Engineering Papers⌕ Search

Engineering topics

Kelly, Dana L

Publications and source records attributed to Kelly, Dana L.

SPAR-H Step-by-Step Guidance

Step-by-step guidance was developed recently at Idaho National Laboratory for the US Nuclear Regulatory Commission on the use of the Standardized Plant Analysis Risk-Human Reliability Analysis (SPAR-H) method for quantifying Human Failure Events (HFEs). This work was done to address SPAR-H user needs, specifically requests for additional guidance on the proper application of various aspects of the methodology. This paper overviews the steps of the SPAR-H analysis process and highlights some of the most important insights gained during the development of the step-by-step directions. This supplemental guidance for analysts is applicable when plant-specific information is available, and goes beyond the general guidance provided in existing SPAR-H documentation. The steps highlighted in this paper are: Step-1, Categorizing the HFE as Diagnosis and/or Action; Step-2, Rate the Performance Shaping Factors; Step-3, Calculate PSF-Modified HEP; Step-4, Accounting for Dependence, and; Step-5, Minimum Value Cutoff.

99 GENERAL AND MISCELLANEOUS↗

Science-Based Simulation Model of Human Performance for Human Reliability Analysis

Human reliability analysis (HRA), a component of an integrated probabilistic risk assessment (PRA), is the means by which the human contribution to risk is assessed, both qualitatively and quantitatively. However, among the literally dozens of HRA methods that have been developed, most cannot fully model and quantify the types of errors that occurred at Three Mile Island. Furthermore, all of the methods lack a solid empirical basis, relying heavily on expert judgment or empirical results derived in non-reactor domains. Finally, all of the methods are essentially static, and are thus unable to capture the dynamics of an accident in progress. The objective of this work is to begin exploring a dynamic simulation approach to HRA, one whose models have a basis in psychological theories of human performance, and whose quantitative estimates have an empirical basis. This paper highlights a plan to formalize collaboration among the Idaho National Laboratory (INL), the University of Maryland, and The Ohio State University (OSU) to continue development of a simulation model initially formulated at the University of Maryland. Initial work will focus on enhancing the underlying human performance models with the most recent psychological research, and on planning follow-on studies to establish an empirical basis for the model, based on simulator experiments to be carried out at the INL and at the OSU.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Minimally Informative Prior Distributions for PSA

A salient feature of Bayesian inference is its ability to incorporate information from a variety of sources into the inference model, via the prior distribution (hereafter simply “the prior”). However, over-reliance on old information can lead to priors that dominate new data. Some analysts seek to avoid this by trying to work with a minimally informative prior distribution. Another reason for choosing a minimally informative prior is to avoid the often-voiced criticism of subjectivity in the choice of prior. Minimally informative priors fall into two broad classes: 1) so-called noninformative priors, which attempt to be completely objective, in that the posterior distribution is determined as completely as possible by the observed data, the most well known example in this class being the Jeffreys prior, and 2) priors that are diffuse over the region where the likelihood function is nonnegligible, but that incorporate some information about the parameters being estimated, such as a mean value. In this paper, we compare four approaches in the second class, with respect to their practical implications for Bayesian inference in Probabilistic Safety Assessment (PSA). The most commonly used such prior, the so-called constrained noninformative prior, is a special case of the maximum entropy prior. This is formulated as a conjugate distribution for the most commonly encountered aleatory models in PSA, and is correspondingly mathematically convenient; however, it has a relatively light tail and this can cause the posterior mean to be overly influenced by the prior in updates with sparse data. A more informative prior that is capable, in principle, of dealing more effectively with sparse data is a mixture of conjugate priors. A particular diffuse nonconjugate prior, the logistic-normal, is shown to behave similarly for some purposes. Finally, we review the so-called robust prior. Rather than relying on the mathematical abstraction of entropy, as does the constrained noninformative prior, the robust prior places a heavy-tailed Cauchy prior on the canonical parameter of the aleatory model.

99 GENERAL AND MISCELLANEOUS↗

Bayesian Inference for Time Trends in Parameter Values: Case Study for the Ageing PSA Network of the European Commission

There is a nearly ubiquitous assumption in PSA that parameter values are at least piecewise-constant in time. As a result, Bayesian inference tends to incorporate many years of plant operation, over which there have been significant changes in plant operational and maintenance practices, plant management, etc. These changes can cause significant changes in parameter values over time; however, failure to perform Bayesian inference in the proper time-dependent framework can mask these changes. Failure to question the assumption of constant parameter values, and failure to perform Bayesian inference in the proper time-dependent framework were noted as important issues in NUREG/CR-6813, performed for the U. S. Nuclear Regulatory Commission’s Advisory Committee on Reactor Safeguards in 2003. That report noted that “industry lacks tools to perform time-trend analysis with Bayesian updating.” This paper describes an application of time-dependent Bayesian inference methods developed for the European Commission Ageing PSA Network. These methods utilize open-source software, implementing Markov chain Monte Carlo sampling. The paper also illustrates the development of a generic prior distribution, which incorporates multiple sources of generic data via weighting factors that address differences in key influences, such as vendor, component boundaries, conditions of the operating environment, etc.

99 GENERAL AND MISCELLANEOUS↗

When the Details Matter – Sensitivities in PRA Calculations That Could Affect Risk-Informed Decision-Making

As the U.S. Nuclear Regulatory Commission (NRC) continues its efforts to increase its use of risk information in decision making, the detailed, quantitative results of probabilistic risk assessment (PRA) calculations are coming under increased scrutiny. Where once analysts and users were not overly concerned with figure of merit variations that were less than an order of magnitude, now factors of two or even less can spark heated debate regarding modeling approaches and assumptions. The philosophical and policy-related aspects of this situation are well-recognized by the PRA community. On the other hand, the technical implications for PRA methods and modeling have not been as widely discussed. This paper illustrates the potential numerical effects of choices as to the details of models and methods for parameter estimation with three examples: 1) the selection of the time period data for parameter estimation, and issues related to component boundary and failure mode definitions; 2) the selection of alternative diffuse prior distributions, including the constrained noninformative prior distribution, in Bayesian parameter estimation; and 3) the impact of uncertainty in calculations for recovery of offsite power.

99 GENERAL AND MISCELLANEOUS↗

A Mid-Layer Model for Human Reliability Analysis: Understanding the Cognitive Causes of Human Failure Events

The Office of Nuclear Regulatory Research (RES) is sponsoring work in response to a Staff Requirements Memorandum (SRM) directing an effort to establish a single human reliability analysis (HRA) method for the agency or guidance for the use of multiple methods. As part of this effort an attempt to develop a comprehensive HRA qualitative approach is being pursued. This paper presents a draft of the method’s middle layer, a part of the qualitative analysis phase that links failure mechanisms to performance shaping factors. Starting with a Crew Response Tree (CRT) that has identified human failure events, analysts identify potential failure mechanisms using the mid-layer model. The mid-layer model presented in this paper traces the identification of the failure mechanisms using the Information-Diagnosis/Decision-Action (IDA) model and cognitive models from the psychological literature. Each failure mechanism is grouped according to a phase of IDA. Under each phase of IDA, the cognitive models help identify the relevant performance shaping factors for the failure mechanism. The use of IDA and cognitive models can be traced through fault trees, which provide a detailed complement to the CRT.

99 GENERAL AND MISCELLANEOUS↗

Impact of Uncertainty on Calculations for Recovery from Loss of Offsite Power

Uncertainty, both aleatory and epistemic, can have a significant impact on estimated probabilities of recovering from loss of offsite power within a specified time window, and such probabilities are an input to risk-informed decisions as to the significance of inspection findings in the U.S. Nuclear Regulatory Commission’s Reactor Oversight Process. In particular, the choice of aleatory model for offsite power recovery time can have a significant impact on the estimated nonrecovery probability, especially if epistemic uncertainty regarding parameters in the aleatory model is accounted for properly. In past and current analyses, such uncertainty has largely been ignored. This paper examines the impact of both aleatory and epistemic uncertainty on the results, using modern open-source Bayesian inference software, which implements Markov chain Monte Carlo sampling. It includes examples of time-dependent convolution calculations to show the impact that uncertainty can have on this increasingly frequent type of calculation, also. The results show that the “point estimate” result, which is an input to risk-informed decisions, can easily be uncertain by a factor of 10 if both aleatory and epistemic uncertainties are considered. The paper also illustrates the use of Bayesian model selection criteria to aid in the choice of aleatory model.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

A mid-layer model for human reliability analysis : understanding the cognitive causes of human failure events.

The Office of Nuclear Regulatory Research (RES) at the US Nuclear Regulatory Commission (USNRC) is sponsoring work in response to a Staff Requirements Memorandum (SRM) directing an effort to establish a single human reliability analysis (HRA) method for the agency or guidance for the use of multiple methods. As part of this effort an attempt to develop a comprehensive HRA qualitative approach is being pursued. This paper presents a draft of the method's middle layer, a part of the qualitative analysis phase that links failure mechanisms to performance shaping factors. Starting with a Crew Response Tree (CRT) that has identified human failure events, analysts identify potential failure mechanisms using the mid-layer model. The mid-layer model presented in this paper traces the identification of the failure mechanisms using the Information-Diagnosis/Decision-Action (IDA) model and cognitive models from the psychological literature. Each failure mechanism is grouped according to a phase of IDA. Under each phase of IDA, the cognitive models help identify the relevant performance shaping factors for the failure mechanism. The use of IDA and cognitive models can be traced through fault trees, which provide a detailed complement to the CRT.

42 ENGINEERING↗

Bayesian Inference in Probabilistic Risk Assessment -- The Current State of the Art

Markov chain Monte Carlo approaches to sampling directly from the joint posterior distribution of aleatory model parameters have led to tremendous advances in Bayesian inference capability in a wide variety of fields, including probabilistic risk analysis. The advent of freely available software coupled with inexpensive computing power has catalyzed this advance. This paper examines where the risk assessment community is with respect to implementing modern computational-based Bayesian approaches to inference. Through a series of examples in different topical areas, it introduces salient concepts and illustrates the practical application of Bayesian inference via Markov chain Monte Carlo sampling to a variety of important problems.

99 GENERAL AND MISCELLANEOUS↗

Common-Cause Failure Analysis in Event Assessment

This paper describes the approach taken by the U. S. Nuclear Regulatory Commission to the treatment of common-cause failure in probabilistic risk assessment of operational events. The approach is based upon the Basic Parameter Model for common-cause failure, and examples are illustrated using the alpha-factor parameterization, the approach adopted by the NRC in their Standardized Plant Analysis Risk (SPAR) models. The cases of a failed component (with and without shared common-cause failure potential) and a component being unavailable due to preventive maintenance or testing are addressed. The treatment of two related failure modes (e.g., failure to start and failure to run) is a new feature of this paper. These methods are being applied by the NRC in assessing the risk significance of operational events for the Significance Determination Process (SDP) and the Accident Sequence Precursor (ASP) program.

99 GENERAL AND MISCELLANEOUS↗

Practical Issues in Component Aging Analysis

This paper examines practical issues in the statistical analysis of component aging data. These issues center on the stochastic process chosen to model component failures. The two stochastic processes examined are repair same as new, leading to a renewal process, and repair same as old, leading to a nonhomogeneous Poisson process. Under the first assumption, times between failures can treated as statistically independent observations from a stationary process. The common distribution of the times between failures is called the renewal distribution. Under the second process, the times between failures will not be independently and identically distributed, and one cannot simply fit a renewal distribution to the cumulative failure times or the times between failures. The paper illustrates how the assumption made regarding the repair process is crucial to the analysis. Besides the choice of stochastic process, other issues that are discussed include qualitative graphical analysis and simple nonparametric hypothesis tests to help judge which process appears more appropriate. Numerical examples are presented to illustrate the issues discussed in the paper.

99 GENERAL AND MISCELLANEOUS↗