Sensor fault diagnosis in water-steam power plant: A combined observer-based/pattern-recognition approach

This work deals with the problem of model-based sensor FDI in water-steam power plants where, due to extreme pressure and temperature conditions, measurement sensors are prone to failures. Faults in the measurement devices of output variables (water flow and level) and of input variable (steam flow) are considered. When both the output and input measurements are subject to faults it is hard to detect and estimate them. To overcome this limitation and achieve FDI, we propose to use a sliding mode observer (SMO) and to make an appropriate signature analysis on the resulting output injection terms in order to identify a “distinguishing” signature for each fault. The performance of the proposed scheme has been evaluated off-line using real-data taken from the TEKO B1 Thermal Power Plant of Kostolac (Serbia) whose nominal power is 330 MW.