ABSTRACT
Probabilistic risk assessment (PRA) was developed to facilitate the quantification of risks associated with complex engineered systems. It is particularly appropriate for analyzing the frequencies of extremely rare events, such as core melts in nuclear reactors, for which little if any accident data will be available . To the extent possible, PRA models are hierarchical in nature. This provides a way of structuring the vast quantities of information that go into a risk analysis. In particular, two reliability analysis techniques are commonly used for quantifying the likelihood of an accident: fault trees and event trees. Many PRAs use event trees to model major plant systems and fault trees to quantify the failure probabilities of the various systems.
A variety of different types of data are needed to support PRA quantification. This includes data on initiating event frequencies (i.e., the frequency of departures from normal operation), component failure rates, common cause failure rates (i.e., the frequency with which two or more components fail during a short period of time for the same reason), component maintenance frequencies and durations, component fragilities (i.e., component failure probabilities as a function of exogenous stresses such as earthquakes, fires, floods, or high temperatures), and human error rates. For most of these data needs, several different types of information may be available, including not only component-specific information (e.g., the number of observed failures of each component), but also expert opinion and data on the failure frequencies of similar components at other plants. Bayesian data analysis is often used to combine generic and component-specific information.
68Once a plant-specific risk model has been developed and quantified, the model can then be used for risk management purposes, and in fact, the use of PRA has resulted in a number of examples of successful risk management involving relatively inexpensive but highly effective risk reduction options. In recent years, such risk management applications have increasingly been undertaken based on risk analyses performed by plant staff rather than consultants, reflecting the successful diffusion of risk analysis technology into the mainstream of engineering applications.
