Methodology for modeling system evolution to support decisionmaking with an application in severe accident management

A new methodology is developed for modeling a system's evolution to support decisionmaking on alternative actions which affect the system's evolution. The methodology is based upon the goal tree success tree method for hierarchal decomposition of the system's objective. The goal tree success tree structure is used to identify the system's objective and supporting functions and the events and actions which affect realization of these functions. The events are logically arranged using an event tree format and potential event sequences and end states are identified. The probability of the occurrence of the events are determined via condition-based logical diagrams (CLDs) and functional logic diagrams (FLDs) which model the propagation of binary logic (in the case of CLDs) and incorporate equations representing the more complex hierarchy process is then used to evaluate the alternative actions affecting the end states and provide a recommendation on the preferred actions for meeting the system objective. The CLDs and FLDs provide a mechanism for modeling diverse knowledge, i.e., deterministic, probabilistic, logic based, as well as temporal, in a uniform manner. This methodology is used to develop a model for analyzing severe accident phenomena and alternative actions for mitigating a severe accident at a nuclear power plant. This severe accident model is then used as the knowledge base for a real-time expert system. Tests run using this expert system demonstrate its capability to calculate the probability of various accident progression end states and provide recommendations on the priority for implementing severe accident mitigation strategies. In addition, research was performed to evaluate the use of the FLD concept to model containment failure from direct containment heating as analyzed using the Risk Oriented Accident Analysis Methodology. A system of hierarchically linked FLDs were developed which illustrate the relative effect of input parameters on intermediate and anal level parameters. This research has illustrated the usefulness of this modeling methodology to integrate diverse knowledge in a manner useful in supporting decisionmaking.