Adaptive fuzzy inference causal graph approach to fault detection and isolation of field devices in nuclear power plants

Abstract

An adaptive fuzzy inference causal graph is presented as an integrated approach for fault detection and isolation of field devices including sensors, actuators, and controllers in nuclear power plants. In this approach, nuclear plant systems are represented as a causal graph consisting of individual process variables connected with adaptive fuzzy inference system models. The adaptive fuzzy inference system models generated from historical data are used to characterize the relationships among process variables during normal operation. Fault detection and isolation is achieved by monitoring and cause-effect reasoning on the residuals. This unique cause-effect reasoning strategy for fault isolation can avoid seeking signatures patterns from fault data, which are usually very difficult to obtain for a large system. The most parsimonious model structure, which is a decisive factor in building robust data driven models, is achieved through the system decomposition that inherent in a causal graph. The developed approach has been demonstrated for a simulated pressurized water reactor steam generator system. Both simple faults and complex faults with fault propagation can be successfully isolated during their incipient fault stages, regardless of fault magnitudes and initial power level.