Research Papers by CRA in PSAM12

CRA staff and students have written several research articles due to be published in PSAM12. The abstracts for these articles are reproduced below, and full copies can be obtained on request. (See the end of article for email addresses)

Propagating Uncertainty in Phenomenological Analysis into Probabilistic
Safety Analysis
Ashraf El - Shanawany
Abstract: The operation of nuclear power plants is supported by numerous analyses, both
computational and experimental. Probabilistic risk analysis models attempt to quantify the risk of
power plants, and implicitly use the supporting analyses during this process. The way in which these
analyses are used in risk models is usually conservative, but could instead be represented as an
uncertainty distribution. The conservatisms are often hidden, but affect every aspect of risk models;
for example in the definition of success criteria. This paper uses operator reliability as an example to
quantitatively demonstrate how conservative interpretations of supporting analyses can affect risk
model predictions.
The influence of human factors is recognised to be crucially important to risk models for nuclear
power plants. Human error probability quantification is a key aspect in determining the relative risk
importance of human actions in the context of a holistic probabilistic safety analysis model. However,
there are large degrees of uncertainty in numerous aspects of human factors analysis and in the
resulting quantification, many of which can be traced back to supporting transient analyses, such as
thermal hydraulic and neutronic analyses. Risk models have historically used conservative
judgements resulting from these analyses as an input into human reliability assessment. This paper
presents a method for incorporating uncertainty distributions arising from phenomenological analyses
into human reliability quantification. The method is illustrated using uncertainty in the timescale
available to the operator for performing specified actions. This paper shows how to include
uncertainty distributions over the time available to the operator and provides updated quantitative
analysis. An illustrative example of operator initiated long term hold down of reactivity is presented.

----------------------------------------------------------------------------------------------------------------------
Understanding Relative Risk: An Analysis of Uncertainty and Time at Risk
Ashraf El - Shanawany
Abstract: Risk at nuclear facilities in the UK is managed through a combination of the ALARP
principle (As Low As Reasonably Practicable), and numerical targets. The baseline risk of a plant is
calculated through the use of Probabilistic Safety Assessment (PSA) models, which are also used to
estimate the risk in various plant states, including maintenance states. Taking safety equipment out of
service for maintenance yields a temporary increase in risk. Software tools such as RiskWatcher can
be used to monitor the real time level of risk at plant. In combination with software tools to estimate
the instantaneous risk, time at risk arguments are frequently employed to justify safety during plant
modifications or maintenance activities. In this paper we consider the effect of using conservative
estimates for the probability of failure on demand of safety critical components compared to using a
full uncertainty distribution. It is found that conservatism in the base case model translates to a hidden
optimism when used in time at risk arguments. While it is known and accepted that quantified risks are
necessarily approximate, useful insights can be gained through risk modelling by considering relative
risks. Anything that distorts relative risks impacts on the usefulness of the risk modelling. The
important point of the effect discussed here is that it has the potential to distort relative risks. The
mapping between the base case conservatism and the time at risk optimism is characterised, and the
effect is illustrated using simple hypothetical examples. These simple examples show that the shape of
the full uncertainty distributions of model parameters have important and direct consequences for time
at risk arguments, and must be considered in order to avoid distorting the risk profile.

----------------------------------------------------------------------------------------------------------------------
UK Experience of Developing Alpha Factors for Use in Nuclear PRA Models
Garth Rowlands and Ashraf El - Shanawany
Abstract: Modelling Common Cause Failures (CCFs) is an essential part of Probabilistic Risk Assessment (PRA). In the UK, the normal approach for the Advanced Gas-cooled Reactors (AGRs) is to use the beta factor approach with these parameters determined using the Unified Partial Method (UPM). However, there has been recent impetus to consider the feasibility of using a more detailed CCF approach for the AGRs such as the Alpha Factor method. The AGRs share some component types with water cooled reactors. For these it is possible to obtain alpha factors from international databases (such as the US Nuclear Regulatory Commission (NRC) CCF Parameter Estimates and the International Common-Cause Data Exchange (ICDE)). However, AGRs contain many unique components which are not listed in these databases. An additional difficulty is the small AGR fleet size and consequently a potential lack of operating experience. This paper presents the experience to date in deriving alpha factors for AGR components, and presents a Bayesian method which can be used in cases where comparative prior data is sparse. Insights and experiences from the process are discussed.

----------------------------------------------------------------------------------------------------------------------
Can we quantify human reliability in Level 2 PSA?
Lavinia Raganelli, Barry Kirwan
Abstract: In current safety practice in the nuclear power domain, the demand for Level Two PSA by regulatory organizations has become mandatory, and this has received greater priority after the Fukushima-Daiichi accident in Japan in March 2011. However, there are many challenges in the process of performing a Level Two PSA. Most of the challenges are related to uncertainties in the plant state in such accident scenarios. However, even assuming that it is possible to know the exact extent of damage in a selected scenario, a key question remains: “What level of detail is required for describing the human response?” In reality, damage to equipment and the exact plant status are not predictable; therefore Severe Accident Management Guidelines (SAMGs) and Emergency Operating Procedures (EOPs) offer guidelines for operator behaviour rather than specifying the procedural details of actions. In this paper the appropriate level of detail for the analysis of operator action in Level Two PSA models is discussed, as are the difficulties in conducting Human Reliability Assessment (HRA) for vaguely defined actions. It is found that most current HRA approaches for Level 2 PSA rely heavily on expert judgment, but is such expertise valid? This paper explores potential ways forward for HRA in Level 2 PSA.
----------------------------------------------------------------------------------------------------------------------

Ashraf El-Shanawany - ashanawany@c-risk-a.co.uk

Lavinia Raganelli - lraganelli@c-risk-a.co.uk