151

Chapter 9

Trend and XY Plots

I n Chapter {7} we discussed the Amplitude-PHase-Time (APHT) plots, a form

of polar and Bode plots, which are used for trending vector data. Besides vec-
tor data, there are many other parameters that we would like to trend, vibration
levels, position data, process data, or any other parameter that can be useful for
machine condition monitoring. The Trend plot shows changes in and the rate of
change of parameters that may signal a developing or impending problem in a
machine. This information can be used to set limits or thresholds for action.
We may also wish to examine how any of these variables change with respect
to others. Such correlation is the heart of a good diagnostics methodology. No
single variable or plot type can reveal everything about a machine. Data is usu-
ally correlated with many other pieces of information to arrive at a diagnosis.
This kind of correlation is often done with multiple trend plots, where several
variables are plotted against the same time scale.
Correlation can also be done with XY plots, where two parameters are plot-
ted against each other.
This chapter will deal with the construction and uses of the Trend plot and
provide several examples. We will briefly discuss problems that can arise when
the data sample rate is too low. Finally, we will discuss XY plots, a special type of
trend plot used for correlation of two parameters.

Trend Plots
The Trend plot is a simple, rectangular plot on which the value of a meas-
ured parameter is plotted versus time. Trend plots can be used to display any
kind of data versus time: direct vibration, nX amplitude, nX phase (the APHT
152 Data Plots

plot is a trend plot that displays both), gap voltage (radial or thrust position),
rotor speed, and process variables, such as pressure, temperature, flow, or power.
Trend plots are used to detect changes in these important parameters. They are
used for both long and short term monitoring of machinery in all types of serv-
ice and are, typically, an example of a steady state (constant speed) plot.
The data for a trend plot can be collected by computer or by hand. Figure 9-
1 shows a trend plot of hand-logged gap voltage from a fluid-film bearing at the
discharge end of a refrigeration compressor. Due to improper grounding, elec-
trostatic discharge gradually eroded 280 µm (11 mil) of the bearing, allowing the
rotor shaft to slowly move into the babbitt. The trend plot alerted the operators
to the fact that something was wrong, and they scheduled a shutdown in time
to prevent serious damage. This is an example of how a very simple data set pro-
vided valuable information that saved the plant from an expensive failure.
Even though a trend plot may look like a continuous history, it is not.
Parameters are assumed to be slowly changing, so the data to be trended is sam-
pled at intervals that depend on the importance of the machinery and the data.
If a sudden change in behavior of the parameter occurs between samples, the
data will be missed.
Some data values may fluctuate periodically. For example, 1X amplitude and
phase may change periodically due to a thermal rub. The period of change for
this kind of malfunction can be on the order of minutes to hours. Amplitude and
phase modulation can occur in induction motors, due to uneven air gap, at twice

Figure 9-1. A trend of hand-logged gap

Vertical position (mil)

voltage from a fluid-film bearing at the 0

discharge end of a refrigeration compres- Scheduled shutdown
sor. Due to improper grounding, electro- -5
static discharge gradually eroded 280 µm
-10
(11 mil) of the bearing, allowing the rotor
shaft to slowly move into the babbitt. The -15 Bearing backing
trend plot provided early warning of the
problem and identified the rate of
0 5 10 15 20 25
change, allowing a scheduled shutdown.
Time (days)
See Reference [1] for a detailed discussion
of this machine problem.
 Chapter 9 Trend and XY Plots 153

the slip frequency; the period here is usually a fraction of a second. If the sam-
pling frequency is less than twice the frequency of interest (does not satisfy the
Nyquist criterion), then the frequency of the changes in the trend plot will be
incorrect, an effect known as aliasing.
The trend plot from an induction motor (Figure 9-2) looks like, at first
glance, a timebase plot. However, it is a trend of unfiltered, peak-to-peak vibra-
tion that is changing periodically. This motor, which drives a boiler feed pump,
has an uneven air gap problem. The vibration amplitude is modulated at a beat
frequency equal to twice the slip frequency of the motor.
The data in blue was sampled very rapidly, at about 10 samples per second.
This produced a high resolution trend plot. The data in red is a portion of anoth-
er trend plot from about two hours earlier, when the motor was experiencing the
same problem. The sample rate was one sample every 10 seconds, a factor of 100
slower. Note that the frequency of the change of the red is much lower, the result
of aliasing. This would not be obvious unless the data taken at the higher sam-
ple rate was available. The observed modulation frequency is much lower than
the true modulation frequency. Parameters which change periodically like this
are relatively rare, but this example demonstrates how the sample rate can pro-
duce a misleading picture of machine behavior. This effect can also happen
when, to increase the trend interval on the horizontal axis, the database of sam-
pled data is decimated (samples are thrown out).

Figure 9-2. Aliasing in a trend plot. This is

not a timebase plot! This boiler feed
Unfiltered
pump motor has an uneven air gap. The 3
vibration amplitude is modulated at a
Amplitude (mil pp)

beat frequency equal to twice the slip fre-

2
quency of the motor. The data in blue was
sampled at about 10 samples per second.
The data in red is a portion of a trend plot 1
that was sampled about two hours earlier,
at one sample every 10 seconds, a factor
of 100 slower. When the two trends are 0
11:33:20 :30 :40 :50 11:34:00
compared, it is clear that the red data is
Time
aliased; the sample rate was not fast
enough to accurately capture the changes
in the vibration.
154 Data Plots

Multiple parameters can also be displayed on the same plot, or on several

plots with the same time scale, to correlate changes that occur in a machine.
Figure 9-3 shows a trend plot of a compressor with a center seal instability prob-
lem. Both 1X (blue) and unfiltered (red) vibration amplitude are shown on the
same plot and to the same scale. While the machine was running at full speed,
the suction pressure was varied, and a fluid-induced instability appeared or dis-
appeared depending on the level of suction pressure. Note that the 1X vibration
decreased when the unfiltered vibration increased. This can happen when high
amplitude, Not-1X vibration moves the rotor to a region of where the Dynamic
Stiffness is different, changing the 1X response. A plot of the actual pressure
data, which was unavailable, would highlight the correlation of the variables.
Figure 9-4 shows trend data from a 30 000 hp, synchronous electric motor
driving an axial flow air compressor. The plot shows vibration amplitude from a
displacement probe (blue, right scale) and a velocity transducer (green, left
scale). This motor had a weak foundation system that slowly loosened, allowing
the vibration amplitude, measured by both transducers, to slowly increase.
When sets of chock wedges were installed and the foundation bolts retightened
around 30 August, the vibration dropped sharply, followed by another slow rise.
Near the end of the record, the machine was shut down, and the foundation was
completely overhauled. The motor itself was found to have problems, and the
unit was replaced. After restarting, the vibration returned to very low levels.

XY Plots
Where the Trend plot displays one or more parameters versus time, the XY
plot (not to be confused with XY axes or XY transducers) can be used to display
any two parameters versus each other. Correlations between the parameters will
show a diagonal relationship. A complete lack of correlation will show either a
horizontal or vertical relationship.
Figure 9-5 shows an XY plot of vibration amplitude versus gap voltage from
a 125 MW steam turbine generator, HP/IP unit running at 3600 rpm. The plot
starts at point 1 when the machine undergoes a load change. Between points 1
and 2 (red), the plot clearly shows a correlation between changing shaft position
(measured by the gap voltage) and 1X vibration amplitude. As the shaft moves
(blue) to point 3 during the next three hours, the vibration decreases while the
gap voltage remains approximately constant. It is possible that a gravity bow
may be working itself out or the measurement probe may be viewing a different
section of the shaft as the machine reaches thermal equilibrium. The shaft takes
only about 35 minutes to move from point 3 to point 4 (green), most likely in
response to another load change.
 Chapter 9 Trend and XY Plots 155

Figure 9-3. A trend plot of a compressor Suction

pressure
with a center-seal instability problem. Both
1X (blue) and unfiltered (red) vibration 2

Amplitude (mil pp)

amplitudes are shown on the same plot, at
the same scale. While the machine was run- Unfiltered

ning at full speed, the suction pressure var-

ied; a fluid-induced instability appeared and 1
disappeared, depending on the level of suc- 1X

tion pressure. Note that the 1X vibration

decreased when the unfiltered vibration 0
increased (see text for further explanation). 12:26 12:27 12:28 12:29 12:30
Time

0.5 2.0
Chock wedges inserted
Displacement

0.4 1.6

Displacement (mil pp)

Velocity (in/s pk)

0.3 After 1.2

foundation
repair
0.2 0.8

0.1 0.4
Velocity

0.0 0.0
31-Jan

27-Mar

25-Apr

10-May

23-May

12-Jun

28-Jun

14-Jul

27-Jul

15-Aug

30-Aug

12-Oct

30-Nov

05-Feb

15-Apr

22-Jul

23-Sep

Date

Figure 9-4. Trend data from a 30 000 hp, synchronous electric motor driving
an axial flow air compressor. The plot shows vibration amplitude from a dis-
placement probe (blue, right scale) and a velocity transducer (green, left
scale). This motor had a weak foundation system that slowly loosened. The
vibration amplitude slowly increased on both transducers until around 30
August, when sets of chock wedges were installed and foundation bolts
retightened. The vibration shows a sharp drop, followed by a slow rise. Near
the end of the record, the machine was shut down, and the foundation was
completely overhauled.
156 Data Plots

Figure 9-5. An XY plot of 1X vibration

amplitude versus gap voltage from a 125
MW steam turbine generator, HP/IP unit
running at 3600 rpm. The plot clearly 4.0 2
shows (red) the correlation between 10:53

Amplitude (mil pp)

changing shaft position (measured by the 3.5
gap voltage) and an increase in1X vibra-
tion amplitude when the load changes. As 3.0 1
10:28
the shaft moves to point 3 during the next 4
14:33
three hours, the vibration decreases while 2.5
3
the gap voltage remains approximately 13:58
constant (no correlation). After that, the 2.0
shaft takes only about 35 minutes to move -9 -8 -7 -6 -5 -4
Gap (Vdc)
to point 4 (green), most likely in response
to another load change.

Summary
The Trend plot is a rectangular plot of a measured parameter versus time.
Trend plots can be used to display any kind of data versus time: direct vibration,
nX amplitude, nX phase, gap voltage, rotor speed, and process variables such as
pressure, temperature, flow, or power. Data can be correlated by plotting several
variables on the same time scale.
Trended data is sampled at periodic intervals, and it is assumed to be slow-
ly changing. If data values can change periodically, the sampling frequency must
be at least twice the frequency of change (the Nyquist criterion). If not, the data
will be aliased; the displayed amplitude and frequency of the trend data will not
be correct.
The XY plot can be used to display any two parameters against each other.
Correlations between the parameters will show a diagonal relationship. A com-
plete lack of correlation will show either a horizontal or vertical relationship.

References
1. Eisenmann, Robert C., Sr., and Eisenmann, Robert C., Jr., Machinery
Malfunction Diagnosis and Correction (Upper Saddle River: Prentice-Hall,
Inc., 1998), pp. 751-758.