SFDA-Single Sensor Faults
In this chapter, a NN-based and EKF-based sensor fault detection and accommodation (SFDA) scheme are proposed and compared under different levels of unknown inputs and fault types. The schemes are tested on a nonlinear UAV model. The EKF is chosen as a representative of nonlinear (fixed) modelbased SFDA schemes which rely on a mathematical description of the real system. On the other hand, the NN is chosen due to its adaptive structure and online training capabilities. To test their robustness to unknown inputs, different levels of system and measurement noise are considered in the UAV model. Parameter uncertainties are also included in the EKF equations to investigate the performance of such methods to modelling errors.
KeywordsFalse Alarm False Alarm Rate Extend Kalman Filter Unknown Input Sensor Fault
Unable to display preview. Download preview PDF.