Using a detailed Probabilistic Risk/Safety Assessment (PRA/PSA) model to estimate the risk level of a nuclear facility is a well established practice, though the scope and granularity of such models continue to evolve. Over the past several decades, model based methods for development of fault tree models have become available and increasingly adopted. Modern tools can now fully automate the generation of fault trees—directly from systems engineering artifacts such as P&IDs or single line diagrams—if there is a component library for the relevant domain and system type defined. This not only improves efficiency but also enables systems engineering and safety engineering to work from a unified conceptual and visual representation of the system.
The combinatorial description of failure scenarios, which has proven highly effective in PRA/PSA, has also been applied to availability analyses for nuclear power plants, including trip monitors and generation risk assessment (GRA). A trip monitor uses fault/event tree models to estimate the frequency of a plant trip or derate based on the current operational state. GRA aims to forecast generation losses by estimating how often and how long trips or derates occur as a result of equipment degradation or failure. GRA requires additional data beyond standard fault/event tree analysis, such as repair durations and downtime. Current GRA methods typically rely on ad hoc adaptations of PRA/PSA tools to obtain the necessary results. Also, PRA/PSA analyses focus on point estimates, whereas financial and production related assessments commonly incorporate uncertainty, often through Monte Carlo simulation techniques.
In this paper, we explore how model based approaches can streamline and improve GRA model development and output. First, modeling can be carried out at a more suitable level of abstraction—closer to systems engineering—allowing parts of the design that do not require detailed treatment to be grouped into macro components. Second, dynamic behaviors such as repairs, reconfigurations, and derates can be naturally represented and evaluated through simulation. We also examine how component importance measures can be computed while accounting for uncertainty. Finally, we illustrate these ideas using a small example model for GRA.