Condition based maintenance (CBM) is based on
analysis and data collection monitored by sensors on the aircraft.
The knowledge discovery, about the performance of different
parameters, by using these data will provide new ways of
diagnosing and predicting the state of aircraft engines. However,
a single flight produces a huge amount of data that characterize
aircraft engine behaviour. The use of algorithms for the
simultaneous processing of these data is a difficult and sometimes
impossible task. The objective of this work is to choose the best
way to select instances for a sample. There should be no loss of
relevant information in the sample to identify the state of the
engine. We use five methods to select the sample and through
clustering techniques and sensibility analysis we choose the best
way to select the sample.