Large industrial utilities with a limited amount of critical failures, such as core damage or large early release in nuclear power plants, give rise to a vast number of different scenarios leading from an accident initiating event to an undesired consequence. The industrial practice, standardized as Probabilistic Risk/Safety Assessment (PRA or PSA), uses event trees and fault trees for accident scenario modeling. Great scalability of the analysis of event tree/fault tree models comes for the price of restricted modeling power. Relatively simple and natural concepts for describing an accident progression, such as repairs of components or cold spare safety systems being started when the primary alternative fails (triggering), can be expressed only with limitations and for an often prohibitive modeling cost.
On the other hand, full analysis of models with repairs and spares such Boolean-Driven Markov Processes or as Dynamic Fault Trees becomes quickly computationally intractable even for medium sized system descriptions. We propose a novel method for a scalable analysis based on the following approximation: the number of repairs taken into account in the analysis is bounded by a number determined by the modeler and cold spares are not switched off even when the primary system is repaired. We show how this method handles failures on-demand and failures in-operation, repairs and cold spares.