Fault detection using dynamic principal component analysis by average estimation

This paper presents a fault detection method based on pattern classification in which the stationary property of the time series of a dynamic system is not required. In particular the dynamic principal component analysis, DPCA, method for feature extraction in diagnostics issues is extended to the case of nonstationary data. The idea is to improve the DPCA performance introducing on-line average estimations of the input and output of the dynamic systems. These new parameters of the data allow extending the method for no stationary time series reducing the false alarms during the detection stage. As study case the detection of faults in a flow control valve has been used, in which it is assumed that the control signal and stem displacement are measured signals. Simulator Data has been used to adjust the procedure and show the effectiveness of the proposed methodology.