Page 291 - Intro Predictive Maintenance

282 An Introduction to Predictive Maintenance

resolution vibration data, incoming product characteristics, all pertinent process data,
and actual operating control parameters.

Vibration Data
For steady-state operation, high-resolution, single-channel vibration data can be used
to evaluate a system’s operating dynamics. If the system is subject to variables, such
as incoming production, operator control inputs, or changes in speed or load, multi-
channel, real-time data may be required to properly evaluate the system. In addition,
for systems that rely on timing or have components where response time or response
characteristics are critical to the process, these data should be augmented with time-
domain vibration data.

Data Normalization
In all cases, vibration data must be normalized to ensure proper interpretation. Without
a clear understanding of the actual operating envelope that was present when the vibra-
tion data were acquired, it is nearly impossible to interpret the data. Normalization is
required to eliminate the effects of process changes in the vibration profiles. 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 com-
parison of a series of profiles generated over time.

Regardless of the machine’s 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 (e.g., 5¥, 10¥), and
bearing-frequency locations are generally noninteger multiples (e.g., 0.5¥, 1.5¥). Plot-
ting 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. At a minimum, the
vibration data must be normalized to correct for changes in speed, load, and other
process variables.

Speed. When normalizing data for speed, all machines should be considered to be
variable-speed—even those classified as constant-speed. Speed changes caused by
load occur even with simple “constant-speed” machine-trains, such as electric-
motor–driven centrifugal pumps. Generally, the change is relatively minor (between
5 to 15 percent), 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 analyst’s 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