© 2007 by François Gagnon
Calibration issues in a condition monitoring context have been a point of contention for as long as anyone can remember. Since a high percentage of vibration measurements are made for comparative purposes, the industry vendor’s main tenet has been to altogether avoid such calibration issues, relying on the one hand on a contracted “annual” service or on the other on an “on the spot” handheld shaker calibration adjustment.
The rationale relies on the largely correct understanding that amplitudes may be of limited significance (in comparison to frequency contents) and on saying that the spectral contents comparisons were still valid in spite of instrument drift, environmental effects and transducer error (mounting and calibration) when taking other measurement inequities and control factors into consideration.
Calibration relies on a comparative to a known standard, and the given comparative at any moment in time can be “qualified” in terms of accuracy and precision. Others might apply but these two remain our principal concerns. Accuracy is itself the comparison to a reference: are we close or far from the required value? Precision has more to do with “grouping”: are we consistently near, on reference value target, or consistently far? Are we “all over the map”?
Drift occurs slowly, so even years of accumulating calibration error will leave mostly significant and comparable data, so the very nature of the periodic data collection task has meant a very matter of fact attitude on most users’ part. Since absolute value measurements are not USUALLY a requirement within vibration-based predictive maintenance. There are exceptions, such as the nuclear industry. Otherwise, we accept a certain error percentage as inevitable and therefore tolerable. For a normal data collection routine performed by a seasoned practitioner of data collection, an eight hour day can in some instances mean as many as 800 FFT spectra accompanied by 2200 overalls, a number of high frequency demodulated spectra and undoubtedly some process parameters (temperature, pressure, load, amperage) for correlation purposes. This example supposes a relatively compact and clean workplace where the close to seventy monitored machines covered in the span of one day are close to one another and are all four-bearing machines (driver and driven rotor) as opposed to complex (such as geared or variable speed) machines.
In the course of finishing that day’s work, the analyst or person assigned to data collection must perform a rapid fire sequence of sub-tasks that we can list as:
- applying the transducer (preferably clean) magnet / torquing the accelerometer in place
- avoiding transducer-banging shocks in doing so / ensuring target cleanliness
- waiting for autoranging to take place (or risking an error if this function is bypassed)
- collecting the overall parameters and the relevant FFT spectra and/or waveform
- rejecting improper / spurious signals or data
- choosing and adjusting a zoom FFT on certain frequencies of interest (better resolution FFT, varying waveform digitalization density, performing supplemental tests as needed)
- gathering demodulated information on a “need to know” or alarm basis (implying the proper setting of high-frequency high-pass or bandpass filters for the specific machine under review) / some will prefer to use the Spike Energy / Acceleration Enveloping / PeakVue / Shock Pulse
- for the previous, many will gather that High-Frequency / Ultrasonic data at a longer interval than the main parameters; our opinion? Always!
- getting relevant process data when and where pertinent or available
Finally, the personnel assigned to data collection must do all of the above in between the time spent walking to and from the various machine points.
To acquire relevant data, the measurement locations must remain unchanging (as much as possible) from one data collection interval to the next. For this purpose, it is customary to mark or paint the target area. Permanent mounted targets (brass, stainless, other) are also sometimes used. If personnel changes occur in the course of repeated collections, a wider measurement fluctuation is expected. A good and well-trained team aware of all issues can keep the variations within a 5% to 10% range for overall values. A lesser team could obtain wild fluctuations reaching percentages in the order of 30% to 50%. Most errors perpetrated in spectral data would arise from transducer or magnet mounting rocking, interfering contaminants, etc.
The solutions sought to compensate for such field measurement inequities have been the quick-connect, permanently mounted transducer coupling, or fully prepared stud mounting lodging. The use of mounting blocks, often referred to as NATO blocks, has also been popularized in certain circles.
Given mostly unavoidable measurement uncertainties arising from personnel error and field conditions, the calibration error has been deemed small and even negligible in relation to other factors affecting the readings. After all, given a nominal transducer output value (within 1% of reality), an undamaged cable and a reliable instrument (perhaps calibration drift has reached 1%), the accelerometer mounting and mounting repeatability issues remain blatantly dominant over calibration, introducing a precision issue as opposed to an accuracy issue (as defined above).
Whenever the instrument / sensor sets or the persons collecting data in a specific area exceed one in number, the errors can become divergent (imagine one measurement coming in at an inordinately low value, and another coming in high). In such cases, tightening precision through calibration will deliver much better results. Again, a precision issue where the values obtained become inconsistent.
Most instruments and their attachments are entirely factory sealed and leave little to be calibration adjusted for the end user. Digital instruments allow for adjustment of the exact transducer output calibration (name plate actual value) as opposed to the nominal or generic value. This is a small improvement and will yield improvements in the order of 1% to 2%. If several accelerometers are used in conjunction with a variety of data collectors to amass relevant condition monitoring route data on the same machines, which also implies many different persons performing the same collection task in turn, a tightening of the accelerometer discrepancies is highly recommendable.
The matter is further complicated by official calibration requirements versus occasional verification. A standard demanding traceable references back to the National Institute for Standards and Technology (NIST) makes for a very expensive calibration and one where few end-users are compliant. Vendors do rely on such standards and abide by their terms, but a predictive maintenance program rarely does save in ultrasensitive areas. Once the calibration equipment is compliant and traceable to NIST, the calibration performance should follow an accepted standard. Many sources recommend MIL-STD-45662A (qv), but the military have since realigned to ISO 10012-1 or for calibration workstations, ISO 16063-21 (there are others).
As a rule of thumb, instrumentation used to calibrate field measurement devices calls for one order of precision improvement. This means calibration references should be capable of 10 times more frequency or amplitude precision than the devices to be calibrated.
A shaker table could possibly be checked for total vertical displacement using gage blocks (precision 0.000 1″ or 2,5 micrometer) and for frequency precision through an oscilloscope. Would this satisfy requirements? No. The displacement of the shaker table would need to be maintained at a one Mil level (one thousandth of an inch or 2,5 micrometer) over the entire calibration-test frequency range to make the previous suggestion acceptable. Such a displacement would generate altogether unreasonable velocities for frequencies above 5000 or 6000 CPM (when approaching 100 Hz).
For accelerometers using a separate charge amplifier, vendors have vaunted the merits of rapid calibration checks and adjustments through the use of a small, hand-held calibrated shaker and adjustment of the charge amplifier input/output ratio. Since a variation in cable length considerably alters calibration for small, low output accelerometers connected to an external charge amplifier (while doubling the probability of cable mishap due to the presence of two cables), the procedure was and remains a necessity to amass valid data for this type of equipment. For high precision measurements using magnet-mounts on NATO blocks, immediate and long term repeatability of measurements has left something to be desired in spite of being one of the few immediately and repeatedly calibrated field measurements offered to the vibration professional.
In this writer’s opinion based on field experimentation, a precision measurement repeated for validity over a short time span tends to show variations in calibration or precision from one measurement to the next, having led to questions about the higher drift of adjustable calibration when compared to “fix” or settled calibrations.
For permanently installed transducers, practice has typically only called for measurement calibration validity when the data seems to be entirely spurious or when the transducer fails to send a signal. Quick hardware replacements have been more readily implemented than transducer verification.
Accepted current practice involves returning the data collector / analyzer to the vendor once a year at a set date for a full instrument calibration procedure.
In a context such as recognized / acceptable NDT recommended practice, an on-site calibration test rig may be necessary for any application involving life or health affecting systems. In that category, any site dealing with air transport (helicopter, aircraft, space shuttle, etc) essential systems qualifies.
Any system for which failure would result in catastrophic environmental conditions (the nuclear industry being one such example) also would require on-site calibration capacity. We nuance “catastrophic” from “merely” hazardous based on potential lethality.
All other sites where mechanical and/or electrical systems are monitored for purposes of maintaining corporate profitability through continued production can merely rely on recommended vendor practice.
Some writers comment on the inability of some programs to capture and identify a developing problem due to instrument calibration problems. While it may be possible for an undetected instrument or sensor failure to permit the previous to occur, it is unlikely that a mere calibration problem is the cause for a monitoring program being defeated by mechanical or electro-mechanical component deterioration.
Repeated observations have led to the conclusion that a number of problems are far more likely to be at the source of detection inequities. Improper training (or lack thereof), unprocessed or insufficiently processed data, improper measurement techniques, incomplete measurement routes, omitted measurements, obsolete instrumentation, unsuitable measurement programming and failure to report detected changes or problems are far more likely causes. Unadjusted program collection following a “missed opportunity” also resonates as a possible cause.
In concluding, for the practitioner, what is the best approach?
a) Comparative measurement of all the sensor’s to be used on a specific job (back-to-back is nice, but usually not readily feasible, so we’ll settle for side-by-side).
b) Identification of amplitude sensitive tasks: standard conformance, acceptance testing (really only an issue when hovering near the limit), ultrasensitive applications (tiny observable values, nuclear industry, semiconductors, microsurgery environments, etc)
c) Reliance on true values, such as those obtained for high-density waveform or extracted from interpolated FFT spectra; alternately, it would be advisable to have Flat-Top windowed FFT to obtain the amplitude values from spectra, but the interpolated Hanning windowed spectra will generally be far more practical or feasible due to frequency peak separation concerns.
Addendum
It goes without saying that consultants should expect the unexpected by being fully prepared for any contingency by having ALL instruments calibrated to the best standard at all times, with occasional side-by-side comparisons executed regularly. The concluding suggestions apply far more to the in-plant practitioner than the visiting expert. Should any dispute arise when performing acceptance testing of newly arrived equipment due to certificate requests and the like, keep in mind that the threshold established as a LIMIT for a fully-loaded, worst-case scenario machine almost psychologically implies that we would be some distance from the “Reject” line. In stringent applications, the operative concept becomes how little will be tolerated at the beginning of service, so “close to the limit” also becomes more palatable. It is an unwritten rule developed from reactions and impressions expressed or demonstrated in many meetings and report presentations over the years.
© 2007 by François Gagnon
