Fault Diagnosis of a Sensor Network: A Distributed Filtering Approach
This paper proposes a model-based approach to develop a novel fault diagnosis scheme for a sensor network of a cascade, parallel and feedback combination of subsystems. The objective is to detect and isolate a fault in any of the subsystems and measurement sensors which are subject to disturbances and/or measurement noise. Our approach hinges on the use of a bank of Kalman filters (KF) to detect and isolate faults. Each KF is driven by either a pair (a) of consecutive sensor measurements or (b) of a reference input and a measurement. It is shown that the KF residual is a reliable indicator of a fault in subsystems and sensors located in the path between the pair of the KF's input. The simple and efficient procedure proposed here analyzes each of the associated paths and leads to both the detection and isolation of any fault that occurred in the paths analyzed. The scheme is successfully evaluated on several simulated examples and on a physical fluid system exemplified by a benchmarked laboratory-scale two-tank system to detect and isolate faults including sensor, actuator and leakage ones.