Learning Curves in Control Rooms: Skill? Rule? Knowledge?

Abstract

Reactor safety and risk are dominated by the potential and major contribution for human error in the design, operation, control, management, regulation and maintenance of the plant, and hence to all accidents. We need to determine the outcome (error) probability, or the chance of failure. Time and again we are faced with the same situation and question: how to predict the risk or chance of a mistake by an operating crew or team. Conventional reliability engineering is associated with the failure rate of components, or systems, or mechanisms, not of human beings in and interacting with a technological system. The probability of failure requires a prior knowledge of the total number of outcomes, which for any predictive purposes we do not know or have. Analysis of failure rates due to human error based on the Learning Hypothesis allows a new determination of the dynamic human error rate in technological systems, consistent with and derived from the available world data. The basis for the analysis is that humans learn from experience, both prior to and during a transient or event, and consequently the accumulated experience defines the failure rate. Our “best” equation for the probability of human error, outcome or failure rate, which has been validated against the full spectrum of the world’s outcome data, allows for calculation and prediction of the probability of human error. In nuclear probabilistic risk assessment, the modeling of nuclear plant operator actions and transient control behavior is extremely important, and is a requirement according to industry standards. The human error probability (HEP) often classified according to Skill, Rule and Knowledge based behavior. We examine the data and results observed in transients in both plants and simulators, available from France and the USA. We demonstrate that the human error probability (HEP) is dynamic, and that it may be predicted using the Learning Hypothesis and the minimum failure rate, and can be utilized for probabilistic risk analysis purposes.