Development of a surrogate model for quenching estimation of ex-vessel debris beds and its coupling with MELCOR

In the severe accident management (SAM) strategy for Nordic boiling water reactors (BWRs), a flooded reactor cavity is conceived to receive corium in case of vessel failure, with the hope that the discharged corium will fragment and form a coolable particulate debris bed in the deep water pool. The so-formed debris bed on the cavity basement is supposed to be very hot at the beginning and therefore its quenching is a prerequisite for long-term coolability. In previous study the coupled MELCOR/COCOMO simulation was employed to simulate quench process of ex-vessel debris beds in severe accident scenarios. Although it successfully extended the MELCOR capability, the calculation was dramatically slowed down by explosive computational cost of COCOMO. To overcome the limitation, the present study is to develop a surrogate model (SM) which can replace the me-chanical code COCOMO and realize quick estimations of the quench process of ex-vessel debris beds. It was then coupled with MELCOR code for integral severe accident analyses of a Nordic BWR with cooling of ex-vessel debris beds. The SM was developed based on a database generated from COCOMO calculations of various one-dimension (1D) debris beds quenched in the reactor cavity, using artificial neural networks (ANNs). Finally, the coupled MELCOR/SM simulation was applied to safety analyses of postulated severe accident scenarios due to station blackout (SBO) in the BWR, where MELCOR performs integral analysis of accident progression while SM predicts the consequences (e.g. energy transfer) of debris bed quench. The simulation results show that the coupled MELCOR/SM simulation can predict the trends of containment pressure and pool temperature similar to those of the coupled MELCOR/COCOMO simulation. Compared with MELCOR standalone simulation, the coupled MELCOR/SM simulation predicted earlier pool saturation and containment venting.

Automated summary

In Nordic BWRs, a flooded reactor cavity is used to receive corium in the event of vessel failure, with the aim of forming a coolable debris bed. Previous simulation studies used the MELCOR/COCOMO model for quenching ex-vessel debris beds, but were hindered by the slow computational cost of COCOMO. This study developed a surrogate model (SM) based on artificial neural networks (ANNs) to replace COCOMO, allowing for quick estimations of the quenching process. The coupled MELCOR/SM simulation was then used for integral severe accident analyses, showing similar trends to the MELCOR/COCOMO simulation in predicting containment pressure and pool temperature.