Fault Class Identification Through Applied Data Mining of AH-64 Condition Indicators

The current deployment of Health and Usage Monitoring Systems (HUMS) on US Army rotorcraft has generated large volumes of mechanical vibration and other forms of data in support of Condition-Based Maintenance (CBM) practices, yet relatively little study has been performed on the collected datasets as a whole. Vibration data collected from 54 AH-64 helicopters was analyzed to determine whether a data mining approach to fault type classification was possible. Several characteristics of the vibration data were examined, and it was found that approximately 40% of the information contained within the current set of condition indicators is redundant. However, he precision and repeatability of the measurements was found to be sufficiently high that a single acquisition was able to uniquely identify the aircraft from which it came 85% of the time. Since there was not sufficient historical information to create a labeled training data set, a fault class identifier was trained on laboratory test data in which three AH-64 tail rotor gearboxes failed under identical fault conditions. When cross validation was performed, the classifier was able to predict the gearbox life fraction with a correlation coefficient as high as 0.8859; however, the results were less accurate when testing the same classifier on healthy gearbox data or actual aircraft data.