o

.-

A-.-OO3-O13

RLAERO-PROP-R-163

J.t,AUSThALIA ,-

DEPARTMENT OF DEFENCE
DEFENCE SCIENCE AND TECHNOLOGY ORGANIKATON
AERONAUTICAL RESEARCH LABORATORIES
MELBOURNE. VICTORIA

AERO PROPULSION REPORT 163

APPLICATIONS OF VIBRATION ANALYSIS TO THE

CONDITION MONITORING OF ROLLING
ELEMENT BEARINGS
by
N. S. SWANSSON and S. C. FAVALORO

FOR PUBLIC.RELEASE

.APPROVED

COPY No

COMMONWEALTH OF AUSTRAUA 1984

JANUAiRY 1984

AR-003-010
DEPARTMENT OF DEFENCE
DEFENCE SCIENCE AND TECHNOLOGY ORGANISATION
AERONAUTICAL RESEARCH LABORATORIES

AERO PROPULSION REPORT 163

APPLICATIONS OF VIBRATION ANALYSIS TO THE

CONDITION MONITORING OF ROLLING
ELEMENT BEARINGS
by
N. S. SWANSSON and S. C. FAVALORO

SUMMARY
Condition monitoring ine.lhods lor rolling element bearings, which utilize the high
frequency ibrations generated bl' bearing damiage, hare been investigated a~id compared
experitnenially. Kurt-osis value~,s,sjA'ctra of v-ibration sign-al envelopes and pulsve measvurement methods provided ej/i'ctive -detection of early'damage, but kurtosis in particular
was not' effctive in evaluating e.vtcnsive damtage, where a combination of methods is
required.
Comparison of di/jrent acceleromneter itvpes used as v'ibration vensors gave yen', similar
results, but an acoustic emission transducer provided indications of incipient damage signiJicantly earlier than the ac~clerom!ters.
Tests were conducted under reasonably./javourable conditions of measurement So
further testing under less favourable cotiditions is proposed, and the efji'ct of such conditions~ is briefly considered.

eCommonwealth of' Austal~ia 1984

POSTAL ADDRESS: Director. Aeronautical Research Labor-atories.
Bo~x 41331. P.O., Melbourne. Victoria. 3W)0, A#A~rafia'

REPRODUCTION QUALITY NOTICE

This document is the best quality available. The copy furnished
to DTIC contained pages that may have the following quality
problems:
"* Pages smaller or larger than normal.
"* Pages with background color or light colored printing.
"* Pages with small type or poor printing; and or
"* Pages with continuous tone material or color

photographs.
Due to various output media available these conditions may or
may not cause poor legibility in the microfiche or hardcopy output
you receive.

El

If this block is checked, the copy furnished to DTIC

contained pages with color printing, that when reproduced in
Black and White, may change detail of the original copy.

SUMMARY
1. INTRODUCTION
2. SOURCES O! VIBRATION, IN ROLLING ELEMENT BEARINGS

.0

2.1 Imperfect Rolling Motion

L.1.1 Manufacturing Faults
2.1.2 Faulty Application, Installation or Operation
2.1.3 Deterioration of Bearing Components in Service
2.2 Other Vibration Sources
2.2.1 Acoustic Emission
2.2.2 External Vibrations

2
2
2
2
2

3. CHARACTERISI1CS OF VIBRATION OF DAMAGED BEARINGS

3.1 Impact Action

3.2 Stress Wave Propagation and Response
3.3 Contact Frequencies and Modulation

3
5

4. VIBRATION ANALYSIS METHODS FOR BEARING MONITORING

4.1
4.2
4.3
4.4
4.5
4.6

Conditions of Measurement
Signal/Noise Ratio Improvement
Simple Level Measurement-Time and Frequency Domains
Pulse Measurement
Pattern Recognition v. Level Measurement
Probability Distribution

7
7
7
8
8
9

5. EXPERIMENTAL INVESTIGATIONS

12

5.1 Beating Test Rig

5.2 Test Bearings

12
12
14

5.3 Instrmenation
5.3.1 Kurtosis Measurement

15

5.3.2 Transducers
5.3.3 Other Instrumentatlon

16
19

6. RESULTS AND DISCUSSION

6.1 Angular Contact Bearing Tests

6.2 Deep Groove Bearing Tests
6.2.1 Kurtosls and Level Trends
6.2.2 Shock Palse Distributions
6.2.3 Envelope Spectra
6.2.4 Time Domain Signals
6.2.5 Comparison of Accelerometers
6.2.6 Acoustic Emission

"

r19

19
24
25
25
,.25
27
27
27

'

'

"

'

"

"

-;

7. CONCLUSIONS
8. APPENDICES
A-Kurtosis trend plots

n-Rms acceleration
C-Acoustic Emission trend plots
D-Sbock Pulse Distribution Curves
E--Envelope Spectra
F-Time Domain Signals
G-Pi-;.,eive Damage Details
DISTRIBUTION
DOCUMENT CONTROL DATA

27

1. INTRODUCTION

"Toobtain the greatest effectiveness in the operation of complex machines and mechanical
systems, on-condition maintenance strategies are being widely employed to minimise operating
and overhaul costs, without adversely affecting reliability and availability. Access to, and'use of
effective methods for monitoring the condition of these systems and their component parts are
essential prerequisites to the adoption of on-condition maintenance, and developments in condition monitoring techniques are currently of considerable interest, especially in areas where
machine availabiiity is vital. Service equipment, such as helicopter transmissions and turbine
engines, is one important area where monitoring methods are being sought in order to detect
performance deterioration, damage or incipient failure, and if possible to diagnose the source
of the problem and predict its sequence of development.
Data sources for condition monitoring incluue direct physical inspection (provided the
machine can be shut down periodically for this purpose), non-destructive material inspection
techniques, examination of lubricating oil and oil borne wear debris, and the analysis of dynamic
values of various parameters generated while the machine is in operation. Of the latter, the
most commonly used parameter is machine vibration, although noise, shaft torque and other
parameters can be used when they provide significant data., Analysis methods for dynamic signals,
such as vibration, aim to derive the maximum condition information from the available data,
and methods to enhance the usefulness of this information are constantly sought.
Vibration analysis has been extensively used for evaluating the condition of gear transmissions since distinct, discrete meshing frequencies can be identified; spectrum analysis using
the Fourier Transform analyser has virtually become a standard technique for this purpose.
Rolling element bearings are equally vital machine components, but the vibrations generated
by both sound and damaged bearings are much more complex and discrete, frequencies ara not
so readily, identified and distinguished. This paper examines some of the vibration analysis
techniques particularly applicable to the monitoring of rolling element bearings, and describes
th, findings of some supporting experimental investigations.

2. SOURCES OF VIBRATION IN ROLLING ELEMENT BEARINGS

Under normal operating conditions rolling element bearings always generate some degree
of vibration and noise, albeit at very low level for a properly installed bearing in good condition. Sources of vibration include imperfect rolling r-otion' during rotation, acoustic emission,
and external vibration tran~smitted via the bearings.

S2.I mperfect Rolling Modem

Motion which is not perfectly regular is the major source of bearing vibration. Causes of
irregularity can be classified into areas of manufacture, application and operation, and
deterioration in service.

"*

2.1.1 Mamifacturing Faults

Departures from perfect geometry are the main items in this category. These include out of
roundness of inner or oute race, lack of sphericity of balls, and unequal ball diameters. A
possible secondary cause is lack of uniformity of bearing materials; e.g. varying local modulus
due to inclusions etc.

2.1.2 Faulty Application, Installation or Operation

Some faults in this classification result directly in imperfect rolling motion; others have
obvious mechanical effects. The former category includes assembly misalignment and race
distortion, as well as brinelling of the bearing races (caused, by overloading or impact either
during assembly or operation). Looseness or similar faults produce direct mechanical impact
effects.
Poor or inadequate lubrication can lead to friction and sticking with irregular motion or
vibration. At the highly stressed Hertzian contact area, if there is no separating layer or lubricant,
the contacting surfaces will weld together and then be broken or sheared apart. Since the contact
area is very small only a trace of lubricant is required but this trace is essential (the main purpose
"ofgreater oil supply is to remove heat from the bearing, preventing deterioration of compon'ents
due to overheating). Wear particles or other minute debris in the oil produce vibration effects
similar to defective lubrication, since very small particles can penetrate the lubricant: film,
effectively resulting in metal to metal contact.

".
"-

2.1.3 Deterioration of Bearing Components in Service

Continued repetitive loading of a bearing will iead to surface fatigue, manifested as pitting,
flaking of races, balls or rollers. Other progressive surface damage mechanisms
include corrosion,'fretting, abrasive wear (due to contaminants in the oil or some other external
source), and other wear mechanisms associated with defective lubrication. All these faults can
be regarded as departures from ideal rolling geometry and, depending on whether they occur
in the inner race, outer race, ball or rol!er, generate vibration with the same characteristic
periodicities, or repetition frequm.'cies, as geometric irregular~ities.
Early detection and evaluation of this progressive type of damage. 'which if continued
leads to total bearing failure, is the prime reason for monitoring bearing vibration.

"spalling or

2.2 Other Vibration Sources

2.2.1 Acoustic Emission
Acoustic emission is generated by very small scale plastic deformation or crack propagation
in high stress regions. Slip, dislocation movements, or fracture release energy manifested as a
propagating stress wave. In rolling bearings, the high stress levels are found in Hertzian stress
regions just below the contacting surfaces bet,,ecn ball and race. Significant acoustic emission
levels thus provide an indication of sub-iurface fatigue crack growth, likely to'be followed soon
after by surface failure such as pitting or spalling.
2.2.2 External Vibration

.[
.

Measurements of bearing vibration aie complicated because part of the signal detected
may be generated elseN here idithe machine. It is often diflicult to distinguish the external sources
from intrinsic bearing vibration, especially as the bearings form the transmission path between
vibrations originating on the rotating assembly and the stationary bearing'housing or machipe
casing whei'e sensors are located. Furthermore; both external and intrinsi: sources miy excite
"resonances of races characteristic purely of the bearing,, making their discrimination virtually
impossible.

3. CHARACTERISTICS OF VIBRATION OF lDAMAGEDI-BEARINGS

The various types of surface damage in hearinigs. associated with deterioration proctssci
result in geometric changes in the surfmce f'orm or topology of the bearing elements. Manufacturing errors result in similar departures from true geometric shape: the major difference between
such errors and damage being one of scale. Initially at least; surface damage due to fatigue and
similar processes is at a local, almost microscopic, scale: for example A typical initial size of the
fatigue pit is about 0 2 mm. Damage caused by corrosion or fretting covers a larger area, but
local surface roughness remains of small size.

The nature of surface irregularities produced by damage governs the mechanism and
characteristics of vibration generated by thesc featurcs. The smai' size of the irregularities means
that contawl between rolling element and irregularity is of very short duration; effectively it
is in impact, action. A simple model of this contact action first given in [1] clarifies the significant
parameters.
3.1 Impact Action
Fig. 3.1 shows a ball which is assumed to roll into a defect with instantaneous rotation
centre fixed at A, until it impacts the far side of the defect at B with velocity u.. If the ball rolls
without slip on the inner race at C, then for a small defect size x,
=
x dix
Us =V

2 dB'

where
ti = velocity at point C

wi = rotational velocity of inner race

di = inner race diameter
dB = ball diameter.

Ul

The energy transferred to the outer race at impact point B is proportional to the square of
the velocity. Hence,
d12
"U12=
W2 X2
x4 dB2

(3.1)

The response of the impacted bearing race, as measured by stress, particle motion (e.g.
acceleration) or other parameter, is proportional to the energy at impact. Equation (3.1) shows
that the parameters governing this response include the defect size, bearing geometry and speed.
Bearing load does not directly affect. the response, provided that the ball/defect contact conditions shown in Fig. 3.1 are maintained, i.e. the ball must fully engage the defect. Under some
conditions, such as light load, full contact between ball and track may not be maintained, preventing full engagement of the defect, and the impulse response will then be reduced.
A measure of impact duratidn is the total time in which the ball is in contact with the impact
point B. During this phase it is assumed that the ball rolls out of the defect with fixed centre of
rotation B, so to clear the defect the ball centre P. having velocity vp, must roll a distance xp along
the pitch circle dp (see Fig. 3.1).

"-"

2 4
*

rd

*XP= -j

"Thetime of contact ta is:

xP
,

xdp 4
I?
2do ,d4

"or,since
42

d4dt approx.

"ta= "widp

(3.2)

For typical rolling bearing applications in engines and transmissions tR is of the order of
tens of microseconds, e.g. for the bearing tests described in this report with a defect size
x = 0.2 mm, t a = 15 micioseconds. The actual rise-time of the impact will be a small fraction
of the total time of contact, and consequently stress wave frequencies excited by, the impact can
extend into the megahertz region.

Ii3

xp

.do

!:Vi
V".

,1d,

."0.

FIG 3.1

BALL CONTACT WITH DEFECT

13.2 Stress Wave Propagation and Response

When a rolling element contacts a defect it produces an impact which transfers energy into
the bearing race. This impact energy, in the form cf a stress wave, propagates outwards from
the impact point at sonic velocity. When the stress wage reaches the interface between the
bearing race and another component, the major part of the incident energy is reflected back
into the bearing race and a small proportion is transmitted through the interface into the adjacent
component. Damped r:sonan, cscillations of the bearing race and adjacent component, and
other components in the transmission path, are set up by multiple reflections of the propagating
wav'e.
The short impact duration and resulting brief rise-time of the stress wave frort means that
a wide range of natural modes and frequencies can be excited in t'e bearing race. The simpler
modes of the race are most readily excited, and because the race compoients are physically
small, corresponding natural frequencies are high relative to general machine resonances.,
The transmitted signal is attenuatel as it passes through each interface, especially at high
frequencies (50 kHz or greater) where the attenuation becomes severe. High frequency attenuation
vaties a great deal with details of design, fitting of components and installation practices. One
reason for this is that at an interface contact is made at only a few engaging asperities and the
true contact area is only a small fraction of the apparent area. Elsewhere the components are
separated, and at high frequencies the separation distance can exceed the vibration amplitude,
;o that most of the surfaces do not make contact. T.ie constraint applied by the contact depends
on a fortuitous combination of factors. including surface finish, accuracy and tightness of fit,
presence of grease or lubricant (which can act as a couplant transmitting pressure waves between
surfaces), so that the transmissibility is not readily predictable.
The foregoing discussion has referred to stress waves generated by the impulhe action of a
"rolling element contacting a defect. It applies equally when acoustic emission mechanisms
"produce stress waves. As previously discussed, propagating fatigue cracks can generate acoustic
emission, and since the motions involved are very small with correspondingly short durations,
frequencies produced ;an extend into the megahertz range, albeit at low signal level. The low
amplitude means that attenuation in the transmission path is a highly important consideration,
and satisfactory measurement of acoustic emission requires that the transducer be located in a
position with a favourable transmission path from tCe bearing.

*.

3.3 Contact Frequencies and Load Modulation

So far, phenomena following an impact when a single ball or roller contacts a defect have
"been desribed. In an operating bearing a series of impacts occurs as successive balls contact the
defect, generating a sequence of damped oscillation bursts superimposed on a random vibration
"background as shown in Fig. 3.2(a). The repetition fiequency of the impacts and associated
oscillation bursts depend on whether the defect is in the outer or inner race, or ball. If it is assumed
that the balls roll without slip, repetition frequencies can be predicted theoretically [21 from the
TABLE I
Contact Frequmdes is Ball W, Roiler Berinp
ft

':::.,.
t

S..

..

frequency of rotation cf inner race

Other notation as in Fig. 5.2

Outer rqc~e defect

Inner ruce defect

Nf

Ball or roller defect

.i

(i-:Cos
:

d.Hos(
t.2

r I''I rrr i r
!'lll l U~~Pd

SL

00

.0

4-.m

00

wr

-1
foirmulae given in Table I (slip modifies the predicted values). Manufa,turing irregularities
and faults such as lobing or ovality, can produce additional freguencies with interaction effects
leading to numnerous possible sum and difference frequencies as described in [3].
If the orientation of the load on the bearing remains fixed relative to the defect, successive
pulse bursts are of roughly equal amp itude, as shown in Fig. 3.2(a). However, if the load
orientation rotates -elative to the defect, the impact amplitude is reduced in the unloaded regicn,
so successive pulse amplitudes are modulated as shown in Fig. 3.2(b). A similar effect occurs
in the less common event of a ball or roller defect, although the resulting modulation pattern is
more complicated.

4. VIBRATION ANALVSIS METHODS FOR BEARING MONITORING

4.1 Condiions of Measurement
The most effective vibration analysis methods for rolling element bearings are those which
ufilise the particular cnaraceristics of bearing vibration. Selection of such methods is of greatest
impo'rtance when measurements must be performed under unfavourable conditions. Ray [41
states that for favourable conditions of measurement:
(i) The machine mu:;t not have any source of impulsive vibration other than bearings.
(ii) Bearing speed and load must be of sufficient magnitude to generate impulses.
(iii) There should be a direct transmission path from the bearing to the transducer mounting,
with only one interface between bearing race and mounting face on the casing, and
the mounting should be reasonably central relative to the bearing load zone.
Favourable conditior- may be realized on simple machines such as pedestal bearings and
pumps. Most complex machines however lack favourable conditions in one or more respects,
with measurement difiulties ranging from slight to severe. In particular, adverse conditions are
found in aircraft gas turbine "ngines and helicopter transmissions.
4.2' Signal/Noise Rati3 Improvement
Adverse conditions result in a poor signal to noise ratio, since they may lead to:
(i) Low signal-caused by excessive attenuation in the transmission path. or low bea ng
load or speed,
(ii, High noise (i.e. unwanted interfering vibration) from sources in the machine other than
bearings.
Unfortunately techniques such as synchronous time averaging, which are of great value in
U

"improvingsignallnoise in gear transmissions where meshing frequencies are an integral multiple

of shaft speeds, are not applicable to, bearings.

The most useful available technique is banlpass filtering which entails viewing the signal
through a frequency 'window'. The filter window is selected to encompass frequencies generated
by impulses ard to exclude other machine vibralion as far as possible. For maximum sighal/noise
ratio, the window should preferably be centred on one of the system resonances.(which will be
excited by impulses) or the resonance of the measuring transducer. The-latter method is evidently
independ.nt of applicatior, and has been adopted in some commercial measuring systems.
4.3 Level Measurement-Time and Frequency Domains

'

"Because vibration generated by rolling elemert bearings is highly complex, simple measurement of total (rms) vibration level is userul foe monitoring bearings-only when measuring con"ditionsare highly favourable. The extra discrimination of octave or 113 octave analysis does not
significantly enhance the effectiseness of level measurement.
.71

Standard narrowband sp,.ctrum analysis techniques, normally implemented on a Fourier

trarisfoim analyser, do not provide the clarification dhat might be expected. Most of the bearing
vibration energy occurs in a band of frequencies around the pulse excited resonances; these
frequencies depend on the structure or transducer and are not related to the bearing balI passing
frequencies. Spectral lines at ball passing frequencies can be identified in some cases, but often
tend to be submerged in the general noise level. The small defects associated with early bearing
damage do not produce major changes in energy n any particular part of the spectrum, so spectrum analysis, except in favourable conditions, is only capable of detecting substantial bearing
damage.
If after bandpass filtering, which removes extraneous low frequency and noise signals, the
vibration signal is enveloped, high frequency structural or transducer resonances are removed,
but impule repetition frequencies remain. Frequency spectrum analysis of the enveloped signal
clearly reveals ball passing frequencies with the amplitude of these components a function of
the impulse energy. Trends in the amplitude of such components can reveal change or deterior-.
ation in the bearing, and the frequency value enables the origin of the signal to be attributed to
particular bearings.

4.4 Pulse Measurement

A direct indication of bearing condition implemented in a number of instruments, is obtained
by measuring the total energy of the impulse bursts, using a frequency window matched to the
pulse oscillations. In the shock pulse measurement system of SKF, the oscillations in each
pulse are enveloped and then integrated, the total area or energy is termed shock value. A peak
hold circuit indicates maximum shock value of a pulse train. The system also incorporates a
comparator with an adjustable thireshold setting for shock value, so that the rate of shocks
exceediag the present level can be determined--effectively giving a cumulative distribution of
shock values. This distribution ha's been found to give a better correlation with bearing condition
than maximum shock value [5].
Other methods used for measuring pulses determine average energy in a pulse train.
Background noise is excluded by measuring only the signal exceeding a certain threshold level,
which may be set either manually or automatically.

4.5 Pattern Recognition v. Level Measurement

All methods so far described essentially measure the amplittide of the vibration signal.

This applies irrespective of the prior processing used to enhance the significance of the signal,
and irrespective of whether it is measured in the time or frequency domain. Amplitude or level
measurement has two ignificant limitations:
(i) Vibration level varies with sexeral independent parameters such As bearing load and
speed. For different records to be comparable they must be acquired under the same
operating conditions.
0i) Some vibration is always present, even with a. machine whose bearings are in new condition, so deterioration or damage can only be determined by comparison of current
records with those tatn from a similar machine whose bearings are known to be
in good condition. 'Cenerally this is achieved by taking 'baseline' records when the
machine is virtually new, the'n observing trend's as successive records are taken during

the I~fe of the machine.

Pattern recognition Methods, unlike level measurement. characterize intrinsic features of
the vibration signal which forms its own refererice. Generally, vibration patterns are relatively
independent of machine operating parameters, so reproducing identical test conditions for'comparison of records becomes less critical. In addition, the need for a baseline signature is obviated,
as the methods are inherently independent of signal level..
Experience has shown that while~pattern recognition methods are very sensitive to-change
and thus provide.early indication of damage, their major limitation, when'applied to bearing

"*

"8'

signatures, arises when damage is increasing. Under these circumstances, the pattern tends to
revert to that of the undamaged signal, albeit with much higher vibration levels. To cope with
this situation, pattern analysis may need to be complemented by level methods.

"4.6 Probability

Distribution

The distinctive feature of vibration generated by damaged bearings is a periodic series of

impulse generated osciHation bursts, sometimes modulated because of load variation effects.
A random background vibration combines with the impulsive oscillations, tending to obscure
the impulsive pattern as signal/noise ratio deteriorates.

Direct measurement of pulse energy is satisfactory when signal/noise is adequate but with
a less favourable ratio more sensitive methods are required. Such methods include spectrum
analysis of the signal envelope, which determines pulse amplitude and periodicity, and analysis
of statistical properties (in particular those derived from the probability density distribution) of
the vibration signal.

Al

Probability density p(x) of a time varying signal x(t) is defined such that the probability
of the signal value lying between x and x + dx is p(x)dx. Since the vibration of an undamaged
bearing consists of the combination of numbers of sepaiate independent effects, the central limit'
theorem indicates that its probability density will tend towards Gaussian. This theoretical
prediction is confirmed in practice; Fig. 4.1(a) shows the probability density distribution of a
bearing in good condition, with a superimposed Gaussian curve. The same data is plotted-in
Fig. 4.1(b) with a logarithmic probability scale, and the Gaussian distribution appears as a
parabola.
Deterioration and damage in a bearing lead to changes in the probability density. Damped
impulsive oscillations in the time domain signal, of significant magnitude relative to the Gaussian
noise of the undamaged bearing, lead to greater occurrence of extreme signal levels manifested
by changes in the tails of the probability density distribution. The higher probability of extreme
vaiues is shown in Fig. 4.2(a) and more clearly in the logarithmic plot in Fig. 4.2(b).
Of the various statistical measures used to quantify the shape of probability density curves,,
one most easily measured is the crest factor-the ratio of peak value to standard deviation (rms).
For a periodic signal this provides an absolute definition; but for a signal with random components peak value must be defined as the value exceeded at some defined relatively low
"probability level. (The existence of this pr;.3bability may not be apparent in analogue peak or
crest factor measurement, but it is implicit in the time decay constants of the measuring circuit).
Crest factor measurement provides a useful preliminary indication, but more sensitive
methods are availalble for use in adverse conditions, An approach which is simple but sensitive
utilizes the statistical moments Mn of the probability distribution p(x". The moment MA of

order n is defined as:

JM

(xOr.p(x) x

14.1)

As defined the first moment is zeio and the second moment is familiar as the variance, s2 . Odd

*"of

moments provide a measure of the asymmetry or skewness of the distribution; even moments
measure its spread with the higher order moments responding to occurrence of extrtme values
"inthe tails of the distribution. Th.- fourth moment M4 has been found to be sufficiently sensitive
without responding excessively to occasional measurement perturbations. This moment is
normalised relative to standard deviation; the ratio M4/M22 = M4/s 4 is called kurtosis, K.
The kurtosis of a sional with Gaussian probability distribution is 3; vibration measurements
on good bearings give kurtosis values close to 3. As bearing damage develops, with the generation
impulsive oscillation bursts (e.g. Fig. 3.2) kurtosis values rise, as shown in te corresponding
probability density plots (Figs 4.1 and 4.2),<
9

p(x)
0.5'-

p(xJ

Jr.

-I-.-

0.1

'0.2H-

.1.
0.001

U
*

--

(a)

Ib)

3.2

FIG 4.1

PROBABILITY DENSITY DISTRIBUTIONS

iO

SOUND BEARING

0.

pp"W

0.r
0.1
0.

~--

'--01

.'".

01

0.1'

t,,

~,

..

'..i

-"""
-

...Q

--

0.001

'-

"(a)

(b)
K = 6.5

,11

FIG 4'.2

PROBABILITY DENSITY DISTRIBUTIONS

-DAMAGED

BEARING

0
"5.EXPERIMENTAL INVESTIGATIONS
A number of bearing and gear rigs have been in use for some time for the purpose of
evaluating condition monitoring methods. Initially, kurtosis measurements were made on
angular contact bearings, which were being used in other tests for the evaluation of shock pulse
measurement [5]. After promising initial results, a program was initiated to test a number of
deep groove bearings to failure, with the aim of eva!uating kurtosis measurement and comparing its effectiveness as a condition monitoring technique for bearings with other methods
"such as shock pulse, rms acceleration and envelope spectrum measurement. In addition comparison of tra ,sducers with different resonances, including acoustic emission, would be made.

5.1 Bearing Test Rig

The rig comprises two bearing housings, one containing the 30 mm test bearing, the other
containing two angular contact bearings arranged in tandem which react the test load. Axial
"load is applied to the outer race of the test bearing by screwing a large nut into the housing,
thereby putting the shaft in compression with a balancing tension in a surrounding tube, which
"Isfitted with a strain gauge bridge for load measuremrnt.
Cooling is obtained by jet impingement of a synthetic oil, and power is supplied to the rig
by a Ward-Leonard set with a DC motor providing a variable speed drive.
A photograph of the rig is shown in Fig. 5.1, highlighting the principal features and the
location of various transducers.

S-

5.2 Test Bearings

"Two types of ball bearings, of 30 mm bor-, were used:

(i) Angular contact bearing-ABEC 9 precision class (Barden 106Bx48).
(ii) Deep groove ball bearing (SKF 6006/C3).
The principal dimensions of both bearings are shown in Fig. 5.2. Tible 5.1 gives contact or
repetition frequencies for tde bearings at the test speed of 6000 rpm, calculated from the
formulae given in Table I.

TABLE 5.1
Ball Passing Frequencies
Frequency (Hz)
Component

Angular Contact

Deep Groove

Ba'l Passing-Outer Race

587

453

b all Passing-Inner Race

813

647

Ball Rolling

583

549

Based on shaft frequency 100 Hz (6000 rpm)

'

12.

""
*

C.M.L. & B & K

accelerometers
on this face

Loading
nut

Location of
acoustic emission
transducer

Test bearing
housing

Kistler
accelerometer

Strain gauge
bridge/tension tube

Shock pulse
transducer

FIG 5.1

Location of PCB
accelerometer

BEARING RIG

13

Reaction
housing

The inner and outer races of the bearings are bonded to steel sleeves, which are supported in
the housing and on the shaft at three contact points as shown in Fig. 5.3- Resonant frequencies
hame been calculated for the bearing components and the bearing/sleeve assembly (assuming
ideal bonding) for lower modes using formulae in [6], and are given in Table 5.2.

TABLE 5.2
Natural Frequencies (kHz) of Bearing Components

"No. of
diametral
nodes

Mode type

Angular contact
Alone
With sleeve
Inner Outer Inner Outer

Deep groove
Alone
With sleeve
Inner Outer Inner Outer

RaceIn-plane
Flexure

0
2
3

49-1I
8-2
23.1-

32-1
3.5
9-8

40-6
16-4
46-5

27-8
8-4
23.8

48-7
8-7
24-5

32-3
3-8
10-8

40-3
16-6
47-1

27-9
8.7
24-6

RaceTransverse

2
3

7-9
22-7

3-4
9.7

15.9
45.7

8-1
23-4

8.4
24-1

3-7
10-6

16.1
46-3

8-4
24-2

Flexure

BallEllipsoidal

437-0

414-0

5.3 Instrumentation
A diagram of the instrumentation arrangement is shown in Fig. 5.4. Not all equipment
illustrated was in use at the one time; in the preliminary tests. on the angular contact bearings,
a single accelerometer was used and its output, after signal conditioning and capture on the
digital waveform ecorder, was processed on the ARL DECsysteml0 site computer: In the
later tests on the dt.p groove bearings, additional accelerometers, including an acoustic emission
transducer, a shock pulse measuring system and an on-line kurtosis meterwere used. Much
"fasterdata processing was provided with the addition of a dedicated VTI03 microcomputer to
the system.

5.3.1 Kurtosis NMeasurement

Kurtosis was calculated as follows on the DECsysteml0 and VT103 computers using
S'ita records captured by the digital recorder. Ifa random sample of N points,
-V,
,aken from the time signal X(t) (uniform sample i-terva!s may be used provided
.......
Veriod is unrelated to any periodicities in the signil) the formula for statistical moments
o. equation (4.1) becomes:

N
."

_(I-,)15I)

Mn =~~iN

14

N(5.

to l

0Ii
ti

ti

-p

-p

I'I

(a)
Angular. contact (BARDEN)

(b)
Deep groove (SKF)

Dimensions (mm)
Bore
*Outside

di
D
dp42.5

diameter
Pitch diameter
Mean thickness
Inner and outer
Width
Ball diameter
Contact angle
No. of balls

t1

to
b,13
0
N

FIG ,.2

(a)
30
55
3.6
3.6
7.1
150
14

BEARING DIMENSIONS

(b)
30
55
42.5
3.9
3.9
13
.7.5
1

and so kurtosis is:

N

N
K

(Xi -X)4
i=1

M4

(Ml2)2

(5.2)

~(Xi_,f)

--

At high sample rates (i.e. when examining spectra of high frequency windows) the total sample
"time was less than one shaft revolution-sometimes only one or two impulse events-so concatenated records were averaged until acceptable convergence was obtained, using the formula:
L
L

M4j

".-

tj=- 1

j~l(5.3)

(for records of equal length)

The Kurtosis meter (K-meter) implements this measurement process in one instrument,
Input signals are bandpass filtered and then digitized. The limits of each frequency window or
band (the same limits as those used on the bandpass filter prior to computer calculation of
"kurtosis) are given in Table 5.3. The K-meter incorporates a microprocessor which computes
kurtosis, rms acceleration and velocity. Output of the envelope of each band is provided for
external spectrum analysis, but the gain of the envelope circuit is variable and is not calibrated.
K-meter operation and readout is oriented toward the non-technical user: using kurtosis
and rms acceleration values, an algorithm in the micro-processor gives a one-shot diagnosis of
bearing condition-*Good', 'Early', 'Advanced' or 'High Damage'.
TABLE 5.3
Frequency Bands
Band No.

"I

,*

Frequency Cut-off kHz

2.5

2
3

5
10

10
20

20

40

"5

40

80

5.3.2 Transducers
A secondary objective formulated after the initial tests was to compare the effectiveness of
various transducers in detecting bearing damage; so in the main veries of tests four transducers
were used. measuring acceleration and acoustic emission. Characteristics of these transducers
are given in Table 5.4. Full characteristics were not, available for the transducer supplied with the
Shock Pulse Meter, but its nominal resonance frequency is known to be about 32 kHz.
16

CL~

w
_

0r.

c-J
LU

0o

ZL

17z

~TEST BEARING

-,.

S~HOUSING

'I
"SKF

rF"

DUNEGAN

KISTLER
MOEL
818

B& K
MODEL
4339

PCB
MODEL
309A

PIEZOTRON
MODEL I
1583C3
L

4
CHARG
MODE
A

MODEL
464A
,CP MODE

CML

SKF
'MODEL21AI
SHOCK PULSE

DLNEGAN
MODEL
D9203

S1
CMIL
KURTOSIS

ROCKLAND
MO'EL 452

LDUAL HI/LO FILTER

MET2R

.,Envelope

"signal

BIOMTION
MODEL 1010
WAVEFORM
RECORDER

HEWLETT
PACKARD

::

5308A I
IMODEL
COUNTER

ENELO;E
ARR
C

INTERFACING

"DECIVT-

103,
COMPUTER

FIG 5.4
'S
,.

10

OR
'

.INSTRUMENTATION SYSTEM

>,

"

TABLE 5.4
Transducer Characteristics

Manufacturer

Kistler
Bruel & Kjaer
PCB
Dunegan;'Endevco
CML (K-meter)

Type

Model

Sensitivity

Nominal Mounted
Resonance-kHz

Accelerometer
Accelerometer
Miniature
Accelerometer
Acoustic
Enission
Accelerometer

818
4339
309A

9- 85 mV/g
iI -0 pc/g
5 mV/g

34
57
120

D9203

-64dB re I V/hbar

> 200

100 mV/g

28

5.3.3 Other Instrumentation

Other instruments shown in Fig. 5.4, excepting the Shock Pulse Meter, are standard types
and need no detailed description. The principle of shock pulse measurement is outlined in
Section 4.4; a more detailed description of its implementation is given in 15].

6. RESULTS AND DISCUSSION

6.1 Angular Contact Bearing Tests

Several high precision angular contact ball bearings, as shown in Fig. 5.2. which had already
sustained varying degrees of damage, were available for kurtosis measurement. The results of
these tests are presented in Figures 6.1-6.4. which include plots of kurtosis in various frequency
"bands under different conditions of rotational speed and axial load, and for increasing degrees
"ofdamage.
One bearing was tested at a speed of 6000 rpm and a load of 4500 N. After 65 hours the
bearing had sustained spalling damage of the kind illustrated in Fig. 6.1. Such damage causes
substantial impulsive action and these effects are well illustrated in the kurtosis trend plots
in Fig. 6.2.
A second bearing, which had already sustained spalling damage to the inner race, was tested
Uwider low load of 445 N and varying speed conditions. Measurements were also taken from
sound bearings located in the reaction housing and the results are shown in Fig. 6.3.. Testing
continued at 6000 rpm and increased load of 4500 N: these .results are shown in Fig. 6.4.
It can be seen that sound bearings produce kurtosis values closeto 3,. regardless of load and
speed, conditions. For damaged bearings, kurtosis varies with these conditiohs in a complex
manner, though in all' cases there is a definite indication of damage with kurtosis values significantly greater than 3 in at least some of the frequency bands.
At low speeds impact energies are smaller (eqn 3.1), and impulse amplitudes may a!so he
reduced because of imperfect ball/track contact at low loads. Together these effects may lead to
impulse levels being significantly reduced relative to background noise. so in the low frequency
bands kurtosis valies reveal no damage (Fig. 6.3). Continuing higher kurtosis values in the high
frequency bands, under the same conditions, are more dillicult to explain. Other impulse
generating mechanisms associated with propagating bearing dlamage. and having greater energy
"content in the high frequency hands, may be significant in this situation.
Figs 6. 4 (a) and 6.4(b) show that, starting in the lower frequency bands and progressing to
the higher, the kurtosis of a severely damaged bearing tends to revert to the undamag-d value
of 3, albeit with a much higher general vibration level (not shown on these plots). This
characteristic has been widely observed (it is seen in many of the results later in this report)
and must be taker into account when kurtosis measurement is used for bearing condition
monitoring.,
*

' 19

*(a)

Inner race spallI size:

.75 mm x.5 mm

(b) Outer race spall size: 2.5 x 2.0 mm

*FIG

6.1

DAMAGE TO BEARING AFTER 65 HOURS RUNTIME

20

53 hours

63.5 hrs.

5
T
0

'~

*I

I,

III

.
2

I
4,

Frequency band

64 hrs.

64.5 hrt.

we.

4-4

.1

Frequency band

FIG 6.2

KURTOSIS Vs. FREQUENCY BAND PLOTS FOR A'NEW BEARING AT TIME

0 HOURS UNTIL 65 HOURS, RUNNING-AT 4500 N LOAD AND 6000 RPM.
21

500 rpm

'4

--

--

40

""0

2000 rpm

Frequency band

Key:
(1) Frequency band
1
2-5 kHz
5-10kHz
2
3
10-20 kHz
4 = 20-50 kHz

rpm

(2) Bearings
,o.
o = sound

5000 rpm

0-0

66

m'

(3) Axii' ,oad

4.

131.5 hours

2xl mm spall
inner race

Frequency band

FIG 6.3

KURTOSIS (K) Vs. FREQUENCY BAND AT VARIOUS-SPEEDS FOR SOUND

AND DAMAGED BEARINGS.
22

445N

Damage(
bearing

55

21

-- 0=3

0....

1,

-- o

Frequency band
Wb

(a)

-7w

Id(a)

FIG. 6.4
Damaged bearinig of Fig 6.3: Load - 4500N, Speed
(b) Same bearing, I5 muniotes later (131.75 hours),
(c) Inner race of bearing at 131.75 Itours
2a4

6000 rpm

Sound
bearing
0--- -0

When damage consists only of a few discrete pits or spalls, the oscillations generated by one
impulse decay back to the background noise level before the next impulse generating defect is
encountered, and this impulsiveness is manifested by high kurtosis values of the signal. But with
severe damage, such as Fig. 6.4(c), there are multiple closely spaced impulse actions, and the
oscillations produced by one impulse do not have sufficient time to decay back to the noise level
before the next impulse is encountered. The total vibration signal consists of the cumulative
response to this impulse sequence, which liecause of the random nature of the impulses tends
towards a Gaussian probability distribution with kurtosis approaching 3. This effect is most
pronounced in the low frequency bands, because these signals have the longest decay times.
The existence of transducer and structural resonances affects results in certain frequency
bands. Impulse action will excite resonances both in the rig and in the transducer, and relatively
high vibration signals can be detected at structural and transducer resonances. Herein is the main
significance of these resonant frequencies; frequency bands or 'windows' centred on these values
should record reasonable signal levels for kurtosis measurement. For example, Fig. 6.2 suggests
the presence of a structural resonance in band 2 (5-10 kHz), since increased kurtosis values were
first detected in this band.
Values of natural frequency for the bearing components have been computed (Table 5.2),
but which frequency will be most readily excited depends on. a precise, almost microscopic
definition of the conditions of constraint and so cannot be predicted. A similar computation
was made for the bearing housing using a finite element model, revealing a number of possible
modes in the frequency range (Table 6.1), but again it is not possible to predict which modes
will be excited.

TABLE 6.1
Bearing Housing Assembly Natural Frequencies (kHz)
from Finite Element Model
Boundary Conditions-Base Restrained in Plane
Symmetric Mode

Antisymmetric Mode

"7.3

7-2
13"0
14-8
15.4

11-5

"-148

-'"15.1

"-"19"8

.15 7

18.1
20-8

20'5

21-5

22"4

23 - I

To resolve this uncertainty, experimental measurements were made of the impact response
of the bearing/housing system. 'Two different types of transducer were used, enabling structural
and transducer resonaices to be distinguished. Results are shown in Fig. 6.5. and a major
structural resonance at 8.5 kHz is evident, corresponding to the relatively simple bearing race
"mode with 2 diametral nodes (Table 5.2). Other structural resonances seen in these response
curves could also be useful in selecting frequency windows ror vibration measurement.
S.4

"6.2 Deep Groove Ball Bearing Tests

Four standard commercial quality deep groove ball bearings, as shown in Fig. 5.2. were
available fqr testing. At first only shock pulse measurement and kurtosis estimation by computer
were available, but the' K-meter became available early in the tes. series, thus providing an extra
b"
42

24

transducer with alternative kurtosis and level estimation, and a signal envelope output for spectrum analysis. Since the envelope output has a variable gain, no significance could be attached
to the absolute level of spectral lines, but late in the test program, a locally developed circuit
was obtained, giving a calibrated envelope output for spectrum analysis.
Kurtosis, g rms levels, shock pulse distributions and envelope spectra for these bearings
(designated A, B, C, D) are plotted in Appendices A, B, C, D and E. A description of bearing
condition at times of inspection is given in Appendix G, with accompanying photographs.
The four bearings, which were operated at the same speed and load (6000 rpm and 6600 N),
had widely different lives. Life to the first indication of damage varied from 1*4 to 136 hours,
and to the conclusion of testing from 2 -5 to 152 hours. However, damage patterns during the
tests were similar. for all bearings. The following discussion is primarily centrted on the results.
from bearing C, since this bearing was fitted with the greatest range of transducers and
measuiring systems. Generally, results for the one bearing are typical of all, but comparisons
with other be.irings are made where appropriate.

6.2.1 Kurtosis and Level Trends

r.

--

Changes in kurtosis values as bearings deteriorated in operation foflowed a fairly typical

pattern. First damage indications were increasing kurtosis in low frequency bands, with progressive damage leading to a fairly steady plateau with high kurtosis values in all bands, and
finally, kurtosis values reducing, especially in the low bands, with extensive damage. This pattern
is illustrated'by bearing C where kurtosis values greater than 3 were first. observed at about
I1I hours with the Kistler transducer (Fig. A.7) and the K-meter (Fig. A.8) and with tile PCB
transducer (Fig. A.9) when first fitted at 12 hours. Physical inspection of the bearing at this time
25
revealed a spall about I x 3 mm in the inner race.
With further testing, kurtosis values remained
high, but after 17 -7 hours they tendcd to reduce to undamaged values, primarily in the lower
frequency bands. Inspections at 15 hours revealed that the inner race spall had grown to about
2 x 3 -5 mM~. with 'bruise' marks (due to crushing of re-entrained wear particles) all round the
outer 'race. At 18 -5 hours, there were two large spall areas, 8 mm and II mm long separated
by about 2 mm on the inner face, with severe bruising over the remainder of the inner and outer
races. The test wvas concluded at 20 hours with about 180' of the inner race completely failed,
two more spalls beginning on the remaining section of the race, severe bruising and one spall
on the outer race. a&nd spalling and pi, ..g apparent on the balls (Figs G.5-G.9).
Simultaneous measurements of total (rms) vibration showed levels increasing slowly in
the early damage phase, and much more'rapidly as bearing' damage became severe (Figs 8.7,
8.8, 8.9). The first noticeabec increase in level with bearing C occurred after 14 hours (Fig. 8.8),
some 3 hours after the first registered change in kurtosis.
6.2.2 Shock Pulse Distributions
Shock pulse distribution curves revealed a definite and consistert change with early damage
(Fig. D.3-bearing C at II -5 hours) and progressive changes with increasing and, severe damage.

6.2.3 Envelope Specra

Envelope spectra of deteriorating bearings followed a pattern sirnilar to kurtosis. Spectra
of undamaged beariigs revealed random spectral distributions with no identifiable frequency
peaks. Once early damage occurred, the spectra changed markedly and the various characteristic
ball passing frequencies became apparent. With progressive damage, amplitudes at these
frequencies increased relative to background noiw,. When the damage became severe, the threshold noise increased but unlike kurtosis, which at this stage tended to revert to undamaged valuies,
spectral lines at the characteristic frequencies were u.ill clearly evident. albeit at reduced magnitudes. Fig. E.2 clearly iiistrates the pattern of rise and f'ail in relative magnitudes of spectral
lines at ball, passing frequencies, While Figs E-3-.E.5 show a similar pattern for abs~lute
magnitudes when the locally made circuit was used ont bearing P

Amp (g)
0.3-

3.

IKistler transducer

29.5
0.2H

~35.5

24.5

10

33 kHz resonance

20

30
kHz

40

Amp (g)
0.2-2.

50

60

B & K transducer
kHz resonance
-~57

8.5
29.5

01-

15.5

5.

0.1

YL'40.0

'1

010

20,

30r

0s

kHz

FIG3 6.5

HOUSING IMPULSE RESPONSE

"26

60

Nearly all spectra show contact frequencies for outer or inner races (ref. Table 5.1). In
addition, a series of other frequencies separated by shaft frequency (100 Hz) intervals are evident,
especially when damage becomes extensive. This indicates strong shaft frequency modulation
of the signal, possibly due to eccentric shaft rotation as extensive damage has been found to loosen

the bearu'g.
The last spectrum with bearing B severely damaged (105 hours-Fig. E.l(c)) reveals some
race frequency component evident in Figs E. I(b) and E. 1(c), has disappeared
and a strong frequency component at 530 Hz has appeared. Although this frequency very nearly
"corresponds to the ball rolling frequency, examination of the bearing at the conclusion of the
test revealed little or no damage to the balls. This tends to exclude the ball rolling frequency
as the source of the 530 Hz component and a satisfactory explanation for its appearance and the
disappearance of the outer race component requires further investigation.

"anomalies. The outer

6.2.4 Time Domain Signals

Figs F.I-F.4 show time traces of signals from bearing C at I and 14-5 hours runtime, in
various frequency bands. Differences between the undamaged and damaged states are evident,
with structural, accelerometer and event frequencies apparent in :he latter state. One feature is
the larger number of oscillations following each impulse in the 5-10'kHz and 20-40 kHz bands
(where there are major structural and accelerometer resonances) compared with the number
in the 2-5 kHz and 10-20 kHz bands.
6.2.5 Comparison, of Accelerometers
Comparison of kurtosis and rms values from the four different accelerometers used did not
reveal major differences. Any effects which may be due to the different resonance frequencies
of these transducers are largely submerged in the structural resonances of the test rig and inherent
scatter of measurement. All are adequately effective in detecting early damage provided the pattern
over all bands is used. Envelope spectra are also very little effected by the particular accelerometer
and frequency band %Nhich is used.

*"
*

6.2.6 Acoustic Emission

.
"

Unlike the accelerometers, the acoustic emission transducer produced some significantly
different results. Differences %ere not noted with the first bearings teste4 (A and B) despite
periodical sampling of all signals; sampling intervals were not appropriate and shock pulse
measurement was relied .on for the earliest indication of damage. Results for bearings C and D
show that the acoustic transducer produced increased kurtosis values significantly before any
indication from the accelerometers.
On hearing C the acoustic transducer indicated higher kurtosis at about 5 hours operating
time, some 6 hours before the first indications from other sensors (Fig. C.3 cf. Fig. A.8).
Subsequ'ently bearing D was observed very carefully. Unfortunately it failed very quickly but
even so acoustic indications at 0-5 hours preceded K-meter indications by over 0.5 hours (Fig.
C.4 cf. Fig. A.IO0).
'.Another
feature of the acoustic signals is the reduction of kurtosis values below Gaussian
levels '(to about 1 8) in conditions of severe damage. This effect occurred with all bearings, but
only when.general signal levels %ere %eryhigh. Although the equipment appeared to be operating
within. nominal limits, it is suspected that the transducer and measuring system may not be
linear under extreme conditions.

7. CONCLUSIONS
, Vibration analysis methods which utilize high frequency impulsively excited signals provide
an effecti%.e means of detecting early damage in rolling element bearings. Such methods have an
inherent discriminatory capability since in many machines rolling bearings are the only com"pbrnents producing impulses which excite the high frequency resonances.

S',,

27.

":
"-

*"established

In a series of tests investigating alternative signal processing inethods-kurtosis, spectral

analysis of the enveloped signal and shock pulse distribution-all were fPund to produce clear
indications of early or moderate damage. With damage increasing to severe, kurtosis values
tended to revert to undamaged levels, envelope spectra became more random and difficult to
interpret and shock pulse distribution curves showed a progressive, though gradual change.
Total vibration levels, which did not respond significantly to early damage, increased substantially with severe damage and so usefully complemented the other measurements.
Comparison of a number of accelerometers having different resonance frequencies indicated
"no significant difference in effectiveness in detecting bearing damage. However, an acoustic
emission transducer provided indications of damage sooner than the accelerometers. In the
early stages of.failure fatigue cracks develop in the region of greatest Hertzian stress just below
the surface and subsequent propagation of these cracks generates acoustic emission signals.
It is most likely that the acoustic emission transducer detected these signals prior to the development of surface pits or spalls which produce the characteristic impulsive vibrations capable of
being detected by accelerometers. It should be realised that this promising result was obtained
on a simple bearing rig with a reasonably direct transmission path from bearing to transducer,
and further inv:stigation is required to establish the effectiveness of acoustic emission neasurement in practical situations where the transmission path is less direct. Once such procedures are
it may be possible to use acoustic emission methods to locate the precise source of
bearing failure when this cannot be achieved by other means.
The bearing rig used in the series of tests provided favourable measurement conditions,
and though all monitoring methods investigated utilize the discriminatory capacity of high frequency signals, there are differences in selectivity which become more -.nportant when endeavouring to monitor in the adverse conditions frequently encountered in practice.
Showk pulse measurement is less discriminating as it measures the area of the impulse peaks,
comparing the distribution of energies with a baseline distribution from a similar bearing in good
condition. Kurtosis calculation, by using. the fourth power of amplitude, emphasizes the 'peakiness' characteristic of the signal pattern, and requires no baseline since bearings in good condition generate a Gaussian signal which forms a standard reference.
Spectra of the signal envelope make the greatest use of the characteristics of -the signal
pattern. The amplitude of characteristic spectral lines is significant, and in addition a clear
distinction is seen between the discrete lines produced by damaged bearings and the relatively
"random spectra of sound bearings. The significance of individual spectral lines is examined
elsewhere (7], but regardless of the details of the pattern, the emergence of discrete lines provides
the clearest confirmation of bearing damage, so that this method is arguably the most effective
and vcrsatile of all methods investigated in this paper.

28

--

ACKNOWLEDGEMENTS
The authors wish to acknowledge the assistance given by Mr A. J. Davis in the operation.

S~of

the rig and recording of experimental data.

REFERENCES
[1] Boto, P. A. (1971). 'Detection of Bearing Damage by Shock Pulse Measurement', Ball
Bearing Journal (SKF) Vol. 167, pp. 1-7.
[2] Taylor, J. 1. (1980). 'Identification of Bearing Defects by Spectral Analysis'. Trans. ASME
Journ. Mech. Des. Vol. 102, pp. 199-204, April 1980.
[3] Meyer, L. D., Ahlgren, F. F., and Weichbrodt, B. (1980). 'An Analytical Model for Ball
Bearing Vibrations to Predict Vibration Response to Distributed Defects', Trans. ASME
Journ. Mech. Des., Vol. 102, pp. 205-2W0, April 1980.
[4] Ray, A. G. (1980). 'Monitoring Rolling Contact Bearings under Adverse Conditions',
2n, ..nternational Conference, Vibrations in Rotating Machinery, I. Mech. E. Conference
Publiczrions 1980-84, pp. 187-194.
[5] Parmington, B. (1983). 'Application of the Shock Pulse Method to Monitoring R3ifing
ElementBearing Deterioration', ARL Mechanical Engineering Report--to be published.
[6) Ensor, L. C., and Febg, C. C. (1975). 'A Study of the Use of Vibration and Stress Wave
Sensing for the Detecion of Bearing Failure', Engineering Report No. 75--18C, EN DEVCO,
San Juan Capistrano.

p.

[7] McFadden, P. D., and Smith, J. D. (1984). 'Model for the Vibration Produced by a Single
Point Defect in a Rolling Element Bearing', Journal of Sound and Vibration, Vol. 96, No. I,
1984.

APPENDIX A
Kurtoais Trend Plots
Kurtosis trends: bearings A-D with

"(1) Kistler accelerometer

(2) Bruel & Kjaer accelerometer
(3) PCB accelerometer
(4) K-meter

KEY: Frequency bands:

. .

2-5 kHz*'

2.. . 5-10kHz
3 . . . 10-20kHz
4 . . . 20-40kHz
5 . . . 40-80kHz
S2.5-5-0
['.

i.

'."

I:.

kHz for K-meter.

Ku rtosis
(K)

10

120

Ko

60

Rntim~

(hs.

444.

'0i

LnW

w<0

zw
<CIO

ulN

.71
6

S
>4
3

2
0I
152-2

24

FIG A.3
KURTOSIS -,BEARING A
-K- METER

w
ow

-o,

9-"
T

C/)cc

Lu
U)Lu

l1ul

-*

.m

.4

.cc

<U3

(qi-

ratD

.V4

Iu

0L

w)
0I-

ujw-j
ab

LIU
L)-

'-0

K
7
6
5
4.

3
2
1L4
12
6

3
00

6
5

45

3I
KU2~SBAIG
0

.MEE

Hrs.

Ntt00 .
0

0Ox

tLU

w
I'U

0
cc

co

-4

U.'U

co

00

o<MA

CLJ

Cc0t

Runtime

Frequency band (kHz)

(hrs.)

2-5

5-10

10-20*

20-40

40-80

80-110

110 H.P.

0.1
1.5

3.02
4.28

3.03
3.67

2.98
3.86

3.31
6.01

3.0
-6.22

3.19
9.13

3.2
15.0

FIG A.11
KURTOSIS -- BEARING D
-

PCB TRANSDUCER-

APPENDIX B
RMS Acceleration Trend Plots
r
g rms trends: bearings A-D with
(1) Kistler accelerometer
(2) Bruel & Kjaer accelerometer

ft

PCB accelerometer
(3)
(4) K-meter

KEYY: (Same as Appendix A)

i.

g rms

12

288
366

ggm-BARNG

ITE

CEEOEE

- -- -- -- -

- - - ----------

in-

ow

U.

0E

LI-

cc

co~~

Ln

.0%a

56

41

4
4

3'
2-

ItJ

FIG 8.3
grns -,BEARING A
K -METER

Ln0

ivw

cr-

- w
-<.

LA

Ln

-j

03c
QU

L.U

I-i

-,

-a

-~~L

----

N/cc
N

~-

o
I-u

rms

22

Hrs.

g rms

Ie

'.0

51-

opl

11
Hrs
Band

17

FIG 8.7?%
g rdis-BEARING C

KISTLER ACCELEROMETER

g rms
-J2

21
21.

88

grrmERIG

J..~~

8KME~

oo

.. .....

0 '-

'-..
,

rI

I
*

LA.

-*LU,

rc

N>7

uj
.

~cLJ

(A~~0

,w

*40

co

0
0L

C)c

UO-W

C4<
Ul% C,

/
/

Runtime

"(.hrs.)

2-5

5-10

0.1
1.5

.2
.72

.74
1.25

Frequency band (kHz)

10-20
20-40

.69
.90

.67
.95

FIG 8.11
rms -BEAR INGD0
-PCB, ACCELEROMETER

40-80

80-110

110 H.P.

2.61
4.08

1.89
4.63

.41
4.94

APPENDIX C
Acoustic Emission Trend Plots

J"

0x

2 5.
00.

.0

t-

en D

cc

C4

(4'-

0~

Ul

LL

LA
0)

9)

M
0
.~~'-

'0

ol

06

wUcn

.00.0

CI

U.

.00,
(A0

0
LU

'C

/%

4,o

C4

c 4

o
SL

CC

No

*o~

(JZ(3(U

q\

::

\LA

,_

,.

.':

O'-

".c

ii t.ijt

-.
-o

,'U

""

Ui

APPENDIX D
Shock Pulse Distribution Curves
Shock value determined as in paragraph 4.4.

IIiI

Ia

In

In
'I

-OP

0I

.mop

I ii

u-b

/
/

Ifri
I

I
1
I

I!

'I

//

Iu'i

II
I
I
/

-<
0

/
I,
00

I.

* /y/
II,,,

.-

-E

-0
0

Ii>
I

*1,
IV
IL

/00

UU

cn>

APPENDIX E
Envelope Spectra

Enrelope Spectra using

(1) ARL developed circuit
(2) K-meter

0.47

Amp

103 hrs.

'"

a).
a)

0.94

1.4

IL

1.5

1.0

0.5
0.47

0
Arop

kHz

2.0

z.5

2.0

2.5

104 hrs.

0.94
-

b)

"0

1.5

1.0

0.15
0.53

Amp

kHz

.r.

"_

105 hrs.

c)
0.3

0.63

1.27

0.1

"0

.k
0.5

kHz
1.0

1.S

FIG E.1
Bearing B
K -. meter: 0-110 kHz band

2.0

2.5

Amp

2.2 hrs.

p , ,' +

Ii' +L(+

:.

1.5

1.0

0.5

0
Amp

'ikHz

2.0

2.5

0.64

12.5 hrs.
1.27

' 0.1
,..-

~b)P"
0.419

1.9

"

L(

2.5

2.0

1.5

1.0

0'5

0
Amp

- .kHz

0.65

14.1 hrs.

'
0.

"1.3

C)
0.1

1.9

0.4

I
0.5

1.0

1.5

FIG E2
Bearing C
K -meter. 0-10 kHz band

Hz

~k

~L~L..L4.

~I~J
~
"0.

2.0

2.S

0.64

Amp

16.8 hrs.

,:

~d)

1.28

0.84

0.2

t:

2.5

1.5.0

1.0

0.5

0.63

Amp

hrs.

i18.25

0.23

"1.26
1.49

1.92

0.43
10.1'

0
-

2.0

2.S

2.0

2.5

19.5 hrs.

I
0.43

S0

1.5

0.63

S~1.9

f)

-tl-'" kHz

1.0

0.5

Amp
, 0.1
AO.1

Sl.... I

..

...

0.5

1.6

1.07

0P62.2S

1.0

1.5

"FIG E.2 cont'd.

Bearing C

K - meter- 0-10 kHz band

mV
- 12 L

0.47

Band: 80-110kHz

0.94

1..4

1.87

I)

i
2.5

2.0

1.5

1.0

0.5

(a)
mV

""

~0.47
04

",250

Band: 110 kHz high pass

0.94

1251.4

"1.87
0 1'I

'

...

.."

I " """ "1...

.,

1.5

1.0

0.5

"(b)

FIG E.3
Bearing D - 1 hr, runtime,.
Acoustic emission transducer,
Amplitude in millivolts (mV)

.. "1"
1 . *'.
2.0

. ..

T--b

2.5

, kHz

mV
6
Band: 80-110kHz

*
01 0.44 0.64

1.28

y~,

80t~

0.64

0.

1.2

0.84,

1.8

1.5,

1.0

0.5

2.0

(b)

k~
-

Band: 110 kHz high pass

0.44

0.1

.
2.5

()z.
10
10.52.0

0.5

FIG E.4
Bearing D -2 hrs runtime
Acoustic emission transducer
in'miflivolts (mV)
-Amplitude

2.5

Band: 5-10 kHz

0.001

a) 0.0005

'

II

0 F ~ ~ i---1.0
0.5

0
0.003

~
1.5

0.58

-kHz

2.0

2.5

Band: 10-20 kHz kHz

0.2
b)

0.001S57

O.O~.1.17
0.1
g

Band: 20-40 kHz

0.63

047
0

1.27

M
.4

0.5

1.0

1.5

FIG E.5
Bearing D - 0.5 hrs runtirne
PCB Accelerometer

2.0

2.5

010.47

Band: 40-80 kHz

0.94
1.75

d)

0.05-

000

0.5

1.0

1.5

0.47Band:

2.0

80-110 kHz

064
g

e)

0.'94

0.015-

00.5

1.0

FIG E.5cor~t'd

1.5

2.025

2.5

S*

Hand: b-lUKMZ

g
00.8

.0

1.5

..

0.5

gI

0.1

., .

1.0

. . ..

.I...

1.5

0.65

b)

--

US0

0.1-

'

.I

2.0

..

5.I. kHz-

2.5

".~

Bad 10-20 kif:

It
:

..

;J

'

,U

0.5

1.0

1.5

2.0

2.5

0.65
Band: 20-40 kHz

b)

0.05

IiI
:

' '

U ;' '

,',

,' ... . , ,.,

.,

41"

I,..

0.20.8

'
.

0.5

'

,',
I

"

1.0

6FIG

Band:
"
-"

1.5

E.6

20-40k1
1

'kHz

2.0-

2.5

Bearing 0 - 2 hrs. tuntime

PCB Accelerometer

" .-

S~0.64
1.2-

Band: 40-80kHz

1.28
d)

08

0.8

0.61ol
0.1

'"

voL

2.5

2.0

1.5

1.0

0.5

0.85
Band: 80-110kHz

128

064

0.2
e)

0.5 -

0.5

1.0

1.5

2.5

2.0

0.64Band: E olkHzhighpass

01..

.. . .

..

..

..

-...-I

. ---

I1.28

f)

0.35 -..0-1
0.7.

05

.0..

FIG E.6 cont'd

2.0

2.5

APPENDIX F
Time D~omain Signals
Tithe Domain signals of undamaged and damaged (single spall) bearings using Kistler
accelerometer.

At(A

-c

-~

E
--

Cu

-.

.Ch

IC
4-O

7-

f-7,

.............

-A

Ct

E vi

Er
-4-

Cr

Cc
C

coa

--

'cm

>

*o

4
APPENDIX G,
Progressive Damage Details

Progressive damage details and photographs of deep groove ball bearings.

Progressive Damage Details

Runtime
(hrs)

151

152-25

Photographs
(Figures)
(Divisions
cm)

Bearing A,
'Inner Race: spall, I x 3mm
Outer Race: slight pitting and reentrainment.
End Test
Inner Race: 80'-90' failure

Fig. G.A

Outer Race: 360' reentrainment/pitting

102

105

Bearing B
Inner Race: slight bruising
Outer Race: spall, 2 x 2 mm
End Test
"Inner Race: 1200 failure
"spall,4 x 3 mm
2 spalls, 10 x 3 mm
Outer Race: spall, 4 x 3 mm

Fig. G.3
Fig. G.4
Fig. G.2

reentrainment, 360'

S11
15

Bearing C
Inner Race: spall, I x 3 mm
Outer Race: slight bruising
Inner Race: spall, 2 x 3.5mm
slight bruising
Outer Race: bruising/reentrainment 360'

18.5

Inner Race: spall, 8 x 3 mm

spall, II x 3 mm (separated by 2 mm)
severe bruising over remainder
Outer Race: severe bruising, 360*

20

End Test
"Inner Race: 1800 failure
initiation of 2 spalls
severe bruising overall
Outer Race: initiation of one spall
bruising/pitting. worsened
Balls:
6ach ball is damaged with spalls and pitting

1.5

Bearing D
Inner Race: spall, 2 x I mm
slight bruising

2"5

End Test
Inner Race: spall, 15 x 3 mm
spall, 5 x 3 mm (5 mm separation)

"

Outer Race: bruising 3600

Figs G.5, G.6, G.7

Fig. G.8

Fig. G.9

Fig. G.IO

pp

FFIG C.2

*
A

_____________________________

B
U

FIG G.3

UL.inLinini

'B.
FIGG.4

FIG G.5

I-Ki

'4

FIG G.7

Hu. G.8

FIG G.9

FIG G.10

DISTRI.BUTION
AUSTRALIA
DEPARTMENT OF DEFENCE
Centra] Office
Chief Defence Scientist
Deputy Chief Defence Scientist
copy)
Supcrintendent, Science and Technology Programmes
Controller, Projects and Analytical Studies
Defence Science Representative (U.K.) (Doc. Data sheet only)
Counsellor, Defence Science (U.S.A.) (Doc. Data sheet only)
Defence Central Library
Document Exchange Centre, D.I.S.B. (18 copies)
Joint Intelligence Organisation
Librarian ,i Block, Victoria Barracks, Melbourne
Director General-Army Development (NSO) (4 copies)
Defence Inoustry and Material Policy, FAS

Aeronautical Research Laboratories

Director
Library
Superintendent-Aero Propulsion
Divisional File-Aero Propulsion
Authors: N. S. Swansson
S. C. Favaloro
C. N. King
M. L. Atkin
B. Rebechi
Materials Research Laboratories
Director/Library
Defence Research Centre

"Library
Navy Office
Navy Scientific Adviser
RAN Aircraft Maintenance and Flight Trials Unit
RAN Tactical School, Library
Directorate of Naval Aircraft Engineering
Directorate of Naval Aviation Policy
Superintendent. Aircraft Maintenance and Repair
Directorate of Naval Ship Design

"Iz

Army Office
Army Scienti=z Adviser
Engineering Development Establishment, Library
Royal Military College Library
US Army Research, Development and Standardisation Group

Air Force Office

Air Force Scientific Adviser
Aircraft Research and Development Unit
Scientific Flight Group
Library
Technical Division Library
Director General Aircraft Engineering-Air Force
Director, General Operational Requirements-Air Force
HQ Operational Command (SMAINTSO)
HQ Support Command (SLENGO)
RAAF Academy, Point Cook

Central Studies Establishment

Information Centre

DEPARTMENT OF DEFENCE SUPPORT

Government Aircraft Factories
Manager
Library

DEPARTMENT OF AVIATION
Library

Flying Operations and Airworthiness Division

STATUTORY AND STATE AUTHORITIES AND INDUSTRY

Australian Atomic Energy Commission, Director
Trans-Australia Airlines, Library
Q .ntasAirways Limited
Gas and Fuel Corporation of Victoria, Manager Scientific Services
SEC of'Vic., Herman Research Laboratory, Library
Ansett Airlines of Australia, Library
Australian Coal Industry Research Labs Ltd, Director
B.H:P., Melbourne Research Laboratories
B.P. Australia Ltd, Library
Commonwealth Aircraft Corporation, Library
Hawker de Havilland Aust. Pty Ltd, Bankstown, Library
Institute of Fuel, Australian Branch, Secretary
Australian Institute of Petroleum Ltd
Rolls Royce of Australia Pty Ltd, Mr C. G. A. Bailey

Io

UNIVERSITIES AND COLLEGES

Barr Smith Library
Adelaide
Professor of Mechanical Engineering
Library
Flinders
Library
Latrobe
Engineering Library
Melbourne
Hargrave Library
Monash
Dr R. Alfredson, Mechanical Engineering Department
Library
Newcastle
Engineering Library
Sydney
Head, School of Civil Engineering
Physical Sciences Library
N.S.W.
Professor R; A. A. Bryant, Mechanical Engineering
Professor G. D. Sergeant, Fuel Technoigy
Assoc, Professor R. W. Traill-Nash, Civil Engineering
Library
Que:nslazid
Engineering Library
Tasmania
Library
Western Australia
Associate Professor J. A. Cole, Mechanical Engineering
Library
R.M.I.T.
Dr H. Kowalski, Mech. & Production Engineering
Library
Chisholm I.T.
The Tribology Group

CANADA
International Civil Aviation Organization, Library
NRC
Aeronautical & Mechanical Engineering Library
Division of Mechanical Engineering, Director

FRANCE
ONERA, Library
GERMANY
Fachiaiform-tionszentrum: Energie. Physic, Mathematik GMBH
INDIA
Defence Ministry, Aero Development EstablIshment, Library
Gas Turbine Research Establishment, Director
Hindustan Aeronautics Ltd, Library
National Aeronautical Laboratory, Information Centre
ISRAEL
Technion-Israel Institute of Technology
Professor J. Singer,

ITALYo
n
Professor Ing. Guiseppe Gabricilli

JAPAN
National Research Institute fer Metals, Fatigue Testing Division
Universities
Kagawa University

Professor H. ishikawa

NETHERLANDS
National Aecospace Laboratory (NLR), Library

NEW ZEALAND
Defence Scientific Establishment, Library
Transport Ministry, Airworthiness Brand' Library
RNZAF, Vice Consul (Defence Liaison)
Universities
Canterbury

Library
Professor D. Stevenson, Mechanicai Lngineering

SWEDEN
Aeronautical Research Institute, Library
Swedish National Defense Research institute (FOA)

SWITZERLAND
Armament Technology and Procurement Group
F + W (Swiss Federal Aircraft Factory).

UNITED KINGDOM
CAARC, Secretary (NPL)
Royal Aircraft Establishment
Bedford, Library
Commonwealth Air Transport Council Secretariat
Admiralt'y Marine Technology Establishment
Holton Heath, Dr N. J. Wadsworth
St Leonard's Hill, Superintendent
National Gas Turoine Ee:::.shment
Director, Pyestock North
National Engineering L.aboratory, Library
British Libraqy, Lending Division
CAARC Co.ordinator. Structures
A;rcraft Research Association, Library
British Ship Research Association
National Maritime Institute, Library
Elctrical Power Engineering Co. Ltd
GEC Gas Turbines Ltd, Managing Director
Fulmer Research Instiiute Ltd, Research Director
Motor Industry Research Association, Director
Ricardo & Co. Engineers (1927) Ltd, Manager
'

-"1

Rolls-Royce Ltd
Aero Division Bristol, Library
British Aerospace
Hatficld-Chester Division, Library
British Hovercraft Corporation Ltd, Library
Short Brothers Ltd, Technical Library
Stewart-Hughes Ltd, Manager
Westland Helicopters Ltd, Manager
Natal Air Materials Laboratory, Director
Universities and Colleges
Bristol
Cambridge
Manchester
Nottingham
Southampton
Strathclyde
Cranfield Institute
of Technology
University College
of Swansea

Engineering Library
Whittle Library
Professor N. Johannesen, Fluid Mechanics
Science Library
Library
Library
Library
Professor B. L. Clarkson

UNITED STATES OF AMERICA

NASA Scientific and Te:hnical Information Facility
Applied Mechanics Reviews
Metals Information
The John Crerar Library
Allis Chalmers Corporation. Library
Boeing Co.
Mr J. C. McMillan
Kentex Research Library
United Technologies Corporation, Library
Lockheed-California Company
Lockheed Missiles and Space ompany
Lockheed Georgia
McDonnell Aircraft Compan', Library
Nondestructive Testing Infor ation Analysis Center'
Universities and Colleges
Johns Hopkins
Iowa State
Massachusetts Inst.
of Technology
SPARES (20 copies)
TOTAL (190 copies)

Profes or S. Corisin, Engineering

Dr G. K. Serovy, Mechanical Engineering
M.I.T L'braries

Departnmnt 46f Defenca

DOCUMENT CONTROL DATA

L. b. Establishment No.

1. a. AR No.

ARL-AERO-PROP-R--163

AR-003-0'0

2. Document Date

January 1984

3. Task No.

Ds r

82,t051

4. Title

S. Security

6. No. Pages

APPLICATIONS OF VIBRATION ANALYSIS

TO THE CONDITION MONITORING OF

a.document

28

8. Author(s)

9. Downgrading instructions

BARIGSb.
ROLINGELMEN

Ucasfe
c. abstract
title

7. No. Refs

N. S. Swansson
S. C. Favaloro
ILI Authority (as appropriate)

10. Corporate Author and Address

Aeronautic^al Research Laboratsries

a.Sponsior

b. Security

c.Downgrading
d. Approva

P.O. Box 4331. MELBOURNE, VIC., 3001

.2. Secondary Distributi',n (of this docuament)

Approved ior public release

Overseas enquirers outside stated limitations should be referred through ASOIS. Defence Information Sarvices
Branch. Department of Defence. Campbell Park, CANBERRA. ACT, 2601.
13. a. This document may be. AN%0.UNCED in catalogues and awareness services available to ...

No limitations
13. b. Citation for other purposes (;.e. casual annouincement) may be (s'tiect) unrestricted (or) as fo~r 13 a.
14. Descriptors

Bearings
Monitors
Vibration meters.
Acoustic emissions
Wear

Engine components
Antifriction bearings
C-ondition inolnitoring

I5. COSATi
13090
14040
21100

Group

16. Abstract

Condition monitoring mnethods for rolling element bearings, whirh utilise the high frequency
iibratjonsv generate'd hv bearing cwn~age. hare been investigated and compured exvperimentally.
Kurtosis ivalues, spectra of' vibration signal envelopes and pulse measurement methods provided
vdamage, but kuriosis in particular itas not ejfrctiive in evaluating
elf ective detection of eatrlt'
extensive lwntage, ivohere a combbnation of methods is required.

Comparison of difterent accelerometer ti-pes used as tibratioh sensors gate very vimilar
results, but an acouv ic em~ission transducer pro videdl ind(ic~ationas of incipient dlamage significantly
earlier thant the accc-lerorneters.
Tests itt-ere ronduicted under reasonably la tourable conditions of measurement so further
testing under less 14%tourable conditions is proposed. and the e/firet o such conditions, is briefi-;'
conside~red.

This paipe is to be uased to record Informwatlon which Is required by the EsteblIshment for Its oS uQ& but
which witi me~t be added to theo DISTIS daft bare uaiin sperifically requested.
W~Abmtac (Coned)

17. Impift

Aeronautica.1 Research Laboratories, Melbourne

It DeruPuwe Seuim Nd Nur+w

19. Cos Cede

Aero Propulsion Report 163

417430

21. Computer Programms Used

22.

mtblishmart Nol PsA#

2M. Typo 4~ eporr are iv- ~ Covered

I.

FILMED
9-85

* DTI'C