Maturation of Detection Functions by Performances Benchmark. Application to a PHM Algorithm
Abstract
To monitor the start sequence of turbofan engines, a PHM algorithm has been developed. It is called ESC for Engine Start Capability. The anomaly detection phase is performed using a global abnormality criterion. This criterion is obtained by computing the Mahalanobis-distance on standardized residuals (z-scores). In the maturation (performance assessment) perspective of the ESC PHM algorithm, two detection approaches are compared. The first approach (approach 1) is based on the crossing of a threshold determined from the moving average over n starts of the global abnormality criterion. The second approach (approach 2) is based on a minimum number (k) of the global abnormality criterion observations that cross a threshold out of a fixed one (n). The two thresholds are respectively determined in accordance with the two detection approaches. The performances of each approach have been estimated, in terms of probability of detection, P(Detection|Degradation), and No Fault Found ratio (NFF), P(Healthy|Detection).