Abstract

The paper presents two readily implementable approaches for Sensor Fault Detection, Identification (SFD/I) and faulted sensor data reconstruction in complex systems, in real-time. Specifically, Principal Component Analysis (PCA) and Self-Organizing Map Neural Networks (SOMNNs) are demonstrated for use on industrial turbine systems. In the first approach, Squared Prediction Error (SPE) based on the PCA residual space is used for SFD. SPE contribution plot is employed for SFI. A missing value approach from an extension of PCA is applied for faulted sensor data reconstruction. In the second approach, SFD is performed by SOMNN based Estimation Error (EE), and SFI is achieved by EE contribution plot. Data reconstruction is based on an extension of the SOMNN algorithm. The results are compared in each examining stage. The validation of both approaches is demonstrated through experimental data during the commissioning of an industrial 15MW turbine.