Neurally-augmented Immnity-based Detection and Identification of Aircraft Sub-system Failures

This paper presents the development and testing through simulation of an integrated scheme for aircraft sub-system failure detection and identification (FDI) based on the artificial immune system (AIS) paradigm augmented with artificial neural networks. The features that define the self within the AIS paradigm include neural estimates of the angular accelerations produced by the abnormal conditions. The simulation environment integrates the NASA Generic Transport Model interfaced with FlightGear. A hierarchical multi-self strategy was investigated for developing FDI schemes capable of handling malfunctions of a variety of aircraft sub-systems. The performance of the FDI scheme has been evaluated in terms of false alarms and successful detection and identification over a wide flight envelope and for several actuator and aerodynamic surface failures. For all cases considered, the performance was very good, confirming the potential of the AIS paradigm augmented with the proposed neural network-based approach for feature definition to offer a comprehensive solution to the aircraft sub-system FDI problem.