Graph-based exergy FDI sensitivity analysis on the Tennessee Eastman process

Abstract

This study presents work on identifying the location and options possible for functional parameters associated with each energy graph-based visualisation (EGBV) method. Each of the EGBV methods were analysed to determine whether sensitivity to functional parameters exists during FDI of fault conditions using the Tennessee Eastman process (TEP) model. Additionally, validation of this sensitivity analysis was performed to confirm if initial results held true under different model operation modes. Literature was studied on existing Heterogeneous Euclidean-Overlap Metric (HEOM) based distance functions as well as previous implementations of the respective EGBV methods to identify possible alternative options for each functional parameter. Following the identification and compiling a structure of functional parameters for each EGBV method, the Tennessee Eastman process model was implemented to acquire data which demonstrate a range of various fault conditions during system operation. The structure utilised to acquire multiple datasets at the default base case mode later served useful in providing an opportunity to acquire datasets for validation purposes. Initial datasets that were acquired were utilised in transforming process measurements into the attributed graphs utilised for fault detection and isolation (FDI) of operational conditions observed. Each EGBV method was implemented as a range of functional parameter permutations in delivering FDI results that were quantitatively measured and analysed to identify possible sensitivity demonstrated. The execution of the functional parameter permutations for each EGBV method delivered gave a clear indication of the impact demonstrated by functional parameter selection on each category of performance measured. A clear deviation in performance was observed when comparing alternate functional parameter options to the default methodology for each EGBV method. The sensitivity analysis further confirmed that sensitivity in performance was demonstrated towards the functional parameter locations investigated. The validation of the initial sensitivity analysis performed consisted in reconfiguring the TEP model to mode 5 operation, in presenting new conditions for sensitivity to be observed. This validation delivered both a confirmation of sensitivity and new information on how sensitivity demonstrated by the EGBV methods varied between model operation modes. This study presents evidence to the deviation in FDI performance resulting from EGBV method functional parameters and sensitivity in performance demonstrated towards each functional parameter. This sensitivity is validated for a single model application, and presents an initial case to investigate functional parameters at a larger scale in future research.