Page 352 - Intro Predictive Maintenance

Establishing a Predictive Maintenance Program 343

Transducers

The final portion of a predictive maintenance system is the transducer that will be
used to acquire data from plant equipment. Becaise we have assumed that a micro-
processor-based system will be used, we will limit this discussion to those sensors that
can be used with this type of system.

Acquiring accurate vibration and process data will require several types of transduc-
ers. Therefore, the system must be capable of accepting input from as many different
types of transducers as possible. Any restriction of compatible transducers can become
a serious limiting factor. This should eliminate systems that will accept inputs from a
single type of transducer. Other systems are limited to a relatively small range of trans-
ducers that will also prohibit maximum utilization of the system. Selection of the spe-
cific transducers required to monitor the mechanical condition (e.g., vibration, flow,
pressure) also deserves special consideration and will be discussed later.

15.5 DATABASE DEVELOPMENT
Each of the predictive maintenance technologies requires a logical method of acquir-
ing, storing, evaluating, and trending massive amounts of data over an extended
period. Therefore, a comprehensive database that is based on the actual requirements
of critical plant systems must be developed for the predictive maintenance program.
At a minimum, these databases should include the following capabilities:

• Establishing data acquisition frequency
• Setting up analysis parameters
• Setting boundaries for signature analysis
• Defining alert and alarm limits
• Selecting transducers

15.5.1 Establishing Data Acquisition Frequency
During the implementation stage of a predictive maintenance program, all classes of
machinery should be monitored to establish a valid baseline data set. Full vibration
signatures should be acquired to verify the accuracy of the database setup and deter-
mine the initial operating condition of the machinery. Because a comprehensive
program will include trending and projected time-to-failure, multiple readings are
required on all machinery to provide sufficient data for the microprocessor to develop
trend statistics. During this phase, measurements are usually acquired every two
weeks.

After the initial or baseline evaluation of the machinery, the frequency of data col-
lection will vary depending on the classification of the machine-trains. Class I
machines should be monitored on a two- to three-week cycle; Class II on a three- to
four-week cycle; Class III on a four- to six-week cycle; and Class IV on a six- to ten-
week cycle. This frequency can, and should, be adjusted for the actual condition of