In Nordic design of boiling water reactors (BWRs) a deep water pool under the reactor vessel is employed as a severe accident management strategy for the core melt fragmentation and the long term cooling of corium debris. The height and shape of the debris bed are among the most important factors that determine if decay heat can be removed from the porous debris bed by natural circulation of water. The debris bed geometry is formed as a result of melt release, fragmentation, sedimentation and settlement on the containment basemat. After settlement, the shape can change with time due to movement of particles promoted by the coolant flow (debris bed self-leveling process). Both aleatory (accident scenario, stochastic) and epistemic (modeling, lack of knowledge) uncertainties are important for assessing the risks.
The present work describes a preliminary risk analysis of debris bed coolability for Nordic BWRs under severe accident conditions. It was assumed that once debris remelting starts containment failure becomes imminent. Such assumption allows to estimate the containment failure probability by calculating the probability that the time necessary for the spreading debris bed to achieve a coolable configuration will be shorter than the onset time of debris bed re-melting. An artificial neural network was employed as a surrogate model (SM) for the mechanistic full model (FM) of the debris spreading in order to achieve computationally efficient propagation of uncertainties. The effect of uncertainty in the ranges and probability density functions (PDFs) of the input parameters was addressed. Parameters defining shapes of the PDFs were varied for three different distribution families (beta, truncated normal and triangular). The results of the risk analysis were reported as complementary cumulative distribution functions (CCDFs) of the conditional containment failure probability (CCFP). It is demonstrated that CCFP can vary in wide ranges depending on the randomly selected combinations of the PDFs of the input parameters. Given the selected ranges of the input parameters, sensitivity analyses identified: the effective particle diameter and the debris bed porosity as the largest contributors to the CCFP uncertainty. It was shown that the self-leveling phenomenon reduces sensitivity of debris coolability to the initial shape of the bed. However, the initial shape remains an important uncertainty factor for the most likely values of the particle size and porosity. Importance of the initial shape increases when the effectiveness of the self-leveling is small (e.g. in case of high initial temperature or heat up rate of the debris). Findings of this work in combination with consideration of the necessary efforts can be used for prioritization of the future research on obtaining new information on the uncertain parameters.