Rapid detection of faults for safety critical aircraft operation

Fault diagnosis typically assumes a sufficiently large fault signature and enough time for a reliable decision to be reached. However, for a class of safety critical faults on commercial aircraft engines, prompt detection is paramount within a millisecond range to allow accommodation to avert undesired engine behavior. At the same time, false positives must be avoided to prevent inappropriate control action. To address these issues, several advanced features were developed that operate on the residuals of a model based detection scheme. We show that these features pick up system changes reliably within the required time. A bank of binary classifiers determines the presence of the fault as determined by a maximum likelihood hypothesis test. We show performance results for four different faults at various levels of severity and show performance results throughout the entire flight envelope on a high fidelity aircraft engine model.