Back To Basics CMVA 4
Back To Basics CMVA 4
Back To Basics CMVA 4
ABSTRACT
With all the recent advancements in microprocessors and digital signal processing, there is a
tendency to simply look at pattern recognition (or signatures) of captured data in the frequency
domain and to make some quick decisions on root-cause problem solving. In some cases, even
the fundamental thinking (processing) on the data is taken over by the instrument itself.
Modern-day maintenance folks have become dependent on advanced technology to quickly
diagnose complex machinery problems by simply placing a sensor on a machine component and
awaiting a report of the fundamental problem along with a recommendation for fixing it i.e.,
balance, align, tie down loose parts, avoid resonance, or replace a defective bearing.
With all these functions now being viewed as somewhat automatic and accepted practice over
the past decade, it is easy to forget about the basics of the measurement and the large amount of
processing that has to be performed. For those more seasoned analysts in this field, there is a
recollection of the time when only a signal trace of the vibration level was available on a scope
or chart. A large amount of hand calculations were then required to correlate the information to
possible faults associated with specific machinery components and operations.
As the frequency data are derived from a transducer providing a raw time waveform electrical
signal of the vibration measurement, whether it be a displacement probe, velocity pickup, or an
accelerometer, it is worth going back to basics for a moment to see how this raw data can be a
big help in analyzing and pinpointing some fundamental vibration issues. Oftentimes, the raw
time information is only viewed as being a stepping stone to get to a quick problem resolution
and the data are actually discarded.
This paper is intended to show the value of retaining a sample of the collected time waveform
and to describe how the raw data should be properly collected and analyzed. This additional
information will help one to arrive at a solid PdM strategy for pinpointing and confirming
complex vibration problems with a high degree of confidence.
1.
INTRODUCTION
In recent years, there has been a bit of resurgence in the use of time waveform analysis in
condition monitoring and predictive maintenance. It had been avoided for years as modern Fast
Fourier Transform (FFT) processes have prevailed as being more simplistic and fast, as the name
suggests.
The analysis of time waveform data is certainly not a new technique. In the early days of
vibration analysis, time waveform data was viewed on strip charts and oscilloscopes and
2
frequency components were calculated by hand. The basic relationship between frequency and
time measurement is illustrated in the graph in Figure 1.
Note that dual differential cursor marks can show the approximate period (0.02701 sec.) and be
used to estimate frequency of the fundamental incoming signal. Calculation here from Hz to
CPM shows 2222 RPM, where an actual tachometer on the shaft confirmed the running speed at
2116 RPM.
Oftentimes, raw vibration sensor signals and patterns are not so clear in how to perform a quick
analysis. This is illustrated with the input from a sensor attached to a gearbox, as shown in
Figure 3.
2. DISCUSSION
Now that it is more evident that there is true value in the raw time waveform data, there are a few
fundamental concepts and considerations that need to be explored.
2.1 Time Waveform Considerations
Time waveform can be used effectively to enhance and to confirm spectral information in the
following applications:
While it is true that time waveform can be applied to a variety of vibration problems, some
situations show that spectral and phase data provide better indications on the source of the
problem. Two primary examples are:
Experience has shown that there are a number of machinery problems where time waveform
analysis truly reigns supreme over conventional spectral methods. These include:
There are also a number of areas where time waveform data proves to be reinforcing to
spectrum. These include:
Time waveform charts can best illustrate true peak values, phase relationships, transient
conditions, system impacts, damping, modulation, and beating.
Care must be taken when setting up parameters of the instrument for getting proper time
waveform data. Of special interest are the following measurement parameter settings:
Units of Measurement.
Time period being sampled.
Resolution.
Averaging.
Sampling window.
While some instruments do not allow the setting of the desired time period in terms of
revolutions or time, it is necessary to set up an equivalent frequency max (or Fmax) value. This
value is determined by the following formula:
Fmax (CPM) = (lines of resolution) x RPM / (# of revolutions desired)
Remember the inverse relationship between time and frequency in data collection, where the
lower the Fmax, the longer data collection time.
Assuming that lines of resolution are fixed at 1600, Table 2 provides applicable Fmax values.
Table 2 Fmax values for 8 and 10 shaft revolutions
MACHINE RPM
3600
1800
1200
900
300
100
2.4 Resolution
For detailed time waveform analysis, it is common practice that 4096 samples (1600 lines) are
used. This degree of resolution ensures that the data collected have sufficient accuracy and that
key vibration events are captured.
2.5 Averaging
In most data collectors, a type of averaging is performed during the FFT process. Unless
synchronous time averaging is employed (to be discussed later), the time waveform presented on
the screen will be the last average taken even if multiple averages are selected in the instrument
setup. It is normal therefore to take just a single average. Overlap averaging should be disabled
for time waveform analysis.
With spectral analysis, it is quite common to employ linear averaging, where each instantaneous
spectrum is added to the next and the sum contribution at each data point is divided by the
number of averages taken. This method is great for fault finding, and it is used in most
predictive maintenance programs. It is also a useful technique in averaging out random
background vibrations. Oftentimes, overlapping averages are used in FFT calculations to get
better frequency and amplitude representations from windowing sampling on the captured time
waveform.
Peak hold is sometimes referred to as averaging, but is not a true averaging technique. It simply
captures, holds, and displays the highest sampled peak value. This technique is quite useful in
viewing transient events and in stress and resonance analysis calculations.
Exponential averaging takes the most recent sample of data and weights it more heavily than
previous samples. It is most useful in observing conditions that are changing slowly with respect
to sampling time i.e. a steady-state process.
Time synchronous averaging (TSA) can be used to synchronize data acquisition to a particular
rotating component. This technique can be useful on gears where broken teeth are suspected to
assist in the location of the defective teeth relative to a reference mark. It is also useful on
reciprocating equipment to time and correlate events to a particular crank angle.
2.6 Windowing
A variety of windows can be applied to the time waveform prior to performing the FFT. The
purpose of these windows is to shape the spectrum and minimize leakage errors, thereby
providing a more accurate frequency determination (but sacrificing absolute accuracy of
amplitude). Some instruments allow these windows to be applied to time waveform data as well.
This would force the data to zero at the start and end of the time sample potentially losing data.
To eliminate this effect, a uniform (rectangular) window is recommended.
10
increased amplitude in 1X, 2X, and/or 3X RPM. If one takes a look at the acceleration
waveform and compares the information to the velocity spectrum, there are some methods to
distinguish one problem from the other.
The acceleration waveform from a misalignment will display regular, periodic spacing between
the major peaks and follow a nice pattern (i.e., one of three major peaks may be consistently
higher than the others with each shaft rotation). This is something that will be evident in a time
display and not so obvious in frequency. In addition, in the time waveform, peaks will normally
be less than 2 gs, indicating very little impacting. In the spectral display, the noise floor for
misalignment will be quite low, again indicating negligible impacting.
On the other hand, time waveform data from machines with looseness are characterized by an
irregular spacing between major peaks. In addition, there is no real pattern to the occurrences of
these peaks. They appear to be random in variation. In some cases, looseness will show impacts
in the time waveform exceeding 6 gs. These kinds of impacts also show a significant noise
floor in the spectral plot.
The relative phase angle between the 1X RPM and 2X RPM components actually determines the
shape (pattern) of the plot associated with misalignment.
11
12
13
The waveform pattern in Figure 14 indicates a non-repetitive pattern characteristic of nonsynchronous vibration. The vertical lines are spaced at 1X RPM.
14
Care must be exercised when determining symmetry of the time axis. 1X RPM markers are
available in most software programs and they should be used to avoid confusion.
At first glance the captured waveform in Figure 16 appears to have large impacts occurring with
somewhat similar spacing. The horizontal axis is scaled in time units.
15
An FFT captured on this machine provided the spectral display in Figure 18.
16
Next, lets compare this to a waveform exhibiting a beating sensation, as shown in Figure 20.
17
18
Note the spectral velocity plot shown in Figure 24. The vibration is less than 0.04 IPS, and this
data by itself would not warrant any further action.
19
Also, because of the nature of the measurement, it is important that the tachometer pickup be
aimed at the correct target on a reasonably steady machine speed, with little jitter.
2.9 Crest Factor
There has been considerable discussion so far on the ability to detect peak impacts and possible
defects and influences from certain machinery defects. One would be remiss in these discussions
20
without touching on the term Crest Factor to assist in quantifying the contribution of impacting
signals in a captured time waveform.
Crest Factor is defined as the peak amplitude of a time waveform divided by the RMS value.
(i.e., Crest Factor = Peak / RMS)
The main purpose of the crest factor calculation is to give the analyst a quick idea of how much
impacting is occurring in a waveform. Impacting is often associated with roller bearing wear,
cavitation, and gear tooth wear. Figure 27 provides an example of a waveform with a crest
factor of 3.
21
22
Crest Factor is a quick and useful calculation that gives the analyst an idea of how much
impacting is occurring in a time waveform. This is useful information that is lost for the most
part if one is only viewing a spectrum, as the displayed FFT data cannot differentiate between
impacting and random noise.
Impacting in a time waveform may indicate rolling element bearing wear, gear tooth wear or
cavitation.
As a valuable tool in time waveform analysis, Crest Factor is sometimes trended over time in
order to see if the amount of impacting in a machine component is increasing or not.
3.
CONCLUSIONS
It is hoped that the information conveyed in this paper will be utilized as an additional tool in
performing vibration analysis and assessing machinery performance in a condition-based
maintenance program.
Time waveform data has indeed proven to be an excellent analysis tool. It is not recommended
that it be taken on all measurement locations on a regular basis. This practice would add
significantly to the time required and the data storage requirements.
Table 3 helps to show when frequency, phase, and time data are best for collection and analysis
in arriving at root-cause faults of machinery problems.
Table 3 Where frequency, phase, and waveform data are best for analysis
Application/Problem
Unbalance
Misalignment
Resonance
Rolling Elements
Sleeve Bearings
Gears
Electrical Faults
Looseness
Flow Problems
Very Low Frequency
Cyclical Vibration
Variable Speed
Spectrum/FFT
X
X
X
X
X
X
X
X
X
X
Phase
X
X
X
X
X
Time Waveform
X
X
X
X
X
X
X
X
X
Because of uncertainty of phase shifts in integrating raw measurement data, it is best to use
transducer native units in conducting time waveform analysis usually acceleration.
As was stated earlier, no windowing or overlapping should be performed in collecting the
waveform sample. In fact, as we are looking for a snap-shot, 1 average will suffice.
23
To obtain adequate resolution, 4096 samples are usually enough (equivalent to a 1600-line
spectrum), and the time period should be sufficient to collect 6-10 repetitions of the event of
interest, or cycles of the shaft under review.
Time synchronous averaging can be used if, and only if, it is necessary to isolate data to the
particular shaft under review. This is particularly useful for looking for cracked gear teeth and in
looking for reciprocating machines and tying events back to a specific crank angle.
Transducer mounting must be firm, with no rocking or movement, to provide the required time
domain data for pin-point analysis work to be performed.
As a general rule, time waveform data should be captured for the following selected analysis
situations to enhance and support FFT data:
Time waveform data has proven to enhance and confirm observations in spectral data. It is not
intended as a replacement for anything, but as yet an additional tool for providing better analysis
and confidence in established PdM programs.
Analyzing patterns and following a rigid set of guidelines prevents the analyst from cutting
corners and skipping steps that might contain vital clues to the machinery faults. Looking at a
recorded waveform display and thinking about the orientation of the sensor can be a big help in
visualizing the motion in a pattern and identifying possible sources of the problem.
24
REFERENCES
1.
2.
3.
Machinery Vibration Analysis & Predictive Maintenance, Cornelius Scheffer and Paresh
Gidhar, 2004
4. Signal Processing for Effective Vibration Analysis, Dennis H. Shreve, IRD Mechanalysis,
Inc. 1995
5.
The Time Waveform, Stuart Courtney, Entek IRD International Inc., 1998
6.
Time Waveform Analysis, James E. Berry, Technical Associates of Charlotte, P.C., 1993
BIOGRAPHY
Dennis H. Shreve holds B.E.E. and M.Sc. E.E. degrees from The Ohio State University,
with specialization in high-speed data communications. He has 42 years of experience in
designing and developing electronics and software systems and leading projects for realtime industrial process monitoring and control applications. Over the past 23 years, he has
specialized in predictive maintenance (PdM) technologies and vibration detection, analysis,
and correction methods for maintaining machinery health. Dennis is certified by Technical
Associates and ANST as a Level III Vibration Analyst, and he is a Certified Maintenance
and Reliability Professional (CMRP). He is an active member of several professional
societies, including Vibration Institute, CMVA, SMRP, ISA, and I.E.E.E., where he has written several articles and
conducted public seminars. Dennis is currently employed with Commtest, Inc. as Senior Staff Engineer for the
Channel Partner Sales organization.