Uncertainty and sensitivity analysis of a fire-induced accident scenario involving binary variables and mechanistic codes

Abstract

In response to the transition by the United States Nuclear Regulatory Commission (NRC) to a risk-informed, performance-based fire protection rulemaking standard, Fire Probabilistic Risk Assessment (PRA) methods have been improved, particularly in the areas of advanced fire modeling and computational methods. As the methods for the quantification of fire risk are improved, the methods for the quantification of the uncertainties must also be improved. In order to gain a more meaningful insight into the methods currently in practice, it was decided that a scenario incorporating the various elements of uncertainty specific to a fire PRA would be analyzed. The NRC has validated and verified five fire models to simulate the effects of fire growth and propagation in nuclear power plants. Although these models cover a wide range of sophistication, epistemic uncertainties resulting from the assumptions and approximations used within the model are always present. The uncertainty of a model prediction is not only dependent on the uncertainties of the model itself, but also on how the uncertainties in input parameters are propagated throughout the model. Inputs to deterministic fire models are often not precise values, but instead follow statistical distributions. The fundamental motivation for assessing model and parameter uncertainties is to combine the results in an effort to calculate a cumulative probability of exceeding a given threshold. This threshold can be for equipment damage, time to alarm, habitability of spaces, etc. Fire growth and propagation is not the only source of uncertainty present in a fire-induced accident scenario. Statistical models are necessary to develop estimates of fire ignition frequency and the probability that a fire will be suppressed. Human Reliability Analysis (HRA) is performed to determine the probability that operators will correctly perform manual actions even with the additional complications of a fire present. Fire induced Main Control Room (MCR) abandonment scenarios are a significant contributor to the total Core Damage Frequency (CDF) estimate of many operating nuclear power plants. Many of the resources spent on fire PRA are devoted to quantification of the probability that a fire will force operators to abandon the MCR and take actions from a remote location. However, many current PRA practitioners feel that effect of MCR fires have been overstated. This report details the simultaneous application of state-of-the-art model and parameter uncertainty techniques to develop a defensible distribution of the probability of a forced MCR abandonment caused by a fire within a MCR benchboard. These results are combined with the other elements of uncertainty present in a fire-induced MCR abandonment scenario to develop a CDF distribution that takes into account the interdependencies between the factors. In addition, the input factors having the strongest influence on the final results are identified so that operators, regulators, and researchers can focus their efforts to mitigate the effects of this class of fire-induced accident scenario.