Applied Sensor Fault Detection, Identification and Data Reconstruction

Abstract

Sensor fault detection and identification (SFD/I) has attracted considerable attentions in military applications, especially where the safety issues present priority drivers. This paper presents two readily implementable approaches for SFD/I in complex systems. Firstly, hierarchical clustering (HC) approach is demonstrated for use on gas turbines. SFD/I is achieved by monitoring cluster changes occurring in the resulting dendrograms, by comparing to HC fingerprints that are developed for ‘normal’ operation. In the second approach, using self organizing map neural networks (SOMNNs), SFD/I is performed according to the classification of changes from neural network outputs, and again comparing to fingerprints for ‘normal’ operation. The proposed methods are capable of detecting sensor faults from large groups of sensors. In contrast to many traditional methods of SFD/I that require the detection and identification of faulted sensors in multiple stages, the proposed methods are capable of achieving SFD/I in a single stage.  It is shown that the SOMNN classifications and the HC dendrograms provide commensurate detection and identification outputs, whilst also allowing for convenient operator use through graphical interpretations of their outputs.  Furthermore, after identifying a faulted sensor, it is possible in some circumstances to reconstruct the measurements expected from that sensor (using the remaining non-faulted sensor information) and thereby facilitate improved unit availability. The efficacy of the proposed approaches is demonstrated through the use of experimental measurements from gas turbines, although ultimately the underlying principles are applicable to other complex industrial and military systems. The presented techniques are now fully operational and monitoring a fleet of gas turbines in real-time.