Proceedings of the Eighth International Conference on Probabilistic Safety Assessment & Management (PSAM)
27 Multidimensional Confusability Matrices Enhance Systematic Analysis of Unsafe Actions and Human Failure Events Considered in PSAs of Nuclear Power Plants (PSAM-0334)
Download citation file:
- Ris (Zotero)
- Reference Manager
In conducting a probabilistic safety assessment of a nuclear power plant, it is important to identify unsafe actions (UAs) and human failure events (HFEs) that can lead to or exacerbate conditions during a range of incidents initiated by internal or external events. Identification and analysis of UAs and HFEs during a human reliability analysis can be a daunting process that often depends completely on subject matter experts attempting to divine a list of plant conditions and performance shaping factors (PSFs) that may influence incident outcomes. Key to this process of including the most important UAs and resulting HFEs is to speculate upon deviations of specific circumstances from a base case definition of a scenario that may present confusion regarding system diagnosis and appropriate actions (i.e., due to procedures, training, informal rules, etc.). Intuiting the location and impact of such system weaknesses is challenging and careful organization of analyst's approach to this process is critical for defending any argument for completeness of the analysis.
Two dimensional distinguishability-confusability matrices were introduced as a tool to test symbol distinguishability for information displays. This paper expands on the tool by presenting multidimensional confusability matrices as a very helpful, pragmatic tool for organizing the process of combining expert judgment regarding system weaknesses, human performance and highly targeted experimentation in a manner that strengthens the quantitative justification for why particular UAs and HFEs were incorporated into a PSA. Furthermore, the particular approach presented here helps to strengthen the justification for specific likelihood determinations (i.e., human error probabilities) that end up being inserted into a probabilistic risk assessment (PRA) or other numerical description of system safety.
This paper first introduces the multidimensional confusability matrix (MCM) approach and then applies it to a set of hypothetical loss of coolant accidents (LOCAs) for which a detailed human reliability analysis is desired. The basic structure of the MCM approach involves showing how actual plant states can be mapped to information available to the operators, and then mapping the information available to operator diagnoses and responses. Finally, there is a mapping of actual plant states to operator performance-each mapping is shown to vary along temporally grounded levels of dominant PSFs (e.g., stress, time available, procedures, training, etc.). MCM facilitates comprehensive analysis of the critical signals/information guiding operator diagnoses and actions. Particular manipulations of plant states, available information and PSFs and resulting operator performance may be experimentally gathered using targeted simulator studies, table top exercises with operators, or thought experiments among analysts. It is suggested that targeted simulator studies will provide the best quantitative mappings across the surfaces generated using the MCMs and the best aid to uncovering unanticipated pieces of ‘critical’ information used by operators. Details of quantifying overall operator performance using the MCM technique are provided. It is important to note that the MCM tool should be considered neutral regarding the issue of so-called ‘reductionist’ HRA methods (e.g., THERP-type) versus ‘holistic’ HRA methods (e.g., ATHEANA, MERMOS). If the analyst's support ‘reductionist’ approaches, then the MCM will represent more of a traditional interval-type, quantitative response surface in their analysis (i.e., more quantitative resolution and generalizability). If the analysis team places more emphasis on ‘holistic’ approaches, then the MCM will represent more of a nominal cataloging or ordinal ranking of factors influencing their specific analysis. In both types of analyses, the MCM tool helps in organizing, documenting and facilitating quantification of expert judgments and, when resources allow, targeted experimental data to support human reliability analyses.