Evaluation of a Condition Monitoring Method's Fault Detection Reliability for Condition-Based Maintenance Applications

摘要

When implementing Condition-Based Maintenance (CBM) for safety-critical systems such as aircraft, it is necessary to demonstrate the effectivity of the applied Condition Monitoring System (CMS) in detecting potential faults before failure. A quantitative fault detection reliability assessment for a given monitoring method and fault type, adapted from the field of Non-Destructive Testing (NDT), is proposed. It involves first modeling the relationship between the Condition Indicator values obtained with the CMS, for example vibration levels, and the corresponding fault severity in a signal response model. After defining an acceptable Probability of False Alarm (PoFA), the Probability of Detection (PoD) of the fault can be computed from the signal response model as a function of fault severity. Vibration monitoring presents particular challenges with regard to fault detection reliability, due to the high variability often encountered in vibration measurements. Further, a large number of individual Condition Indicators are typically computed from a vibration measurement and used for diagnostics. To systematically compare the detection performance of various indicators and to identify those indicators with the most significant response to a given fault, a discriminability measure derived from the signal response model is introduced. The fault detection reliability evaluation is demonstrated with vibration monitoring data from a series of Seeded Fault Tests (SFT) on helicopter tail rotor drive shaft bearings. Multiple bearings were first tested in an undamaged state, and then retested after the insertion of a local defect on the bearing inner ring race. The size of the defects was artificially increased in several steps, to obtain data over a range of fault severities. The evaluation shows that a Condition Indicator intended to detect bearing inner race faults indeed exhibits the most significant signal response, and achieves a high PoD at small defect sizes.