Graph-based fault detection for a gas-to-liquids process: an exergy approach

Abstract

As there are many safety and financial risks within modern process plants, process monitoring is said to be indispensable. Process monitoring aids operators in ensuring reliable and efficient operation of the plant. Fault detection and Isolation (FDI), which make up a large portion of a process monitoring protocol, is a sophisticated scheme which aims to detect and isolate anomalies that occur within the plant. For the past 50+ years, much work has been done on developing FDI schemes for a vast array of different applications. In recent years, novel energy-based FDI techniques were proposed, as energy is seen as a unifying parameter of different domains. These energy-based approaches also endeavour to capture causal (or structural) information of the physical system. Keeping with this theme, this study will determine, after some alterations, the applicability and performance of some of the previously developed energy-based approaches, especially compared to one another, when applied to a single, larger-scale petrochemical process. The petrochemical process, a gas-to-liquids (GTL) process, is not seen within the FDI literature, and could arguably be well-suited to being used as a benchmark to evaluate the performance of proposed FDI schemes. A such, this study systematically documents the process specifics and modelling effort to allow easy recreation thereof. As the model was simulated within the commercial process simulator, Aspen HYSYS®, and the FDI approaches require energy data, user variables were created to compute the desired energy data automatically. The different techniques investigated were a fixed-threshold approach, a graph matching approach using a distance parameter, and graph matching approach utilising eigendecomposition. The approaches and their methodologies are shown and applied accordingly; using the same normal and faulty energy data of the GTL model. The results are then interpreted in terms of the approaches’ ability to detect and isolate the pre-defined faults. Finally, the performance of the approaches is compared to determine the best performing technique. This study shows the degrees of applicability of the examined energy-based FDI approaches. It also found the graph-based approach utilising the distance parameter showing the most promise, as fault locations could be distinguished. This, therefore, confirms not only the usefulness of expressing the system in terms of energy but that structural information is also retained.