107

Chapter 8

Spectrum Plots

I n previous chapters, we have discussed the use of filtering to obtain vibration

vectors. These vectors are filtered to multiples of running speed, 1X, 2X, 3X,
etc., and provide us with information about the behavior of a machine at one of
those particular frequencies. This information is presented to us as the ampli-
tude and phase of the vibration at that frequency.
However, machines can vibrate at many different frequencies simultaneous-
ly. These frequencies can be related or unrelated to running speed and include
both subsynchronous and supersynchronous frequencies. Since these frequen-
cies are associated with the operating condition of the machine, the machinery
diagnostician must have some way to determine the frequency content of a
vibration signal in order to make an accurate diagnosis.
Vibration frequencies sometimes appear in the form of a harmonic series. A
harmonic series consists of the lowest frequency in the series, called the funda-
mental, and a number of frequencies at integer multiples of the fundamental. In
typical harmonic series, the amplitude of higher order frequencies will decline
rapidly. To avoid confusion, in this book we will define a harmonic as any fre-
quency that is an integer multiple of the fundamental. The first harmonic is the
fundamental, the second harmonic has a frequency of twice the fundamental,
the third harmonic has a frequency of three times the fundamental, etc. Often,
the term harmonics will be used as a general term to indicate integer multiple
frequencies that are above the fundamental.
The fundamental vibration frequency of a harmonic series can be any vibra-
tion frequency. Often, the fundamental is 1X, but it can also be any subsynchro-
nous or supersynchronous frequency. For example, a harmonic series could be
108 Data Plots

based on a fundamental at ½X and include 1X, ³⁄₂X, 2X, ⁵⁄₂X, etc. In this series,
the ½X is the fundamental or first harmonic, 1X is the second harmonic, ³⁄₂X is
the third harmonic, etc. Such a series can be generated by a ½X rub, as we will
see later in this chapter.
While the timebase and orbit can be used to evaluate frequency informa-
tion, the most convenient plot for this purpose is the spectrum plot. The spec-
trum plot is created from the signal of a single transducer. It is the basic display
of a spectrum analyzer and has been a mainstay of machinery diagnostics for
many years.
In the past few years, an important new tool, the full spectrum plot, has been
developed that uses the signals from a pair of orthogonal, shaft relative, vibra-
tion transducers. The full spectrum plot compares to the conventional spectrum
plot in the same way that the orbit plot compares to a timebase plot. The full
spectrum plot contains much more information than the spectrum plot, includ-
ing vibration precession direction and orbit ellipticity. It is so important for
machinery diagnostics that we now refer to the conventional spectrum plot as
the half spectrum plot.
Spectrum plots are used to identify the frequency components that are pres-
ent in complex vibration signals and to trend changes in the amplitude of fre-
quency components. These frequencies include running speed, multiples of run-
ning speed, line frequency electrical noise, gear mesh frequencies, gear defect
frequencies, rolling element bearing frequencies, and vane and blade pass fre-
quencies. Rotor system natural frequencies that are excited will also show up on
the spectrum plot. Subsynchronous frequencies that are often associated with
fluid-induced instability, compressor rotating stall, compressor surge, or rub,
and supersynchronous frequencies that are often associated with rubs and shaft
cracks can also be identified.
In this chapter, we will start with a discussion of the complex vibration sig-
nal, its frequency content, and how that information is displayed on the half
spectrum plot. We will then discuss some technical aspects of spectrum signal
processing, followed by the meaning of and enhanced information content in
the full spectrum plot.
After full spectrum, we will discuss plot formats that present spectrum data
versus speed and versus time: the spectrum cascade plot for startup or shutdown
data and the spectrum waterfall plot, which is used primarily for steady state
trending. Both of these plots can be generated in half and full spectrum formats.
These plots are similar in structure, but have important differences in their
application.
 Chapter 8 Spectrum Plots 109

The Spectrum Plot

Let’s start with a machine operating at a constant speed. Because of a com-
bination of unbalance and other effects, the machine is vibrating in a compli-
cated way. The unfiltered vibration signal from a transducer on this machine will
reveal all of the complexity of the machine vibration that falls within the band-
width of the data collection system. A filtered signal from this transducer will be
a sine wave with some amplitude and phase.
Figure 8-1 shows a complex, timebase vibration signal (red) and a series of
sine waves (blue) that add up to produce the timebase signal. Using the Fourier
Transform, the frequency, amplitude, and phase of these sine waves can be com-
puted from a digital sample of the original timebase signal. The phase for each
signal (a component of the original) is measured with respect to the trigger sig-
nal that starts the sampling process at time t0. The Fourier Transform output is

f4 = 4 f1

f 3 = 3 f1 Displacement

Frequency
components
f 2 = 2 f1
Tim
e

ency
Frequ
f
1

Spectrum plot
Amplitude (pp)
Displacement

Raw
timebase
signal

f1 f f f4
Time 2 3
Frequency

Figure 8-1. Time and frequency domain display of a complex signal. The Fourier Transform
applied to a complex, periodic, timebase signal (red) produces a set of harmonic sine waves, or
components. The signal components are grouped at upper right by adding the third dimension,
frequency. The plot is rotated to produce a two-dimensional plot of amplitude versus frequency,
the spectrum plot. The components now appear as a series of vertical lines, whose lengths repre-
sent the peak-to-peak amplitudes of the signal components. Because the timing (phase) informa-
tion is lost, it is not possible to reconstruct the original waveform from the components of the
spectrum plot.
110 Data Plots

equivalent to the output of a series of band-pass filters that have been set to
integer multiples of the lowest frequency signal, f1.
The component sine waves in the figure are plots of displacement versus
time. If we also plot them versus frequency, we can create a three-dimensional
plot (upper right). With this perspective, we can see each component’s frequen-
cy, amplitude, and phase.
This plot is somewhat repetitious along the time axis. After all, we know
these are sine waves; why continue repeating the same data over and over? If we
rotate the plot so that the time axis disappears, we will see a two-dimensional
plot of amplitude versus frequency (bottom right). Note that the component sig-
nals now appear as a series of vertical lines; each line represents a single fre-
quency and its height is the amplitude of the signal. This is a conventional, or
half spectrum plot.
By hiding the time axis, we have obtained a relatively simple plot format that
allows us to clearly see the frequencies and amplitudes of the component sig-
nals. Unfortunately, we have hidden an important piece of information, the
phase of the component signals. Because of this, is not possible to reconstruct
the original waveform from the component frequencies and amplitudes. This is
a drawback of all types of spectrum plots.
The amplitude scale can be either linear or logarithmic. Logarithmic scaling
is useful when there is a need to compare signals with both very large and very
small amplitudes. This scaling will clearly display all signal components and the
noise floor. However, when applied to rotating machinery work, logarithmic
scaling makes it more difficult to quickly discriminate between significant and
insignificant vibration components. Linear scaling has the advantage of show-
ing the most significant components; weak, insignificant, and low-level noise
components are greatly reduced in scale or eliminated. Because of its advantage
for machinery work, linear scaling will be used for the spectrum plots in this
book.
The frequency scale can be displayed in several frequency units. Most spec-
trum analyzers display the frequency in hertz (Hz). This is useful when compar-
ing machine vibration frequencies to line frequencies, such as in induction
motor or steam turbine generator diagnostics. Some software packages can dis-
play units of Hz, cpm, or orders of running speed. Cpm is a very convenient unit
when working with machinery, because it is easy to compare a frequency in cpm
to the running speed of the machine in rpm. Spectrum plots that are displayed
in orders of running speed (1X, 2X, etc.) are also easy to interpret.
 Chapter 8 Spectrum Plots 111

Technical Issues
While, on the surface, the spectrum plot appears simple to interpret, it is
important to understand its limitations and to recognize bad or questionable
data. The discussion here is a brief summary of the most important points. See
Chapter {Signal Processing} for more information.
The Fourier Transform (and its computational cousin, the Fast Fourier
Transform, or FFT) assumes that the unfiltered vibration signal has always been,
and will always be, unchanged. In other words, the machine vibration state does
not change, and the signal repeats in exactly the same way forever. This assump-
tion is an adequate approximation for most machines that operate at a steady
state speed or change speed slowly.
This assumption can break down badly for machines that experience sud-
denly changing vibration conditions (such as crushers or wood chippers) or for
machines that accelerate (induction motors) or decelerate very rapidly.
Spectrum plots under these circumstances can have significant errors in ampli-
tude and frequency. Spectra of rapidly changing data can exhibit broadened
spectral lines that are significantly shifted in frequency.
In practice, the Fast Fourier Transform (FFT) calculates the spectrum from
a sample record, which contains a specific number of digital waveform samples.
Because the sample record has a finite length, part of the algorithm involves
extending its length by, in effect, repeatedly wrapping the signal around on
itself. Unless the number of cycles of the signal exactly matches the length of the
sample record (which is improbable), there will be a discontinuity at the junc-
tion. This discontinuity introduces noise (leakage) into the spectrum, which
broadens the frequency lines, reduces the calculated amplitudes, and increases
the noise floor.
This problem is reduced by windowing. A window function applied to the
sample record forces the signal to zero at the endpoints, typically in a gradual
and smooth manner. This has the effect of eliminating the step discontinuity in
the extended signal. Depending on the software, several types of windowing
functions are available, each with its own advantages and disadvantages. A
Hanning window function is usually the best compromise for rotating machin-
ery work, providing good amplitude and frequency resolution. Note that
different windowing functions will produce different results for amplitude and
frequency when applied to the same data.
Figure 8-2 shows a vibration waveform, containing a mixture of 1X and ½X
frequencies, and two examples of conventional (half) spectrum plots. For the
top spectrum plot (the middle plot in the figure), no windowing function was
applied to the sample record. Note that the spectrum “lines” are not quite lines;
instead, they have some finite width and widen at the bottom. A small noise
112 Data Plots

floor is also visible. This is an example of leakage due to the discontinuity at the
sample record endpoints and the limitations of digital sampling. The bottom
plot shows the spectrum when a Hanning window is applied to the sample
record. Note that the ½X spectral line is narrower and higher, and the residual
noise floor has virtually disappeared.
A digitally calculated spectrum consists of discrete frequency bins (lines) of
finite width. Thus, resolution (line width) is also an important issue. The maxi-
mum resolution of a spectrum is determined by the ratio of the spectrum span,

Parent waveform

-2

-4

0 20 40 60
Time (ms)
Figure 8-2. A timebase plot with two exam- No window
8
ples of conventional (half ) spectrum plots.
For the middle plot, no windowing function
Amplitude (mil pp)

6 1/2X
was applied to the sample record. The spec-
trum “lines” are not quite lines; instead, they
4 1X
have some finite width and widen at the
bottom. The bottom plot shows the spec-
2
trum calculated using a Hanning window.
The 1/2X spectral line is narrower and high-
er, and the residual noise floor has virtually 0
0 1 2 3 4 5 6
disappeared. Frequency (kcpm)

Hanning window
8
1/2X
Amplitude (mil pp)

4
1X

0
0 1 2 3 4 5 6
Frequency (kcpm)
 Chapter 8 Spectrum Plots 113

the range of displayed frequencies, to the number of spectrum lines that are dis-
played:

Span
Resolution = (8-1)
Number of Lines

The spectrum plot is a collection of these lines, arranged side by side. The
width of each line is equal to the resolution of the spectrum. For example, a 400
line spectrum with a span of zero to 200 Hz will have a resolution of

200 Hz
Resolution = = 0.5 Hz/Line
400 Lines

Thus, each frequency line will, ideally, represent only the spectral energy in a 0.5
Hz (30 cpm) wide band from 0.25 Hz below to 0.25 Hz above the center frequen-
cy of the line. Accuracy in the displayed amplitude and frequency of a spectrum
line will depend on where the actual vibration frequency is with respect to the
center frequency and which window function is used.
The limited resolution of spectrum plots means that there is always an
uncertainty associated with any frequency we wish to measure. In the example
above, a frequency actually located at, for example, 99.75 Hz, is displayed at 100
Hz. A spectrum plot with poor resolution will have a corresponding large uncer-
tainty in the measured frequency. Even good resolution spectra may not be able
to discriminate between vibration frequencies of exactly ½X and 0.49X, an
important distinction for malfunction diagnosis. Higher resolution (zoomed)
spectra can help, but orbits with Keyphasor dots can sometimes be superior to
spectrum plots for making this kind of discrimination (see Chapter {5}).
Noise can be a problem in spectrum plots. The Fourier transform of a spike
is a series of spectrum lines extending to very high frequency. Thus, anything
that produces a sharp corner in the signal will produce a series of spectrum
lines. Sharp corners can result from shaft rebound at a rub contact point or from
an inadequate sampling frequency (causing a corner where a smooth transition
really exists), among other things. Spikes or steps in the signal can originate
from electrical noise problems or from scratches on the shaft. Spectrum plots
are calculated from uncompensated waveforms which may contain significant
slow roll or glitch content. In general, the appearance of a harmonic series of
spectrum lines should be viewed with caution. Use Timebase, Orbit, or Cascade
plots (below) to validate the data.
114 Data Plots

The Full Spectrum

The half spectrum is a spectrum of a single timebase waveform. The full
spectrum is the spectrum of an orbit. It is derived from the waveforms from two,
orthogonal, shaft relative transducers, combined with knowledge of the direc-
tion of rotation. The information from the two transducers provides timing
(phase) information that allows the full spectrum algorithm to determine the
direction of precession at each frequency. Because the timing information is
critical, the two waveforms must be sampled at the same time.
The full spectrum is calculated by performing an FFT on each transducer
waveform. The results are then subjected to another transform that converts the
data into two new spectra that represent frequencies of precession, one spec-
trum for X to Y precession and one for Y to X precession. The last step uses the
direction of rotation information to determine which of the spectra represents
forward and which represents reverse precession frequencies. When this process

Y X
Timebase
waveforms

Y X

FFT FFT

Half spectrum Half spectrum

Complex spectral data

Full Spectrum Transform

Rotation
direction
R F

Full spectrum

Figure 8-3. Calculation of the full spectrum. An FFT is performed on each waveform from an XY
transducer pair. The complex spectral data is then subjected to another transform that generates
two new spectra that represent the frequencies of precession: one spectrum for X to Y precession
and one for Y to X precession. The direction of rotation is used to determine which of the spectra
represents forward precession and which represents reverse precession frequencies.
 Chapter 8 Spectrum Plots 115

is completed, the two spectra are combined into a single plot, the full spectrum
plot (Figure 8-3).
Figure 8-4 shows the relationships among timebase waveforms, half spectra,
the orbit, and the full spectrum. The Y and X timebase waveforms and their half
10
Y Y
4
8
Displacement (mil)

Amplitude (mil pp)

1/2X
2
6
0
4 1X
-2
2
-4
0
0 20 40 60 0 1 2 3 4 5 6
Time (ms) Frequency (kcpm)

10
X X 1X
4
8
Displacement (mil)

Amplitude (mil pp)

2
6
0
4
-2
2
-4 1/2X
0
0 20 40 60 0 1 2 3 4 5 6
Time (ms) Frequency (kcpm)

Full spectrum
Y 10
4
8
Amplitude (mil pp)
Displacement (mil)

2 +1X
6
0 X
+1/2X
4 –1/2X
–1X
-2
2
-4
0
-4 -2 0 2 4 -6 -4 -2 0 2 4 6
4410 rpm Reverse Forward
Precession frequency (kcpm)
Figure 8-4. Timebase waveforms and their half spectra, the orbit and its full spectrum. The two
waveforms combine to produce the orbit. The data that produces the two half spectra is
processed to produce the full spectrum at bottom right, which is the spectrum of the orbit.
116 Data Plots

spectra are at the top. The two waveforms combine to produce the orbit at bot-
tom left. The data used to generate the half spectra are further processed to pro-
duce the full spectrum at bottom right. Essentially, the full spectrum is the spec-
trum of the orbit; it contains some of the information that the orbit displays that
is not available in the individual waveforms. Note that you can not generate the
full spectrum by combining the two half spectra.
In the full spectrum plot, the spectrum of forward precession frequencies is
on the positive horizontal axis and the spectrum of reverse precession frequen-
cies is on the negative horizontal axis. Thus, for each frequency, there are two
possible spectrum lines, one forward, and one reverse. The relative length of the
spectrum lines for each frequency indicates the shape and direction of preces-
sion of the orbit filtered to that frequency.
Figure 8-5 shows four, circular, 1X orbits, with different directions of preces-
sion, indicated by the blank/dot sequence, and different directions of rotor rota-
tion, indicated by the arrow. To the right of each orbit is its full spectrum. Since

Orbit Full spectrum

tn
Ro +1X

Forward
precession

–1X
Reverse Forward

tn
Ro –1X
Figure 8-5. 1X, circular orbits and their full
spectra. The spectrum line appears on the Reverse
precession
forward side of the plot for the two orbits
that are undergoing forward precession. +1X
The spectrum line appears on the reverse Reverse Forward
side of the plot for the two orbits that are
tn
undergoing reverse precession. The relative Ro +1X
direction of vibration precession, not just Forward
the direction of rotation, determines precession
whether the line is forward or reverse.
–1X
Reverse Forward

tn
Ro –1X

Reverse
precession

+1X
Reverse Forward
 Chapter 8 Spectrum Plots 117

Orbit
t n
Ro

Full spectrum
+1X

–1X 2a
2b

Reverse Forward

tn tn 1 tn
Ro Ro Ro
2 b
1
b a
a b
1 a
2

Possible vector descriptions

Figure 8-6. Forward and reverse vector construction of a 1X, elliptical orbit. The full spectrum of
an orbit shows a 1X, forward precession component and a smaller, 1X, reverse precession compo-
nent. The sum of the lengths of the two full spectrum lines is equal to the length (peak-to-peak)
of the major axis of the elliptical orbit. The difference between the lengths of the two full spec-
trum lines is equal to the length (peak-to-peak) of the minor axis of the elliptical orbit. The orbit is
generated by the sum of two, counterrotating vectors with lengths equal to half the spectrum
line length. Because of the lack of phase information in the full spectrum, the orbits may have any
orientation.
118 Data Plots

each orbit is circular, there is only one line, which is the peak-to-peak amplitude
of the orbit. Regardless of the direction of rotation, when the direction of pre-
cession is forward (in the direction of rotation), the line is on the positive axis,
and when the direction of precession is reverse (opposite the direction of rota-
tion) it is on the negative axis.
The path of these circular orbits is generated by a single, rotating vector. The
vector length is the peak amplitude of the orbit (½ the line height, which repre-
sents peak-to-peak amplitude). It rotates at the frequency of the line and in the
direction of precession indicated by the line. As this vector rotates, the tip of the
vector will trace out the path of the orbit. The full spectrum line defines this
rotating, forward vector, and, consequently, the orbit. As with the half spectrum,
there is no phase information in the full spectrum plot, so the Keyphasor dot
location is arbitrary.
For the 1X, elliptical orbit in Figure 8-6, the direction of precession is X to Y,
the same as the rotation direction, so the orbit is another example of forward
precession. Its full spectrum shows a 1X, forward precession component and a
smaller, 1X, negative precession component. The vectors a and b are defined by
½ the value of the positive and negative lines of the full spectrum and rotate in
opposite directions. The sum of the vectors a and b generate the path of the
orbit. Because they rotate in opposite directions, they will alternately add to and
subtract from each other twice per revolution, defining the semimajor and semi-
minor axes of the elliptical orbit. (The sum and difference of the full spectrum
line heights are peak-to-peak values and would define the major and minor
axes).
Thus, the full spectrum lines represent the forward and reverse vectors that
define the orbit. Note that the orientation of the orbit ellipse shown in the figure
is arbitrary. The full spectrum contains information about size, ellipticity, and
direction of precession, but, because there is no phase information, there is no
information about orientation. An orbit with any orientation, but possessing the
same ellipticity, size, and direction of precession will have the same full spec-
trum. Thus, the full spectrum is independent of transducer orientation. This is
different from half spectrum, which does depend on the transducer orientation.
Figure 8-7 shows the progression of 1X orbit shapes from forward circular
through line to reverse circular and the associated full spectra. The correspon-
ding full spectra show the same sort of progression. The relative size of the for-
ward and reverse line heights correlate with the shape and precession direction
of the orbit:

1. A single component, whether forward or reverse, means a circular

orbit.
 Chapter 8 Spectrum Plots 119

2. The largest component determines the direction of precession.

3. The smaller the difference between the components, the more

elliptical the orbit.

4. Equal components means a line orbit.

Note that, although the full spectrum can be derived from the orbit, it is not pos-
sible to derive the orbit solely from a full spectrum; the full spectrum does not
display phase information.

Orbit Full spectrum

tn
Ro +1X

–1X
Reverse Forward

tn
Ro

+1X

–1X
Figure 8-7. Circular and elliptical 1X orbits
with their full spectra. The direction of Reverse Forward
precession of the orbit is indicated by the
tn
dominant line of the forward and reverse Ro

components. The ellipticity is determined

–1X +1X
by the relative size of the forward and
reverse components. Note that when the
orbits are circular, there is only one spec-
trum line, and when the orbit is a line, the Reverse Forward

spectrum components are equal. Ro

tn

–1X

+1X

Reverse Forward

tn
Ro –1X

+1X
Reverse Forward
120 Data Plots

Complicated orbits will have forward and reverse components at many fre-
quencies. Each pair of components represents a set of vectors that rotate in for-
ward and reverse directions at a specific frequency. The most complex orbit can
always be described by set of such vectors and full spectrum lines. The lines in
the full spectrum represent the precessional structure of the orbit. Each pair of
forward and reverse precession frequency components describes an orbital com-
ponent, a suborbit (circular, elliptical, or line) with a particular precession fre-
quency and direction. The entire orbit can be expressed as the sum of its orbital
components in the same way that a timebase waveform can be expressed as the
sum of its sine wave components.
Figure 8-8 shows a complex orbit from a steam turbine with a ½X rub. The
orbit contains ½X, 1X, and some higher order vibration frequencies. The full
spectrum helps clarify the complexity. Note that the 1X spectral line pair shows
that the 1X component is largest, forward, and mildly elliptical. The ½X line pair
shows that this component is nearly a line orbit. Also, there is 2X vibration pres-
ent that is also a line orbit. Some small ³⁄₂X, the third harmonic of the ½X fun-
damental, is also visible.

At first glance, the full spectrum might seem abstract. What is significant
about pairs of vectors with forward and reverse precession? It lets us easily iden-
tify key orbit characteristics that might otherwise be obscured. Precession direc-
tion and ellipticity provide insight into the state of health of a machine. More

10 Full spectrum
10 Y X +1X
Amplitude (mil pp)

0 5
–1/2X +1/2X
–2X –1X +2X
-5

–3/2X +3/2X
-10 0
1 mil/div 2580 rpm -6 -4 -2 0 2 4 6
Reverse Forward
Precession frequency (kcpm)
Figure 8-8. An orbit and full spectrum from a steam turbine with a 1/2X rub. The full spec-
trum helps clarify the complex orbit, which is a sum of 1/2X, 1X, and some higher order
vibration. The 1X component orbit is the largest, forward, and mildly elliptical. The 1/2X and
2X orbits are nearly line orbits. A small 3/2X component is also visible and is the third har-
monic of the 1/2X fundamental.
 Chapter 8 Spectrum Plots 121

Rub Fluid-induced instability

8 20
Y 1/2X Y
0.49X
Amplitude (mil pp)

6 15

1X
4 10
1X

2 5

0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Frequency (kcpm) Frequency (kcpm)

8 20
X and Y Forward X and Y Forward
highly nearly
+1X +0.49X
Amplitude (mil pp)

6 elliptical 15 circular

+1/2X +1X
4 10
–1X –1/2X

2 5
–0.49X
–1X
0 0
-6 -4 -2 0 2 4 6 -6 -4 -2 0 2 4 6
Reverse Forward Reverse Forward
Precession frequency (kcpm) Precession frequency (kcpm)

Y Y
4 20

2 10

0 X 0 X

-2 -10

-4 -20

-4 -2 0 2 4 -20 -10 0 10 20
4410 rpm 5070 rpm
Figure 8-9. Half and full spectrum display of a 1/2X rub (red data) and a fluid-induced instability
(blue data). Note the similarity in appearance of the two half spectrum plots (top). The full spec-
trum plots clearly reveal the difference in the subsynchronous vibration: the rub orbit is extreme-
ly elliptical; the fluid-induced instability orbit is forward and nearly circular. The unfiltered orbits
are shown for comparison.
122 Data Plots

importantly, some rotor system malfunctions can have characteristic signatures

on a full spectrum plot that are not available on half spectrum plots. These char-
acteristics can be used to discriminate between different malfunctions that
produce vibration with similar frequencies.
An example is discrimination between a ½X rub and fluid-induced instabil-
ity. Fluid-induced instability almost always appears as a predominantly forward,
nearly circular, subsynchronous vibration, usually at a frequency below ½X.
Compare the half and full spectrum plots (Figure 8-9) of a ½X rub (red data) to
a fluid-induced instability (blue data). Note the similarity in appearance of the
half spectrum plots (top). The frequencies of the subsynchronous lines (in
orders of machine speed) are very close. It would be very difficult to discriminate
between these two malfunctions given only half spectrum plots.
Compare the full spectrum plots (middle). There is a clear difference in the
relative size of the forward and reverse subsynchronous components. The plot
for the rub (left) shows that the subsynchronous component is extremely ellip-
tical. The full spectrum plot for the instability (right) shows that the subsyn-
chronous vibration is clearly forward and nearly circular. The forward, circular,
subsynchronous behavior is typical of fluid-induced instability and atypical of
rub. The additional information on the full spectrum plot, which may not be
immediately obvious on the orbits (bottom), clearly reveals a difference in
behavior that is valuable for diagnostics.
Full spectrum is a new tool. Once its relationship to the orbit is understood,
half spectrum plots appear limited by comparison. The effort made to master
this important new format will be rewarded by an enhanced understanding of
machinery behavior.

Spectrum Cascade Plots

During a startup or shutdown, spectra can be taken at different speeds.
These spectra can be displayed by adding a third dimension to the spectrum
plot, rotor speed. The spectra are positioned in the order of increasing speed
with the lowest speed spectrum in front. Such a plot is called a Spectrum
Cascade plot, or, more simply, a cascade plot. A cascade plot can be constructed
with a series of half spectra or with a series of full spectra. The Full Spectrum
Cascade plot is preferred because of its higher information content.
The Full Spectrum Cascade plot in Figure 8-10 shows the startup of a
machine with a rub. As with the Full Spectrum plot, the horizontal axis repre-
sents precession frequency. The vertical scale to the right is for amplitude, which
is measured from the baseline of each spectrum. The vertical scale to the left is
for the third dimension, rotor speed. Each spectrum is placed on the plot so that
its baseline corresponds to the rotor speed at which the sample was taken.
 Chapter 8 Spectrum Plots 123

Order line
+1X
–1X –1/2X +1/2X

4810 rpm
5
Speed (krpm)

2 mil pp/div
Amplitude
3

2 4410 rpm

0
-8 -6 -4 -2 0 2 4 6 8
Reverse Forward
1810 rpm
Precession frequency (kcpm)

Figure 8-10. A Full Spectrum Cascade plot of a machine startup. The horizontal axis represents
precession frequency, the amplitude scale is to the right, and the rotor speed scale is on the left.
Order lines, drawn diagonally from the origin, highlight diagonal, tracking relationships: vibration
frequencies that are proportional to running speed. The horizontal and vertical relationships
associated with the machine resonances and a rub are marked with ellipses. See the text for a full
discussion.

Three primary relationships should be examined when viewing a cascade

plot. First is the diagonal relationship; vibration frequencies that change with, or
track, running speed. Order lines, which are drawn diagonally from the origin
(zero speed, zero frequency), are the most typical diagonal relationship (these
are not available when the frequency axis is in orders of running speed). The ±1X
order lines show the points on the spectra where the vibration or precession fre-
quency is equal to running speed. The ±2X order lines identify frequencies equal
to twice running speed, the ±½X lines identify frequencies equal to one-half run-
ning speed, etc.
Order lines quickly establish important tracking frequency relationships.
The most obvious are the 1X order lines. Vibration on these lines is typically
caused by a combination of runout, shaft bow, and unbalance. Normally, the +1X
vibration component will be higher than the –1X component because unbal-
ance usually produces forward orbits. The figure shows two probable balance
resonances, which appear as peaks in amplitude. Because of the lack of phase
124 Data Plots

information on the spectrum plot, suspected resonances should always be

confirmed on a Bode or polar plot.
Remember that rapid machine accelerations and decelerations may cause
smearing and offset of frequencies in the spectra. A good check is to verify that
the 1X vibration is aligned with the 1X order line. Any significant offset implies
that frequencies have been shifted.
The second key relationship is horizontal. Horizontal relationships exist for
different frequencies at the same speed (in the same spectrum). One example of
this relationship is the integer multiples (harmonics) of forcing frequencies pro-
duced by nonlinearities and asymmetries in rotor systems. The primary forcing
frequency, 1X due to unbalance, can produce harmonics at 1X, 2X, 3X, etc.
Another example is the sum and difference frequencies (sidebands) produced
when one signal modulates another.
Harmonics can also be generated by rubs. Anything in a machine that caus-
es a sudden change in direction of the shaft will produce harmonics in the spec-
trum. The sharper the turn, the richer the harmonic spectrum. Subsynchronous
vibration frequencies can also produce harmonics; for example, a ½X rub can
produce harmonics at 1X, ³⁄₂X, 2X, ⁵⁄₂X, etc (Figure 8-8). Scratches on shafts
cause sudden changes in displacement signals that result in harmonics (these
signals also track running speed and stay constant in amplitude over a wide
speed range).
Horizontal relationships can involve more than harmonics. The ½X vibra-
tion in Figure 8-10 is caused by a rub that is the result of high amplitude
response in the second balance resonance. The horizontal relationship is impor-
tant because of the correlation of the ½X vibration with the resonance, and it is
marked with the horizontal ellipse.
The third key relationship is vertical. Vertical relationships exist for things
that happen at the same frequency over different speeds (multiple spectra).
Many rotor system natural frequencies remain constant over speed and appear
on cascade plots as a series of lines on different spectra, arranged vertically. As
a machine changes speed, if the 1X forcing function coincides with (crosses on
a cascade plot) a natural frequency of the rotor, a resonance will occur. If there
are harmonics, the harmonic vibration is also available to excite any resonance.
Thus, if a 2X vibration coincides with a natural frequency, then it will excite that
natural frequency as it passes through. The resonance will show on the 2X order
line as a peak of amplitude that will be vertically aligned with the peak of ampli-
tude on the 1X line.
Similarly, 3X, 4X, etc., vibration, for example, that produced by a blade- or
vane-pass mechanism, can also excite a resonance. These vertical relationships
can be important for malfunction diagnosis. Often the only evidence of a low
 Chapter 8 Spectrum Plots 125

level of higher order vibration will be when the higher order line crosses a rotor
system natural frequency.
In the figure, the vertical ellipse shows the nearly vertical relationship
between the rub vibration frequency and the first balance resonance frequency
of this machine. The offset in frequency is predictable, because the rotor system
has been stiffened by the rub contact, and the rotor system natural frequency
that was excited during the resonance has been moved to a higher frequency.
The Full Spectrum Cascade plot is a good tool to use when a significant
amount of Not-1X activity is detected. The components of the Not-1X vibration
can be quickly identified by using this plot, which will aid in identifying the
source of the vibration.
Unbalance response will usually produce negligible vibration at slow roll
speeds. Because spectrum plots are generated from uncompensated data, a
large slow roll runout will produce significant vibration along the 1X order line
down to very low speeds, and it can be easily recognized. Figure 8-10 shows this
behavior. Note that the +1X vibration is clearly visible at slow roll speed; this
machine had a significant rotor bow.
Shaft scratches can produce a rich harmonic spectrum. Because the parent
waveforms are uncompensated, the scratch spectrum will appear in all of the
spectra on the cascade plot. This behavior makes it fairly easy to recognize. If the
same set of harmonics are visible at all speeds, then they are probably due to one
or more shaft scratches. Multiple scratches can appear as a mixture of forward
and reverse harmonics on a full spectrum, depending on the number of scratch-
es and their spacing on the shaft.

Spectrum Waterfall Plots

While spectrum cascade plots are designed to display multiple spectra ver-
sus speed, from transient, startup or shutdown data, spectrum waterfall plots
are designed to display multiple spectra versus time, usually during constant
speed operation. Waterfall plots substitute time for rotor speed on the third axis
(although rotor speed may change). The horizontal and vertical relationships of
the cascade plots are maintained, but the diagonal relationships are, typically,
distorted; order lines may not be straight.
Waterfall plots are commonly used to examine how machine vibration
changes with a change in an operating parameter. The full spectrum waterfall
plot in Figure 8-11 shows the response of a compressor with a center seal insta-
bility problem. As suction pressure varied, vibration behavior changed dramati-
cally, from normal, mostly 1X behavior, to fluid-induced instability, and back
again. Orbits are shown for different suction pressure conditions. Note that the
full spectrum clearly shows that the subsynchronous vibration is predominant-
126 Data Plots

ly forward at that frequency; this is a strong indicator that the problem is a fluid-
induced instability.
At times, machines will be started, run for a short time, and then be shut
down. This is an ideal application for a waterfall plot. The waterfall plot can pro-
vide good visibility of these three different regions on one plot. Figure 8-12
shows a full spectrum waterfall of the startup, steady state, and shutdown of an
induction motor with an electrical noise problem. In the startup region (red),
the very high amplitude set of lines at ±60 Hz is line frequency noise. The ±1X
motor rotor vibration appears on the diagonal as the motor accelerates. When
the motor reaches operating the speed of around 3590 rpm, electrical power is
reduced, and the 1X rotor response and 60 Hz line frequency noise combine into
a single line (green region). When power is cut around 11:29, the ±60 Hz noise

Suction pressure: High Low High

12:30:08 14252

12:29:32 14316

12:28:34 14345

12:27:56 14338
1 mil pp/div
Amplitude

12:26:58 14296

12:26:05 14195

12:25:42 14168

12:24:53 13816

12:24:17 12465

12:23:36 11205

12:23:27 9888 rpm

-8 -6 -4 -2 0 2 4 6 8
Reverse Forward
Precession frequency (kcpm)

Figure 8-11. A Full Spectrum Waterfall plot. Waterfall plots display spectra with respect to time,
while the machine speed is either constant or changing. It is an excellent plot for correlating
response to operating parameters. The running speed is displayed on the right side of the plot,
and the amplitude scale is at the extreme right. In this example, a pipeline compressor has an
instability that appears when suction pressure is high (red). Significant full spectrums are identi-
fied with their orbits
 Chapter 8 Spectrum Plots 127

component disappears suddenly, leaving the ±1X coastdown vibration (blue).

The APHT plot from this machine showed a periodic ripple in amplitude and
phase caused by the beat frequency between the 60 Hz (3600 cpm) line noise and
the 1X rotor frequency at 3590 cpm. This waterfall plot proved that the problem
in the data was noise-related and not rotor-related.

Power
cut
11:30:16 380

+1X
X
11:29:50 970

–1
Coastdown
11:29:40 1870

11:29:35 3520

10:51:21 3582

1 mil pp/div
Amplitude
10:21:21 3591

09:51:21 3583
Operating
09:05:08 3584

08:25:08 3591

08:05:08 3584

07:48:14 X 2930
–1

Startup +1
X

07:48:08 604 rpm

-200 -100 0 100 200
60 Hz 60 Hz

Precession Frequency (Hz)

Figure 8-12. A full spectrum waterfall of an induction motor with an electrical noise problem. The
very high amplitude lines at ±60 Hz in the startup region (red) are line frequency noise. When the
motor reaches operating the speed of around 3590 rpm (green), electrical power is reduced, and
the ±1X lines become a mixture of 1X rotor response and 60 Hz line noise. When power is cut at
around 11:29, the 60 Hz lines disappear suddenly, leaving the 1X, coastdown vibration (blue).
128 Data Plots

Summary
The conventional (half) spectrum plot displays amplitude of vibration on
the vertical axis versus frequency of vibration on the horizontal axis. It is con-
structed using the sampled timebase waveform from a single transducer.
Spectrum plots can be used to identify the frequencies of running speed,
harmonics of running speed, sub- and supersynchronous vibration frequencies,
gear mesh frequencies, gear defect frequencies, rolling element bearing defect
frequencies, vane and blade pass frequencies, sidebands, glitch, and line fre-
quency noise.
The full spectrum uses the waveforms from an orthogonal pair of vibration
transducers (usually shaft relative). The full spectrum displays frequency and
direction of precession on the horizontal axis. Forward precession frequencies
are displayed to the right of the origin and reverse precession frequencies are
displayed to the left of the origin.
The full spectrum is the spectrum of an orbit, and the forward and reverse
frequency component pairs represent orbit components (filtered orbits). The
ratio of the amplitudes of full spectrum component pairs gives information
about the ellipticity and direction of precession of the components, important
characteristics for malfunction diagnosis. However, there is no information
about the orientation of the orbit.
Spectrum cascade plots are sets of spectra that are collected during the
startup or shutdown of a machine. Cascade plots can be constructed from either
half spectra, or full spectra. Cascade plots have important information associat-
ed with vertical, horizontal, and diagonal relationships.
Waterfall plots are collections of spectra obtained, usually, during steady
state operating conditions and plotted versus time. They also can use either half
spectra or full spectra.

The spectrum plot is a powerful tool when carefully applied. Because of its
wide availability, there is a temptation to use the spectrum plot to the exclusion
of other plot formats. But the spectrum, however powerful, is not a substitute
for the information that can only be obtained in other plots: the filtered ampli-
tude and phase in polar and Bode plots, the shaft position information in
Average Shaft Centerline plots, the shape and frequency information in the
Orbit, and the waveform information in the Timebase plot. All of this informa-
tion is needed for comprehensive machinery management.