Page 292 - Intro Predictive Maintenance

Operating Dynamics Analysis 283

assigned a speed of 1,780 revolutions per minute (rpm) during setup. The analyst then
assumes that all data sets were acquired at this speed. In actual practice, however, the
motor’s 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
1,750rpm and 1,800rpm, a difference of 50rpm. This variation is enough to distort
data normalized to 1,780rpm. 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 micro-
processor-based systems to determine the machine speed to use for data normaliza-
tion. 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 software program. Because the cursor position does not represent the true fre-
quency of displayed peaks, it cannot be used. The displayed cursor position is an
average value. The graphics packages in most of the programs use an average of four
or five data points to plot each visible peak. This technique is acceptable for most data
analysis purposes, but it can skew the results if used to normalize the data. The ap-
proximate machine speed obtained from such a plot is usually within 10 percent 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 amplitude. For example, the vibration amplitude of a centrifugal compressor taken
at 100 percent load is substantially lower than the vibration amplitude in the same
compressor operating at 50 percent 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 percent with a corresponding 50
percent load variation. Instead, it tends to follow more of a quadratic relationship.
A 50 percent load variation can create a 200 percent, or a factor of four, change in
vibration energy.

None of the comparative trending or diagnostic techniques used by traditional vibra-
tion analysis can be used on variable-load machine-trains without first normalizing
the data. Again, since even machines classified as constant-load operate in a variable-
load condition, it is good practice to normalize all data to compensate for load varia-
tions using the proper relationship for the application.

Other Process Variables. Other variations in a process or system have a direct effect
on the operating dynamics and vibration profile of the machinery. In addition to
changes in speed and load, other process variables affect the stability of the rotating
elements, induce abnormal distribution of loads, and cause a variety of other abnor-
malities that directly impact diagnostics. Therefore, each acquired data set should