Answers are grounded in this paper and cite their sources.
The paper presents two approaches for fault detection and discrimination based on extensions of principal component analysis (PCA). The first approach proposes the concept of y-indices through use of a transposed formulation of the data matrices utilized in traditional PCA. The y-indices are introduced to measure the differences between sensor reading datasets in the ‘sensor domain’ rather than as traditionally employed in the time domain. This addresses problems associated with traditional PCA methods when measurement data is subject to bias and drifting due to transient power demand or load changes, for instance, which can lead to excessive false alarms. Residual errors (REs) and faulty sensor identification indices (FSIIs) are introduced in the second approach, where REs are generated from the residual sub-space of PCA, which is used to detect abnormal operating conditions, and FSIIs are introduced to supplement the REs in order to classify sensor- or component-faults. The methods are employed in a time rolling window on field data from a gas turbine system during commissioning. It is shown that in-operation sensor faults can be detected through use of both y-indices and REs and FSIIs, and sensor faults and component faults can be discriminated. The techniques are generic, and will find use in many military systems with complex, safety critical control and sensor arrangements.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2013) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: