Asymptotic Local Approach in Fault Detection Based on Predictive Filters

Abstract

Faults diagnosis techniques are important to improve the reliability of practical systems. A fault diagnosis technique is proposed based on a nonlinear predictive filter (PF) that estimates the modeling errors and uses them to correct the estimate of the system states. Under normal operation conditions, the modeling errors contain mainly the modeling uncertainties. However, if a component, an actuator, or a sensor fault occurs, the modeling errors will depart from the normal pattern. Therefore, fault detection is reduced to detecting irregularities in the modeling errors generated by the PF. The asymptotic local approach is used here to detect faults from the modeling errors because of its ability to detect small or incipient faults. The threshold for the fault diagnosis test can be obtained from the x 2 table for a selected confidence level. To demonstrate the performance of the proposed fault diagnosis technique, it is applied to detect abrupt and incipient faults in the satellite attitude measurement system.