Application of Bayesian methods and networks to ignition hazard event prediction in nuclear waste decommissioning operations

The major purpose of the study is to examine how Bayesian networks can be used to represent and understand potential ignition scenarios in nuclear waste decommissioning. This is illustrated using a network to represent a situation with stacked storage boxes containing pyrophoric material removed from waste storage silos. Corrosion of this material during storage produces hydrogen which is released through a filter medium into the gap between the boxes. The probabilistic relationships used to indicate dependence between network nodes are expressed by conditional probability tables or C++ coded equations that relate to UK nuclear industry corrosion and storage data. The study focuses on optimal prediction of the likelihood of a flammable hydrogen atmosphere arising in the gap between stacked boxes and the conditions necessary to exceed the lower flammable limit. It is concluded that the approach offers a useful means of easily determining the manner in which varying the controlling parameters affects the possibility of an ignition event. The effect of data variation can be examined at first hand using the supplementary Bayesian Network that accompanies the article.