Probability of Detection (POD)-Oriented View to Fault Diagnosis for Reliability Assessment of FDI Approaches

Reliability assessment of Structural Health Monitoring systems applied to the diagnosis of faults in elastic structures is discussed in this contribution. In the field of Non Destructive Testing (NDT), Probability of Detection (POD) is used as a performance measure for quantifying the reliability of NDT approaches. However, reliability measures applied to Fault Detection and Isolation (FDI) is not discussed much in this research area. In this contribution the reliability of vibration-based monitoring approaches with respect to their principal ability to detect changes, realize diagnosis, and isolate causes (as FDI) is discussed. Using eigenfrequency and band power as attributes, a novel feature-based POD is proposed and implemented as a reliability measure for vibration-based FDI. The a_90/95 criteria which represents 90% probability of detecting a fault with 95% level of confidence is successfully implemented to an experimentally realized monitoring system. Emphases is made on improving the detection quality through sensor/information fusion. Acceleration, strain, and deflection measurement paths are utilized for diagnosis purposes based on experimental results.