Fault tolerant adaptive estimation of nonlinear processes using redundant measurements

This paper presents a dual layer approach for robust fault tolerant estimation of nonlinear processes using a combined adaptive extended Kalman filter and fault detection and filter reconfiguration. From the one hand, the filter is made robust in face of environment uncertainty using adaptive filtering. To this end, the filter identifies the measurement covariance by means of recursive estimation, upon which the adaptation relies, to suppress the effect of sporadic variations in the quality of measurements as well as compensates for incipient sensor faults. From the other hand, fault monitoring is continuously applied to the filter's innovation in an attempt to initiate filter reconfiguration when the adaptation mechanism alone is not able to overcome the failure situation. The discussion of the results is embedded in the application framework of state estimation of a batch distillation process.