Risk in technical and scientific studies: general introduction to uncertainty management and the concept of risk

George Apostolakis (MIT) presented an introduction to the concept of risk and uncertainty management and their use in technical and scientific studies. He noted that Quantitative Risk Assessment (QRA) provides support to the overall treatment of a system as an integrated socio-technical system. Specifically, QRA aims to answer the questions: - What can go wrong (e.g., accident sequences or scenarios)? - How likely are these sequences or scenarios? - What are the consequences of these sequences or scenarios? The Quantitative Risk Assessment deals with two major types of uncertainty. An assessment requires a 'model of the world', and this preferably would be a deterministic model based on underlying processes. In practice, there are uncertainties in this model of the world relating to variability or randomness that cannot be accounted for directly in a deterministic model and that may require a probabilistic or aleatory model. Both deterministic and aleatory models of the world have assumptions and parameters, and there are 'state-of-knowledge' or epistemic uncertainties associated with these. Sensitivity studies or eliciting expert opinion can be used to address the uncertainties in assumptions, and the level of confidence in parameter values can be characterised using probability distributions (pdfs). Overall, the distinction between aleatory and epistemic uncertainties is not always clear, and both can be treated mathematically in the same way. Lessons on safety assessments that can be learnt from experience at nuclear power plants are that beliefs about what is important can be wrong if a risk assessment is not performed. Also, precautionary approaches are not always conservative if failure modes are not identified. Nevertheless, it is important to recognize that uncertainties will remain despite a quantitative risk assessment: e.g., is the scenario list complete, are the models accepted as reasonable, and are parameter probability distributions representative of the current state of knowledge? As a consequence, decisions should be risk-informed, not risk-based, and traditional approaches still have a role to play. As an example of a risk-informed approach, guidelines from the USNRC for the acceptance of proposed changes at NPPs were presented. These acknowledge that the level of technical review and management attention given to proposals will vary according to both the estimated core damage frequency (CDF) and the change in frequency resulting from the proposed change (ΔCDF). Different criteria apply to different regions on a plot of CDF vs. ΔCDF, but the numerical values defining the regions are indicative and the boundaries between them are not interpreted definitively. Finally, two interpretations of the defence-in-depth (DiD) approach were presented. (author)