Fault monitoring and diagnosis

A technique for both monitoring and diagnosing a range of subtle faults in an arbitrary dynamic system is presented. With minor modification, the technique can be extended to encompass sensor fault diagnosis and is well suited to accommodation within a real-time Al management system. It is based on signal processing of signals from conventional sensors taking measurements from a dynamic system. This ensures that no disruption is applied to the system in fitting conditioning sensors. The method is based on the techniques of system identification, Kalman filtering and statistical classification in which the faults are diagnosed by checking the Kalman filter residuals against previously learned patterns