On the evaluation of human error probabilities for post-initiating events

Abstract

Quantification of human error probabilities (HEPs) for the purpose of human reliability assessment (HRA) is very complex. Because of this complexity, the state of the art includes a variety of HRA models, each with its own objectives, scope and quantification method. In addition to varying methods of quantification, each model is replete with its own terminology and categorizations, therefore making comparison across models exceedingly difficult. This paper demonstrates the capabilities and limitations of two prominent HRA models: the Electric Power Research Institute (EPRI) HRA Calculator (using the HRC/ORE and Cause Based Decision Tree methods), used widely in industry, and A Technique for Human Error Analysis (ATHEANA), developed by the US Nuclear Regulatory Commission. This demonstration includes a brief description of the two models, a comparison of what they incorporate in HEP quantification, a "translation" of terminologies, and examples of their capabilities via the Halden Task Complexity experiments. Possible ways to incorporate learning from simulator experiments, such as those at Halden, to improve the quantification methods are also addressed. The primary difference between ATHEANA and the EPRI HRA Calculator is in their objectives. EPRI's objective is to provide a method that is not overly resource intensive and can be used by a PRA analyst without significant HRA experience. Consequently, EPRI quantifies HEPs using time reliability curves (TRCs) and cause based decision trees (CBDT). ATHEANA attempts to find contexts where operators are likely to fail without recovery and quantify the associated HEP. This includes finding how operators can further degrade the plant condition while still believing their actions are correct. ATHEANA quantifies HEPs through an expert judgment elicitation process.

(cont.) ATHEANA and the EPRI Calculator are very similar in the contexts they consider in HEP calculation: both factor in the accident sequence context, performance shaping factors (PSFs), and cognitive factors into HEP calculation. However, stemming from the difference in objectives, there is a difference in how deeply into a human action each model probes. ATHEANA employs a HRA team (including a HRA expert, operations personnel and a thermo-hydraulics expert) to examine a broad set of PSFs and contexts. It also expands the accident sequences to include the consequences of a misdiagnosis beyond simple failures in implementing the procedures (what will the operator likely do next given a specific misdiagnosis?) To limit the resource burden, the EPRI Calculator is prescriptive and limits the PSFs and cognitive factors for consideration thus enhancing consistency among analysts and reducing needed resources. However, CBDT and ATHEANA have the same approach to evaluating the cognitive context. The Halden Task Complexity experiments looked at different factors that would increase the probability of human failures such as the effects of time pressure/information load and masked events. EPRI and ATHEANA could use the design of the Halden experiments as a model for future simulations because they produced results that showed important differences in crew performance under certain conditions. Both models can also use the Halden experiments and results to sensitize the experts and analysts to the real effects of an error forcing context.