FDI of process faults based on PCA and cluster analysis

TLDR: In this paper, a new approach to fault detection and isolation that combines Principal Component Analysis (PCA), Clustering and Pattern Recognition is presented, which is tested on experimental data from an integrated gasification and combined cycle (IGCC) section of an oil refinery plant to monitor a compression's process.

Abstract: A new approach to fault detection and isolation that combines Principal Component Analysis (PCA), Clustering and Pattern Recognition is presented. Single, multiple faults which may cause errors in the sensor readings and/or in the actuators as well as process faults are considered. Determination of the number of principal components is based on the statistical test ANOVA following the approach proposed by the authors in previous works. To overcome to the growth of complexity in the analysis of process faults that typically involve many variables, an automatic procedure for the isolation of the principal known faults has been developed. The proposed methodology which is based on Clustering and Pattern Recognition Analysis represents the new contribution of the present paper. The method is tested on experimental data from an IGCC (Integrated Gasification & Combined Cycle) section of an oil refinery plant to monitor a compression's process. Results show the goodness and effectiveness of the proposed approach on process faults detection and isolation.