Yes - No - Maybe: How probabilistic methods can be used to study the influence of human actions on the safety of NPPs


© ra2studio

When assessing the safety of nuclear installations, not only the probability of failure of technical components but also the reliability of human actions plays a role. In contrast to physical processes, which can be described by regularities and mathematical equations, human actions elude such a description due to their complexity and dynamics. They are subject to various influencing factors and can therefore lead to a multitude of possible results.

Can the reliability of human actions at all be included as a factor in a safety assessment? And if so, in what form? GRS researchers have looked into these questions in more detail in a study and have developed a new model for the reliability analysis of human actions.

Evaluation of human actions: What aspects play a role?

When investigating the reliability of human actions, aspects such as personnel qualification, stress level, ergonomics of the man-machine interface and the design quality of equipment are considered as well as more "abstract" parameters. These include, for example, the time required to perform an action. But the interaction between several persons who take over certain parts of an action is also considered here. The resulting action sequences are correspondingly complex and are dynamic due to their possible interactions and temporal dependencies.

This means that they can change over the course of the action. As part of conventional methods of reliability analysis, human actions can therefore only be considered very roughly and incompletely, if at all.

Three Mile Island (left) and Fukushima Daiichi (right): reactor accidents whose course was influenced by human error (Source: not copyrighted, Digital Globa CC BY-SA 3.0

“Crew-Modul“: GRS method for the reliability analysis of human actions

GRS experts also deal repeatedly with the determination of the reliability of human actions as part of their expert opinions and research activities in connection with safety assessments. In order to be able to model and simulate human actions as dynamic processes as accurately as possible in the future, they have developed the so-called Crew Module. It is used in conjunction with the MCDET (Monte Carlo Dynamic Event Tree) method developed by GRS for probabilistic dynamics analysis.

The main difference compared with previous models is that the Crew Module can also take into account dependencies on random events, system states, and temporal interactions. The basic concept of the methodology is to divide a complex action sequence into several small, less complex individual actions, which are then evaluated with regard to potential error possibilities. Such an individual action may e.g. be the operation of a switch, the reading of a display, or the covering of a distance from A to B. The finer the breakdown into individual actions, the more detailed the action sequence of the measure to be evaluated and the dependencies and interactions taking place in it can be simulated and analysed in the Crew Module.

Example: fire-fighting

In order to test the functionality of the Crew Module, a fire fighting measure was modelled. The following persons are involved in the modelling: shift supervisors (SL), fire guard (BL), fire brigade operations manager (ELFW) and fire brigade or fire-fighting group personnel (LG). The individual actions are summarised in so-called action lists (HL), which can result in different action sequences.

The model calculations using the example "fire" were evaluated with regard to various problems. The main interest was not on determining the probability of extinction, but on within what timeframe and by what means the fire is extinguished. Time played an important role here in that the damage to safety-relevant components depends essentially on the duration of the fire.

The example of the fire-fighting measure shows in which form the stress development can be modelled with the Crew Module in a course of action as a dynamic variable as a function of process conditions and random influences. The example also illustrates how different decision-making processes occur as a function of stress development and which (alternative) courses of action result from the decisions. In addition, it was assumed in the example that high stress will not necessarily affect all persons involved but may only affect individual persons. It was assumed e.g. that if the shift supervisor acts erroneously, only the shift supervisor himself and in the fire guard will experience high stress, while the firefighters will only assume a normal to slightly increased stress level in this situation due to their special training.

Results of the GRS investigations

With the methodology used in the Crew Module, not only the influence of stress on so-called omission errors can be modelled, i.e. errors that result from a lack of action, but also its influence on wrong decisions and on errors in execution. It could be shown that both the influencing variables "stress development in general" and "decisions made under high stress" can be quantified probabilistically with the crew methodology and can thus be used for the reliability assessment of a human action.

On the basis of previous work, it could be shown that the method is generally suitable for the analysis of the dynamics of human actions. The use of the Crew Module is also conceivable in other systems outside nuclear technology where the reliability of human actions plays a role, such as in large industrial plants or in aviation.

Find out more

GRS Report 516: Human reliability assessment in PSA for common-cause initiators (in German)