A functional modelling based methodology for testing the predictions of fault detection and identification systems

Fault detection and identification (FDI) systems, which are based on data mining and artificial intelligence techniques, cannot guarantee a perfect success rate or provide analytical proof for their predictions. This characteristic is problematic when such an FDI system is monitoring a safety-critical process. In these cases, the predictions of the FDI system need to be verified by other means, such as tests on the process, to increase trust in the diagnosis. This paper contributes an extension of the Hierarchical Functional Fault Detection and Identification (HFFDI) system, a combination of a plant-wide and multiple function-specific FDI modules, developed in past research. A test preparation and test-based verification phase is added to the HFFDI methodology. The functional decomposition of the process and the type of the faulty components guides the preparation of specific tests for every fault to be identifiable by the HFFDI system. These tests have the potential to confirm or disprove the existence of the fault(s) in the target process. The target is minor automation faults in redundant systems of the monitored process. The proposed extension of the HFFDI system is applied to a case study of a generic Nuclear Power Plant model. Two HFFDI predictions are tested (a successful and an incorrect prediction) in single fault scenarios and one prediction is tested in a in a two fault scenario. The results of the case study show that the testing phase introduced in this paper is able to confirm correct fault predictions and reject incorrect fault predictions, thus the HFFDI extension presented here improves the confidence of the HFFDI output.