Sensor fault diagnosis of a power plant: an approach based on state estimation techniques

The design of fault diagnosis and detection devices for the input and output sensors of dynamic systems requires the knowledge of an accurate mathematical model of the process since modeling uncertainty affects the sensitivity to the faults and increases the false-alarm probability. For this reason, the technique used in this paper gives weight to the identification procedure which exploits equation error and errors-in-variables models in connection with the values of the signal to noise ratios concerning the input and output measurements. The fault detection is performed by analyzing residuals, which are generated by a bank of dynamic observers and unknown input observers or, when the measurement noises are not negligible, by a bank of classical Kalman filters and Kalman filters with unknown inputs. The effectiveness of the procedure has been tested on real data acquired from the 120MW power plant of Pont sur Sambre.