice. It presents a paper
machine case study that illustrates how a rolling
element bearing defect develops. Also some case
studies showing the strength of analysis in a
modulating environment will be discussed.
Measurement setups are very important for good
analysis and ease of recognition of symptoms. This
will be illustrated with a case study too.

ANDRE J. SMULDERS
SKF Condition Monitoring

SKF Condition Monitoring

Andre J. Smulders holds a master degree in


Electrical Engineering and a bachelor degree in
Mechanical Engineering. He worked in the
computer industry, the semi-conductor industry and
the sensor industry before joining SKF in 1981.
Has developed the Condition Monitoring
technologies as applied by SKF Condition
Monitoring today. He is the co-inventor of SEE
technology and holds patents in the fields of semiconductors, sensor technologies, measurement
techniques and in the field of signal analysis. He has
been a part time professor at a technical high school
for a number of years. He was involved of the start
up of SKF Condition Monitoring in 1989. He has
been involved in the development of techniques and
applications in the field of Condition Monitoring and
Quality Monitoring.

Technical Association of the


Pulp and Paper Industry
Optimized Application of
Feature Extraction Techniques
by Andre J. Smulders SKF Condition Monitoring
Copyright 2000 by SKF Condition Monitoring, Inc. ALL RIGHTS RESERVED

Abstract
Recent advancements in envelope enhancement
techniques as applied to acceleration and acoustics
emissions signals have led to new measurement
solutions for many vibration problems. This paper
discusses the theory of enveloping and how it is
implemented in practice. It presents a paper
machine case study that illustrates how a rolling
element bearing defect develops. Also some case
studies showing the strength of analysis in a
modulating environment will be discussed.
Measurement setups are very important for good
analysis and ease of recognition of symptoms. This
will be illustrated with a case study too.

Condition Monitoring An Historical


Review
The main reason for condition monitoring is to
prolong machinery life with the least overall cost.
There are several measurement parameters that
contribute to the evaluation of machinery health
such as vibration, bearing temperature, lubrication,
oil conditions, pressure that are measured on a
periodic basis to assess the long term prognosis of
the operating machine life. These machine condition
parameters as applied to a monitoring program, are
not the only factors in the attempt to achieve
maximum reliability with minimum cost. The
simplistic periodic visual inspection and the
experienced technicians ear are more often equally
important as diagnostic measures to augment the
predictive maintenance program and avert
catastrophic failures. An important aspect of the
data loggers programmed route is to assure periodic

SKF Condition Monitoring

visits to every machine based on a critical machine


priority that not only measures the assigned
vibration points, but to perform visual and acoustic
inspections as well. These machine conditions that
are subjective evaluations are entered as
maintenance notes to be reviewed later in the
machine history file. In a sense, the periodic data
logging sequence as defined by the predictive
maintenance schedule serves inherently as a
watchmans clock to assure a prioritized organized
visitation by experienced personnel to every
machine. The assessment of machine conditions,
operating performance and status of auxiliary
components valves piping, packing, loose bolts,
flange leaks are then considered in total for
recommended corrective maintenance actions.
Condition monitoring has always existed where
engine room personnel have felt, smelled or listened
to machine sounds as symptomatic of abnormal
machine performance. In these times of higher
speeds, design limit operations, complex processes
involving large populations of finite life machine
components, more automatic controls resulting in
minimum operations staff combined with
spiralling maintenance costs and extreme down time
production loss the need was created for warning
diagnostic systems employing hardware and
software sophisticated technology. These modern
condition monitoring systems now include new
measurement techniques which were untried and
unproven in the immediate past. These modern
methods are known to be viable as evidenced by the
case studies that are incorporated in the last section
of this article.

Technical Association of the Pulp and


Paper Industry Optimized Application
of Feature Extraction Techniques
T E C H N I C A L

The Ultimate Goal


In simple terms the main aim of the monitoring
system is to first significantly reduce unexpected
mechanical failures, thereby minimizing downtime
production losses. This objective has been achieved
generally by predictive maintenance programs that
rely principally on a periodic data logging schedule,
involving instruments measuring overall vibration
data. These instruments often operate in conjunction
with computer based programs that will trend the
historical data of each measurement point and also
allows the development of routes that can be
downloaded to field instruments. The next level of
the program goal is to recognize problems early
enough to schedule repairs with minimum disruption
to the operations.
Such maintenance decisions require a sufficient
assessment of the problem to assume a risk of failure
with the consequences of an unscheduled shutdown.
Todays predictive maintenance programs with the
many vibration measurement tools available and the
various analysis methods inherent in the associated
software, give the maintenance engineer the
opportunity to corrective measures that prolong
machine service life, improve product quality and
reduce production costs by running process speeds
closer to the design limits.
Although multiparameter measurements are
required for a complete assessment of operating
characteristic, vibration is the best measurement
parameter for evaluating machine dynamic
conditions that affect machine performance and
service life.
The effects of imbalance, misalignment, mechanical
looseness, bearing defects, ineffective lubrication,
shaft rubs are revealed as vibration characteristics
that are often identified by some spectrum signature.
The methods of feature extraction for problem

P A P E R

recognition have been developed and continue to be


refined for both manually derived and automatic
diagnostic decisions. The probable accuracy of
these maintenance recommendations that is derived
from these problem recognition methods, increases
with more available detailed information concerning
specific machine characteristics and its associated
mechanical components.
Oil condition monitoring provides an estimate of the
deteriorating lubricating properties that can
contribute to machine damage.
Viscosity changes, contaminants and metal particles
are some of lube oil detectable trends, that will over
time affect, the wear of bearing surfaces.
In addition to automatic vibration measurements and
data transposition to organized files of data trending
diagnostic analysis and new software extensions
allow for expert analysis. These software additions
to the traditional predictive maintenance software
include programs that scan machine historical data
with feature extraction algorithms to generate
symptom files. These files are compared against
resident diagnostic rules that are used to estimate the
probable machine failure modes. The expert system
then recommends maintenance actions based on
these severity estimates.

Traditional Vibration Parameter


Velocity
In the past field vibration measurements were
usually performed using velocity transducers that did
not require excitation. The electronic
instrumentation measured overall values and the data
was manually recorded. Vibration trends were
plotted manually to determine the machine health
status.
Velocity measurements remain today an important
measurable parameter since it is essentially related
to vibration energy.

Technical Association of the Pulp and


Paper Industry Optimized Application
of Feature Extraction Techniques
T E C H N I C A L

Overall vibration velocity measurements are often


compared to standardized alarm levels based on
accumulated experience. These velocity alarms are
constant levels applied over a wide frequency range.
Velocity is a parameter linear to vibration and is
proportional to sound pressure so correlated with the
sound impression generated in a machine
environment.
The more universal transducer in use today is the
piezoelectric accelerometer. The velocity
measurement parameter is obtained by simple
integration of acceleration.

Feature Extraction Techniques


For Optimized Analysis and Ease of Diagnostics
capabilities the strongest techniques known today is
Acceleration Enveloping (or Demodulation in
general as it can also be applied on other signals like
motor currents or pressure signals).
Enveloping addresses the problem of isolating small
but significant impulse perturbations that are
summed, during measurement, with larger, low
frequency, stationary vibration signals, such as
imbalance and misalignment. These small impulse
signals come from the accelerometer response to
impulsive forces from bearing race defects, from
roller flat spots, from gear teeth interaction.
Specifically related to a paper machine press
sections, these signals may come from felt joint
connectivity and/or felt dewatering anomalies.
Although normal FFT spectrum analysis separates
these signals into their fundamental and harmonics,
the individual amplitudes are often too small to see
above the instrumentation noise level. Because of
this low signal-to-noise ratio, these small spectral
components are not generally measurable in the
early onset of a bearing or other machine fault.
A small, narrow, repetitive impact signal, when
converted to the frequency domain, results in a plot

P A P E R

of small harmonic amplitudes with a frequency


separation equal to the repetitive rate. Compare the
different amplitude/frequency relationships between
a sinusoidal pure tone signal and a repetitive
impulse.
The impulse signal amplitude is proportional to the
pulse width (-t) and pulse cycle interval (T). The
smaller this ratio is that is, the narrower the pulse
width the smaller are the spectrum amplitudes.
This ratio is, of course, related to the width of the
bearing defect.
Initially, an accelerometer response signal is small in
amplitude and narrow in time as each ball rolls over
a newly developing fault. An acceleration spectrum
plot at this early stage of defect growth would
probably not show the defect as its amplitude is
below the dynamic range of the measuring
instrument. Vibration components identifying an
incipient bearing failure are then not seen in an
acceleration spectrum plot. However, enveloping
technology, now implemented in many dataloggers
and on-line systems that incorporate FFT analysis,
has proven to be an effective measurement tool
because it modifies the raw vibration signal so as to
enhance the rolling element bearing defect signal
and other comparable signal.

The Basics of Enveloping


The envelope method separates a repetitive impulse
from a complex vibration signal by using a band
pass filter that rejects low frequency components
that are synchronous with vibration.
Although there are signal enhancements that result
from structural resonances, the envelope method is
completely independent on local resonance to isolate
rolling element defect signals. This is very
important as resonance frequencies and the damping
at resonance are often not stable so not useful for a
trend type analysis.

Technical Association of the Pulp and


Paper Industry Optimized Application
of Feature Extraction Techniques
T E C H N I C A L

Filter criteria selection is based on suitable rejection


of the low frequency sinusoids while optimizing the
passband of the defect harmonics. This also creates
the possibility for separation of phenomenon. This
is illustrated in the following figure. The figure
provides the table of filter selections based on
rotational speeds and shows the optimal band of
analysis.
After filtering the vibration signal, the resultant
signal is enveloped by means of a circuit that
approximates the squaring process of the signal.
The enveloping process demodulates the signal
which approximates a squaring function. This
translates the signal in the frequency domain to a

P A P E R

baseband display of the repetition rate harmonic


components, where the component amplitude versus
frequency is equivalent to a sin x over x distribution.
These displays would only be seen if there are
repetitive impulse components in a part of the
overall raw vibration signal.
Another way of understanding this translation to
baseband is to consider the bandpass filtered signal
as only comprising the higher frequency harmonic
components of the repetitive impulse. When this
harmonic series is squared, sum and difference
components are created. The difference components
fold back into the analysis range while all of the
summed components are outside the analysis range.

Enveloping Settings Microlog/Multilog


PAPER MACHINE PRESS/FELT MONITORING
Filters

Enveloping Frequency

Speed Range

Analyzing Range

Felts/Press Rolls
5 100 Hz

0 50 RPM

0 10 Hz

ROLL BEARINGS
50 1,000 Hz

25 500 RPM

0 100 Hz

ROLL BEARINGS / GEARS


500 10,000 Hz

250 5,000 RPM

0 1,000 Hz

GEARS
5,000 40,000 Hz

2,500 RPM

0 10,000 Hz

High Pass Filter: 24 dB/octave Bessel


Low Pass Filter: 12 dB/octave Bessel
REMARK: In a application where gears are involved sometimes a lower envelope filter needs to be chosen
to suppress the noise from gearmesh frequencies that are commonly very dominant.

Technical Association of the Pulp and


Paper Industry Optimized Application
of Feature Extraction Techniques
T E C H N I C A L

Fundamental Properties of
Acceleration Enveloping/
Demodulation/Rectification
This feature extraction technique has a number of
principles advantages and properties that make it
ideal for signal extraction of non-sinusoidal signals
and signals that are modulated by some carrier
phenomenon.
1. SELECTIVE FILTERING so excludes specific
sinusoidal signals.
2. DISCRIMINATION OF PHENOMENON by energy
estimation in a specific selected frequency band.
3. PULSE ENHANCEMENT VERSUS SINUSOIDAL SIGNALS.
Energy estimation focuses on peak phenomenon
with correlated phase characteristics versus
wavy type phenomenon.
4. SIGNAL-TO-NOISE IMPROVEMENT. An energy
estimation enhances localized energy,
concentrating FFT distributed peaks into its
basebands.
5. Speed varying compensation as small phase shifts
during rotation (non-constant rotational velocity)
will be averaged-out.
6. INSTANTANEOUS SYNCHRONOUS TIME AVERAGING.
Bringing energy to baseband frequency
components enables the time record to be longer
and so inherently does better synchronous
averaging.

Preconditions for Optimal Enveloping

P A P E R

2. Pre-filtering with Constant Time Delay filters for


good Peak reproduction.
3. Large bandwidth for optimal summation of
energy.
4. Signal source separation by Optimal Pre-filter
selection.
5. Time domain analysis so extraction is done
without separation of coherent frequency
components.
6. Low pass filter selection after Enveloping for
rejection of Out-of-Band components.

Conclusion
The acceleration enveloping technique is emerging
as a very practical measurement tool for assessing
initial problems associated with bearings, rollers,
and felt rotation. The very low speeds at which
these measurements occur are often at sensitivity
limits of transducers and electronics. In the past,
synchronous time averaging over very long intervals
was required to isolate problems to a particular roll
by establishing external trigger references.
Enveloping has proven its capabilities to extract
impact force signals developed by roll eccentricity,
flat spots, rolling element bearing defects and many
other impulse type or modulating type signals.
Although enveloping is not the panacea for
diagnosing all machine problems, it is proving to be
an adaptable and effective measurement method in
the tool box of analysis techniques.

1. Sufficient signal-to-noise ratio in the measuring


chain before Enveloping is performed.

Technical Association of the Pulp and


Paper Industry Optimized Application
of Feature Extraction Techniques
T E C H N I C A L

P A P E R

Case Studies
BEARING TEST RIG
DEMONSTRATION
Figure 1.
Standard Velocity Measurement
with defective bearing.
Although bearing defect
frequencies noticeable no clear
indication as still many other
frequency components are of
the same level.

Figure 1.

Figure 2.
Zoomed Velocity spectrum with
rotational components visible
but no significant bearing defect
pattern.

Figure 2.

Technical Association of the Pulp and


Paper Industry Optimized Application
of Feature Extraction Techniques
T E C H N I C A L

P A P E R

Case Studies
BEARING TEST RIG
DEMONSTRATION
Figure 3.
Enveloped Acceleration
showing a clear discriminative
spectrum of an Inner-race
Defect Pattern.

Figure 3.

Figure 4.

SEE spectrum (Enveloped


Acoustic Emission spectrum)
also showing the bearing defect
pattern as indicating friction
(progress of wear). The extra
sidebands around the bearing
defect modulation frequency
peaks indicate a modulation by
a low-frequency phenomenon
likely uneven coupling loading.

Figure 4.

Technical Association of the Pulp and


Paper Industry Optimized Application
of Feature Extraction Techniques
T E C H N I C A L

P A P E R

Case Studies
BEARING DEFECT DEVELOPMENT
ON A DRYER FELT ROLL
Figure 5.
Trend Plot of the Standard
Velocity Measurement. No
indication of a bearing defect
visible.

Figure 5.

Figure 6.
Velocity Spectrum showing a
number of harmonic patterns
but no clear indication of an
Inner-race Defect Pattern.

Figure 6.

Technical Association of the Pulp and


Paper Industry Optimized Application
of Feature Extraction Techniques
T E C H N I C A L

P A P E R

Case Studies
BEARING DEFECT DEVELOPMENT
ON A DRYER FELT ROLL
Figure 7.
Trend Plot of the Acceleration
Enveloping Measurement.
Good indication of a bearing
defect development.

Figure 7.

Figure 8.
Enveloped Acceleration
Spectrum showing a clear
discriminative spectrum of an
Innerrace Defect Pattern.

Figure 8.

Technical Association of the Pulp and


Paper Industry Optimized Application
of Feature Extraction Techniques
T E C H N I C A L

P A P E R

Case Studies
OPTIMAL MEASUREMENT SETUP
ON DRYER CAN
Figure 9.
Spectrum Plot of the
Acceleration Enveloping
Measurement. Although the
bearing defect is visible the
pattern is not extremely clear.
The measurement
TIMELENGTH was too short.
This is defined by the selected
Bandwidth versus the chosen
RESOLUTION (LINES).
Timelength = Lines / Bandwidth
Optimal timelength is 10 15X
the time for one shaft rotation.

Figure 9.

Figure 10.
Time plot belonging to Figure 9.
The measurement Timelength
does not contain sufficient
revolutions of the shaft to built a
clear spectral pattern.

Figure 10.

10

Technical Association of the Pulp and


Paper Industry Optimized Application
of Feature Extraction Techniques
T E C H N I C A L

P A P E R

Case Studies
OPTIMAL MEASUREMENT SETUP
ON DRYER CAN
Figure 11.
Spectrum Plot of the
Acceleration Enveloping
Measurement. The bearing
defect is clearly visible with a
clear sideband pattern so
indicative for an innerrace
defect pattern. The selected
measurement Timelength is
optimally chosen.

Figure 11.

Figure 12.
Time plot belonging to Figure
11. The measurement
Timelength does contain
sufficient revolutions of the
shaft (modulation) to built a
clear spectral pattern.

Figure 12.

11

Technical Association of the Pulp and


Paper Industry Optimized Application
of Feature Extraction Techniques
T E C H N I C A L

P A P E R

Case Studies
PRESS SECTION FELT ANOMALY
WITH MODULATION DRIVE TRAIN
PATTERN
Figure 13.
Time plot indicating the Felt
repetition pattern (see also
Figure 14) modulated by a
DRIVE TRAIN control loop
problem.

Figure 13.

Figure 14.
Zoomed Time plot indicating
the Felt repetition pattern.
These patterns are indicative of
uneven dewatering
characteristics in the felt.

Figure 14.

12

Technical Association of the Pulp and


Paper Industry Optimized Application
of Feature Extraction Techniques
T E C H N I C A L

P A P E R

Case Studies
PRESS SECTION FELT ANOMALY
WITH MODULATION DRIVE TRAIN
PATTERN
Figure 15.
Spectrum Plot of the
Acceleration Enveloping
Measurement. The FELT
pattern is clearly visible. The
sideband pattern indicative for a
modulation pattern becomes
clearer after zooming (see
Figure 16).

Figure 15.

Figure 16.
Zoomed Spectrum Plot of the
Acceleration Enveloping
Measurement.
The modulation caused by the
drive train driving the Fourth
press is clearly positioned
around the spectral Felt Pattern.

Figure 16.

13

You might also like