Publications Details

Coherent sampling of multiple branch event tree questions

Payne Jr., A.C.; Wyss, G.D.

In the detailed phenomenological event trees used in recent Level III PRA analyses questions arise about the possible outcomes of events for which the underlying physics is not well understood and where the initial and boundary conditions are uncertain. Examples of the types of events being analyzed are: What is the containment failure mode?, Is them a large in-vessel steam explosion?, How much H{sub 2}, CO, and CO{sub 2} are produced during core-concrete interactions? The outcomes of each of these questions must be defined based on an understanding of the basic physics of the phenomena and the level of detail of the probabilistic analysis. Many of these phenomena have never occurred since severe reactor accidents are extremely rare events. The only information we have about these phenomena comes from four basic sources: general theoretical knowledge, limited experimental results a few actual events, and various models of the phenomena. All of these phenomena have significant uncertainty arising from three basic sources: level of detail, initial and boundary conditions, and lack of knowledge. Since it is not possible to conduct enough full scale tests to generate a set of ``objective`` relative frequencies, the probabilities, therefore, will have to be ``subjective`` and generated based on expert knowledge. In assessing the conditional probabilities of the various possible outcomes of an event during an accident, the expert must amalgamate his knowledge with the level of detail being used in the PRA analysis to generate a set of probabilities for the defined set of outcomes. It is often convenient for an expert to formulate his opinion in terms of expecting to see n{sub i} occurrences of outcome E{sub i} in N occurrences of event E. The order of the outcomes is typically not important because the individual trials are viewed as being independent of one another.