Fault diagnosis for a class of nonlinear digital sensors based on output tracking

This paper focuses on fault diagnosis for a class of digital sensors. The first derivative and second derivative of these sensors' output signal under normal conditions will not involve a great jump due to physical limitations. It is similar to maneuvering targets which do not exhibit particularly jump in velocity and acceleration. So, a real-time random sensor fault diagnosis is transformed into a maneuvering target tracking problem. And a fault diagnosis method independent on system models is proposed. An improved unscented Kalman filter (UKF) is employed to track the output and estimate the value of various states. A mean-adaptive acceleration (MAA) model is established to find the faults of digital sensors online. According to the analysis of the failure characteristics in different sampling conditions, a method is proposed to isolate the faults. Theoretical analysis and experimental results show that the method can diagnose and isolate digital sensor fault accurately in real applications.