Current Status of the SAPHIRE Risk Analysis Software

Abstract

As a supporting organization for risk activities at a variety of government agencies including the U.S. Nuclear Regulatory Commission (NRC), National Aeronautics and Space Administration (NASA), and Department of Energy (DOE), the Idaho National Laboratory (INL) has been a principal developer of probabilistic risk and reliability analysis tools for over 35 years. The current state-of-the-practice has evolved to the use of the SAPHIRE software. This tool started in the mid-1980s as part of the NRC’s general risk activities. In 1986, work commenced on the precursor to the SAPHIRE software — this software package was named IRRAS. While limited to the analysis of only fault trees of medium size, version 1 of IRRAS was the initial step in the progress that today has led to the SAPHIRE software, software that is capable of running on multiple processors simultaneously and is able to handle extremely complex analyses. SAPHIRE has been designed to handle large fault trees, where a tree may have up to 64,000 basic events and gates. To handle the fault trees, two mechanisms for developing and modifying the fault tree are available — a graphical editor and a hierarchical logic editor. In risk applications, there are two predominant event tree analysis methods, the “large event tree” approach and the “fault tree linking” approach. A couple of key identifying attributes of the large event tree approach is that the number of accident sequences becomes very large (measured in the millions or more) and the event tree branches are represented by probability values. For the fault tree linking approach, the number of sequences is low and fault trees represent the event tree branch points. The INL has designed SAPHIRE to handle both event tree methods. SAPHIRE uses logic models to determine minimal cut sets. Once the dominant cut sets are determined, the group of cut sets must be quantified to determine the overall probability of frequency. After the cut sets are generated, they are used to obtain the importance measures for each basic event in the cut sets. The resultant cut sets are also used to propagate (using either Monte Carlo or Latin Hypercube sampling) the basic event’s epistemic uncertainty.