See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224624969

Envelope bearing analysis: Theory and practice

Conference Paper  in  IEEE Aerospace Conference Proceedings · April 2005

DOI: 10.1109/AERO.2005.1559671 · Source: IEEE Xplore

CITATIONS READS
17 2,100

2 authors, including:

Eric Bechhoefer
GPMS Inc
109 PUBLICATIONS   980 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Prognostics and Health Management of Wind Turbine Generators View project

Improving the Safety Management System through Robust HFDM View project

All content following this page was uploaded by Eric Bechhoefer on 18 August 2016.

The user has requested enhancement of the downloaded file.

 Envelope Bearing Analysis: Theory and PracticeΩ
David Hochmannπ and Eric Bechhoefer
Goodrich Corporation
Fuel Utility Systems
100 Panton Road
Vergennes, VT 05491
USA

Abstract— f Frequency Independent Variable

y ( x) Signal to Envelope
The rolling element bearing is an important
element for power transmission within the helicopter drive s ( x) Heterodyned Signal
train system. Monitoring the condition of the rolling bearing
element’s component provides advantages in the operation, r ( x) Low Passed Signal
safety, and maintenance areas.
An overview of the diagnostic process for a rolling
Y(f) Fourier representation of Signal to Envelope
element bearing with pit type fault from the mechanical S( f ) Fourier representation of Heterodyned Signal
defect, to the defect signature, to the detection process is
presented. The detection process applies the enveloping R( f ) Fourier representation of Low Passed Signal
technique.. The work presented is demonstrated with an
analytical description of the enveloping process, analytical z Center frequency
examples, a simple numerical simulation and the results c Cut off frequency
from an operational helicopter. b Number of rolling elements
The benefits of understanding the fundamental f Ball bearing inner race rotational speed
bearing analysis technology include improved diagnostics
capability and prognostic capability. d Ball bearing diameter
e Bearing pitch diameter
β Bearing Contact angle
δ ( x) δ (0) = 1
Ωπ
1 TABLE OF CONTENTS Impulse function,
Θ (t ) Short term transient
1 Table of Contents ....................................................... 1
2 Nomenclature ............................................................. 1 ℑ Fourier Series
3 Introduction ................................................................ 1 * Complex Conjugate
4 Discrete Rolling Element Bearing Faults ................... 2 k , p, q Integer index
5 Envelope Detection Theory........................................ 4
6 Examples .................................................................... 5
6.1 One Tone Analytical Example.......................... 5
6.2 Three Tone Analytical Example ....................... 6
6.3 Numerical Example .......................................... 7 3 INTRODUCTION
6.4 Aircraft Example .............................................. 7
7 Conclusions ................................................................ 9
8 References .................................................................. 9 The rolling element bearing is an important
9 Acknowledgements .................................................... 9 component for power transmitting systems within the
10 Biography................................................................... 9 helicopter drive train. Monitoring the condition of the
bearing component provides advantages in the safety,
operation and maintenance areas. A commonly used method
is envelope detection.
2 NOMENCLATURE The envelope detection process is demonstrated
through a simple analytical model and through operational
x, t Independent domain variables acquired data. The content of this paper is several fold:

1. Explain a mechanism of a discrete bearing fault from

Ω
the detection perspective.
0-7803-8870-4/05/$20.00  2005 IEEE 2. Analytically describe the envelope signal.
π
Corresponding Author: [email protected]
1
 3. Explain the meaning of the results in the envelope
signal The lack of a priori information about the nature of the
4. Present analytical, numerical and experimental transient is an important issue with respect to the detection
examples. of defect. The exact shape of the transient or the natural
frequency of the ring is difficult to determine ahead of time.
The goal of this paper is to present the health Though transients are difficult to know exactly, a priori,
monitor and diagnostic problem in a mechanistic approach. they share some properties that are useful for detection. One
The defect is related to a physical phenomenon that is a useful property of transients is that relationship between
measurable quantity. The diagnostics algorithm is examined time and frequency, [ 9 ]. Transients in time are local
in terms of the physical defect and how it relates to changes features whereas transients in the frequency domain are
in the diagnostics algorithm output. distributed, [ 8 ]. To demonstrate, three transient signals are
examined. The three transients presented in this case are the
perfect impulse, a short term impulse like event with a non-
4 DISCRETE ROLLING ELEMENT BEARING zero duration T, and a damped sinusoidal signal. These
FAULTS transients are presented in the time domain in, Figure 2,
Figure 3, and Figure 4 respectively. Figure 5 presents the
magnitude of the frequency spectrum for the three
Bearing diagnostics traditionally assumes a small fault in transients. Not absolutely, but typically Figure 2 and Figure
one of the rolling element bearing’s components which will 3 may be considered as the detect signature at the fault
cause a short time duration transient [ 1 ]. Assuming fixed source. Similarly, Figure 4 may be considered the bearing
outer race, moving inner race and no slippage for the rolling race, component housing or an alternative component’s
elements , Figure 1, equations ( 1 ), ( 2 ), and ( 3 ) provide response to the excitation.
the inner race defect rate, the outer race defect rate, and the
ball fault rate for a single discrete defect, respectively, [ 1]. ideal impulse
1
Equation ( 4 ) presents the cage unbalance frequency, [ 1].
0.9

bf  d  0.8
 1 + cos ( β )  (1) 0.7
2  e 
displacement

0.6

0.5
bf  d 
 1 − cos ( β )  (2) 0.4
2  e  0.3

0.2

ef   d  
2

 1 −   cos ( β ) 
0.1
2
(3)
d  e 
0
0 1 2 3 4 5 6
time

Figure 2 Idealized Impulse in Time Domain, Input

f  d 
 1 − cos ( β )  (4)
2 e  short term discrete event
1

Outer Race 0.9

0.8

0.7

Ball
displacement

0.6

0.5

0.4

0.3

0.2

Cage 0.1

0
0 1 2 3 4 5 6
time
Inner Race
Figure 3 Short Discrete Event of duration T, Input
Figure 1 Rolling Element Bearing
2
 the spectral representation of Figure 6 up to the Nyquist
damped sinusoid sampling frequency. Adjusting Figure 7’s scaling
1 parameters results in Figure 8 which is the spectral result of
0.8 Figure 6 about 50 Hz. Figure 8 presents the fundamental
0.6 premise for bearing detection for many techniques. The
presence of a signal indicates a defect and the modulation of
0.4
the signal identifies quasi-uniquely the source. The actual
displacement

0.2 identification capability is limited by the uniqueness of the

0 various possible defect sources. Two identical bearings are
-0.2
both possible defect source when only the generating period
is considered.
-0.4
The envelope detection method turns the above qualitative
-0.6 description into a qualitative technique. Two common
-0.8 methods, Figure 9, of envelope detection is typically known
-1
as bandpass rectification and Hilbert transform, [ 2], which
0 1 2 3
time
4 5 6 produce results that are proportional to each other. The
envelope detection method is explained through the
Figure 4 Damped Sinusoidal in Time Domain, Response bandpass rectification path in this paper.

Transient detection is also a key diagnostics technique for

certain gear faults, [ 9 ]. Unlike gear faults that are y ( x '+ nT ) = y ( x ' )
synchronous to the rotation rate which allows isolation n ∈ K , -2,-1,0,1,2, K (5)
through synchronous averaging techniques, [ 9 ] and [ 10 ],
the periodic nature of bearing fault defect rates are not x ' ∈ [0, T )
typically synchronous to the rotational rate, f . Though the
rolling element bearing transient defect signal is not
synchronous to the shaft speed it is assumed periodic ( 5 ) in  n
the first approximation. ℑ ( y ( x ) ) = ℑ ( y ( x ' ) ) ∑δ  f −  (6)
The Fourier series of a periodic signal may be described as ( ∞  T
6 ). The Fourier series of a periodic signal is the product of
the Fourier series of one cycle of the periodic signal and an
impulse train which is a function of the cycle period. The
idea is demonstrated numerically in Figure 6, Figure 7 and
Figure 8. Figure 6 is an ideal pulse train that may be
representative of a defect signature in a bearings with a
period of T = 10 s . In operational situations the bearing
defect signature is obscured by the signatures of normally
operating components such as shafts and gears. Figure 7 is

δ (t )
Magnitude

Θ (t ) e −αt sin( f ot )

fo
f
1
T Ψ (α )

Figure 5 Frequency Domain representation of three transient signals, an ideal impulse, δ(t), a short term transient, Θ(t),
of duration T, and a damped sinusoidal signal
3
 y ( x)
10 s

BandPass Hilbert Transform

Rectification Technique

r2 ( x )

Figure 6 Time/Spatial Domain Damped Harmonic Signal Figure 9 Envelope Detection Method
Train, y(x’)=sin(2π50x’)e-4x’

5 ENVELOPE DETECTION THEORY

The analytical examination of the enveloping technique

follows the bandpass rectification method. The bandpass
method is presented as a flow chart in Figure 10. The
bandpass method involves a heterodyne operation followed
by a low pass filter operation. The magnitude squared of the
complex output is computed through the multiplication with
the complex conjugate.

y ( x)

Figure 7 Fourier Series of Figure 6, y(x’)=sin(2π50x’)e-4x’

Frequency Shift

0.1 Hz
Low Pass Filter

RR*

Figure 8 Fourier Series of Figure 6 scaled about 50 Hz,

r2 ( x )
y(x’)=sin(2π50x’)e-4x’

Figure 10 Band Pass Process for Enveloping

4
A real value signal y ( x ) such as one that may be acquired r ( x ) = r ( x ) r* ( x )
2

for bearing analysis is represented as a complex coefficient

Fourier series, ( 7 ). For a real valued signal the Fourier = ∑ Ap e j 2π ( p − z ) x ∑ Aq*e − j 2π ( q − z ) x
( 11 )
coefficient Ak possess the following property Ak = A−* k p q

while implies that the magnitude spectrum of y ( x ) is

= ∑∑ Ap Aq*e j 2π ( p − q ) x
p q
symmetric about the zero frequency.

y ( x ) = ∑ Ak e j 2π kx (7) 5.1 Examples

k

Section 5 presented the analytical operations to compute the

A heterodyne operation is applied to y ( x ) , ( 8 ), resulting
r ( x ) , from the real valued signal
2
envelope signal,
in signal, s ( x ) , whose spectrum is no longer symmetric
y ( x ) . Further insight may be gathered by examining
about the zero frequency but about the frequency z .
Traditionally, the frequency z is selected near a resonance several examples.
that is unlikely to be excited by harmonic motion.

y ( x ) e − j 2π zx = ∑ Ak e j 2π kx e − j 2π zx 5.2 One Tone Analytical Example

k

= ∑ Ak e
j 2π ( k − z ) x The first analytical example is an extremely simple case of a
(8)
k single harmonic signal, ( 12 ). Such an example may be used
to begin the familiarity process in a numerical environment
= s ( x) with the envelope algorithm. Equation ( 13 ) is the Fourier
representation of ( 12 ). The heterodyne operation or
Analytically, an ideal low pass filter with a cut off frequency shift of ( 12 ) is presented as ( 14 ) in the spectral
domain. Note that ( 14 ) is now conjugate symmetric about
frequency of c is applied to the signal s ( x ) . From the
frequency z . Assuming that ( Ω − z ) < c , the cut off
original frequencies, k , the components that pass through
frequency of the low pass filter, processing ( 14 ) through
the low pass filter are defined by the inequality ( 9 )
the low pass filter results in ( 15 ). Computing the inverse
resulting in the complex valued envelope signal r ( x ) Fourier series of ( 15 ) results in a complex values signal (
presented as ( 10 ). 16 ). Computing the magnitude square of ( 16 ) results in (
17 ), the envelope signal. The envelope of a single harmonic
signal is a constant value which is a function of the
p∈k ∴ p − z < c (9) amplitude only.
This example does not provide significant insight into the
envelope signal meaning but does verify, a common
r ( x ) = ∑ Ap e
j 2π ( p − z ) x
( 10 ) observation by individuals numerically experimenting with
p the envelope algorithms. The next analytical example is a
signal with three harmonic tones.
The magnitude of the complex valued function r ( x )
through the multiplication of the complex conjugate of y ( x ) = Φ sin ( 2πΩt ) ( 12 )
r ( x ) resulting in, (11), the squared magnitude of r ( x ) .
Further examination of envelope signal as r ( x ) or
2
ℑ ( y ( x )) = Y ( f )
Φj Φj ( 13 )
r ( x)
2
has been examined by Ho, [ 2 ]. = δ ( f + Ω) − δ ( f − Ω)
2 2

5
 Φj ℑ ( y ( x )) = Y ( f )
S( f )= δ ( f + (Ω − z ))
2 Φ0 j Φ j
Φj
( 14 ) = δ ( f + Ω0 ) − 0 δ ( f − Ω0 )
− δ ( f − (Ω + z )) 2 2
2 Φ1 j Φ1 j
+ δ ( f + Ω1 ) − δ ( f − Ω1 ) ( 19 )
2 2
Φj
R( f ) = δ ( f + (Ω − z )) ( 15 ) Φ j Φ j
+ 2 δ ( f + Ω2 ) − 2 δ ( f − Ω2 )
2 2 2

ℑ −1 ( R ( f ) ) = r ( x )
Φ0 j Φ j
Φ j j 2π ( Ω− z ) x ( 16 ) S( f )= δ ( f + ( Ω0 − z ) ) − 0 δ ( f − ( Ω0 + z ) )
= e 2 2
2 Φ j Φ j
+ 1 δ ( f + ( Ω1 − z ) ) − 1 δ ( f − ( Ω1 + z ) )
2 2 ( 20 )
Φ2 j Φ2 j
δ ( f + ( Ω2 − z ) ) − δ ( f − ( Ω2 + z )
*
 Φ j j 2π ( Ω− z ) x   Φ j j 2π ( Ω− z ) x  +
r ( x)
2
= e  e  2 2
 2  2 
 Φ j j 2π ( Ω− z ) x   −Φ j − j 2π ( Ω− z ) x 
= e  e  ( 17 ) Φ0 j
 2  2  R( f ) = δ ( f + ( Ω0 − z ) )
2
Φ2
= Φ j
4 + 1 δ ( f + ( Ω1 − z ) ) ( 21 )
2
Φ j
+ 2 δ ( f + ( Ω2 − z ) )
2
5.3 Three Tone Analytical Example
ℑ −1 ( R ( f ) ) = r ( x )
The next analytical example is a simple case of the
superposition of three harmonic signals, ( 18 ). Equation ( Φ 0 j j 2π ( Ω0 − z )
19 ) is the Fourier representation of ( 18 ). The heterodyne
= e
2
operation or frequency shift of ( 18 ) is presented as ( 20 ) in ( 22 )
Φ j
the spectral domain. Note that ( 20 ) is now conjugate + 1 e j 2π ( Ω1 − z )
symmetric about frequency z . Assuming that 2
(Ω q − z ) < c , the cut off frequency of the low pass filter, Φ j
+ 2 e j 2π ( Ω 2 − z )
processing ( 20 ) through the low pass filter results in ( 21 ). 2
Computing the inverse Fourier series of ( 21 ) results in a
complex values signal ( 22 ). Computing the magnitude
r ( x ) = r ( x ) r* ( x )
2
square of ( 22 ) results in ( 23 ) an envelope signal. Noting
the relationship between the complex exponential functions 1 2
and the trigonometric function defined by ( 24 ) and = Φ 0 + Φ12 + Φ 22
substituting into ( 23) results in ( 25). Equations ( 23) and ( 4
25) provide the initial insight to the envelope signal. The
envelope signal is a sum of the product of differential
(
+Φ 0Φ1 e
j 2π ( Ω0 −Ω1 ) x
+e
− j 2π ( Ω0 −Ω1 ) x
) ( 23 )

frequencies. This result explains the frequency invariance of +Φ Φ ( e1 2

j 2π ( Ω1 −Ω2 ) x
+ e − j 2π ( Ω1 −Ω2 )
x
)
the envelope signal for perturbations of frequency shift
variable z and the cutoff frequency c . +Φ Φ ( e )
j 2π ( Ω0 −Ω2 ) x − j 2π ( Ω0 −Ω2 x ) 
0 2 +e

y ( x ) = Φ 0 sin ( 2πΩ0 t )
+Φ1 sin ( 2πΩ1t ) ( 18 )
e j 2π fx + e − j 2π fx = 2 cos ( 2π fx ) ( 24 )

+Φ 2 sin ( 2πΩ 2 t )

6
 1 2
r ( x) =
2
Φ 0 + Φ12 + Φ 22
4
(
+2Φ 0Φ1 cos ( 2π ( Ω0 − Ω1 ) x ) ) ( 25 )
+2 Φ 1 Φ 2 ( cos ( 2π ( Ω − Ω ) x ) )
1 2

+2 Φ 0 Φ 2 ( cos ( 2π ( Ω − Ω ) x ) )
0 2 Figure 13 104 Hz Spacing of Side Bands Reflecting Impulse
Rate

5.4 Numerical Example

Insight into the performance of the envelope algorithm is

easily made by generating a numerical example that could
be representative of a real world signal. Figure 11 is a
numerically generate exponentially decaying impulse
oscillating at 17 KHz. The pulse rate is 104 Hz and the Figure 14 PSD of Enveloped Signal
sample rate is 104KHz. The impulse signal is corrupted
with white nose such that the resulting test signal has an
SNR of 0 dB. The spectrum Figure 12 of Figure 11 clearly
shows the oscillation frequency centered at 17 KHz. Note
that there is essential no detectable signal at the impulse rate 5.5 Aircraft Example
of 104 Hz in Figure 11. Figure 13 scales the plot of Figure
12 around the oscillation frequency of 17 kHz. which shows
the 104 Hz spacing, as may be predicted by ( 6 ). The time The envelope algorithm based on, Section 5, is implemented
domain signal of Figure 11 is processed through ( 11 ) on the Goodrich Corporation (FUS) IMD HUMS, and is
which results in Figure 14 with z = 17.5 kHz and currently on number of helicopters. Time domain data may
be recorded for further analysis. The data is sampled at a
c = 2.5 kHz . The spectral results of the envelope signal high sample rate and 215 samples are typically recorded.
clearly identifies the 104 Hz impulse rate. The SNR of the
The first time domain example is Figure 15 which presents a
resulting signal is nominally 35 dB
single acquisition of a operating helicopter. No impulse
trains are readily apparent in the time domain. Figure 16
displays the spectral series of Figure 15. Significant spectral
activity is centered around 20 kHz. The high level may be
an indicator of a resonance. The envelop algorithm was
performed with a window between 17.5 KHz to 22.5 KHz.
The spectrum of the envelop is presented in Figure 17.
Figure 17 highlights impulses train at 69 Hz, 284 Hs and
401 Hz. The 69 Hz frequency is associated with a shaft,
while the 284 Hz and 401 Hz frequency are the Outer and
Inner bearing rate frequency supporting the 69 Hz shaft.
Figure 11 Time Domain Representation of Impulse with 0 These elevated bearing energies are indicative of a
dB SNR White Noise propagating bearing fault.

Figure 12 Spectrum of Test Signal Figure 15 Time Domain Signal from Aircraft

7
 Figure 20 Spectrum of Envelope Signal from Figure 18
Figure 16 Spectrum of Aircraft Signal from Figure 15

The last time domain example is Figure 21 which presents a

single acquisition of a operating helicopter. The acquisition
was investigated due to poor data quality indications
including high peak to peak voltage levels, which may
indicate a loose mounting bracket. The time domain
measurements clearly displayed a series of impulses trains
with an approximate rate of 3.7 x 10-3 seconds, or 270 Hz.
The spectrum of Figure 21 is presented in Figure 22. Figure
Figure 17 Spectrum of Envelope Signal from Figure 15
22 displays the numerous spectral lines associated with shaft
and gear mesh at the lower frequency band. A potential
The next time domain acquisition is from a different aircraft resonance is evident between 17 KHz and 27 KHz.
but the same sensor location as the prior example with Performing an envelope computation on Figure 21 between
“good” components. Figure 18 presents the time domain 17 KHz and 27 KHz results in a envelope spectrum
acquisition, Figure 19 presents the spectral results of Figure presented in Figure 23.
18, and Figure 20 the spectral results of the envelope of From Figure 23 a 268 Hz rate is evident which matches the
Figure 18. The results of Figure 18, Figure 19, and Figure 270 Hz time domain estimation. Further investigation
20 is presented at the same scaling as Figure 15, Figure 16, suggested that the pump, being driven at 67 Hz, is of a 4
and Figure 17 respectively.. The spectrum of this envelop lobe design may be the source. This example further
shows no discernable signal, which is expected for demonstrates the capability of component and fault
nominally health components. identification.

Figure 18 Nominal Aircraft Time Domain Signal Figure 21 Time Domain Signal from Aircraft

Figure 22 Spectrum of Aircraft Signal from Figure 21

Figure 19 Spectrum of Aircraft Signal from Figure 18

8
 and Management System)”, IEEE Aerospace
Conference, Big Sky, MT., 2003.
[ 10 ] Hochmann, D., Sadok, M., “Theory of Synchronous
Averaging”, IEEE Aerospace Conference, Big Sky,
MT, 2004.

Figure 23 Spectrum of Envelope Signal from Figure 21 8 ACKNOWLEDGEMENTS

The authors would like to thank Mokhtar Sadok of

6 CONCLUSIONS Goodrich Corporation (FUS), Vergennes Vermont, for his
technical assistance The authors would also like to thank Al
Duke, Robert Hess, and Bob Zeliff of Goodrich Corporation
The envelope detection method leverages off the spectral
(FUS), Vergennes Vermont, for their support
domain property that time domain periodic signals modulate
in the frequency domain spectrum. The modulation in the
frequency domain provides the envelope detection method
to discriminate between various modulation rates. This 9 BIOGRAPHY
allows the envelope detection method to detect not only
faults on different bearings from the same time domain data Dr. David Hochmann is a senior technologist for Goodrich
but also the potential source, whether it be an inner race, Aerospace, working on the Health and Usage Monitoring
outer race, or ball fault. System. Dr. Hochmann has a Ph.D. from The Ohio State
University, Columbus Ohio studying in the Mechanical
Engineering Department.
7 REFERENCES
Dr. Eric Bechhoefer is a retired
Naval aviator with a M.S. in
[1] Goldman, S., “Vibration Spectrum Analysis”, Operation Research and a Ph.D. in
Industrial Press, New York, NY., 1999. General Engineering, with a focus on
[2] Ho, D., Randall, R.B., “Optimization of Bearing Statistics and Optimization. Dr.
Diagnostics Techniques using Simulated and Actual Bechhoefer has worked at Goodrich
Bearing Fault Signals”, Mechanical Systems and Aerospace since 2000 as a
Signal Processing, Vol. 14, Issue 5, pp. 763-788, Diagnostics Technical Lead. He has
2000. previously worked at The MITRE
Corporation in the Signal Processing
[3] Harting. D.R., “Demodulated Resonance Analysis- A Center.
Powerful Incipient Failure Detection Technique”.
ISA Transactions, Vol. 17, No. 1, pp.. 35-40.
[4] McFadden, P., D., “Model for the Vibration
Produced by a Single Point Defect in a Rolling
element Bearing”, Journal of Sound and Vibration,
Vol. 96, Issue 1, pp. 69-82, 1984.
[5] Randall, R.B, “Hilbert Transform Techniques in
Machine Diagnostics”
[6] Mitchell, J., “Introduction to Machinery Analysis and
Monitoring”, Penn Well Books, Tulsa, OK., 1993.
[7] Goldman, S., “Vibration Spectrum Analysis”,
Industrial Press, New York, NY., 1999.
[8] A., V., Oppenheim, R., W., Schafer, “Digital Signal
Processing”, Prentice-Hall Inc., New Jersey, 1975.
[9] Hochmann, D., Bechhoefer, E., “Gear Tooth Crack
Signals and their Detection via the FM4 Measure in
Application for a Helicopter HUMS (Health Usage

View publication stats