20 | MOBLEY.

FM Page i Wednesday, March 3, 1999 2:29 PM

ROOTCAUSEFAILURE
ANALYSIS
20 | MOBLEY.FM Page ii Wednesday, March 3, 1999 2:29 PM
PLANT ENGINEERING MAINTENANCE SERIES
Vibration Fundamentals
R. Keith Mobley
Root Cause Failure Analysis
R. Keith Mobley
Maintenance Fundamentals
R. Keith Mobley
20 | MOBLEY.FM Page iii Wednesday, March 3, 1999 2:29 PM
ROOTCAUSEFAILURE
ANALYSIS
R. Keith Mobley
Boston Oxford Auckland Johannesburg Melbourne New Delhi
20 | MOBLEY.FM Page iv Wednesday, March 3, 1999 2:29 PM
Newnes is an imprint of ButterworthHeinemann.
Copyright 1999 by ButterworthHeinemann
A member of the Reed Elsevier group
All rights reserved.
No part of this publication may be reproduced, stored in a retrieval system, or transmitted in
any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, with-
out the prior written permission of the publisher.
Recognizing the importance of preserving what has been written, ButterworthHeinemann
prints its books on acid-free paper whenever possible.
LibraryofCongressCataloging-in-PublicationData
Mobley, R. Keith, 1943-
Root cause failure analysis / by R. Keith Mobley.
p. cm. (Plant engineering maintenance series)
Includes index.
ISBN 0-7506-7158-0 (alk. paper)
1. Plant maintenance. 2. System failures (Engineering)
I. Title. II. Series.
TS192.M625 1999
658.202dc21 98-32097
CIP
BritishLibraryCataloguing-in-PublicationData
A catalogue record for this book is available from the British Library.
The publisher offers special discounts on bulk orders of this book.
For information, please contact:
Manager of Special Sales
ButterworthHeinemann
225 Wildwood Avenue
Woburn, MA 018012041
Tel: 781-904-2500
Fax: 781-904-2620
For information on all Newnes publications available, contact our World Wide Web home page
at: http://www.newnespress.com
10 9 8 7 6 5 4 3 2 1
Printed in the United States of America
Part I THEORY: INTRODUCTION
TO VIBRATION ANALYSIS............. 1
Chapter 1 INTRODUCTION ......... 2
Chapter 2 VIBRATION
Chapter 3 VIBRATION
Chapter 4 VIBRATION
Chapter 5 VIBRATION
Chapter 6 MACHINE
Chapter 7 VIBRATION DATA
Chapter 9 ANALYSIS
ANALYSIS APPLICATIONS......... 3
ANALYSIS OVERVIEW................ 6
SOURCES.................................... 13
THEORY....................................... 17
DYNAMICS................................... 26
TYPES AND FORMATS............... 42
Chapter 8 DATA ACQUISITION... 49
TECHNIQUES .............................. 60
Part II FREQUENCY-DOMAIN
VIBRATION ANALYSIS................... 65
Chapter 10 OVERVIEW ............... 66
Chapter 11 MACHINE-TRAIN
Chapter 12 DATABASE
Chapter 13 VIBRATION DATA
Chapter 14 TRENDING
Chapter 15 FAILURE-MODE
MONITORING PARAMETERS..... 71
DEVELOPMENT .......................... 97
ACQUISITION .............................. 112
ANALYSIS .................................... 125
ANALYSIS .................................... 138
Chapter 16 SIGNATURE
ANALYSIS .................................... 181
Chapter 17 ROOT-CAUSE
ANALYSIS .................................... 189
Part III RESONANCE AND
CRITICAL SPEED ANALYSIS......... 200
Chapter 18 INTRODUCTION ....... 201
Chapter 19 TYPES OF
Chapter 20 EXAMPLES OF
Chapter 21 TESTING FOR
RESONANCE ............................... 202
RESONANCE ............................... 208
RESONANCE ............................... 213
Chapter 22 MODE SHAPE ........... 222
Part IV REAL-TIME ANALYSIS....... 224
Chapter 23 OVERVIEW ............... 225
Chapter 25 DATA
Chapter 27 TRANSIENT
Chapter 28 SYNCHRONOUS
Chapter 30 TORSIONAL
Chapter 24 APPLICATIONS......... 230
ACQUISITION .............................. 235
Chapter 26 ANALYSIS SETUP .... 246
(WATERFALL) ANALYSIS ........... 255
TIME AVERAGING....................... 259
Chapter 29 ZOOM ANALYSIS...... 265
ANALYSIS .................................... 267
GLOSSARY...................................... 286
LIST OF ABBREVIATIONS ............. 291
INDEX............................................... 293
01.Mobley.1-6 Page 1 Friday, February 5, 1999 9:44 AM
Part I
THEORY: INTRODUCTION TO
VIBRATION ANALYSIS
Part I is an introduction to vibration analysis that covers basic vibration theory. All
mechanical equipment in motion generates a vibration prole, or signature, that
reects its operating condition. This is true regardless of speed or whether the mode
of operation is rotation, reciprocation, or linear motion. Vibration analysis is applica-
ble to all mechanical equipment, although a commonyet invalidassumption is
that it is limited to simple rotating machinery with running speeds above 600 revolu-
tions per minute (rpm). Vibration prole analysis is a useful tool for predictive main-
tenance, diagnostics, and many other uses.
1
01.Mobley.1-6 Page 2 Friday, February 5, 1999 9:44 AM
Chapter 1
INTRODUCTION
Several predictive maintenance techniques are used to monitor and analyze critical
machines, equipment, and systems in a typical plant. These include vibration analysis,
ultrasonics, thermography, tribology, process monitoring, visual inspection, and other
nondestructive analysis techniques. Of these techniques, vibration analysis is the
dominant predictive maintenance technique used with maintenance management pro-
grams.
Predictive maintenance has become synonymous with monitoring vibration character-
istics of rotating machinery to detect budding problems and to head off catastrophic
failure. However, vibration analysis does not provide the data required to analyze
electrical equipment, areas of heat loss, the condition of lubricating oil, or other
parameters typically evaluated in a maintenance management program. Therefore, a
total plant predictive maintenance program must include several techniques, each
designed to provide specic information on plant equipment.
2
01.Mobley.1-6 Page 3 Friday, February 5, 1999 9:44 AM
Chapter 2
VIBRATION ANALYSIS APPLICATIONS
The use of vibration analysis is not restricted to predictive maintenance. This tech-
nique is useful for diagnostic applications as well. Vibration monitoring and analysis
are the primary diagnostic tools for most mechanical systems that are used to manu-
facture products. When used properly, vibration data provide the means to maintain
optimum operating conditions and efciency of critical plant systems. Vibration anal-
ysis can be used to evaluate uid ow through pipes or vessels, to detect leaks, and to
perform a variety of nondestructive testing functions that improve the reliability and
performance of critical plant systems.
Some of the applications that are discussed briey in this chapter are predictive main-
tenance, acceptance testing, quality control, loose part detection, noise control, leak
detection, aircraft engine analyzers, and machine design and engineering. Table 2.1
lists rotating, or centrifugal, and nonrotating equipment, machine-trains, and continu-
ous processes typically monitored by vibration analysis.
Table 2.1 Equipment and Processes Typically Monitored by Vibration Analysis
Centrifugal Reciprocating Continuous Process
Pumps
Compressors
Blowers
Fans
Motor/generators
Ball mills
Chillers
Pumps
Compressors
Diesel engines
Gasoline engines
Cylinders
Other machines
Continuous casters
Hot and cold strip lines
Annealing lines
Plating lines
Paper machines
Can manufacturing lines
Pickle lines
continued
3
01.Mobley.1-6 Page 4 Friday, February 5, 1999 9:44 AM
4 Vibration Fundamentals
Table 2.1 Equipment and Processes Typically Monitored by Vibration Analysis
Centrifugal Machine-Trains Continuous Process
Product rolls Boring machines Printing
Mixers Hobbing machines Dyeing and nishing
Gearboxes Machining centers Roong manufacturing lines
Centrifuges Temper mills Chemical production lines
Transmissions Metal-working machines Petroleum production lines
Turbines Rolling mills, and most Neoprene production lines
Generators machining equipment Polyester production lines
Rotary dryers Nylon production lines
Electric motors Flooring production lines
All rotating machinery Continuous process lines
Source: Integrated Systems, Inc.
PREDICTIVE MAINTENANCE
The fact that vibration proles can be obtained for all machinery that has rotating or
moving elements allows vibration-based analysis techniques to be used for predic-
tive maintenance. Vibration analysis is one of several predictive maintenance tech-
niques used to monitor and analyze critical machines, equipment, and systems in a
typical plant. However, as indicated before, the use of vibration analysis to monitor
rotating machinery to detect budding problems and to head off catastrophic failure is
the dominant predictive maintenance technique used with maintenance management
programs.
ACCEPTANCE TESTING
Vibration analysis is a proven means of verifying the actual performance versus
design parameters of new mechanical, process, and manufacturing equipment. Preac-
ceptance tests performed at the factory and immediately following installation can be
used to ensure that new equipment performs at optimum efciency and expected life-
cycle cost. Design problems as well as possible damage during shipment or installa-
tion can be corrected before long-term damage and/or unexpected costs occur.
QUALITY CONTROL
Production-line vibration checks are an effective method of ensuring product qual-
ity where machine tools are involved. Such checks can provide advanced warning
that the surface nish on parts is nearing the rejection level. On continuous pro-
cess lines such as paper machines, steel-nishing lines, or rolling mills, vibration
01.Mobley.1-6 Page 5 Friday, February 5, 1999 9:44 AM
5 Vibration Analysis Applications
analysis can prevent abnormal oscillation of components that result in loss of
product quality.
LOOSE OR FOREIGN PARTS DETECTION
Vibration analysis is useful as a diagnostic tool for locating loose or foreign objects in
process lines or vessels. This technique has been used with great success by the
nuclear power industry and it offers the same benets to non-nuclear industries.
NOISE CONTROL
Federal, state, and local regulations require serious attention be paid to noise levels
within the plant. Vibration analysis can be used to isolate the source of noise gener-
ated by plant equipment as well as background noises such as those generated by
uorescent lights and other less obvious sources. The ability to isolate the source of
abnormal noises permits cost-effective corrective action.
LEAK DETECTION
Leaks in process vessels and devices such as valves are a serious problem in many
industries. A variation of vibration monitoring and analysis can be used to detect leak-
age and isolate its source. Leak-detection systems use an accelerometer attached to
the exterior of a process pipe. This allows the vibration prole to be monitored in
order to detect the unique frequencies generated by ow or leakage.
AIRCRAFT ENGINE ANALYZERS
Adaptations of vibration analysis techniques have been used for a variety of specialty
instruments, in particular, portable and continuous aircraft engine analyzers. Vibration
monitoring and analysis techniques are the basis of these analyzers, which are used
for detecting excessive vibration in turboprop and jet engines. These instruments
incorporate logic modules that use existing vibration data to evaluate the condition of
the engine. Portable units have diagnostic capabilities that allow a mechanic to deter-
mine the source of the problem while continuous sensors alert the pilot to any devia-
tion from optimum operating condition.
MACHINE DESIGN AND ENGINEERING
Vibration data have become a critical part of the design and engineering of new
machines and process systems. Data derived from similar or existing machinery can
be extrapolated to form the basis of a preliminary design. Prototype testing of new
machinery and systems allows these preliminary designs to be nalized, and the
vibration data from the testing adds to the design database.
01.Mobley.1-6 Page 6 Friday, February 5, 1999 9:44 AM
Chapter 3
VIBRATION ANALYSIS OVERVIEW
Vibration theory and vibration prole, or signature, analyses are complex subjects that
are the topic of many textbooks. The purpose of this chapter is to provide enough the-
ory to allow the concept of vibration proles and their analyses to be understood
before beginning the more in-depth discussions in the later sections of this module.
THEORETICAL VIBRATION PROFILES
A vibration is a periodic motion or one that repeats itself after a certain interval of
time. This time interval is referred to as the period of the vibration, T. A plot, or pro-
le, of a vibration is shown in Figure 3.1, which shows the period, T, and the maxi-
1
mum displacement or amplitude, X
0
. The inverse of the period, --- , is called the
T
frequency, f, of the vibration, which can be expressed in units of cycles per second
(cps) or Hertz (Hz). A harmonic function is the simplest type of periodic motion and
is shown in Figure 3.2, which is the harmonic function for the small oscillations of a
simple pendulum. Such a relationship can be expressed by the equation:
X = X
0
sin(t) ,
where
X = Vibration displacement (thousandths of an inch, or mils)
X
0
= Maximum displacement or amplitude (mils)
= Circular frequency (radians per second)
t = Time (seconds).
6
01.Mobley.1-6 Page 7 Friday, February 5, 1999 9:44 AM
7 Vibration Analysis Overview
Figure 3.1 Periodic motion for bearing pedestal of a steam turbine.
Figure 3.2 Small oscillations of a simple pendulum, harmonic function.
01.Mobley.1-6 Page 8 Friday, February 5, 1999 9:44 AM
8 Vibration Fundamentals
ACTUAL VIBRATION PROFILES
The process of vibration analysis requires the gathering of complex machine data,
which must then be deciphered. As opposed to the simple theoretical vibration curves
shown in Figures 3.1 and 3.2 above, the prole for a piece of equipment is extremely
complex. This is true because there are usually many sources of vibration. Each
source generates its own curve, but these are essentially added and displayed as a
composite prole. These proles can be displayed in two formats: time domain and
frequency domain.
Time Domain
Vibration data plotted as amplitude versus time is referred to as a time-domain data
prole. Some simple examples are shown in Figures 3.1 and 3.2. An example of the
complexity of these type of data for an actual piece of industrial machinery is shown
in Figure 3.3.
Time-domain plots must be used for all linear and reciprocating motion machinery.
They are useful in the overall analysis of machine-trains to study changes in operating
conditions. However, time-domain data are difcult to use. Because all of the vibra-
tion data in this type of plot are added to represent the total displacement at any given
time, it is difcult to determine the contribution of any particular vibration source.
The French physicist and mathematician Jean Fourier determined that nonharmonic
data functions such as the time-domain vibration prole are the mathematical sum of
Figure 3.3 Example of a typical time-domain vibration prole for a piece of machinery.
01.Mobley.1-6 Page 9 Friday, February 5, 1999 9:44 AM
9 Vibration Analysis Overview
Figure 3.4 Discrete (harmonic) and total (nonharmonic) time-domain vibration curves.
simple harmonic functions. The dashed-line curves in Figure 3.4 represent discrete
harmonic components of the total, or summed, nonharmonic curve represented by the
solid line.
These type of data, which are routinely taken during the life of a machine, are directly
comparable to historical data taken at exactly the same running speed and load. How-
ever, this is not practical because of variations in day-to-day plant operations and
changes in running speed. This signicantly affects the prole and makes it impossi-
ble to compare historical data.
Frequency Domain
From a practical standpoint, simple harmonic vibration functions are related to the
circular frequencies of the rotating or moving components. Therefore, these frequen-
cies are some multiple of the basic running speed of the machine-train, which is
expressed in revolutions per minute (rpm) or cycles per minute (cpm). Determining
01.Mobley.1-6 Page 10 Friday, February 5, 1999 9:44 AM
10 Vibration Fundamentals
Figure 3.5 Typical frequency-domain vibration signature.
these frequencies is the rst basic step in analyzing the operating condition of the
machine-train.
Frequency-domain data are obtained by converting time-domain data using a mathe-
matical technique referred to as a fast Fourier transform (FFT). FFT allows each
vibration component of a complex machine-train spectrum to be shown as a discrete
frequency peak. The frequency-domain amplitude can be the displacement per unit
time related to a particular frequency, which is plotted as the Y-axis against frequency
as the X-axis. This is opposed to the time-domain spectrum, which sums the velocities
of all frequencies and plots the sum as the Y-axis against time as the X-axis. An exam-
ple of a frequency-domain plot or vibration signature is shown in Figure 3.5.
Frequency-domain data are required for equipment operating at more than one run-
ning speed and all rotating applications. Because the X-axis of the spectrum is fre-
quency normalized to the running speed, a change in running speed will not affect the
plot. A vibration component that is present at one running speed will still be found in
the same location on the plot for another running speed after the normalization,
although the amplitude may be different.
01.Mobley.1-6 Page 11 Friday, February 5, 1999 9:44 AM
11 Vibration Analysis Overview
Interpretation of Vibration Data
The key to using vibration signature analysis for predictive maintenance, diagnostic,
and other applications is the ability to differentiate between normal and abnormal
vibration proles. Many vibrations are normal for a piece of rotating or moving
machinery. Examples of these are normal rotations of shafts and other rotors, contact
with bearings, gear-mesh, etc. However, specic problems with machinery generate
abnormal, yet identiable, vibrations. Examples of these are loose bolts, misaligned
shafts, worn bearings, leaks, and incipient metal fatigue.
Predictive maintenance utilizing vibration signature analysis is based on the following
facts, which form the basis for the methods used to identify and quantify the root
causes of failure:
1. All common machinery problems and failure modes have distinct vibra-
tion frequency components that can be isolated and identied.
2. A frequency-domain vibration signature is generally used for the analysis
because it is comprised of discrete peaks, each representing a specic
vibration source.
3. There is a cause, referred to as a forcing function, for every frequency
component in a machine-trains vibration signature.
4. When the signature of a machine is compared over time, it will repeat until
some event changes the vibration pattern (i.e., the amplitude of each dis-
tinct vibration component will remain constant until there is a change in
the operating dynamics of the machine-train).
While an increase or a decrease in amplitude may indicate degradation of the
machine-train, this is not always the case. Variations in load, operating practices, and
a variety of other normal changes also generate a change in the amplitude of one or
more frequency components within the vibration signature. In addition, it is important
to note that a lower amplitude does not necessarily indicate an improvement in the
mechanical condition of the machine-train. Therefore, it is important that the source
of all amplitude variations be clearly understood.
VIBRATION MEASURING EQUIPMENT
Vibration data are obtained by the following procedure: (1) Mount a transducer onto
the machinery at various locations, typically machine housing and bearing caps, and
(2) use a portable data-gathering device, referred to as a vibration monitor or analyzer,
to connect to the transducer to obtain vibration readings.
Transducer
The transducer most commonly used to obtain vibration measurements is an acceler-
ometer. It incorporates piezoelectric (i.e., pressure-sensitive) lms to convert mechan-
ical energy into electrical signals. The device generally incorporates a weight
01.Mobley.1-6 Page 12 Friday, February 5, 1999 9:44 AM
12 Vibration Fundamentals
suspended between two piezoelectric lms. The weight moves in response to vibra-
tion and squeezes the piezoelectric lms, which sends an electrical signal each time
the weight squeezes it.
Portable Vibration Analyzer
The portable vibration analyzer incorporates a microprocessor that allows it to con-
vert the electrical signal mathematically to acceleration per unit time, perform a FFT,
and store the data. It can be programmed to generate alarms and displays of the data.
The data stored by the analyzer can be downloaded to a personal or a more powerful
computer to perform more sophisticated analyses, data storage and retrieval, and
report generation.
01.Mobley.1-6 Page 13 Friday, February 5, 1999 9:44 AM
Chapter 4
VIBRATION SOURCES
All machinery with moving parts generates mechanical forces during normal opera-
tion. As the mechanical condition of the machine changes due to wear, changes in the
operating environment, load variations, etc., so do these forces. Understanding
machinery dynamics and how forces create unique vibration frequency components is
the key to understanding vibration sources.
Vibration does not just happen. There is a physical cause, referred to as a forcing
function, and each component of a vibration signature has its own forcing function.
The components that make up a signature are reected as discrete peaks in the FFT or
frequency-domain plot.
The vibration prole that results from motion is the result of a force imbalance. By
denition, balance occurs in moving systems when all forces generated by, and acting
on, the machine are in a state of equilibrium. In real-world applications, however,
there is always some level of imbalance and all machines vibrate to some extent. This
chapter discusses the more common sources of vibration for rotating machinery, as
well as for machinery undergoing reciprocating and/or linear motion.
ROTATING MACHINERY
A rotating machine has one or more machine elements that turn with a shaft, such as
rolling-element bearings, impellers, and other rotors. In a perfectly balanced machine,
all rotors turn true on their centerline and all forces are equal. However, in industrial
machinery, it is common for an imbalance of these forces to occur. In addition to
imbalance generated by a rotating element, vibration may be caused by instability in
the media owing through the rotating machine.
13
01.Mobley.1-6 Page 14 Friday, February 5, 1999 9:44 AM
14 Vibration Fundamentals
Rotor Imbalance
While mechanical imbalance generates a unique vibration prole, it is not the only
form of imbalance that affects rotating elements. Mechanical imbalance is the condi-
tion where more weight is on one side of a centerline of a rotor than on the other. In
many cases, rotor imbalance is the result of an imbalance between centripetal forces
generated by the rotation. The source of rotor vibration also can be an imbalance
between the lift generated by the rotor and gravity.
Machines with rotating elements are designed to generate vertical lift of the rotating
element when operating within normal parameters. This vertical lift must overcome
gravity to properly center the rotating element in its bearing-support structure. How-
ever, because gravity and atmospheric pressure vary with altitude and barometric
pressure, actual lift may not compensate for the downward forces of gravity in certain
environments. When the deviation of actual lift from designed lift is signicant, a
rotor might not rotate on its true centerline. This offset rotation creates an imbalance
and a measurable level of vibration.
Flow Instability and Operating Conditions
Rotating machines subject to imbalance caused by turbulent or unbalanced media
ow include pumps, fans, and compressors. A good machine design for these units
incorporates the dynamic forces of the gas or liquid in stabilizing the rotating ele-
ment. The combination of these forces and the stiffness of the rotor-support system
(i.e., bearing and bearing pedestals) determine the vibration level. Rotor-support stiff-
ness is important because unbalanced forces resulting from ow instability can deect
rotating elements from their true centerline, and the stiffness resists the deection.
Deviations from a machines designed operating envelope can affect ow stability,
which directly affects the vibration prole. For example, the vibration level of a cen-
trifugal compressor is typically low when operating at 100% load with laminar air-
ow through the compressor. However, a radical change in vibration level can result
from decreased load. Vibration resulting from operation at 50% load may increase by
as much as 400% with no change in the mechanical condition of the compressor. In
addition, a radical change in vibration level can result from turbulent ow caused by
restrictions in either the inlet or discharge piping.
Turbulent or unbalanced media ow (i.e., aerodynamic or hydraulic instability) does
not have the same quadratic impacts on the vibration prole as that of load change,
but it increases the overall vibration energy. This generates a unique prole that can
be used to quantify the level of instability present in the machine. The prole gener-
ated by unbalanced ow is visible at the vane or blade-pass frequency of the rotating
element. In addition, the prole shows a marked increase in the random noise gener-
ated by the ow of gas or liquid through the machine.
01.Mobley.1-6 Page 15 Friday, February 5, 1999 9:44 AM
15 Vibration Sources
Mechanical Motion and Forces
A clear understanding of the mechanical movement of machines and their compo-
nents is an essential part of vibration analysis. This understanding, coupled with the
forces applied by the process, are the foundation for diagnostic accuracy.
Almost every unique frequency contained in the vibration signature of a machine-
train can be directly attributed to a corresponding mechanical motion within the
machine. For example, the constant end play or axial movement of the rotating ele-
ment in a motor-generator set generates an elevated amplitude at the fundamental
(1), second harmonic (2), and third harmonic (3) of the shafts true running
speed. In addition, this movement increases the axial amplitude of the fundamental
(1) frequency.
Forces resulting from air or liquid movement through a machine also generate unique
frequency components within the machines signature. In relatively stable or laminar-
ow applications, the movement of product through the machine slightly increases
the amplitude at the vane or blade-pass frequency. In more severe, turbulent-ow
applications, the ow of product generates a broadband, white noise prole that can
be directly attributed to the movement of product through the machine.
Other forces, such as the side-load created by V-belt drives, also generate unique fre-
quencies or modify existing component frequencies. For example, excessive belt ten-
sion increases the side-load on the machine-trains shafts. This increase in side-load
changes the load zone in the machines bearings. The result of this change is a marked
increase in the amplitude at the outer-race rotational frequency of the bearings.
Applied force or induced loads can also displace the shafts in a machine-train. As a
result the machines shaft will rotate off-center which dramatically increases the
amplitude at the fundamental (1) frequency of the machine.
RECIPROCATING AND/OR LINEAR-MOTION MACHINERY
This section describes machinery that exhibits reciprocating and/or linear motion(s)
and discusses typical vibration behavior for these types of machines.
Machine Descriptions
Reciprocating linear-motion machines incorporate components that move linearly in a
reciprocating fashion to perform work. Such reciprocating machines are bidirectional
in that the linear movement reverses, returning to the initial position with each com-
pleted cycle of operation. Nonreciprocating linear-motion machines incorporate com-
ponents that also generate work in a straight line, but do not reverse direction within
one complete cycle of operation.
01.Mobley.1-6 Page 16 Friday, February 5, 1999 9:44 AM
16 Vibration Fundamentals
Few machines involve linear reciprocating motion exclusively. Most incorporate a
combination of rotating and reciprocating linear motions to produce work. One exam-
ple of such a machine is a reciprocating compressor. This unit contains a rotating
crankshaft that transmits power to one or more reciprocating pistons, which move lin-
early in performing the work required to compress the media.
Sources of Vibration
Like rotating machinery, the vibration prole generated by reciprocating and/or lin-
ear-motion machines is the result of mechanical movement and forces generated by
the components that are part of the machine. Vibration proles generated by most
reciprocating and/or linear-motion machines reect a combination of rotating and/or
linear-motion forces.
However, the intervals or frequencies generated by these machines are not always
associated with one complete revolution of a shaft. In a two-cycle reciprocating
engine, the pistons complete one cycle each time the crankshaft completes one 360-
degree revolution. In a four-cycle engine, the crank must complete two complete rev-
olutions, or 720 degrees, in order to complete a cycle of all pistons.
Because of the unique motion of reciprocating and linear-motion machines, the level
of unbalanced forces generated by these machines is substantially higher than those
generated by rotating machines. As an example, a reciprocating compressor drives
each of its pistons from bottom-center to top-center and returns to bottom-center in
each complete operation of the cylinder. The mechanical forces generated by the
reversal of direction at both top-center and bottom-center result in a sharp increase in
the vibration energy of the machine. An instantaneous spike in the vibration prole
repeats each time the piston reverses direction.
Linear-motion machines generate vibration proles similar to those of reciprocating
machines. The major difference is the impact that occurs at the change of direction
with reciprocating machines. Typically, linear-motion-only machines do not reverse
direction during each cycle of operation and, as a result, do not generate the spike of
energy associated with direction reversal.
01.Mobley.1-6 Page 17 Friday, February 5, 1999 9:44 AM
Chapter 5
VIBRATION THEORY
Mathematical techniques allow us to quantify total displacement caused by all vibra-
tions, to convert the displacement measurements to velocity or acceleration, to sepa-
rate these data into their components through the use of FFT analysis, and to
determine the amplitudes and phases of these functions. Such quantication is neces-
sary if we are to isolate and correct abnormal vibrations in machinery.
PERIODIC MOTION
Vibration is a periodic motion, or one that repeats itself after a certain interval of time
called the period, T. Figure 3.1 illustrated the periodic motion time-domain curve of a
steam turbine bearing pedestal. Displacement is plotted on the vertical, or Y-axis, and
time on the horizontal, or X-axis. The curve shown in Figure 3.4 is the sum of all
vibration components generated by the rotating element and bearing-support structure
of the turbine.
Harmonic Motion
The simplest kind of periodic motion or vibration, shown in Figure 3.2, is referred to
as harmonic. Harmonic motions repeat each time the rotating element or machine
component completes one complete cycle.
The relation between displacement and time for harmonic motion may be expressed by:
t X = X
0
sin ( )
The maximum value of the displacement is X
0
, which is also called the amplitude.
The period, T, is usually measured in seconds; its reciprocal is the frequency of the
vibration, f, measured in cycles-per-second (cps) or Hertz (Hz).
17
01.Mobley.1-6 Page 18 Friday, February 5, 1999 9:44 AM
18 Vibration Fundamentals
Figure 5.1 Illustration of vibration cycles.
1
f = ---
T
Another measure of frequency is the circular frequency, , measured in radians per sec-
ond. From Figure 5.1, it is clear that a full cycle of vibration (t) occurs after 360 degrees
or 2 radians (i.e., one full revolution). At this point, the function begins a new cycle.
= 2f
For rotating machinery, the frequency is often expressed in vibrations per minute
(vpm) or
01.Mobley.1-6 Page 19 Friday, February 5, 1999 9:44 AM
19 Vibration Theory
Figure 5.2 Two harmonic motions with a phase angle between them.

VPM = ----

By denition, velocity is the rst derivative of displacement with respect to time. For
a harmonic motion, the displacement equation is:
t X = X
0
sin ( )
The rst derivative of this equation gives us the equation for velocity:
dX
t v = ------- = X

= X
0
cos ( )
dt
This relationship tells us that the velocity is also harmonic if the displacement is har-
monic and has a maximum value or amplitude of X
0
.
By denition, acceleration is the second derivative of displacement (i.e., the rst
derivative of velocity) with respect to time:
2
d X
a = ---------- = X

=
2
X
0
sin ( ) t
2
dt
2
This function is also harmonic with amplitude of X
0
.
t Consider two frequencies given by the expression X
1
= asin ( ) and
X
2
= bsin (t + ) , which are shown in Figure 5.2 plotted against t as the X-axis.
The quantity, , in the equation for X
2
is known as the phase angle or phase difference
between the two vibrations. Because of , the two vibrations do not attain their maxi-

mum displacements at the same time. One is ---- seconds behind the other. Note that

01.Mobley.1-6 Page 20 Friday, February 5, 1999 9:44 AM

20 Vibration Fundamentals
Figure 5.3 Nonharmonic periodic motion.
these two motions have the same frequency, . A phase angle has meaning only for
two motions of the same frequency.
Nonharmonic Motion
In most machinery, there are numerous sources of vibrations, therefore, most time-
domain vibration proles are nonharmonic (represented by the solid line in Figure
5.3). While all harmonic motions are periodic, not every periodic motion is harmonic.
Figure 5.3 is the superposition of two sine waves having different frequencies, and the
dashed lines represent harmonic motions. These curves are represented by the follow-
ing equations:
X
1
= a sin ( )
1
t
X
2
= b sin ( )
2
t
The total vibration represented by the solid line is the sum of the dashed lines. The
following equation represents the total vibration:
X = X
1
+ X
2
= a sin ( ) + b sin ( )
1
t
2
t
Any periodic function can be represented as a series of sine functions having frequen-
cies of , 2, 3, etc.:
01.Mobley.1-6 Page 21 Friday, February 5, 1999 9:44 AM
21 Vibration Theory
f t ( ) = A
0
+ A
1
sin (t +
1
) + A
2
sin (2t +
2
) + A
3
sin (3t +
3
) +
This equation is known as a Fourier series, which is a function of time or f(t). The
amplitudes (A
l
, A
2
, etc.) of the various discrete vibrations and their phase angles
(
1
,
2
,
3
) can be determined mathematically when the value of function f(t) is
known. Note that these data are obtained through the use of a transducer and a porta-
ble vibration analyzer.
The terms, 2, 3, etc., are referred to as the harmonics of the primary frequency, .
In most vibration signatures, the primary frequency component is one of the running
speeds of the machine-train (1 or 1). In addition, a signature may be expected to
have one or more harmonics, for example, at two times (2), three times (3), and
other multiples of the primary running speed.
MEASURABLE PARAMETERS
As shown previously, vibrations can be displayed graphically as plots, which are
referred to as vibration proles or signatures. These plots are based on measurable
parameters (i.e., frequency and amplitude). Note that the terms prole and signature
are sometimes used interchangeably by industry. In this module, however, prole is
used to refer either to time-domain (also may be called time trace or waveform) or fre-
quency-domain plots. The term signature refers to a frequency-domain plot.
Frequency
Frequency is dened as the number of repetitions of a specic forcing function or
vibration component over a specic unit of time. Take for example a four-spoke wheel
with an accelerometer attached. Every time the shaft completes one rotation, each of
the four spokes passes the accelerometer once, which is referred to as four cycles per
revolution. Therefore, if the shaft rotates at 100 rpm, the frequency of the spokes pass-
ing the accelerometer is 400 cycles per minute (cpm). In addition to cpm, frequency is
commonly expressed in cycles per second (cps) or Hertz (Hz).
Note that for simplicity, a machine elements vibration frequency is commonly
expressed as a multiple of the shafts rotation speed. In the preceding example, the
frequency would be indicated as 4X, or four times the running speed. In addition,
because some malfunctions tend to occur at specic frequencies, it helps to segregate
certain classes of malfunctions from others.
Note, however, that the frequency/malfunction relationship is not mutually exclusive
and a specic mechanical problem cannot denitely be attributed to a unique fre-
quency. While frequency is a very important piece of information with regard to iso-
lating machinery malfunctions, it is only one part of the total picture. It is necessary to
evaluate all data before arriving at a conclusion.
01.Mobley.1-6 Page 22 Friday, February 5, 1999 9:44 AM
22 Vibration Fundamentals
Amplitude
Amplitude refers to the maximum value of a motion or vibration. This value can be
represented in terms of displacement (mils), velocity (inches per second), or accelera-
tion (inches per second squared), each of which is discussed in more detail in the fol-
lowing section on Maximum Vibration Measurement.
Amplitude can be measured as the sum of all the forces causing vibrations within a
piece of machinery (broadband), as discrete measurements for the individual forces
(component), or for individual user-selected forces (narrowband). Broadband, compo-
nent, and narrowband are discussed in a later section titled Measurement Classica-
tions. Also discussed in this section are the common curve elements: peak-to-peak,
zero-to-peak, and root-mean-square.
Maximum Vibration Measurement
The maximum value of a vibration, or amplitude, is expressed as displacement, veloc-
ity, or acceleration. Most of the microprocessor-based, frequency-domain vibration
systems will convert the acquired data to the desired form. Because industrial vibra-
tion-severity standards are typically expressed in one of these terms, it is necessary to
have a clear understanding of their relationship.
Displacement
Displacement is the actual change in distance or position of an object relative to a
reference point and is usually expressed in units of mils, 0.001 inch. For example,
displacement is the actual radial or axial movement of the shaft in relation to the nor-
mal centerline usually using the machine housing as the stationary reference. Vibra-
tion data, such as shaft displacement measurements acquired using a proximity
probe or displacement transducer should always be expressed in terms of mils, peak-
to-peak.
Velocity
Velocity is dened as the time rate of change of displacement (i.e., the rst derivative,
dX
------- or X

) and is usually expressed as inches per second (in./sec). In simple terms,
dt
velocity is a description of how fast a vibration component is moving rather than how
far, which is described by displacement.
Used in conjunction with zero-to-peak (PK) terms, velocity is the best representation
of the true energy generated by a machine when relative or bearing cap data are used.
(Note: Most vibration monitoring programs rely on data acquired from machine
housing or bearing caps.) In most cases, peak velocity values are used with vibration
data between 0 and 1000 Hz. These data are acquired with microprocessor-based,
frequency-domain systems.
01.Mobley.1-6 Page 23 Friday, February 5, 1999 9:44 AM
23 Vibration Theory
Acceleration
Acceleration is dened as the time rate of change of velocity (i.e., second derivative of
2
d X
displacement, ---------- or X

) and is expressed in units of inches per second squared (in./

2
dt
sec
2
). Vibration frequencies above 1000 Hz should always be expressed as acceleration.
Acceleration is commonly expressed in terms of the gravitational constant, g, which is
32.17 ft/sec
2
. In vibration analysis applications, acceleration is typically expressed in
terms of g-RMS or g-PK. These are the best measures of the force generated by a
machine, a group of components, or one of its components.
Measurement Classications
At least three classications of amplitude measurements are used in vibration analysis:
broadband, narrowband, and component.
Broadband or Overall
The total energy of all vibration components generated by a machine is reected by
broadband, or overall, amplitude measurements. The normal convention for express-
ing the frequency range of broadband energy is a ltered range between 10 and
10,000 Hz, or 600 and 600,000 cpm. Because most vibration-severity charts are based
on this ltered broadband, caution should be exercised to ensure that collected data
are consistent with the charts.
Narrowband
Narrowband amplitude measurements refer to those that result from monitoring the
energy generated by a user-selected group of vibration frequencies. Generally, this
amplitude represents the energy generated by a ltered band of vibration components,
failure mode, or forcing functions. For example, the total energy generated by ow
instability can be captured using a ltered narrowband around the vane or blade-pass-
ing frequency.
Component
The energy generated by a unique machine component, motion, or other forcing func-
tion can yield its own amplitude measurement. For example, the energy generated by
the rotational speed of a shaft, gear set meshing, or similar machine components gen-
erate discrete vibration components and their amplitude can be measured.
Common Elements of Curves
All vibration amplitude curves, which can represent displacement, velocity, or accel-
eration, have common elements that can be used to describe the function. These com-
mon elements are peak-to-peak, zero-to-peak, and root-mean-square, each of which is
illustrated in Figure 5.4.
01.Mobley.1-6 Page 24 Friday, February 5, 1999 9:44 AM
24 Vibration Fundamentals
Figure 5.4 Relationship of vibration amplitude.
Peak-to-Peak
As illustrated in Figure 5.4, the peak-to-peak amplitude (2A, where A is the zero-to-
peak) reects the total amplitude generated by a machine, a group of components, or
one of its components. This depends on whether the data gathered are broadband, nar-
rowband, or component. The unit of measurement is useful when the analyst needs to
know the total displacement or maximum energy produced by the machines vibration
prole.
Technically, peak-to-peak values should be used in conjunction with actual shaft-dis-
placement data, which are measured with a proximity or displacement transducer.
Peak-to-peak terms should not be used for vibration data acquired using either rela-
tive vibration data from bearing caps or when using a velocity or acceleration trans-
ducer. The only exception is when vibration levels must be compared to vibration-
severity charts based on peak-to-peak values.
Zero-to-Peak
Zero-to-peak (A), or simply peak, values are equal to one-half of the peak-to-peak
value. In general, relative vibration data acquired using a velocity transducer are
expressed in terms of peak.
01.Mobley.1-6 Page 25 Friday, February 5, 1999 9:44 AM
25 Vibration Theory
Root-Mean-Square
Root-mean-square (RMS) is the statistical average value of the amplitude generated
by a machine, one of its components, or a group of components. Referring to Figure
5.4, RMS is equal to 0.707 of the zero-to-peak value, A. Normally, RMS data are used
in conjunction with relative vibration data acquired using an accelerometer or
expressed in terms of acceleration.
01.Mobley.1-6 Page 26 Friday, February 5, 1999 9:44 AM
Chapter 6
MACHINE DYNAMICS
The primary reasons for vibration prole variations are the dynamics of the
machine, which are affected by mass, stiffness, damping, and degrees of freedom.
However, care must be taken because the vibration prole and energy levels gener-
ated by a machine also may vary depending on the location and orientation of the
measurement.
MASS, STIFFNESS, AND DAMPING
The three primary factors that determine the normal vibration energy levels and the
resulting vibration proles are mass, stiffness, and damping. Every machine-train is
designed with a dynamic support system that is based on the following: the mass of
the dynamic component(s), a specic support system stiffness, and a specic amount
of damping.
Mass
Mass is the property that describes how much material is present. Dynamically, it is
the property that describes how an unrestricted body resists the application of an
external force. Simply stated, the greater the mass the greater the force required to
accelerate it. Mass is obtained by dividing the weight of a body (e.g., rotor assembly)
by the local acceleration of gravity, g.
The English system of units is complicated compared to the metric system. In the
English system, the units of mass are pounds-mass (lbm) and the units of weight are
pounds-force (lbf). By denition, a weight (i.e., force) of 1 lbf equals the force pro-
duced by 1 lbm under the acceleration of gravity. Therefore, the constant, g
c
, which
has the same numerical value as g (32.17) and units of lbm-ft/lbf-sec
2
, is used in the
denition of weight:
26
01.Mobley.1-6 Page 27 Friday, February 5, 1999 9:44 AM
27 Machine Dynamics
Mass

g
Weight = --------------------
g
c
Therefore,
Weight

g
c
Mass = ---------------------------
g
Therefore,
Weight

g
c
lbf lbm

ft
Mass = --------------------------- = ---------- ---------------------- = lbm
g ft
----------
lb f

sec
2
2
sec
Stiffness
Stiffness is a spring-like property that describes the level of resisting force that results
when a body undergoes a change in length. Units of stiffness are often given as
pounds per inch (lbf/in.). Machine-trains have more than one stiffness property that
must be considered in vibration analysis: shaft stiffness, vertical stiffness, and hori-
zontal stiffness.
Shaft Stiffness
Most machine-trains used in industry have exible shafts and relatively long spans
between bearing-support points. As a result, these shafts tend to ex in normal opera-
tion. Three factors determine the amount of ex and mode shape that these shafts have
in normal operation: shaft diameter, shaft material properties, and span length. A
small-diameter shaft with a long span will obviously ex more than one with a larger
diameter or shorter span.
Vertical Stiffness
The rotor-bearing support structure of a machine typically has more stiffness in the
vertical plane than in the horizontal plane. Generally, the structural rigidity of a bear-
ing-support structure is much greater in the vertical plane. The full weight of and the
dynamic forces generated by the rotating element are fully supported by a pedestal
cross-section that provides maximum stiffness.
In typical rotating machinery, the vibration prole generated by a normal machine
contains lower amplitudes in the vertical plane. In most cases, this lower prole can
be directly attributed to the difference in stiffness of the vertical plane when compared
to the horizontal plane.
Horizontal Stiffness
Most bearing pedestals have more freedom in the horizontal direction than in the ver-
tical. In most applications, the vertical height of the pedestal is much greater than the
horizontal cross-section. As a result, the entire pedestal can ex in the horizontal
plane as the machine rotates.
01.Mobley.1-6 Page 28 Friday, February 5, 1999 9:44 AM
28 Vibration Fundamentals
Figure 6.1 Undamped spring-mass system.
This lower stiffness generally results in higher vibration levels in the horizontal plane.
This is especially true when the machine is subjected to abnormal modes of operation
or when the machine is unbalanced or misaligned.
Damping
Damping is a means of reducing velocity through resistance to motion, in particular
by forcing an object through a liquid or gas, or along another body. Units of damping
are often given as pounds per inch per second (lbf/in./sec, which is also expressed as
lbf-sec/in.).
The boundary conditions established by the machine design determine the freedom of
movement permitted within the machine-train. A basic understanding of this concept
is essential for vibration analysis. Free vibration refers to the vibration of a damped
(as well as undamped) system of masses with motion entirely inuenced by their
potential energy. Forced vibration occurs when motion is sustained or driven by an
applied periodic force in either damped or undamped systems. The following sections
discuss free and forced vibration for both damped and undamped systems.
Free VibrationUndamped
To understand the interactions of mass and stiffness, consider the case of undamped
free vibration of a single mass that only moves vertically, as illustrated in Figure 6.1.
In this gure, the mass M is supported by a spring that has a stiffness K (also referred
to as the spring constant), which is dened as the number of pounds of tension neces-
sary to extend the spring 1 in.
01.Mobley.1-6 Page 29 Friday, February 5, 1999 9:44 AM
29 Machine Dynamics
The force created by the static deection, X
i
, of the spring supports the weight, W, of
the mass. Also included in Figure 6.1 is the free-body diagram that illustrates the two
forces acting on the mass. These forces are the weight (also referred to as the inertia
force) and an equal, yet opposite force that results from the spring (referred to as the
spring force, F
s
).
The relationship between the weight of mass M and the static deection of the spring
can be calculated using the following equation:
W = KX
i
If the spring is displaced downward some distance, X
0
, from X
i
and released, it will
oscillate up and down. The force from the spring, F
s
, can be written as follows, where
a is the acceleration of the mass:
Ma
F = KX = --------
s
g
c
2
d X
It is common practice to replace acceleration a with
----------
, the second derivative of the
2
dt
displacement, X, of the mass with respect to time, t. Making this substitution, the
equation that denes the motion of the mass can be expressed as:
2 2
M d X M d X
----- ---------- = KX or ----- ---------- + KX = 0
g
c
2
g
c
2
dt dt
Motion of the mass is known to be periodic in time. Therefore, the displacement can
be described by the expression:
t X = X
0
cos ( )
where
X = Displacement at time t
X
0
= Initial displacement of the mass
= Frequency of the oscillation (natural or resonant frequency)
t = Time.
If this equation is differentiated and the result inserted into the equation that denes
motion, the natural frequency of the mass can be calculated. The rst derivative of the
equation for motion given previously yields the equation for velocity. The second
derivative of the equation yields acceleration.
dX
t Velocity = ------- = X

= X
0
sin ( )
dt
01.Mobley.1-6 Page 30 Friday, February 5, 1999 9:44 AM
30 Vibration Fundamentals
2
d X
t Acceleration = ---------- = X

=
2
X
0
cos ( )
2
dt
2
d X
Inserting the above expression for acceleration, or ---------- , into the equation for F
s
2
yields the following:
dt
2
M d X
----- ---------- + KX = 0
g
c
2
dt
M 2
t ----- X
0
cos ( ) + KX = 0
g
c
M 2 M 2
+ ----- X KX = ----- + K = 0
g
c
g
c
Solving this expression for yields the equation:
Kg
c
= ----------
M
where
= Natural frequency of mass
K = Spring constant
M = Mass.
Note that, theoretically, undamped free vibration persists forever. However, this never
occurs in nature and all free vibrations die down after time due to damping, which is
discussed in the next section.
Free VibrationDamped
A slight increase in system complexity results when a damping element is added to
the spring-mass system shown in Figure 6.2. This type of damping is referred to as
viscous damping. Dynamically, this system is the same as the undamped system illus-
trated in Figure 6.1, except for the damper, which usually is an oil or air dashpot
mechanism. A damper is used to continuously decrease the velocity and the resulting
energy of a mass undergoing oscillatory motion.
The system is still comprised of the inertia force due to the mass and the spring force,
but a new force is introduced. This force is referred to as the damping force and is
proportional to the damping constant, or the coefcient of viscous damping, c. The
damping force is also proportional to the velocity of the body and, as it is applied, it
opposes the motion at each instant.
01.Mobley.1-6 Page 31 Friday, February 5, 1999 9:44 AM
31 Machine Dynamics
Figure 6.2 Damped spring-mass system.
In Figure 6.2, the unelongated length of the spring is L
0
and the elongation due to the
weight of the mass is expressed by h. Therefore, the weight of the mass is Kh. Figure
6.2(a) shows the mass in its position of stable equilibrium. Figure 6.2(b) shows the
mass displaced downward a distance X from the equilibrium position. Note that X is
considered positive in the downward direction.
Figure 6.2(c) is a free-body diagram of the mass, which has three forces acting on it.
The weight (Mg/g
c
), which is directed downward, is always positive. The damping
dX

force
[
\
c------- , which is the damping constant times velocity, acts opposite to the direc-
dt
)
tion of the velocity. The spring force, K(X + h), acts in the direction opposite to the
displacement. Using Newtons equation of motion, where

F = Ma , the sum of
01.Mobley.1-6 Page 32 Friday, February 5, 1999 9:44 AM
32 Vibration Fundamentals
the forces acting on the mass can be represented by the following equation, remem-
bering that X is positive in the downward direction:
2
----- ---------- = -------- c------- K X
M d X Mg dX
( + h)
g
c
2
g
c
dt
dt
2
M d X dX
----- ---------- = Kh c------- KX Kh
g
c
2
dt
dt
2
M d X dX
----- ---------- = c------- KX
g
c
2
dt
dt
M
Dividing by ----- :
g
c
d
2
X
cg
c dX
Kg
c
X
---------- = -------- ------- --------------
dt
2
M dt M
To look up the solution to the preceding equation in a differential equations table
(such as in the CRC Handbook of Chemistry and Physics) it is necessary to change
the form of this equation. This can be accomplished by dening the relationships,
cg
c
/M = 2 and Kg
c
/M =
2
, which converts the equation to the following form:
2
d X dX 2
---------- = 2------- X
2
dt
dt
Note that for undamped free vibration, the damping constant, c, is zero and, therefore,
is also zero.
2
d X 2
---------- = X
2
dt
2
d X 2
---------- + X = 0
2
dt
The solution of this equation describes simple harmonic motion, which is given
below:
t t X = A cos ( ) + B sin ( )
dX
Substituting at t = 0, then X = X
0
and ------- = 0 , then
dt
t X = X
0
cos ( )
This shows that free vibration is periodic and is the solution for X. For damped free
vibration, however, the damping constant, c, is not zero.
2
d X dX 2
---------- = 2------- X
2
dt
dt
01.Mobley.1-6 Page 33 Friday, February 5, 1999 9:44 AM
33 Machine Dynamics
or
2
d X dX 2
+ ---------- 2------- + X = 0
2
dt
dt
or
2 2
D + 2D + = 0
which has a solution of:
d
1
t d
2
t
X = Ae + Be
where
2 2
d
1
= +
2 2
d
2
=
There are different conditions of damping: critical, overdamping, and underdamping.
Critical damping occurs when = . Overdamping occurs when > . Under-
damping occurs when < .
The only condition that results in oscillatory motion and, therefore, represents a
mechanical vibration is underdamping. The other two conditions result in aperiodic
motions. When damping is less than critical ( < ), then the following equation
applies:
X
0
X = ------e
t
(
1
cos
1
t + sin
1
t )

1
where
2 2

1
=
Forced VibrationUndamped
The simple systems described in the preceding two sections on free vibration are alike
in that they are not forced to vibrate by any exciting force or motion. Their major con-
tribution to the discussion of vibration fundamentals is that they illustrate how a sys-
tems natural or resonant frequency depends on the mass, stiffness, and damping
characteristics.
The massstiffnessdamping system also can be disturbed by a periodic variation of
external forces applied to the mass at any frequency. The system shown in Figure 6.1
is increased in complexity by the addition of an external force, F
0
, acting downward
on the mass.
01.Mobley.1-6 Page 34 Friday, February 5, 1999 9:44 AM
34 Vibration Fundamentals
In undamped forced vibration, the only difference in the equation for undamped free
vibration is that instead of the equation being equal to zero, it is equal to F
0
sin(t):
2
M d X
t ----- ---------- + KX = F
0
sin ( )
g
c
2
dt
Since the spring is not initially displaced and is driven by the function F
0
sin(t), a
particular solution, X = X
0
sin(t), is logical. Substituting this solution into the above
equation and performing mathematical manipulations yields the following equation
for X:
X
st
X = C
1
sin ( ) + C
2
cos ( ) + ------------------------------
2
- sin ( ) t t t
n n
1 (
n
)
where
X = Spring displacement at time, t
= Static spring deection under constant load, F
0
X
st
= Forced frequency
= Natural frequency of the oscillation
n
t = Time
C
1
,C
2
= Integration constants determined from specic boundary conditions.
In the above equation, the rst two terms are the undamped free vibration, and the
third term is the undamped forced vibration. The solution, containing the sum of two
sine waves of different frequencies, is itself not a harmonic motion.
Forced VibrationDamped
In a damped forced vibration system such as the one shown in Figure 6.3, the motion
of the mass M has two parts: (1) the damped free vibration at the damped natural fre-
quency and (2) the steady-state harmonic motions at the forcing frequency. The
damped natural frequency component decays quickly, but the steady-state harmonic
associated with the external force remains as long as the energy force is present.
With damped forced vibration, the only difference in its equation and the equation for
damped free vibration is that it is equal to F
0
sin(t) as shown below instead of being
equal to zero.
2
M d X dX
t ----- ---------- + c------- + KX = F
0
sin ( )
g
c
2
dt
dt
With damped vibration, damping constant c is not equal to zero and the solution of the
equation gets quite complex assuming the function, X = X
0
sin(t ). In this equa-
tion, is the phase angle, or the number of degrees that the external force, F
0
sin(t),
is ahead of the displacement, X
0
sin(t ). Using vector concepts, the following
01.Mobley.1-6 Page 35 Friday, February 5, 1999 9:44 AM
------
35 Machine Dynamics
Figure 6.3 Damped forced vibration system.
equations apply, which can be solved because there are two equations and two
unknowns:
M 2
Vertical vector component: KX
0
----- X
0
F
0
cos = 0
g
c
Horizontal vector component: cX
0
F
0
sin = 0
Solving these two equations for the unknowns X
0
and :
F
0
F
0
K
X
0
= -------------------------------------------------------- = --------------------------------------------------------------
2 M
[

2

c

2
( ) +
\
[
K
g
----
c
-
2
)

2
c
|
1 ------
j
+
[
2----- ------
\
c
)
\

2
) c n
n
c
2----- ------
c
c
c

n
tan = ----------------------- = -------------------------------
2 2
M 2
1 (
n
)
K -----
g
c
where
c = Damping constant
M
c = Critical damping = 2-----
n
c
g
c
01.Mobley.1-6 Page 36 Friday, February 5, 1999 9:44 AM
36 Vibration Fundamentals
c/c
c
= Damping ratio
F
0
= External force
F
0
/K = Deection of the spring under load, F
0
(also called static deection, X
st
)
= Forced frequency
= Natural frequency of the oscillation
n

/
= Frequency ratio.
n
For damped forced vibrations, three different frequencies have to be distinguished:
the undamped natural frequency, = Kg
c
M ; the damped natural frequency,
n
Kg
c
cg
c
2
q = ----------
[
--------

; and the frequency of maximum forced amplitude, sometimes

M
\
2M
)
referred to as the resonant frequency.
DEGREES OF FREEDOM
In a mechanical system, the degrees of freedom indicate how many numbers are
required to express its geometrical position at any instant. In machine-trains, the rela-
tionship of mass, stiffness, and damping is not the same in all directions. As a result,
the rotating or dynamic elements within the machine move more in one direction than
in another. A clear understanding of the degrees of freedom is important in that it has
a direct impact on the vibration amplitudes generated by a machine or process system.
One Degree of Freedom
If the geometrical position of a mechanical system can be dened or expressed as a
single value, the machine is said to have one degree of freedom. For example, the
position of a piston moving in a cylinder can be specied at any point in time by mea-
suring the distance from the cylinder end.
A single degree of freedom is not limited to simple mechanical systems such as the
cylinder. For example, a 12-cylinder gasoline engine with a rigid crankshaft and a rig-
idly mounted cylinder block has only one degree of freedom. The position of all of its
moving parts (i.e., pistons, rods, valves, cam shafts, etc.) can be expressed by a single
value. In this instance, the value would be the angle of the crankshaft.
However, when mounted on exible springs, this engine has multiple degrees of free-
dom. In addition to the movement of its internal parts in relationship to the crank, the
entire engine can now move in any direction. As a result, the position of the engine
and any of its internal parts require more than one value to plot its actual position in
space.
The denitions and relationships of mass, stiffness, and damping in the preceding sec-
tion assumed a single degree of freedom. In other words, movement was limited to a
01.Mobley.1-6 Page 37 Friday, February 5, 1999 9:44 AM
37 Machine Dynamics
Figure 6.4 Torsional one-degree-of-freedom system.
single plane. Therefore, the formulas are applicable for all single-degree-of-freedom
mechanical systems.
The calculation for torque is a primary example of a single degree of freedom in a
mechanical system. Figure 6.4 represents a disk with a moment of inertia, I, that is
attached to a shaft of torsional stiffness, k.
Torsional stiffness is dened as the externally applied torque, T, in inch-pounds
needed to turn the disk one radian (57.3 degrees). Torque can be represented by the
following equations:
2
d

Torque = Moment of inertia angular acceleration = I --------- = I

2
dt
01.Mobley.1-6 Page 38 Friday, February 5, 1999 9:44 AM
38 Vibration Fundamentals
In this example, three torques are acting on the disk: the spring torque, damping
torque (due to the viscosity of the air), and external torque. The spring torque is minus
()k, where is measured in radians. The damping torque is minus ()c

, where c is
the damping constant. In this example, c is the damping torque on the disk caused by
an angular speed of rotation of one radian per second. The external torque is T
0
sin(t).

t I =

Torque = c

k+ T
0
sin ( )
or

t I + c

+ k = T
0
sin ( )
Two Degrees of Freedom
The theory for a one-degree-of-freedom system is useful for determining resonant or
natural frequencies that occur in all machine-trains and process systems. However,
few machines have only one degree of freedom. Practically, most machines will have
two or more degrees of freedom. This section provides a brief overview of the theo-
ries associated with two degrees of freedom. An undamped two-degree-of-freedom
system is illustrated in Figure 6.5.
The diagram of Figure 6.5 consists of two masses, M
1
and M
2
, which are suspended
from springs, K
1
and K
2
. The two masses are tied together, or coupled, by spring K
3
,
so that they are forced to act together. In this example, the movement of the two
masses is limited to the vertical plane and, therefore, horizontal movement can be
ignored. As in the single-degree-of-freedom examples, the absolute position of each
mass is dened by its vertical position above or below the neutral, or reference, point.
Since there are two coupled masses, two locations (i.e., one for M
1
and one for M
2
)
are required to locate the absolute position of the system.
To calculate the free or natural modes of vibration, note that two distinct forces are
acting on mass, M
1
: the force of the main spring, K
1
, and that of the coupling spring,
K
3
. The main force acts upward and is dened as K
1
X
1
. The shortening of the cou-
pling spring is equal to the difference in the vertical position of the two masses,
X
1
X
2
. Therefore, the compressive force of the coupling spring is K
3
(X
1
X
2
). The
compressed coupling spring pushes the top mass, M
1
, upward so that the force is
negative.
Because these are the only tangible forces acting on M
1
, the equation of motion for the
top mass can be written as:
M
1
------- X

1
= K
1
X
1
K
3
(X
1
X
2
)
g
c
or
01.Mobley.1-6 Page 39 Friday, February 5, 1999 9:44 AM
39 Machine Dynamics
Figure 6.5 Undamped two-degree-of-freedom system with a spring couple.
M
1
------- X

1
+ (K
1
+ K
3
)X
1
K
3
X
2
= 0
g
c
The equation of motion for the second mass, M
2
, is derived in the same manner. To
make it easier to understand, turn the gure upside down and reverse the direction of
X
1
and X
2
. The equation then becomes:
M
2
------- X

2
= K
2
X
2
K
3
( X
1
X
2
)
g
c
or
M
2
------- X

2
+ (K
2
+ K
3
)X
2
K
3
X
1
= 0
g
c
01.Mobley.1-6 Page 40 Friday, February 5, 1999 9:44 AM
----------------
40 Vibration Fundamentals
If we assume that the masses M
1
and M
2
undergo harmonic motions with the same
frequency, , and with different amplitudes, A
1
and A
2
, their behavior can be repre-
sented as follows:
t X
1
= A
1
sin ( )
t X
2
= A
2
sin ( )
By substituting these into the differential equations, two equations for the amplitude
A
1
ratio, ------ , can be found:
A
2
A
1
K
3
------ = -----------------------------------------
A
2
M
1 2
------- K
1
K
3
g
c
M
2 2
------- K
2
K
3
A
1
g
c
------ = -----------------------------------------
A
2
K
3
For a solution of the form we assumed to exist, these two equations must be equal:
M
2 2
------- K
2
K
3
K
3
g
c
----------------------------------------- = -----------------------------------------
K
3
M
1 2
------- K
1
K
3
g
c
or
4 2
|
K
1
+ K
3
K
2
+ K
3
|
K
1
K
2
+ K
2
K
3
+ K
1
K
3


-------------------- + --------------------

+ ------------------------------------------------------- = 0
|
M
1
g
c
M
2
g
c |
M
1
M
2
2
g
c
This equation, known as the frequency equation, has two solutions for
2
. When sub-
stituted in either of the preceding equations, each one of these gives a denite value
A
1
for ------ . This means that there are two solutions for this example, which are of the
A
2
t t form A
1
sin ( ) and A
2
sin ( ) . As with many such problems, the nal answer is
01.Mobley.1-6 Page 41 Friday, February 5, 1999 9:44 AM
41 Machine Dynamics
the superposition of the two solutions with the nal amplitudes and frequencies deter-
mined by the boundary conditions.
Many Degrees of Freedom
When the number of degrees of freedom becomes greater than two, no critical new
parameters enter into the problem. The dynamics of all machines can be understood
by following the rules and guidelines established in the one- and two-degree-of-free-
dom equations. There are as many natural frequencies and modes of motion as there
are degrees of freedom.
02.Mobley.7- Page 42 Friday, February 5, 1999 10:07 AM
Chapter7
VIBRATIONDATATYPESANDFORMATS
There are several options regarding the types of vibration data that can be gathered for
machine-trains and systems and the formats in which it can be collected. However,
selection of type and format depends on the specic application.
The two major data-type classications are time domain and frequency domain. Each
of these can be further divided into steady-state and dynamic data formats. In turn,
each of these two formats can be further divided into single-channel and multiple-
channel formats.
DATA TYPES
Vibration proles can be acquired and displayed in one of two data types: (1) time
domain or (2) frequency domain.
Time-Domain Data
Most of the early vibration analyses were carried out using analog equipment, which
necessitated the use of time-domain data. The reason for this is that it was difcult to
convert time-domain data to frequency-domain data. Frequency-domain capability
was not available until microprocessor-based analyzers incorporated a straightforward
method (i.e., fast Fourier transform) for transforming the time-domain spectrum into
its frequency components.
Actual time-domain vibration signatures are commonly referred to as time traces or
time plots (see Figure 7.1). Theoretical vibration data are generally referred to as
waveforms (see Figure 7.2).
42
02.Mobley.7- Page 43 Friday, February 5, 1999 10:07 AM
43 VibrationDataTypesandFormats
Figure 7.1 Typical time-domain signature.
Figure 7.2 Theoretical time-domain waveforms.
02.Mobley.7- Page 44 Friday, February 5, 1999 10:07 AM
44 VibrationFundamentals
Time-domain data are presented with amplitude as the vertical axis and elapsed time
as the horizontal axis. Time-domain proles are the sum of all vibration components
(i.e., frequencies, impacts, and other transients) that are present in the machine-train
and its installed system. Time traces include all frequency components, but the indi-
vidual components are more difcult to isolate than with frequency-domain data.
The prole shown in Figure 7.2 illustrates two different data acquisition points, one
measured vertically and one measured horizontally, on the same machine and taken at
the same time. Because they were obtained concurrently, they can be compared to
determine the operating dynamics of the machine.
In this example, the data set contains an impact that occurred at 0.005 sec. The impact
is clearly visible in both the vertical (top) and horizontal (bottom) data sets. From
these time traces, it is apparent that the vertical impact is stronger than the horizontal.
In addition, the impact is repeated at 0.015 and 0.025 sec. Two conclusions can be
derived from this example: (1) The impact source is a vertical force and (2) it impacts
the machine-train at an interval of 0.010 sec, or a frequency of 1/0.010 sec = 100 Hz.
The waveform in Figure 7.2 illustrates theoretically the unique frequencies and tran-
sients that could be present in a machines signature. The gure illustrates the com-
plexity of such a waveform by overlaying numerous frequencies. The discrete
waveforms that make up Figure 7.2 are displayed individually in Figures 7.2(b)(e).
Note that two of the frequencies, shown in Figures 7.2(c) and (d), are identical, but
have a different phase angle ().
With time-domain data, the analyst must manually separate the individual frequencies
and events that are contained in the complex waveform. This effort is complicated tre-
mendously by the superposition of multiple frequencies. Note that, rather than over-
laying each of the discrete frequencies as illustrated theoretically in Figure 7.2(a),
actual time-domain data represent the sum of these frequencies as was illustrated in
Figure 7.1.
To analyze this type of plot, the analyst must manually change the timescale to obtain
discrete frequency curve data. The time interval between the recurrence of each fre-
quency can then be measured. In this way, it is possible to isolate each of the frequen-
cies that make up the time-domain vibration signature.
For routine monitoring of machine vibration, however, this approach is not cost effec-
tive. The time required to isolate manually each of the frequency components and tran-
sient events contained in the waveform is prohibitive. However, time-domain data have a
denite use in a total plant predictive maintenance or reliability improvement program.
Machine-trains or process systems that have specic timing events (e.g., a pneumatic
or hydraulic cylinder) must be analyzed using time-domain data format. In addition,
time-domain data must be used for linear and reciprocating motion machinery.
02.Mobley.7- Page 45 Friday, February 5, 1999 10:07 AM
45 VibrationDataTypesandFormats
Figure 7.3 Typical frequency-domain signature.
Frequency-Domain Data
Most rotating machine-train failures result at or near a frequency component associ-
ated with the running speed. Therefore, the ability to display and analyze the vibration
spectrum as components of frequency is extremely important.
The frequency-domain format eliminates the manual effort required to isolate the
components that make up a time trace. Frequency-domain techniques convert time-
domain data into discrete frequency components using a fast Fourier transform
(FFT). Simply stated, FFT mathematically converts a time-based trace into a series
of discrete frequency components (see Figure 7.3). In a frequency-domain plot, the
X-axis is frequency and the Y-axis is the amplitude of displacement, velocity, or
acceleration.
With frequency-domain analysis, the average spectrum for a machine-train signature
can be obtained. Recurring peaks can be normalized to present an accurate represen-
tation of the machine-train condition. Figure 7.4 illustrates a simplied relationship
between the time-domain and frequency-domain methods.
02.Mobley.7- Page 46 Friday, February 5, 1999 10:07 AM
46 VibrationFundamentals
Figure 7.4 Relationship between time domain and frequency domain.
The real advantage of frequency-domain analysis is the ability to normalize each
vibration component so that a complex machine-train spectrum can be divided into
discrete components. This ability simplies isolation and analysis of mechanical deg-
radation within the machine-train. In addition, note that frequency-domain analysis
can be used to determine the phase relationships for harmonic vibration components
in a typical machine-train spectrum. The frequency domain normalizes any or all run-
ning speeds, whereas time-domain analysis is limited to true running speed.
Mathematical theory shows that any periodic function of time, f(t), can be represented
as a series of sine functions having frequencies , 2, 3, 4, etc. Function f(t) is rep-
resented by the following equation, which is referred to as a Fourier series:
f t A
0
( ) = + A
1
sin (t +
1
) + A
2
sin (2t +
2
) + A
3
sin (3t +
3
) + ,
where
A = Amplitude of each discrete sine wave
x
= Frequency
= Phase angle of each discrete sine wave.
x
Each of these sine functions represents a discrete component of the vibration signa-
ture discussed previously. The amplitudes of each discrete component and their phase
02.Mobley.7- Page 47 Friday, February 5, 1999 10:07 AM
47 VibrationDataTypesandFormats
angles can be determined by integral calculus when the function f(t) is known.
Because the subject of integral calculus is beyond the scope of this module, the math
required to determine these integrals is not presented. A vibration analyzer and its
associated software perform this determination using FFT.
DATA FORMATS
Both time-domain and frequency-domain vibration data can be acquired and analyzed
in two primary formats: (1) steady state or (2) dynamic. Each of these formats has
strengths and weaknesses that must be clearly understood for proper use. Each of
these formats can be obtained as single- or multiple-channel data.
Steady-State Format
Most vibration programs that use microprocessor-based analyzers are limited to
steady-state data. Steady-state vibration data assume the machine-train or process
system operates in a constant, or steady-state, condition. In other words, the machine
is free of dynamic variables such as load, ow, etc. This approach further assumes that
all vibration frequencies are repeatable and maintain a constant relationship to the
rotating speed of the machines shaft.
Steady-state analysis techniques are based on acquiring vibration data when the
machine or process system is operating at a xed speed and specic operating param-
eters. For example, a variable-speed machine-train is evaluated at constant speed
rather than over its speed range.
Steady-state analysis can be compared to a still photograph of the vibration prole
generated by a machine or process system. Snapshots of the vibration prole are
acquired by the vibration analyzer and stored for analysis. While the snapshots can be
used to evaluate the relative operating condition of simple machine-trains, they do not
provide a true picture of the dynamics of either the machine or its vibration prole.
Steady-state analysis totally ignores variations in the vibration level or vibration gen-
erated by transient events such as impacts and changes in speed or process parame-
ters. Instruments used to obtain the proles contain electronic circuitry, which are
specically designed to eliminate transient data.
In the normal acquisition process, the analyzer acquires multiple blocks of data. As
part of the process, the microprocessor compares each block of data as it is acquired.
If a block contains a transient that is not included in subsequent blocks, the block con-
taining the event is discarded and replaced with a transient-free block. As a result,
steady-state analysis does not detect random events that may have a direct, negative
effect on equipment reliability.
02.Mobley.7- Page 48 Friday, February 5, 1999 10:07 AM
48 VibrationFundamentals
Dynamic Format
While steady-state data provide a snapshot of the machine, dynamic or real-time data
provide a motion picture. This approach provides a better picture of the dynamics of
both the machine-train and its vibration prole. Data acquired using steady-state
methods would suggest that vibration proles and amplitudes are constant. However,
this is not true. All dynamic forces, including running speed, vary constantly in all
machine-trains. When real-time data acquisition methods are used, these variations
are captured and displayed for analysis.
Single-ChannelFormat
Most microprocessor-based vibration monitoring programs rely on single-channel
vibration data format. Single-channel data acquisition and analysis techniques are
acceptable for routine monitoring of simple, rotating machinery. However, it is impor-
tant that single-channel analysis be augmented with multiple-channel and dynamic
analysis. Total reliance on single-channel techniques severely limits the accuracy of
analysis and the effectiveness of a predictive maintenance or reliability improvement
program.
With the single-channel method, data are acquired in series or one channel at a time.
Normally, a series of data points is established for each machine-train, and data are
acquired from each point in a measurement route. Whereas this approach is more than
adequate for routine monitoring of relatively simple machines, it is based on the
assumption that the machines dynamics and the resultant vibration prole are con-
stant throughout the entire data acquisition process. This approach hinders the ability
to evaluate real-time relationships between measurement points on the machine-train
and variations in process parameters such as speed, load, pressure, etc.
Multiple-ChannelFormat
Multiple-channel data provide the best picture of the relationship between measure-
ment points on a machine-train. Data are acquired simultaneously from all measure-
ment points on the machine-train. With this type of data, the analyst can establish the
relationship between machine dynamics and vibration prole of the entire machine.
In most cases, a digital tape recorder is used to acquire data from the machine.
Because all measurement points are recorded at the same time, the resultant data can
be used to compare the triaxial vibration prole of all measurement points. This capa-
bility greatly enhances the analysts ability to isolate abnormal machine dynamics and
to determine the root cause of deviations.
03.Mobley.8 Page 49 Friday, February 5, 1999 10:27 AM
Chapter8
DATAACQUISITION
It is important for predictive maintenance programs using vibration analysis to have
accurate, repeatable data. In addition to the type and quality of the transducer, three key
parameters affect data quality: the point of measurement, orientation, and transducer-
mounting techniques.
In a predictive and reliability maintenance program, it is extremely important to keep
good historical records of key parameters. How measurement point locations and ori-
entation to the machines shaft were selected should be kept as part of the database. It
is important that every measurement taken throughout the life of the maintenance pro-
gram be acquired at exactly the same point and orientation. In addition, the compres-
sive load, or downward force, applied to the transducer should be exactly the same for
each measurement.
VIBRATION DETECTORS: TRANSDUCERS AND CABLES
A variety of monitoring, trending, and analysis techniques are available that can and
should be used as part of a total plant vibration monitoring program. Initially, such a
program depends on the use of historical trends to detect incipient problems. As the
program matures, however, other techniques such as frequency-domain signature
analysis, time-domain analysis, and operating dynamics analysis are typically added.
An analysis is only as good as the data used, therefore, the equipment used to collect
the data are critical and determine the success or failure of a predictive maintenance
or reliability improvement program. The accuracy and proper use and mounting of
equipment determines whether or not valid data are collected.
Specically, three basic types of vibration transducers can be used for monitoring the
mechanical condition of plant machinery: displacement probes, velocity transducers,
49
03.Mobley.8 Page 50 Friday, February 5, 1999 10:27 AM
50 VibrationFundamentals
Figure 8.1 Displacement probe and signal conditioning system.
and accelerometers. Each has limitations and specic applications for which its use is
appropriate.
Displacement Probes
Displacement, or eddy-current, probes are designed to measure the actual movement,
or displacement, of a machines shaft relative to the probe. Data are normally
recorded as peak-to-peak in mils, or thousandths of an inch. This value represents the
maximum deection or displacement from the true centerline of a machines shaft.
Such a device must be rigidly mounted to a stationary structure to obtain accurate,
repeatable data. Figure 8.1 shows an illustration of a displacement probe and signal
conditioning system.
Permanently mounted displacement probes provide the most accurate data on
machines having a rotor weight that is low relative to the casing and support structure.
Turbines, large compressors, and other types of plant equipment should have dis-
placement transducers permanently mounted at key measurement locations.
The useful frequency range for displacement probes is from 10 to 1000 Hz, or 600 to
60,000 rpm. Frequency components above or below this range are distorted and,
therefore, unreliable for determining machine condition.
03.Mobley.8 Page 51 Friday, February 5, 1999 10:27 AM
51 DataAcquisition
Figure 8.2 Schematic diagram of velocity pickup: (1) pickup case, (2) wire out, (3) damper,
(4) mass, (5) spring, (6) magnet.
The major limitation with displacement or proximity probes is cost. The typical cost
for installing a single probe, including a power supply, signal conditioning, etc., aver-
ages $1000. If each machine to be evaluated requires 10 measurements, the cost per
machine is about $10,000. Using displacement transducers for all plant machinery
dramatically increases the initial cost of the program. Therefore, key locations are
generally instrumented rst and other measurement points are added later.
Velocity Transducers
Velocity transducers are electromechanical sensors designed to monitor casing, or rel-
ative, vibration. Unlike displacement probes, velocity transducers measure the rate of
displacement rather than the distance of movement. Velocity is normally expressed in
terms of inches per second (in./sec) peak, which is perhaps the best method of
expressing the energy caused by machine vibration. Figure 8.2 is a schematic diagram
of a velocity measurement device.
Like displacement probes, velocity transducers have an effective frequency range of about
10 to 1000 Hz. They should not be used to monitor frequencies above or below this range.
The major limitation of velocity transducers is their sensitivity to mechanical and
thermal damage. Normal use can cause a loss of calibration and, therefore, a strict
recalibration program is required to prevent data errors. At a minimum, velocity trans-
ducers should be recalibrated every 6 months. Even with periodic recalibration, how-
ever, velocity transducers are prone to provide distorted data due to loss of calibration.
03.Mobley.8 Page 52 Friday, February 5, 1999 10:27 AM
52 VibrationFundamentals
Figure 8.3 Schematic diagram of accelerometer: (1) base, (2) piezoelectric crystals, (3)
mass, (4) case.
Accelerometers
Acceleration is perhaps the best method of determining the force resulting from
machine vibration. Accelerometers use piezoelectric crystals or lms to convert
mechanical energy into electrical signals and Figure 8.3 is a schematic of such a
device. Data acquired with this type of transducer are relative acceleration expressed
in terms of the gravitational constant, g, in inches/second/second.
The effective range of general-purpose accelerometers is from about 1 to 10,000 Hz.
Ultrasonic accelerometers are available for frequencies up to 1 MHz. In general,
vibration data above 1000 Hz (or 60,000 cpm) should be taken and analyzed in accel-
eration or gs.
A benet of the use of accelerometers is that they do not require a calibration program
to ensure accuracy. However, they are susceptible to thermal damage. If sufcient
heat radiates into the piezoelectric crystal, it can be damaged or destroyed. However,
03.Mobley.8 Page 53 Friday, February 5, 1999 10:27 AM
53 DataAcquisition
Figure 8.4 Types of coiled cables.
thermal damage is rare because data acquisition time is relatively short (i.e., less than
30 sec) using temporary mounting techniques.
Cables
Most portable vibration data collectors use a coiled cable to connect to the transducer
(see Figure 8.4). The cable, much like a telephone cord, provides a relatively compact
length when relaxed, but will extend to reach distant measurement points. For general
use, this type of cable is acceptable, but it cannot be used for all applications.
The coiled cable is not acceptable for low-speed (i.e., less than 300 rpm) applications
or where there is a strong electromagnetic eld. Because of its natural tendency to
return to its relaxed length, the coiled cable generates a low-level frequency that cor-
responds to the oscillation rate of the cable. In low-speed applications, this oscillation
frequency can mask real vibration that is generated by the machine.
A strong electromagnetic eld, such as that generated by large mill motors, acceler-
ates cable oscillation. In these instances, the vibration generated by the cable will
mask real machine vibration.
03.Mobley.8 Page 54 Friday, February 5, 1999 10:27 AM
54 VibrationFundamentals
In applications where the coiled cable distorts or interferes with the accuracy of
acquired data, a shielded coaxial cable should be used. Although these noncoiled
cables can be more difcult to use in conjunction with a portable analyzer, they are
absolutely essential for low-speed and electromagnetic eld applications.
DATA MEASUREMENTS
Most vibration monitoring programs rely on data acquired from the machine housing
or bearing caps. The only exceptions are applications that require direct measurement
of actual shaft displacement to obtain an accurate picture of the machines dynamics.
This section discusses the number and orientation of measurement points required to
prole a machines vibration characteristics.
The fact that both normal and abnormal machine dynamics tend to generate unbal-
anced forces in one or more directions increases the analysts ability to determine the
root-cause of deviations in the machines operating condition. Because of this, mea-
surements should be taken in both radial and axial orientations.
Radial Orientation
Radially oriented measurements permit the analyst to understand the relationship of
vibration levels generated by machine components where the forces are perpendicular
to the shafts centerline.
For example, mechanical imbalance generates radial forces in all directions, but mis-
alignment generally results in a radial force in a single direction that corresponds with
the misaligned direction. The ability to determine the actual displacement direction of
the machines shaft and other components greatly improves diagnostic accuracy.
Two radial measurement points located 90 degrees apart are required at each bearing
cap. The two points permit the analyst to calculate the actual direction and relative
amplitude of any displacement that is present within the machine.
Figure 8.5 illustrates a simple vector analysis where the vertical and horizontal radial
readings acquired from the outboard bearing cap indicate a relative vertical vibration
velocity of 0.5 inches per second peak (IPS-PK) and a horizontal vibration velocity of
0.3 IPS-PK. Using simple geometry, the amplitude of vibration velocity (0.583 IPS-
PK) in the actual direction of deection can be calculated.
Axial Orientation
Axially oriented measurements are used to determine the lateral movement of a
machines shaft or dynamic mass. These measurement points are oriented in-line or
parallel with the shaft or direction of movement.
03.Mobley.8 Page 55 Friday, February 5, 1999 10:27 AM
55 DataAcquisition
Figure 8.5 Resultant shaft velocity vector based on radial vibration measurements.
At least one axial measurement is required for each shaft or dynamic movement. In
the case of shafts with a combination of oat and xed bearings, readings should be
taken from the xed or stationary bearing to obtain the best data.
TRANSDUCER-MOUNTING TECHNIQUES
For accuracy of data, a direct mechanical link between the transducer and the
machines casing or bearing cap is absolutely necessary. This makes the method used
to mount the transducer crucial to obtaining accurate data. Slight deviations in this
link will induce errors in the amplitude of vibration measurement and also may create
false frequency components that have nothing to do with the machine.
Permanent Mounting
The best method of ensuring that the point of measurement, its orientation, and the
compressive load are exactly the same each time is to permanently or hard mount the
transducers, which is illustrated in Figure 8.6. This guarantees accuracy and repeat-
ability of acquired data. However, it also increases the initial cost of the program. The
average cost of installing a general-purpose accelerometer is about $300 per measure-
ment point or $3000 for a typical machine-train.
03.Mobley.8 Page 56 Friday, February 5, 1999 10:27 AM
56 VibrationFundamentals
Figure 8.6 Permanent mounts provide best repeatability.
Quick-Disconnect Mounts
To eliminate the capital cost associated with permanently mounting transducers, a
well-designed quick-disconnect mounting can be used instead. With this technique, a
quick-disconnect stud having an average cost of less than $5 is permanently mounted
at each measurement point. A mating sleeve built into the transducer is used to con-
nect with the stud. A well-designed quick-disconnect mounting technique provides
almost the same accuracy and repeatability as the permanent mounting technique, but
at a much lower cost.
Magnets
For general-purpose use below 1000 Hz, a transducer can be attached to a machine by
a magnetic base. Even though the resonant frequency of the transducer/magnet
assembly may distort the data, this technique can be used with some success. How-
ever, since the magnet can be placed anywhere on the machine, it is difcult to guar-
antee that the exact location and orientation are maintained with each measurement.
Figure 8.7 shows common magnetic mounts for transducers.
Handheld Transducer
Another method used by some plants to acquire data is handheld transducers. This
approach is not recommended if it is possible to use any other method. Handheld trans-
ducers do not provide the accuracy and repeatability required to gain maximum benet
from a predictive maintenance program. If this technique must be used, extreme care
03.Mobley.8 Page 57 Friday, February 5, 1999 10:27 AM
57 DataAcquisition
Figure 8.7 Common magnetic mounts for transducers.
should be exercised to ensure that the same location, orientation, and compressive load
are used for every measurement. Figure 8.8 illustrates a handheld device.
ACQUIRING DATA
Three factors must be considered when acquiring vibration data: settling time, data
verication, and additional data that may be required.
Settling Time
All vibration transducers require a power source that is used to convert mechanical
motion or force to an electronic signal. In microprocessor-based analyzers, this power
source is usually internal to the analyzer. When displacement probes are used, an
external power source must be provided.
03.Mobley.8 Page 58 Friday, February 5, 1999 10:27 AM
58 VibrationFundamentals
Figure 8.8 Handheld transducers should be avoided when possible. (a) Orientation is not 90
degrees to shaft centerline. (b) Measurement-point location is not always consistent. (c)
Compressive load varies and may induce faulty readings.
When the power source is turned on, there is a momentary surge of power into the
transducer. This surge distorts the vibration prole generated by the machine. There-
fore, the data-acquisition sequence must include a time delay between powering up
and acquiring data. The time delay will vary based on the specic transducer used and
type of power source.
Some vibration analyzers include a user-selected time delay that can automatically be
downloaded as part of the measurement route. If this feature is included, the delay can
be preprogrammed for the specic transducer that will be used to acquire data. No
further adjustment is required until the transducer type is changed.
In addition to the momentary surge created by energizing the power source, the
mechanical action of placing the transducer on the machine creates a spike of energy
that may distort the vibration prole. Therefore, the actual data-acquisition sequence
should include a 10- to 20-second delay to permit decay of the spike created by
mounting the transducer.
03.Mobley.8 Page 59 Friday, February 5, 1999 10:27 AM
59 DataAcquisition
Data Verication
A number of equipment problems can result in bad or distorted data. In addition to the
surge and spike discussed in the preceding section, damaged cables, transducers,
power supplies, and other equipment failures can cause serious problems. Therefore,
it is essential to verify all data throughout the acquisition process.
Most of the microprocessor-based vibration analyzers include features that facilitate
verication of acquired data. For example, many include a low-level alert that auto-
matically alerts the technician when acquired vibration levels are below a preselected
limit. If these limits are properly set, the alert should be sufcient to detect this form
of bad data.
Unfortunately, not all distortions of acquired data result in a low-level alert. Damaged
or defective cables or transducers can result in a high level of low-frequency vibra-
tion. As a result, the low-level alert will not detect this form of bad data. However, the
vibration signature will clearly display the abnormal prole that is associated with
these problems.
In most cases, a defective cable or transducer generates a signature that contains a ski-
slope prole, which begins at the lowest visible frequency and drops rapidly to the
noise oor of the signature. If this prole is generated by defective components, it
will not contain any of the normal rotational frequencies generated by the machine-
train.
With the exception of mechanical rub, defective cables and transducers are the only
sources of this ski-slope prole. When mechanical rub is present, the ski slope will
also contain the normal rotational frequencies generated by the machine-train. In
some cases, it is necessary to turn off the auto-scale function in order to see the rota-
tional frequencies, but they will be clearly evident. If no rotational components are
present, the cable and transducer should be replaced.
Additional Data
Data obtained from a vibration analyzer are not the only things required to evaluate
machine-train or system condition. Variables, such as load, have a direct effect on the
vibration prole of machinery and must be considered. Therefore, additional data
should be acquired to augment the vibration proles.
Most microprocessor-based vibration analyzers are capable of directly acquiring pro-
cess variables and other inputs. The software and rmware provided with these sys-
tems generally support preprogrammed routes that include almost any direct or
manual data input. These routes should include all data required to analyze effectively
the operating condition of each machine-train and its process system.
04.Mobley.9 Page 60 Thursday, February 4, 1999 2:41 PM
Chapter 9
ANALYSIS TECHNIQUES
Techniques used in vibration analysis are trending, both broadband and narrowband;
comparative analysis; and signature analysis.
TRENDING
Most vibration monitoring programs rely heavily on historical vibration-level ampli-
tude trends as their dominant analysis tool. This is a valid approach if the vibration
data are normalized to remove the inuence of variables, such as load, on the recorded
vibration energy levels. Valid trend data provide an indication of change over time
within the monitored machine. As stated in preceding sections, a change in vibration
amplitude is an indication of a corresponding change in operating condition that can
be a useful diagnostic tool.
Broadband
Broadband analysis techniques have been used for monitoring the overall mechanical
condition of machinery for more than 20 years. The technique is based on the overall
vibration or energy from a frequency range of zero to the user-selected maximum fre-
quency, F
MAX
. Broadband data are overall vibration measurements expressed in units
such as velocity-PK, acceleration-RMS, etc. This type of data, however, does not pro-
vide any indication of the specic frequency components that make up the machines
vibration signature. As a result, specic machine-train problems cannot be isolated
and identied.
The only useful function of broadband analysis is long-term trending of the gross
overall condition of machinery. Typically, a set of alert/alarm limits is established
to monitor the overall condition of the machine-trains in a predictive maintenance
60
04.Mobley.9 Page 61 Thursday, February 4, 1999 2:41 PM
61 Analysis Techniques
program. However, this approach has limited value and, when used exclusively,
severely limits the ability to achieve the full benet of a comprehensive program.
Narrowband
Like broadband analysis, narrowband analysis also monitors the overall energy, but
for a user-selected band of frequency components. The ability to select specic
groups of frequencies, or narrowbands, increases the usefulness of the data. Using this
technique can drastically reduce the manpower required to monitor machine-trains
and improve the accuracy of detecting incipient problems.
Unlike broadband data, narrowband data provide the ability to directly monitor, trend,
and alarm specic machine-train components automatically by the use of a micropro-
cessor for a window of frequencies unique to specic machine components. For exam-
ple, a narrowband window can be established to directly monitor the energy of a gear
set that consists of the primary gear mesh frequency and corresponding side bands.
COMPARATIVE ANALYSIS
Comparative analysis directly compares two or more data sets in order to detect
changes in the operating condition of mechanical or process systems. This type of
analysis is limited to the direct comparison of the time-domain or frequency-domain
signature generated by a machine. The method does not determine the actual dynam-
ics of the system. Typically, the following data are used for this purpose: (1) baseline
data, (2) known machine condition, or (3) industrial reference data.
Note that great care must be taken when comparing machinery vibration data to
industry standards or baseline data. The analyst must make sure the frequency and
amplitude are expressed in units and running speeds that are consistent with the
standard or baseline data. The use of a microprocessor-based system with software
that automatically converts and displays the desired terms offers a solution to this
problem.
Baseline Data
Reference or baseline data sets should be acquired for each machine-train or pro-
cess system to be included in a predictive maintenance program when the machine
is installed or after the rst scheduled maintenance once the program is estab-
lished. These data sets can be used as a reference or comparison data set for all
future measurements. However, such data sets must be representative of the normal
operating condition of each machine-train. Three criteria are critical to the proper
use of baseline comparisons: reset after maintenance, proper identication, and
process envelope.
04.Mobley.9 Page 62 Thursday, February 4, 1999 2:41 PM
62 Vibration Fundamentals
Reset After Maintenance
The baseline data set must be updated each time the machine is repaired, rebuilt, or
when any major maintenance is performed. Even when best practices are used,
machinery cannot be restored to as-new condition when major maintenance is per-
formed. Therefore, a new baseline or reference data set must be established following
these events.
Proper Identication
Each reference or baseline data set must be clearly and completely identied. Most
vibration-monitoring systems permit the addition of a label or unique identier to any
user-selected data set. This capability should be used to clearly identify each baseline
data set.
In addition, the data-set label should include all information that denes the data set.
For example, any rework or repairs made to the machine should be identied. If a new
baseline data set is selected after the replacement of a rotating element, this informa-
tion should be included in the descriptive label.
Process Envelope
Because variations in process variables, such as load, have a direct effect on the vibra-
tion energy and the resulting signature generated by a machine-train, the actual oper-
ating envelope for each baseline data set must also be clearly identied. If this step is
omitted, direct comparison of other data to the baseline will be meaningless. The label
feature in most vibration monitoring systems permits tagging of the baseline data set
with this additional information.
Known Machine Condition
Most microprocessor-based analyzers permit direct comparison to two machine-trains
or components. The form of direct comparison, called cross-machine comparison, can
be used to identify some types of failure modes.
When using this type of comparative analysis, the analyst compares the vibration
energy and prole from a suspect machine to that of a machine with known operating
condition. For example, the suspect machine can be compared to the baseline refer-
ence taken from a similar machine within the plant. Or, a machine prole with a
known defect, such as a defective gear, can be used as a reference to determine if the
suspect machine has a similar prole and, therefore, a similar problem.
Industrial Reference Data
One form of comparative analysis is direct comparison of the acquired data to indus-
trial standards or reference values. The vibration-severity standards presented in Table
04.Mobley.9 Page 63 Thursday, February 4, 1999 2:41 PM
Analysis Techniques 63
Table 9.1 Vibration Severity Standards*
Condition Machine Classes
Good operating condition
Alert limit
Alarm limit
Absolute fault limit
I
0.028
0.010
0.156
0.260
II
0.042
0.156
0.396
0.400
III
0.100
0.255
0.396
0.620
IV
0.156
0.396
0.622
1.000
* Measurements are in inches per second peak. Applicable to a machine with running speed between 600
and 12,000 rpm. Narrowband setting: 0.3 to 3.0 running speed.
Machine Class Descriptions:
Class I Small machine-trains or individual components integrally connected with the
complete machine in its normal operating condition (i.e., drivers up to 20 hp).
Class II Medium-sized machines (i.e., 20- to 100-hp drivers and 400-hp drivers on spe-
cial foundations).
Class III Large prime movers (i.e., drivers greater than 100 hp) mounted on heavy, rigid
foundations.
Class IV Large prime movers (i.e., drivers greater than 100 hp) mounted on relatively
soft, lightweight structures.
Source: Derived by Integrated Systems, Inc. from ISO Standard #2372.
9.1 were established by the International Standards Organization (ISO). These data
are applicable for comparison with ltered narrowband data taken from machine-
trains with true running speeds between 600 and 12,000 rpm. The values from the
table include all vibration energy between a lower limit of 0.3 true running speed
and an upper limit of 3.0. For example, an 1800-rpm machine would have a ltered
narrowband between 540 (1800 0.3) and 5400 rpm (1800 3.0). A 3600-rpm
machine would have a ltered narrowband between 1080 (3600 0.3) and 10,800
rpm (3600 3.0).
FAST FOURIER TRANSFORM SIGNATURE ANALYSIS
The phrase full fast Fourier transform signature is usually applied to the vibration
spectrum that uniquely identies a machine, component, system, or subsystem at a
specic operating condition and time. It provides specic data on every frequency
component within the overall frequency range of a machine-train. The typical fre-
quency range can be from 0.1 to 20,000 Hz.
In microprocessor systems, the FFT signature is formed by breaking down the total
frequency spectrum into unique components, or peaks. Each line or peak represents
a specic frequency component that, in turn, represents one or more mechanical
components within the machine-train. Typical microprocessor-based predictive
04.Mobley.9 Page 64 Thursday, February 4, 1999 2:41 PM
64 Vibration Fundamentals
maintenance systems can provide signature resolutions of at least 400 lines and
many provide up to 12,800 lines.
A set of full-signature spectra can be an important analysis tool, but requires a tre-
mendous amount of microprocessor memory. It is impractical to collect full, high-res-
olution spectra on all machine-trains on a routine basis. Data management and storage
in the host computer is extremely difcult and costly. Full-range signatures should be
collected only if a conrmed problem has been identied on a specic machine-train.
This can be triggered automatically by exceeding a preset alarm limit in the historical
amplitude trends.
Broadband and Full Signature
Systems that utilize either broadband or full signature measurements have limitations
that may hamper the usefulness of the program. Broadband measurements usually do
not have enough resolution at running speeds to be effective in early problem diag-
nostics. Full-signature measurement at every data point requires a massive data acqui-
sition, handling, and storage system that greatly increases the capital and operating
costs of the program.
Normally, a set of full-signature spectra is needed only when an identied machine-
train problem demands further investigation. Please note that, although full signatures
generate too much data for routine problem detection, they are essential for root-cause
diagnostics. Therefore, the optimum system includes the capability to utilize all tech-
niques. This ability optimizes the programs ability to trend, do full root-cause failure
analysis, and still maintain minimum data management and storage requirements.
Narrowband
Typically, a machine-trains vibration signature is made up of vibration components
with each component associated with one or more of the true running speeds within
the machine-train. Because most machinery problems show up at or near one or more
of the running speeds, the narrowband capability is very benecial in that high-resolu-
tion windows can be preset to monitor the running speeds. However, many of the
microprocessor-based predictive maintenance systems available do not have narrow-
band capability. Therefore, care should be taken to ensure that the system utilized
does have this capability.
05.Mobley.10 Page 65 Thursday, February 4, 1999 2:49 PM
PartII
FREQUENCY-DOMAINVIBRATION
ANALYSIS
This module provides the basic knowledge and skills required to implement a com-
puter-based vibration monitoring program. It discusses the following topics: (1) typi-
cal machine-train monitoring parameters, (2) database development, (3) data-
acquisition equipment and methods, and (4) data analysis.
Although each of the commercially available computer-based vibration monitoring
systems has unique features and formats, the information contained in this training
module is applicable to all of the systems. However, the manual provided by the ven-
dor should be used in conjunction with this module to ensure proper use of the micro-
processor-based data collection analyzer and the computer-based software.
65
05.Mobley.10 Page 66 Thursday, February 4, 1999 2:49 PM
Chapter10
OVERVIEW
During the past 10 years, vibration monitoring and analysis instrumentation have
improved dramatically. During this period, a number of new microprocessor-based
systems have been introduced that greatly simplify the collection, data management,
long-term trending, and analysis of vibration data. While these advancements permit
wider use of vibration monitoring as a predictive-maintenance tool, use is generally
limited to relatively simple, steady-state rotating machinery. Typically, these systems
collect single-channel, frequency-domain data.
Table 10.1 lists the typical machinery that can be monitored using these microproces-
sor-based systems. All machines included in the centrifugal column are ideal applica-
tions for this technology. Those in the columns headed Reciprocating and Machine-
Trains can be evaluated, but the limitations of the technology preclude full diagnostic
capability. The remaining machinery and process systems included under the heading
Continuous Process are much more difcult to analyze using this technology, but it
can be done.
Table 10.1 Typical Machinery Monitored by Vibration Analysis
Centrifugal Reciprocating ContinuousProcess
Pumps Pumps Continuous casters
Compressors Compressors Hot and cold strip lines
Blowers Diesel engines Annealing lines
Fans Gasoline engines Plating lines
Motor-generators Cylinders and other machines Paper machines
Ball mills Can manufacturing lines
Chillers Pickle lines
Product rolls Printing
continued
66
05.Mobley.10 Page 67 Thursday, February 4, 1999 2:49 PM
67 Overview
Table 10.1 Typical Machinery Monitored by Vibration Analysis
Centrifugal Machine-Trains ContinuousProcess
Mixers Grinders Dyeing and nishing
Gearboxes Boring machines Roong manufacturing lines
Centrifuges Hobbing machines Chemical production lines
Transmissions Machining centers Petroleum production lines
Turbines Temper mills Neoprene production lines
Generators Metal-working machines Polyester production lines
Rotary dryers Rolling mills Nylon production lines
Electric motors Most machining equipment Flooring production lines
All rotating machinery Continuous process lines
ADVANTAGES
The automatic functions provided by most of the new systems have greatly reduced
the time and manpower required to monitor critical plant equipment. These functions
have virtually eliminated both the human errors and the setup time normally associ-
ated with older vibration-monitoring techniques.
Simplied Data Acquisition and Analysis
With the combined power of the data collector and system software, data acquisition
has been reduced to simple measurement routes that require limited operator input.
The technicians role is to temporarily mount a transducer at the proper measurement
point and push a button. The microprocessor automatically acquires conditions, eval-
uates, and stores the vibration data.
Automated Data Management
Before computer-based systems were developed, a major limitation of vibration mon-
itoring programs was the labor required to manage, store, retrieve, and analyze the
massive amount of data generated. However, the computer-based systems in use today
virtually eliminate this labor requirement. These systems automatically manage data
and provide almost instant data retrieval for analysis.
LIMITATIONS
There are several limitations of the computer-based systems and some system charac-
teristics, particularly simplied data acquisition and analysis, provide both advantages
and disadvantages. Other limitations arise because only single-channel, steady-state,
05.Mobley.10 Page 68 Thursday, February 4, 1999 2:49 PM
68 VibrationFundamentals
frequency-domain data greater than 600 cycles per minute (cpm) or 10 Hertz (Hz) can
be collected. Note that cpm also is referred to as revolutions per minute (rpm).
Simplied Data Acquisition and Analysis
While providing many advantages, simplied data acquisition and analysis also can
be a liability. If the database is improperly congured, the automated capabilities of
these analyzers will yield faulty diagnostics that can allow catastrophic failure of crit-
ical plant machinery.
Because technician involvement is reduced to a minimum level, the normal tendency
is to use untrained or partially trained personnel for this repetitive function. Unfortu-
nately, the lack of training results in less awareness and knowledge of visual and audi-
ble clues that can, and should be, an integral part of the monitoring program.
Single-Channel Data
Most of the microprocessor-based vibration monitoring systems collect single-chan-
nel, steady-state data that cannot be used for all applications. Single-channel data are
limited to the analysis of simple machinery that operates at relatively constant speed.
While most of the microprocessor-based instruments are limited to a single input
channel, in some cases, a second channel is incorporated in the analyzer. However,
this second channel generally is limited to input from a tachometer, or a once-per-rev-
olution input signal. This second channel cannot be used for vibration-data capture.
This limitation prohibits the use of most microprocessor-based vibration analyzers for
complex machinery or machines with variable speeds. Single-channel data-acquisi-
tion technology assumes that the vibration prole generated by a machine-train
remains constant throughout the data-acquisition process. This is generally true in
applications where machine speed remains relatively constant (i.e., within 5 to 10
rpm). In this case, its use does not severely limit diagnostic accuracy and can be effec-
tively used in a predictive maintenance program.
Steady-State Data
Most of the microprocessor-based instruments are designed to handle steady-state
vibration data. Few have the ability to reliably capture transient events such as rapid
speed or load changes. As a result, their use is limited in situations where these occur.
In addition, vibration data collected with a microprocessor-based analyzer is ltered
and conditioned to eliminate nonrecurring events and their associated vibration pro-
les. Antialiasing lters are incorporated into the analyzers specically to remove
spurious signals such as impacts. While the intent behind the use of antialiasing lters
is valid, however, their use can distort a machines vibration prole.
05.Mobley.10 Page 69 Thursday, February 4, 1999 2:49 PM
69 Overview
Figure 10.1 Vibration is dynamic and amplitudes constantly change.
Because vibration data are dynamic and the amplitudes constantly change as shown in
Figure 10.1, most predictive maintenance system vendors strongly recommend aver-
aging the data. They typically recommend acquiring 3 to 12 samples of the vibration
prole and averaging the individual proles into a composite signature. This approach
eliminates the variation in vibration amplitude of the individual frequency compo-
nents that make up the machines signature. However, these variations, referred to as
beats, can be a valuable diagnostic tool. Unfortunately, they are not available from
microprocessor-based instruments because of averaging and other system limitations.
Frequency-Domain Data
Most predictive maintenance programs rely almost exclusively on frequency-domain
vibration data. The microprocessor-based analyzers gather time-domain data and
automatically convert it using fast Fourier transform (FFT) to frequency-domain data.
A frequency-domain signature shows the machines individual frequency compo-
nents, or peaks.
While frequency-domain data analysis is much easier to learn than time-domain data
analysis, it does not provide the ability to isolate and identify all incipient problems
within the machine or its installed system. Because of this, additional techniques
(e.g., time-domain, multichannel, and real-time analysis) must be used in conjunction
with frequency-domain data analysis to obtain a complete diagnostic picture.
05.Mobley.10 Page 70 Thursday, February 4, 1999 2:49 PM
70 VibrationFundamentals
Low-Frequency Response
Many of the microprocessor-based vibration monitoring analyzers cannot capture
accurate data from low-speed machinery or machinery that generates low-frequency
vibration. Specically, some of the commercially available analyzers cannot be used
where frequency components are below 600 cpm or 10 Hz.
Two major problems restricting the ability to acquire accurate vibration data at low
frequencies are electronic noise and the response characteristics of the transducer. The
electronic noise of the monitored machine and the noise oor of the electronics
within the vibration analyzer tend to override the actual vibration components found
in low-speed machinery.
Analyzers specially equipped to handle noise are required for most industrial applica-
tions. There are at least three commercially available microprocessor-based analyzers
capable of acquiring data below 600 cpm. These systems use special lters and data-
acquisition techniques to separate real vibration frequencies from electronic noise. In
addition, transducers with the required low-frequency response must be used.
06.Mobley.11 Page 71 Thursday, February 4, 1999 2:51 PM
Chapter 11
MACHINE-TRAIN MONITORING
PARAMETERS
This chapter discusses normal failure modes, monitoring techniques that can prevent
premature failures, and the measurement points required for monitoring common
machine-train components. Understanding the specic location and orientation of
each measurement point is critical to diagnosing incipient problems.
The frequency-domain, or fast Fourier transform (FFT), signature acquired at each
measurement point is an actual representation of the individual machine-train compo-
nents motion at that point on the machine. Without knowing the specic location and
orientation, it is difcultif not impossibleto correctly identify incipient problems.
In simple terms, the FFT signature is a photograph of the mechanical motion of a
machine-train in a specic direction and at a specic point and time.
The vibration-monitoring process requires a large quantity of data to be collected,
temporarily stored, and downloaded to a more powerful computer for permanent stor-
age and analysis. In addition, there are many aspects to collecting meaningful data.
Data collection generally is accomplished through the use of microprocessor-based
data-collection equipment referred to as vibration analyzers. However, before analyz-
ers can be used, it is necessary to set up a database with the data-collection and analy-
sis parameters. The term narrowband refers to a specic frequency window that is
monitored because of the knowledge that potential problems may occur due to known
machine components or characteristics in this frequency range.
The orientation of each measurement point is an important consideration during the
database setup and during analysis. There is an optimum orientation for each mea-
surement point on every machine-train in a predictive maintenance program. For
example, a helical gear set creates specic force vectors during normal operation. As
the gear set degrades, these force vectors transmit the maximum vibration compo-
71
06.Mobley.11 Page 72 Thursday, February 4, 1999 2:51 PM
72 Vibration Fundamentals
Figure 11.1 Recommended measurement-point logic: AO, axial orientation; HO, horizontal
orientation; VO, vertical orientation.
nents. If only one radial reading is acquired for each bearing housing, it should be ori-
ented in the plane that provides the greatest vibration amplitude.
For continuity, each machine-train should be set up on a common-shaft with the
outboard driver bearing designated as the rst data point. Measurement points should
be numbered sequentially starting with the outboard driver bearing and ending with
the outboard bearing of the nal driven component. This is illustrated in Figure 11.1.
Any numbering convention may be used, but it should be consistent, which provides
two benets:
1. Immediate identication of the location of a particular data point during
the analysis/diagnostic phase.
2. Grouping the data points by common shaft enables the analyst to evalu-
ate all parameters affecting each component of a machine-train.
DRIVERS
All machines require some form of motive power, which is referred to as a driver. This
section includes the monitoring parameters for the two most common drivers: electric
motors and steam turbines.
Electric Motors
Electric motors are the most common source of motive power for machine-trains. As
a result, more of them are evaluated using microprocessor-based vibration monitoring
systems than any other driver. The vibration frequencies of the following parameters
are monitored to evaluate operating condition. This information is used to establish a
database.
06.Mobley.11 Page 73 Thursday, February 4, 1999 2:51 PM
73 Machine-Train Monitoring Parameters
Bearing frequencies
Imbalance
Line frequency
Loose rotor bars
Running speed
Slip frequency
V-belt intermediate drives.
Bearing Frequencies
Electric motors may incorporate either sleeve or rolling-element bearings. A narrow-
band window should be established to monitor both the normal rotational and defect
frequencies associated with the type of bearing used for each application.
Imbalance
Electric motors are susceptible to a variety of forcing functions that cause instability
or imbalance. The narrowbands established to monitor the fundamental and other har-
monics of actual running speed are useful in identifying mechanical imbalance, but
other indices also should be used.
One such index is line frequency, which provides indications of instability. Modula-
tions, or harmonics, of line frequency may indicate the motors inability to nd and
hold magnetic center. Variations in line frequency also increase the amplitude of the
fundamental and other harmonics of running speed.
Axial movement and the resulting presence of a third harmonic of running speed is
another indication of instability or imbalance within the motor. The third harmonic is
present whenever there is axial thrusting of a rotating element.
Line Frequency
Many electrical problems, or problems associated with the quality of the incoming
power and internal to the motor, can be isolated by monitoring the line frequency.
Line frequency refers to the frequency of the alternating current being supplied to the
motor. In the case of 60-cycle power, monitoring of the fundamental or rst har-
monic (60 Hz), second harmonic (120 Hz), and third harmonic (180 Hz) should be
performed.
Loose Rotor Bars
Loose rotor bars are a common failure mode of electric motors. Two methods can be
used to identify them.
The rst method uses high-frequency vibration components that result from oscil-
lating rotor bars. Typically, these frequencies are well above the normal maximum
06.Mobley.11 Page 74 Thursday, February 4, 1999 2:51 PM
74 Vibration Fundamentals
frequency used to establish the broadband signature. If this is the case, a high-pass lter
such as high-frequency domain can be used to monitor the condition of the rotor bars.
The second method uses the slip frequency to monitor for loose rotor bars. The pass-
ing frequency created by this failure mode energizes modulations associated with slip.
This method is preferred since these frequency components are within the normal
bandwidth used for vibration analysis.
Running Speed
The running speed of electric motors, both alternating current (AC) and direct current
(DC), varies. Therefore, for monitoring purposes, these motors should be classied as
variable-speed machines. A narrowband window should be established to track the
true running speed.
Slip Frequency
Slip frequency is the difference between synchronous speed and actual running speed
of the motor. A narrowband lter should be established to monitor electrical line fre-
quency. The window should have enough resolution to clearly identify the frequency
and the modulations, or sidebands, that represent slip frequency. Normally, these
modulations are spaced at the difference between synchronous and actual speed, and
the number of sidebands is equal to the number of poles in the motor.
V-Belt Intermediate Drives
Electric motors with V-belt intermediate drives display the same failure modes as
those described previously. However, the unique V-belt frequencies should be moni-
tored to determine if improper belt tension or misalignment is evident.
In addition, electric motors used with V-belt intermediate drive assemblies are suscep-
tible to premature wear on the bearings. Typically, electric motors are not designed to
compensate for the side loads associated with V-belt drives. In this type of applica-
tion, special attention should be paid to monitoring motor bearings.
The primary data-measurement point on the inboard bearing housing should be
located in the plane opposing the induced load (side load), with the secondary point at
90 degrees. The outboard primary data-measurement point should be in a plane oppo-
site the inboard bearing with the secondary at 90 degrees.
Steam Turbines
There are wide variations in the size of steam turbines, which range from large utility
units to small package units designed as drivers for pumps, etc. The following section
describes in general terms the monitoring guidelines. Parameters that should be mon-
itored are bearings, blade pass, mode shape (shaft deection), and speed (both run-
ning and critical).
06.Mobley.11 Page 75 Thursday, February 4, 1999 2:51 PM
75 Machine-Train Monitoring Parameters
Bearings
Turbines use both rolling-element and Babbitt bearings. Narrowbands should be
established to monitor both the normal rotational frequencies and failure modes of the
specic bearings used in each turbine.
Blade Pass
Turbine rotors are comprised of a series of vanes or blades mounted on individual
wheels. Each of the wheel units, which is referred to as a stage of compression, has a
different number of blades. Narrowbands should be established to monitor the blade-
pass frequency of each wheel. Loss of a blade or exing of blades or wheels is
detected by these narrowbands.
Mode Shape (Shaft Deection)
Most turbines have relatively long bearing spans and highly exible shafts. These fac-
tors, coupled with variations in process ow conditions, make turbine rotors highly
susceptible to shaft deection during normal operation. Typically, turbines operate in
either the second or third mode and should have narrowbands at the second (2) and
third (3) harmonics of shaft speed to monitor for mode shape.
Speed
All turbines are variable-speed drivers and operate near or above one of the rotors
critical speeds. Narrowbands should be established that track each of the critical
speeds dened for the turbines rotor. In most applications, steam turbines operate
above the rst critical speed and in some cases above the second. A movable narrow-
band window should be established to track the fundamental (1), second (2), and
third (3) harmonics of actual shaft speed. The best method is to use orders analysis
and a tachometer to adjust the window location.
Normally, the critical speeds are determined by the mechanical design and should not
change. However, changes in the rotor conguration or a buildup of calcium or other
foreign materials on the rotor will affect them. The narrowbands should be wide
enough to permit some increase or decrease.
INTERMEDIATE DRIVES
Intermediate drives transmit power from the primary driver to a driven unit or units.
Included in this classication are chains, couplings, gearboxes, and V-belts.
Chains
In terms of its vibration characteristics, a chain-drive assembly is much like a gear set.
The meshing of the sprocket teeth and chain links generates a vibration prole that is
almost identical to that of a gear set. The major difference between these two
06.Mobley.11 Page 76 Thursday, February 4, 1999 2:51 PM
76 Vibration Fundamentals
machine-train components is that slack in the chain tends to modulate and amplify the
tooth-mesh energy. Most of the forcing functions generated by a chain-drive assembly
can be attributed to the forces generated by tooth-mesh. The typical frequencies asso-
ciated with chain-drive assembly monitoring are those of running speed, tooth-mesh,
and chain speed.
Running Speed
Chain drives are normally used to provide positive power transmission between a
driver and driven unit where direct coupling cannot be accomplished. Chain drives
generally have two distinct running speeds: driver or input speed and driven or output
speed. Each of the shaft speeds is clearly visible in the vibration prole and a discrete
narrowband window should be established to monitor each of the running speeds.
These speeds can be calculated using the ratio of the drive to driven sprocket. For
example, where the drive sprocket has a circumference of 10 in. and the driven
sprocket a circumference of 5 in., the output speed will be two times the input speed.
Tooth-mesh narrowband windows should be created for both the drive and driven
tooth-meshing frequencies. The windows should be broad enough to capture the side-
bands or modulations that this type of passing frequency generates. The frequency of
the sprocket-teeth meshing with the chain links, or passing frequency, is calculated by
the following formula:
Tooth-mesh Frequency = Number of Sprocket Teeth Shaft Speed
Unlike gear sets, there can be two distinctive tooth-mesh frequencies for a chain-drive
system. Because the drive and driven sprockets do not directly mesh, the meshing fre-
quency generated by each sprocket is visible in the vibration prole.
Chain Speed
The chain acts much like a driven gear and has a speed that is unique to its length. The
chain speed is calculated by the following equation:
Number of Drive Sprocket Teeth Shaft Speed
Chain Speed = ------------------------------------------------------------------------------------------------------------------
Number of Links in Chain
For example:
25 teeth 100 rpm 2500
Chain Speed = --------------------------------------------- = ------------ = 10 cpm = 10 rpm
250 links 250
Couplings
Couplings cannot be monitored directly, but they generate forcing functions that
affect the vibration prole of both the driver and driven machine-train component.
Each coupling should be evaluated to determine the specic mechanical forces and
failure modes they generate. This section discusses exible couplings, gear couplings,
jackshafts, and universal joints.
06.Mobley.11 Page 77 Thursday, February 4, 1999 2:51 PM
77 Machine-Train Monitoring Parameters
Flexible Couplings
Most exible couplings use an elastomer or spring-steel device to provide power
transmission from the driver to the driven unit. Both coupling types create unique
mechanical forces that directly affect the dynamics and vibration prole of the
machine-train.
The most obvious force with exible couplings is endplay or movement in the axial
plane. Both the elastomer and spring-steel devices have memory, which forces the
axial position of both the drive and driven shafts to a neutral position. Because of
their exibility, these devices cause the shaft to move constantly in the axial plane.
This is exhibited as harmonics of shaft speed. In most cases, the resultant prole is a
signature that contains the fundamental (1) frequency and second (2) and third
(3) harmonics.
Gear Couplings
When properly installed and maintained, gear-type couplings do not generate a
unique forcing function or vibration prole. However, excessive wear, variations in
speed or torque, or overlubrication results in a forcing function.
Excessive wear or speed variation generates a gear-mesh prole that corresponds to
the number of teeth in the gear coupling multiplied by the rotational speed of the
driver. Since these couplings use a mating gear to provide power transmission, varia-
tions in speed or excessive clearance permit excitation of the gear-mesh prole.
Jackshafts
Some machine-trains use an extended or spacer shaft, called a jackshaft, to connect
the driver and a driven unit. This type of shaft may use any combination of exible
coupling, universal joint, or splined coupling to provide the exibility required mak-
ing the connection. Typically, this type of intermediate drive is used either to absorb
torsional variations during speed changes or to accommodate misalignment between
the two machine-train components.
Because of the length of these shafts and the exible couplings or joints used to trans-
mit torsional power, jackshafts tend to ex during normal operation. Flexing results in
a unique vibration prole that denes its operating mode shape.
In relatively low-speed applications, the shaft tends to operate in the rst mode or
with a bow between the two joints. This mode of operation generates an elevated
vibration frequency at the fundamental (1) turning speed of the jackshaft. In higher
speed applications, or where the exibility of the jackshaft increases, it deects into
an S shape between the two joints. This S or second mode shape generates an ele-
vated frequency at both the fundamental (1) frequency and the second harmonic (2)
of turning speed. In extreme cases, the jackshaft deects further and operates in the
06.Mobley.11 Page 78 Thursday, February 4, 1999 2:51 PM
78 Vibration Fundamentals
Figure 11.2 Typical double-pivot universal joint.
third mode. When this happens, it generates distinct frequencies at the fundamental
(1), second harmonic (2), and third harmonic (3) of turning speed.
As a general rule, narrowband windows should be established to monitor at least these
three distinct frequencies, that is, 1, 2, and 3. In addition, narrowbands should be
established to monitor the discrete frequencies generated by the couplings or joints
used to connect the jackshaft to the driver and driven unit.
Universal Joints
Various types of universal joints are used to transmit torsional power. In most cases,
this type of intermediate drive is used where some misalignment between the drive
and driven unit is necessary. Because of the misalignment, the universals pivot points
generate a unique forcing function that inuences both the dynamics and vibration
prole generated by a machine-train.
Figure 11.2 illustrates a typical double-pivot universal joint. This type of joint, which
is similar to those used in automobiles, generates a unique frequency at four times
(4) the rotational speed of the shaft. Each of the pivot-point bearings generates a
passing frequency each time the shaft completes a revolution.
Gearboxes
Gear sets are used to change speed or rotating direction of the primary driver. The
basic monitoring parameters for all gearboxes include bearings, gear-mesh frequen-
cies, and running speeds.
Bearings
A variety of bearing types is used in gearboxes. Narrowband windows should be
established to monitor the rotational and defect frequencies generated by the specic
type of bearing used in each application.
06.Mobley.11 Page 79 Thursday, February 4, 1999 2:51 PM
79 Machine-Train Monitoring Parameters
Special attention should be given to the thrust bearings, which are used in conjunc-
tion with helical gears. Because helical gears generate a relatively strong axial force,
each gear shaft must have a thrust bearing located on the backside of the gear to
absorb the thrust load. Therefore, all helical gear sets should be monitored for shaft
runout.
The thrust, or positioning, bearing of a herringbone or double-helical gear has little or
no normal axial loading. However, a coupling lockup can cause severe damage to the
thrust bearing. Double-helical gears usually have only one thrust bearing, typically on
the bull gear. Therefore, the thrust-bearing rotor should be monitored with at least one
axial data-measurement point.
The primary data-measurement point on each shaft should be in a plane opposing the
preload created by the gear mesh. A secondary data-measurement point should be
located at 90 degrees to the primary point.
Gear-Mesh Frequencies
Each gear set generates a unique prole of frequency components that should be mon-
itored. The fundamental gear-mesh frequency is equal to the number of teeth in the
pinion or drive gear multiplied by the rotational shaft speed. In addition, each gear set
generates a series of modulations, or sidebands, that surround the fundamental gear-
mesh frequency. In a normal gear set, these modulations are spaced at the same fre-
quency as the rotational shaft speed and appear on both sides of the fundamental gear
mesh.
A narrowband window should be established to monitor the fundamental gear-mesh
prole. The lower and upper limits of the narrowband should include the modulations
generated by the gear set. The number of sidebands will vary depending on the resolu-
tion used to acquire data. In most cases, the narrowband limits should be about 10%
above and below the fundamental gear-mesh frequency.
A second narrowband window should be established to monitor the second harmonic
(2) of gear mesh. Gear misalignment and abnormal meshing of gear sets result in
multiple harmonics of the fundamental gear-mesh prole. This second window pro-
vides the ability to detect potential alignment or wear problems in the gear set.
Running Speeds
A narrowband window should be established to monitor each of the running speeds
generated by the gear sets within the gearbox. The actual number of running speeds
varies depending on the number of gear sets. For example, a single-reduction gearbox
has two speeds: input and output. A double-reduction gearbox has three speeds: input,
intermediate, and output. Intermediate and output speeds are determined by calcula-
tions based on input speed and the ratio of each gear set. Figure 11.3 illustrates a typ-
ical double-reduction gearbox.
06.Mobley.11 Page 80 Thursday, February 4, 1999 2:51 PM
80 Vibration Fundamentals
Figure 11.3 Double-reduction gearbox.
If the input speed is 1800 rpm, the intermediate and output speeds are calculated
using the following equation:
Input Speed Number of Input Gear Teeth
Intermediate Speed = -------------------------------------------------------------------------------------------------------
Number of Intermediate Gear Teeth
Intermediate Speed Number of Intermediate Gear Teeth
Output Speed = -------------------------------------------------------------------------------------------------------------------------------------------
Number of Output Gear Teeth
V-Belts
V-belts are common intermediate drives for fans, blowers, and other types of machin-
ery. Unlike some other power transmission mechanisms, V-belts generate unique forc-
ing functions that must be understood and evaluated as part of a vibration analysis.
06.Mobley.11 Page 81 Thursday, February 4, 1999 2:51 PM
81 Machine-Train Monitoring Parameters
Table 11.1 Belt Drive Failure: Symptoms, Causes, and Corrective Actions
Symptom Cause Corrective Action
High 1 rotational frequency in radial Unbalanced or Balance or replace
direction. eccentric sheave. sheave.
High 1 belt frequency with harmonics. Defects in belt. Replace belt.
Impacting at belt frequency in waveform.
High 1 belt frequency. Sinusoidal Unbalanced belt. Replace belt.
waveform with period of belt frequency.
High 1 rotational frequency in axial Loose, misaligned, Align sheaves; retension
plane. 1 and possibly 2 radial. or mismatched belts. or replace belts as needed.
Source: Integrated Systems, Inc.
The key monitoring parameters for V-belt-driven machinery are fault frequency and
running speed.
Most of the forcing functions generated by V-belt drives can be attributed to the elas-
tic or rubber band effect of the belt material. This elasticity is needed to provide the
traction required to transmit power from the drive sheave (i.e., pulley) to the driven
sheave. Elasticity causes belts to act like springs, increasing vibration in the direction
of belt wrap, but damping it in the opposite direction. As a result, belt elasticity tends
to accelerate wear and the failure rate of both the driver and driven unit.
Fault Frequencies
Belt-drive fault frequencies are the frequencies of the driver, the driven unit, and the
belt. In particular, frequencies at 1 the respective shaft speeds indicate faults with the
balance, concentricity, and alignment of the sheaves. The belt frequency and its har-
monics indicate problems with the belt. Table 11.1 summarizes the symptoms and
causes of belt-drive failures, as well as corrective actions.
Running Speeds
Belt-drive ratios may be calculated if the pitch diameters (see Figure 11.4) of the
sheaves are known. This coefcient, which is used to determine the driven speed
given the drive speed, is obtained by dividing the pitch diameter of the drive sheave
by the pitch diameter of the driven sheave. These relationships are expressed by the
following equations:
Drive Sheave Diameter
Drive Reduction = -----------------------------------------------------------
Driven Sheave Diameter

Drive Sheave Diameter

Driven Speed, rpm = Drive Speed, rpm -----------------------------------------------------------

Driven Sheave Diameter

06.Mobley.11 Page 82 Thursday, February 4, 1999 2:51 PM

82 Vibration Fundamentals
Figure 11.4 Pitch diameter and center-to-center distance between belt sheaves.

Driven Sheave Diameter

Drive Speed, rpm = Driven Speed, rpm -----------------------------------------------------------

Drive Sheave Diameter

Using these relationships, the sheave rotational speeds can be determined. However,
obtaining the other component speeds requires a bit more effort. The rotational speed
of the belt cannot be directly determined using the information presented so far. To
calculate belt rotational speed (rpm), the linear belt speed must rst be determined by
nding the linear speed (in./min) of the sheave at its pitch diameter. In other words,
multiply the pitch circumference (PC) by the rotational speed of the sheave, where:
Pitch Circumference (in) = Pitch Diameter (in)
Linear Speed (in/min) = Pitch Circumference (in) Sheave Speed (rpm)
To nd the exact rotational speed of the belt (rpm), divide the linear speed by the
length of the belt:
Linear Speed (in/min)
Belt Rotational Speed (rpm) = -----------------------------------------------------
Belt Length (in)
To approximate the rotational speed of the belt, the linear speed may be calculated
using the pitch diameters and the center-to-center distance (see Figure 11.4) between
the sheaves. This method is accurate only if there is no belt sag. Otherwise, the belt
rotational speed obtained using this method is slightly higher than the actual value.
In the special case where the drive and driven sheaves have the same diameter, the for-
mula for determining the belt length is as follows:
06.Mobley.11 Page 83 Thursday, February 4, 1999 2:51 PM
83 Machine-Train Monitoring Parameters
Belt Length = Pitch Circumference + (2 Center Distance)
The following equation is used to approximate the belt length where the sheaves have
different diameters:
Drive PC Driven PC +
Belt Length = ------------------------------------------------------ + (2 Center Distance)
2
DRIVEN COMPONENTS
This chapter cannot effectively discuss all possible combinations of driven compo-
nents that may be found in a plant. However, the guidelines provided here can be used
to evaluate most of the machine-trains and process systems that are typically included
in a microprocessor-based vibration monitoring program.
Compressors
The two basic types of compressors are (1) centrifugal and (2) positive displacement.
Both of these major classications can be further divided into subtypes, depending on
their operating characteristics. This section provides an overview of the more com-
mon centrifugal and positive-displacement compressors.
Centrifugal
The two types of commonly used centrifugal compressors are (1) in-line and (2)
bullgear.
In-Line
The in-line centrifugal compressor functions in exactly the same manner as a centrif-
ugal pump. The only difference between the pump and the compressor is that the
compressor has smaller clearances between the rotor and casing. Therefore, in-line
centrifugal compressors should be monitored and evaluated in the same manner as
centrifugal pumps and fans. As with these driven components, the in-line centrifugal
compressor is comprised of a single shaft with one or more impeller(s) mounted on
the shaft. All components generate simple rotating forces that can be monitored and
evaluated with ease. Figure 11.5 shows a typical in-line centrifugal compressor.
Bullgear
The bullgear centrifugal compressor (Figure 11.6) is a multistage unit that utilizes a
large helical gear mounted on the compressors driven shaft and two or more pinion
gears, which drive the impellers. These impellers act in series, whereby compressed
air or gas from the rst-stage impeller discharge is directed by ow channels within
the compressors housing to the second-stage inlet. The discharge of the second stage
is channeled to the inlet of the third stage. This channeling occurs until the air or gas
exits the nal stage of the compressor.
06.Mobley.11 Page 84 Thursday, February 4, 1999 2:51 PM
84 Vibration Fundamentals
Figure 11.5 Typical in-line centrifugal compressor.
Generally, the driver and bullgear speed is 3600 rpm or less, and the pinion speeds are
as high as 60,000 rpm (see Figure 11.7). These machines are produced as a package
with the entire machine-train mounted on a common foundation that also includes a
panel with control and monitoring instrumentation.
Positive Displacement
Positive-displacement compressors, also referred to as dynamic-type compressors,
conne successive volumes of uid within a closed space. The pressure of the uid
increases as the volume of the closed space decreases.
Reciprocating
Reciprocating compressors are positive-displacement types having one or more cylin-
ders. Each cylinder is tted with a piston driven by a crankshaft through a connecting
rod. As the name implies, compressors within this classication displace a xed vol-
ume of air or gas with each complete cycle of the compressor.
06.Mobley.11 Page 85 Thursday, February 4, 1999 2:51 PM
85 Machine-Train Monitoring Parameters
A
B
Figure 11.6 (a) Cut-away of bullgear centrifugal compressors (b) Bullgear centrifugal com-
pressors have built-in supervisory systems.
Reciprocating compressors have unique operating dynamics that directly affect their
vibration proles. Unlike most centrifugal machinery, reciprocating machines com-
bine rotating and linear motions that generate complex vibration signatures.
Crankshaft Frequencies
All reciprocating compressors have one or more crankshaft(s) that provide the motive
power to a series of pistons, which are attached by piston arms. These crankshafts
rotate in the same manner as the shaft in a centrifugal machine. However, their
06.Mobley.11 Page 86 Thursday, February 4, 1999 2:51 PM
86 Vibration Fundamentals
Figure 11.7 Internal bullgear drives pinion gears at each stage.
dynamics are somewhat different. The crankshafts generate all of the normal frequen-
cies of a rotating shaft (i.e., running speed, harmonics of running speed, and bearing
frequencies), but the amplitudes are much higher.
In addition, the relationship of the fundamental (1) frequency and its harmonics
changes. In a normal rotating machine, the 1 frequency normally contains between
60 and 70% of the overall, or broadband, energy generated by the machine-train. In
reciprocating machines, however, this prole changes. Two-cycle reciprocating
machines, such as single-action compressors, generate a high second harmonic (2)
and multiples of the second harmonic. While the fundamental (1) is clearly present,
it is at a much lower level.
Frequency Shift Due to Pistons
The shift in vibration prole is the result of the linear motion of the pistons used to
provide compression of the air or gas. As each piston moves through a complete
06.Mobley.11 Page 87 Thursday, February 4, 1999 2:51 PM
87 Machine-Train Monitoring Parameters
Figure 11.8 Two-cycle, or single-action, air compressor cylinder.
cycle, it must change direction two times. This reversal of direction generates the
higher second harmonic (2) frequency component.
In a two-cycle machine, all pistons complete a full cycle each time the crankshaft
completes one revolution. Figure 11.8 illustrates the normal action of a two-cycle, or
single-action, compressor. Inlet and discharge valves are located in the clearance
space and connected through ports in the cylinder head to the inlet and discharge con-
nections. During the suction stroke, the compressor piston starts its downward stroke
and the air under pressure in the clearance space rapidly expands until the pressure
falls below that on the opposite side of the inlet valve (point B). This difference in
pressure causes the inlet valve to open into the cylinder until the piston reaches the
bottom of its stroke (point C).
During the compression stroke, the piston starts upward, compression begins, and at
point D has reached the same pressure as the compressor intake. The spring-loaded
inlet valve then closes. As the piston continues upward, air is compressed until the
pressure in the cylinder becomes great enough to open the discharge valve against the
pressure of the valve springs and the pressure of the discharge line (point E). From
this point, to the end of the stroke (point E to point A), the air compressed within the
cylinder is discharged at practically constant pressure.
The impact energy generated by each piston as it changes direction is clearly visible
in the vibration prole. Since all pistons complete a full cycle each time the crank-
shaft completes one full revolution, the total energy of all pistons is displayed at the
fundamental (1) and second harmonic (2) locations.
06.Mobley.11 Page 88 Thursday, February 4, 1999 2:51 PM
88 Vibration Fundamentals
Figure 11.9 Horizontal reciprocating compressor.
In a four-cycle machine, two complete revolutions (720 degrees) are required for all
cylinders to complete a full cycle.
Piston Orientations
Crankshafts on positive-displacement reciprocating compressors have offsets from
the shaft centerline that provide the stroke length for each piston. The orientation of
the offsets has a direct effect on the dynamics and vibration amplitudes of the com-
pressor. In an opposed-piston compressor where pistons are 180 degrees apart, the
impact forces as the pistons change directions are reduced. As one piston reaches top
dead center, the opposing piston also is at top dead center. The impact forces, which
are 180 degrees out of phase, tend to cancel or balance each other as the two pistons
change directions.
Another conguration, called an unbalanced design, has piston orientations that are
neither in phase nor 180 degrees out of phase. In these congurations, the impact
forces generated as each piston changes direction are not balanced by an equal and
opposite force. As a result, the impact energy and the vibration amplitude are greatly
increased.
Horizontal reciprocating compressors (see Figure 11.9) should have X-Y data points
on both the inboard and outboard main crankshaft bearings, if possible, to monitor the
connecting rod or plunger frequencies and forces.
06.Mobley.11 Page 89 Thursday, February 4, 1999 2:51 PM
89 Machine-Train Monitoring Parameters
Figure 11.10 Screw compressorssteady-state applications only.
Screw
Screw compressors have two rotors with interlocking lobes and act as positive-dis-
placement compressors (see Figure 11.10). This type of compressor is designed for
baseload, or steady-state, operation and is subject to extreme instability should either
the inlet or discharge conditions change. Two helical gears mounted on the outboard
ends of the male and female shafts synchronize the two rotor lobes.
Analysis parameters should be established to monitor the key indices of the compres-
sors dynamics and failure modes. These indices should include bearings, gear mesh,
rotor passing frequencies, and running speed. However, because of its sensitivity to
process instability and the normal tendency to thrust, the most critical monitoring
parameter is axial movement of the male and female rotors.
Bearings
Screw compressors use both Babbitt and rolling-element bearings. Because of the
thrust created by process instability and the normal dynamics of the two rotors, all
screw compressors use heavy-duty thrust bearings. In most cases, they are located on
the outboard end of the two rotors, but some designs place them on the inboard end.
The actual location of the thrust bearings must be known and used as a primary mea-
surement-point location.
Gear Mesh
The helical timing gears generate a meshing frequency equal to the number of teeth
on the male shaft multiplied by the actual shaft speed. A narrowband window should
be created to monitor the actual gear mesh and its modulations. The limits of the win-
dow should be broad enough to compensate for a variation in speed between full load
and no load.
06.Mobley.11 Page 90 Thursday, February 4, 1999 2:51 PM
90 Vibration Fundamentals
Figure 11.11 Major fan classications.
The gear set should be monitored for axial thrusting. Because of the compressors
sensitivity to process instability, the gears are subjected to extreme variations in
induced axial loading. Coupled with the helical gears normal tendency to thrust, the
change in axial vibration is an early indicator of incipient problems.
Rotor Passing
The male and female rotors act much like any bladed or gear unit. The number of
lobes on the male rotor multiplied by the actual male shaft speed determines the rotor-
passing frequency. In most cases, there are more lobes on the female than on the male.
To ensure inclusion of all passing frequencies, the rotor-passing frequency of the
female shaft also should be calculated. The passing frequency is equal to the number
of lobes on the female rotor multiplied by the actual female shaft speed.
Running Speeds
The input, or male, rotor in screw compressors generally rotates at a no-load speed of
either 1800 or 3600 rpm. The female, or driven, rotor operates at higher no-load
speeds ranging between 3600 and 9000 rpm. Narrowband windows should be estab-
lished to monitor the actual running speed of the male and female rotors. The win-
dows should have an upper limit equal to the no-load design speed and a lower limit
that captures the slowest, or fully loaded, speed. Generally, the lower limits are
between 15 and 20% lower than no-load.
Fans
Fans have many different industrial applications and designs vary. However, all fans
fall into two major categories: (1) centerline and (2) cantilever. The centerline cong-
uration has the rotating element located at the midpoint between two rigidly sup-
ported bearings. The cantilever or overhung fan has the rotating element located
outboard of two xed bearings. Figure 11.11 illustrates the difference between the
two fan classications.
06.Mobley.11 Page 91 Thursday, February 4, 1999 2:51 PM
91 Machine-Train Monitoring Parameters
The following parameters are monitored in a typical predictive maintenance program
for fans: aerodynamic instability, running speeds, and shaft mode shape, or shaft
deection.
Aerodynamic Instability
Fans are designed to operate in a relatively steady-state condition. The effective con-
trol range is typically 15 to 30% of their full range. Operation outside of the effective
control range results in extreme turbulence within the fan, which causes a marked
increase in vibration. In addition, turbulent ow caused by restricted inlet airow,
leaks, and a variety of other factors increases rotor instability and the overall vibration
generated by a fan.
Both of these abnormal forcing functions (i.e., turbulent ow and operation outside of
the effective control range) increase the level of vibration. However, when the insta-
bility is relatively minor, the resultant vibration occurs at the vane-pass frequency. As
it becomes more severe, there also is a marked increase in the broadband energy.
A narrowband window should be created to monitor the vane-pass frequency of each
fan. The vane-pass frequency is equal to the number of vanes or blades on the fans
rotor multiplied by the actual running speed of the shaft. The lower and upper limits
of the narrowband should be set about 10% above and below (10%) the calculated
vane-pass frequency. This compensates for speed variations and it includes the broad-
band energy generated by instability.
Running Speeds
Fan running speed varies with load. If xed lters are used to establish the bandwidth
and narrowband windows, the running speed upper limit should be set to the synchro-
nous speed of the motor, and the lower limit set at the full-load speed of the motor.
This setting provides the full range of actual running speeds that should be observed
in a routine monitoring program.
Shaft Mode Shape (Shaft Deection)
The bearing-support structure is often inadequate for proper shaft support because of
its span and stiffness. As a result, most fans tend to operate with a shaft that deects
from its true centerline. Typically, this deection results in a vibration frequency at
the second (2) or third (3) harmonic of shaft speed.
A narrowband window should be established to monitor the fundamental (1), second
(2), and third (3) harmonic of shaft speed. With these windows, the energy associ-
ated with shaft deection, or mode shape, can be monitored.
06.Mobley.11 Page 92 Thursday, February 4, 1999 2:51 PM
92 Vibration Fundamentals
Generators
As with electric motor rotors, generator rotors always seek the magnetic center of
their casings. As a result, they tend to thrust in the axial direction. In almost all cases,
this axial movement, or endplay, generates a vibration prole that includes the funda-
mental (1), second (2) and third (3) harmonic of running speed. Key monitoring
parameters for generators include bearings, casing and shaft, line frequency, and run-
ning speed.
Bearings
Large generators typically use Babbitt bearings, which are nonrotating, lined metal
sleeves (also referred to as uid-lm bearings) that depend on a lubricating lm to
prevent wear. However, these bearings are subjected to abnormal wear each time a
generator is shut off or started. In these situations, the entire weight of the rotating
element rests directly on the lower half of the bearings. When the generator is started,
the shaft climbs the Babbitt liner until gravity forces the shaft to drop to the bottom of
the bearing. This alternating action of climb and fall is repeated until the shaft speed
increases to the point that a uid lm is created between the shaft and Babbitt liner.
Subharmonic frequencies (i.e., less than the actual shaft speed) are the primary evalu-
ation tool for uid-lm bearings and they must be monitored closely. A narrowband
window that captures the full range of vibration frequency components between elec-
tronic noise and running speed is an absolute necessity.
Casing and Shaft
Most generators have relatively soft support structures. Therefore, they require shaft
vibration monitoring measurement points in addition to standard casing measurement
points. This requires the addition of permanently mounted proximity, or displace-
ment, transducers that can measure actual shaft movement.
The third (3) harmonic of running speed is a critical monitoring parameter. Most, if
not all, generators tend to move in the axial plane as part of their normal dynamics.
Increases in axial movement, which appear in the third harmonic, are early indicators
of problems.
Line Frequency
Many electrical problems cause an increase in the amplitude of line frequency, typi-
cally 60 Hz, and its harmonics. Therefore, a narrowband should be established to
monitor the 60-, 120-, and 180-Hz frequency components.
Running Speed
Actual running speed remains relatively constant on most generators. While load
changes create slight variations in actual speed, the change in speed is minor. Gener-
06.Mobley.11 Page 93 Thursday, February 4, 1999 2:51 PM
93 Machine-Train Monitoring Parameters
ally, a narrowband window with lower and upper limits of 10% of design speed is
sufcient.
Process Rolls
Process rolls are commonly found in paper machines and other continuous process
applications. Process rolls generate few unique vibration frequencies. In most cases,
the only vibration frequencies generated are running speed and bearing rotational
frequencies.
However, rolls are highly prone to loads induced by the process. In most cases, rolls
carry some form of product or a mechanism that, in turn, carries a product. For exam-
ple, a simple conveyor has rolls that carry a belt, which carries product from one loca-
tion to another. The primary monitoring parameters for process rolls include bearings,
load distribution, and misalignment.
Bearings
Both nonuniform loading and roll misalignment change the bearing load zones. In
general, either of these failure modes results in an increase in outer-race loading. This
is caused by the failure mode forcing the full load onto one quadrant of the bearings
outer race.
Therefore, the ball-pass outer-race frequency should be monitored closely on all pro-
cess rolls. Any increase in this unique frequency is a prime indication of a load, ten-
sion, or misaligned roll problem.
Load Distribution
By design, process rolls should be uniformly loaded across their entire bearing span
(see Figure 11.12). Improper tracking and/or tension of the belt, or product carried by
the rolls, will change the loading characteristics.
The loads induced by the belt increase the pressure on the loaded bearing and
decrease the pressure on the unloaded bearing. An evaluation of process rolls should
include a cross-comparison of the overall vibration levels and the vibration signature
of each rolls inboard and outboard bearing.
Misalignment
Misalignment of process rolls is a common problem. On a continuous process line,
most rolls are mounted in several levels. The distance between the rolls and the
change in elevation make it extremely difcult to maintain proper alignment.
In a vibration analysis, roll misalignment generates a signature similar to classical
parallel misalignment. It generates dominant frequencies at the fundamental (1) and
second (2) harmonic of running speed.
06.Mobley.11 Page 94 Thursday, February 4, 1999 2:51 PM
94 Vibration Fundamentals
Figure 11.12 Rolls should be uniformly loaded: (a) proper and (b) improper.
Pumps
A wide variety of pumps are used by industry and they can be grouped into two types:
centrifugal and positive displacement. Pumps are highly susceptible to process-
induced or installation-induced loads. Some pump designs are more likely to have
axial- or thrust-induced load problems. Induced loads created by hydraulic forces also
are a serious problem in most pump applications.
Recommended monitoring for each type of pump is essentially the same, regardless
of specic design or manufacturer. However, process variables such as ow, pressure,
load, etc., must be taken into account.
Centrifugal
Centrifugal pumps can be divided into two basic types: end-suction and horizontal
split-case. These two major classications can be broken further into single-stage and
multistage pumps. Each of these classications has common monitoring parameters,
but each also has unique features that alter their forcing functions and the resultant
vibration prole. The common monitoring parameters for all centrifugal pumps
include axial thrusting, vane-pass, and running speed.
Axial Thrusting
End-suction and multistage pumps with in-line impellers are prone to excessive axial
thrusting. In the end-suction pump, the centerline axial inlet conguration is the pri-
mary source of thrust. Restrictions in the suction piping, or low suction pressures, cre-
ate a strong imbalance that forces the rotating element toward the inlet.
06.Mobley.11 Page 95 Thursday, February 4, 1999 2:51 PM
95 Machine-Train Monitoring Parameters
Multistage pumps with in-line impellers generate a strong axial force on the outboard
end of the pump. Most of these pumps have oversized thrust bearings (e.g., Kingsbury
bearings) that restrict the amount of axial movement. However, bearing wear caused
by constant rotor thrusting is a dominant failure mode. Monitoring of the axial move-
ment of the shaft should be done whenever possible.
Hydraulic Instability (Vane Pass)
Hydraulic or ow instability is common in centrifugal pumps. In addition to the
restrictions of the suction and discharge discussed previously, the piping congura-
tion in many applications creates instability. Although ow through the pump should
be laminar, sharp turns or other restrictions in the inlet piping can create turbulent
ow conditions. Forcing functions such as these result in hydraulic instability, which
displaces the rotating element within the pump.
In a vibration analysis, hydraulic instability is displayed at the vane-pass frequency of
the pumps impeller. Vane-pass frequency is equal to the number of vanes in the
impeller multiplied by the actual running speed of the shaft. Therefore, a narrowband
window should be established to monitor the vane-pass frequency of all centrifugal
pumps.
Running Speed
Most pumps are considered constant speed, but the true speed changes with variations
in suction pressure and back-pressure caused by restrictions in the discharge piping.
The narrowband should have lower and upper limits sufcient to compensate for these
speed variations. Generally, the limits should be set at speeds equal to the full-load
and no-load ratings of the driver.
There is a potential for unstable ow through pumps, which is created by both the
design-ow pattern and the radial deection caused by back-pressure in the dis-
charge piping. Pumps tend to operate at their second-mode shape or deection pat-
tern. This mode of operation generates a unique vibration frequency at the second
harmonic (2) of running speed. In extreme cases, the shaft may be deected further
and operate in its third (3) mode shape. Therefore, both of these frequencies should
be monitored.
Positive Displacement
A variety of positive-displacement pumps are commonly used in industrial applica-
tions. Each type has unique characteristics that must be understood and monitored.
However, most of the major types have common parameters that should be monitored.
With the exception of piston-type pumps, most of the common positive-displacement
pumps utilize rotating elements to provide a constant-volume, constant-pressure out-
put. As a result, these pumps can be monitored with the following parameters:
hydraulic instability, passing frequencies, and running speed.
06.Mobley.11 Page 96 Thursday, February 4, 1999 2:51 PM
96 Vibration Fundamentals
Hydraulic Instability (Vane Pass)
Positive-displacement pumps are subject to ow instability, which is created either by
process restrictions or by the internal pumping process. Increases in amplitude at the
passing frequencies, as well as harmonics of both shaft running speed and the passing
frequencies, typically result from instability.
Passing Frequencies
With the exception of piston-type pumps, all positive-displacement pumps have one
or more passing frequencies generated by the gears, lobes, vanes, or wobble plates
used in different designs to increase the pressure of the pumped liquid. These passing
frequencies can be calculated in the same manner as the blade or vane-passing fre-
quencies in centrifugal pumps (i.e., multiplying the number of gears, lobes, vanes, or
wobble plates times the actual running speed of the shaft).
Running Speeds
All positive-displacement pumps have one or more rotating shafts that provide power
transmission from the primary driver. Narrowband windows should be established to
monitor the actual shaft speeds, which are in most cases essentially constant. Upper
and lower limits set at 10% of the actual shaft speed are usually sufcient.
07.Mobley.12 Page 97 Thursday, February 4, 1999 3:03 PM
Chapter12
DATABASEDEVELOPMENT
Valid data are an absolute prerequisite of vibration monitoring and analysis. Without
accurate and complete data taken in the appropriate frequency range, it is impossible
to interpret the vibration proles obtained from a machine-train.
This is especially true in applications that use microprocessor/computer-based sys-
tems. These systems require a database that species the monitoring parameters, mea-
surement routes, analysis parameters, and a variety of other information. This input is
needed to acquire, trend, store, and report what is referred to as conditioned vibra-
tion data.
The steps in developing such a database are (1) collection of machine and process
data and (2) database setup. Input requirements of the software are machine and pro-
cess specications, analysis parameters, data lters, alert/alarm limits, and a variety
of other parameters used to automate the data-acquisition process.
MACHINE AND PROCESS DATA COLLECTION
Database development can be accelerated and its accuracy improved by rst creating
detailed equipment and process information sheets that fully describe each machine
and system to be monitored.
Equipment Information Sheets
The rst step in establishing a database that denes the operating condition of each
machine-train or production system is to generate an equipment information sheet
(EIS) for each machine-train. The information sheet must contain all of the machine-
specic data such as type of operation and information on all of the components that
make up the machine-train.
97
07.Mobley.12 Page 98 Thursday, February 4, 1999 3:03 PM
98 VibrationFundamentals
TypeofOperation
The EIS should dene the type of operation (i.e., constant speed or variable speed) that
best describes the normal operation of each machine-train. This information allows the
analyst to determine the best method of monitoring and evaluating each machine.
Constant-Speed Machinery
Few, if any, machines found in a manufacturing or production plant are truly constant
speed. While the nameplate and specications may indicate that a machine operates at
a xed speed, it will vary slightly in normal operation.
The reason for speed variations in constant-speed machinery is variation of process
load. For example, a centrifugal pumps load will vary due to the viscosity of the uid
being pumped or changes in suction or discharge pressure. The pump speed will
change as a result of these load changes.
As a general rule, the speed variation in a constant-speed machine is about 15%. For
an electric motor, the actual variation can be determined by obtaining the difference
between the amperage drawn under full-load and no-load conditions. This difference,
taken as a percentage of full-load amperage draw, provides the actual percentage of
speed-range variation that can be expected.
Variable-Speed Machinery
For machinery and process systems that have a wide range of operating speeds, the
data sheet should provide a minimum and maximum speed that can be expected dur-
ing normal operation. In addition, a complete description of other variables (e.g.,
product type) that affect the machines speed should be included. For example, a pro-
cess line may operate at 500 ft/min with product A and 1000 ft/min with product B.
Therefore, the data sheet must dene all of the variables associated with both product
A and product B.
Constant Versus Variable Load
As with constant-speed machines, true constant-load machines are rare. For the few
that may be specied as having constant load, there are factors that cause load
changes to occur. These factors include variations in product, operating conditions,
and ambient environment. These variations will have a direct, and often dramatic,
impact on a machines vibration prole.
Variations in load, no matter how slight, alter the vibration prole generated by a
machine or system. The relationship between load and the vibration energy generated
by a machine can be a multiple of four. In other words, a 10% change in load may
increase or decrease the vibration energy by 40%.
When using vibration data as a diagnostic tool, you must always adjust or normalize
the data to the actual load that was present when the data set was acquired.
07.Mobley.12 Page 99 Thursday, February 4, 1999 3:03 PM
99 DatabaseDevelopment
MachineComponents
The EIS should provide information on all components (e.g., bearings, gears, gear-
boxes, electric motors, pumps) that make up the machine-train. Because these compo-
nents generate vibration energy and unique frequency components, this information is
essential for proper analysis. At a minimum, the information sheet must include
detailed bearing information, passing frequencies, and nameplate data.
Bearings
All bearings in a machine-train must be identied. For example, rolling-element bear-
ing data must include the manufacturers part number, bearing geometry, and the
unique rotational frequencies that it will generate. Rotational frequencies can be
determined by using the bearing part number to look them up in the database that is
included in most vibration monitoring software programs. They also can be obtained
from the bearing vendor. Babbitt or sleeve bearing data must include type (e.g., plain,
tilting-pad), as well as manufacturer and part number.
Passing Frequencies
All components that generate a passing frequency must be included on the informa-
tion sheet. Such components include fan or compressor blades, vanes on pump impel-
lers, rotor bars in electric motors, and gear teeth on both the pinion and bullgear in a
gear set. The number of vanes, blades, and gear teeth must be recorded on the infor-
mation sheet. The passing frequency is the number of vanes, blades, etc., times the
rotation speed of the shaft on which they are mounted.
Nameplate Data
Each machine-train component has a vendors nameplate permanently attached to its
housing. The EIS should include all nameplate data, including the serial number that
uniquely identies a machine or component.
Knowing the serial number allows detailed information on a machine or component to
be obtained. This is possible because machinery manufacturers must maintain records
for their products. These records, which are usually identied by serial number, con-
tain complete design and performance data for that specic unit. For example, it is
possible to obtain a performance curve or complete bill of materials for each pump
found in a plant.
Process Information Sheets
A process information sheet (PIS) should be developed for each machine-train and
production process that is to be included in a predictive maintenance program. These
data sheets should include all process variables that affect the dynamics and vibration
proles of the monitored components.
07.Mobley.12 Page 100 Thursday, February 4, 1999 3:03 PM
100 VibrationFundamentals
Many production and process systems handle a wide range of products. They typi-
cally have radically different machine and system operating parameters, as well as
variable speeds and loads for each of the products they process.
Each process parameter directly affects both the machinery dynamics and the vibra-
tion proles. For example, the line tension, strip width, and hardness of the incoming
strip radically affect the vibration prole generated by a continuous process line in a
steel mill. With few exceptions, process variations such as these must be considered in
the vibration analysis.
DATABASE SETUP
The input-data requirements and steps needed to set up the database for a computer-
based vibration monitoring program vary depending on the analyzer/software vendor
and the systems capabilities. This section discusses the input required for such a data-
base. However, this information should be used in conjunction with the vendors
users manual to ensure proper implementation.
The key elements of database setup discussed in this section are analysis parameter
sets, data lters (i.e., bandwidths, averaging, and weighting), limits for alerts and
alarms, and data-acquisition routes.
Analysis Parameter Sets
The software used to manage the data incorporate what are referred to as analysis
parameter sets (APSs). APSs dene and specify machine dynamics, components, and
failure modes to be monitored.
Most microprocessor-based systems permit a maximum of 256 APSs per database.
This limit could be restrictive if the analyst wishes to establish a unique APS for each
machine-train. To avoid this problem, APSs should be established for classes of
machine-trains. For example, a group of bridle gearboxes that are identical in both
design and application should share the same APS.
This approach provides two benets. One is that it simplies database development
because one parameter set is used for multiple machine-trains. Therefore, less time is
required to establish them. The other is that this approach permits direct comparison
of multiple machine-trains. Since all machine-trains in a class share a common APS,
the data can be directly compared. For example, the energy generated by a gear set is
captured in a narrowband window established to monitor gear mesh. With the same
APS, the gear mesh narrowbands can be used to compare all gear sets within that
machine-train classication.
07.Mobley.12 Page 101 Thursday, February 4, 1999 3:03 PM
101 DatabaseDevelopment
Data Filters
When selecting the bandwidth frequency range to use for data collection in a vibra-
tion-monitoring system, one might be tempted to select the broadest range available.
If enough computing power were available, we could simply gather data over an in-
nite frequency range, analyze the data, and be assured that no impending failures were
missed. However, practicalities of limited computing power prevent us from taking
this approach.
Therefore, it is necessary to lter or screen the data we collect using our knowledge
of the machinery being monitored. This is necessary to make collection, storage, and
analysis of the data manageable with the equipment available. Electronic lters screen
the quantity and quality of data that are collected. Mathematical ltering techniques
such as resolution, averaging, and weighting are used on the data that are collected.
Bandwidth
Bandwidth frequency range settings are crucial to obtaining meaningful data for a
vibration-based predictive maintenance program. Because of limits inherent with
computer-based data collection and analysis systems (i.e., limited storage and data-
handling capacity), these settings must be properly specied to obtain frequency data
in the range generated by machine components where failures occur. Improper set-
tings will likely yield data in frequency ranges where problems do not exist and miss
critical clues to serious problems with the machinery.
Analysis Type
As discussed previously, data-collection analyzers incorporate analysis parameter sets
that allow the user to control the data-gathering process. APSs provide the option of
selecting either frequency analysis for xed-speed machinery or orders analysis for
variable-speed machinery.
Constant Speed: Frequency Analysis
Constant-speed machinery generates a relatively xed set of frequency components
within its signature. Therefore, specic APSs can be established to monitor using fre-
quency analysis. Because speed is relatively constant, the location of specic fre-
quency components (e.g., running speed) will not change greatly. Therefore, the
broadband and each narrowband window can be established with a constant minimum
and maximum frequency limit, which are referred to as xed lters.
The position of these xed lters should be set to ensure capture of information that is
needed. The lter settings are determined from the speed range (i.e., no-load to full-
load range) of the primary driver. In addition, the lower and upper limits of each lter
should be adjusted by 10 to 15% to allow for slight variations in speed.
07.Mobley.12 Page 102 Thursday, February 4, 1999 3:03 PM
102 VibrationFundamentals
Variable Speed: Orders Analysis
In a variable-speed machine, the unique frequencies generated by components such as
bearings and gear sets do not remain constant. As the speed changes, the unique fre-
quency components vary in direct proportion to the speed change. For this type of
machinery, the analyzers orders analysis option is used to automatically adjust each
of the lters used to set the bandwidth and narrowbands for each data set to the true
machine speed.
The analyzer automatically moves the lters that designate the lower and upper limits
of each narrowband window to correspond with the actual running speed at the time
the data are collected. To activate this function, the technician must either manually
enter the running speed or use a tachometer input to trigger data acquisition.
Boundary Conditions and Resolution
The frequency boundary conditions and resolution for the full FFT signature depend
on the specic system being used. Typically, the full-signature capability of various
predictive maintenance systems has a lower frequency limit of 10 Hz and an upper
limit of 10 to 30 kHz. A few special low-frequency analyzers have a lower limit of 0.1
Hz, but retain the upper limit of 30 kHz. Typical resolutions are 100 to 12,800 lines.
Maximum Frequency
The dynamics of each machine-train determines the maximum frequency, F
MAX
, that
should be used for both data acquisition and analysis. The frequency must be high
enough to capture and display meaningful data, but not so high that resolution is lost
and meaningful data ltered out. As a general rule, when setting F
MAX
, it is necessary
to take into account the harmonics of running speed and frequencies of components
such as rolling-element bearings and gear mesh. Therefore, F
MAX
should be set to the
maximum frequency encountered in any of these.
Gear Mesh
The bandwidth should include at least the second harmonic of the calculated gear-
mesh frequency. This permits early detection of misalignment, gear wear, and other
abnormal operating dynamics. If excessive axial thrusting is expected, the third har-
monic of gear mesh also should be included. For example, helical gears are prone to
generate axial movement that increases as the gears wear.
Rolling-Element Bearings
The ability to monitor rolling-element or antifriction bearing defects requires the
inclusion of multiples of their rotating frequency. For example, with ball-pass inner-
race bearings, the bandwidth should include the second harmonic (2).
Running-Speed Harmonics
When setting bandwidth, at least three harmonics of running speed should be
included to ensure the ability to quantify the operating-mode shape of the shaft. This
is accomplished by setting F
MAX
to at least three times the running speed.
07.Mobley.12 Page 103 Thursday, February 4, 1999 3:03 PM
103 DatabaseDevelopment
Generally, shafts deect from their true centerline during normal, as well as abnor-
mal, operation. In normal operation, this deection is slight and results in a rst
mode that creates a frequency component at the actual running speed, or rst har-
monic (1). As instability increases, the shaft deects more and more. Under abnor-
mal operating conditions, the shaft typically deforms into either the second mode or
third mode. This deection creates unique frequencies at the second (2) and third
(3) harmonics of running speed. In addition, some failure modes (e.g., parallel mis-
alignment) increase the energy level of one or all three of these frequencies.
Minimum Frequency
The data-collection hardware and software permit the selection of the minimum fre-
quency, F
MIN
, or the low-frequency cutoff below which no data are acquired and
stored during the monitoring process. In most applications, however, this option
should not be used.
Selecting a low-frequency cutoff does not improve resolution and is strictly an arbi-
trary omission of visible frequency components within a vibration signature. The FFT
is calculated on a bandwidth having a lower limit of zero and an upper limit equal to
the maximum frequency, F
MAX
, which is selected by the user. The only reason for
selecting a minimum frequency other than zero is to remove unneeded low-frequency
components from the signature display.
Resolution
Resolution is the degree of spacing of visible frequency components in the vibration
signature and is proportional to the bandwidth. The equation for resolution follows:
Bandwidth
F
MAX
F
MIN
Resolution = ----------------------------------------------- = -----------------------------------------------
Lines of Resolution Lines of Resolution
From this equation, it is apparent that the bandwidth should be as small as possible to
minimize the spacing and avoid missing important data. The typical number of lines
of resolution is 100 to 12,800, but it is important to make sure that the selected band-
width and resolution include all pertinent frequency components and that they are vis-
ible in the signature.
For routine monitoring, 800 lines of resolution are recommended. Higher resolution
may be needed for root-cause analysis, but this requires substantially more memory in
both the analyzer and host computer. While the latter is not a major problem, higher
resolution reduces the number of measurement points that can be acquired with an
analyzer without transferring acquired data to the host computer. This can greatly
increase the time required to complete a measurement route and should be avoided
when possible.
The combination of bandwidth and lines of resolution selected for each machine-train
must affect separation of the unique frequency components that represent a machines
operating dynamics. Resolution can be improved by reducing F
MAX
, increasing the
lines of resolution, or a combination of both.
07.Mobley.12 Page 104 Thursday, February 4, 1999 3:03 PM
104 VibrationFundamentals
For example, if the analyzer can provide a 400-line FFT, the resolution of a signature
taken with a bandwidth of 0 to 20,000 Hz will be 50 Hz, or 3000 rpm, for each dis-
played line. The same 400-line FFT will provide a resolution of 2.5 Hz, or 150 rpm,
with an F
MAX
of 1000 Hz.
It is important to remember that the rst two and last two lines of resolution are lost
when the FFT is calculated. In the example just described (F
MAX
= 1000 Hz), the rst
visible speed is 450 rpm and the highest visible speed is 59,700 rpm. Because the FFT
always drops the rst two and last two lines of resolution, the rst visible frequency is
three times the calculated resolution (3 150 = 450) and the highest visible frequency
is lowered by two lines (59,700 is visible, but 59,850 and 60,000 are not shown).
Narrowbands
Analysis using narrowbands is based on specifying a series of ltered windows for a
machine component or failure mode. The analyst can establish up to 12 narrowbands
for each measurement point on each machine-train.
This concept reduces the manual analysis required for each data set. The analyst can
scan the documentation that is generated (i.e., the exception report and trend charts)
for each of the selected narrowbands to determine if further analysis is required.
Before implementing the analyzers narrowband capability, the analyst should rst
understand the dynamics of each machine-train. Once this is done, establish narrow-
bands that bracket each of the bandwidths identifying each of the major components.
At a minimum, establish a narrowband window around the following:
Each primary running speed
Each gear-mesh frequency (including sidebands)
Each set of bearing frequencies
Each blade/vane-pass frequency
Each belt frequency
Constant-Speed Machinery
If a machine-train operates at constant speed, the best method is to set the windows
using the F
MIN
and F
MAX
frequencies associated with the specic component. For
example, a narrowband window could be established to monitor the energy generated
by a gear set by dening the minimum and maximum frequencies bounding the gear
mesh. The bandwidth of the narrowband should be broad enough to include the mod-
ulations, or sidebands, generated by the meshing.
For example, a gear with 50 teeth generates a gear mesh at 50 times the running speed
of its shaft (50). If the shaft turns at 100 rpm, the gear mesh frequency is 5000 cpm
(also rpm). The modulations of a normal gear will occur at multiples of shaft speed
(100 rpm). Therefore, in order to capture ve sidebands on each side of the gear mesh
07.Mobley.12 Page 105 Thursday, February 4, 1999 3:03 PM
105 DatabaseDevelopment
frequency, the narrowband window should be established with lters set at 4500 and
5500 cpm [5000 (5 100)].
In actual practice, the narrowband lters should be somewhat greater than those in the
example. Since constant-speed machines tend to have a slight variation in speed due
to load variations, the narrowbands should be adjusted to compensate for these varia-
tions. In the example given previously, the limit of the lower lter should be decreased
by 10% and the upper limit raised by 5% to compensate for speed variation.
Variable-Speed Machinery
Variable-speed machine-train narrowband windows should be converted to their rela-
tionship to the running speed (1). For example, if the frequency of the ball-pass
inner-race rolling-element bearing is calculated to be 5.9 times the primary shaft run-
ning speed, then the narrowband window should be set as 5.3 to 6.2. This allows
the microprocessor to track the actual bearing rotational frequency regardless of the
variation in running speed.
As a general rule, the bandwidth of each narrowband should be just enough to capture
the energy generated by the monitored component. Because orders analysis automati-
cally adjusts the lters used to acquire narrowband energy data, these windows can be
somewhat tighter than those used for frequency analysis.
Antialiasing Filters
Vibration data collected with a microprocessor-based analyzer can be ltered and
conditioned to eliminate nonrecurring events and their associated vibration proles.
Antialiasing lters are incorporated into data-collection analyzers specically to
remove spurious signals such as impacts. While the intent behind the use of antialias-
ing lters is valid, their use can distort a machines vibration prole.
Averaging
All machine-trains are subject to random, nonrecurring vibration as well as periodic
vibration. Therefore, it is advisable to acquire several sets of data and average them to
eliminate the spurious signals. Averaging also improves the repeatability of the data
since only the continuous signals are retained.
Number of Averages
Typically, a minimum of three samples should be collected for an average. However,
the factor that determines the actual number is time. One sample takes 3 to 5 sec, a
four-sample average takes 12 to 20 sec, and a 1000-sample average takes 50 to 80 min
to acquire. Therefore, the nal determination is the amount of time that can be spent
at each measurement point.
In general, three to four samples are acceptable for good statistical averaging and
keeping the time required per measurement point within reason. Exceptions to this
include low-speed machinery, transient-event capture, and synchronous averaging.
07.Mobley.12 Page 106 Thursday, February 4, 1999 3:03 PM
106 VibrationFundamentals
Table 12.1 Overlap Averaging Options
Overlap(%) Description
0 No overlap. Data trace update rate is the same as the block processing rate.
This rate is governed by the physical requirements that are internally driven
by the frequency range of the requested data.
25 Terminates data acquisition when 75% of each block of new data is
acquired. The last 25% of the previous sample (of the 75%) will be added to
the new sample before processing is begun. Therefore, 75% of each sample
is new. As a result, accuracy may be reduced by as much as 25% for each
data set.
50 The last 50% of the previous block is added to a new 50% or half-block of
data for each sample. When the required number of samples is acquired and
processed, the analyzer averages the data set. Accuracy may be reduced to
50%.
75 Each block of data is limited to 25% new data and the last 75% of the previ-
ous block.
90 Each block contains 10% new data and the last 90% of the previous block.
Accuracy of average data using 90% overlap is uncertain. Since each block
used to create the average contains only 10% of actual data and 90% of a
block that was extrapolated from a 10% sample, the result cannot be repre-
sentative of the real vibration generated by the machine-train.
Source: Integrated Systems, Inc.
Overlap Averaging
Many of the microprocessor-based vibration monitoring systems offer the ability to
increase their data-acquisition speed. This option is referred to as overlap averaging.
While this approach increases speed, it is not generally recommended for vibration
analysis. Overlap averaging reduces the accuracy of the data and must be used with
caution. Its use should be avoided except where fast transients or other unique
machine-train characteristics require an articial means of reducing the data-acquisi-
tion and processing time.
When sampling time is limited, a better approach is to reduce or eliminate averaging
altogether in favor of acquiring a single data block, or sample. This reduces the acqui-
sition time to its absolute minimum. In most cases, the single-sample time interval is
less than the minimum time required to obtain two or more data blocks using the max-
imum overlap-averaging sampling technique. In addition, single-sample data are
more accurate.
Table 12.1 describes overlap-averaging options. Note that the approach described in
this table assumes that the vibration prole of monitored machines is constant.
07.Mobley.12 Page 107 Thursday, February 4, 1999 3:03 PM
107 DatabaseDevelopment
WindowSelection:WeightingorSignalConditioning
The user can select the type of signal conditioning, or weighting used to display the
vibration signature. For routine monitoring, the Hanning window should be selected.
The at-top window should be used for waterfall analysis, which modies the prole
of each frequency component so that the true amplitude is displayed.
Hanning Correction
The Hanning correction provides the best capture of the individual frequency compo-
nents of a signature. However, this weighting factor may distort the actual amplitude
of the frequency components. Nevertheless, it is used for routine monitoring using
FFT analysis.
Flat-Top Weighting
Flat-top weighting provides the best representation of the actual amplitude of each
frequency component. However, it may distort the actual location (i.e., frequency) of
each component.
Flat-top weighting is useful when doing waterfall analysis. Even though the actual
location of each frequency component may be slightly out of position, the prole is
more visible when closely packed into a waterfall display. However, it is not normally
used for single-channel FFT analysis.
Alert/Alarm Limits
All microprocessor-based predictive maintenance systems are designed to automati-
cally evaluate degradation of the machine-train by monitoring the change in ampli-
tude of vibration using trending techniques. By establishing a series of alert/alarm
limits in the database, the system can automatically notify the analyst that degradation
is occurring. At least three levels of alert/alarms should be established: 1) low-limit
alert, (2) maximum rate of change, and (3) absolute fault.
Low-LimitAlert
The rst alert, the low-limit alert, should be set at the lowest vibration amplitude that
will be encountered from a normally operating machine-train. This value is needed to
ensure that valid data are taken with the microprocessor. If this minimum amplitude is
not reached, the system alerts the operator, who can retake or verify the data point.
Low-limit selection is arbitrary, but should be set slightly above the noise oor of the
specic microprocessor used to acquire data.
MaximumRateofChangeAlert
The second alert, the maximum rate of change alert, is used to automatically notify
the operator that, based on statistical data, the rate of degradation has increased above
the preselected norm. Since the vibration amplitudes of all machine-trains increase as
07.Mobley.12 Page 108 Thursday, February 4, 1999 3:03 PM
108 VibrationFundamentals
normal wear occurs, the statistical rate of this normal increase should be trended. A
drastic change in this rate is a major indication that a problem is developing.
The system should be able to establish the norm based on trends developed over time.
However, the analyst must establish the level of deviation that triggers the alarm. The
level of deviation in rate depends on the mechanical condition of each machine-train.
For a new machine in good operating condition, the limit is typically set at two times
the norm. However, this must be adjusted based on the actual baseline of the machine.
Note that it is better to set the limit too low initially and adjust it later.
Absolute-FaultAlarm
The third limit, the absolute-fault alarm, is the most critical of the alert/alarm limits.
When this limit is reached, the probability of catastrophic failure within 1000 operat-
ing hours is greater than 90%.
Absolute-fault limits are typically based on industrial standards for specic classica-
tions of machinery. Generally, these standards are based on a ltered broadband limit
and are not adjusted for variables such as speed, load, or mounting conguration.
However, vibration amplitude and its severity depend on speed and load. Therefore,
alert/alarm limits must be adjusted for variations in both of these critical factors.
Effect of Speed on Limit
Vibration-severity charts, such as the Rathbone chart illustrated in Figure 12.1, pro-
vide a basis for establishing the absolute-fault limit for machinery. Note, however,
that the Rathbone chart does not adjust the maximum limit for speedsomething that
can cause a serious problem in most industrial applications.
A 2-mil shaft displacement at 600 rpm is acceptable for most applications, but the
same displacement at 1800 rpm is considered severe. Therefore, alert/alarm limits
must be established based on the actual speed range of each machine-train. When cas-
ing severity is used (i.e., data are taken from the bearing caps rather than actual shaft
displacement), the limits can be grouped into three basic speed ranges: less than 299
rpm, 300 to 1199 rpm, and 1200 through 3600 rpm. Tables 12.212.4 provide the
alert/alarm limits for each speed range.
Effect of Load on Limit
Load has a direct impact on the vibration energy generated by a machine-train. For
example, a centrifugal compressor operating at full load will have a lower level of
vibration than the same compressor operating at 50% load. This change in vibration
energy is the direct result of a corresponding change in the spring stiffness of the
rotating element under varying load conditions.
The rated, or design, load of a machine establishes the following elements: (1) spring
constant, (2) stiffness of the rotating element, and (3) damping coefcient of its sup-
07.Mobley.12 Page 109 Thursday, February 4, 1999 3:03 PM
109 DatabaseDevelopment
Figure 12.1 Rathbone chart.
Table 12.2 Alarm Limits for 1200 rpm and Higher
Bandwidth Alert Alarm AbsoluteFault
Overall 0.15 IPS-PK 0.30 IPS-PK 0.628 IPS-PK
1 Narrowband 0.10 IPS-PK 0.20 IPS-PK 0.40 IPS-PK
2 Narrowband 0.05 IPS-PK 0.10 IPS-PK 0.20 IPS-PK
3 Narrowband 0.04 IPS-PK 0.08 IPS-PK 0.15 IPS-PK
1 Gear mesh 0.05 IPS-PK 0.10 IPS-PK 0.2 IPS-PK
Rolling-element bearing 0.05 IPS-PK 0.10 IPS-PK 0.2 IPS-PK
Blade/vane pass 0.05 IPS-PK 0.10 IPS-PK 0.2 IPS-PK
Source: Integrated Systems, Inc.
07.Mobley.12 Page 110 Thursday, February 4, 1999 3:03 PM
110 VibrationFundamentals
Table 12.3 Alarm Limits for 300 to 1199 rpm
Bandwidth Alert Alarm AbsoluteFault
Overall 0.10 IPS-PK 0.15 IPS-PK 0.30 IPS-PK
1 Narrowband 0.05 IPS-PK 0.10 IPS-PK 0.20 IPS-PK
2 Narrowband 0.02 IPS-PK 0.05 IPS-PK 0.10 IPS-PK
3 Narrowband 0.01 IPS-PK 0.03 IPS-PK 0.06 IPS-PK
1 Gear mesh 0.02 IPS-PK 0.04 IPS-PK 0.08 IPS-PK
Rolling-element bearing 0.03 IPS-PK 0.05 IPS-PK 0.10 IPS-PK
Blade/vane pass 0.03 IPS-PK 0.05 IPS-PK 0.10 IPS-PK
Source: Integrated Systems, Inc.
Table 12.4 Alarm Limits for 299 rpm and Below
Bandwidth Alert Alarm AbsoluteFault
Overall 0.05 IPS-PK 0.10 IPS-PK 0.20 IPS-PK
1 Narrowband 0.03 IPS-PK 0.06 IPS-PK 0.12 IPS-PK
2 Narrowband 0.01 IPS-PK 0.03 IPS-PK 0.06 IPS-PK
3 Narrowband 0.01 IPS-PK 0.02 IPS-PK 0.04 IPS-PK
1 Gear mesh 0.02 IPS-PK 0.03 IPS-PK 0.05 IPS-PK
Rolling-element bearings 0.02 IPS-PK 0.03 IPS-PK 0.05 IPS-PK
Blade/vane pass 0.02 IPS-PK 0.03 IPS-PK 0.05 IPS-PK
Source: Integrated Systems, Inc.
port system. Therefore, when load varies from design, the stiffness of the rotor and
the rotor-support system also must change.
This change in vibration energy can be clearly observed in trend data acquired from
machine-trains. A sawtooth trend is common to most predictive maintenance pro-
grams, which can be directly attributed to variations in load. The only way to compen-
sate for load variations is to track the actual load associated with each data set.
Mounting Conguration and Operating Envelope
Industrial standards, such as the Rathbone severity chart, assume that the machine is
rigidly mounted on a suitable concrete foundation. Machines mounted on deck-plate
or on exible foundations have higher normal vibration proles and cannot be evalu-
ated using these standards.
In addition, industrial standards assume a normal operating envelope. All machines
and process systems have a nite operating range and must be operated accordingly.
Deviations from either the design operating envelope or from best operating practices
will adversely affect the operating condition and vibration level of the machine.
07.Mobley.12 Page 111 Thursday, February 4, 1999 3:03 PM
111 DatabaseDevelopment
Figure 12.2 Typical measurement route.
Data-Acquisition Routes
Most computer-based systems require data-acquisition routes to be established as part
of the database setup. These routes specically dene the sequence of measurement
points and, typically, a route is developed for each area or section of the plant. With
the exception of limitations imposed by some of the vibration monitoring systems,
these routes should dene a logical walking route within a specic plant area. A typi-
cal measurement is shown in Figure 12.2.
Most of the computer-based systems permit rearrangement of the data-acquisition
sequence in the eld. They provide the ability to skip through the route until the
appropriate machine, or the measurement point of a machine, is located. However,
this manual adjustment of the preprogrammed route is time consuming and should be
avoided whenever possible.
08.Mobley.13 Page 112 Thursday, February 4, 1999 3:10 PM
Chapter13
VIBRATIONDATAACQUISITION
Limitations on data acquisition arise due to the use of portable handheld, micropro-
cessor-based analyzers to obtain data. Limits are cost, weight restrictions, and the fact
that they are generally designed for a technician to manually take a series of single
measurements directly from individual machine-trains or machine-train components.
Therefore, this discussion is limited to the best practices for acquiring single-channel,
frequency-domain data using portable, handheld analyzers. It does not address multi-
channel or other nonserial data-acquisition techniques.
Data-acquisition and vibration-detection equipment (i.e., analyzers and transducers)
are critical factors that determine the success or failure of a vibration monitoring and
analysis program. Their accuracy, proper usage, and mounting determine if valid data
are collected. An optimum program is based on the accuracy and repeatability of the
data, both of which are negatively affected by the use of the wrong transducer or
mounting technique.
TRANSDUCERS
The three basic types of transducers that can be used for monitoring the mechanical
condition of plant machinery are (1) displacement probes (measures movement), (2)
velocity transducers (measures energy due to velocity), and (3) accelerometers (mea-
sures force due to acceleration). Each has specic applications in a monitoring pro-
gram, while each also has limitations.
MEASUREMENT ORIENTATION
Most vibration-monitoring programs rely on data acquired from machine housings or
bearing caps. The only exceptions are applications that require direct measurement of
112
08.Mobley.13 Page 113 Thursday, February 4, 1999 3:10 PM
113 VibrationDataAcquisition
Figure 13.1 Measurement points should provide shortest direct mechanical link to shaft.
actual shaft displacement to gain an accurate picture of the machines dynamics.
Transducers used to acquire the data are mounted either radially or axially.
MEASUREMENT LOCATIONS
Each measurement point, typically located on the bearing housing or machine casing,
should provide the shortest direct mechanical link to the shaft. Figure 13.1 illustrates
such a location oriented in both the axial and radial planes. If a transducer is not
mounted in an appropriate location, the data will be distorted by noise such as uid
ow in the bearing reservoir or through the machine.
Measurement locations should be permanently marked to ensure repeatability of data.
If transducers are permanently mounted, the location can be marked with a center-
punch, paint, or any other method that identies the point. The following sections give
the recommended locations and orientations of measurement points for the following
common machines or machine components: compressors, electric motors, fans and
blowers, gearboxes, process rolls, and pumps.
Compressors
In most cases, measurement point locations for compressors are identical to those of
pumps and fans. If a compressor is V-belt driven, the primary measurement point
08.Mobley.13 Page 114 Thursday, February 4, 1999 3:10 PM
114 VibrationFundamentals
should be in a plane opposing the side load created by the belts on the inboard and
outboard bearings. The secondary point should be at 90 degrees to the primary point.
At least one axial measurement point should be located on each compressor shaft.
Axial data are helpful in identifying and quantifying thrust (i.e., induced) loads cre-
ated by both the process and any potential compressor-element problems, such as
imbalance, cracked blade, etc.
In applications where numerous compressors are in proximity, an additional measure-
ment point on the base is useful for identifying structural resonance or cross-talk
between the units.
Centrifugal
The two major types of centrifugal compressors used in industrial applications are in-
line and bullgear compressors.
In-Line Centrifugal Compressors
Measurement locations for in-line centrifugal compressors should be based on the
same logic as discussed for pumps. Impeller design and orientation, as well as the
inlet and discharge congurations, are the dominant reasons for point location. (Fig-
ure 11.5 illustrates a typical multistage, in-line compressor.)
The in-line impeller conguration generates high axial thrusting, which increases the
importance of the axial (Z-axis) measurement point. That point should be on the xed
bearing and oriented toward the driver.
In addition, this type of compressor tends to have both the suction and discharge ports
on the same side of the compressors housing. As a result, there is a potential for aero-
dynamic instability within the compressor. Orientation of the primary (X-axis) radial
measurement point should be opposite the discharge port and oriented toward the dis-
charge. The secondary (Y-axis) radial point should be in the direction of shaft rotation
and 90 degrees from the primary radial point.
Bullgear Compressors
Because of the large number of these machines being manufactured, proper locations
for displacement, or proximity, probes have been established by the various machine
manufacturers. Nearly all of these compressors are supplied by the original equipment
manufacturer (OEM) with one or two proximity probes already mounted on each pin-
ion shaft and, sometimes, one probe on the bullgear shaft. These probes can be used to
obtain vibration data with the microprocessor-based portable analyzers. However,
they must be augmented with casing measurements acquired from suitable acceler-
ometers. This is necessary because there are two problems with the proximity data.
First, most of the OEM-supplied data-acquisition systems perform signal condition-
ing on the raw data acquired from the probes. If the conditioned signal is used, there is
08.Mobley.13 Page 115 Thursday, February 4, 1999 3:10 PM
115 VibrationDataAcquisition
a bias in the recorded amplitude. This bias may increase the raw-data signal by 30 to
50%. If data from the proximity probes are to be used, it is better to acquire it before
signal conditioning. This can be accomplished by tapping into the wiring between the
probe and display panel.
The second problem is data accuracy. The pinions on most bullgear compressors
rotate at speeds between 20,000 and 75,000 rpm. While these speeds, for the most
part, are within the useful range of a proximity probe (600 to 60,000 rpm), the fre-
quencies generated by common components (i.e., tilting-pad bearings and impeller
vane-pass) are well outside this range. In addition, proximity probes depend on a
good sight picture, which means a polished shaft that has no endplay or axial move-
ment. Neither of these conditions is present in a bullgear compressor.
Primary (X-axis) and secondary (Y-axis) radial measurements should be acquired
from both bearings on the bullgear shaft. If the shaft has Babbitt bearings, it is a good
practice to periodically acquire four radial readings, one at each quadrant of the bear-
ing, to determine the load zones of the bearing. Normal vertical and horizontal loca-
tions are acceptable for the routine readings, but primary (X-axis) measurement points
should be in the horizontal plane (i.e., 90 degrees from vertical in the direction of
rotation). For clockwise rotation, the primary should be on the right side and for coun-
terclockwise on the left.
Because a bullgear compressor incorporates a large helical gear, the shaft displays
moderate to high axial thrusting. Therefore, an axial (Z-axis) measurement point
should be acquired from the thrust (outboard) bearing oriented toward the driver.
The pinion shafts in this type of compressor are inside the housing. As a result, it is
difcult to obtain radial measurements directly. A cross-sectional drawing of the com-
pressor is required to determine the best location and orientation for the measurement
points.
PositiveDisplacement
Two major types of positive-displacement compressors are used in industrial applica-
tions: reciprocating and screw.
Reciprocating Compressors
Limitations of the frequency-domain analysis prevent total analysis of reciprocating
compressors. It is limited to the evaluation of the rotary forces generated by the main
crankshaft. Therefore, time-domain and phase analysis are required for complete
diagnostics.
The primary (X-axis) radial measurement point should be located in a plane opposite
the piston and cylinder. Its orientation should be toward the pistons stroke. This ori-
entation provides the best reading of the impacts and vibration prole generated by
the reversing linear motion of the pistons. The secondary (Y-axis) radial measurement
08.Mobley.13 Page 116 Thursday, February 4, 1999 3:10 PM
116 VibrationFundamentals
Figure 13.2 Typical cross-section of a reciprocating compressor.
point should be spaced at 90 degrees to the primary point and in the direction of rota-
tion at the main crankshaft. This conguration should be used for all accessible main
crankshaft bearings. Figure 13.2 provides a typical cross-section of a reciprocating
compressor, which will assist in locating the best measurement points. Similar draw-
ings are available for most compressors and can be obtained from the vendor. There
should be little axial thrusting of the main crankshaft, but an axial (Z-axis) measure-
ment point should be established on the xed bearing, oriented toward the driver.
If the vibration analyzer permits acquisition of time-domain data, additional time-
waveform data should be obtained from the intermediate guide as well as the inlet and
discharge valves. The intermediate guide is located where the main crankshaft lever
arm connects to the piston rod. Time waveforms from these locations detect any bind-
ing or timing problems that may exist in the compressor.
Screw Compressors
Figure 11.10 illustrates a typical single-stage screw compressor. Radial measurements
should be acquired from all bearing locations in the compressor. The primary bearing
locations are the inboard, or oat, bearings on the driver side of the compressor hous-
ing and the xed bearing located on the outboard end of each shaft.
In most cases, the outboard bearings are not directly accessible and measurement
points must be located on the compressors casing. Extreme care must be taken to
ensure proper positioning. A cross-sectional drawing facilitates selection of the best,
most direct mechanical link to the these bearings.
The primary (X-axis) radial measurement point should be located opposite the mesh
of the rotors and oriented toward the mesh. In the illustration, the primary point is on
08.Mobley.13 Page 117 Thursday, February 4, 1999 3:10 PM
117 VibrationDataAcquisition
the top of the housing and oriented in the downward direction. The secondary (Y-axis)
radial measurement point should be in the direction of rotation and 90 degrees from
the primary.
Because of the tendency for screw compressors to generate high axial vibration when
subjected to changes in process conditions, the axial (Z-axis) measurement point is
essential. The ideal location for this point is on the outboard, or xed, bearing and ori-
ented toward the driver. Unfortunately, this is not always possible. The outboard bear-
ings are fully enclosed within the compressors housing and an axial measurement
cannot be obtained at these points. Therefore, the axial measurement must be
acquired from the oat, or inboard, bearings. While this position captures the axial
movement of the shaft, the recorded levels are lower than those acquired from the
xed bearings.
Electric Motors
Both radial (X- and Y-axis) measurements should be taken at the inboard and outboard
bearing housings. Orientation of the measurements is determined by the anticipated
induced load created by the driven units. The primary (X-axis) radial measurement
should be positioned in the same plane as the worst anticipated shaft displacement.
The secondary (Y-axis) radial should be positioned at 90 degrees in the direction of
rotation to the primary point and oriented to permit vector analysis of actual shaft dis-
placement.
Horizontal motors rely on a magnetic center generated by its electrical eld to posi-
tion the rotor in the axial (Z-axis) plane between the inboard and outboard bearings.
Therefore, most electric motors are designed with two oat bearings instead of the
normal conguration incorporating one oat and one xed bearing. Vertical motors
should have an axial (Z-axis) measurement point at the inboard bearing nearest the
coupling and oriented in an upward direction. This data point monitors the downward
axial force created by gravity or an abnormal load.
Electric motors are not designed to absorb side loads, such as those induced by V-belt
drives. In applications where V-belts or other radial loads are placed on the motor, the
primary radial transducer (X-axis) should be oriented opposite the direction of
induced load and the secondary radial (Y-axis) point should be positioned at 90
degrees in the direction of rotation. If, for safety reasons, the primary transducer can-
not be positioned opposite the induced load, the two radial transducers should be
placed at 45 degrees on either side of the load plane created by the side load.
Totally enclosed, fan-cooled, and explosion-proof motors present some difculty
when attempting to acquire data on the outboard bearing. By design, the outboard
bearing housing is not accessible. The optimum method of acquiring data is to perma-
nently mount a sensor on the outboard-bearing housing and run the wires to a conve-
nient data-acquisition location. If this is not possible, the X-Y data points should be as
close as possible to the bearing housing. Ensure that there is a direct mechanical path
08.Mobley.13 Page 118 Thursday, February 4, 1999 3:10 PM
118 VibrationFundamentals
to the outboard bearing. The use of this approach results in some loss of signal
strength from motor-mass damping. Do not obtain data from the fan housing.
Fans and Blowers
If a fan is V-belt driven, the primary measurement point should be in a plane opposing
the side load created by the belts on the inboard and outboard bearings. The second-
ary point should be at 90 degrees to the primary in the direction of rotation.
Bowed shafts caused by thermal and mechanical effects create severe problems on
large fans, especially overhung designs. Therefore, it is advantageous to acquire data
from all four quadrants of the outboard bearing housing on overhung fans to detect
this problem.
At least one axial measurement point should be located on each fan shaft. This is
especially important on fans that are V-belt driven. Axial data are helpful in identify-
ing and quantifying thrust (induced) loads created by the process and any potential
fan element problems such as imbalance, cracked blade, etc.
In applications where numerous fans are in proximity, an additional measurement
point on the base is useful for identifying structural resonance or cross-talk between
the fans.
Gearboxes
Gearbox measurement point orientation and location should be congured to allow
monitoring of the normal forces generated by the gear set. In most cases, the separat-
ing force, which tends to pull the gears apart, determines the primary radial measure-
ment point location. For example, a helical gear set generates a separating force that is
tangential to a centerline drawn through the pinion and bullgear shafts. The primary
(X-axis) radial measurement point should be oriented to monitor this force and a sec-
ondary (Y-axis) radial should be located at 90 degrees to the primary. The best loca-
tion for the secondary (Y-axis) radial is opposite the direction of rotation. In other
words, the secondary leads the primary transducers.
With the exception of helical gears, most gear sets should not generate axial or thrust
loads in normal operation. However, at least one axial (Z-axis) measurement point
should be placed on each of the gear shafts. The axial point should be located at the
xed, or thrust, bearing cap and oriented toward the gearbox.
In complex gearboxes, it may be difcult to obtain radial measurements from the
intermediate or idler shafts. In most cases, these intermediate shafts and their bearings
are well inside the gearbox. As a result, direct access to the bearings is not possible. In
these cases, the only option is to acquire axial (Z-axis) readings through the gearbox
housing. A review of the cross-sectional drawings allows the best location for these
08.Mobley.13 Page 119 Thursday, February 4, 1999 3:10 PM
119 VibrationDataAcquisition
Figure 13.3 Typical process-roll conguration and wrap-force vectors.
measurements to be determined. The key is to place the transducer at a point that will
provide the shortest, direct link to the intermediate shaft.
Process Rolls
Process rolls are widely used by industry. As with other machine components, two
radial (X- and Y-axis) and one axial (Z-axis) measurement should be acquired from
each roll. However, the orientation of these measurement points is even more critical
for process rolls than for some of the other machine components.
The loading on each roll is generated by the belt, wire mesh, and/or transported prod-
uct. The amount and distribution of the load varies depending on the wrap of the car-
ried load. Wrap refers to the angular distance around the roll that touches the belt,
wire mesh, or product. In most conveyor systems, the load is relatively uniform and is
in a downward direction. In this case, the traditional vertical, horizontal, and axial
mounting positions are acceptable.
Figure 13.3 represents a typical process-roll conguration. The arrows indicate the
force vectors generated by the wire, belt, or product wrap around these rolls. The left
roll has a force vector at 45 degrees down to the left; the right roll has a mirror image
force vector; and the bottom roll has a vertical vector.
The primary (X-axis) radial measurement for the bottom roll should be in the vertical
plane with the transducer mounted on top of the bearing cap. The secondary radial (Y-
axis) measurement should be in the horizontal plane facing upstream of the belt.
08.Mobley.13 Page 120 Thursday, February 4, 1999 3:10 PM
120 VibrationFundamentals
Since the belt carried by the roll also imparts a force vector in the direction of travel,
this secondary point should be opposite the direction of belt travel.
The ideal primary (X-axis) point for the top right roll is opposite the force vector. In
this instance, the primary radial measurement point should be located on the right of
the bearing cap facing upward at a 45-degree angle. Theoretically, the secondary (Y-
axis) radial point should be at 90 degrees to the primary on the bottom-left of the
bearing cap. However, it may be difcult, if not impossible, to locate and access a
measurement point here. Therefore, the next best location is at 45 degrees from the
anticipated force vector on the left of the bearing cap. This placement still provides
the means to calculate the actual force vector generated by the product.
Pumps
Appropriate measurement points vary by type of pump. In general, pumps can be
classied as centrifugal or positive displacement, and each of these can be divided
into groups.
CentrifugalPumps
The location of measurement points for centrifugal pumps depends on whether the
pump is classied as end suction or horizontal splitcase.
End Suction Pumps
Figure 13.4 illustrates a typical single-stage, end-suction centrifugal pump. The suc-
tion inlet is on the axial centerline, while the discharge may be either horizontal or
vertical. In the illustration, the actual discharge is horizontal and is anged in the ver-
tical downstream.
The actual discharge orientation determines the primary radial (X-axis) measurement
point. This point must be oriented in the same plane as the discharge and opposite the
direction of ow. In the illustration, the primary point should be in the horizontal
plane facing the discharge.
Restrictions or other causes of back-pressure in the discharge piping deect the shaft
in the opposite direction. Referring back to the illustration, the shaft would be
deected toward the front of the picture. If the discharge were vertical and in the
downward direction, the primary radial measurement point would be at the top of the
pumps bearing cap looking downward.
A second radial (Y-axis) measurement point should be positioned at 90 degrees to the
primary in a plane that captures secondary shaft deection. For the pump illustrated in
Figure 13.4, the secondary (Y-axis) radial measurement point is located on top of the
pumps bearing cap and oriented downward. Since the pump has a clockwise rotation,
back-pressure in the discharge piping forces the shaft both downward and horizon-
tally toward the front of the picture.
08.Mobley.13 Page 121 Thursday, February 4, 1999 3:10 PM
121 VibrationDataAcquisition
Figure 13.4 Typical end-suction, single-stage centrifugal pump.
Because this type of pump is susceptible to axial thrusting, an axial (Z-axis) measure-
ment point is essential. This point should be on the xed bearing housing oriented
toward the driver.
Horizontal Splitcase Pumps
The ow pattern through a horizontal splitcase pump is radically different than that
through an end-suction pump. Inlet and discharge ow are in the same plane and
almost directly opposite one another. This conguration, illustrated in Figure 13.5,
greatly improves the hydraulic-ow characteristics within the pump and improves its
ability to resist ow-induced instability.
The location of the primary (X-axis) radial measurement point for this type of pump is
in the horizontal plane and on the opposite side from the discharge. The secondary (Y-
axis) radial measurement point should be 90 degrees to the primary point and in the
direction of rotation. If the illustrated pump has a clockwise rotation, the measure-
ment point should be on top, oriented downward. For a counterclockwise rotation, it
should be on bottom, oriented upward.
Single-stage pumps generate some axial thrusting due to imbalance between the dis-
charge and inlet pressures. The impeller design provides a means of balancing these
forces, but it cannot absolutely compensate for the difference in the pressures. As a
result, there will be some axial rotor movement. In double volute, or multistage,
08.Mobley.13 Page 122 Thursday, February 4, 1999 3:10 PM
122 VibrationFundamentals
Figure 13.5 Typical horizontal splitcase pump.
pumps, two impellers are positioned back to back. This conguration eliminates most
of the axial thrusting when the pump is operating normally.
An axial (Z-axis) measurement point should be located on the xed bearing housing.
It should be oriented toward the driver to capture any instability that may exist.
Multistage Pumps
Multistage pumps may be either end-suction or horizontal splitcase pumps. They have
two basic impeller congurations, in-line or opposed, as shown in Figure 13.6. In-line
impellers generate high thrust loads.
The impeller conguration does not alter the radial measurement locations discussed
in the preceding sections. However, it increases the importance of the axial (Z-axis)
measurement point. The in-line conguration drastically increases the axial loading
on the rotating element and, therefore, the axial (Z-axis) measurement point is critical.
Obviously, this point must be in a location that detects axial movement of the shaft.
However, since large, heavy-duty xed bearings are used to withstand the high thrust
loading generated by this design, direct measurement is difcult. A cross-sectional
08.Mobley.13 Page 123 Thursday, February 4, 1999 3:10 PM
123 VibrationDataAcquisition
Figure 13.6 In-line and opposed impellers on multistage pumps.
drawing of the pump may be required to locate a suitable location for this measure-
ment point.
Positive-DisplacementPumps
Positive-displacement pumps can be divided into two major types: rotary and recipro-
cating. All rotary pumps use some form of rotating element, such as gears, vanes, or
lobes to increase the discharge pressure. Reciprocating pumps use pistons or wobble
plates to increase the pressure.
Rotary Pumps
Locations of measurement points for rotary positive-displacement pumps should be
based on the same logic as in-line centrifugal pumps. The primary (X-axis) radial
measurement should be taken in the plane opposite the discharge port. The secondary
08.Mobley.13 Page 124 Thursday, February 4, 1999 3:10 PM
124 VibrationFundamentals
(Y-axis) radial should be at 90 degrees to the primary and in the direction of the rotors
rotation.
Since most rotary positive-displacement pumps have inlet and outlet ports in the same
plane and opposed, there should be relatively little axial thrusting. However, an axial
(Z-axis) measurement should be acquired from the xed bearing, oriented toward the
driver.
Reciprocating Pumps
Reciprocating pumps are more difcult to monitor because of the combined rotational
and linear motions that are required to increase the discharge pressure. Measurement
point location and orientation should be based on the same logic as that of reciprocat-
ing compressors.
09.Mobley.14 Page 125 Thursday, February 4, 1999 3:21 PM
Chapter14
TRENDINGANALYSIS
Long-term vibration trends are a useful diagnostic tool. Trending techniques involve
graphically comparing the total energy, which is the sum of the frequency compo-
nents amplitude over some consistent, user-selected frequency range (i.e., F
MIN
to
F
MAX
), over a long period of time to get a historical perspective of the vibration pat-
tern. Plots of this sum against time (e.g., days) provide a means of quantifying the rel-
ative condition of the monitored machine (see Figure 14.1). Most predictive
maintenance systems provide automatic-trending capabilities for recorded data. This
is not to be confused with time-domain plots, which are instantaneous measures of
total vibration amplitude plotted against time measured in seconds.
Figure 14.1 Trend data are plotted versus time and provide historical trends.
125
09.Mobley.14 Page 126 Thursday, February 4, 1999 3:21 PM
126 VibrationFundamentals
Figure 14.2 Typical broadband measurements.
Used properly, this feature greatly enhances a predictive maintenance program. The
real value of trending techniques is that they provide the capability of automatically
scanning large amounts of data (both broadband and narrowband) and reporting any
change in preselected values.
TYPES OF TRENDS
The three primary categories of trends are broadband, narrowband, and combinations
of the two.
Broadband Trends
Most microprocessor-based vibration-monitoring systems acquire and record a l-
tered broadband energy level for each data point included in the program. The band-
width of the energy band is determined by the minimum, F
MIN
, and maximum, F
MAX
,
frequencies that were established as part of the database setup. In most applications,
the minimum frequency should be zero, but the maximum varies, depending on the
specic machine-train. Figure 14.2 illustrates typical broadband data.
09.Mobley.14 Page 127 Thursday, February 4, 1999 3:21 PM
127 TrendingAnalysis
POINT SHAFT INSTABILITY VANEPASS BEARING DEFECTS
V01 0.10 0.09 0.001
V02 0.10 0.09 0.002
V03 0.13 0.10 0.002
V04 0.11 0.09 0.010
Figure 14.3 Narrowband data.
Broadband data cannot be used to identify specic machine components (e.g., bear-
ing, gears) or failure modes (e.g., imbalance, misalignment). The data acquired using
broadband lters are limited to the total energy value contained within the user-
selected frequency window or bandwidth, F
MIN
to F
MAX
.
At best, broadband energy provides a gross approximation of the machines condition
and its relative rate of degradation. Since the only available data are overall energy
values, broadband data do not provide enough detail to permit diagnosis of machine-
train condition. Without discrete identication of the specic frequencies that make
up the overall energy, the failure mode or failing component cannot be determined.
Narrowband Trends
Like broadband data, narrowband data also reect the total energy, but it reects a
more restricted user-selected range or window. Narrowband monitoring generally is
used to trend and evaluate one selected machine-train component rather than several.
Filtered narrowband windows are typically set up around the unique frequency com-
ponents generated by specic machine-train components so that the energy in each
ltered window can be directly attributed to that specic machine component. How-
ever, even though narrowband analysis improves the diagnostic capabilities of a pre-
dictive maintenance program, it is not possible to isolate and identify specic failure
modes within a machine-train. Figure 14.3 illustrates the added information provided
by narrowband data.
09.Mobley.14 Page 128 Thursday, February 4, 1999 3:21 PM
128 VibrationFundamentals
Figure 14.4 Narrowband trends provide energy histories of specic components.
Figure 14.4 illustrates narrowband data trends. In addition to the overall or broadband
energy, narrowband trends indicate the relative energy in select machine-train compo-
nents. In effect, this type of analysis is a series of mini-overall energy readings.
Composite Trends
Most microprocessor-based systems permit composite trending (i.e., simultaneous
displays) of both ltered broadband and narrowband data. Figure 14.5 illustrates a
composite trend that includes both broadband and narrowband data. This type of plot
is quite benecial because it permits the analyst to track the key indicators of machine
condition on one plot.
EVALUATION METHODS
Trend data can be used in the following ways: (1) to compare with specic reference
values, (2) mode-shape comparisons, and (3) cross-machine comparisons.
Comparison to Reference Values
Three types of reference-value comparisons are used to evaluate trend data: baseline
data, rate of change, and industrial standards.
BaselineData
A series of baseline or reference data sets should be taken for each machine-train
included in a predictive maintenance program. These data sets are necessary for
future use as a reference point for trends, time traces, and FFT signatures that are col-
09.Mobley.14 Page 129 Thursday, February 4, 1999 3:21 PM
129 TrendingAnalysis
Figure 14.5 Combined trends (i.e., composite) provide both broadband and narrowband
data.
lected over time. Such baseline data sets must be representative of the normal operat-
ing condition of each machine-train to have value as a reference.
Three criteria are critical to the proper use of baseline comparisons: reset after main-
tenance, proper identication, and process envelope.
Reset After Maintenance
The baseline data set must be updated each time the machine is repaired, rebuilt, or
when any major maintenance is performed. Even when best practices are used,
machinery cannot be restored to as-new condition when major maintenance is per-
formed. Therefore, a new baseline or reference data set must be obtained following
these events.
Proper Identication
Each reference or baseline data set must be clearly and completely identied. Most
vibration-monitoring systems permit the addition of a label or unique identier to any
user-selected data set. This capability should be used to clearly identify each baseline
data set.
09.Mobley.14 Page 130 Thursday, February 4, 1999 3:21 PM
130 VibrationFundamentals
In addition, the data set label should include all dening information. For example,
any rework or repairs made to the machine should be identied. If a new baseline data
set is obtained after the replacement of a rotating element, this information should be
included in the descriptive label.
Process Envelope
Since variations in process variables, such as load, have a direct effect on the vibration
energy and signature generated by a machine-train, the actual operating envelope for
each baseline data set must also be clearly identied. If this step is omitted, direct
comparison of other data to the baseline will be meaningless. The label feature in
most vibration monitoring systems permits tagging the baseline data set with this
additional information.
RateofChange
Rate of change is the most often used trend analysis. Because most of the micropro-
cessor-based systems provide the ability to automatically display both broadband and
narrowband data trends, analysts tend to rely on this means of comparative analysis.
Rate of change is a valid means of dening the relative condition of rotating machin-
ery. As a general rule, there must be a change in mechanical condition before there
can be a change in the vibration energy generated by a machine. Therefore, monitor-
ing the rate that the energy levels change, either up or down, is a useful tool (see Fig-
ure 14.6). Caution: Broadband and narrowband data must be normalized for changes
in load before being valid. Normal variations in machine load destroy the validity of
non-normalized trend data and little can be gained from its use.
IndustrialStandards
There are a number of published standards that dene acceptable levels of vibration in
machinery. These standards are valuable reference tools, but they must be clearly
understood and properly used. Industrial-standard data can be obtained from a Rath-
bone chart and from the American Petroleum Institute.
Rathbone Vibration-Severity Chart
The Rathbone chart provides levels of vibration severity that range from extremely
smooth, which is the best possible operating condition, to very rough or absolute-fault
limit, which is the maximum level where a machine can operate (see Figure 14.7).
This chart is useful, although it is often misused. Four factors must be understood
before using the chart:
1. Data included in the Rathbone chart are valid for machines with running
speeds between 1200 and 3600 rpm. The chart cannot be used for low-
speed or turbomachinery.
09.Mobley.14 Page 131 Thursday, February 4, 1999 3:21 PM
131 TrendingAnalysis
Figure 14.6 Rate of change of trend data indicates condition change.
2. The data presented in the chart are relative vibration levels (i.e., taken from
a bearing pedestal using either an accelerometer or velocity probe) in
inches per second (in./sec) peak.
3. Data are peak values of velocity (in./sec) for a ltered broadband from 10
to 10,000 Hz.
4. The severity levels are relative, not absolute. For example, when a machine
reaches the absolute-fault limit, it has a 90% probability of failure within
its next 1000 hours of operation (i.e., it is not going to fail tomorrow).
09.Mobley.14 Page 132 Thursday, February 4, 1999 3:21 PM
132 VibrationFundamentals
Figure 14.7 Rathbone chart-relative vibration severity.
American Petroleum Institute Standards
The American Petroleum Institute (API) has established standards for vibration lev-
els. Unlike the Rathbone chart, which presents relative vibration data, the API stan-
dards are actual shaft displacement as measured with a displacement probe. Based on
the API data, acceptable vibration can be dened by:
12,000
Vibration Severity = ---------------- mils, peak-to-peak
rpm
The API equivalent of the absolute-fault limit can be dened by:
09.Mobley.14 Page 133 Thursday, February 4, 1999 3:21 PM
133 TrendingAnalysis
12,000
Absolute-fault Limit = 1.3 ---------------- mils, peak-to-peak
rpm
The API standards are reasonable for turbomachinery, but are unacceptable for lower
speed machines. The standards are applicable for rotating machinery with speeds
above 1800 rpm; marginal for speeds between 600 and 1800 rpm; and should not be
used for speeds below 600 rpm.
Mode Shape (Shaft Deection)
A clear understanding of the mode shape, or shaft deection, of a machines rotating
element is a valuable diagnostic tool. Both broadband and narrowband ltered energy
windows can be used at each measurement point and orientation across the machine.
The resultant plots, one in the vertical plane and one in the horizontal plane, provide an
approximation of the mode shape of the complete machine and its rotating element.
Unfortunately, these plots must be developed manually. The microprocessor-based
systems generally do not automate this function, but they are easily constructed on
graph paper. The following steps are used to construct such a plot:
1. The rst step is to draw two horizontal lines on the graph paper. One is
used to plot the vertical data and the other to plot the horizontal. These two
lines show the location of each measurement point in inches with the out-
board motor bearing being at zero.
2. Next, draw vertical lines that intersect the left-hand end of the two hori-
zontal lines. These vertical lines form the amplitude scale for the two plots.
Establish the amplitude scale based on the maximum energy level recorded
in the broadband or narrowband windows.
3. The nal step is to plot the actual measured amplitude at each measure-
ment point on the machine-train. Start with the outboard motor bearing
and move across the machine until the nal data set is plotted.
BroadbandPlots
The overall energy from the ltered broadband plotted against measurement location
provides an approximation of the mode shape of the installed machine. Figure 14.8
illustrates a vertical broadband plot taken from a Spencer blower. Note that the motor
appears to be exing in the vertical direction. Extremely high amplitudes are present
in the motors outboard bearing and the amplitudes decrease at subsequent measure-
ment points across the machine.
A mode curve exhibiting the shape shown in Figure 14.8 could indicate that the motor
mountings, or the baseplate under the motor, are loose and that the motor is moving
vertically. In fact, in the example from which this gure was taken, this is exactly
what was happening. The blowers baseplate oats on a 1-in. thick cork pad, which
09.Mobley.14 Page 134 Thursday, February 4, 1999 3:21 PM
134 VibrationFundamentals
Figure 14.8 Vertical broadband mode shape for Spencer blower indicates potential failure.
is normally an acceptable practice. However, in this example, an inlet lter/silencer
was mounted without support directly to the inlet located on the right end of the
machine. The weight of the lter/silencer compressed the cork pad under the blower,
which lifted the motor-end of the baseplate off of the cork pad. In this mode, the
motor has complete freedom of movement in the vertical plane. In effect, it acts like a
tuning fork and creates the high overall energy recorded on the mode plot.
NarrowbandPlots
Narrowband plots permit the same type of evaluation for major vibration components
such as fundamental running speed (1) or gear mesh. The plots are constructed in the
same way as for the broadbands, except that the amplitude values are for user-selected
windows, or bands.
Using the previously mentioned example of the Spencer blower, Figure 14.9 is a plot
of the fundamental (1) frequency of the motor-blower shaft versus measurement
location. Note that the vertical mode (see Figure 14.9) appears to be relatively normal,
09.Mobley.14 Page 135 Thursday, February 4, 1999 3:21 PM
135 TrendingAnalysis
Figure 14.9 Horizontal narrowband (1) mode shape indicates shaft deection.
except for the motor looseness problem. The horizontal plot seems to indicate that the
shaft is being severely deected from its true centerline. In addition, the plot suggests
that the deection is outboard (i.e., toward the rotor) from the two blower-support
bearings. This outboard deection eliminates misalignment between the motor and
blower as a possible source of the deection. We must now determine what could
cause the rotor to be deected and why only in the horizontal direction.
The Spencer blower in this example provides air to a drying process in a metal-coating
line. Its conguration includes an end-suction inlet that is in line with the shaft and a
horizontal discharge that is perpendicular to the shaft. In this particular example, the
source of the shaft deection observed in the mode plot is aerodynamic instability.
The reason for this instability is that the blowers are incorrectly sized for the applica-
tion and are running well outside their performance curve. In effect, the blowers have
no back-pressure and are operating in a runout condition. The result of operating in
09.Mobley.14 Page 136 Thursday, February 4, 1999 3:21 PM
136 VibrationFundamentals
this condition is that the design load intended to stabilize the rotor is no longer
present. This causes the shaft to deect or ex, generating the high amplitudes
observed in the horizontal mode plot.
The problem is eliminated by restricting the discharge air ow from the blowers. By
increasing the back-pressure, the blowers are able to operate within their normal enve-
lope and the shaft deection disappears.
Cross-Machine Comparisons
Cross-machine comparison is an extremely benecial tool to the novice analyst. Most
vibration-monitoring systems permit direct comparison of vibration data, both ltered
window energy and complete signatures, acquired from two machines. This capability
permits the analyst to compare a machine that is known to be in good operating condi-
tion directly with one that is perceived to have a problem. There are several ways that
cross-machine comparisons can be made using microprocessor-based systems: multi-
ple plots, ratio, and difference.
LIMITATIONS
Although quite valuable when used properly, trends do not allow the analyst to con-
rm that a problem exists or to determine the cause of incipient problems. Another
limitation is the limited number of values the system can handle. Further, the data
need to be normalized for speed, load, and process variables.
Number of Values
Some of the vibration-monitoring systems limit the number of data sets and duration
of data that can be automatically trended. In most systems, the number of values that
can be trended is limited to 8 to 12 data sets. Although this limitation prevents trend-
ing the machine over its useful life, it does not eliminate trending as a vibration-mon-
itoring tool.
Data Normalization
Trend data that are not properly normalized for speed, load, and process variables are
of little value. Because load and process-variable normalization require a little more
time during the data-acquisition process, many programs do not perform these adjust-
ments. If this is the case, it is best to discontinue the use of trends altogether.
As an example, Figure 14.10 illustrates the impact of load on vibration trends. The
solid line represents the recorded raw broadband vibration levels. The dashed line is
the same data adjusted for changes in load.
09.Mobley.14 Page 137 Thursday, February 4, 1999 3:21 PM
137 TrendingAnalysis
Figure 14.10 Trends must be adjusted or normalized for load changes.
Figure 14.11 Baselines must be reset following repair.
Trend data also must be adjusted for maintenance and repair activities. Figure 14.11
illustrates an average trend curve that indicates a sharp rise in vibration levels. It also
reects that, after repair, the levels drop radically. At this point, all baseline and refer-
ence values should be reset. If this does not occur, the automatic trending capabilities
of the computer-based system do not function properly.
10.Mobley.15 Page 138 Friday, February 5, 1999 10:38 AM
Chapter 15
FAILURE-MODE ANALYSIS
All of the analysis techniques discussed to this point have been methods to determine
if a potential problem exists within the machine-train or its associated systems. Fail-
ure-mode analysis is the next step required to pinpoint specically the failure mode
and identify which machine-train component is degrading.
Although failure-mode analysis identies the number and symptoms of machine-train
problems, it does not always identify the true root cause of problems. Root cause must
be veried by visual inspection, additional testing, or other techniques such as operat-
ing dynamics analysis.
Failure-mode analysis is based on the assumption that certain failure modes are com-
mon to all machine-trains and all applications. It also assumes that the vibration pat-
terns for each of these failure modes, when adjusted for process-system dynamics, are
absolute and identiable.
Two types of information are required to perform failure-mode analysis: (1)
machine-train vibration signatures, both FFTs and time traces; and (2) practical
knowledge of machine dynamics and failure modes. A number of failure-mode
charts are available that describe the symptoms or abnormal vibration proles that
indicate potential problems exist. An example is the following description of the
imbalance failure mode, which was obtained from a failure-mode chart: Single-
plane imbalance generates a dominant fundamental (1) frequency component with
no harmonics (2, 3, etc.). Note, however, that the failure-mode charts are simplis-
tic since many other machine-train problems also excite, or increase the amplitude
of, the fundamental (1) frequency component. In a normal vibration signature, 60
to 70% of the total overall, or broadband, energy is contained in the 1 frequency
component. Any deviation from a state of equilibrium increases the energy level at
this fundamental shaft speed.
138
10.Mobley.15 Page 139 Friday, February 5, 1999 10:38 AM
139 Failure-Mode Analysis
COMMON GENERAL FAILURE MODES
Many of the common causes of failure in machinery components can be identied by
understanding their relationship to the true running speed of the shaft within the
machine-train.
Table 15.1 is a vibration troubleshooting chart that identies some of the common
failure modes. This table provides general guidelines for interpreting the most com-
mon abnormal vibration proles. These guidelines, however, do not provide positive
verication or identication of machine-train problems. Verication requires an
understanding of the failure mode and how it appears in the vibration signature.
The sections that follow describe the most common machine-train failure modes: crit-
ical speeds, imbalance, mechanical looseness, misalignment, modulations, process
instability, and resonance.
Critical Speeds
All machine-trains have one or more critical speeds that can cause severe vibration
and damage to the machine. Critical speeds result from the phenomenon known as
dynamic resonance.
Critical speed is a function of the natural frequency of dynamic components such as a
rotor assembly, bearings, etc. All dynamic components have one or more natural fre-
quencies that can be excited by an energy source that coincides with, or is in proxim-
ity to, that frequency. For example, a rotor assembly with a natural frequency of 1800
rpm cannot be rotated at speeds between 1782 and 1818 rpm without exciting the
rotors natural frequency.
Critical speed should not be confused with the mode shape of a rotating shaft. Deec-
tion of the shaft from its true centerline (i.e., mode shape) elevates the vibration ampli-
tude and generates dominant vibration frequencies at the rotors fundamental and
harmonics of the running speed. However, the amplitude of these frequency compo-
nents tends to be much lower than those caused by operating at a critical speed of the
rotor assembly. Also, the excessive vibration amplitude generated by operating at a crit-
ical speed disappears when the speed is changed. Vibrations caused by mode shape tend
to remain through a much wider speed range or may even be independent of speed.
The unique natural frequencies of dynamic machine components are determined by
the mass, freedom of movement, support stiffness, and other factors. These factors
dene the response characteristics of the rotor assembly (i.e., rotor dynamics) at vari-
ous operating conditions.
Each critical speed has a well-dened vibration pattern. The rst critical excites the
fundamental (1) frequency component; the second critical excites the secondary (2)
component; and the third critical excites the third (3) frequency component.
10.Mobley.15 Page 140 Friday, February 5, 1999 10:38 AM
140 Vibration Fundamentals
Table 15.1 Vibration Troubleshooting Chart
Source: Predictive Maintenance for Process Machinery, R. Keith Mobley, Technology for Energy Corp., 1988.
10.Mobley.15 Page 141 Friday, February 5, 1999 10:38 AM
141 Failure-Mode Analysis
The best way to conrm a critical-speed problem is to change the operating speed of
the machine-train. If the machine is operating at a critical speed, the amplitude of the
vibration components (1, 2, or 3) will immediately drop when the speed is
changed. If the amplitude remains relatively constant when the speed is changed, the
problem is not critical speed.
Imbalance
The term balance means that all forces generated by, or acting on, the rotating ele-
ment of a machine-train are in a state of equilibrium. Any change in this state of equi-
librium creates an imbalance. In the global sense, imbalance is one of the most
common abnormal vibration proles exhibited by all process machinery.
Theoretically, a perfectly balanced machine that has no friction in the bearings would
experience no vibration and would have a perfect vibration prolea perfectly at,
horizontal line. However, there are no perfectly balanced machines in existence. All
machine-trains exhibit some level of imbalance, which has a dominant frequency
component at the fundamental running speed (1) of each shaft.
An imbalance prole can be excited due to the combined factors of mechanical imbal-
ance, lift/gravity differential effects, aerodynamic and hydraulic instabilities, process
loading, and, in fact, all failure modes.
Mechanical Imbalance
It is incorrect to assume that mechanical imbalance must exist to create an imbalance
condition within the machine. Mechanical imbalance, however, is the only form of
imbalance that is corrected by balancing the rotating element. When all failures are
considered, the number of machine problems that are the result of actual mechanical
rotor imbalance is relatively small.
Single-Plane Mechanical Imbalance
Single-plane mechanical imbalance excites the fundamental (1) frequency compo-
nent, which is typically the dominant amplitude in a signature. Because there is
only one point of imbalance, only one high spot occurs as the rotor completes each
revolution. The vibration signature also may contain lower level frequencies reect-
ing bearing defects and passing frequencies. Figure 15.1 illustrates single-plane
imbalance.
Because mechanical imbalance is multidirectional, it appears in both the vertical and
horizontal directions at the machines bearing pedestals. The actual amplitude of the
1 component generally is not identical in the vertical and horizontal directions and
both generally contain elevated vibration levels at 1.
10.Mobley.15 Page 142 Friday, February 5, 1999 10:38 AM
142 Vibration Fundamentals
Figure 15.1 Single-plane imbalance.
The difference between the vertical and horizontal values is a function of the bearing-
pedestal stiffness. In most cases, the horizontal plane has a greater freedom of move-
ment and, therefore, contains higher amplitudes at 1 than the vertical plane.
Multiplane Mechanical Imbalance
Multiplane mechanical imbalance generates multiple harmonics of running speed.
The actual number of harmonics depends on the number of imbalance points, the
severity of imbalance, and the phase angle between imbalance points.
Figure 15.2 illustrates a case of multiplane imbalance in which there are four out-of-
phase imbalance points. The resultant vibration prole contains dominant frequencies
at 1, 2, 3, and 4. The actual amplitude of each of these components is deter-
mined by the amount of imbalance at each of the four points, but the 1 component
should always be higher than any subsequent harmonics.
Lift/Gravity Differential Imbalance
Lift, which is designed into a machine-trains rotating elements to compensate for the
effects of gravity acting on the rotor, is another source of imbalance. Because lift does
not always equal gravity, there is always some imbalance in machine-trains. The
vibration component due to the lift/gravity differential effect appears at the funda-
mental or 1 frequency.
10.Mobley.15 Page 143 Friday, February 5, 1999 10:38 AM
143 Failure-Mode Analysis
Figure 15.2 Multiplane imbalance generates multiple harmonics.
Other
In fact, all failure modes create some form of imbalance in a machine, as do aerody-
namic instability, hydraulic instability, and process loading. The process loading of
most machine-trains varies, at least slightly, during normal operations. These vibra-
tion components appear at the 1 frequency.
Mechanical Looseness
Looseness, which can be present in both the vertical and horizontal planes, can create
a variety of patterns in a vibration signature. In some cases, the fundamental (1) fre-
quency is excited. In others, a frequency component at one-half multiples of the
shafts running speed (0.5, 1.5, 2.5, etc.) is present. In almost all cases, there are
multiple harmonics, both full and half.
Vertical Looseness
Mechanical looseness in the vertical plane generates a series of harmonic and half-
harmonic frequency components. Figure 15.3 is a simple example of a vertical
mechanical looseness signature.
10.Mobley.15 Page 144 Friday, February 5, 1999 10:38 AM
144 Vibration Fundamentals
Figure 15.3 Vertical mechanical looseness has a unique vibration prole.
In most cases, the half-harmonic components are about one-half of the amplitude of
the harmonic components. They result from the machine-train lifting until stopped by
the bolts. The impact as the machine reaches the upper limit of travel generates a fre-
quency component at one-half multiples (i.e., orders) of running speed. As the
machine returns to the bottom of its movement, its original position, a larger impact
occurs that generates the full harmonics of running speed.
The difference in amplitude between the full harmonics and half-harmonics is caused
by the effects of gravity. As the machine lifts to its limit of travel, gravity resists the
lifting force. Therefore, the impact force that is generated as the machine foot con-
tacts the mounting bolt is the difference between the lifting force and gravity. As the
machine drops, the force of gravity combines with the force generated by imbalance.
The impact force as the machine foot contacts the foundation is the sum of the force
of gravity and the force resulting from imbalance.
Horizontal Looseness
Figure 15.4 illustrates horizontal mechanical looseness, which is also common to
machine-trains. In this example, the machines support legs ex in the horizontal
plane. Unlike the vertical looseness illustrated in Figure 15.3, gravity is uniform at
each leg and there is no increased impact energy as the legs direction is reversed.
10.Mobley.15 Page 145 Friday, February 5, 1999 10:38 AM
145 Failure-Mode Analysis
Figure 15.4 Horizontal looseness creates rst and second harmonics.
Horizontal mechanical looseness generates a combination of rst (1) and second
(2) harmonic vibrations. Since the energy source is the machines rotating shaft, the
timing of the ex is equal to one complete revolution of the shaft, or 1. During this
single rotation, the mounting legs ex to their maximum deection on both sides of
neutral. The double change in direction as the leg rst deects to one side then the
other generates a frequency at two times (2) the shafts rotating speed.
Other
There are a multitude of other forms of mechanical looseness (besides vertical and
horizontal movement of machine legs) that are typical for manufacturing and process
machinery. Most forms of pure mechanical looseness result in an increase in the
vibration amplitude at the fundamental (1) shaft speed. In addition, looseness gener-
ates one or more harmonics (i.e., 2, 3, 4, or combinations of harmonics and half-
harmonics).
However, not all looseness generates this classic prole. For example, excessive bear-
ing and gear clearances do not generate multiple harmonics. In these cases, the vibra-
tion prole contains unique frequencies that indicate looseness, but the prole varies
depending on the nature and severity of the problem.
10.Mobley.15 Page 146 Friday, February 5, 1999 10:38 AM
146 Vibration Fundamentals
With sleeve or Babbitt bearings, looseness is displayed as an increase in subharmonic
frequencies (i.e., less than the actual shaft speed, such as 0.5). Rolling-element bear-
ings display elevated frequencies at one or more of their rotational frequencies.
Excessive gear clearance increases the amplitude at the gear-mesh frequency and its
sidebands.
Other forms of mechanical looseness increase the noise oor across the entire band-
width of the vibration signature. While the signature does not contain a distinct peak
or series of peaks, the overall energy contained in the vibration signature is increased.
Unfortunately, the increase in noise oor cannot always be used to detect mechanical
looseness. Some vibration instruments lack sufcient dynamic range to detect
changes in the signatures noise oor.
Misalignment
This condition is virtually always present in machine-trains. Generally, we assume
that misalignment exists between shafts that are connected by a coupling, V-belt, or
other intermediate drive. However, it can exist between bearings of a solid shaft and at
other points within the machine.
How misalignment appears in the vibration signature depends on the type of mis-
alignment. Figure 15.5 illustrates three types of misalignment (i.e., internal, offset,
and angular). These three types excite the fundamental (1) frequency component
because they create an apparent imbalance condition in the machine.
Internal (i.e., bearing) and offset misalignment also excite the second (2) harmonic fre-
quency. Two high spots are created by the shaft as it turns though one complete revolu-
tion. These two high spots create the rst (1) and second harmonic (2) components.
Angular misalignment can take several signature forms and excites the fundamental
(1) and secondary (2) components. It can excite the third (3) harmonic frequency
depending on the actual phase relationship of the angular misalignment. It also cre-
ates a strong axial vibration.
Modulations
Modulations are frequency components that appear in a vibration signature, but can-
not be attributed to any specic physical cause, or forcing function. Although these
frequencies are, in fact, ghosts or articial frequencies, they can result in signicant
damage to a machine-train. The presence of ghosts in a vibration signature often leads
to misinterpretation of the data.
Ghosts are caused when two or more frequency components couple, or merge, to
form another discrete frequency component in the vibration signature. This generally
occurs with multiple-speed machines or a group of single-speed machines.
10.Mobley.15 Page 147 Friday, February 5, 1999 10:38 AM
147 Failure-Mode Analysis
Figure 15.5 Three types of misalignment.
Note that the presence of modulation, or ghost peaks, is not an absolute indication of
a problem within the machine-train. Couple effects may simply increase the ampli-
tude of the fundamental running speed and do little damage to the machine-train.
However, this increased amplitude will amplify any defects within the machine-train.
Coupling can have an additive effect on the modulation frequencies, as well as being
reected as a differential or multiplicative effect. These concepts are discussed in the
sections that follow.
Take as an example the case of a 10-tooth pinion gear turning at 10 rpm while driving
a 20-tooth bullgear having an output speed of 5 rpm. This gear set generates real fre-
quencies at 5, 10, and 100 rpm (i.e., 10 teeth 10 rpm). This same set also can gener-
ate a series of frequencies (i.e., sum and product modulations) at 15 rpm (i.e., 10 rpm
+ 5 rpm) and 150 rpm (i.e., 15 rpm 10 teeth). In this example, the 10-rpm input
speed coupled with the 5-rpm output speed to create ghost frequencies driven by this
articial fundamental speed (15 rpm).
Sum Modulation
This type of modulation, which is described in the preceding example, generates a
series of frequencies that includes the fundamental shaft speeds, both input and out-
put, and fundamental gear-mesh prole. The only difference between the real fre-
quencies and the ghost is their location on the frequency scale. Instead of being at the
10.Mobley.15 Page 148 Friday, February 5, 1999 10:38 AM
148 Vibration Fundamentals
Figure 15.6 Sum modulation for a speed-increaser gearbox.
actual shaft-speed frequency, the ghost appears at frequencies equal to the sum of the
input and output shaft speeds. Figure 15.6 illustrates this for a speed-increaser gear-
box.
Difference Modulation
In this case, the resultant ghost, or modulation, frequencies are generated by the dif-
ference between two or more speeds (see Figure 15.7). If we use the same example as
before, the resultant ghost frequencies appear at 5 rpm (i.e., 10 rpm 5 rpm) and 50
rpm (i.e., 5 rpm 10 teeth). Note that the 5-rpm couple frequency coincides with the
real output speed of 5 rpm. This results in a dramatic increase in the amplitude of one
real running-speed component and the addition of a false gear-mesh peak.
This type of coupling effect is common in single-reduction/increase gearboxes or
other machine-train components where multiple running or rotational speeds are rela-
tively close together or even integer multiples of one another. It is more destructive
than other forms of coupling in that it coincides with real vibration components and
tends to amplify any defects within the machine-train.
Product Modulation
With product modulation, the two speeds couple in a multiplicative manner to create a
set of articial frequency components (see Figure 15.8). In the previous example,
product modulations occur at 50 rpm (i.e., 10 rpm 5 rpm) and 500 rpm (i.e., 50 rpm
10 teeth).
10.Mobley.15 Page 149 Friday, February 5, 1999 10:38 AM
149 Failure-Mode Analysis
Figure 15.7 Difference modulation for a speed-increaser gearbox.
Figure 15.8 Product modulation for a speed-increaser gearbox.
10.Mobley.15 Page 150 Friday, February 5, 1999 10:38 AM
150 Vibration Fundamentals
Beware that this type of coupling often may go undetected in a normal vibration anal-
ysis. Since the ghost frequencies are relatively high compared to the expected real fre-
quencies, they are often outside the monitored frequency range used for data
acquisition and analysis.
Process Instability
Normally associated with bladed or vaned machinery such as fans and pumps, process
instability creates an imbalanced condition within the machine. In most cases, it
excites the fundamental (l) and blade-pass/vane-pass frequency components. Unlike
true mechanical imbalance, the blade-pass and vane-pass frequency components are
broader and have more energy in the form of sideband frequencies.
In most cases, this failure mode also excites the third (3) harmonic frequency and
creates strong axial vibration. Depending on the severity of the instability and the
design of the machine, process instability also can create a variety of shaft-mode
shapes. In turn, this excites the 1, 2, and 3 radial vibration components.
Resonance
Resonance is dened as a large-amplitude vibration caused by a small periodic stimu-
lus having the same, or nearly the same, period as the systems natural vibration. In
other words, an energy source with the same, or nearly the same, frequency as the nat-
ural frequency of a machine-train or structure will excite that natural frequency. The
result is a substantial increase in the amplitude of the natural frequency component.
The key point to remember is that a very low amplitude energy source can cause mas-
sive amplitudes when its frequency coincides with the natural frequency of a machine
or structure. Higher levels of input energy can cause catastrophic, near instantaneous
failure of the machine or structure.
Every machine-train has one or more natural frequencies. If one of these frequencies
is excited by some component of the normal operation of the system, the machine
structure will amplify the energy, which can cause severe damage.
An example of resonance is a tuning fork. If you activate a tuning fork by striking it
sharply, the fork vibrates rapidly. As long as it is held suspended, the vibration decays
with time. However, if you place it on a desk top, the fork could potentially excite the
natural frequency of the desk, which would dramatically amplify the vibration energy.
The same thing can occur if one or more of the running speeds of a machine excites
the natural frequency of the machine or its support structure. Resonance is a very
destructive vibration and, in most cases, it will cause major damage to the machine or
support structure.
10.Mobley.15 Page 151 Friday, February 5, 1999 10:38 AM
151 Failure-Mode Analysis
Figure 15.9 Resonance response.
There are two major classications of resonance found in most manufacturing and
process plants: static and dynamic. Both types exhibit a broad-based, high-amplitude
frequency component when viewed in a FFT vibration signature. Unlike meshing or
passing frequencies, the resonance frequency component does not have modulations
or sidebands. Instead, resonance is displayed as a single, clearly dened peak.
As illustrated in Figure 15.9, a resonance peak represents a large amount of energy.
This energy is the result of both the amplitude of the peak and the broad area under
the peak. This combination of high peak amplitude and broad-based energy content is
typical of most resonance problems. The damping system associated with a resonance
frequency is indicated by the sharpness or width of the response curve,
n
, when mea-
sured at the half-power point. R
MAX
is the maximum resonance and R
MAX
/ 2 is the
half-power point for a typical resonance-response curve.
Static Resonance
When the natural frequency of a stationary, or nondynamic, structure is energized, it
will resonate. This type of resonance is classied as static resonance and is considered
to be a nondynamic phenomenon. Nondynamic structures in a machine-train include
casings, bearing-support pedestals, and structural members such as beams, piping, etc.
Since static resonance is a nondynamic phenomenon, it is generally not associated
with the primary running speed of any associated machinery. Rather, the source of
static resonance can be any energy source that coincides with the natural frequency of
any stationary component. For example, an I-beam support on a continuous annealing
line may be energized by the running speed of a roll. However, it also can be made to
10.Mobley.15 Page 152 Friday, February 5, 1999 10:38 AM
152 Vibration Fundamentals
Figure 15.10 Typical discrete natural frequency locations in structural members.
resonate by a bearing frequency, overhead crane, or any of a multitude of other energy
sources.
The actual resonant frequency depends on the mass, stiffness, and span of the excited
member. In general terms, the natural frequency of a structural member is inversely
proportional to the mass and stiffness of the member. In other words, a large tur-
bocompressors casing will have a lower natural frequency than that of a small end-
suction centrifugal pump.
Figure 15.10 illustrates a typical structural-support system. The natural frequencies of
all support structures, piping, and other components are functions of mass, span, and
stiffness. Each of the arrows on Figure 15.10 indicates a structural member or station-
ary machine component having a unique natural frequency. Note that each time a
structural span is broken or attached to another structure, the stiffness changes. As a
result, the natural frequency of that segment also changes.
While most stationary machine components move during normal operation, they are
not always resonant. Some degree of exing or movement is common in most station-
ary machine-trains and structural members. The amount of movement depends on the
10.Mobley.15 Page 153 Friday, February 5, 1999 10:38 AM
153 Failure-Mode Analysis
Figure 15.11 Dynamic resonance phase shift.
spring constant or stiffness of the member. However, when an energy source coin-
cides and couples with the natural frequency of a structure, excessive and extremely
destructive vibration amplitudes result.
Dynamic Resonance
When the natural frequency of a rotating, or dynamic, structure (e.g., rotor assembly
in a fan) is energized, the rotating element resonates. This phenomenon is classied as
dynamic resonance and the rotor speed at which it occurs is referred to as the critical.
In most cases, dynamic resonance appears at the fundamental running speed or one of
the harmonics of the excited rotating element. However, it also can occur at other fre-
quencies. As in the case of static resonance, the actual natural frequencies of dynamic
members depend on the mass, bearing span, shaft and bearing-support stiffness, and a
number of other factors.
Conrmation Analysis
In most cases, the occurrence of dynamic resonance can be quickly conrmed. When
monitoring phase and amplitude, resonance is indicated by a 180-degree phase shift
as the rotor passes through the resonant zone. Figure 15.11 illustrates a dynamic reso-
nance at 500 rpm, which shows a dramatic amplitude increase in the frequency-
domain display. This is conrmed by the 180-degree phase shift in the time-domain
plot. Note that the peak at 1200 rpm is not resonance. The absence of a phase shift,
coupled with the apparent modulations in the FFT, discount the possibility that this
peak is resonance related.
Common Confusions
Vibration analysts often confuse resonance with other failure modes. Because many
of the common failure modes tend to create abnormally high vibration levels that
appear to be related to a speed change, analysts tend to miss the root cause of these
problems.
10.Mobley.15 Page 154 Friday, February 5, 1999 10:38 AM
154 Vibration Fundamentals
Figure 15.12 Dynamic resonance plot.
Dynamic resonance generates abnormal vibration proles that tend to coincide with
the fundamental (1) running speed, or one or more of the harmonics, of a machine-
train. This often leads the analyst to incorrectly diagnose the problem as imbalance or
misalignment. The major difference is that dynamic resonance is the result of a rela-
tively small energy source, such as the fundamental running speed, that results in a
massive amplication of the natural frequency of the rotating element.
Function of Speed
The high amplitudes at the rotors natural frequency are strictly speed dependent. If
the energy source, in this case speed, changes to a frequency outside the resonant
zone, the abnormal vibration will disappear.
In most cases, running speed is the forcing function that excites the natural frequency
of the dynamic component. As a result, rotating equipment is designed to operate at
primary rotor speeds that do not coincide with the rotor assemblys natural frequen-
cies. Most low- to moderate-speed machines are designed to operate below the rst
critical speed of the rotor assembly.
Higher speed machines may be designed to operate between the rst and second, or
second and third, critical speeds of the rotor assembly. As these machines accelerate
through the resonant zones or critical speeds, their natural frequency is momentarily
excited. As long as the ramp rate limits the duration of excitation, this mode of opera-
10.Mobley.15 Page 155 Friday, February 5, 1999 10:38 AM
155 Failure-Mode Analysis
tion is acceptable. However, care must be taken to ensure that the transient time
through the resonant zone is as short as possible.
Figure 15.12 illustrates a typical critical-speed or dynamic-resonance plot. This gure
is a plot of the relationship between rotor-support stiffness (X-axis) and critical rotor
speed (Y-axis). Rotor-support stiffness depends on the geometry of the rotating ele-
ment (i.e., shaft and rotor) and the bearing-support structure. These are the two domi-
nant factors that determine the response characteristics of the rotor assembly.
FAILURE MODES BY MACHINE-TRAIN COMPONENT
In addition to identifying general failure modes that are common to many types of
machine-train components, failure-mode analysis can be used to identify failure
modes for specic components in a machine-train. However, care must be exercised
when analyzing vibration proles, because the data may reect induced problems.
Induced problems affect the performance of a specic component, but are not caused
by that component. For example, an abnormal outer-race passing frequency may indi-
cate a defective rolling-element bearing. It also can indicate that abnormal loading
caused by misalignment, roll bending, process instability, etc., has changed the load
zone within the bearing. In the latter case, replacing the bearing does not resolve the
problem and the abnormal prole will still be present after the bearing is changed.
Bearings: Rolling Element
Bearing defects are one of the most common faults identied by vibration monitoring
programs. Although bearings do wear out and fail, these defects are normally symp-
toms of other problems within the machine-train or process system. Therefore,
extreme care must be exercised to ensure that the real problem is identied, not just
the symptom. In a rolling-element, or antifriction, bearing vibration prole, three dis-
tinct sets of frequencies can be found: natural, rotational, and defect.
Natural Frequency
Natural frequencies are generated by impacts of the internal parts of a rolling-element
bearing. These impacts are normally the result of slight variations in load and imper-
fections in the machined bearing surfaces. As their name implies, these are natural
frequencies and are present in a new bearing that is in perfect operating condition.
The natural frequencies of rolling-element bearings are normally well above the max-
imum frequency range, F
MAX
, used for routine machine-train monitoring. As a result,
they are rarely observed by predictive maintenance analysts. Generally, these frequen-
cies are between 20 kHz and 1 MHz. Therefore, some vibration-monitoring programs
use special high-frequency or ultrasonic monitoring techniques such as high-fre-
quency domain (HFD).
10.Mobley.15 Page 156 Friday, February 5, 1999 10:38 AM
--------
156 Vibration Fundamentals
Note, however, that little is gained from monitoring natural frequencies. Even in cases
of severe bearing damage, these high-frequency components add little to the analysts
ability to detect and isolate bad bearings.
Rotational Frequency
Four normal rotational frequencies are associated with rolling-element bearings: fun-
damental train frequency (FTF), ball/roller spin, ball-pass outer-race, and ball-pass
inner-race. The following are denitions of abbreviations that are used in the discus-
sion that follows:
BD = Ball or roller diameter
PD = Pitch diameter
= Contact angle (for roller = 0)
n = Number of balls or rollers
f
r
= Relative speed between the inner and outer race (rps).
Fundamental Train Frequency
The bearing cage generates the FTF as it rotates around the bearing races. The cage
properly spaces the balls or rollers within the bearing races, in effect, by tying the
rolling elements together and providing uniform support. Some friction exists
between the rolling elements and the bearing races, even with perfect lubrication. This
friction is transmitted to the cage, which causes it to rotate around the bearing races.
Because this is a friction-driven motion, the cage turns much slower than the inner
race of the bearing. Generally, the rate of rotation is slightly less than one-half of the
shaft speed. The FTF is calculated by the following equation:
1 BD
FTF = --- f 1 --------
2
r
PD
Ball-Spin Frequency
Each of the balls or rollers within a bearing rotates around its own axis as it rolls
around the bearing races. This spinning motion is referred to as ball spin, which gen-
erates a ball-spin frequency (BSF) in a vibration signature. The speed of rotation is
determined by the geometry of the bearing (i.e., diameter of the ball or roller, and
bearing races) and is calculated by:
1 PD 2
BSF = --- -------- x f 1

BD

2
xcos
2 BD
r

PD

Ball-Pass Outer-Race
The ball or rollers passing the outer race generate the ball-pass outer-race frequency
(BPFO), which is calculated by:
n BD
BPFO = --- f

1 -------- xcos

2 PD

10.Mobley.15 Page 157 Friday, February 5, 1999 10:38 AM

157 Failure-Mode Analysis
Ball-Pass Inner-Race
The speed of the ball/roller rotating relative to the inner race generates the ball-pass
inner-race rotational frequency (BPFI). The inner race rotates at the same speed as the
shaft and the complete set of balls/rollers passes at a slower speed. They generate a
passing frequency that is determined by:
n BD
BPFI = --- f

1 + -------- xcos

2 PD

Defect Frequencies
Rolling-element bearing defect frequencies are the same as their rotational frequen-
cies, except for the BSF. If there is a defect on the inner race, the BPFI amplitude
increases because the balls or rollers contact the defect as they rotate around the bear-
ing. The BPFO is excited by defects in the outer race.
When one or more of the balls or rollers have a defect such as a spall (i.e., a missing
chip of material), the defect impacts both the inner and outer race each time one revo-
lution of the rolling element is made. Therefore, the defect vibration frequency is vis-
ible at two times (2) the BSF rather than at its fundamental (1) frequency.
Bearings: Sleeve (Babbitt)
In normal operation, a sleeve bearing provides a uniform oil lm around the sup-
ported shaft. Because the shaft is centered in the bearing, all forces generated by the
rotating shaft, and all forces acting on the shaft, are equal. Figure 15.13 shows the bal-
anced forces on a normal bearing.
Lubricating-lm instability is the dominant failure mode for sleeve bearings. This
instability is typically caused by eccentric, or off-center, rotation of the machine shaft
resulting from imbalance, misalignment, or other machine or process-related prob-
lems. Figure 15.14 shows a Babbitt bearing that exhibits instability.
When oil-lm instability or oil whirl occurs, frequency components at fractions (i.e.,
1/4, 1/3, 3/8, etc.) of the fundamental (1) shaft speed are excited. As the severity of
the instability increases, the frequency components become more dominant in a band
between 0.40 and 0.48 of the fundamental (1) shaft speed. When the instability
becomes severe enough to isolate within this band, it is called oil whip. Figure 15.15
shows the effect of increased velocity on a Babbitt bearing.
Chains and Sprockets
Chain drives function in essentially the same basic manner as belt drives. However,
instead of tension, chains depend on the mechanical meshing of sprocket teeth with
the chain links.
10.Mobley.15 Page 158 Friday, February 5, 1999 10:38 AM
158 Vibration Fundamentals
Figure 15.13 A normal Babbitt bearing has balanced forces.
Figure 15.14 Dynamics of Babbitt bearing that exhibits instability.
10.Mobley.15 Page 159 Friday, February 5, 1999 10:38 AM
159 Failure-Mode Analysis
Figure 15.15 Increased velocity generates an unbalanced force in a Babbitt bearing.
Gears
All gear sets create a frequency component referred to as gear mesh. The fundamental
gear-mesh frequency is equal to the number of gear teeth times the running speed of
the shaft. In addition, all gear sets create a series of sidebands or modulations that are
visible on both sides of the primary gear-mesh frequency.
Normal Prole
In a normal gear set, each of the sidebands is spaced by exactly the 1 running speed
of the input shaft and the entire gear mesh is symmetrical as seen in Figure 15.16. In
addition, the sidebands always occur in pairs, one below and one above the gear-mesh
frequency, and the amplitude of each pair is identical (Figure 15.17).
If we split the gear-mesh prole for a normal gear by drawing a vertical line through
the actual mesh (i.e., number of teeth times the input shaft speed), the two halves
would be identical. Therefore, any deviation from a symmetrical prole indicates a
gear problem. However, care must be exercised to ensure that the problem is internal
to the gears and not induced by outside inuences.
External misalignment, abnormal induced loads, and a variety of other outside inu-
ences destroy the symmetry of a gear-mesh prole. For example, a single-reduction
gearbox used to transmit power to a mold-oscillator system on a continuous caster
drives two eccentric cams. The eccentric rotation of these two cams is transmitted
10.Mobley.15 Page 160 Friday, February 5, 1999 10:38 AM
160 Vibration Fundamentals
Figure 15.16 Normal gear set prole is symmetrical.
Figure 15.17 Sidebands are paired and equal.
10.Mobley.15 Page 161 Friday, February 5, 1999 10:38 AM
161 Failure-Mode Analysis
Figure 15.18 Typical defective gear-mesh signature.
directly into the gearbox, creating the appearance of eccentric meshing of the gears.
However, this abnormal induced load actually destroys the spacing and amplitude of
the gear-mesh prole.
Defective Gear Proles
If the gear set develops problems, the amplitude of the gear-mesh frequency increases
and the symmetry of the sidebands changes. The pattern illustrated in Figure 15.18 is
typical of a defective gear set, where OAL is the broadband, or total, energy. Note the
asymmetrical relationship of the sidebands.
Excessive Wear
Figure 15.19 is the vibration prole of a worn gear set. Note that the spacing between
the sidebands is erratic and is no longer evenly spaced by the input shaft speed fre-
quency. The sidebands for a worn gear set tend to occur between the input and output
speeds and are not evenly spaced.
Cracked or Broken Teeth
Figure 15.20 illustrates the prole of a gear set with a broken tooth. As the gear
rotates, the space left by the chipped or broken tooth increases the mechanical clear-
ance between the pinion and bullgear. The result is a low-amplitude sideband to the
left of the actual gear-mesh frequency. When the next (i.e., undamaged) teeth mesh,
the added clearance results in a higher energy impact. The sideband to the right of the
mesh frequency has a much higher amplitude. As a result, the paired sidebands have a
10.Mobley.15 Page 162 Friday, February 5, 1999 10:38 AM
162 Vibration Fundamentals
Figure 15.19 Wear or excessive clearance changes the sideband spacing.
Figure 15.20 A broken tooth will produce an asymmetrical sideband prole.
nonsymmetrical amplitude, which is due to the disproportional clearance and impact
energy.
Improper Shaft Spacing
In addition to gear-tooth wear, variations in the center-to-center distance between
shafts create erratic spacing and amplitude in a vibration signature. If the shafts are
too close together, the sideband spacing tends to be at input shaft speed, but the ampli-
tude is signicantly reduced. This condition causes the gears to be deeply meshed
(i.e., below the normal pitch line), so the teeth maintain contact through the entire
mesh. This loss of clearance results in lower amplitudes, but it exaggerates any tooth-
prole defects that may be present.
10.Mobley.15 Page 163 Friday, February 5, 1999 10:38 AM
163 Failure-Mode Analysis
Figure 15.21 Unloaded gear has much higher vibration levels.
If the shafts are too far apart, the teeth mesh above the pitch line, which increases the
clearance between teeth and amplies the energy of the actual gear-mesh frequency
and all of its sidebands. In addition, the load-bearing characteristics of the gear teeth
are greatly reduced. Because the force is focused on the tip of each tooth where there
is less cross-section, the stress in each tooth is greatly increased. The potential for
tooth failure increases in direct proportion to the amount of excess clearance between
the shafts.
Load Changes
The energy and vibration proles of gear sets change with load. When the gear is fully
loaded, the proles exhibit the amplitudes discussed previously. When the gear is
unloaded, the same proles are present, but the amplitude increases dramatically. The
reason for this change is gear-tooth roughness. In normal practice, the backside of the
gear tooth is not nished to the same smoothness as the power, or drive, side. There-
fore, there is more looseness on the nonpower, or back, side of the gear. Figure 15.21
illustrates the relative change between a loaded and unloaded gear prole.
Jackshafts and Spindles
Another form of intermediate drive consists of a shaft with some form of universal
connection on each end that directly links the prime mover to a driven unit (see Fig-
ures 15.22 and 15.23). Jackshafts and spindles are typically used in applications
where the driver and driven unit are misaligned.
10.Mobley.15 Page 164 Friday, February 5, 1999 10:38 AM
164 Vibration Fundamentals
Figure 15.22 Typical gear-type spindle.
Figure 15.23 Typical universal-type jackshaft.
Most of the failure modes associated with jackshafts and spindles are the result of
lubrication problems or fatigue failure resulting from overloading. However, the
actual failure mode generally depends on the conguration of the exible drive.
Lubrication Problems
Proper lubrication is essential for all jackshafts and spindles. A critical failure point
for spindles (see Figure 15.22) is in the mounting pod that provides the connection
between the driver and driven machine components. Mounting pods generally use
10.Mobley.15 Page 165 Friday, February 5, 1999 10:38 AM
165 Failure-Mode Analysis
either a spade-and-slipper or a splined mechanical connector. In both cases, regular
application of a suitable grease is essential for prolonged operation. Without proper
lubrication, the mating points between the spindles mounting pod and the machine-
train components impact each time the torsional power varies between the primary
driver and driven component of the machine-train. The resulting mechanical damage
can cause these critical drive components to fail.
In universal-type jackshafts like the one illustrated in Figure 15.23, improper lubrica-
tion results in nonuniform power transmission. The absence of a uniform grease lm
causes the pivot points within the universal joints to bind and restrict smooth power
transmission.
The typical result of poor lubrication, which results in an increase in mechanical
looseness, is an increase of those vibration frequencies associated with the rotational
speed. In the case of gear-type spindles (Figure 15.22), there will be an increase in
both the fundamental (1) and second harmonic (2). Because the resulting forces
generated by the spindle are similar to angular misalignment, there also will be a
marked increase in the axial energy generated by the spindle.
The universal-coupling conguration used by jackshafts (Figure 15.23) generates an
elevated vibration frequency at the fourth (4) harmonic of its true rotational speed.
This failure mode is caused by the binding that occurs as the double pivot points move
through a complete rotation.
Fatigue
Spindles and jackshafts are designed to transmit torsional power between a driver and
driven unit that are not in the same plane or that have a radical variation in torsional
power. Typically, both conditions are present when these exible drives are used.
Both the jackshaft and spindle are designed to absorb transient increases or decreases
in torsional power caused by twisting. In effect, the shaft or tube used in these designs
winds, much like a spring, as the torsional power increases. Normally, this torque and
the resultant twist of the spindle are maintained until the torsional load is reduced. At
that point, the spindle unwinds, releasing the stored energy that was generated by the
initial transient.
Repeated twisting of the spindles tube or the solid shaft used in jackshafts results in a
reduction in the exible drives stiffness. When this occurs, the drive loses some of its
ability to absorb torsional transients. As a result, damage may result to the driven unit.
Unfortunately, the limits of single-channel, frequency-domain data acquisition pre-
vent accurate measurement of this failure mode. Most of the abnormal vibration that
results from fatigue occurs in the relatively brief time interval associated with startup,
when radical speed changes occur, or during shutdown of the machine-train. As a
result, this type of data acquisition and analysis cannot adequately capture these tran-
10.Mobley.15 Page 166 Friday, February 5, 1999 10:38 AM
166 Vibration Fundamentals
Figure 15.24 Load zones determined by wrap.
sients. However, the loss of stiffness caused by fatigue increases the apparent
mechanical looseness observed in the steady-state, frequency-domain vibration signa-
ture. In most cases, this is similar to the mechanical looseness.
Process Rolls
Process rolls commonly encounter problems or fail due to being subjected to induced
(variable) loads and from misalignment.
Induced (Variable) Loads
Process rolls are subjected to variable loads that are induced by strip tension, tracking,
and other process variables. In most cases, these loads are directional. They not only
inuence the vibration prole, but determine the location and orientation of data
acquisition.
Strip Tension or Wrap
Figure 15.24 illustrates the wrap of the strip as it passes over a series of rolls in a con-
tinuous-process line. The orientation and contact area of this wrap determines the load
zone on each roll. In this example, the strip wrap is limited to one-quarter of the roll
circumference. The load zone, or vector, on the two top rolls is on a 45-degree angle
to the passline. Therefore, the best location for the primary radial measurement is at
45 degrees opposite to the load vector. The secondary radial measurement should be
90 degrees to the primary. On the top-left roll, the secondary measurement point
10.Mobley.15 Page 167 Friday, February 5, 1999 10:38 AM
167 Failure-Mode Analysis
Figure 15.25 Load from narrow strip concentrated in center.
should be to the top left of the bearing cap; on the top-right roll, it should be at the
top-right position.
The wrap on the bottom roll encompasses one-half of the roll circumference. As a
result, the load vector is directly upward, or 90 degrees, to the passline. The best loca-
tion for the primary radial-measurement point is in the vertical-downward position.
The secondary radial measurement should be taken at 90 degrees to the primary.
Since the strip tension is slightly forward (i.e., in the direction of strip movement), the
secondary measurement should be taken on the recoiler-side of the bearing cap.
Because strip tension loads the bearings in the direction of the force vector, it also
tends to dampen the vibration levels in the opposite direction, or 180 degrees, of the
force vector. In effect, the strip acts like a rubber band. Tension inhibits movement
and vibration in the direction opposite the force vector and amplies the movement in
the direction of the force vector. Therefore, the recommended measurement-point
locations provide the best representation of the rolls dynamics.
In normal operation, the force or load induced by the strip is uniform across the rolls
entire face or body. As a result, the vibration prole in both the operator- and drive-
side bearings should be nearly identical.
Strip Width and Tracking
Strip width has a direct effect on roll loading and how the load is transmitted to the
roll and its bearing support structures. Figure 15.25 illustrates a narrow strip that is
tracking properly. Note that the load is concentrated on the center of the roll and is not
uniform across the entire roll face. The concentration of strip tension or load in the
center of the roll tends to bend the roll. The degree of deection depends on the fol-
lowing: roll diameter, roll construction, and strip tension. Regardless of these three
factors, however, the vibration prole is modied. Roll bending, or deection,
10.Mobley.15 Page 168 Friday, February 5, 1999 10:38 AM
168 Vibration Fundamentals
Figure 15.26 Roll loading.
increases the fundamental (1) frequency component. The amount of increase is
determined by the amount of deection.
As long as the strip remains at the true centerline of the roll face, the vibration prole
in both the operator- and drive-side bearing caps should remain nearly identical. The
only exceptions are bearing rotational and defect frequencies. Figures 15.26 and
15.27 illustrate uneven loading and the resulting different vibration proles of the
operator- and drive-side bearing caps. This is an extremely important factor that can
be used to evaluate many of the failure modes of continuous-process lines. For exam-
ple, the vibration prole resulting from the transmission of strip tension to the roll and
its bearings can be used to determine proper roll alignment, strip tracking, and proper
strip tension.
Alignment
Process rolls must be properly aligned. The perception that they can be misaligned
without causing poor quality, reduced capacity, and premature roll failure is incorrect.
In the case of single rolls (e.g., bridle and furnace rolls), they must be perpendicular to
the passline and have the same elevation on both the operator- and drive-side. Roll
pairs such as scrubber/backup rolls must be absolutely parallel to each other.
10.Mobley.15 Page 169 Friday, February 5, 1999 10:38 AM
169 Failure-Mode Analysis
Figure 15.27 Typical vibration prole with uneven loading.
Figure 15.28 Vertically misaligned roll.
Single Rolls
With the exception of steering rolls, all single rolls in a continuous-process line must
be perpendicular to the passline and have the same elevation on both the operator- and
drive-side. Any horizontal or vertical misalignment inuences the tracking of the strip
and the vibration prole of the roll.
Figure 15.28 illustrates a roll that does not have the same elevation on both sides (i.e.,
vertical misalignment). With this type of misalignment, the strip has greater tension
on the side of the roll with the higher elevation, which forces it to move toward the
lower end. In effect, the roll becomes a steering roll, forcing the strip to one side of
the centerline.
10.Mobley.15 Page 170 Friday, February 5, 1999 10:38 AM
170 Vibration Fundamentals
Figure 15.29 Scrubber roll set.
The vibration prole of a vertically misaligned roll is not uniform. Because the strip
tension is greater on the high side of the roll, the vibration prole on the high-side
bearing has lower broadband energy. This is the result of damping caused by the strip
tension. Dominant frequencies in this vibration prole are roll speed (1) and outer-
race defects. The low end of the roll has higher broadband vibration energy and dom-
inant frequencies include roll speed (1) and multiple harmonics (i.e., the same as
mechanical looseness).
Paired Rolls
Rolls that are designed to work in pairs (e.g., Damming or Scrubber rolls) also must
be perpendicular to the passline. In addition, they must be absolutely parallel to each
other. Figure 15.29 illustrates a paired set of Scrubber rolls. The strip is captured
between the two rolls and the counter-rotating brush roll cleans the strip surface.
Due to the designs of both the Damming and Scrubber roll sets, it is quite difcult to
keep the rolls parallel. Most of these roll sets use a single pivot point to x one end of
the roll and a pneumatic cylinder to set the opposite end.
Other designs use two cylinders, one attached to each end of the roll. In these designs,
the two cylinders are not mechanically linked and, therefore, the rolls do not maintain
their parallel relationship. The result of nonparallel operation of these paired rolls is
evident in roll life.
10.Mobley.15 Page 171 Friday, February 5, 1999 10:38 AM
171 Failure-Mode Analysis
Figure 15.30 Result of misalignment or nonparallel operation on brush rolls.
For example, the Scrubber/backup roll set should provide extended service life. How-
ever, in actual practice, the brush rolls have a service life of only a few weeks. After
this short time in use, the brush rolls will have a conical shape, much like a bottle
brush (see Figure 15.30). This wear pattern is visual conrmation that the brush roll
and its mating rubber-coated backup roll are not parallel.
Vibration proles can be used to determine if the roll pairs are parallel and, in this
instance, the rules for parallel misalignment apply. If the rolls are misaligned, the
vibration signatures exhibit a pronounced fundamental (1) and second harmonic
(2) of roll speed.
Multiple Pairs of Rolls
Because the strip transmits the vibration prole associated with roll misalignment, it
is difcult to isolate misalignment for a continuous-process line by evaluating one
single or two paired rolls. The only way to isolate such misalignment is to analyze a
series of rolls rather than individual (or a single pair of) rolls. This approach is consis-
tent with good diagnostic practices and provides the means to isolate misaligned rolls
and to verify strip tracking.
Strip Tracking
Figure 15.31 illustrates two sets of rolls in series. The bottom set of rolls is properly
aligned and has good strip tracking. In this case, the vibration proles acquired from
10.Mobley.15 Page 172 Friday, February 5, 1999 10:38 AM
172 Vibration Fundamentals
Figure 15.31 Rolls in series.
the operator- and drive-side bearing caps are nearly identical. Unless there is a dam-
aged bearing, all of the proles contain low-level roll frequencies (1) and bearing
rotational frequencies.
The top roll set also is properly aligned, but the strip tracks to the bottom of the roll
face. In this case, the vibration prole from all of the bottom bearing caps contains
much lower level broadband energy and the top bearing caps have clear indications of
mechanical looseness (i.e., multiple harmonics of rotating speed). The key to this type
of analysis is the comparison of multiple rolls in the order in which they are con-
nected by the strip. This requires comparison of both top and bottom rolls in the order
of strip pass. With proper tracking, all bearing caps should be nearly identical. If the
strip tracks to one side of the roll face, all bearing caps on that side of the line will
have similar proles. However, they will have radically different proles compared to
those on the opposite side.
Roll Misalignment
Roll misalignment can be detected and isolated using this same method. A misaligned
roll in the series being evaluated causes a change in the strip track at the offending
roll. The vibration proles of rolls upstream of the misaligned roll will be identical on
10.Mobley.15 Page 173 Friday, February 5, 1999 10:38 AM
173 Failure-Mode Analysis
Figure 15.32 Bends that change shaft length generate axial thrust.
both the operator- and drive-side of the rolls. However, the proles from the bearings
of the misaligned roll will show a change. In most cases, they will show traditional
misalignment (i.e., 1 and 2 components), but also will indicate a change in the uni-
form loading of the roll face. In other words, the overall or broadband vibration levels
will be greater on one side than the other. The lower readings will be on the side with
the higher strip tension and the higher readings will be on the side with less tension.
The rolls following the misalignment also show a change in vibration pattern. Since
the misaligned roll acts as a steering roll, the loading patterns on the subsequent rolls
show different vibration levels when the operator- and drive-sides are compared. If
the strip track was normal prior to the misaligned roll, the subsequent rolls will indi-
cate off-center tracking. In those cases where the strip was already tracking off-center,
a misaligned roll either improves or amplies the tracking problem. If the misaligned
roll forces the strip toward the centerline, tracking improves and the vibration proles
are more uniform on both sides. If the misaligned roll forces the strip further off-cen-
ter, the nonuniform vibration proles will become even less uniform.
Shaft
A bent shaft creates an imbalance or a misaligned condition within a machine-train.
Normally, this condition excites the fundamental (1) and secondary (2) running-
speed components in the signature. However, it is difcult to determine the differ-
ence between a bent shaft, misalignment, and imbalance without a visual inspection.
10.Mobley.15 Page 174 Friday, February 5, 1999 10:38 AM
174 Vibration Fundamentals
Figure 15.33 Bends that do not change shaft length generate radial forces only.
Figures 15.32 and 15.33 illustrate the normal types of bent shafts and the force pro-
les that result.
V-Belts
V-belt drives generate a series of dynamic forces, and vibrations result from these
forces. Frequency components of such a drive can be attributed to sheaves and belts.
The elastic nature of belts can either amplify or damp vibrations that are generated by
the attached machine-train components.
Sheaves
Even new sheaves are not perfect and may be the source of abnormal forces and
vibration. The primary sources of induced vibration due to sheaves are eccentricity,
imbalance, misalignment, and wear.
Eccentricity
Vibration caused by sheave eccentricity manifests itself as changes in load and rota-
tional speed. As an eccentric drive (Figure 15.34) sheave passes through its normal
rotation, variations in the pitch diameter cause variations in the linear belt speed. An
eccentric driven sheave causes variations in load to the drive. The rate at which such
variations occur helps to determine which is eccentric. An eccentric sheave also may
10.Mobley.15 Page 175 Friday, February 5, 1999 10:38 AM
175 Failure-Mode Analysis
Figure 15.34 Eccentric sheaves.
Figure 15.35 Light and heavy spots on an unbalanced sheave.
appear to be unbalanced. However, performing a balancing operation will not correct
the eccentricity.
Imbalance
Sheave imbalance may be caused by several factors, one of which may be that it was
never balanced to begin with. The easiest problem to detect is an actual imbalance of
the sheave itself. A less obvious cause of imbalance is damage that has resulted in loss
of sheave material. Imbalance due to material loss can be determined easily by visual
inspection, either by removing the equipment from service or using a strobe light
while the equipment is running. Figure 15.35 illustrates light and heavy spots that
result in sheave imbalance.
10.Mobley.15 Page 176 Friday, February 5, 1999 10:38 AM
176 Vibration Fundamentals
Figure 15.36 Angular sheave misalignment.
Figure 15.37 Parallel sheave misalignment.
Misalignment
Sheave misalignment most often produces axial vibration at the shaft rotational fre-
quency (1) and radial vibration at one and two times the shaft rotational frequency (1
and 2). This vibration prole is similar to coupling misalignment. Figure 15.36 illus-
trates angular sheave misalignment and Figure 15.37 illustrates parallel misalignment.
Wear
Worn sheaves also may increase vibration at certain rotational frequencies. However,
sheave wear is more often indicated by increased slippage and drive wear. Figure
15.38 illustrates both normal and worn sheave grooves.
10.Mobley.15 Page 177 Friday, February 5, 1999 10:38 AM
177 Failure-Mode Analysis
Figure 15.38 Normal and worn sheave grooves.
Belts
V-belt drives typically consist of multiple belts mated with sheaves to form a means of
transmitting motive power. Individual belts, or an entire set of belts, can generate
abnormal dynamic forces and vibration. The dominant sources of belt-induced vibra-
tions are defects, imbalance, resonance, tension, and wear.
Defects
Belt defects appear in the vibration signature as subsynchronous peaks, often with
harmonics. Figure 15.39 shows a typical spectral plot (i.e., vibration prole) for a
defective belt.
Imbalance
An imbalanced belt produces vibration at its rotational frequency. If a belts perfor-
mance is initially acceptable and later develops an imbalance, the belt has most likely
lost material and must be replaced. If imbalance occurs with a new belt, it is defective
and also must be replaced. Figure 15.40 shows a spectral plot of shaft rotational and
belt defect (i.e., imbalance) frequencies.
Resonance
Belt resonance occurs primarily when the natural frequency of some length of the belt
is excited by a frequency generated by the drive. Occasionally, a sheave also may be
excited by some drive frequency. Figure 15.41 shows a spectral plot of resonance
excited by belt-defect frequency.
Belt resonance can be controlled by adjusting the span length, belt thickness, and belt
tension. Altering any of these parameters changes the resonance characteristics. In
most applications, it is not practical to alter the shaft rotational speeds, which also are
possible sources of the excitation frequency.
Resonant belts are readily observable visually as excessive deection, or belt whip. It
can occur in any resonant mode so inection points may or may not be observed
10.Mobley.15 Page 178 Friday, February 5, 1999 10:38 AM
178 Vibration Fundamentals
Figure 15.39 Typical spectral plot (i.e., vibration prole) of a defective belt.
Figure 15.40 Spectral plot of shaft rotational and belt defect (i.e., imbalance) frequencies.
10.Mobley.15 Page 179 Friday, February 5, 1999 10:38 AM
179 Failure-Mode Analysis
Figure 15.41 Spectral plot of resonance excited by belt-defect frequency.
Figure 15.42 Examples of mode resonance in a belt span.
10.Mobley.15 Page 180 Friday, February 5, 1999 10:38 AM
180 Vibration Fundamentals
along the span. Figure 15.42 illustrates rst-, second-, and third-mode resonance in a
belt span.
Tension
Loose belts can increase the vibration of the drive, often in the axial plane.
In the case of multiple V-belt drives, mismatched belts also aggravate this condition.
Improper sheave alignment can also compromise tension in multiple-belt drives.
Wear
Worn belts slip and the primary indication is speed change. If the speed of the driver
increases and the speed of the driven unit decreases, then slippage is probably occur-
ring. This condition may be accompanied by noise and smoke, causing belts to over-
heat and be glazed in appearance. It is important to replace worn belts.
11.Mobley.16 Page 181 Friday, February 5, 1999 11:25 AM
Chapter16
SIGNATUREANALYSIS
Most failures of rotating and reciprocating machinery exhibit characteristic vibration
proles that are associated with specic failure modes. This phenomenon is due to the
forcing function, caused by a developing defect, having a unique characteristic signa-
ture. None of the ltered bandwidth monitoring methods provides the means to detect
and evaluate these unique proles. Signature analysis provides this capability and its
use is required in a comprehensive predictive maintenance program.
CHARACTERISTIC VIBRATION SIGNATURES
A vibration signature provides a clear, accurate snapshot of the unique frequency
components generated by, or acting on, a machine-train. Such a signature is obtained
by converting time-domain data into its unique frequency components using a fast
Fourier transform (FFT). Such a vibration signature, referred to as frequency-domain
data, is used in signature analysis to evaluate the dynamics of the machine.
Frequency-domain vibration signatures form the basis for any predictive maintenance
program designed to detect, isolate, and verify incipient problems within a machine-
train. These signatures are the basic tools used for in-depth analysis methods such as fail-
ure-mode, root-cause, and operating dynamics analyses. Operating dynamics analysis
TM
,
which is beyond the scope of this module, uses vibration data and other process
parameters, such as ow rate, pressure, and temperature, to determine the actual oper-
ating condition of critical plant systems.
TYPES OF SIGNATURE ANALYSIS
In general, new or immature predictive maintenance programs are limited to compar-
ative analysis or waterfall trending. Although these comparative techniques provide
181
11.Mobley.16 Page 182 Friday, February 5, 1999 11:25 AM
182 VibrationFundamentals
the ability to detect severe problems, they cannot be used to isolate and identify the
forcing functions or failure modes. These methods also are limited in their ability to
provide early detection of incipient problems.
As the predictive maintenance program matures, root-cause analysis and operating
dynamics analysis
TM
methods can be used. With the addition of these more advanced
diagnostic tools, vibration signatures become an even more valuable process perfor-
mance improvement tool.
Automatic Trending Analysis
A predictive maintenance program utilizing a microprocessor-based vibration ana-
lyzer and a properly congured database automatically trends vibration data on each
machine-train. In addition, it compares the data to established baselines and generates
trend, time-to-failure, and alert/alarm status reports.
The use of just these standard capabilities greatly reduces unscheduled failures. How-
ever, these automated functions do not identify the root causes behind premature
machine-train component failures. In most cases, more in-depth analysis allows the
predictive analyst to identify the reason for pending failure and to recommend correc-
tive actions to prevent a recurrence of the problem. Again, the specic microproces-
sor-based system used determines how much manual effort is required for more in-
depth analysis.
More In-Depth Trending Analysis
More in-depth analysis is called for when the automatic trending analysis described in
the previous section indicates that a machine-train is exhibiting excessive vibration.
Obviously, machine-trains that are operating within acceptable boundaries do not
require further investigation. Care should be taken, however, to ensure that the auto-
mated functions of the predictive maintenance system report abnormal growth trends
as well as machine-trains that are actually in alarm.
ComparativeAnalysis(WaterfallTrending)
FFT signatures that are collected on a regular schedule provide a means of trending
that can help the analyst identify changes in machine condition. Changes in the oper-
ating parameters, such as load, will directly affect the signatures generated by a
machine.
Unlike trending analysis, which is based on broadband and narrowband data, compar-
ative analysis is a visual comparison of the relative change of the machine-trains full
vibration signature and its discrete frequency components over a period of time.
Because vibration signatures are acquired at regular intervals in a predictive mainte-
nance program, this form of trending is very effective in identifying changes in
machine condition.
11.Mobley.16 Page 183 Friday, February 5, 1999 11:25 AM
183 SignatureAnalysis
Figure 16.1 Illustration of a waterfall plot.
Displaying the signatures in a waterfall or multiple-spectra display (sequentially by
data-acquisition time) allows the analyst to easily see the relationship of each fre-
quency component generated by the machine (see Figure 16.1). Any signicant
change in the amplitude of any discrete frequency is clearly evident in this type of dis-
play, which is used in many of the gures in subsequent sections.
Although comparative analysis can be used to help the analyst identify specic
changes that are generated by process changes, each signature must be normalized for
process variations. Therefore, as part of the acquired data set, the analyst must record
the specic process conditions for each data set. With this information and the water-
fall display of vibration signatures, the analyst can quantify the changes that result
from variations of these parameters.
Developing problems within a machine-train can be identied by comparing the FFT
signature to the following: (1) a baseline or reference signature, (2) previous signa-
tures, or (3) industrial standards. This method determines if a potential problem exists
and can be used to isolate within the machine-train the probable source of developing
problems.
11.Mobley.16 Page 184 Friday, February 5, 1999 11:25 AM
184 VibrationFundamentals
Figure 16.2 Comparison to baseline reference.
Baseline or Reference Signatures
A series of baseline or reference data sets should be taken for each machine-train
included in a predictive maintenance program (Figure 16.2). These data sets are nec-
essary to compare with trends, time traces, and FFT signatures that are collected over
time. Therefore, baseline data sets must be representative of the normal operating
condition of each machine-train in order to have value as a reference.
In integrated process plants where most machines are subject to variable operating
conditions, this exercise requires more than one reference data set for each machine-
train. To be of benet, a series of baselines must be acquired from each machine-train,
each of which should accurately represent a specic operating variable (i.e., product,
machine setup, load, etc.). It is important that all data sets (whether baseline data or
current operating data) be clearly identied in order to be useful. Current operations
data must be compared to a reference data set having the same operating conditions
(Figure 16.2).
Note that baseline references must be updated each time a machine-train is over-
hauled, replaced, or when a new process setup is established. A current set of valid
reference data is essential when performing comparative analysis.
11.Mobley.16 Page 185 Friday, February 5, 1999 11:25 AM
SignatureAnalysis 185
Table 16.1 Vibration Severity Standards*
Condition MachineClasses(IPS-PK)
Good operating condition
Alert limit
Alarm limit
Absolute-fault limit
I
0.028
0.010
0.156
0.260
II
0.042
0.156
0.396
0.400
III
0.100
0.255
0.396
0.620
IV
0.156
0.396
0.622
1.000
* Applicable to a machine with running speed between 600 to 12,000 rpm. Narrowband setting: 0.3 to
3.0 running speed.
MachineClassDescriptions:
Class I Small machine-trains or individual components integrally connected with the
complete machine in its normal operating condition (i.e., drivers up to 20 hp).
Class II Medium-sized machines (i.e., 20- to 100-hp drivers and 400-hp drivers on spe-
cial foundations).
Class III Large prime movers (i.e., drivers greater than 100 hp) mounted on heavy, rigid
foundations.
Class IV Large prime movers (i.e., drivers greater than 100 hp) mounted on relatively
soft, lightweight structures.
Source: Derived by Integrated Systems, Inc., from ISO Standard 2372.
Nonbaseline Signatures
Visual comparison of two signatures can enable the analyst to determine if a problem
is developing. As with the case of ltered energy data, all signatures must be normal-
ized for process variables such as speed, load, etc., in order for comparisons to be
valid. Direct comparison is useful only when both data sets reect the same operating
conditions or parameters.
Common-shaft analysis is used to identify the strongest vibration by visually compar-
ing the signatures of all measurement points on a common shaft. It is a useful tech-
nique for isolating the source of abnormal vibrations. Although this method does not
absolutely identify the problem, it reduces the number of machine components that
must be inspected or evaluated to correct the problem.
Industrial Standards
One form of comparative analysis is direct comparison of the acquired data to indus-
trial standards or reference values. The vibration severity standards presented in Table
16.1 were established by the International Standards Organization (ISO). These data
are applicable for comparison with ltered narrowband data taken from machine-
trains with true running speeds between 600 and 12,000 rpm. The values from the
table include all vibration energy between a lower limit of 0.3 true running speed
11.Mobley.16 Page 186 Friday, February 5, 1999 11:25 AM
186 VibrationFundamentals
and an upper limit of 3.0. For example, an 1800-rpm machine would have a ltered
narrowband between 540 (1800 0.3) and 5400 rpm (1800 3.0). A 3600-rpm
machine would have a ltered narrowband between 1080 (3600 0.3) and 10,800
rpm (3600 3.0).
MicroprocessorComparisons
Many of the microprocessor-based predictive maintenance systems also allow direct
comparisons of the relative strengths of each frequency component. Such micropro-
cessor comparisons do not require knowledge of the machine-train or vibration analy-
sis techniques, but both data sets must be acquired under the same operating
conditions. Increases in relative strength indicate more vibration and a developing
problem in the machine-train.
Cross-machine comparison is an extremely benecial tool to the novice analyst. Most
vibration monitoring systems permit direct comparison of vibration data, both ltered
window energy and complete signatures, acquired from two machines. This capability
permits the analyst to directly compare a machine that is known to be in good operat-
ing condition with one that is perceived to have a problem. There are several ways that
cross-machine comparisons can be made using microprocessor-based systems: multi-
ple plots, ratio, and difference.
Multiple Plots
Two or more signatures can be shown on a single display. This method permits the
analyst to directly compare the actual signatures generated at each measurement
point on both the suspect and a reference machine-train. This multiple-signature dis-
play permits direct comparison of each frequency component within the signatures
(Figure 16.3).
Ratio Analysis
With this technique, the signature from the suspect machine is divided by the signa-
ture of the reference machine, frequency by frequency. The resultant display shows
the relative amplitude, both positive and negative, of each frequency component in the
suspect machine-train (Figure 16.4). As an example, the display may indicate that the
gear-mesh energy in the suspect machine is 40% higher than that in the reference
machine (i.e., ratio = 1.4). With this information, the analyst can isolate specic
machine components that are potential problems.
Difference Analysis
With the difference analysis technique, the signature of the reference machine is sub-
tracted from that of the suspect machine, frequency by frequency. The resultant plot
displays the difference value, positive and negative, of each frequency component
within the two (see Figure 16.5).
11.Mobley.16 Page 187 Friday, February 5, 1999 11:25 AM
187 SignatureAnalysis
Figure 16.3 Multiple-signature display.
Figure 16.4 Ratio of two signatures.
11.Mobley.16 Page 188 Friday, February 5, 1999 11:25 AM
188 VibrationFundamentals
Figure 16.5 Difference of two signatures.
12.Mobley.17 Page 189 Friday, February 5, 1999 11:31 AM
Chapter17
ROOT-CAUSEANALYSIS
Root-cause analysis provides the means to isolate and identify specic causes of
abnormal vibration components observed for a machine-train. The trending and anal-
ysis methods discussed in the previous sections generally identify symptoms rather
than causes of problems.
Traditional predictive maintenance programs do not typically incorporate root-cause
analysis. Most programs are considered successful if they identify incipient problems
in time to prevent severe damage or forced downtime. However, programs that do not
include this type of analysis leave two questions unanswered: (1) Why did the prob-
lem occur? (2) Will it recur?
To achieve maximum benet from a predictive maintenance program, it is not enough
to predict failure in time to repair it without affecting production. In most instances,
problems can be solved at minimal expense through the use of root-cause analysis.
However, problems are often ignored for quite some time and machine failures
allowed to recur. The net result is that machine life can be drastically shortened, main-
tenance cost is increased, and available production time is reduced. In short, ignoring
chronic problems, no matter how small they seem, costs a typical plant a substantial
amount of actual, or potential, revenue every year.
DETERMINING WHY PROBLEMS OCCUR
Machine-train failures do not occur without a reason. To solve the problem, the reason
for failure must be found and corrected. Root-cause analysis is based on machine-
train operation and how its dynamics affect the vibration spectrum. However, the ana-
lyst may have to evaluate the entire process system to determine the reason behind a
chronic machine-train failure. If the chronic problem is system related rather than
machine-train related, the knowledge of process dynamics required to perform the
189
12.Mobley.17 Page 190 Friday, February 5, 1999 11:31 AM
190 VibrationFundamentals
analysis may be beyond the capabilities of the predictive maintenance staff. In this
case, the assistance of the plant engineering staff is required.
In extreme cases, the experience of the plant engineering staff will not be sufcient to
cost effectively resolve a chronic problem. Most plant engineers are generalists within
a specic discipline (e.g., mechanical, electrical). As such, they are not, and should
not be expected to be, specialists in machinery or process design and application.
Many of the chronic machinery problems that occur daily require special knowledge
of the specic machinery and system to nd the solution. In most cases, the rst line
of support for this expertise is the original equipment manufacturer. The original man-
ufacturer should certainly have the required machine design knowledge. The manu-
facturer may, however, be biased. In many cases, an unbiased consultant can provide a
cost-effective solution to a chronic problem.
WHEN TO PERFORM THE ANALYSIS
True root-cause analysis is not necessary for many machine-train problems. However,
if the problem is chronic and recurs more often than expected, such an analysis is
essential.
In many cases, recurring machine-train problems can be traced to a system or process
rather than something mechanical. However, chronic mechanical problems do exist
and usually can be traced to improper installation, application, or maintenance.
A classic example of a process-related problem is a recurring centrifugal air compres-
sor problem experienced by one of our clients. In this instance, three 4-stage centrifu-
gal compressors, each rated at 2150 cfm at 100 psig on a common header, exhibited
chronic failure problems. In a period of approximately 12 months, all three compres-
sors required major rebuilds and several minor repairs.
The predictive maintenance program identied each failure and a failure-mode analy-
sis isolated and identied the specic failure point. Nevertheless, the problem contin-
ued to recur. A root-cause analysis indicated that the real problem was not mechanical
(i.e., compressor/motor), but was a system (i.e., plant air) problem. A redesign of the
plant air system eliminated the problem and normal machine life was restored.
METHODOLOGY
The methodology used in root-cause analysis is basically the same as that used for
failure-mode analysis. The major difference is that for root-cause analysis the vibra-
tion signature is only one part of the data required to isolate the problem. In addition
to vibration data, complete process data (e.g., ow, pressure, temperature), detailed
process system-design data, and machine-design data are required. Sometimes, test-
ing utilizing techniques other than vibration analysis is required. However, the con-
12.Mobley.17 Page 191 Friday, February 5, 1999 11:31 AM
191 Root-CauseAnalysis
cept of root-cause analysis is to understand the relationship of all components within
the system so that the cause of a problem can be identied.
The premise of root-cause analysis, like failure-mode analysis, is that there is an iden-
tiable reason or reasons for each abnormal occurrence in machine-train or system
operation. However, each application is unique and the methods used vary with the
specic process system or application. Therefore, root-cause analysis requires a com-
plete knowledge of machine and process system design, operation, and typical failure
modes. While many chronic problems are unique to a particular plant, many are com-
monly found in other applications, plants, and/or industries.
Unfortunately, obtaining a good grasp of root-cause analysis techniques requires
much training and experience. In addition to formal technical education and plant
experience, a good experience base in both normal and abnormal systems operation is
required to cost effectively and accurately identify and correct most typical chronic
problems.
Because of this, it is generally more cost-effective to use a consultant to resolve many
chronic problems rather than training in-house staff. However, simple root-cause
analysis can and should be included in most predictive maintenance programs. Semi-
nars that provide the basic knowledge required to perform simple analyses are avail-
able from a number of consultants and predictive maintenance system manufacturers.
Simple root-cause analysis is certainly within the capabilities of most predictive
maintenance personnel and, with some technical support from outside consultants,
most chronic problems can be resolved.
SUSCEPTIBILITY TO CHRONIC PROBLEMS
Some machinery and process systems are more susceptible to chronic process- or
application-induced problems than are others. Compressors, pumps, and fans are
examples of machinery that commonly exhibit chronic problems as a direct result of
process or application problems. These machines are extremely sensitive to process
variables.
Chronic bearing failure is probably the most common machine/system-induced prob-
lem. By design, bearings are the weakest link in most machinery. A high failure rate is
a denite indication of a system/machine problem and chronic failures should not be
ignored because they are denite indicators that machine-train life is being adversely
affected. If bearings fail, other mechanical damage also may be present.
DIAGNOSTIC TECHNIQUES
The following diagnostic and data techniques are discussed in this section: common-
shaft analysis, shaft deection, and data normalization.
12.Mobley.17 Page 192 Friday, February 5, 1999 11:31 AM
192 VibrationFundamentals
Common-Shaft Analysis
Vibration analysis requires the use of proper evaluation techniques that permit the
analyst to understand fully the operating condition of each machine-train and its asso-
ciated system. However, the natural tendency is to evaluate each measurement point
individually or to limit evaluation to a single machine component, such as an electric
motor. While this approach provides some insight, it cannot provide a complete pic-
ture of the machine-trains dynamics.
A fundamental requirement of good analysis technique is to evaluate each machine-
train using a common-shaft approach rather than using data from individual measure-
ment points or machine-train components not found on a common shaft. With this
approach, all data acquired from the entire machine-trainfrom the rst outboard
motor bearing to the outboard bearing of the nal driven componentare evaluated
as a series of common shafts.
WhatIsaCommonShaft?
The term common shaft refers to each individual shaft that exists in a machine-train.
For example, there are two common shafts in a machine-train consisting of an electric
motor that is directly coupled to a single-reduction gearbox that is directly coupled to
a fan or other machine component. The rst extends from the outboard motor bearing
to the outboard bearing of the high-speed (input) gearbox shaft. The second common
shaft extends from the outboard bearing of the low-speed (output) gearbox shaft to the
outboard bearing of the fan or driven machine.
Figure 17.1 illustrates a double-reduction gearbox drive-train. In this conguration,
there are three common shafts. The rst begins at the outboard motor bearing and
ends at the outboard bearing of the gearbox input shaft. The second begins at the
inboard bearing of the intermediate gear shaft and ends with the intermediate gear
shaft outboard bearing. The nal common shaft begins with the inboard (left) bearing
of the gearbox output shaft and continues through the pump impeller.
DataOrganization
Data obtained from a common-shaft analysis should be organized so that the deec-
tion and dynamics of all shafts within the extended machine-train can be evaluated as
a single unit. For example, the primary radial data points from all bearing caps on a
common shaft are evaluated as one unit; the secondary data points as a second unit;
and the axial data points as the nal unit. After each of the individual common shafts
has been evaluated, the analyst can compare the dynamics of each to gain a clear
understanding of the overall dynamics of the entire machine-train.
Advantages
The use of common-shaft analysis greatly enhances an analysts ability to detect and
diagnose machine-train problems. The advantages of this approach include the ability
12.Mobley.17 Page 193 Friday, February 5, 1999 11:31 AM
193 Root-CauseAnalysis
Figure 17.1 Example of common shafts in machine-trains.
to conrm common failure modes; obtain a clear picture of the shaft displacement,
both internal and external; and isolate any abnormal forcing functions that may exist.
Comparing the amplitude of individual frequency components at the various measure-
ment points along a common shaft is a simple means of isolating the location of
developing machine-train problems. The energy of each vibration component is great-
est at the machine component or source of problem. This is because the machine-train
housing absorbs the energy as the data-measurement location moves away from the
source.
Therefore, comparing the strength of each frequency component at various points
along the common shaft is a means of isolating the source. Since most of the unique
frequency components have a direct relationship to the running speed of a specic
shaft, monitoring all measurement points on a common shaft simplies comparative
analysis of the unique frequency components created by that shaft.
12.Mobley.17 Page 194 Friday, February 5, 1999 11:31 AM
194 VibrationFundamentals
Even though this technique does not identify the specic failure mode or problem
(e.g., misalignment, imbalance), it does isolate the source or location of the problem.
With this information, specic problems are much easier to isolate and identify. For
example, a comparison of the 1, 2, and 3 components at each measurement point
on a common shaft can be used to locate potential misalignment problems. The
amplitude of each abnormal component is greatest at the source of vibration and
decreases as the distance from the source increases.
Common-shaft analysis provides the ability to conrm or eliminate potential failure
modes. For example, an electric motor may exhibit an abnormal second-harmonic
(2) frequency that is normally associated with parallel misalignment. If analysis is
limited to individual machine components, in this case the motor, there is no way to
conrm misalignment between the motor and its driven unit. When common-shaft
techniques are used, data from the driven unit is used to conrm or eliminate mis-
alignment. In this instance, the driven unit must also contain abnormal 2 vibration to
conrm misalignment between the motor and driven unit.
MachinesinSeries
The common-shaft approach can be expanded to include analysis of machines in
series, for example, the furnace roll of a continuous-process line or mill stands in a
hot strip or tandem mill. After completing an analysis of the individual machine-trains
in this example, the analyst should compare the dynamics of each machine-train to
those connected to them (Figure 17.2). In many cases, the only point of connection is
the strip, but this is more than enough to provide a direct mechanical connection and
to interlace the dynamics of the machines in series.
This approach is essential on all continuous-process lines and where two or more
machines are installed in series. By analyzing the entire system as a unit, the analyst
can better understand the dynamics of each machine. The use of this approach can be
illustrated with the misalignment that is frequently observed in single-reduction gear-
boxes that drive rolls found in the drive-train of a continuous-process line. Often, roll
misalignment or improper strip tracking or tension is the real source of the problem
that is diagnosed as misalignment in the gear set.
The abnormal loading of the rolls that occurs as a result of roll misalignment or
improper strip tracking is transmitted into the gearbox and, in effect, misaligns the
gears. By evaluating the interaction of these machines in series, the analyst can clearly
see that the source of the misalignment is outside of the gearbox and can track the
problem to its source.
MachinesinParallel
In applications where two or more machines are installed and operated in parallel, a
similar common-shaft approach should be used. After analyzing each of the machine-
trains, the analyst should compare the results to determine the interaction of the com-
12.Mobley.17 Page 195 Friday, February 5, 1999 11:31 AM
195 Root-CauseAnalysis
Figure 17.2 Machines in series should be analyzed as a unit.
plete system (Figure 17.3). In many cases, systems that use multiple machines in par-
allel are subject to moderate to severe process instability. This instability is created by
improper piping congurations, improper isolation of nonrunning or standby
machines, inlet starvation, and a variety of other causes. The only means of isolating
these problems is to analyze the entire system, not just the individual machine-trains.
The probability of two or more machines generating the exact same vibration prole
is extremely unlikely, but many system problems create a common prole in the
machines that make up the system. By comparing the proles of all machines in par-
allel, the analyst can quickly see that the problem is system related rather than
machine related.
Shaft Deection
Analysis of shaft deection is a fundamental diagnostic tool. If the analyst can estab-
lish the specic direction and approximate severity of shaft displacement, it is much
easier to isolate the forcing function. For example, when the discharge valve on an
end-suction centrifugal pump is restricted, the pumps shaft is displaced in a direction
opposite to the discharge volute. Such deection is caused by the back-pressure gen-
erated by the partially closed valve. Most of the failure modes and abnormal operating
dynamics that affect machine reliability force the shaft from its true centerline. By
12.Mobley.17 Page 196 Friday, February 5, 1999 11:31 AM
196 VibrationFundamentals
Figure 17.3 Machines in parallel should be analyzed as a group.
using common-shaft diagnostics, the analyst can detect deviations from normal oper-
ating condition and isolate the probable forcing function.
Data Normalization
Data acquired as part of a regular vibration-monitoring program must be normalized
before an effective analysis can be performed. It is virtually impossible to properly
evaluate the machine condition or to detect abnormal behavior without normalizing
the data.
Normalization is required to eliminate the effects of process changes in the vibration
proles. At a minimum, each data set must be normalized for speed, load, and the
other standard process variables. Normalization allows the use of trending techniques,
or the comparison of a series of proles generated over time.
Regardless of the machines operating conditions, the frequency components should
occur at the same location when comparing normalized data for a machine. Normal-
ization allows the location of frequency components to be expressed as an integer
multiple of shaft running speed, although fractions sometimes result. For example,
gear-mesh frequency locations are generally integer multiples (5, 10, etc.) and
12.Mobley.17 Page 197 Friday, February 5, 1999 11:31 AM
197 Root-CauseAnalysis
bearing-frequency locations are generally noninteger multiples (0.5, 1.5, etc.).
Plotting the vibration signature in multiples of running speed quickly differentiates
the unique frequencies that are generated by bearings from those generated by gears,
blades, and other components that are integers of running speed.
Speed
When normalizing data for speed, all machines should be considered to be variable
speedeven those classied as constant speed. Speed changes due to load occur even
with simple constant-speed machine-trains, such as electric-motor-driven centrifu-
gal pumps. Generally, the change is relatively minor (between 5 and 15%), but it is
enough to affect diagnostic accuracy. This variation in speed is enough to distort
vibration signatures, which can lead to improper diagnosis.
With constant-speed machines, an analysts normal tendency is to normalize speed to
the default speed used in the database setup. However, this practice can introduce
enough error to distort the results of the analysis because the default speed is usually
an average value from the manufacturer.
For example, a motor may have been assigned a speed of 1780 rpm during setup. The
analyst then assumes that all data sets were acquired at this speed. In actual practice,
however, the motors speed could vary the full range between locked rotor speed (i.e.,
maximum load) to synchronous (i.e., no-load) speed. In this example, the range could
be between 1750 and 1800 rpm, a difference of 50 rpm. This variation is enough to
distort data normalized to 1780 rpm. Therefore, it is necessary to normalize each data
set to the actual operating speed that occurs during data acquisition rather than using
the default speed from the database.
Take care when using the vibration analysis software provided with most micropro-
cessor-based systems to determine the machine speed to use for data normalization. In
particular, do not obtain the machine speed value from a display screen plot (i.e., on-
screen or print-screen) generated by a microprocessor-based vibration analysis soft-
ware program. This value is not to be used because the cursor position does not repre-
sent the true frequency of displayed peaks, but is instead an average value (see Figure
17.4). The graphics packages in most of the programs use an average of four or ve
data points to plot each visible peak. This technique is acceptable for most data-anal-
ysis purposes, but can skew the results if used to normalize the data. The approximate
machine speed obtained from such a plot is usually within 10% of the actual value,
which is not accurate enough to be used for speed normalization. Instead, use the peak
search algorithm and print out the actual peaks and associated speeds.
Load
Data also must be normalized for variations in load. Where speed variations result in a
right or left shift of the frequency components, variations in load change the ampli-
tude. For example, the vibration amplitude of a centrifugal compressor taken at 100%
12.Mobley.17 Page 198 Friday, February 5, 1999 11:31 AM
198 VibrationFundamentals
Figure 17.4 Screen display cursor position does not provide true frequency.
load is substantially lower than the vibration amplitude in the same compressor oper-
ating at 50% load.
In addition, the effect of load variation is not linear. In other words, the change in
overall vibration energy does not change by 50% with a corresponding 50% load vari-
ation. Instead, it tends to follow more of a quadratic relationship. A 50% load varia-
tion can create a 200%, or a factor of 4, change in vibration energy.
Note that none of the comparative trending or analysis techniques can be used on vari-
able-load machine-trains without rst normalizing the data. Again, since even
machines classied as constant-load machines operate in a variable-load condition, it
is good practice to normalize all data to compensate for load variations utilizing the
proper relationship for the application.
OtherProcessVariables
Other variations in a process or system have a direct effect on the operating dynamics
and vibration prole of the machinery. In addition to changes in speed and load, other
process variables affect the stability of the rotating elements, induce abnormal distri-
bution of loads, and cause a variety of other abnormalities that directly impact diag-
12.Mobley.17 Page 199 Friday, February 5, 1999 11:31 AM
199 Root-CauseAnalysis
nostics. Therefore, each acquired data set should include a full description of the
machine-train and process system parameters.
As an example, abnormal strip tension or traction in a continuous-process line
changes the load distribution on the process rolls that transport a strip through the
line. This abnormal loading induces a form of misalignment that is visible in the roll
and its drive-trains vibration prole.
13.Mobley.18 Page 200 Friday, February 5, 1999 11:37 AM
PartIII
RESONANCEANDCRITICALSPEED
ANALYSIS
200
13.Mobley.18 Page 201 Friday, February 5, 1999 11:37 AM
Chapter18
INTRODUCTION
Resonance is a large-amplitude vibration caused by a small periodic stimulus having
the same, or nearly the same, period as the systems natural vibration. In other
words, a low-level energy source may excite a natural frequency and cause a sub-
stantial increase in its amplitude. Every machine-train has at least one natural vibra-
tion frequency.
Resonance is a very destructive vibration and, in most cases, will cause major damage
to the machine or support structure. An example of resonance is a tuning fork. If you
activate a tuning fork by striking it sharply, the fork vibrates rapidly. As long as it is
held suspended, the vibration decays with time. However, if you place it on a desk
top, the fork may excite the natural frequency of the desk, which dramatically ampli-
es the vibration energy.
The same thing can occur if one or more of the running speeds of a machine excites
the natural frequency of the machine or its support structure. The key point to remem-
ber is that a very low amplitude energy source can cause a large vibration amplitude
when its frequency coincides with the natural frequency of a machine or structure.
Higher levels of input energy can cause catastrophic, near-instantaneous failure of the
machine or structure.
201
13.Mobley.18 Page 202 Friday, February 5, 1999 11:37 AM
Chapter19
TYPESOFRESONANCE
Two major classications of resonance are found in most manufacturing and process
plants: static and dynamic. Both types exhibit a broad-based, high-amplitude fre-
quency component when viewed in a fast Fourier transform (FFT), or frequency-
domain, vibration signature. Unlike meshing or passing frequencies, the resonance
frequency component does not have modulations or sidebands. Instead, resonance is
displayed as a single, clearly dened peak, which represents a large quantity of
energy. Such a peak is illustrated in Figure 19.1.
The high-energy content results from the amplitude of the peak as well as the broad
area under the peak. This combination of high peak amplitude and area is typical of
most resonance problems. The damping system associated with a resonance fre-
quency is indicated by the sharpness or width of the response curve,
n
, when mea-
sured at the half-power point. R
MAX
is the maximum resonance and R
MAX
/ 2 is the
half-power point for a typical resonance response curve.
To determine system damping, we must determine the maximum response. This is the
response at the resonant frequency as indicated by the maximum value of R
velocity
or
(R
v
)
MAX
(a dimensionless velocity response factor), which is dened by Q as shown in
the following equation:
1
R Q = ( )
MAX
= ---
v
2
where is fraction of critical damping. The maximum dimensionless acceleration fac-
tor, (R
a
)
MAX
, and dimensionless displacement factor, (R
d
)
MAX
, responses are slightly
larger and can be calculated as:
R
v
( )
MAX
R
d
= R = ---------------------------
a

( )
MAX
( )
MAX
2
1 2
(1 )
202
13.Mobley.18 Page 203 Friday, February 5, 1999 11:37 AM
203
2
TypesofResonance
Figure 19.1 Resonance response.
The damping in a system is indicated by the sharpness or width of the response curve
in the vicinity of a resonant frequency
n
. The width is designated as a frequency
increment, , measured at the half-power point where the value of R is equal to
R
MAX
--------------- . This also is illustrated in Figure 19.1. Where the values of are less than 0.1,
the damping of the system can be approximated by:
1
-------- = ----2
Q
n
STATIC RESONANCE
Static resonance is a function of the natural frequency of nondynamic, or stationary,
machine components (e.g., casings and bearing support pedestals) and structural
members (i.e., beams, piping, etc.). When one or more of the natural frequencies of a
stationary structure is energized or excited, it resonates.
Because static resonance is a nondynamic phenomenon, it is generally not associated
with the primary running speed of any associated machinery. Rather, the source of
static resonance can be any energy source that coincides with the natural frequency of
any stationary component. For example, an I-beam support on a continuous annealing
13.Mobley.18 Page 204 Friday, February 5, 1999 11:37 AM
204 VibrationFundamentals
Figure 19.2 Typical discrete natural frequency locations in structural members.
line may be energized by the running speed of a roll. However, it also can be made to
resonate by a bearing frequency, overhead crane, or other such energy source.
The resonant frequency depends on the mass, stiffness, and span of the excited mem-
ber. In general terms, the natural frequency of a structural member is inversely pro-
portional to its mass and stiffness. In other words, a large turbocompressors casing
will have a lower natural frequency than that of a small end-suction centrifugal pump.
Figure 19.2 illustrates a typical structural support system and the discrete natural fre-
quency locations. Each of the arrows indicates a structural member or stationary
machine component having a unique natural frequency. Note that each time a struc-
tural span is broken or attached to another structure, the stiffness changes. As a result,
the natural frequency of that segment also changes.
While most stationary machine components move during normal operation, they are
not always resonant. Some degree of exing or movement is common in stationary
machine-trains and structural members. The amount of movement depends on the
spring constant, or stiffness, of the member.
13.Mobley.18 Page 205 Friday, February 5, 1999 11:37 AM
205 TypesofResonance
Figure 19.3 Rotor support stiffness versus critical rotor speed.
DYNAMIC RESONANCE
When the natural frequency of a rotating (i.e., dynamic) structure, such as a bearing or
a rotor assembly in a fan, is energized, the rotating machine element resonates. This
phenomenon is called dynamic resonance and the rotor speed at which it occurs is the
critical speed.
Figure 19.3 illustrates a typical critical speed, or dynamic resonance, plot. The graph
shows the relationship between rotor-support stiffness (X-axis) and rotor speed (Y-
axis). Rotor-support stiffness depends on the geometry of the rotating element (i.e.,
shaft and rotor) and the bearing-support structure. These are the two dominant factors
that determine the response characteristics of the rotor assembly.
In most cases, running speed is the forcing function that excites the natural frequency
of the dynamic component. As a result, rotating equipment is designed to operate at
primary rotor speeds that do not coincide with the rotor assemblys natural frequen-
cies. As with static components, dynamic machine components have one or more nat-
ural frequencies that can be excited by an energy source that coincides with, or is in
proximity to, that frequency.
High amplitudes of the rotors natural frequency are strictly speed dependent. If the
frequency of the energy source, in this case speed, changes to a value outside the res-
onant zone, the abnormal vibration disappears.
13.Mobley.18 Page 206 Friday, February 5, 1999 11:37 AM
206 VibrationFundamentals
As with static resonance, the actual natural frequencies of dynamic members depend
on the mass, bearing span, shaft and bearing-support stiffness, freedom of movement,
and other factors that dene the response characteristics of the rotor assembly (i.e.,
rotor dynamics) under various operating conditions.
In most cases, dynamic resonance appears at the fundamental running speed or one of
the harmonics of the excited rotating element. However, it also can occur at other fre-
quencies. For example, a rotor assembly with a natural frequency of 1800 rpm cannot
operate at speeds between 1980 and 1620 rpm ( 10%) without the possibility of
exciting the rotors natural frequency.
Most low- to moderate-speed machinery is designed to operate below the rst critical
speed of the rotor assembly. Higher speed machines may be designed to operate
between the rst and second, or second and third, critical speeds of the rotor assem-
bly. As these machines accelerate through the resonant zones or critical speeds, their
natural frequency is momentarily excited. As long as the ramp rate limits the duration
of excitation, this mode of operation is acceptable. However, care must be taken to
ensure that the transition time through the resonant zone is as short as possible.
Note that critical speed should not be confused with the mode shape of a rotating
shaft. Deection of the shaft from its true centerline (i.e., mode shape) elevates the
vibration amplitude and generates dominant vibration frequencies at the rotors fun-
damental and harmonics of the running speed.
However, the amplitude of these frequency components tends to be much lower than
those caused by operating at a critical speed of the rotor assembly. Also, the excessive
vibration amplitude generated by operating at a critical speed disappears when the
speed is changed, but those caused by mode shape tend to remain through a much
wider speed range or may even be independent of speed.
Conrmation Analysis
In most cases, the occurrence of dynamic resonance can be quickly conrmed. When
monitoring phase and amplitude, resonance is indicated by a 180-degree phase shift
as the rotor passes through the resonant zone. Figure 19.4 illustrates a dynamic reso-
nance at 500 rpm, which shows a dramatic amplitude increase in the frequency-
domain display. Resonance is conrmed by the 180-degree phase shift in the time-
domain plot. Note that the peak at 1200 rpm is not resonance. The absence of a phase
shift, coupled with the apparent modulations in the FFT, eliminates the possibility that
this peak is resonance related.
Common Confusions
Vibration analysts often confuse resonance with other failure modes. Many of the
common failure modes tend to create abnormally high vibration levels that appear to
13.Mobley.18 Page 207 Friday, February 5, 1999 11:37 AM
207 TypesofResonance
Figure 19.4 Dynamic resonance phase shift.
be related to a speed change. Therefore, analysts tend to miss the root cause of these
problems.
Dynamic resonance generates abnormal vibration proles that tend to coincide with
the fundamental (1) running speed or one or more of the harmonics. This often leads
the analyst to incorrectly diagnose the problem as imbalance or misalignment.
14.Mobley.20 Page 208 Friday, February 5, 1999 11:45 AM
Chapter20
EXAMPLESOFRESONANCE
STATIC RESONANCE
Many machine-trains and production systems are subject to static and dynamic reso-
nance. This section discusses some specic examples for each type.
Examples of machinery that exhibit static resonance are variable-speed machines,
continuous-process lines, and deck-mounted machine trains.
Variable-Speed Machines
A variety of variable-speed process machinery, such as four-high rolling mills used by
the steel industry (Figure 20.1) are operated over a wide range of running speeds.
Because their normal mode of operation tends to excite one or more of the machines
natural frequencies, these machines are prime examples of equipment that experi-
ences static resonance.
Waterfall data, such as that taken from a typical cold reduction mill, clearly displays
the transitions through the resonance zones of a four-high mill. These zones occur as
the mill accelerates from dead-stop, and decelerates from full speed to dead-stop.
Some of the resonance zones are caused by excitation of the natural frequencies of the
mill stand or other stationary members of the mill. Others are the result of dynamic
resonance created by the excitation of the natural frequency of a roll or other rotating
member within the mill.
Without a clear understanding of the specic natural frequencies of a process system,
it is difcult to separate the static and dynamic resonance exhibited by a waterfall plot
such as the one shown in Figure 20.2. This gure is a typical waterfall plot of a com-
plete production cycle for a cold reduction mill. Note how the running speed of the
208
14.Mobley.20 Page 209 Friday, February 5, 1999 11:45 AM
209 ExamplesofResonance
Figure 20.1 Variable-speed four-high rolling mill.
Figure 20.2 Waterfall or cascade plot.
14.Mobley.20 Page 210 Friday, February 5, 1999 11:45 AM
210 VibrationFundamentals
Figure 20.3 Continuous-process line.
rolls, gears, and other mill components passes through a number of resonant zones as
the mill accelerates. These resonance zones, displayed as broad-based peaks, are
clearly visible as mill speed increases from left to right of the horizontal axis.
Continuous-Process Lines
All continuous-process lines, such as plating lines, paper machines, etc., are subject to
resonance due to the excitation of one or more natural frequencies of their support
structure. Figure 20.3 illustrates a continuous-process line.
Deck-Mounted Machine-Trains
Any machine-train that is mounted on a deck-plate rather than a solid concrete foun-
dation is subject to resonance problems. If the stiffness of the deck-plate is inadequate
or the support span is too great, the normal result is static resonance created by the
excitation of the deck-plates natural frequency.
Others
At one time or another, all machine-trains in a plant are subject to resonance prob-
lems. The proximity of other machines and transients caused by variable running
speeds greatly increases the potential for periodic, momentary excitation of one or
more of the natural frequencies. As long as these transients are short lived, they nor-
mally do not cause serious problems. However, sustained excitation of a natural fre-
quency can, and often does, result in severe damage.
In most cases, resonance is limited to the casing or support structure of the machine-
train. The resulting vibration typically has a low frequency and may exhibit extremely
high amplitudes. Gearboxes, compressors, pumps, and other machine types are partic-
ularly susceptible to this form of resonance. Because the source of excitation is often
external to the monitored machine, static resonance is generally difcult to isolate.
14.Mobley.20 Page 211 Friday, February 5, 1999 11:45 AM
211 ExamplesofResonance
DYNAMIC RESONANCE
Most of the machine-trains used in a plant are susceptible to dynamic resonance. It is
especially prevalent in variable-speed machine-trains that are operated over a wide
range of speeds. However, even constant-speed machines, such as fans and blowers,
are prime candidates for resonance problems. Rolling mills, which are variable-speed
machines, also are prime candidates for dynamic resonance.
Fans and Blowers
Dynamic resonance is one of the most common failure modes of fans and blowers.
While most fans are operated at or near constant speed, it is possible to create situa-
tions where the speed of rotation coincides with the rotors natural frequency.
Although all fans and blowers are susceptible, cantilevered or overhung designs are
the most likely candidates for resonance or critical speed problems.
Typical fans and blowers are designed to operate at speeds 10 to 15% below the
rotors rst critical speed. As long as the fans speed and the rotors mass remain con-
stant, this design practice does not create a problem. However, when either speed or
mass changes, serious problems may result.
Many fans are belt driven. As a result, the sheave ratio may be changed to increase
speed. In some cases, this change in ratio and, hence, speed is unintentional. For
example, a millwright might replace a damaged sheave with one of different diameter.
In other cases, the speed may be raised in an attempt to increase ow or pressure. In
either case, the result is the same. The new fan speed may coincide with the rst criti-
cal speed of the rotor assembly and severe, potentially destructive vibration may
occur.
Another common problem associated with fans and blowers is an increase in rotor
mass. In the dirty plant environment, the rotor assemblies in fans and blowers tend to
accumulate dirt, moisture, and other contaminants. This phenomenon, called plate-
out, increases the mass of the rotating element. Because the natural frequency of the
rotor is dependent on its mass, this increase changes the natural frequency. As the
mass increases, the natural frequency becomes lower. If the mass changes enough, the
rst critical of the rotor assembly may coincide with the design running speed. The
result is an increase in vibration amplitude at running speed.
Rolling Mills
As mentioned in the variable-speed machine discussion all hot and cold reduction
rolling mills are highly susceptible to dynamic resonance. Each of the rolls has a nat-
ural frequency determined by its installed conguration. The natural frequency of
each roll depends on a number of variables that change during normal operation of the
mill.
14.Mobley.20 Page 212 Friday, February 5, 1999 11:45 AM
212 VibrationFundamentals
Variables, such as roll bending, roll force, and balancing force, will change the natural
frequency of each roll. As a result, it is extremely difcult to isolate the specic roll
that is being affected by resonance. Many of the chatter problems associated with cold
reduction and temper mills are caused by dynamic resonance. Chatter is caused by
gauge deviation in the strip. Third and fth octave chatter problems are, in many
cases, the excitation of the natural frequencies of the work and backup rolls (dynamic
resonance) or mill stand (static resonance).
15.Mobley.21 Page 213 Friday, February 5, 1999 11:48 AM
Chapter21
TESTINGFORRESONANCE
The purpose of resonance testing is to isolate the machine component that is being
excited and to determine the source of the excitation force.
STATIC RESONANCE
Static resonance testing is limited to structural members or machine components that
do not have dynamic physical properties (i.e., properties that change with speed or
time). Such structures include piping, machine casings, machine supports, deck-
plates, and other structural members.
During testing, the natural frequencies of the entire system are compared with the
vibration, or forcing, frequencies on an interference (i.e., Campbell) diagram to deter-
mine if the system is resonant. Figure 21.1 illustrates such a diagram.
In most cases, evidence of a potential static resonance problem will be found in the
routine frequency- and time-domain vibration data that are collected as part of a pre-
dictive maintenance program. These data will contain high-amplitude, high-energy
frequency components that cannot be explained or identied as a specic dynamic
force generated by the machine-train or its systems. The component generated by
potential static resonance may be at any frequency from 1 Hz to 30 kHz, but will
rarely fall at the fundamental (1) or any harmonic of running speed.
Isolating the Natural Frequency
In most cases, identifying the specic structural member or static machine component
being excited is very difcult. In a typical structure, there are a large number of natu-
ral frequencies with each corresponding to a specic structural member or span. As a
213
15.Mobley.21 Page 214 Friday, February 5, 1999 11:48 AM
214 VibrationFundamentals
Figure 21.1 Campbell diagram.
Figure 21.2 Simple machine support system.
15.Mobley.21 Page 215 Friday, February 5, 1999 11:48 AM
215 TestingforResonance
result, it is time consuming to test each component. Unfortunately, this is the only
positive means of isolating the offending component.
In a simple support system such as the one illustrated in Figure 21.2, the natural fre-
quencies of the structure can be isolated by mounting multiple sensors on the struc-
ture and then hitting, or ringing, the structure with a zero-impact hammer. This
approach is valid, but may not identify all natural frequencies of the structure. For
example, the natural frequency of the right leg may not be the same as the left.
Therefore, the only accurate method to identify all natural frequencies is to isolate
and ring or excite each structural member or static machine component. The method-
ology to be used depends on the type of machine component or structure to be tested
and the source of excitation energy.
The excitation energy source can be difcult to determine, but in many cases it can be
traced directly to one or more dynamic forces in proximity to the test structure. The
possible sources of forcing functions, or excitation energies, include machine running
speed(s), imbalance of a rotating or reciprocating element, misalignment, gear mesh,
hydraulic/aerodynamic noise, and a variety of other abnormal dynamics that may be
generated by machine-trains or process systems.
A number of energy sources can be used during testing to excite natural frequencies
of stationary machine components or structural members. These sources include sinu-
soidal and nonsinusoidal vibration forces and ringing.
SinusoidalVibrationForces
Sinusoidal vibration forces can be used to excite the natural frequency of stationary
components. However, these forces must be swept through the frequency range until
they match the natural frequencies of the structure being tested.
NonsinusoidalVibrationForces
A nonsinusoidal force generates orders of the fundamental forcing frequency that, in
turn, excites the structures natural frequency. This phenomenon often occurs in nor-
mal machine operation.
Sometimes a shaker is used in conjunction with a power amplier and a wave genera-
tor to excite the natural frequencies. The natural frequencies are determined by
sweeping through the frequency range of interest. An example of a mechanical shaker
is a variable-speed motor having a double-ended shaft with offset disks for mass
unbalance. Other excitation sources, including random forces, can be added to the
shaker to excite natural frequencies instantaneously.
Some frequencies generated by random forces coincide with, and thus excite, the
machine-trains natural frequencies. This phenomenon can be seen in signatures taken
15.Mobley.21 Page 216 Friday, February 5, 1999 11:48 AM
216 VibrationFundamentals
from the operating machines where random energy in the system can excite the natu-
ral frequencies.
Ringing
The most popular method for exciting natural frequencies is to strike a machine or
structure with a timber, or hammer. The range of frequencies excited depends on the
duration of the impact.
Hard-faced steel hammers tend to bounce off a structure, thus providing a short-dura-
tion impact. As a result, this device only excites high frequencies. To excite natural
frequencies of 10 Hz or less, a soft tip must be used on the hammer.
Isolating the Forcing Function
The preceding sections describe how to identify specic machine components or
structures that are experiencing static resonance. In addition, they discuss how to iso-
late the unique natural frequencies. The next step is to nd the sources of the forcing
functions or energy sources.
The rst step is to dene clearly the specic frequency being excited. There must be
an energy source with a frequency identical to the resonant frequency or some fre-
quency within its generated broadband energy. The source of excitation energy can be
determined either by calculation or by mapping.
Calculation
An analyst can easily calculate certain unique frequencies generated by a machine-
train or process system. In many cases, the excitation energy source will be within the
same machine-train or in proximity to the point of resonance. Therefore, the analyst
should start with the machinery closest to the point of resonance. The following are
examples of easily determined unique frequencies for a centrifugal pump rotating at
1800 rpm and having 10 vanes on the impeller:
Fundamental frequency is equal to the rotating speed, or 1800 rpm.
Vane-pass frequency (cycles per minute) is equal to the number of vanes on
the impeller multiplied by the rotating speed (i.e., 10 vanes 1800 rpm
= 18,000 cpm).
While the calculation method does not conrm the location of an energy source, it
provides a list of most likely sources of excitation. Direct measurement of these
sources using a vibration analyzer can then be used to isolate the forcing function.
Mapping
Since the specic resonant frequency is known, the analyzer can be used to track the
source of that unique frequency. If the excitation source is within the machine, the
15.Mobley.21 Page 217 Friday, February 5, 1999 11:48 AM
217 TestingforResonance
meter can be used to record the amplitude of the resonant frequency at various points
around the machine. When a source of that specic frequency is found, the machine
component adjacent to that measurement location is a probable source of excitation.
The same approach can be used to locate sources of excitation energy outside the res-
onant machine or structural member. Data can be acquired at regular intervals around
the resonant member. If an energy source that coincides with the resonant frequency
is observed, it can be tracked to its origination point using this method.
Natural frequencies are not always excited by an energy source that is a unique, con-
current frequency. Broadband energy sources also can be the source of excitation. For
example, an unbalanced motor may not generate a measurable frequency that coin-
cides with the observed resonance, but its broadband output may contain energy that
coincides with, or is an integer multiple of, the natural frequency. Therefore, any high
broadband energy is a potential source of excitation.
Testing Conditions
For resonance testing, the structure, piping, or machine should be as close as possible
to its normal operating state. Parts of a machine cannot arbitrarily be removed and
tested. For example, the natural frequencies of a gear that is not mounted on its shaft
differ from those of a mounted gear. Similarly, the natural frequencies of a machine
that is mounted for shop testing differ from those of a machine mounted on its normal
foundation.
The level of sophistication and detail of resonance testing varies. A simple resonance
test often provides the necessary structural natural frequency information. However,
better information and greater detail can be obtained from more sophisticated tests
and instrumentation.
Test Equipment
Tunable lter analyzers and storage oscilloscopes (analog or digital) can be used to
observe the instantaneous vibration resulting from an impact or excitation of a structure.
A trigger level must be set on the oscilloscope in the single-sweep mode. The vibra-
tion signal from the impact will be held on the screen of the oscilloscope. The natural
period is determined by counting the divisions in one period of vibration and multi-
plying that number by the time-per-division setting on the time base. The procedure
using a digital oscilloscope is similar, except that the period is read directly with a
cursor.
Rough estimates of the natural frequencies of a structure can be obtained with a tun-
able lter analyzer in the lter-out mode by observing the frequency meter after
impact. The frequency meter will indicate the natural frequency as long as the struc-
ture is ringing.
15.Mobley.21 Page 218 Friday, February 5, 1999 11:48 AM
218 VibrationFundamentals
The described methods evaluate the vibration response to a single impact. When such
a test is conducted, the sweep lter is started and the structure is repeatedly bumped
until the entire frequency range is scanned. Note, however, that the decay time for
each bump is so rapid that they do not interact. The analyzer responds as the lter
sweeps through the natural frequencies and the peaks of the envelope of the responses
indicate the natural frequencies.
This bump test also can be performed using a FFT analyzer and a single impact rather
than a series of impacts. In this case, the trigger on the analyzer is set to respond to
impact from the hammer.
DYNAMIC RESONANCE
Rather than a Hanning window, the setting should be a uniform window. A variety of
test methods can be used to identify dynamic resonance. Each has proven capability
for its specic application, but generally cannot be used in other applications.
Constant-Speed Machines
This section reviews the best techniques for most common applications.
Constant-speed machines that are being operated at their design speed and load
should not be affected by dynamic resonance. In most cases, dynamic resonance
problems in this class of machine result from a radical change in the operating enve-
lope (i.e., speed, load, etc.) or a modication of the machine-train.
Identifying dynamic resonance in a constant-speed machine-train is sometimes dif-
cult. Routine monitoring such as that conducted as part of a predictive maintenance
program detects the abnormal vibration levels that result from dynamic resonance, but
do not clearly isolate resonance as the source of the problem.
In the previous fan example, the fundamental running speed of the fan shaft is the
forcing function and the rst critical speed is the resonant frequency. Both appear at
the running-speed frequency. Because imbalance and most other failure modes also
result in an increase in the fundamental (1) frequency component, the question is
how to separate these failure modes from resonance.
The major difference between resonance and other failure modes is the amplitude of
the frequency component. Normally, common failure modes such as imbalance and
misalignment increase the amplitude of the fundamental (1) frequency, but the
increase is small to moderate when compared to prior readings or to other frequency
components.
Dynamic resonance will dramatically increase the amplitude of the natural frequency
component. Typically, the relationship between the energy levels at the resonant or
15.Mobley.21 Page 219 Friday, February 5, 1999 11:48 AM
219 TestingforResonance
natural frequency is an order of magnitude or more higher than any of the normal
rotational frequencies associated with the machine-train. In addition, the resonance
frequency tends to have a broader base than normal rotational frequencies with no
modulations.
The only positive means of isolating resonance in a constant-speed machine is to per-
form a coast-down or run-up transient analysis. This type of analysis can be con-
ducted with the vibration analyzer in the transient-capture mode. Mounting
transducers and a tachometer on the suspect machine, then recording the amplitude
and phase of the suspected resonant frequency performs the test.
In a coast-down test, the machine-trains driver component is turned off and the sus-
pect frequency is recorded as the speed decreases. If the problem is critical speed, the
amplitude decreases rapidly as the speed decreases. In addition, there is a 180-degree
phase shift as the machines speed passes through the resonant zone.
A run-up test uses the same setup, but records the amplitude and phase as the machine
accelerates from dead-stop to full speed. The amplitude of the natural frequency dra-
matically increases as the speed coincides with the rotors natural frequency. In addi-
tion, the natural frequencys phase shifts 90 degrees as the machines speed enters the
resonant zone. If the speed continues to increase and leaves the upper limit of the res-
onant zone, the phase will shift an additional 90 degrees.
Variable-Speed Machines
Isolation of resonance in variable-speed machines (e.g., cold reduction mills) can be
accomplished by acquiring a waterfall plot of signatures during either run up or coast
down of the machine (see Figure 21.3). The waterfall plot consists of a series of fre-
quency-domain vibration signatures that are captured periodically throughout these
two transients. The frequency of data capture can be controlled by either a tachometer
input or time interval, but should be high enough to ensure complete coverage of the
acceleration and deceleration of the machine-train.
Typically, signatures should be captured at least once per second or every ve revolu-
tions of the primary drive-shaft. This frequency is, in most cases, sufcient to capture
the transition of the running speed through any natural frequencies or resonant zones.
The waterfall plot clearly displays resonant zones that are crossed by any of the
rotational frequencies within the machine-train. If the resonance is excited by the
major components (e.g., running speed, gear mesh), the waterfall plot will, in most
cases, clearly display the resonance as a major increase in amplitude of the resonant
frequency.
For example, the gear-mesh energy from a gearbox or a pinion stand in a cold-reduc-
tion mill typically passes through one or more resonant zones as the mill accelerates
or decelerates. The transients are displayed as momentary increases in the amplitude
15.Mobley.21 Page 220 Friday, February 5, 1999 11:48 AM
220 VibrationFundamentals
Figure 21.3 Waterfall or cascade plot.
of the gear-mesh frequency as it passes through the resonant zone. In the same fash-
ion, the running speed of the mill also passes through resonant zones during accelera-
tion and deceleration.
Most variable-speed machines exhibit this same type of momentary resonance. Nor-
mally, it is not harmful and does not require corrective action. As long as the transi-
tion through the resonant zone is made quickly and smoothly, there is no adverse
effect on the machine-train. However, if the excitation force (i.e., running speed or
gear mesh) is maintained for some time within a resonant zone, serious damage may
occur. Severe chatter or even catastrophic failure of the mill or one of its components,
such as a roll, could occur.
Dynamic resonance is not always driven by an obvious excitation source. Because
extremely low levels of vibration energy can excite a natural frequency, other machine
components, such as bearings, can be the source of extremely destructive resonance.
Using the cold mill as an example, one of the most common sources of resonance is
the outer-race passing frequency of a work-roll bearing. Because of roll-bending prac-
tices, the load zone in the work-roll bearing is shifted to the outer race. The low-level
energy created by the balls or rollers passing the outer race is sufcient to excite the
natural frequency of the work roll. When this occurs, the resonance of the roll results
15.Mobley.21 Page 221 Friday, February 5, 1999 11:48 AM
221 TestingforResonance
in roll bounce or eccentric rotation. In turn, this bounce causes gauge deviation in the
strip, or chatter.
The most important point to remember when searching for the excitation source of
resonance is that the amplitude or strength of the source does not have to be high.
Amplitudes of less than 0.03 IPS-PK can be enough to excite the rolls or mill stands
and result in a measurable gauge deviation in the strip.
15.Mobley.21 Page 222 Friday, February 5, 1999 11:48 AM
Chapter22
MODESHAPE
Machinery rotor assemblies, such as shafts, are designed to rotate on their true center-
line and in a state of equilibrium. In practical applications, however, few machines
achieve optimum or design conditions. As a result, most machines operate with a
slight amount of imbalance where the shaft rotates slightly off of its true centerline.
Such rotor deection is referred to as mode shape.
Even in a perfectly installed system, imbalance often results from the difference
between the lift that was designed into the rotor assembly and the force of gravity that
pulls the rotor downward. Because the effects of gravity vary with altitude, and the lift
varies with temperature, humidity, and barometric pressure, there is always some dif-
ferential. As a result, all machine trains have some imbalance and rotate offset from
their true centerline. The slightly eccentric or off-center rotation of the shaft generates
a low-level frequency component that coincides with the rotating speed of the shaft.
This type of rotor dynamics is called the rst mode.
In addition, all rotating shafts are exible and change shape throughout their normal
operating range. The shape or shapes a shaft takes in actual operation are a function of
shaft stiffness, bearing span, rotor weight, and the action of outside forces such as
aerodynamic or hydrodynamic forces generated by the system. In ideal systems, the
shaft retains a relatively straight shape, but does not rotate on its true centerline. This
offset or eccentric rotation is one form of the rst mode of the rotor assembly. If the
rotor shaft deforms and bends, but maintains its original node points (refer back to
Figure 19.3), it will continue to generate a vibration frequency at the fundamental
running speed of the shaft (1). The amplitude of this fundamental frequency is
directly proportional to the amount of bend or deection between the shaft bearings.
If the shaft exes into a double bend that crosses its true centerline, it creates a node
point at the point at which it crosses the neutral or centerline. As the shaft rotates, the
222
15.Mobley.21 Page 223 Friday, February 5, 1999 11:48 AM
ModeShape 223
double-bend shape creates two high spots as it passes the vibration transducer. These
high spots are interpreted as the fundamental (1) and second-harmonic (2) frequen-
cies of running speed. This prole describes the second mode of the rotor assembly.
In some applications, the shaft can ex or deform into mode shapes that generate third
(3) and fourth (4) running-speed harmonics. Cantilevered rotating machine ele-
ments, such as on overhung fans, are typical examples of machinery subject to shaft
deections that force operation in the third or fourth mode.
16.Mobley.23 Page 224 Friday, February 5, 1999 11:55 AM
PartIV
REAL-TIMEANALYSIS
224
16.Mobley.23 Page 225 Friday, February 5, 1999 11:55 AM
Chapter23
OVERVIEW
Real-time analysis (RTA) is an advanced diagnostic technique. It is especially useful
with complex machinery, for evaluating transient events, such as rapid speed changes,
or where the steady-state vibration data gathered by conventional vibration analyzers
are not sufcient to isolate the root cause of a problem.
HARDWARE REQUIREMENTS
Real-time vibration analyzers provide many benets that are essential for a compre-
hensive vibration-monitoring and analysis program. They provide the ability to ana-
lyze a dynamic signal and they permit the analyst to view the actual dynamics of a
machine or process system in real time. As a result, the analyst can evaluate the actual
cause-and-effect relationship of variables, such as load, speed, and product parame-
ters, as changes take place. This unique ability permits the analyst to isolate specic
changes in the machines or systems operating envelope that directly affect the oper-
ating condition, reliability, and useful life of the machine or system.
The diagnostic logic used for RTA is based on the same concepts as traditional time-
and frequency-domain vibration analysis. However, it provides the ability to view the
interactions of the machines vibration components dynamically rather than statically.
This greatly improves analysis accuracy and the ability to isolate the true root cause
of problems. As illustrated in Figure 23.1, real-time analyzers process and display
multiple data formats in real time.
A real-time vibration analyzer is essentially a conventional microprocessor-based
vibration analyzer with expanded capability. These analyzers acquire a time-domain
data signal and convert it to frequency-domain by performing a fast Fourier transform
(FFT) on the data. Both conventional and real-time instruments have lters and sig-
225
16.Mobley.23 Page 226 Friday, February 5, 1999 11:55 AM
226 VibrationFundamentals
Figure 23.1 Real-time analyzers process and display multiple data formats.
nal-conditioning logic that convert the mechanical forces within the machine-train
into vibration data that are displayed in units, such as inches per second peak (i.e.,
velocity data), mils peak-to-peak (i.e., displacement data), or gs (i.e., acceleration
data). Most of these instruments provide easy data acquisition and display of fre-
quency-domain vibration signatures.
The major difference between real-time analyzers and routine vibration-monitoring
instruments is processing speed. A real-time analyzer can evaluate complex vibrations
and noise hundreds of times faster than conventional analyzers. These instruments
process the acquired raw signal in a fraction of the time that it would take a general-
purpose instrument.
Additional data ports allow real-time analyzers to acquire and process simultaneous
channels of data, which is a major advantage over conventional analyzers. Unlike
conventional vibration-monitoring instruments, real-time analyzers have multiple,
independent circuits that can process incoming data in parallel. They include synchro-
nizing logic to ensure that all channels of recorded data are real time and absolutely
parallel. The ability to evaluate multiple channels of data that are absolutely synchro-
nized in real time is a major advantage of this diagnostic technology. This capability
permits the analyst to evaluate the interactions of the entire machine-train, or the
machine-train and its installed system. Because the operating dynamics of a machine
are dependent on these interactions, this type of analysis (i.e., multichannel analysis)
is an invaluable diagnostic tool.
16.Mobley.23 Page 227 Friday, February 5, 1999 11:55 AM
Overview 227
LIMITATIONS OF MICROPROCESSOR-BASED DATA COLLECTORS
Although real-time microprocessor-based, vibration-monitoring instruments are less
limited in their functional capabilities than conventional instruments, they share some
limitations, but are designed to overcome others. Conventional microprocessor-based,
frequency-domain data collectors are designed to acquire only a single channel of
vibration data while real-time analyzers acquire multichannel data. As a result, con-
ventional units cannot provide all of the capabilities necessary for all diagnostic appli-
cations. This limitation is discussed later as are others, including signal-conditioning
methods (i.e., bandwidth limitations, spurious signal rejection logic, and antialiasing
logic), display types, data-processing speed, and near real-time data.
Single-Channel Data Limitations
For more complex machine-trains or in applications where speed, load, or other pro-
cess variables change frequently, the use of single-channel data collection has limited
value. For example, a variable-speed machine may alter speed during the data-acqui-
sition process. If this occurs, the single-channel data collector will either smear the
fundamental frequency or delete the blocks of data that contain the speed change.
Either of these actions will distort the vibration prole of the machine-train and
severely restrict the analysts ability to evaluate accurately the machines operating
condition.
Although there are many applications where single-channel, frequency-domain data
acquisition is insufcient, it is acceptable for relatively steady-state operating condi-
tions, such as those found with constant-speed motors, pumps, and other simple
machine-trains. In these cases, even though the machine-train data are acquired in
series (i.e., one point at a time with some time lag between readings), the total data set
provides a reasonably accurate prole of the machine-trains operating condition.
Signal-Conditioning Methods
All microprocessor-based data collectors contain signal-conditioning logic designed
to ensure accuracy of acquired data. While these lters and the logic used to condition
the input signal are valid for relatively steady-state machine-trains, they can distort
data from variable-speed/variable-load machines. In particular, three signal-condi-
tioning logic factors can cause problems: bandwidth limitations, spurious signal
rejection, and antialiasing logic.
BandwidthLimitations
The signal logic used to condition the incoming signal from a transducer limits the
bandwidths that can be used for data collection. While microprocessor-based instru-
ments permit a select range of maximum frequencies, F
MAX
, the minimum frequency
limit, F
MIN
, for signal conditioning is always zero. This limitation is not obvious on
some systems because, although the software permits the user to select a F
MIN
other
than zero, the input value is not used during the data-acquisition sequence.
16.Mobley.23 Page 228 Friday, February 5, 1999 11:55 AM
228 VibrationFundamentals
The disadvantage of a xed F
MIN
of zero is that it prevents acquisition of a high-reso-
lution vibration prole centered around a unique frequency band. This ability, called
zoom, is available in most real-time analyzers and is a necessary diagnostic tool for
complex machinery.
SpuriousSignalRejectionLogic
Signal-conditioning logic eliminates periodic vibration frequencies that do not repeat
within each block of data acquired by the microprocessor. This logic is used to elimi-
nate spurious frequencies, such as impacts, transients, or electronic noise, that might
distort the vibration signature. The disadvantage of such logic is that failure modes
due to such events cannot be detected when using this data-acquisition technique.
The microprocessors signal-rejection logic automatically evaluates each block of
data as it is acquired by directly comparing it to preceding blocks. If the rejection
logic detects frequencies in the new block of data that were not present in the preced-
ing block, it automatically rejects the new block. Because the microprocessor auto-
matically acquires an additional block of data until each of the acquired blocks
contains nearly identical frequencies, the technician may be unaware that this is
occurring. In extreme cases, the instrument will abort the data-acquisition process due
to radical, repeated uctuations in speed, load, or other variables that change either
the vibration frequencies or amplitude.
Antialiasing Logic
Most single-channel vibration instruments include an antialiasing lter in their signal-
conditioning logic. These lters are designed to prevent vibration signature distortion
caused by fold-over of both low- and high-frequency components. Fold-over is where
high frequencies invert or fold and appear as higher frequencies. While prevention
of distortion is important, the use of preset lters can limit the diagnostic capability of
these instruments.
Display Type
Conventional microprocessor-based systems used for vibration monitoring in tradi-
tional predictive maintenance programs acquire, condition, and store an averaged
snapshot of the machines vibration prole. This snapshot is later downloaded to a
desktop computer for analysis. This snapshot view severely limits the analysts ability
to understand the dynamics of the machine-train because the vibrations generated by
a machine are not static, but are dynamic.
Both the overall vibration energy and that of each frequency component are con-
stantly changing. The phenomenon, referred to as beats or beating, is indicative of
the machines operating condition and can provide a powerful tool for analysis.
However, data-processing speed, display logic, and diagnostic logic used in conven-
tional vibration-monitoring instruments prevent them from being used to monitor
this phenomenon.
16.Mobley.23 Page 229 Friday, February 5, 1999 11:55 AM
Overview 229
Real-time analyzers are required to take advantage of the data provided by the phe-
nomenon. These analyzers permit direct or indirect data acquisition and provide a
dynamic display of the time-domain waveform, frequency-domain signature, or both.
Dynamic data allow the analyst to better view and interpret the condition of the
machine-train. Dynamic displays show the interaction of the various frequencies that
make up a machines signature. This is especially important when two or more fre-
quencies couple, which can signicantly inuence the overall signature.
Data-Processing Speed
Most conventional data-acquisition instruments that are used for routine vibration
monitoring and analysis do not have the data-processing speed needed to perform
RTA.
17.Mobley.24 Page 230 Friday, February 5, 1999 11:58 AM
Chapter24
APPLICATIONS
There are many potential applications for real-time analysis, ranging from advanced
vibration analysis to structural and process analyses. However, the major applications
for this diagnostic technique are transient analysis, complex-machine analysis using
synchronous time averaging and narrowband zoom, multichannel analysis, and tor-
sional analysis.
TRANSIENT ANALYSIS
Many of the machines and process systems found in manufacturing and process
plants are periodically subjected to events that directly affect their operating dynam-
ics, reliability, and useful life. These events, called transients, may be part of the nor-
mal operating mode or they may be an external incident or variable not uniquely
associated with the machine or system. Specic applications that require transient
analysis include variable-speed machines, process or product variations, and random
or periodic impacts.
A transient is a short-duration event that periodically occurs in a machine or process
system. As an example, when a variable-speed machines running speed coincides
with its rotors critical speed, that event is considered to be a transient. In normal
practice, the duration of this type of event isor should berelatively short. Because
they generally occur so quickly, conventional vibration-monitoring instruments do not
have the processing speed required to capture the impact of the event and, therefore, a
real-time analyzer is required to perform the proper diagnostics.
Figure 24.1(a) illustrates a transient captured in a time waveform and Figure 24.1(b)
illustrates the same transient in a waterfall or cascade frequency-domain plot. These
capture-and-display capabilities account for much of the diagnostic power of RTA.
230
17.Mobley.24 Page 231 Friday, February 5, 1999 11:58 AM
Applications 231
A
B
Figure 24.1 (a) Time waveform of transient event. (b) Same transient in waterfall format.
Variable-Speed Machines
Paper machines and steam turbines are examples of variable-speed machines that
operate over a wide range of speeds and are excellent applications for RTA. Because
the dynamics of the machine and its components change during the transitional period
associated with any substantial speed change, the ability to measure and evaluate
these changes is critical to any reliability program.
17.Mobley.24 Page 232 Friday, February 5, 1999 11:58 AM
232 VibrationFundamentals
RTA permits accurate capture of real-time, multichannel vibration and process param-
eter data as the machine accelerates to, or decelerates from, full speed. Because these
variations are rapid, the increased data-processing speed of a real-time analyzer is
essential.
Process or Product Variations
Machines or process systems that are subject to a range of product- or process-related
variables are also ideal applications for RTA. Proper use of data-acquisition and anal-
ysis techniques permits the analyst to fully dene and quantify the cause-and-effect of
the variables on the operating dynamics, reliability, and useful life of the machine or
system component.
Random or Periodic Impacts
Many of the machine-trains and process systems that make up a plant are subjected to
apparently random or periodic impacts, or transients, that may affect their reliability
or useful life. Conventional vibration monitoring will not detect these events except
by accident.
The combination of digital tape recorders and RTA can be used to document and
evaluate these occurrences. Using a digital tape recorder to acquire vibration and
process data over a long period of time is a cost-effective way to capture those ran-
dom or long-interval periodic events that may affect a machines reliability. When
combined with the tremendous data-processing speed of a real-time analyzer, this
provides the analyst with the means to evaluate these events quickly to isolate their
root cause.
COMPLEX-MACHINE ANALYSIS
Not all machines or process systems within a plant are simple. Many are comprised of
complex components that all contribute to the vibration waveform or signature. As a
result, analysis of the vibration prole is often very difcult. For example, a triple-
reduction gear box, coupled to a variable-speed steam turbine and variable-pitch fan,
generates a very complex vibration prole. Each of the components in the driver,
gearbox, and fan contributes to the time waveform or frequency-domain signature.
Conventional vibration meters cannot separate the individual contributions of these
components from the composite prole.
Real-time analyzers can be used to break up the individual frequencies generated by
various machine or system components because of its data-processing speed and sig-
nal-conditioning logic. Two RTA techniques are especially useful for complex
machines: synchronous time averaging and narrowband zoom.
17.Mobley.24 Page 233 Friday, February 5, 1999 11:58 AM
Applications 233
Synchronous Time Averaging
When synchronous time averaging is coupled with the real-time analyzers ability to
provide parallel, multiple data channels and real-time processing, RTA can be used to
evaluate extremely complex vibration proles. This technique removes unwanted fre-
quencies that may be generated by other components either within the same machine,
from other machines, or from unknown outside sources. Simply stated, this technique
nullies or removes any vibration frequencies that are not absolutely synchronous to
the shaft or turning speed being investigated.
Narrowband Zoom
One of the limitations of conventional, general-purpose vibration instruments is their
inability to provide high-resolution signatures of specic frequency components (e.g.,
gear-mesh, vane-pass, or bearing modulation). Because these systems have a xed
lower data acquisition frequency limit (i.e., zero), they cannot provide resolution on
any set of frequencies that has a lower limit greater than zero. As a result, the analyst
may not be able to visually fully evaluate the prole generated by a specic machine
component.
Most real-time analyzers can provide a true high-resolution signature for any set of
frequency components within the machine vibration prole. This ability, called nar-
rowband zoom, is a powerful diagnostic tool. It is especially useful for complex
machines where multiple components may require close evaluation. As an example,
narrowband zoom can be used to radically increase the resolution of the gear-mesh
frequency of a gear set. The analyst can establish a narrowband-zoom window by
selecting the gear-mesh as the center frequency and dening a bandwidth that
includes the gears modulation frequencies. The resultant display provides a high-res-
olution signature that can be used to fully analyze the gears condition.
MULTICHANNEL ANALYSIS
Multichannel analysis is not unique to RTA. It can be performed with conventional,
general-purpose vibration instruments using a digital tape recorder and good disci-
pline. However, this approach is more difcult and does not provide the diagnostic
power of RTA.
Conventional vibration-monitoring systems are limited by their processing speed
and functional capabilities. Therefore, they cannot provide all of the tools needed
for full multichannel analysis. In addition, using them requires the raw data to be
acquired using multichannel tape recorders and then played back through the gen-
eral-purpose meter. As a result, playback timing becomes a real problem. There is
no guarantee that all data are time synchronized or that the displayed data are
meaningful.
17.Mobley.24 Page 234 Friday, February 5, 1999 11:58 AM
234 VibrationFundamentals
Use of a real-time analyzer, with or without a digital tape recorder, eliminates these prob-
lems. The multichannel, parallel-processing capabilities of the analyzer provide a quick,
positive means of retrieving and displaying data that are absolutely time synchronized.
TORSIONAL ANALYSIS
Torsional vibration of rotating elements is the rapid uctuation of angular shaft veloc-
ity, and its basic units are either radians or degrees. A machine will often increase or
decrease speed over some period of weeks, days, or seconds. As a machine changes
speed, torque is applied to the shaft in one direction or the other.
Torsional vibration is not a simple parameter to analyze. Transducer requirements are
stringent and shaft access may be limited. Above all, however, there is a peculiar mys-
tique engulng torsional vibration. Therefore, this module attempts to dispel its mys-
tique by providing a basic understanding of torsional motion, what it means, and how
it can be interpreted.
18.Mobley.25 Page 235 Friday, February 5, 1999 12:02 PM
Chapter25
DATAACQUISITION
This section provides the basic information needed to acquire accurate real-time data.
It assumes that the analyst or technician is familiar with microprocessor-based real-
time spectrum analyzers, digital tape recorders, and other appropriate instrumenta-
tion. The users manuals for the actual instruments to be used should be consulted in
conjunction with this training module.
Regardless of whether direct or taped data acquisition is selected, the approach used
to gather real-time data is the same as for single-channel (i.e., route) acquisition. The
same rules are used for measurement point location and orientation, analysis parame-
ter set selection, measurement point denition, etc. The only exception is that all data
are broadband, rather than both broadband and narrowband.
Before using a real-time analyzer as part of the periodic monitoring program, the
technician or analyst should review the instructions provided for data acquisition. All
of the rules and methods used in the routine monitoring program apply to real-time
data acquisition. In addition, he should thoroughly review the users manuals for all
other instruments to be used for data acquisition and analysis. However, recently pur-
chased real-time spectrum analyzers use a Microsoft Windows-based operating sys-
tem, which greatly simplies their use. Like a personal computer, all functions of the
analyzer can be accessed from the main menu using standard Windows protocol.
Input for all data elds on the acquisition setup must be included for all active chan-
nels before attempting data acquisition. Care must be taken to ensure that all data are
consistent. Unlike a single-channel system, a real-time analyzer provides exactly what
is requested. If errors or inconsistencies are made in the acquisition setup, it will per-
form the preprogrammed statistical or mathematical functions. It will not question
errors or inconsistent formats between the various data elds.
235
18.Mobley.25 Page 236 Friday, February 5, 1999 12:02 PM
236 VibrationFundamentals
For example, if the user selects acceleration data (e.g., 100 mV/g) as the calibration
factor and velocity units of inches per second as the engineering units name, the
real-time analyzer will acquire and display the vibration data as velocity readings
even though it has not integrated the acceleration data into velocity. As a result, the
displayed data will have no value as a diagnostic tool.
METHODS
There are two ways to acquire the data needed to perform a RTA: direct acquisition
and tape recording.
Direct Acquisition
In direct-acquisition mode, the real-time analyzer can be used to acquire a variety of
vibration and nonvibration, process system data, which are stored directly in its on-
board memory. Therefore, the primary limitations of direct acquisition are the on-
board memory capacity and the inexibility of the stored data.
The advantage of direct acquisition is that monitoring of machine-train or process
system operating conditions can occur as the data are acquired. This allows the ana-
lyst to adjust the data-acquisition parameters as needed to ensure accuracy. In applica-
tions where a quick diagnosis is needed, this approach provides a means of isolating
and solving simple problems.
The disadvantage is that it extends the time and manpower required. Unlike the pre-
programmed, microprocessor-based analyzers used for routine vibration monitoring,
real-time analyzers must be manually congured for the specic type of data before
each acquisition. For example, it can either be congured to acquire time waveforms,
frequency-domain signatures, high-resolution narrowbands, or a variety of others.
The analyzer acquires, conditions, and displays a continuous prole in the user-
selected format. If the analyst wants to look at a different data format, she must abort
the data acquisition and reset the instrument for the new data format. In addition, the
acquired time- and frequency-domain data are not time synchronized, but are taken in
series. Data taken in series eliminate the ability to compare the time trace with the fre-
quency-domain signature.
However, the biggest disadvantage of the real-time analyzer is that data, once cap-
tured, cannot be converted to a different format. For example, time traces cannot be
converted to frequency-domain data. This limitation can greatly restrict the diagnostic
capability of the analyzer or increase the analysis time where multiple data formats
are required for proper analysis. Because the user cannot view real-time data in more
than one format, he must reacquire it each time a different format is required. The
problem with this technique is that each data set is new. As a result, subsequent data-
acquisition runs may not duplicate transients or the operating condition of the
machine-train found in a previous run.
18.Mobley.25 Page 237 Friday, February 5, 1999 12:02 PM
237 DataAcquisition
Tape-Recorded Data
With this approach, the analyst uses a tape recorder to acquire the data, which is cap-
tured and stored on tape. This approach permits quicker acquisition of data that can be
analyzed quickly in parallel, or an in-depth analysis of the machine-train or process
condition can be performed at a later date.
TypesofRecorders
Two major types of tape recorders are used to acquire vibration and process parameter
data: analog and digital. Each type has advantages and disadvantages that should be
understood before using them for RTA. The major difference between the two types
of recorders is that, while they both take an analog signal as input, the digital recorder
incorporates an analog-to-digital converter. This is a device that translates continuous
analog signals into proportional discrete digital signals.
Analog Recorder
Analog signals are nominally continuous electrical signals that vary in amplitude or
frequency in response to changes in sound, light, heat, position, or pressure. Analog
recording is any method in which some characteristic of the recording signal, such as
amplitude or frequency, is continuously varied in a manner analogous to the time vari-
ations of the input signal. The two major types of analog tape recorders used to
acquire vibration and process parameter data are direct-record and frequency-modu-
lated units. The major difference between these devices is in their ability to record
low-frequency signals.
Direct-Record Tape Recorder
With direct-record analog units, the signal amplitude is captured directly by the tapes
magnetic eld. Therefore, variations in tape quality and ambient conditions (i.e., heat,
light, and stray magnetic elds) directly affect the data obtained with this type of
recorder. This type of device cannot record frequencies below 25 Hz, or 1500 rpm.
This is because playback is based on the rate of change of tape magnetization.
Frequency-Modulated Tape Recorder
With frequency-modulated analog units, the signal amplitude is recorded as the differ-
ence between a base or carrier frequency and the frequency recorded. As a result, the
frequency-modulated recorder is much less sensitive to variations in ambient condi-
tions and the magnetic properties of the tape used for data acquisition. Frequency-
modulated recording can be used with low frequencies down to the physical limits of
the transducer, signal conditioning, and cable that are used.
Digital Recorder
Digital recorders have the ability to condition and lter the raw input signal in much
the same way as single-channel vibration analyzers and multichannel real-time ana-
lyzers. In this type of tape recorder, the incoming signal is passed through an analog-
to-digital converter and stored in a digital medium as a series of digital values. Most
18.Mobley.25 Page 238 Friday, February 5, 1999 12:02 PM
238 VibrationFundamentals
of these instruments can lter the analog data to prevent aliasing and to condition the
output to user-selected values.
RecordingtheData
Because it is difcult to anticipate the exact formats and data that will be required to
resolve a machine-train or process problem, full-range tape recording of data is the
recommended practice. Storing the data on tape ensures that the raw data will be
available for complete, comprehensive analysis.
Data-AcquisitionPractices
Unlike vibration data that are collected with traditional microprocessor-based predic-
tive maintenance programs, real-time data collection does not use preprogrammed
acquisition routes. Therefore, the acquisition route for obtaining each data set must be
set up and performed manually. As a result, acquiring this type of data requires more
time, discipline, and expertise than for routine vibration monitoring.
The following sections discuss the practices that should be followed to ensure that
accurate, meaningful data are obtained. In particular, the following topics are dis-
cussed: hardware setup for transducers, cables, and power supplies; channel integrity;
test plans; and eld notes for channel data, transducer data, gain, and sequence of
events.
Hardware Setup
RTA is generally used in conjunction with multichannel data acquisition, which com-
plicates the hardware setup requirements. Therefore, the required hardware setup is
quite different than that used for routine vibration monitoring. This section discusses
the setup requirements for the transducers, cables, and power supplies that are needed.
Transducers
Transducers, which are used to obtain vibration or process data, must be selected with
care. In particular, they should be compatible with the specic measurement parame-
ters of an analysis. Generally, accelerometers should be used to acquire the vibration
data for a RTA. This type of transducer is better suited for most applications because
it is less sensitive to mechanical damage and temperature.
The accelerometers should be of the low-mass variety and have a positive means of
mounting to the machine-train (e.g., stud, epoxy, or magnet). In addition, they must
have the linear-response characteristics needed for the specic application. Each
accelerometer should have a certied specication sheet that denes its operating
range and response characteristics. It also should have a current calibration test.
Cables
Unlike general-purpose vibration monitoring, RTA typically requires massive cable
runs to connect the multiple channels to a digital or analog tape recorder, or directly to
the real-time analyzer. Both the number of cables and the average run length create
18.Mobley.25 Page 239 Friday, February 5, 1999 12:02 PM
239 DataAcquisition
unique problems with this type of analysis. Generally, two types of cable are used for
a RTA: microdot and coaxial.
Microdot cables are normally required to make the initial connection between a low-
mass accelerometer, power supply, and tape recorder or analyzer. The cable is a small-
diameter (i.e., about 1/16 in.) assembly that includes threaded connections. Because
of its size, microdot cable is extremely sensitive to misuse or physical damage. There-
fore, care must be taken to ensure that it is protected throughout the data-acquisition
sequence.
The use of microdot cable assemblies should be minimized as much as possible. In
addition to their sensitivity to damage, the resistance within the cable may distort the
electrical signal. Wherever possible, total microdot runs should be less than 5 ft.
Longer runs may cause attenuation or distortion of the signal.
Coaxial cables are used for the long runs that connect the transducer to either a tape
recorder or real-time analyzer. These cables have a larger diameter than microdot cable
and are almost immune to damage. They are similar to those used for cable television
connections and provide a reasonably reliable way to make critical connections.
Total runs between the transducer and recorder should not exceed 70 ft. Signal attenu-
ation beyond this distance has a severe effect on data quality. If longer runs are
required, a signal amplier can be added to each cable to boost the signal strength and
permit the longer run.
Power Supplies
All transducers require a power source to operate properly. In general-purpose vibra-
tion monitoring, the power source is usually part of the analyzer. In many real-time
applications, however, an external power supply must be provided for each acceler-
ometer or transducer.
The external power supply must be matched to the transducer. For example, most
accelerometers require a 4-mV power supply to function properly. In addition to their
compatible rating, power supplies must provide constant, reliable power throughout
the data-acquisition sequence. Because many of the power supplies that are normally
used in this type of application are battery powered, care must be taken to ensure that
fresh batteries are installed at the beginning of each data-acquisition sequence.
Many power supplies include an amplier, or gain, that can be used to increase the
raw signal strength of the transducer. While this ability is helpful with weak signals, it
can lead to serious diagnostic errors. Typically, the gains provided by power supplies
are in steps of 10, ranging from 0 to 100. For example, if the user selects an amplica-
tion factor of 100, the signal strength recorded by the analyzer will be 100 times
higher than the actual vibration energy.
18.Mobley.25 Page 240 Friday, February 5, 1999 12:02 PM
240 VibrationFundamentals
Channel Integrity
In all RTA applications, extreme care must be taken to ensure data accuracy. This is
especially true when the analysis is combined with multichannel data-collection tech-
niques. It is imperative for the analyst to be able to identify absolutely each of the
channels as data are acquired.
Permanently numbering components used for each data-acquisition channel is the
best assurance of this ability. Everything from the accelerometer to the nal connec-
tion on the coaxial cable should be numbered. Permanently afxed cable tags should
be on both ends of all cable assemblies, as well as other channel components.
The entire cable run for each channel should be inspected and veried prior to a data-
acquisition sequence. In addition, a continuity test should be conducted on each chan-
nel to ensure a distortion-free channel.
Test Plans
Applications that require RTA techniques are generally more complex than those that
are appropriate for traditional vibration monitoring and analysis. Typically, RTA is
used for complex applications, such as torsional problems, and a series of well-planned
data acquisitions and analyses are required. Therefore, a detailed test plan is essential.
The test plan should concisely dene the specic tests that will be performed. For
each of these tests, the plan should include the setup data that will be needed to install
and connect the transducers, power supplies, cables, and other instruments.
Field Notes
The analyst must document the exact events, timing, and data-acquisition methods
used to record the information for each data channel. Because analysis may take place
at some time after data acquisition, the analyst must have sufcient documentation to
fully understand exactly when, where, and how the data were recorded.
Many of the digital tape recorders provide a voice-over channel that permits direct
verbal commentary that can be played back during analysis. However, detailed, writ-
ten notes also are essential. Documentation should include the following: channel
data, transducer data, gain, and sequence of events.
Channel Data
The test log for each data set should clearly identify the location and orientation of
each transducer. This information should be veried during the data-acquisition
sequence to make sure that it is accurately recorded.
Transducer Data
The test log should include the specic type and setup of each transducer used for
data acquisition. As a minimum, the log should include model number, serial number,
and engineering conversion unit (i.e., 500 mV/g, 1000 mV/g, etc.).
18.Mobley.25 Page 241 Friday, February 5, 1999 12:02 PM
241 DataAcquisition
Gain
In most cases, an external power supply or signal-conditioning instrument is used in
conjunction with the transducers. Both the power supply and signal-conditioning
units have the capability, called gain, to increase the strength of the raw signal. For
example, a typical gain from a power supply is 10. When this setting is selected, the
raw signal strength is increased by a factor of 10.
The gain that is used must be recorded so the analyst can accurately evaluate signal
strength. If the analyst is unaware of the actual gain, she will believe that the signal
strength is 10 times higher than the actual value.
Sequence of Events
Included in the documentation needed to dene the data set should be a concise
description of the test, channels recorded, and the start-to-end timing of the data-
acquisition process. The information should include all known variables and any
assumptions that may have affected the data.
PARAMETERS
Most analyzers have up to eight channels that can be used for data acquisition. Each
of the active channels to be used for data input, processing, and display must be set up
manually at the beginning of each data-set analysis. Therefore, extreme care must be
taken to ensure that all active channels are properly set up and that both the data-
acquisition and data-analysis parameters are consistent. The parameters required for
proper data acquisition include channel coupling, full-scale voltage, calibration factor,
engineering units name, and trigger group.
Channel Coupling
Coupling is selected on a channel-by-channel basis and denes how the input signal is
conditioned during the data-acquisition sequence. There are three choices for channel
coupling: alternating current (ac), direct current (dc), and internal power supply.
AlternatingCurrent
When the signal source is ac, the dc component is rejected and only the ac component
is acquired by the analyzer. When real-time vibration data are to be acquired, this is
the normal mode of signal conditioning.
This is not the case when the analyzer is used for direct acquisition of data. Selection
of the ac-coupling mode will not provide power to the accelerometers or other trans-
ducers used as part of the direct-acquisition mode of operation. Therefore, the ac-cou-
pling option should not be used for direct data acquisition unless external power
sources are used to drive the transducers.
18.Mobley.25 Page 242 Friday, February 5, 1999 12:02 PM
242 VibrationFundamentals
DirectCurrent
When the dc-coupling mode is selected, both the ac and dc components of the
machines vibration prole are acquired by the analyzer. In most cases, the dc compo-
nent is comprised of electronic noise that distorts the vibration prole acquired from
the machine-train. When a real-time analyzer is being used purely as a vibration ana-
lyzer, this option should not be selected.
Internal Power Supply
Many real-time analyzers have an internal power supply. Unless an external power
source is used, this option should be selected for all direct data-acquisition applica-
tions. It provides a 4-mA/4-V dc power source that can be used to power a compatible
accelerometer or other transducer.
This option should not be used when tape-recorded data are transferred into the ana-
lyzer. Transferring taped data to an analyzer requires an ac coupling.
Full-Scale Voltage
Unlike single-channel, microprocessor-based vibration analyzers, real-time analyzers
do not automatically autoscale the input vibration signal to establish the maximum
signal amplitude. Therefore, the user must select a maximum input voltage before
acquiring data. The full-scale (FS) voltage option presets the maximum vibration
level to be recorded by the analyzer.
The full-scale value, which is usually expressed as root mean square (RMS) must be
selected on a channel-by-channel basis. Care must be exercised to ensure that selec-
tion for all channels is completed before acquiring data.
Most analyzers permit selection of an amplitude scale between 1 mV and 20 V set in
increments of 1, 2, 5, or 10 mV. This range is more than adequate for most applica-
tions, but care must be taken to ensure that the input signal is not amplied above the
FS voltage.
Care must be taken when selecting the FS RMS. Too low a value will clip the fre-
quency components and not provide a true indication of the total amplitude of individ-
ual components or the overall, or broadband, energy represented by the data point.
Loss of the actual amplitudes prevents proper analysis of the data and, hence, the
machine-trains condition.
Most analyzers autoscale function will not override the FS RMS scale selection in
either the data-acquisition or analysis mode. When data are clipped by a low FS RMS
selection, it cannot be recovered.
If the FS RMS scale is too high, it may exceed the analyzers dynamic range. In this
instance, the amplitude of the major frequency components is displayed, but the lower
18.Mobley.25 Page 243 Friday, February 5, 1999 12:02 PM
243 DataAcquisition
level frequency components may be lost in the noise oor. While most analyzers have
a good dynamic range, the potential for masking important frequency components is
high when the maximum FS RMS (20 volts) is selected.
Calibration Factor
The calibration factor is used by the real-time analyzer to convert channel voltage to
the more convenient engineering units (EU). This option is used to convert the raw
voltage reading (in millivolts) into more usable units of measurement, such as those
for velocity, acceleration, or displacement.
The user must enter the appropriate calibration factor for the accelerometer, velocity
transducer, or displacement transducer used to collect data. This conversion factor
must be entered for both direct or tape-recorded data. In most cases, the conversion
factor will be 100 mV/g for a general-purpose accelerometer, or 500 mV/g for a low-
frequency accelerometer. However, the user must dene the actual response charac-
teristics of the transducer used in each application.
Vendors generally include certication curves and specication sheets for transduc-
ers, including accelerometers. This documentation, which should have been retained
upon purchase, provides both the conversion factor and the response characteristics of
the transducer. This information is required to perform a RTA.
Engineering Units Name
In routine vibration-monitoring equipment, the preprogrammed measurement routes
include a conversion factor from raw input voltage (in millivolts) to a user-selected
value, such as velocity, peak, or mils peak-to-peak, and do not require this parameter
to be input.
Most real-time analyzers, however, do not offer this automatic conversion. The engi-
neering unit (EU) name setup parameter identies by name the type of unit (i.e., psi,
mils, speed, etc.) that is needed for each data channel.
The EU name can be set using the standard keyboard in the same manner as the cali-
bration factor. The analyzer will accept any string from one to six characters in length,
but the units should be the same as for the calibration factor. For example, an acceler-
ometer with 100 mV/g response should have an EU name of gs or accel. Consis-
tency between the calibration factor and EU name will prevent confusion and improve
diagnostic accuracy.
Trigger Group
Many of the diagnostic techniques used in RTA rely on the ability to synchronize the
event under investigation to some internal or external event. The trigger group setup
parameter is used to dene the specic event or variable that starts, or triggers, the
18.Mobley.25 Page 244 Friday, February 5, 1999 12:02 PM
244 VibrationFundamentals
data-acquisition sequence. Triggers, such as a once-per-revolution input from a
tachometer, preselected time interval, or a variety of other sources, may be used to
start the data-acquisition sequence.
When using an internal or external event to trigger data acquisition, the analyzer does
not begin processing data until that event occurs. At that time, the analyzer acquires
either a single block of data, or the requested number of samples (i.e., blocks) are col-
lected.
The user can set parameters to perform sampling either (1) on the rst trigger only or
(2) on every trigger received. The user also can specify the characteristics of the trig-
ger such as a signal coming from an external tachometer input or an analog signal
coming from one of the channels. The user can control how soon data will be col-
lected before or after the trigger occurs. The following information is required to set
the trigger and data-acquisition characteristics: source, slope, threshold, and source-
channel delay.
Source
The trigger source is selected using the source-combo box. There are four source
options: free run, external, reference channel, and repetitive.
Free Run
In the free-run mode, data are acquired constantly at the maximum rate allowable
given the number of tasks being processed by the program, which permits multitask-
ing or running more than one application at a time. The number of applications
directly affects the speed of data acquisition and display. For fast transient applica-
tions, the number of simultaneous applications should be kept to the minimum
required to complete the process.
External
With external triggering, data are acquired time-relative to a TTL input signal on the
dedicated trigger channel. The real-time analyzer has a dedicated channel for condi-
tioned TTL tachometer input. This channel is in addition to the two to eight channels
available for data acquisition. As an example, the Scientic Atlanta SC390 unit must
have a TTL input to trigger data acquisition in the external trigger mode.
Reference or Internal Channel
With internal channel triggering, data are acquired relative to the input signal on the
specied reference channel. When operating in two-channel, 100-kHz mode, the rst
channel must be the reference. All other applications can use any channel as the refer-
ence or trigger channel.
Repetitive
The repetitive check box triggering option determines if the data will be collected
only on the rst trigger or on each successive trigger. This option can be used for
18.Mobley.25 Page 245 Friday, February 5, 1999 12:02 PM
245 DataAcquisition
either the external or reference channel source selection. It cannot be enabled for the
free-run option.
Slope
Slope denes the type of edge, either rising or falling, to be used for the trigger. Used
in conjunction with the trigger threshold, the slope eliminates ambiguity in the speci-
cation for analog signal triggering.
With a rising-edge slope, increasing voltage of the signal at the specied threshold
level serves as the trigger. With a falling-edge slope, decreasing voltage of the signal
at the specied threshold level serves as the trigger.
Threshold
The threshold is used to set the trigger point for analog signal triggering. It is usually
specied in terms of a percentage of the channels full-scale RMS value. Any signal
with the appropriate slope exceeding the threshold voltage will act as a trigger. The
threshold is usually an integer percentage in the range of 99%, adjustable in incre-
ments of 1%.
Source-Channel Delay
The source-channel delay text box option is used to set the delay of the sample count
between the trigger and the start of data collection. This delay setting applies to all
channels and is an integer not to exceed the extended recorder memory size option.
A positive delay causes post-triggering, whereby data acquisition is delayed for some
period after the trigger event. A negative delay results in pre-triggering or acquisition
at some selected interval before the anticipated trigger event.
19.Mobley.26 Page 246 Friday, February 5, 1999 12:04 PM
Chapter26
ANALYSISSETUP
In addition to the data-acquisition parameters discussed earlier, the analyst also must
establish the parameters that will be used to analyze the data. Care must be taken to
ensure compatibility between the acquisition and analysis setups.
Analysis mode can be used in conjunction with acquisition mode to view real-time
data during the data-acquisition sequence. In this way, the user can monitor the vibra-
tion characteristics of the machine-train in real time. In addition, the user can verify
the validity of data as they acquired it.
As with the acquisition mode, the RTA program requests specic inputs to dene the
user-selected analysis parameters used to condition and display the data. The menu-
driven template requires user inputs for the basic setup, as well as the display setup.
BASIC SETUP
This section describes the basic setup required for a microprocessor-based, real-time
analyzer. It must be completed each time a data set is evaluated or any time the active
parameters change. Setup includes the following parameters: active channels, refer-
ence channel(s), block size, overlap, process weighting, and average group.
Active Channels
The active channels check boxes are used to select which channels will be used for
data collection, conditioning, and display. Those channels not designated as active
will be ignored, thus freeing memory for use by active channels. There must be at
least one active channel at any given time, but the number is limited only by the ana-
lyzer hardware conguration. There are typically up to eight active channels.
246
19.Mobley.26 Page 247 Friday, February 5, 1999 12:04 PM
247 AnalysisSetup
The analyzer automatically resets the maximum frequency (F
MAX
) to 100 kHz (two-
channel operation) or 40 kHz (three- to eight-channel operation) when the user does
not specify a F
MAX
below these values.
Reference Channels
The reference channel selector only appears when cross spectrum is chosen as the
analysis method. The reference channel option is used to select which of the active
channels will be used as the reference channel for multichannel analyses, such as
transfer functions and cross-products. Note that only a channel already designated as
active can be used as a reference channel.
Block Size
The analyzer divides the continuous stream of data it collects into blocks to facili-
tate processing. The block size selection determines (1) lines of resolution when the
magnitude spectrum (FFT) option is selected or (2) sample size when the time
traces or compressed time options are selected. The block size options include the
following:
512 samples 200 lines
1024 samples 400 lines
2048 samples 800 lines
4096 samples 1600 lines
Overlap
The overlap parameter is used to determine the percentage of overlap that will be used
to speed up the data-acquisition and -processing time. As with the conventional sin-
gle-channel, data-acquisition system, overlap averaging truncates the acquisition of
one block of data and starts the acquisition of the next. Most analyzers permit the fol-
lowing overlap percentage selections: 0, 25, 50, 75, and 90.
Overlap averaging reduces the accuracy of acquired data and must be used with cau-
tion. Except in those cases where fast transients or other unique machine-train charac-
teristics require articial means of reducing the data acquisition and processing time,
overlap averaging should be avoided.
A logical approach is to reduce or eliminate averaging altogether. Acquiring a single
block or sample of data reduces the data-acquisition time to its minimum. In most
cases, this time interval is less than the best time required to acquire two or more
blocks using the maximum overlap sampling techniques. Eliminating averaging gen-
erally provides more accurate data.
19.Mobley.26 Page 248 Friday, February 5, 1999 12:04 PM
248 VibrationFundamentals
NoOverlap
When zero or no overlap is selected, the real-time analyzer always acquires complete
blocks of new data. The data trace update rate is the same as the block processing rate.
This rate is governed by the physical requirements that are internally driven by the
frequency range of the requested data.
25Percent
When 25% overlap is selected, the analyzer truncates data acquisition when 75% of
each block of new data is acquired. The last 25% of the previous sample is added to
the new sample before processing is begun. As a result, data accuracy may be reduced
by as much as 25% for each data set.
50Percent
When 50% overlap is selected, the analyzer adds the last 50% of the previous block to
a new 50%, or half-block, of data for each sample. When the required number of sam-
ples is acquired and processed, the analyzer averages the data set. Accuracy may be
reduced by 50%.
75Percent
When 75% overlap is selected, each block of data is limited to 25% new data and the
last 75% of the previous block. At 75% overlap, there is a potential for distortion of
data.
90Percent
When 90% overlap is selected, each block contains 10% new data and the last 90% of
the previous block. Accuracy of average data using 90% overlap is highly question-
able because each block used to create the average contains only 10% actual data and
90% of one or more blocks that was extrapolated from a 10% sample.
Process Weighting
The process weighting option controls the type of weighting function to be applied in
performing FFT on blocks of time data. Weighting typically is used to reduce sources
of analytical error.
The length of the signal represented by the block of time data may not be an integral
multiple of the signals period. Because the FFT is meaningful only on periodic sig-
nals, this means that the resultant FFT may not accurately represent the actual fre-
quency of the machine-train. In this case, weighting can be used to modify the time
block to articially produce periodicity so that the resultant spectrum is much closer
to the actual signature generated by the machine-train. Weighting options include the
following: rectangular, Hanning, at-top, and response.
19.Mobley.26 Page 249 Friday, February 5, 1999 12:04 PM
249 AnalysisSetup
RectangularWeightingOption
The rectangular option does not weight the input signal. The values displayed by the
real-time analyzer are identical to the raw signal generated by the transducer. With
this option, raw time-waveform data are converted directly into a frequency-domain
signature through FFT.
HanningWeightingOption
The Hanning weighting option provides best capture of the individual frequency com-
ponents that make up a signature. However, it may distort the actual amplitude of the
frequency components. This weighting factor is normally used for magnitude spectra
or normal FFT analysis.
Flat-TopWeightingOption
Flat-top weighting provides the best representation of the actual amplitude of each fre-
quency component within a FFT. However, it may distort the actual location (i.e., fre-
quency) of each and, therefore, is not normally used for magnitude spectra analysis.
Flat-top weighting is useful with cascade, or waterfall, analysis. Its conversion meth-
odology modies the prole of each frequency component so the true amplitude is
displayed. Even though the actual location (i.e., frequency) of each component may
be slightly out of position, the prole is more visible when closely packed in a water-
fall, or cascade, display. The primary advantage of at-top weighting is that it pro-
vides the ability to see the dominant frequencies in a waterfall format.
ResponseWeightingOption
Response is a special weighting factor that should not be used for vibration analysis.
This method of weighting acquired data is limited to those applications where the
impact response characteristics of various materials or structures are the subject of
analysis. This methodology provides the ability to dampen signals caused by postim-
pact ringing of the object being tested.
Average Group
As with other types of vibration analyzers, most real-time analyzers provide the abil-
ity to average multiple blocks of data to derive a display of time-waveform or fre-
quency-domain signatures that best represent the vibration generated by a machine.
The averager is used to set the analyzers ensemble averaging parameters. Its setup
template requires user selection of the following options: domain, method, stop crite-
rion, and stop time/count.
DomainOption
The domain information box identies which averaging method is activated. The
domain is set by the analysis mode and the averager partially determines the type of
19.Mobley.26 Page 250 Friday, February 5, 1999 12:04 PM
250 VibrationFundamentals
analysis and display that can be produced. (See Data Sources in the Display dialog
box in the Help menu for more information.)
Time Domain Option
With the time domain setting, the averager is congured for averaging time-domain
data. This setting must be used for synchronous spectrum displays and is selected by
enabling the sync spectrum analysis mode. Time domain must also be used for com-
pressed and normal time trace displays.
Spectrum Option
The averager is congured for power spectrum domain data by enabling the spectrum
analysis mode. When this option is selected, the analyzer automatically captures the
selected blocks of data, averages their power content, and displays a frequency-
domain signature.
Cross-Properties Option
The averager is congured for power spectrum and cross-properties (i.e., transfer
functions, cross-power spectrum, etc.) data by enabling the cross spectrum analysis
mode.
None Option
When rapid transients, impacts, and a variety of other events are suspected, the use of
averaging can distort data and prevent proper analysis. As a result, most real-time ana-
lyzers have the ability to acquire single blocks of data. By selecting the None option,
the averager is disabled and the analyzer does not perform any averaging.
Method
Ensemble averaging is analogous to traditional number averaging. It combines a
sequence of traces to produce a single trace that is an average of the sequence. It is
much like the averaging used in single-channel meters. Averaging reduces or elimi-
nates spurious signals or distortion caused by outside inuences or nonrecurring
events. The use of averaging techniques in RTA has the same advantages and disad-
vantages as in single-channel systems.
Averaging can be used to smooth out variations in the signal spectrum to determine
typical or average signal behavior. In a similar manner, time-domain averaging also
can help eliminate the effects of random signal noise. During time-domain averaging,
the meaningful signal component tends to reinforce itself while the noise component
tends to cancel itself.
The disadvantage of averaging is that transients that may be useful in isolating a prob-
lem are likely to be eliminated. The real-time analyzer, like the single-channel ana-
lyzer, rejects or averages out nonrepetitive data components, even those that may be
key to the diagnostic being performed. Therefore, the use of averaging techniques for
RTA should be limited to those applications where averaged data permits separation
19.Mobley.26 Page 251 Friday, February 5, 1999 12:04 PM
251 AnalysisSetup
of extraneous noise or inuences from the data set. The real-time analyzer supports
three types of averaging: linear, exponential, and peak hold.
Linear Averaging
With linear averaging, each trace contributes equally. The advantage is that it is fast to
compute, but is suitable only for shorter trace sequences. It is not effective for longer
sequences because the average tends to stabilize due to the analyzers nite resolution.
Exponential Averaging
With exponential averaging, traces do not contribute equally to the average. New
traces are weighted more heavily than old ones. The advantage is that it can be used
indenitely. Because the average does not converge to some value and then remain
relatively constant, this averaging method is not limited by sequence size or duration.
The average dynamically responds to the inuence of new traces and gradually
ignores the effects of older traces.
If averaging is required, exponential averaging should be used for long-duration tran-
sients (e.g., complete coil run in a cold-reduction mill). Linear averaging should be
used for short-duration transients (e.g., spindle acceleration).
Peak Hold Averaging
Peak hold is not a true averaging method. Instead, the average produced by this
method is the highest value recorded in a select number of data blocks. Peak hold is
useful for maintaining a record of the highest values attained at each point throughout
the sequence of traces. The ability to capture and display these maximum values is a
useful diagnostic tool that can be used to improve the accuracy of analysis.
StopCriterion
The stop criterion combo box is used to select whether the averager stops on the basis
of elapsed time (i.e., time), number of samples collected (i.e., count), or if a single
average is performed (i.e., single).
Time
With a time stop criterion, averaging stops after the specied number of seconds
have elapsed. This is the preferred criterion where the interval or duration of the
acquisition is known (e.g., taped data).
Count
With a count stop criterion, averaging stops after the selected number of samples
has been collected. This is the preferred method when the time data are continuous or
when the measurement size is dened.
Single
With the single stop criterion, averaging is performed manually on a trace-by-trace
basis. This is the preferred method for interactively selecting ensembles based on
19.Mobley.26 Page 252 Friday, February 5, 1999 12:04 PM
252 VibrationFundamentals
more complex criteria. Single averaging must be used with the single trigger acquisi-
tion option.
StopTime/Count
The stop time/count scroll bar is used to set the number of seconds or ensembles that
will terminate the acquisition and averaging of data. The range that can be selected for
time is 1 to 86,400 sec (24 hr). The user can scroll upward or downward in increments
of 1 sec.
The range that can be selected for count is 1 to 10,000 ensembles. The up-arrow and
down-arrow icons permit scrolling in either direction in increments of 1 ensemble.
DISPLAY SETUP
The nal setup function required to use a microprocessor-based, real-time analyzer
for analysis is the display format. The display group is used to set up the type and data
source for the analyzer displays.
The Windows-based operating system used by most of these instruments permits mul-
tiple simultaneous displays in any combination of formats. However, the number of
active displays has a direct effect on the speed of both data acquisition and display
update. Therefore, caution should be used to limit the number of active displays used
in order to limit the reduction of real-time functionality.
The built-in screen on the real-time analyzer is somewhat limited when multiple dis-
plays are activated. Where the full-display functionality of the analyzer is needed for
analysis, it is better to use an external monitor instead of the built-in screen.
Display Type
The display type option allows the user to select which type of trace to display from
data currently in the analyzers memory. A pop-up menu, which is activated by click-
ing on the type eld, provides the display options that are available. Only valid types
are listed and these vary depending on averager domain and source.
AveragerDomain/StoreContents
Averager domain is specied in the average group selected by the user in the analysis
setup. Store contents is dened by the current status of store memory.
Source
The source button group is used to determine if the trace data are live, come from the
averager, or come from the storage memory. If a particular source is invalid for the
selected display type, the corresponding button will be grayed out and will not permit
the user to select it.
19.Mobley.26 Page 253 Friday, February 5, 1999 12:04 PM
253 AnalysisSetup
X-Axis Display
The analyzers X-axis display controls the annotation and scaling of the displays X-
axis. However, it is somewhat limited in the options that are available to the user.
Unlike the single-channel system, the real-time analyzer does not permit orders (i.e.,
order tracking) as an optional X-axis value. However, certain spectrum analyzers (e.g.,
Scientic Atlanta SD390) provide an optional signal ratio analysis (SRA) module that
permits order tracking. Because the ability to display spectral FFT data in orders or
multiples of running speed is a useful diagnostic tool, users of the real-time analyzer
may elect to add the SRA option to their existing analyzers.
With the exception of time-domain traces, orbits, Nyquist, and the correlation func-
tions, the real-time analyzer permits X-axis displays in either Hertz (Hz) or cycles per
minute (cpm). The user may select either of these units or change the units at any time
to aid diagnostics.
Time-domain displays, both normal and compressed time, must be displayed in sec-
onds. Orbit displays are limited to raw voltage (V) or engineering units (EU). Nyquist
functions must be displayed in either volts/volts or engineering units/engineering
units. Correlations, both auto and cross, are limited to seconds.
The scaling button group determines the axis scale. The user must select either linear
or logarithmic (base 10). If one of the scaling types is invalid for the selected display
type, it defaults to the valid scaling type and the control corresponding to the invalid
type is grayed to make it unavailable as a selection.
Most domestic users will elect to use the linear scale. Single-channel systems nor-
mally default to the linear scale, so users tend to be more comfortable with this type
of display.
Y-Axis Display
The Y-axis display options permit the user to select the vertical, or amplitude, scale
and display type. Help messages and users manuals describe the authorized display
types for each of the data types that can be selected.
While the analyzer is somewhat more exible in the options available for the Y-axis
display, this is not universal. The user should review the options provided in the docu-
mentation to ensure proper selection. In addition, care must be taken to ensure that the
data acquisition and analysis setups agree with the units and scaling selected for the
display. The analyzer may display data in formats that are not technically correct
when there is a conict between the various user-selected setup options.
Units
The units option allows the user to select the units for the displays Y-axis. Selection
depends on the display type and the axis scaling. The analyzer will default to normal
19.Mobley.26 Page 254 Friday, February 5, 1999 12:04 PM
254 VibrationFundamentals
units for each data type unless another option is selected. Note that the analyzer will
not convert acquired units into other terms unless the unit types option is initialized. If
the acquisition setup established that the data conversion was 100 mV/g, the analyzer
will retain that conversion factor and display data in either gs or EUs.
UnitType
The unit type option is available only for magnitude spectrum (FFTs) and provides
the ability to automatically convert acquired data into other unit types. For example,
data acquired as acceleration units (i.e., gs) can be integrated into velocity units (i.e.,
in./sec) and double integrated into displacement units (i.e., mils). The user may select
the appropriate conversion method from the options displayed in the pop-up window
adjacent to the unit type display. These options are normal, integrate, and differenti-
ate. With the normal option, no change is made to the acquired data. If the integrate
option is selected, the acquired data are integrated, but the user must select the resul-
tant units (i.e., gs to in./sec, gs to mm/sec). The differentiate option converts data to
the desired terms, but the user must select the resultant units (i.e., in./sec to gs, mm/
sec to gs).
A real-time analyzer is not as exible with units as a single-channel system. While the
single-channel system permits selection of RMS, peak, or peak-to-peak scaling fac-
tors for each unit type, the real-time analyzer is limited to its default values. Most of
the unit types are limited to RMS values, which can result in confusion. Many of the
vibration-severity charts use a combination of unit qualiers (RMS, peak, etc.) and
the limited display capabilities of the real-time analyzer may result in the inability to
compare displayed data to vibration severity charts.
Scaling
While the user may select either linear or logarithmic scaling, in most cases, the linear
selection should be used for clarity. In addition, the user should always select the
same vertical and horizontal scale. Mixing linear and log scales may distort the vibra-
tion proles and cause misdiagnosis of the machines condition.
20.Mobley.27 Page 255 Friday, February 5, 1999 12:06 PM
Chapter27
TRANSIENT(WATERFALL)ANALYSIS
RTA is ideally suited to both data capture and analysis of the short-duration events, or
transients, that directly or indirectly affect machine reliability. Transient analysis (TA) is
a common application of RTA. The most popular method of TA, waterfall analysis, cap-
tures a series of closely timed signatures to evaluate the transient. A waterfall, or cascade,
plot like the one in Figure 27.1 is simply a display of multiple frequency-domain vibra-
tion signatures. In RTA, each signature is acquired and displayed as the event happens.
Pseudo-waterfall analysis can be conducted with some general-purpose vibration
instruments. However, the real-time analyzers combination of parallel multichannel
signals and fast data-processing time makes it an excellent diagnostic tool.
SETUP
Most real-time analyzers provide the ability to capture and display a variety of water-
fall or multiple-spectra plots that facilitate transient diagnostics. The setup procedures
for this type of analysis may vary, depending on the type of instrument used. How-
ever, the following information is usually required by all: input group, load-control
group, load method, and number of records.
Input Group
The input group setting denes which of the recorded data channels will be used for
analysis. Typically, a real-time analyzer has between two and eight channels that can
be used to record, condition, and display waterfall data.
ActiveChannels
Active channels are used for waterfall data collection and analysis. Those channels not
designated as active are ignored. It is important to use only those channels actually
255
20.Mobley.27 Page 256 Friday, February 5, 1999 12:06 PM
256 VibrationFundamentals
Figure 27.1 Multiple frequency-domain vibration signatures.
required for analysis, because the number of active channels directly affects the speed of
data acquisition and processing, as well as the extended memory required.
ReferenceChannels
A reference channel is needed for multichannel analyses, such as transfer functions. Only
a channel already designated as an analysis reference channel can be used as the reference
channel in waterfall analysis. To add additional reference channels in the waterfall mode,
reset the acquisition and analysis setup templates to include the additional channels.
FunctionGroup
The function group setting allows selection of the waterfall analysis function that will
be loaded into waterfall memory.
Mode
Mode determines which types of analysis functions are available to be waterfalls. This
parameter cannot be changed on this template.
Function
Function is a pull-down box displaying the choices of functions that can be loaded
into waterfall memory for the selected mode. Only one function can be selected.
Source
The source setting denes the source of data to be included in the waterfall. One
option is live or real-time data, which are either directly from the machine-train or
from prerecorded (taped) data. These data are single block and weighted using the
selections of prior setup steps. The second option is averager data. If selected, this
option provides averaged data for the waterfall display.
20.Mobley.27 Page 257 Friday, February 5, 1999 12:06 PM
257 Transient(Waterfall)Analysis
In most waterfall applications, real-time data are preferred. Because the method of
analysis is used predominantly for transient analysis, real-time, single-block data pro-
vides the best diagnostic capability.
Load Control
Load control options determine how data are transferred from the source-data area to
the waterfall memory. The choices reect the selected analyzer mode, data, source,
and acquisition choices that have already been established as part of the previous
setup functions. Options also include continuous, average recycle, delta time, percent
amplitude, and delta rpm.
Continuous
Continuous control has two different functions based on whether the data source is
live or averaged. Live data are transformed into the desired function and transferred to
the waterfall memory as fast as the analyzer can collect and process the data. The rate
of processing depends on the function, selected number of channels, and number of
active displays. Data are always block processed, but with very small overlap. Aver-
aged (AVG) data defaults to the averager control parameters (e.g., overlap, weighting,
etc.). When one block of data is processed according to the averager overlap factor,
the desired function is transferred to waterfall memory.
For most transient-capture applications, the averaged-data option is used for the
waterfall. The default values for live data are used to provide continuous data for the
waterfall memory and display update.
AverageRecycle
Average recycle load control is only available when the data source is from the aver-
ager. Each time the averager target count reaches the specied number, the data in the
averager are copied to the waterfall memory. Each copy produces one record for each
selected waterfall channel. The averager is cleared and the process begins again until
the requested number of records has been copied to the waterfall memory. Average
recycle forces the averager to use the linear average and count stop method. Because a
linear average is used, this method is not suitable for long-duration events and should
not be used except in fast-transient capture events.
DeltaTime
When delta T is selected, the time increment eld appears to allow the user to enter
the desired time increment (i.e., seconds) in the appropriate eld. When the selected
time increment elapses, data in the source memory are processed and transferred to
the waterfall memory. The granularity of the time increment is 10 msec.
PercentAmplitude
When the amplitude load setting is selected, an amplitude value in percent of full
scale can be entered. This determines when the data from the selected source are
20.Mobley.27 Page 258 Friday, February 5, 1999 12:06 PM
258 VibrationFundamentals
transformed and transferred to the waterfall memory. When the amplitude (must be
RMS value) of the rst selected waterfall channel from the selected waterfall source
exceeds the indicated percentage of full-scale amplitude, data from all channels are
processed and copied to the waterfall memory.
This option can be used to trigger FFT generation at selected energy levels. For exam-
ple, this technique could be used to capture repetitive occurrences of resonance or
other abnormally high vibration levels without processing all signatures that contain
normal energy levels.
DeltaRPM
The delta RPM option permits selection of specic rpm intervals where the data in the
selected source are processed and transferred to the waterfall memory. This method is
typically used to perform runup and coastdown transient analysis. For example, the
user could elect to capture a FFT for each X rpm increase during runup and each X
rpm decrease during coastdown.
Load Method
The load method (i.e., continuous or stop when full) determines if waterfall mem-
ory will be loaded continuously or if loading will stop when waterfall memory is full.
When loading is continuous, data wraps when waterfall memory is full. When this
happens, the rst FFT is dropped from the display and a new signature is added. This
process continues until the user manually stops the process by pressing the stop
waterfall load or hold button.
Number of Records
This parameter species the number of records per channel that will be collected
before loading stops when using the stop when full load method, or before data
wraps when using the continuous load method.
ANALYSIS
The diagnostic logic used to perform waterfall analysis is identical to that used in fre-
quency-domain analysis. The only true difference is in the quality and timing of the
data that are used.
In traditional frequency-domain analysis, data are acquired periodically over long
intervals (i.e., weeks or months) of time. A series of signatures is then displayed so
that a comparative analysis can be made.
RTA uses data that are acquired as the event happens. As a result, there is typically
much more data, but its total time span is relatively short. For example, a typical tran-
sient through a critical speed, or resonant, zone may last less than 15 sec. However,
the real-time data may contain 50 or more individual signatures.
20.Mobley.27 Page 259 Friday, February 5, 1999 12:06 PM
Chapter28
SYNCHRONOUSTIMEAVERAGING
The signal to be measured and analyzed can sometimes be corrupted by unwanted
contributions from noise, line hum, or other machines running nearby. When this
occurs, spectrum averaging can be used to smooth out the total signal, but the tech-
nique does not necessarily help with analyzing a particular machine.
Other techniques are required to extract only the signal harmonics related to the
desired rotating element from a composite measured signal.
SEPARATING SIGNAL FROM NOISE
Several techniques are available for cleaning up a signal surrounded by noise. With
most techniques, some type of spectrum averaging is used. However, for the most
part, the enhancement provided by standard averaging is limited. If the level of the
periodic component is below that of the surrounding noise, increasing the number of
spectrum averages will not reveal the buried signal. It will only make the adjacent
noise smoother.
Figure 28.1 illustrates a periodic signal that has been buried in noise. The display on
the left is the unaltered waveform. The periodic signal also was available as a refer-
ence waveform, but was not used in this rst measurement. The display on the right is
that of a standard averaged spectrum. In this case, 1000 averages were used and the
total signal power was averaged. In the standard averaged display, only the fundamen-
tal spectrum component at 350 Hz is visible. From this display, it is not clear if there
are any harmonics or their distribution.
With synchronous time averaging, blocks of time records are triggered from the
desired reference signal and averaging takes place in the time domain. Waveforms
259
20.Mobley.27 Page 260 Friday, February 5, 1999 12:06 PM
260 VibrationFundamentals
Figure 28.1 Unaltered time waveform (left) and standard averaged spectrum (right) after
1000 averages.
Figure 28.2 Sync averaged time waveform (left) and sync spectrum (right) after 10,000 averages.
that are synchronous with the reference tend to be reinforced frame after frame,
whereas those that are random or synchronous at a different rate are not reinforced
and average to zero. After sufcient averaging in the time domain, a single FFT is per-
formed and the result is referred to as a sync spectrum.
With this method, the signal-to-noise ratio (S/N) improves proportional to the square
root of the number of averages taken. Theoretically, all noise components could be
eliminated by averaging for a long period of time. The result of synchronous spectrum
averaging should be only a periodic spectrum or nothing at all. Because the reference
signal is available for the waveform shown in Figure 28.1, it can be used as an exter-
nal trigger input to the real-time analyzer and a Synchronous Average performed (see
Figure 28.2).
Two important factors should be remembered when performing a synchronous time
average:
1. A reference signal is crucial and must be mechanically linked to the physi-
cal shaft or other rotating element to be isolated. An external signal gener-
ator set, for example, at approximately the same frequency as the desired
rotating element turning speed may appear to momentarily offer some sig-
20.Mobley.27 Page 261 Friday, February 5, 1999 12:06 PM
261 SynchronousTimeAveraging
nal cleanup. However, since it is not phase locked with the machine, aver-
aging enhancement soon ceases and no long-term improvement results.
2. Averaging is done in the time domain and must be followed by a single
FFT. This is important because the FFT typically performs a power
average, including all coefcients that are present. However, with syn-
chronous averaging, we care only about terms that are phase locked
with the reference signal. Thus, we need to average in the time domain
by creating blocks of data that always begin (or end) when the tachome-
ter triggers. In this way, nonsynchronous, or random, events average to
zero due to bipolar cancellation. In nonsynchronous or standard spec-
trum averaging, all components (both synchronous and other periodic
and random data) are averaged. This smooths the spectrum, but does not
enhance the S/N.
APPLICATIONS
There are many applications of synchronous time averaging, but the more common
uses include (1) obtaining multiple-order reference, (2) eliminating beating responses,
and (3) recovering signals below the noise level.
Multiple-Order Reference
In some cases, the primary signal of interest may not be at the fundamental shaft-
speed frequency, but rather at some multiple of that frequency. For example, the ana-
lyst may want to focus on the condition of a rotating element by removing all vibra-
tion except for specic passing frequencies, such as the number of blades on a turbine
wheel, the number of impellers in a pump, gear ratio, etc.
If a multiple tachometer signal of the desired rate is available, it can be used as the
reference trigger input. By using multiples of the tachometer input, only the trigger
multiple and its harmonics will be preserved. An example of this is seen in Figure
28.3 below, where a 3 running speed signal was used as the trigger input to the time
synchronous averaging process. With this trigger, the time waveform and resulting
Figure 28.3 Sync averaged time waveform (left) and sync spectrum (right) using 3 as trigger.
20.Mobley.27 Page 262 Friday, February 5, 1999 12:06 PM
262 VibrationFundamentals
Figure 28.4 Long-term (top) and expanded (bottom) time recording of beat.
spectrum will contain only the third order of running speed and each of its harmonics
(3, 6, 9, etc.).
If the contribution of a particular component (e.g., 35-tooth gear) is of interest, it also
is feasible to multiply a known tachometer signal by a phase-locked ratio to recover
just that signal from a composite waveform. In this case, the higher harmonics may
not be of interest and, if they are, there may be a conict between spectrum resolution
and waveform stability. Remember that time synchronous averaging depends on the
stability of the systems generating the multiple signals, especially on the stability of
the recovered waveform. The more this waveform wanders in frequency or speed, the
more difcult the synchronous averaging process will be.
Eliminating Beating Responses
In some cases, two machines or elements operating near the same running speed can
create a phenomenon referred to as beating, which can confuse an otherwise
straightforward measurement. This is another case where having the tachometer refer-
ence for the specic rolling element can accurately separate out the desired compo-
nent. An example of this is seen in Figure 28.4, which shows long-term and expanded
time recordings of beat.
20.Mobley.27 Page 263 Friday, February 5, 1999 12:06 PM
263 SynchronousTimeAveraging
Figure 28.5 A 2400-cpm, 40-Hz track (left) and 3-D waterfall (right) over a 1-min interval.
Figure 28.6 Sync spectrum using 2400-cpm reference, 200 averages; beating component at
2376 cpm.
Figure 28.5 illustrates a 3-D waterfall and 2400-cpm, 40-Hz track over a 1-min inter-
val. Note that beating causes the spectrum level of the 2400-cpm, 40-Hz component
to vary from nearly 0 to more than 8.6 V.
However, as shown in Figure 28.6, the actual level of the 2400-cpm component, as
derived through sync averaging with respect to the actual 2400-cpm reference, is
4.174 V peak.
Recovering Signals Below the Noise Level
A benet of synchronous time averaging is that no theoretical limit exists for the
amount of signal cleanup that can be performed. It is one of the few signal-process-
ing techniques available where the S/N enhancement is proportional to the square root
of the number of time-averaged blocks processed. Thus, the longer the averaging
20.Mobley.27 Page 264 Friday, February 5, 1999 12:06 PM
264 VibrationFundamentals
Figure 28.7 Standard free-run spectrum analysis performed with 1000 spectrum averages.
Figure 28.8 Sync averaged spectrum of same signal shown in Figure 28.7.
takes place, theoretically, the cleaner the signal gets until the only signal left is the
trigger, or synchronizing frequency, and its harmonics.
The spectrum shown in Figure 28.7 is the result of 1000 averages of a free-running
signal input. The only apparent signal is a peak at 40 Hz. However, it is suspected that
there might be a signal contributing at approximately 37 times the nominal 40-Hz
component. To verify this, a synchronous time average was performed with a refer-
ence signal of 1466 Hz.
The time synchronous spectrum shown in Figure 28.8 was performed with 3400 time
averages followed by a single FFT. This technique makes it apparent that a clear sig-
nal exists at 1466 Hz. Also note that the amplitude of this 1466 Hz component is less
than half that of the amplitude of the 1466 Hz component in Figure 28.7. This indi-
cates that the desired signal is at least 6 dB below the level of the surrounding noise in
the original broadband spectrum.
21.Mobley.29 Page 265 Friday, February 5, 1999 12:40 PM
Chapter29
ZOOMANALYSIS
Zoom analysis provides the means to separate quickly machine-train components,
such as gear sets, from a complex vibration signature. The technique lets the user
select a specic range of vibration frequencies, which the real-time analyzer converts
to a high-resolution, narrowband signature. This capability is unique to real-time ana-
lyzers and is not available in general-purpose, single-channel vibration analyzers.
Real-time zoom analysis can be performed with no data gaps up to a range of 10 kHz
with most microprocessor-based, real-time analyzers. However, the center frequency
plus one-half of the selected frequency span cannot exceed 10 kHz.
Above this range, pseudo-real-time processing occurs, which means that data
required to perform the zoom transform are acquired until the extended recorded
memory of the analyzer is full. When this occurs, the acquired data are processed
before additional data are gathered. However, this may result in data gaps that can
adversely affect the accuracy of the zoomed spectra. The gaps will be proportional to
the time required to perform the zoom transform for each channel, which in some
cases can be between 5 and 10 sec.
When using the zoom mode, the extended recorder memory should be set to the max-
imum available to obtain the best zoom accuracy and resolution. Reducing the num-
ber of active channels and lines of resolution also increases the speed and minimizes
the data gaps.
FREQUENCY SPAN
The frequency span parameter allows the user to select the frequency span for spec-
trum (FFT-based) and octave (digital lter-based) acquisition and analysis. For spec-
trum analysis, the frequency span can be set to any frequency notch from 1 Hz to 100
265
21.Mobley.29 Page 266 Friday, February 5, 1999 12:40 PM
266 VibrationFundamentals
kHz (usually limited to two-channel operation only) or 1 Hz to 40 kHz (for three- to
eight-channel operation). In real-time zoom mode, the frequency span can be set to
any frequency notch from 5 Hz to 10 kHz, as long as the new frequency span is in the
range of zoom capabilities.
CENTER FREQUENCY
The center frequency setting is used to set the center frequency for zoom mode opera-
tion. The center frequency can be set to any value in the range up to 100 kHz (Fre-
quency span/2) for two-channel operation or 40 kHz (Frequency span/2) for three-
to eight-channel operation.
21.Mobley.29 Page 267 Friday, February 5, 1999 12:40 PM
Chapter30
TORSIONALANALYSIS
Torsional vibration is not a simple parameter to analyze because transducer require-
ments are stringent and shaft access may be limited. In addition, there is a peculiar
mystique engulng torsional vibration. This chapter attempts to clarify the process of
its analysis through experimental examples and descriptions of the basic fundamen-
tals of torsional motion and how it can be interpreted.
WHAT IS TORSIONAL VIBRATION?
Torsional vibration of a rotating element is the rapid uctuation of angular shaft
velocity. As a machine changes speed, torque is applied to the shaft in one direction or
the other. A machine often increases or decreases speed over some period: weeks,
days, or seconds. However, when the rotational speed of the machine uctuates dur-
ing one rotation of the shaft, it is considered torsional vibration. Because this type of
vibration involves angular motion, the basic units are either radians or degrees.
Figure 30.1 shows the end view of a shaft in a bearing with a position marker, called a
key-phasor. An angular reference grid that is marked in 10-degree divisions surrounds
the shaft. In this example, an operating speed of 0.1667 rpm is assumed. This is equiv-
alent to a rotational rate of one revolution in 6 min, or 1 degree/sec (true only if there is
no torsional vibration). If the shaft turns at a constant rate of 1 degree/sec, then the
angular velocity is constant. No torsional vibration can be present under this condition.
As an example of a shaft experiencing sinusoidal angular velocity changes, assume a
rotating shaft increases to a maximum turning rate of 1.06 degree/sec during the rst
10 sec of rotation. Also assume that it slows to a minimum rate of 0.94 degree/sec
during the next 10-sec period. Under this condition, this shaft experiences the tor-
sional vibration, frequency, and amplitude shown in Figure 30.2.
267
21.Mobley.29 Page 268 Friday, February 5, 1999 12:40 PM
268 VibrationFundamentals
Figure 30.1 End view with position marker of shaft in bearing.
Figure 30.2 Torsional vibration graph.
21.Mobley.29 Page 269 Friday, February 5, 1999 12:40 PM
269 TorsionalAnalysis
Figure 30.3 Hookes joint.
In this example, both shafts complete one rotation in 1 sec. If we could look at an rpm
readout for each shaft, we would see they are both turning at the same speed. The rst
shaft turns at a constant rate of 1 degree/sec. The second shaft turns at an average rate
of 1 degree/sec. The torsional vibration waveform (see Figure 30.2) goes through one
complete cycle every 20 sec. This is 18 cycles per revolution, which corresponds to a
frequency of 0.05 Hz or 3 rpm.
STANDARDS
The basic standard for analyzing torsional vibration is the universal joint (see Figure
30.3), which also is known as Hookes joint. This device produces a predictable value
of angular velocity, rpm
B
, for output shaft, B, referenced to a constant speed, rpm
A
, of
input shaft, A, and the angle, , between the centerlines of the two shafts. It may be
difcult to comprehend, but at the instant of time portrayed in Figure 30.3, shaft B is
turning at its maximum rate, which is faster than shaft A. After shaft A rotates 90
degrees, shaft B turns at its minimum rate, which is slower than shaft A. Two com-
plete maximum/minimum cycles of shaft B occur every 360-degree rotation of shaft
A. This means the torsional frequency is twice the rotating speed.
Figures 30.4 and 30.5 illustrate how the angular velocity vibration level of the output
shaft changes with input shaft speed, rpm
A
, and shaft angle, . Notice the linear rela-
tionship between angular velocity vibration level and shaft speed in Figure 30.4. In
Figure 30.5 the angular velocity vibration level increases exponentially as the shaft
angle, , increases at a linear rate. Angular velocity vibration levels are expressed in
units of degree/sec, peak.
Now that we have discussed changing shaft velocity, we will look at the rate of the
changes. The rate at which the shaft changes its angular velocity is the measure-
ment of angular acceleration (not normally used to express levels of torsional
vibration). Angular acceleration is harder to comprehend and produces very large
numbers. The results shown in Figures 30.6 and 30.7 are obtained by differentiat-
21.Mobley.29 Page 270 Friday, February 5, 1999 12:40 PM
270 VibrationFundamentals
Figure 30.4 Torsional vibration versus input shaft speed.
Figure 30.5 Torsional vibration versus shaft angle.
21.Mobley.29 Page 271 Friday, February 5, 1999 12:40 PM
271 TorsionalAnalysis
Figure 30.6 Differentiated data from Figure 30.4.
Figure 30.7 Differentiated data from Figure 30.5.
21.Mobley.29 Page 272 Friday, February 5, 1999 12:40 PM
272 VibrationFundamentals
Figure 30.8 Angular displacement versus input shaft speed.
ing the data in Figures 30.4 and 30.5. Angular acceleration is displayed in units of
degree per second squared, peak.
The most commonly used parameter for expressing torsional motion is angular dis-
placement, whose units are degrees, peak-to-peak. There are several reasons to
express torsional motion in terms of angular displacement:
Has a small numerical value.
Tends to remain constant during speed deviations.
Is easier to visualize motion.
Figures 30.8 and 30.9 show the angular displacement values produced by the Hookes
joint relative to input shaft speed and U-joint angle. It is obvious from the data that the
torsional vibration is completely independent of input shaft speed. The torsional
vibration amplitude of induced angular displacement depends solely on the U-joint
angle, . Even at angles up to 30 degrees, the vibration level is always in single-digit
quantities.
DETERMINING TORSIONAL MOTION
Determining the torsional response of shafts or other components requires a positive
means of measuring or calculating the movement of two reference points, one on each
21.Mobley.29 Page 273 Friday, February 5, 1999 12:40 PM
273 TorsionalAnalysis
Figure 30.9 Angular displacement versus shaft angle.
end of the shaft. The following methods are used to measure the motion: optical
encoder, gear teeth, charting and graphic art tape.
Optical Encoder
Torsional motion is a deviation in shaft speed during one revolution, which must be
sensed instantaneously in order to be detected. This requires a signal of many pulses
per revolution (ppr). The best way to measure abrupt changes in shaft velocity is with
an optical shaft encoder, a device that consists of a spinning disk with very accurate
markings.
The encoder normally connects directly to the end of a shaft. As the disk spins, the
marks produce a pulse output each time they pass a photocell. The number of pulses
per revolution depends on the application. The following are the main factors to con-
sider with these devices:
Machine speed. Optical encoders have frequency limitations and, for a
specied pulse rate, there is a maximum turning speed.
Torsional component frequency. This is a matter of resolution and, when
selecting the proper encoder, the Nyquist sampling rate must be used. This
problem must be dealt with any time digital sampling of analog data is
undertaken. The Nyquist sampling theorem states: Data must be sampled at
21.Mobley.29 Page 274 Friday, February 5, 1999 12:40 PM
274 VibrationFundamentals
a rate greater than two times the highest frequency content of the data being
sampled. This keeps the data free of extraneous aliasing terms.
Torsional phase. This measurement requires two signals, one from each end
of the shaft. To measure static twist, signals having one pulse per revolution
are sufcient. When measuring the phase of dynamic frequencies, a multi-
ple pulse rate is required. The same sampling constraints for torsional com-
ponent frequencies apply when making phase measurements.
Gear Teeth
The use of gear teeth is sometimes the only method that can be used to detect tor-
sional motion. The advantages make this technique worth considering, but watch out
for the pitfalls.
Charting and Graphic Art Tape
Graphic art tape is a photo tape with very accurate black and white bars running
across it. It is available in many art supply, and some stationary stores. This tape can
be used with an optical sensor to detect shaft speed changes.
The major advantage of the tape is that it can be wrapped around a shaft that lacks
exposed ends. A serious drawback is the discontinuity point where the two ends of the
tape join. This introduces a torsional component at the shaft speed frequency along
with its associated harmonics.
Measured Versus Calculated Data
Measured data and calculated responses compare very well until the input shaft speed
reaches 1560 rpm, at which point the curves begin to deviate. Why does this deviation
occur? The Hookes joint data must be acquired at low input shaft speeds to be reli-
able because, as speeds increase and angles get greater, the mass of the system cannot
respond to the rapidly changing shaft velocities.
Measured Data
Figure 30.10 shows the data-acquisition arrangement from a Hookes joint test stand
in which a variable-speed dc motor drives input shaft A through a exible coupling
and optical encoder. Both the input and output shafts are 0.375 in. in diameter. Two
bearings support shaft A and eight ywheels, each weighing 1 lb. The ywheels aid in
maintaining a constant angular velocity for shaft A. The driving force for shaft B is
supplied by shaft A through the U-joint. Two bearings support shaft B. The angle, ,
between shafts A and B can be set to 0, 15, or 30 degrees. Shaft B drives a second
optical encoder used to detect the torsional motion introduced by the U-joint.
Both optical encoders receive power from a torsional converter with two independent
channels, each of which receives one of the 40-ppr optical encoder signals. The two
21.Mobley.29 Page 275 Friday, February 5, 1999 12:40 PM
275 TorsionalAnalysis
Figure 30.10 First analysis test stand.
Figure 30.11 Angular velocity of shaft B data.
21.Mobley.29 Page 276 Friday, February 5, 1999 12:40 PM
276 VibrationFundamentals
Figure 30.12 Calculated response versus measured data for 15-degree U-joint angle.
conditioned output signals are connected to a two-channel, real-time analyzer. The
torsional converter converts the 40-ppr signal from the shaft A encoder to a 1-ppr
signal to be used as a tachometer signal. The real-time analyzer also uses the 1-ppr
signal to normalize changing frequency components to the input shaft speed for
phase measurements.
Fifteen-Degree Angle
The following describes an example using the arrangement described in the preceding
section with a U-joint angle of 15 degrees. In this example, the analyzer stores the
data using peak-hold averaging. Figure 30.11 represents the angular velocity mea-
surement of shaft B.
Figure 30.12 is generated by picking discrete data points every 120 rpm and superim-
posing them on the calculated response displayed from Figure 30.12. Allowing for
analysis error due to lter width and weighting, plus the continuously changing speed
in the test, the 5% deviation exhibited is acceptable.
Thirty-Degree Angle
Notice that the amplitude of the angular velocity signal shown in Figure 30.13, which
represents a test having a U-joint angle of 30 degrees, has increased approximately
four times the level obtained with the 15-degree angle. This agrees with the calculated
response illustrated in Figure 30.14.
21.Mobley.29 Page 277 Friday, February 5, 1999 12:40 PM
277 TorsionalAnalysis
Figure 30.13 Increased angular velocity signal.
Figure 30.14 Calculated response versus measured data for 30-degree U-joint angle.
21.Mobley.29 Page 278 Friday, February 5, 1999 12:40 PM
278 VibrationFundamentals
TORSIONAL RESONANCE
Like any resonance, torsional resonance can cause fatigue, which in turn leads to shaft
cracking, coupling deterioration, gear failure, etc. Unfortunately, standard transducers
mounted on a machine do not respond to torsional vibrations.
To build torsional resonance into the example we have been using, we must add
another shaft and coupling (see Figure 30.15). Shaft C is now the mass of the resonant
system and the exible coupling connecting shafts B and C acts as a spring. An accel-
erometer is attached to the bearing that supports shaft B between the exible coupling
and the U-joint. This accelerometer detects any radial vibration of the bearing. The
torsional vibration is monitored and displayed as angular displacement. Figure 30.16
shows two data traces. The top trace shows the amplitude and frequency of the bear-
ings radial motion. The bottom trace shows the amplitude and frequency of the tor-
sional vibration. However, we are more interested in the frequencies present than their
amplitudes. The accelerometer output shows four structural resonances not reected
in the torsional data.
The important point is that the torsional resonance at 5760 rpm is not detected by the
accelerometer, therefore, it is not affecting the bearings. However, the amplitude of
9.56 degrees PK-PK is severe. Referring to Figure 30.8 you will nd the torsional
force applied by a U-joint at 15 degrees is 2 degrees PK-PK. This means that the tor-
sional resonance amplies the vibration level by a factor of 4.78 to 1.
MASS DAMPERS
The addition of mass dampers can help solve certain problems. However, adding
them changes the torsional resonant frequency, which can cause problems if the res-
onant frequencies are moved closer to operating speeds or other torsional forcing
functions. Keep in mind that, while the added mass changes the resonant frequency
of shaft C, it does not change the torsional vibration level produced by the U-joint.
Shaft C with the added mass only responds to lower excitation frequencies. The ex-
ible coupling between shafts B and C absorbs most of the energy from the torsional
input. This mass-damping principle is the technique used in torsional dampers and
harmonic balancers.
To illustrate the concept of mass dampers, we will add a 1-lb ywheel to shaft C. A
torsional displacement measurement can be used to gauge the impact of the use of the
ywheel as a damping mass. Figure 30.17 reects the angular displacement of shaft C
as the test accelerates from 180 to 6000 rpm. In this example, the added mass moves
the torsional resonance from 5760 rpm to 720 rpm, and increases its amplitude from
9.56 to 17.3 degrees PK-PK. Torsional vibration excitation forces from the U-joint are
still just 2 degrees PK-PK.
21.Mobley.29 Page 279 Friday, February 5, 1999 12:40 PM
279 TorsionalAnalysis
Figure 30.15 Second analysis test stand.
Figure 30.16 Bearings radial motion (top) and torsional vibration (bottom).
21.Mobley.29 Page 280 Friday, February 5, 1999 12:40 PM
280 VibrationFundamentals
Figure 30.17 Angular displacement of shaft C with a one-pound mass damper.
Note that the frequency of these forces is twice the frequency of the shaft speed. If the
machine operates above this frequency, the resonance will have no effect and the
angular displacement observed in Figure 30.17 (9.56 degrees PK-PK at 5760 rpm) is
reduced in amplitude to 0.0261 degrees PK-PK.
TORSIONAL PHASE
To sense the phase deviation from one end of a shaft to the other, it is necessary to
measure the angular position of the shaft at each end. In Figure 30.18, an optical
encoder is attached to both ends of a shaft.
The frequency of the encoders is 40 times the shaft speed and, for every 9 degrees the
shaft turns, the encoder puts out one pulse (360(/40 ppr). The phase relationship
between the two encoders at 855 rpm is shown in Figure 30.19. A phase reading of 0
degrees is ideal, but is next to impossible to achieve. It depends on the alignment of
the two encoders. In the case of the test model, the reference phase measurement is
113 degrees when the two ends of the shaft are in phase. If a 9-degree twist develops
21.Mobley.29 Page 281 Friday, February 5, 1999 12:40 PM
281 TorsionalAnalysis
Figure 30.18 Encoders at both shaft ends.
in the shaft, the encoder signals will rotate through 360 degrees and return to 133
degrees.
The count depends on the torsional frequency to be measured and the severity of the
twist. Higher torsional frequencies require more pulses per revolution, while greater
twists require fewer pulses per revolution. If the shaft under discussion in Figure
30.18 experiences more than 9 angular degrees of twist, an optical encoder of fewer
pulses per revolution is needed.
LOOKING AT ENCODER OUTPUTS DIRECTLY
For this next example, we will run the test at 45 rpm (see Figure 30.19). The fre-
quency of each encoder is 1800 rpm (encoder count, 40, times shaft speed, 45). Under
ideal conditions, the phase offset should be constant. Figure 30.20 shows the phase
between the two encoders over a 7-sec period. Notice that the phase shift at 45 rpm is
the same as it was at 855 rpm and the phase varies between 130 and 137 degrees. This
shows that there is a 133.5-degree offset between the two encoders and a 7-degree
phase modulation present. However, the phase modulation is not as bad as it may
appear. Based on the phase sensitivity, there are 0.175 angular degrees of error in our
model due to shaft eccentricities, bearing drag, etc. (Angular phase = frequency
phase shift/encoder count = (137 130)/40 = 0.175.)
21.Mobley.29 Page 282 Friday, February 5, 1999 12:40 PM
282 VibrationFundamentals
Figure 30.19 Phase relationship between encoders.
Figure 30.20 Phase over 7 sec.
21.Mobley.29 Page 283 Friday, February 5, 1999 12:40 PM
283 TorsionalAnalysis
Figure 30.21 Third analysis test stand.
REASONS TO MEASURE
There are two reasons to measure torsional phase: static and dynamic twist. Although
torsional phase can be used to pinpoint torsional resonance, this is not very important
because it can be spotted just as easily using torsional amplitudes.
Static Twist
Shafts twist when under load. This is especially true during rapid speed changes, such
as startup acceleration or braking on deceleration. As the shaft ages, the amount of
twist increases and the shafts ability to return to a neutral position decreases.
Take as an example the system shown in Figure 30.21. In this test, the ywheel on
shaft B weighs 8 lb. The test will be run at idle condition before applying full power.
During idle, shaft B neither leads nor lags shaft A and, by denition, the two shafts
are in phase, although the phase angle may not read zero.
Adding power increases the speed. The phase reading between shaft A and B changes
as the mass of shaft B resists the angular acceleration. A phase shift is generated by
the twist that develops in the exible coupling between the two shafts.
21.Mobley.29 Page 284 Friday, February 5, 1999 12:40 PM
284 VibrationFundamentals
Figure 30.22 First-order phase data.
For data normalization, the signature ratio adapter built into the real-time analyzer
tracks the various frequencies associated with the running speed of the test. This is
necessary to compensate for any speed changes in the test.
The phase angle between the two encoders can be measured using the cross power
spectrum and phase modes of the analyzer. Figure 30.22 is obtained by transferring
the phase information to the analyzer waterfall and proling the rst order.
The data reect a constant phase angle of 156 degrees during idle at 1353 rpm. Maxi-
mum torque occurs at 1886 rpm as the phase shifts to 344 degrees. The speed contin-
ues to increase to 2658 rpm, where the phase stabilizes at 256 degrees. There is a
slight oscillation in the phase as the machine reaches full speed. This is caused by the
accumulated torque stored in the ex coupling being released back into the shafts.
Because damping is minimal, some overshoot occurs.
The nal steady-state phase measurement differs from the original because of the added
drag and friction caused by the new speed. The maximum amount of twist seen by the
shaft as the torque is applied is calculated as follows: angular phase (AP) = frequency
shift/encoder count = (344 156)/40 = 4.7 degrees of maximum twist.
21.Mobley.29 Page 285 Friday, February 5, 1999 12:40 PM
285 TorsionalAnalysis
Figure 30.23 Results of varying load test.
Dynamic Twist
To measure a dynamic phase shift, a load is added between the bearings on shaft B
and the test speed is adjusted to 540 rpm. The load is a manually activated mechanical
brake operated at a random rate during the 7-sec data-acquisition period. Figure 30.23
shows the results of this test. Load variance causes varying amounts of twist to
develop in the shaft at a rapid rate. This is a principal cause of fatigue in drive-train
components. In this example, maximum deviation was 60 degrees, representing 1.5
degrees of twist. Note that all of the frequencies involved in shaft-twist deviations will
be reected in a torsional vibration spectrum.
21.Mobley.29 Page 286 Friday, February 5, 1999 12:40 PM
GLOSSARY
Absolute fault limit The maximum recommended level of overall
vibration accepted in machinery.
Acceleration The rate of change of velocity with respect to
time.
Accelerometers Use a piezoelectric crystal to convert mechani-
cal energy into electrical signals.
Amplitude The maximum absolute value attained by the
disturbance of a wave or by any quantity.
Apex The pitch angle is the sum of the pitch lines
extended, which meet at a point.
Balance All forces generated by or acting on the rotating
element of a machine-train are in a state of equi-
librium.
Ball spin frequency (BSF) The spinning motion of the balls or rollers
within a bearing.
Ball-pass inner-race (BPFI) The ball/roller rotating speed relative to the
inner race.
Ball-pass outer-race (BPFO) The relative speed between the balls or rollers in
a rolling-element bearing and the outer race.
Broadband A band with a wide range of frequencies.
Broadband energy Provides a gross approximation of machines
condition and its relative rate of degradation.
Broadband trending Vibration analysis technique that plots the
change in the overall or broadband vibration of a
machine-train.
286
21.Mobley.29 Page 287 Friday, February 5, 1999 12:40 PM
Glossary 287
Centrifugal pump A machine for moving a liquid, such as water,
by accelerating it radially outward in an impel-
ler to a surrounding volute casing.
Chatter An irregular alternating motion of the parts of a
relief valve due to the application of pressure
where contact is made between the valve disk
and the seat.
Coastdown This occurs when the machines driver is turned
off and the suspect frequency is recorded as the
speed decreases.
Common shaft The individual shafts that exist in all machine-
trains.
Displacement The change in distance or position of an object
relative to a reference point; the actual distance,
off-centerline, of a rotating shaft as compared to
a stationary reference, usually the machine
housing.
Dynamic resonance When the natural frequency of a rotating or
dynamic structure, such as the rotor assembly in
a fan, is energized, the rotating element will res-
onate.
Fast Fourier transform (FFT) Converts a time-domain plot into its unique fre-
quency components using a mathematical pro-
cess.
First mode The slightly eccentric rotation (off-center) of a
shaft will generate a low-level frequency com-
ponent that coincides with the rotating speed of
the shaft.
Fourth mode A shaft can ex or deform into mode shapes that
will generate running-speed harmonics.
Frequency The number of cycles completed by a periodic
quantity in a unit time.
Frequency domain A plane on which signal strength can be repre-
sented graphically as a function of frequency,
instead of a function of time.
Fundamental or rst critical speed The lowest critical speed.
Fundamental train frequency Generated by the precession of the cage as it
rotates around the bearing races.
21.Mobley.29 Page 288 Friday, February 5, 1999 12:40 PM
288 VibrationFundamentals
Gear mesh Frequency is equal to the number of gear teeth
times the running speed of the shaft.
Gravity The gravitational attraction at the surface of a
planet or other celestial body.
Harmonic motion A periodic motion that is a sinusoidal function
of time, that is, motion along a line given by
equation x = a cos(kt + 0), where t is the time
parameter, and a, k, and 0 are constants.
Harmonics A sinusoidal component of a periodic wave,
having a frequency that is an integral multiple of
the fundamental frequency. Also known as har-
monic component.
Helical gears Gear wheels running on parallel axes, with teeth
twisted oblique to the gear axis.
Herringbone gears The equivalent of two helical gears of opposite
hand placed side by side.
Hertz Unit of frequency; a periodic oscillation has a
frequency of n hertz if in 1 second it goes
through n cycles.
Hydrodynamic The study of the motion of a uid and of the
interactions of the uid with its boundaries,
especially in the incompressible inviscid case.
Imbalance Any change in the state of equilibrium.
Laminar Arranged in thin layers. Pertaining to viscous
streamline ow without turbulence.
Low frequency cutoff A frequency below which the gain of a system
or device decreases rapidly.
Machine-train A total machine including the driver, drive train,
and machine.
Narrowband The total energy within a user-selected range, or
windows. Referring to a bandwith of 300 hertz
or less.
Narrowband trending Monitors the total energy for a specic band-
width of vibration frequencies.
Natural frequency The natural frequency of free vibration of a sys-
tem. The frequency at which an undamped sys-
tem with a single degree of freedom will
oscillate upon momentary displacement from its
rest position.
21.Mobley.29 Page 289 Friday, February 5, 1999 12:40 PM
Glossary 289
Node points Where the shaft exes into a double bend that
crosses its true centerline.
Oil whip Occurs when the clearance between the rotating
shaft and sleeve bearing is allowed to close to a
point approaching actual metal-to-metal contact.
See also Oil whirl.
Oil whirl An unstable free vibration whereby a uid-lm
bearing has insufcient unit loading. The shaft
centerline dynamic motion is usually circular in
the direction of rotation. Oil whirl occurs at the
oil ow velocity within the bearing, usually 40
to 49% of the shaft speed. Oil whip occurs when
the whirl frequency coincides with (and
becomes locked to) a shaft resonant frequency.
(Oil whirl and whip can occur in any case where
a uid is between two cylindrical surfaces.)
Peak-to-peak Amplitude of an alternating quantity measured
from positive peak to negative peak.
Rate The amount of change of some quantity during a
time interval divided by the length of the time
interval.
Rathbone chart Provides levels of vibration severity that range
from extremely smooth, best possible operating
condition, to an absolute fault limit, or the maxi-
mum level where a machine can operate.
Resonance A vibration of large amplitude in a mechanical
system caused by a small periodic stimulus of
the same or nearly the same period as the natural
vibration period of the system.
Ringing Method for exciting natural frequencies is to
strike or excite a machine or structure with a
timber or hammer.
Root-cause failure Based on machine-train operation and how its
dynamics affect the vibration spectrum.
Rotational frequencies Related to the motion of the rolling elements,
cage, and rings or races.
Running speed The true rotational speed of the shaft or shafts.
Run-up The machines driver is turned on and records
the amplitude and phase as the machine acceler-
ates from dead-stop to full speed.
21.Mobley.29 Page 290 Friday, February 5, 1999 12:40 PM
290 VibrationFundamentals
Sawtooth (waveform) A waveform characterized by a slow rise time
and a sharp fall, resembling a tooth of a saw.
Second mode As the shaft rotates, the double bend shape will
create two high spots as it passes the vibration
transducer.
Signature The characteristic pattern of target as displayed
by detection and classication equipment.
Signature (FFT) Usually applied to the vibration frequency spec-
trum unique to a particular machine or machine
component, system, or subsystem at a specic
location and point of time.
Static A hissing, crackling, or other sudden sharp
sound that tends to interfere with the reception,
utilization, or enjoyment of desired signals or
sounds. Without motion or change.
Static resonance If the natural frequency of a stationary or non-
dynamic structure, such as a casing, bearing
pedestal, piping or other structure, is energized,
the structure will resonate.
Synchronous In step or in phase, as applied to two or more
circuits, devices, or machines.
Third mode A shaft can ex or deform into mode shapes that
will generate running-speed harmonics.
Velocity The time rate of change of position of a body; it
is a vector quantity having direction as well as
magnitude.
Vibration A continuing periodic change in a displacement
with respect to a xed reference. In its general
sense is a periodic motion. The motion will
repeat itself in all its particulars after a certain
interval of time.
Worm gears Used for nonintersecting shafts at 90 degrees.
X-axis A horizontal axis in a system of rectangular
coordinates.
Y-axis A vertical axis in a system of rectangular coor-
dinates.
21.Mobley.29 Page 291 Friday, February 5, 1999 12:40 PM
LISTOFABBREVIATIONS
Contact angle (for roller = 0)
API American Petroleum Institute
BD Ball or roller diameter
BPFI Ball-pass inner-race
cfm gauge pressure
cpm cycles per minute
cps cycles per second
CSI Computational Systems, Inc.
FFT fast Fourier transform
F
MAX
maximum frequency
F
MIN
minimum frequency
f
r
relative speed between the inner- and outer-race
(rps)
FTF fundamental train frequency
Hz hertz
in./sec inches per second
IPS-PK inches per second peak
kHz kilohertz
MHz megahertz
n Number of balls or rollers
OEM original equipment manufacturer
291
21.Mobley.29 Page 292 Friday, February 5, 1999 12:40 PM
292 VibrationFundamentals
PD pitch diameter
psig cubic feet per minute
RMS root mean square
rpm revolutions per minute
VPM vibrations per minute
21.Mobley.29 Page 293 Friday, February 5, 1999 12:40 PM
INDEX
Absolute fault limit, 108
Acceleration, 19, 22, 23, 29
Accelerometers, 52
Acceptance testing, 3, 4
Active channels, 246
Aerodynamic instability, 91, 149, 222
Alarm limits, 107
Alert limits, 107
American Petroleum Institute, 132
Amplitude, 8, 11, 15, 22, 23, 46
Analysis parameter sets, 100
Analysis type, 101
Antialiasing lters, 105
Applied force, 15
Averaging, 105
Axial, 54
Axial movement, 15, 94
Babbitt bearings, 157
Bandwidth, 101, 103, 227
Baseline data, 61, 128, 184
Bearing frequencies, 73, 75, 78, 89, 92, 93,
102, 155
Blade pass frequency, 15, 75
Broadband, 22, 23, 60, 64, 126, 133, 138
Cables, 53
Calibration factor, 243
Centrifugal compressors, 83, 114, 115
Centrifugal pumps, 94, 120
Chain drives, 75, 159
Channel coupling, 241
Circular frequency, 18
Common shaft analysis, 192
Composite trends, 128
Compressors, 83, 113
Constant speed, 98, 104, 218
Couplings, 76, 77
Crankshaft frequency, 85
Critical speeds, 139, 201
Cross-machine comparison, 136, 182
Damping, 26, 28, 30, 33, 36, 280
Data acquisition, 49, 235
Data lters, 101
normalization, 136, 196
verication, 59
Degrees of freedom, 36, 38
Displacement, 8, 17, 22, 29, 50
Dynamic resonance, 152, 205, 211, 218
Dynamic twist, 285
Dynamic vibration, 47, 48
Eddy current probes, 50
Electric motors, 72, 117
Electromagnetic elds, 53
End play, 15
Engine analyzers, 5
Engineering units, 243
Equipment information sheets, 97
Exponential averaging, 251
Failure modes analysis, 138
Fans and blowers, 90, 118, 211
Fast fourier transform, 10, 42, 45, 63, 138,
225
Flat-top weighting, 249
Flow instability, 14
Forced vibration, 33
Fourier, 8
Fourier series, 21
Free run, 244
Free vibration, 28, 30
Frequency, 21, 46
analysis, 101, 138
domain, 8, 9, 10, 22, 45, 49, 65, 69, 181
Full scale voltage, 242
Gearboxes, 78, 118
Gearmesh frequency, 79, 89, 102, 160
293
21.Mobley.29 Page 294 Friday, February 5, 1999 12:40 PM
294 Index
Generators, 92 Periodic motion, 6, 17, 20
Permanent mounting, 55
Handheld transducers, 56 Phase angle, 21, 46, 281
Hanning weighting, 249 Piezoeclectric crystals, 52
Harmonic function, 6, 9, 21, 102 Piston orientation, 88
Harmonic motion, 17, 20, 34 Pitch circumference, 82
Hertz, 21 Positive displacement compressors, 84, 115
Hydraulic instability, 95, 149, 222 Positive displacement pumps, 95, 123
Predictive maintenance, 3, 4
Imbalance, 13, 73, 140, 175, 177 Primary frequency, 21
Induced loads, 15 Process envelope, 62, 130, 198,
Industrial reference data, 62, 130, 185 Process information sheets, 99
Inertia force, 30 Process instability, 149, 222, 232
Process rolls, 93, 119, 166
Jackshafts, 77, 163 Process weighting, 248
Pumps, 94, 120
Leak detection, 5
Line frequency, 73, 92 Quality control, 3, 4
Linear averaging, 251 Quick disconnect, 56
Linear motion, 15
Lines of resolution, 103 Radial, 54
Load, 11, 14, 98, 108, 197 Rathbone chart, 130
Load distribution, 93, 119 Real-time analysis, 225
Loose parts detection, 3, 5 Reciprocating compressors, 84, 115
Loose rotor bars, 73 Reciprocating machines, 15, 84, 124
Low frequency response, 70 Rectangular weighting, 249
Reference channels, 247
Magnets, 56 Resonance, 150, 177, 201
Mass, 26, 33, 36 Revolutions per minute, 9
Maximum frequency, 102 RMS, 25
Mechanical imbalance, 14, 140 Root cause analysis, 189
Mechanical looseness, 143 Root-mean-square, 25
Mechanical motion, 15 Rotating machinery, 13, 18
Minimum frequency, 103 Rotor.
Misalignment, 93, 145, 173, 175 imbalance, 14
Mode shape, 75, 91, 133, 222 passing frequency, 90
Modulations, 146 Running speed, 9, 10, 74, 76, 79, 81, 90, 91,
Monitoring parameters, 71 92, 102
Multi-channel, 48, 233
Multiplane imbalance, 142 Screw compressor, 89, 116
Settling time, 57
Narrowband, 22, 23, 61, 64, 71, 91, 104, Shaft deection, 75, 91, 102, 133, 195
127, 134 Signature analysis, 63, 181
Narrowband zoom, 233 Single-channel, 48, 68
Natural frequency, 29, 34, 213 Single plane imbalance, 141
Noise control, 3, 5 Sleeve bearings, 157
Nonharmonic motion, 20 Slip frequency, 74
Normalization, 10, 136, 196 Speed, 75, 108, 197
Spring force, 31
Operating envelope, 14, 110 Static.
Orders analysis, 101 deection, 29
Overlap averaging, 106, 247 resonance, 151, 203, 213
twist, 284
Passing frequencies, 99 Steady state, 47, 68
Peak hold averaging, 251 Steam turbines, 74
Peak-to-peak, 24 Stiffness, 26, 27, 33, 36
21.Mobley.29 Page 295 Friday, February 5, 1999 12:40 PM
Synchronous time averaging, 233, 259
Time domain, 8, 10, 42, 49, 138
Torsional analysis, 234, 267
resonance, 278
stiffness, 37
Transducer, 11, 49, 112
Transient analysis, 230, 255
Trending, 60, 125
Trigger group, 243
Turbulent ow, 14
Undamped, 28, 33, 38
Universal joints, 78
Vane pass frequency, 15
Index 295
Variable speed, 98, 208, 219, 231
V-belts, 80, 173
Velocity, 19, 22, 29, 51, 54
analysis, 8
monitoring, 3
prole, 13, 16, 21
severity, 22, 63, 109, 110, 130
signature, 11, 15, 21, 63, 138
sources, 16
damping, 30
Waterfall analysis, 182, 220, 255
Waveform. See Time domain
Zero-to-peak, 22, 24
Zoom analysis, 265