Best estimate computer codes are increasingly used in nuclear industry for the accident management procedures and have been planned to be used for the licensing procedures. Contrary to conservative codes which are supposed to give penalizing results, best estimate codes attempt to calculate accidental transients in a realistic way. It becomes therefore of prime importance, in particular for technical organization as IRSN in charge of safety assessment, to know the uncertainty on the results of such codes.
Thus, CSNI has sponsored few years ago (published in 1998) the Uncertainty Methods Study (UMS) program on uncertainty methodologies used for a SBLOCA transient (LSTF-CL-18) and is now supporting the BEMUSE program for a LBLOCA transient (LOFT-L2-5). The large majority of BEMUSE participants (9 out of 10) use uncertainty methodologies based on a probabilistic modelling and all of them use Monte-Carlo simulations to propagate the uncertainties through their computer codes. Also, all of ‘probabilistic participants’ intend to use order statistics to determine the sampling size of the Monte-Carlo simulation and to derive the uncertainty ranges associated to their computer calculations.
The first aim of this paper is to remind the advantages and also the assumptions of the probabilistic modelling and more specifically of order statistics (as Wilks’ formula) in uncertainty methodologies. Indeed Monte-Carlo methods provide flexible and extremely powerful techniques for solving many of the uncertainty propagation problems encountered in nuclear safety analysis. However it is important to keep in mind that probabilistic methods are data-intensive. That means, probabilistic methods cannot produce robust results unless a considerable body of information has been collected. A main interest of the use of order statistics results is to allow to take into account an unlimited number of uncertain parameters and, from a restricted number of code calculations to provide statistical tolerance limits for any code results. A proof and an extension of this statistical theorem will be given. From this proof, it will appear easily why the use of order statistics results requires the Simple Random Sampling method (SRS).
The second aim of this paper is to illustrate the benefit of these techniques from the application of the IRSN uncertainty methodology on the transient LOFT-L2-5. To achieve this aim, we will use the results obtained in the frame of our participation to the BEMUSE program to clarify how to perform and analyse a Monte-Carlo simulation. In particular, it will be shown how order statistics provide valuable results for estimating percentiles of relevant safety quantities.
Finally, from our experience gained during BEMUSE project, we will conclude on the applicability of Monte-Carlo simulation to derive uncertainty ranges for safety purposes.