On-line Learning RBF Neural Networks for Sensor Validation

A Sensor Failure Detection, Identification and Accommodation (SFDIA) scheme for a flight control system with dual physical redundancy is proposed, with the focus on taking advantage of analytical redundancy for sensor failure purposes. The proposed scheme, based on a set of neural on-line approximators, has shown to be able to provide desirable SFDIA capabilities. The designed approximators are based on either static or recurrent Radial Basis Function Neural Networks featuring stable online learning capabilities. Embedded filters are also introduced in the approximator structure as an effective way of increasing the sensitivity to sensor failure while attenuating sensitivity to noise/disturbance. The schemes have been simulated on a non-linear Simulink model mathematical model of a high performance aircraft developed at West Virginia University from a simulation code distributed for the 1991 GNC Design Challenge.