Energy graph-based visualisation for FDI on a thermal power plant with heat storage capabilities

Abstract

This study explores the application of energy graph-based visualisation as a fault detection and isolation tool on the thermal energy storage test facility (THERESA), a complex water-steam system with integrated heat storage capabilities. Energy graph-based visualisation employs energy as a unifying parameter across physical domains to characterise a system using graph theory to maintain the system’s structural integrity. Key components are represented as nodes, with energy interactions between them represented as links, forming a system-attributed graph. The analysis involves computing energy and exergy values to characterise system behaviour, creating node signature matrices from time series data. A sensitivity analysis is conducted to compare two energy graph-based visualisation analysis methods: distance parameter approach and the sample-by-sample singular value decomposition approach. The sensitivity analysis is implemented through the use of matrix perturbations the results of which indicates that the sample-by-sample singular value decomposition method is more sensitive to changes in the node signature matrices. This study also demonstrates the possibility for fault detection and isolation from energy graph-based visualisation on the THERESA system and also implements the distance parameter method and the sample-by-sample singular value decomposition method as energy graph-based visualisation analysis methods. For fault detection and isolation (FDI), thresholds are developed from two normal operational data sets to evaluate a fault data set with two known faults. The results of the FDI performance indicate that the distance parameter approach is superior in fault detection with both faults detected with a true alarm rate of 99.34% with the sample-by-sample singular value decomposition approach utilising three singular values achieving a true alarm rate of 44.46% with singular value 1. The findings indicate that while the sample-by-sample singular value decomposition approach is more sensitive to changes in the system node signature matrix, the distance parameter approach provides better overall FDI performance, making it more suitable for fault detection in complex systems like THERESA.