Definition of a dynamic source term module within RASTEP

Abstract

RASTEP (RApid Source TErm Prediction) is a computerized tool developed by Scandpower AB for use in the fast online diagnosis of accidents at nuclear power plants. The tool is based on a Bayesian Belief Network (BBN) that is used to determine the most likely plant state, which is associated with pre-calculated source terms. In its current design, the source term predictions are not flexible enough. A previous study evaluated different methods for enhancing the source term module of RASTEP. This thesis work follows that approach and explores the integration of a fast running deterministic code within RASTEP in order to make the predictions more realistic. The MARS software, developed by Fauske & Associates, has been chosen as best candidate for this purpose. Literature studies, along with interviews with experts, and conceptual reasoning have been carried out in order to identify the best linking process. Two modes for coupling the BBN with the deterministic code MARS (integrated and iterative) have been proposed and evaluated. In both approaches, the strength of the probabilistic predictions is kept and the capabilities of MARS are proven also in a linked configuration. It is concluded from the study that this method can be used to enhance RASTEP and it is feasible for implementation in the short term.