A hypothesis generation model of initiating events for nuclear power plant operators

The goal of existing alarm-filtering models is to provide the operator with the most accurate assessment of patterns of annunciated alarms. Some models are based on event-tree analysis, such as DuPont's Diagnosis of Multiple Alarms. Other models focus on improving hypothesis generation by deemphasizing alarms not relevant to the current plant scenario. Many such models utilize the alarm filtering system as a basis of dynamic prioritization. The Lisp-based alarm analysis model presented in this paper was developed for the Advanced Controls Program at Oak Ridge National Laboratory to dynamically prioritize hypotheses via an AFS by incorporating an unannunciated alarm analysis with other plant-based concepts. The objective of this effort is to develop an alarm analysis model that would allow greater flexibility and more accurate hypothesis generation than the prototype fault diagnosis model utilized in the Integrated Reactor Operator/System (INTEROPS) model. INTEROPS is a time-based predictive model of the nuclear power plant operator, which utilizes alarm information in a manner similar to the human operator. This is achieved by recoding the knowledge base from the personal computer-based expert system shell to a common Lisp structure, providing the ability to easily modify both the manner in which the knowledge is structured as well as the logic by which the program performs fault diagnosis