Developing a branch and bound method for large post event state spaces under dynamic PRA assessment using LENDIT metrics and S2R2 sets

Traditional Probabilistic Risk Assessment (PRA) methods have been developed and are quite effective in evaluating risk associated with complex systems, but lack the capability to evaluate complex dynamic systems. These time and energy scales associated with the transient may vary as a function of transition time to a different physical state. Dynamic PRA (DPRA) methods provide a more rigorous analysis of complex dynamic systems, while complete, results in issues associated with combinatorial explosion. In order to address the combinatorial complexity arising from the number of possible state configurations and discretization of transition times, a characteristic scaling metric (LENDIT - length, energy, number, distribution, information and time) is proposed as a means to describe systems uniformly and thus provide means to describe relational constraints expected in the dynamics of a complex (coupled) systems. Thus when LENDIT is used to characterize four sets - 'state, system, resource and response' (S2R2) - describing reactor operations (normal and off-normal), LENDIT and S2R2 in combination have the potential to 'branch and bound (B&B)' the state space investigated by DPRA. In this paper we introduce the concept of LENDIT scales and S2R2 sets applied to a B&B algorithm and apply the methods to a station black out transient (SBO). In this work a relevant SBO of a Westinghouse, 4-loop PWR was considered using RELAP5 with the focus on methods development. Thus is not along conventional Level 1 nor 2, PSA We showed that variations in the LENDIT time scale can change the outcome as anticipated. Further concurrent use of energy and response set scales proved useful in post-event management. Specifically, in a feed-and bleed transient, success may be limited by the availability of emergency cooling that requires two high pressure injection pumps. Event tree branching should be limited to cases where these injection pumps are modeled. Thus the proposed method indeed provides a means to 'prune' the potentially large state spaces. The paper presents these results. (author)