Page 111 - Intro Predictive Maintenance

102 An Introduction to Predictive Maintenance

matically convert it using Fast Fourier Transform (FFT) to frequency-domain data. A
frequency-domain signature shows the machine’s individual frequency components,
or peaks.

While frequency-domain data analysis is much easier to learn than time-domain data
analysis, it cannot isolate and identify all incipient problems within the machine or its
installed system. Because of this limitation, 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.

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. Specifically, some of the commercially available analyzers
cannot be used where frequency components are below 600 cycles per minute (cpm)
or 10Hz.

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 floor” of the electronics
within the vibration analyzer tend to override the actual vibration components found
in low-speed machinery.

Analyzers especially equipped to handle noise are required for most industrial
applications. At least three commercially available microprocessor-based analyzers
are capable of acquiring data below 600cpm. These systems use special filters
and data acquisition techniques to separate real vibration frequencies from elec-
tronic noise. In addition, transducers with the required low-frequency response must
be used.

Averaging
All machine-trains are subject to random, nonrecurring vibrations as well as periodic
vibrations. 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
because only the continuous signals are retained.

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 seconds,
a four-sample average takes 12 to 20 seconds, and a 1,000-sample average takes 50
to 80 minutes to acquire. Therefore, the final determination is the amount of time that
can be spent at each measurement point. In general, three to four samples are accept-
able for good statistical averaging and keeping the time required per measurement
point within reason. Exceptions to this recommendation include low-speed machin-
ery, transient-event capture, and synchronous averaging.