How Sensor Mounting

Affects Measurements

Written By
David A. Corelli

pcb.com | 1 800 959 4464

 HOW SENSOR MOUNTING AFFECTS
MEASUREMENTS
DAVID A. CORELLI
IMI SENSORS

How Sensor Mounting Affects Measurements

Abstract: Sensor mounting can significantly affect both overall vibration and FFT (Fast Fourier
Transform) data. This paper describes the differences and confusion in overall measurement techniques and
shows how frequency response can affect these measurements. The frequency response of some common
mounting methods, such as stud, 2-rail magnet, and flat magnet are measured under controlled laboratory
conditions and the results presented. The laboratory data is then correlated with actual machinery data. The
paper also shows the dramatic effect that mounting has on commonly used high frequency overall
measurements such as Spike Energy™ and PeakVue® that are used for early warning of bearing and gear
faults.

Note: Definitions are included at the end of the paper.

Overall Vibration Level - We often hear the term overall vibration level but what does it mean? It is
certainly not a well defined term and one can make the case that it is not defined at all. There are many
ways to measure overall vibration levels and many of today’s vibration data collectors can be setup to
measure overall vibration levels using different techniques. Thus, it is hard to define or compare acceptable
overall levels for machinery unless the measurement method is well defined. Below are several terms that
describe overall vibration measurements.

• Amplitude
• RMS or true RMS
• Derived or calculated peak
• True peak
• “Full” bandwidth analog
• Filtered analog
• Band limited overall (computed from the FFT)

Measures - The typical measures used for overall vibration measurement and analysis (displacement,
velocity, and acceleration) are useful for detecting different types of faults. The first three measures listed
below are actual physical quantities that can be traced back to a standard, typically through NIST in the
United States. Overall HFE (High Frequency Energy), on the other hand, is not a physical quantity, is
measured using data outside the calibrated range of the accelerometer, and is thus not traceable. This adds
another level of complexity not only in trying set vibration standards or limits for HFE measurements but
also in trying to compare results.

• Displacement is used to measure very low frequency faults such as unbalance in slow speed
machines like cooling tower fans. Unfortunately, this can be difficult to measure without non-
contact (proximity) probes. It is also used to monitor relative shaft movement with fluid film
(sleeve type) bearings using non-contact probes.
• Velocity is a primary measure used in most condition monitoring programs to monitor “Balance-
of-Plant” faults such as unbalance, misalignment, and looseness.
 • Acceleration is typically used to detect higher frequency faults such as gear mesh and broken
rotor bars.
• High Frequency Energy (HFE) uses data collected above the calibrated range of the
accelerometer, including data collected at sensor resonance, and provides early detection of high
frequency and impulsive faults such as bearing defects, gear defects, and loss of lubrication
(metal-to-metal contact).

GM Vibration Standard - I often hear vibration analysts complain that measurements made with new
sensors don’t agree with the same measurements made with older sensors. They also are confused why
another analyst is getting different results than they are when they are seemingly making the same
measurements, particularly with overall measurements. To help address this problem, the General Motors
Vibration Standards Committee defined “Band-Limited Overall Amplitude Acceptance Limits” to ensure
consistency and repeatability of results in their Vibration Standard for the Purchase of New and Rebuilt
Machinery and Equipment. Band-limited overall values are computed from a frequency spectrum (FFT)
that must be within ±5% of the calibrated response of the accelerometer, including sensor mounting effects,
over the selected frequency range. Further, when looking at the total energy in a peak, “a minimum of 5
lines of resolution must be used and the peak must be centered in the band.” The following equation is
defined in the GM Standard to compute the band-limited overall when a specified Hanning window is used
(the 1.5 factor).

When this method is used to determine the overall vibration level, it is much more repeatable between
analysts and “transportable” since the band-limited overall is calculated over a well defined frequency
range and from a calibrated spectrum. However, a problem often arises when the analyst does not realize
that their sensor mounting method is altering the frequency response in the selected frequency range so the
spectrum is actually not calibrated.

HFE Problems - The problems of accuracy, consistency, and repeatability of data is worse when making
HFE measurements because overall HFE levels are determined from data collected outside the calibrated
range of the accelerometer, over a frequency range where sensor mounting dynamics affect the
measurements, and there is no physical quantity to trace it back to. Additionally, different accelerometers
have different natural frequencies and amplification factors. Since HFE measurements include data
collected at the sensor’s natural frequency, this further exacerbates the problem. If this isn’t bad enough,
many if not most industrial accelerometers have built in low pass filters to reduce the amplification effects
(to reduce saturation) at the sensor resonance, which also changes the sensor’s high frequency response
(above the specified range of the sensor). In the case of HFE, all bets are off on accuracy or
transportability. In fact, who is to say what is accurate since there is no physical quantity to trace it back to?

Mounting Effects - Frequency response calibrations were run on two IMI Sensors, a Model 603C01 Low
Cost Industrial Accelerometer and a Model 622B01 Precision Industrial Accelerometer. The 603C01 has
about a 25k Hz natural frequency and a 2-pole internal filter while the 622B01 has about a 30k Hz natural
frequency and a single-pole filter. Frequency response tests were run using step sine analysis on an NIST
traceable accelerometer calibration system. The frequency response was tested with the following sensor
mounts: stud, 35# pull flat magnet, 35# pull 2-rail (curved surface) magnet, ¼ twist mount, and 4” SS
Probe (stinger). A typical test setup is shown in Figure 1. The lower sensor in Figure 1 is a back-to-back
calibration standard. Figure 2 shows the stainless steel probe mounted to a 603C01 accelerometer. The
results of the tests are shown in Table 1.

Since the frequency response of most industrial accelerometers are specified at their ±3dB point, it can be
seen that the mounting method dramatically changes the calibrated range of the accelerometer system
(sensor and mount). The flat frequency response of the 603C01 is reduced from about 12.5k Hz with a stud
mount to about 3554 Hz with a 2-rail (curved surface) magnet under ideal conditions. It will be worse on a
curved surface on a machine. The useful frequency range of the sensor with the stainless steel probe is only
750 Hz. When making overall measurements, significantly different values will be recorded with these
different sensor mounting configurations and the analyst will probably not realize it.

Centrifugal Compressor and Traceable Measurements: The above example clearly shows how the
sensor mount can affect the measured data. But how much variation is actually encountered under normal
day-to-day measurement circumstances due to the different sensors and typical mounts?

Data was collected on the centrifugal compressor shown in Figure 3. The acceleration and velocity spectra
collected with the various sensors and mounts are shown in Figures 4 and 5. Even though these are
seemingly low frequency measurements, i.e., well within the specified range of the accelerometers, there
are major differences in the spectral data in both acceleration and velocity. This is due to the effects of the
sensor mount. For the 2-rail magnet, twist mount, and stinger, the differences are significant. This clearly
shows that the mounting methods typically used in day-to-day data collection can have a large effect on the
accuracy of routinely collected data. The plots also show the overall level as calculated from the spectrum.
In velocity, it ranges from 0.0138 ips stud mounted to 0.0229 ips with the probe or a 65.9% difference. If
the probe data is not considered, the difference is still 37.7%. For acceleration, the numbers run between
0.510 g to 0.983 g for a difference of 92.7% or almost 2 to 1. These are all huge errors and as the
frequencies used in the overall level calculation goes higher, they get worse.

Combustion Air Fan and HFE Measurements - The data shown in Table 2 summarizes the overall Spike
Energy™ readings (an HFE measurement) taken on a combustion air fan using an EntekIRD dataPAC™
1500. The measurements were made with 1 kHz, 2 kHz, and 5 kHz High Pass (HP) Corner filters. The Low
Pass Corner frequency of the data collector is fixed at 65 kHz. Thus, the data included in the overall Spike
Energy™ measurements includes frequencies that are well outside of the specified range of the sensor
including data near the sensor resonance. The table is sorted by Overall Spike Energy™ (gSE units) from
highest to lowest. There is a 46 to 1 difference between the highest and lowest readings! In general, the
sensor with the highest frequency response and best mount has the highest values. That does not mean,
however, that the other readings are wrong. They are just different. Remember, there is no physical
quantity to trace these readings back to.

It should be noted that magnetically mounted accelerometers are routinely used in condition monitoring
programs using portable data collectors and that HFE measurements are generally collected as part of the
route data. These measurements are effective in early detection of bearing problems even though the HFE
is attenuated. As long as the sensor, mounting method, and location are consistent and trended, the
measurements are effective; however, they generally can’t be compared to readings taken by someone else
using a different sensor, mounting method or data collection system.

HFE Measurements – Even though there is no traceability for HFE measurements and it is nearly
impossible to get the same results as someone else, why are they so popular. The simple answer is they
work very well in identifying certain types of faults. The user just needs to know how to use them. These
are typically high pass filtered measurements that are very sensitive to spikes in the data, as is typical with
rolling element bearing faults. Since the measurements are not traceable, the data should be trended. And, it
is up to the analyst to determine what amplitudes are acceptable and which are not based on historical data.

“The most meaningful use of Spike Energy™ is to trend gSE readings along with
velocity and acceleration readings.” 1

“[PeakVue®] is a powerful complementary tool that can detect a range of faults and
problem conditions that techniques such as vibration analysis alone might miss under
certain conditions.” 2

The data in Figure 6 was taken using the IMI Sensors Echo® Wireless Vibration Monitoring System on a
motor driving a centrifugal pump. The first plot is a trend of rms velocity, the second rms acceleration, and
the third is a high pass filtered, 2k Hz HPF, true peak acceleration (an overall HFE measurement). It is
clearly seen that the HFE measurement is far more sensitive to the bearing fault than the rms
measurements. The motor was shut down, the bearing changed, and new levels established. Although there
was a small change in the rms velocity and a bit more in the rms acceleration, they had not tripped their
alarms yet. In slow speed machinery, the HFE reading is the most likely to pick up a bearing fault early.

Conclusions: In order to make accurate measurements, the frequency range must not only be within the
specified range of the accelerometer but must also be within the flat range of the sensor mount. Knowing
the range of the mount is a challenge and can only be determined through testing. When sophisticated
calibration equipment is not available, it is possible to get an estimation of frequency response. Mount a
sensor on a machine having high frequency content using various mounts. Comparing the results, as was
done in the centrifugal compressor example, can provide useful information on about the sensor and
mounting responses.

Even when accurate measurements are made, they may not agree with other accurate measurements. When
trying to compare overall measurements with others, make sure you are using similar sensors, mounting
methods, and measurement methodologies as in the GM Vibration Standard.

True high-frequency measurement such as HFE and demodulation (not covered in this paper) are not
physical measures so the amplitudes are arbitrary. Data must be collected consistently and trended to make
use of the amplitudes. These trends should be used as an indicator to look at other vibration data before
diagnosing a problem. Demodulated time waveforms and spectra can often reveal impending faults,
particularly in cases of impacting, much earlier that conventional analysis. They are a powerful
complementary tool for the vibration analyst.
Figures and Tables

Figure 1: Frequency response test setup Figure 2: Stainless steel probe (stinger)

3 dB Freq Resonant Gain at Total Mass

Sensor & Mount (Hz) Freq (Hz) Resonance (gm)
622B01 + Stud 15000 30000 35 to 40 dB 94
603C01 + Stud 12500 25000 15 dB 51
603C01 + Flat Magnet 6000 11245 21.5 dB 100

622B01 + Flat Magnet 5000 8000 20 dB 143

603C01 + 2-Rail Magnet 3554 6322 27.3 dB 128.1
622B01+ 2-Rail Magnet 3308 6000 22.9 dB 171.1
603C01 + 4" SS Probe 720 1300

Table 1: Resonant frequency and 3 dB point for 2-sensors and 4-mounts

Figure 3: Centrifugal compressor

Figure 4: In velocity spectra, where an analyst would typically expect consistent data,
there are significant differences

Figure 5: Acceleration spectra to 5000 Hz have significant differences

 TM
Overall Spike Energy (gSE)
Pos Sensor Filter Mount Surface 1 kHz HPF 2 kHz HPF 5 kHz HPF
083 603A01 None Flat Mag Glue Base 6.95 6.54 6.46
063 603A01 None Stud Glue Base 6.85 6.36 6.45
023 603A01 None Flat Mag Bare 6.72 6.26 6.35
013 603A01 None 2 Pole Mag Bare 3.79 3.50 3.49
073 603A01 None 2 Pole Mag Glue Base 3.77 3.53 3.44
074 603C01 2-Pole 2 Pole Mag Glue Base 2.85 2.63 2.51
014 603C01 2-Pole 2 Pole Mag Bare 2.52 2.32 2.21
084 603C01 2-Pole Flat mag Glue Base 2.24 2.05 2.02
024 603C01 2-Pole Flat Mag Bare 2.11 1.88 1.81
064 603C01 2-Pole Stud Glue Base 1.91 1.65 1.58
103 603A01 None Stinger Bare 0.64 0.42 0.40
104 603C01 2-Pole Stinger Bare 0.39 0.22 0.14

Table 2: Overall Spike Energy™ for various sensors and mounts

collected on the combustion air fan

RMS Velocity RMS Acceleration True Peak Acceleration

Figure 6: RMS velocity, RMS acceleration, and true peak acceleration trend plots
Definitions

Amplification Factor – The gain or “Q” at the natural frequency (resonance) of the sensor. For a 100
mV/g accelerometer vibrating at 1 g, the output will be 100 mV. If the there is a gain of 10 resonance, a 1 g
vibration at the sensors natural frequency will yield a 1 V output making it appear the input is 10 g.

FFT – Fast Fourier Transform is the algorithm typically used to compute the Discrete Fourier Transform or
frequency spectrum of a digitized time signal. This transforms the signal from the time domain (time
waveform) to the frequency domain (frequency spectrum). In technical papers and publications, this
frequency spectrum is often referred to simply as the FFT.

Frequency Response / Frequency Range of an industrial accelerometer is generally specified at the ±3 dB

points. That means the low and high frequency limits for the accelerometer are determined when the
nominal sensitivity (e.g., 100 mV/g) is either up 3 dB or down 3 dB. The following equation can be used to
determine the actual sensitivity of the accelerometer at these points.

At +3 dB and a nominal sensitivity (SREF) of 100 mV/g, the actual sensitivity (S) is 141 mV/g. At -3 dB, it’s
70.7 mV/g. This shows that when an industrial accelerometer is used at its upper or lower frequency limit,
the deviation from nominal sensitivity is not trivial.

HFE – In vibration literature this can mean High Frequency Enveloping or High Frequency Energy, with
Enveloping being most common. Since this paper deals primarily in overall measurement, it is used for
High Frequency Energy, meaning, an overall measurement that filters out lower frequency content and
includes very high frequency content that most often extends beyond the specified range of the
accelerometer. This type of measurement is most typically used for early detection of rolling element
bearing faults. Spike Energy™ and PeakVue® are examples of HFE type of measurements.

Route is a predefined set of machines, points, and measurements to be made on a regular basis. The route
information is loaded into a portable data collector and an analyst walks around a plant making the
preprogrammed measurements. Most routes include a variety of measurements including time waveform,
velocity, acceleration, and HFE.

Traceability – The National Institute of Standards and Technology, NIST, defines traceability as “an
unbroken record of documentation (“documentation traceability”) or an unbroken chain of measurement
and associated uncertainties (“metrological traceability”)5. In the case of displacement, velocity, and
acceleration (typical vibration measures) there is a well defined physical quantity that is being measured,
can be quantified, and is therefore traceable. In the case of HFE measurements, there is no defined physical
quantity, the measurements are made outside of the traceable range of accelerometers, and are therefore not
traceable. In this case, data should be trended to look for changes.

Transportable – This is a term coined by the author meaning a measurement that is well defined, accurate,
and repeatable and thus could be obtained by anyone making the same measurement.
References:

1. Ming Xu, Joseph M. Shea, James A. Knauff. Vibration Analysis of Multi-Spindle Gear-Type Heads
Using Spike Energy™, VI Chapter Training Seminar, 1994

2. Robinson, James C. and James W. Walker, and James E. Berry. Description of PeakVUE and Illustration
of its Wide Array of Applications in Fault Detection and Problem Severity Assessment. Emerson Process
Management reliability Conference: 2001

3. PeakVUE is a registered trademark of Computational Systems Incorporated.

4. Spike Energy and dataPAC are trademarks of EntekIRD International Corporation.

5. NIST. http://www.nist.gov/pml/mercury_traceability.cfm
 3425 Walden Avenue, Depew, NY 14043 USA
pcb.com/imi-sensors | [email protected] | 800 959 4464 | +1 716 684 0003

© 2021 PCB Piezotronics - all rights reserved. PCB Piezotronics is a wholly-owned subsidiary of Amphenol Corporation. Endevco is an assumed name of PCB Piezotronics of North Carolina, Inc., which is a wholly-owned subsidiary of PCB Piezotronics,
Inc. Accumetrics, Inc. and The Modal Shop, Inc. are wholly-owned subsidiaries of PCB Piezotronics, Inc. IMI Sensors and Larson Davis are Divisions of PCB Piezotronics, Inc. Except for any third party marks for which attribution is provided herein, the
company names and product names used in this document may be the registered trademarks or unregistered trademarks of PCB Piezotronics, Inc., PCB Piezotronics of North Carolina, Inc. (d/b/a Endevco), The Modal Shop, Inc. or Accumetrics, Inc.
Detailed trademark ownership information is available at www.pcb.com/trademarkownership.
WPL_68_1121