Using a FRAM model for retrospective analysis (event analysis)
The common purpose of an accident investigation / event analysis is to find the causes for the observed outcome(s) (the injury). (In some methods, the purpose is even to find the root causes.) An accident investigation therefore typically starts from the observed (adverse) outcome, and traces the developments backwards in order to find out what went wrong or malfunctioned, i.e., the conditions (events) that were necessary and sufficient for the outcome to happen. The reasoning is usually contra-factual, such as: 'If X had not happened, then this (the accident) would not have happened. Therefore X is the cause of the accident.'
In consonance with the principles of Safety-II (q.v.), the purpose of a FRAM analysis is to identify how the system should have functioned for everything to succeed (i.e., “everyday” performance), and to understand the variability of functions which alone or in combination prevented that from happening. This is typically the variability that existed in the situation being analysed, but it could also be the variability that possibly might exist under other conditions.
A FRAM model describes a system’s functions and the potential couplings among functions. The model does not describe or depict an actual sequence of events, i.e., an accident scenario. The basic steps in building a FRAM model of the activity (or performance) that is to be analysed have been described here. The development of the model ended by describing the potential variability.
An accident scenario corresponds to an instantiation of the model. A FRAM model includes a description of the potetial variability of the functions. The instantiation is a “map” of how the potential variability became real (or actual) variability in a given situation. An instantiation shows how functions were coupled under given – favourable or unfavourable – conditions. The analysis of the instantiation of the FRAM model comprises the following steps:
Characterise the variability in the actual event of the set of 'foreground' functions and of the set of 'background' functions (context). Consider whether the actual variability was what one could expect ('normal') or whether it was unusual - either by being very large or by being very small ('abnormal'). (In both cases the situation would not be as usually expected, hence potentially affecting the variability of downstream functions.) The variability is expressed by characterising the variability of the Output of each function, either using the simple or the elaborate description.
Identify the dynamic couplings (functional resonance) that played a role during the event. This is the actual instantiation of the model, which can be used to understand why the event developed the way it did. In relation to the traditional accident analysis, this instantiation provides an explanation of what happened, although it does not necessarily identify unique or specific causes. The explanation may, for instance, be found in the couplings of the variability of everyday performance.
Propose ways to monitor and dampen performance variability (indicators, barriers, design / modification, etc.) In the case of unexpected positive outcome, the search is of course for ways to amplify, in a controlled manner, the variability rather than for ways to dampen it.
© Copyright Erik Hollnagel 2016. All Rights Reserved.