Adaptive Neuro-Fuzzy Based Sensor Fault Detection,Isolation and Reconfiguration

Sensor fault detection is performed by estimating the states of plant and comparing them with measured values. A faultis detected when the difference between the estimated and measured values crosses a threshold value. In this paper, an approach to detect and isolate the sensor fault affecting the mean of the Kalman filter innovation sequence is presented. Reconfiguration of isolated sensor faults is carried out using twomethods viz., i) by replacing faulty sensor by an extra healthy sensor(hardware redundancy) and ii) by ignoring measurement from faulty sensor(analytical redundancy). Also the sensor fault detection, isolation and reconfiguration is demonstrated using Adaptive-Neuro Fuzzy Inference System - non model based approach. In the simulation, the longitudinal dynamics of an aircraft control system are considered and the detection and isolation of pitch rate gyroscope faults affecting mean of the innovation sequence are studied. Reconfiguration for sensor faults is carried out so that the estimator provides the controller with an accurate estimate of the system state after the sensor failure.