Exergy graph-based fault detection and isolation of a gas-to-liquids process

With the sheer size of modern process plants, the Fault Detection and Isolation (FDI) field continues to gain popularity. FDI is a sophisticated concept which aims to detect and isolate anomalies that occur within a plant to avoid losses of personnel, damages to the environment, and financial implications. It does so in a way which is more direct, efficient and safer than what human operators are capable of. One approach to FDI is to consider the exergy characterisation of a system. By describing the exergy of the system units and streams, in this case a gas-to-liquids (GTL) process plant, the various process variables are encapsulated under a universal energy-domain parameter. The advantage of this being that it can describe the physical states as well as the chemical characteristics of the process. Previous work which utilised exergy characterisation along with a fixed-threshold approach showed promise. This study however, shows that the approach falls short when presented with 3 % faults. These results motivated the investigation of utilising attributed graphs, which package exergy data into a framework that preserves structural information of the plant. The usefulness of finding similarities (called graph matching) between the graphs constructed of operational conditions and pre-collected fault conditions to detect and isolate faulty conditions, is demonstrated. The technique performs well when considering fault magnitudes bigger than 8 % but deteriorates when applied to smaller, 3 % faults. The poor performance could be ascribed to the graph matching aspect, which is described by a single distance value that discards dimensionality. Future work will therefore look into the graph matching technique specifically, aiming to retain more informative dimensions. Copyright (C) 2020 The Authors.