PROPER SELECTION OF ACCELEROMETER SENSITIVITY

TO AVOID SATURATION IN VIBRATION MEASUREMENTS

By

Ming Xu, Ph.D.

ENTEK IRD International Corporation

ABSTRACT

Accelerometers are often selected for use with instruments based on their frequency range,
size, weight, mounting and cost without enough attention paid to their sensitivities. The
sensitivity of an accelerometer is inversely proportional to the peak amplitude range. In
certain applications, high acceleration amplitudes or high frequency noise may be present to
overload the transducer causing severe signal distortion and large "ski slope" response. In this
paper, the proper selection of accelerometer sensitivity, two types of ski-slope spectrum and
measurement considerations are discussed. Case histories of high peak acceleration amplitude
and ski-slope due to integration noise are presented.

INTRODUCTION

Piezoelectric accelerometers are widely used in industries for vibration measurements and
machinery condition monitoring. These accelerometers employ piezoelectric crystals to
generate an electrical signal proportional to acceleration. Based on their mode of operation,
accelerometers can be categorized as charge mode and voltage mode. Charge mode
accelerometers contain only the sensing element in the accelerometer with no built-in
electronics. Charge mode accelerometers output a high impedance, electrical charge signal
that is generated by the piezoelectric sensing element. A charge amplifier or in-line charge
converter is used between the sensor and instrument to convert the output signal to a low
impedance voltage. Charge mode accelerometers are typically used in high temperature
applications, such as monitoring engine manifolds and superheated turbines. Because of the
high impedance nature of the output signal, use of low noise, coaxial cable between the
accelerometer and charge amplifier is required to reduce the noise effects.

The voltage mode accelerometers are used in most vibration or shock measurements because
of their simplicity, accuracy, broad frequency range and low cost. ICP (Integrated-Circuit-
Piezoelectric) accelerometers are typical voltage mode accelerometers. They incorporate
built-in signal conditioning electronics to convert the high impedance charge signal into low
impedance voltage signal, which can be readily transmitted over ordinary two wire or coaxial
cables to a readout instrument. The low impedance signal can be transmitted over long cable
distances and used in the industrial environment. Since the amplifiers are built inside the
transducers, they do not need low noise cables or external amplifiers for the measurements.

Enteract Home
 Two of the main parameters in selecting accelerometers are the frequency range and the
sensitivity. The frequency range of interest can be determined based on the machine running
speed and the component frequencies, such as bearing frequencies, gear mesh frequency,
blade or vane pass frequency, as well as the harmonics of these frequencies. On the other
hand, the high frequency range of the sensor is constrained by its mounted resonance
frequency and the low frequency range is constrained by the sensor’s amplifier roll-off filter.
The accelerometer with an adequate frequency range should be chosen to meet the
application requirements.

The selection of proper sensitivity is also important to meet the amplitude range requirement.
The sensitivities of most industrial general purpose accelerometers are standardized to 100
mV/g. However, for certain applications, 100 mV/g accelerometer can be saturated during the
measurement causing large ski slope response. To select the correct sensitivity for an
application, it is necessary to know the range of vibration amplitude levels to which the
transducer will be exposed during the measurements. In this paper, the relationship between
the sensitivity and amplitude range is discussed. Two types of ski slope spectrum and their
causes are discussed in detail. The measurement considerations in selecting proper sensor
sensitivity for different applications are also discussed. Case histories of high peak
acceleration amplitude and ski-slope spectrum due to integration noise are presented.

SENSITIVITY AND AMPLITUDE RANGE

The sensitivity of a transducer is defined as the electrical output per unit of applied
acceleration. For voltage mode accelerometers, sensitivity is determined by the type of
sensing element, the amount of seismic mass and the gain of the internal amplifier. In an ICP
type accelerometer, the charge signal (Q) from the crystals accumulates in capacitor (C)
forming a voltage (V) according to the law of electrostatics (V = Q/C). The magnitude of the
capacitance (C) is the sum of the distributed capacitance of the crystals, lead wire, and
amplifier input. Adding more capacitance reduces sensitivity. The internal amplifiers of ICP
type accelerometers generally have a set gain so that output sensitivities are standardized.

The sensitivity of a piezoelectric accelerometer can be expressed as voltage sensitivity or

charge sensitivity. Voltage sensitivity is expressed in units of millivolts per g (mV/g) (where
1 millivolts = 1x10-3 volts and g is the acceleration of gravity, i.e., g = 32.174 ft/s2 or g =
9.807 m/s2). Charge sensitivity usually is expressed in units of picocoulombs per g (pC/g)
(where 1 picocoulomb = 1x10-12 coulombs). The sensitivities of voltage mode industrial
accelerometers are typically range from 10 mV/g to 500 mV/g.

The amplitude range of accelerometers is related to the power supply voltage, as shown in
Figure 1. The output signal from an accelerometer is an AC signal proportional to the
vibration of the machine or structure at the sensor mounting point. This AC signal is
superimposed on a DC bias voltage (also referred to as Bias Output Voltage (BOV) or Rest
Voltage). The DC component of the signal is blocked by a capacitor. The AC component of

Enteract Home Xu Beginning

 the signal passes through the capacitor and inputs to the instrument. FFT Analyzers, data
collectors and monitors usually contain an internal blocking capacitor for AC coupling.

Figure 1. Amplitude range and supply voltage

As shown in Figure 1, the DC bias voltage is usually set halfway between the lower and
upper cutoff voltages. The lower and upper cutoff voltages are typically 2 ~ 4V above ground
and 2 ~ 4V below the supply voltage, respectively. The difference between the bias voltage
and cutoff voltage determines the voltage swing available at the output of the sensor. The
output voltage swing determines the peak vibration amplitude range. Therefore, an
accelerometer with a sensitivity of 100 mV/g and a peak output swing of 5 volts will have an
amplitude range of 50 g peak. Table 1 lists the typical peak amplitude range of industrial
accelerometers with different sensitivities and a peak output swing of 5 volts.

Table 1. Sensitivity versus peak amplitude range

Sensitivity, mV/g 1000 500 100 50 25 10
Range, g peak 5 10 50 100 200 500

The peak amplitude range of an accelerometer is inversely proportional to its sensitivity, as

illustrated in Table 1. High sensitivity accelerometer (≥ 500 mV/g) can differentiate very
small signals from noise. On the other hand, high sensitivity accelerometers have limited
amplitude and high frequency operating range. Therefore, high sensitivity accelerometers are
often used for low frequency and low amplitude measurements. High sensitivity
accelerometers are not suitable for high frequency measurements. For high frequency and
high acceleration amplitude measurements, lower sensitivity sensors should be used to
prevent overload of sensor’s amplifier. Whereas a 100 mV/g accelerometer may overload at
50 g, a 50 mV/g sensor can measure upto 100 g.

TWO TYPES OF SKI SLOPE AND CAUSES

Two types of ski slope may occur in a frequency spectrum. The first type is caused by
saturation of transducer’s amplifier due to high acceleration amplitudes or high frequency
noise. The second type is due to integration noise. In some applications, very high
acceleration may be present and overload the sensor to cause severe signal distortion. In this
particular case, the high acceleration amplitude can quickly saturate the sensor’s internal
amplifier. It takes time for the charge capacitor to gradually discharge and for the amplifier to

Enteract Home Xu Beginning

 return to its normal operating mode. This process is similar to an impulse with a long decay
time. The long duration in the time domain will transfer to low frequency contents in the
frequency domain. This condition is also referred to as “washover” and has an appearance of
ski slope, as shown in Figure 2. The presence of ski slope also causes the loss of high
frequency data due to scaling, which makes the measured vibration data unusable.

Figure 2. Sample of ski slope spectrum

In vibration measurements, it is not uncommon for a machine tool spindle to generate more
than 50 g peak acceleration during the cutting operation. Certain types of signals can also
generate high impacts, such as broken gear teeth and positive displacement pumps. When the
acceleration amplitude exceeds the transducer’s amplitude range, it will be saturated and
results in severe signal distortion. High frequency noise can also overload sensor's amplifier
causing ski slope. Ski slope response can occur in the following cases: high speed spindles
and compressors, steam leak in paper machine dryers, and pump cavitation.

The second type ski slope is mainly due to integration noise. Integration is typically used in
vibration instruments or software to convert the measured data in different units, such as
acceleration to velocity, acceleration to displacement, and velocity to displacement. The
conversion process can be implemented in either hardware by using analog integrators or
software by performing digital integration. In some cases, the amplitudes in the acceleration
or velocity spectrum are quite small or close to noise floor. When integrating from
acceleration to velocity, acceleration to displacement, or velocity to displacement, the ski
slop appears due to 1/f effect (high level of low frequency noise). Example of this type of ski
slope is shown in Figure 3. In general, the ski slope levels in this case are much lower.

Enteract Home Xu Beginning

 3(a). Acceleration spectrum

3(b). Velocity spectrum

3(c). Displacement spectrum

Figure 3. Ski slope due to 1/f effect

Enteract Home Xu Beginning

 For a given frequency f (Hz), the digital integration is a simple scaling process. The scaling
or conversion factors used to convert between acceleration and velocity, or acceleration and
displacement, or velocity and displacement are given in Table 2.

Table 2. Conversion factors

Units Acceleration (A) Velocity (V) Displacement (D)
g, peak in/s, peak mils, peak-to-peak
Acceleration (A)
g, peak 1 A = 0.0163 f V A = 0.0000511 f2 D
Velocity (V)
in/s, peak V = 61.4338 A/f 1 V = 0.0031416 f D
Displacement (D)
mils, peak-to-peak D = 19555 A/ f2 D = 318.31 V/f 1
f = frequency, Hz.

In the case of integration from acceleration (A, g peak) to velocity (V, in/s peak), the peak
acceleration value A is multiplied by the conversion factor 61.4338/f. The factor (1/f)
depends on the frequency value. The factor (1/f) will be quite small at high frequency and
become very large at low frequency. For instance, if A = 0.002 g peak at 0.25 Hz (15 CPM)
and 200 Hz (12000 CPM), the velocity at f = 0.25 Hz is V = 61.4338 x 0.002 / 0.25 = 0.49
in/s peak and at f = 200 Hz, V = 61.4338 x 0.002 / 200 = 0.0006143 in/s peak. For the same
acceleration amplitude, the ratio of the two velocity amplitudes is 0.49 / 0.0006143 = 798.
Therefore, the integration will yield much higher amplitude at low frequency. This is the so
called “1/f” effect. Similar 1/f effect exists when integrating velocity to displacement. In the
case of double integration, i.e., acceleration integrating to displacement, the factor 19555/f2 is
used and the 1/f effect is more significant at low frequencies.

MEASUREMENT CONSIDERATIONS

The appearance of ski slope in a frequency spectrum should be cause for concern when low
frequency amplitudes are not expected. These low frequency anomalies emulate real
vibrations and their amplitudes will be included in either the spectrum based overall (also
called Root-Sum-Square or RSS overall) or broad band analog overall. Although ski slope
shows at the first few frequency bins in a spectrum, it can have significant effect on trending,
band alarm, and overall alarm. Therefore, it is important to select a proper transducer
sensitivity to avoid the transducer saturation.

In general, lower sensitivity accelerometers (”P9JVKRXOGEHXVHGIRUKLJKLPSDFWRU

high speed applications. Typical applications are monitoring machine tool spindles during the
cut, high speed spindles, high speed gear boxes, positive displacement pumps, etc. Lower
sensitivity accelerometers allow measurements of the higher acceleration amplitudes (•J
without saturation of the sensor’s amplifier. Sensor overload problem can be found by
examining time waveform. Truncated time waveforms indicate signal clipping and amplifier

Enteract Home Xu Beginning

 saturation. If signal clipping is detected, lower sensitivity accelerometers should be used to
replace the higher sensitivity ones.

High sensitivity accelerometers are used for low frequency measurements, such as paper
machines, slow speed agitators, mixers, pumps, low speed machine tools, etc. Low frequency
measurements can be very difficult and time consuming. The important measurement
considerations are low frequency response and noise floor. Low frequency response is
governed by low frequency filters contained within the sensor and analyzer. For instance, if
both the sensor and analyzer have a –3 dB high pass filter cutoff frequency at 1 Hz, the
combined filter attenuation would be –6 dB or 50% at 1 Hz. In other words, the amplitude
reading would be reduced by one half at 1 Hz (-6 dB log = 0.5 linear).

In low frequency measurements, very little vibration in terms of acceleration is present at low
frequency. Therefore, signal to noise ratio must be maximized. Low frequency
accelerometers use a larger seismic mass and peizoceramic sensing elements to maximize the
charge output. The amplifier provides high voltage output (•P9JWRLQFUHDVHWKHVLJQDO
level.

High sensitivity accelerometers are not suitable for high frequency measurements, such as
Spike Energy measurement. First, high sensitivity sensors usually have low pass filters
with the roll-off frequency typically between 300 Hz and 700 Hz. The frequency detection
range of Spike Energy is upto 65 kHz. Thus, there is no signal output in most of the Spike
Energy frequency detection range. Second, the amplitude range of high sensitivity
accelerometers is quite low, as shown in Table 1. It is not uncommon to see relatively large
Spike Energy (in gSE units) readings. In this case, high sensitivity sensors can be easily
saturated. For high impact or high-speed applications, accelerometers with 50 mV/g or less
sensitivity are recommended for Spike Energy measurement.

In some cases, transient signals may cause erroneous readings. This is especially important in
low frequency measurements since the transducer requires longer settling time. Auto ranging
in the portable instruments also needs longer time. Acquiring data before the transducer and
analyzer have had a chance to settle can cause high overall readings that are erroneous. If
temperature variations in the transducer cause direct interference during the measurement,
shear mode transducers and thermal jackets should be used to eliminate the thermal transit
problem.

High frequency noise, such as pump cavitation and steam leaks, can overload the amplifier
and cause ski slope. The high frequency noise can be eliminated by choosing sensors with
built-in low pass filters or using mechanical filters to limit high frequency range. To reduce
the undesired low frequency noise, the hardware integrators of FFT analyzers, data collectors,
and monitors incorporate filters with linear response to low frequency corner frequencies.
These corner frequencies can be selected based on applications, as shown in Figure 4. Similar
high pass filters can also be programmed in the software to reduce the 1/f integration noise.

Enteract Home Xu Beginning

 Figure 4. Selectable integrator corner frequencies

There are several other types of problems that can also generate low frequency content. Such
problems include poor or improper mounting of transducers, weak cable and/or connector
connection, and cable movement. Poor mounting can result in loss of high frequency signals
and increase of noise. Weak connection can generate intermittent signals and cause erroneous
readings. Cable movements can induce electrical charges (so-called Tribo-electricity) and
result in an erroneous output signal. To eliminate the cable movements, cables need to be
clamped or taped down to the machine surface.

CASE HISTORIES

Two case histories are presented. The first case showed that lower sensitivity accelerometer is
proper for high impact applications. The second case history illustrated the sky slope
spectrum due to integration noise.

Case History I. High Acceleration Amplitudes Generated During the Cut

The vibration data were taken from a drilling machine. It had three planetary gear heads and
each gear head was equipped with three drills. The input speed of the planetary gear was
2437 RPM and the drill speed was 5654 RPM. The machine was used to drill three holes on
the work piece. The material of the work piece was 775 High Silicon Aluminum. Two sets of
data were obtained at the front of the drill head, as shown in Figure 5 and Figure 6.

The first set of data was measured by using permanently mounted accelerometer with 100
mV/g sensitivity. The time waveforms were shown in Figure 5 (a) and (b). High impact (g)
levels were generated during the cut. The measured time waveforms were truncated
indicating that the accelerometer’s amplifier was saturated. In this particular case, the
amplitude range of the accelerometer is limited to 50 g’s peak, but the real impact force
during the cut was more that 50 g’s peak.

Enteract Home Xu Beginning

 5(a). First measurement

5(b). Second measurement

Figure 5. Time waveform of the drilling machine using 100 mV/g accelerometer

The second set of data was measured by using magnet mounted accelerometer with 25 mV/g
sensitivity. The unclipped time waveforms are shown in Figure 6 (a) and (b). The measured
data show that the maximum impact generated during the cut was greater than 60 g’s peak.
For this application, accelerometer with 50 mV/g (100 g’s peak range) or 25 mV/g (200 g’s
peak range) is a proper choice to avoid the sensor amplifier saturation problem.

6(a). First measurement

Enteract Home Xu Beginning

 6(b). Second measurement
Figure 6. Time waveform of the drilling machine using 25 mV/g accelerometer

Case History II. Ski Slope Spectrum Due to Integration Noise

The data was taken by using a permanently mounted 50 mV/g accelerometer. In this case, the
transducer was stud mounted underneath a three box spindle cluster. The spindle running
speed was 2400 RPM. This particular accelerometer has been installed for several years.
Since the transducer was inside the machine, it was inaccessible unless the machine had
major repair. The measured acceleration spectrum is shown in Figure 7(a) and the velocity
spectrum obtained by integration is shown in Figure 7(b).

In this particular case, only noise appeared in the acceleration spectrum, as shown in Figure
7(a). This may be caused by bad accelerometer or loose cable connection. The transducer can

7(a). Acceleration spectrum

10

Enteract Home Xu Beginning

 7(b). Velocity spectrum
Figure 7. Ski slope due to integration noise

not be checked out until next major repair. The digital integration of this noisy acceleration
spectrum produced the ski slope in the velocity spectrum due to the 1/f effect. Since the noise
level in the acceleration spectrum was very low, the ski slope level after integration was also
quite low. The very noisy low-level acceleration spectrum usually indicates the problem of
transducer, cable or their connection.

CONCLUDING REMARKS

Piezoelectric accelerometers have been widely used in vibration measurements and condition
monitoring. Most general-purpose industrial accelerometers normally have the sensitivity of
100 mV/g. However, in high impact or high frequency noise applications, 100 mV/g
accelerometers can be saturated, as illustrated in the first case history. To avoid sensor
saturation problem, lower sensitivity accelerometers should be used.

The ski slope occurs when integrating from a noisy acceleration spectrum to velocity
spectrum due to 1/f effect, as shown in the second case history. The appearance of ski slope
should be concerned when low frequency amplitudes are not expected. The causes of noisy
acceleration spectrum need to be determined. In high frequency noise applications, low pass
filter or mechanical filter can be used to reduce the high frequency noise.

High sensitivity accelerometers can differentiate very small signals from noise and are often
used for low frequency and low amplitude measurements. On the other hand, high sensitivity
accelerometers have limited amplitude and high frequency range. Therefore, high sensitivity

11

Enteract Home Xu Beginning

 accelerometers are not suitable for high frequency measurements, such as Spike Energy
measurement. For high frequency applications, accelerometers with 50 mV/g or less
sensitivity are recommended for Spike Energy measurement.

ACKNOLEDGEMENT

The author would like to thank Gene Wogoman and Tony Midlam of Delphi Harrison
Radiator, Donn Stoutenburg and Ed Hart of Entek IRD for their support and assistance during
this study.

REFERENCES

1. Abraham I. Dranetz and Anthony W. Orlacchio, “Piezoelectric and Piezoresistive

Pickups,” Chapter 16 of Shock and Vibration Handbook, 2nd edition, Edited by Cyril M.
Harris and Charles E. Crede, ISBN: 0-07-026799-5, McGraw-Hill Book Company, New
York, New York, 1976.

2. Wilcoxon Research 1995 Industrial Vibration Sensors Product Catalog, Wilcoxon

Research, Inc., Gaithersburg, Maryland, 1995.

3. Vibration and Shock Sensor Selection Guide – Piezoelectric Accelerometers, PCB

Piezotronics, Inc., Depew, New York, 1993.

4. Ming Xu, “Spike Energy Measurement and Case Studies in Machinery Condition
Monitoring,” Proceedings of the 23rd Vibration Institute Annual Meeting, Orlando, Florida,
pp.37-47, June 8-10, 1999.

TRADEMARKS

Spike Energy is trademark of Entek IRD International Corporation.

ICP is registered trademark of PCB Piezotronics, Inc.

12

Enteract Home Xu Beginning