VIRTHUALIS Solutions:

Current approach and specific features of A.I “lead analysis”

The analysis of the Accident is a task carried out in team and definitely not an individual activity. The composition as well as the leader of the team can change according to the criticality of the event/accident and the part of the installation concerned. The higher the criticality of an event is, the higher the position of the investigation leader and of the members in the hierarchy. The team is usually made up of people coming from 3 different departments:

The department concerned with the accident;
the safety department;
3rd department implied indirectly in the accident (e.g. the department maintenance when the accident takes place after a maintenance operation, etc).

Furthermore, Accident and Incident investigation is not a full time activity. These are two very important aspects for the tools to be developed since they mean that (1) expertise related to the accident under investigation is distributed among the team’s members, thus the sharing of the same representation can not be assumed definitely and (2) the hopefully low frequency of participation in AI tasks provide no actual opportunity to gain expertise in it. Fundamentally, the activity of the team of investigators during the course of the analysis is to “try and build a story or mental model that explains the accident” (Hollnagel, 2005, p 55). It is a dynamic process that requires modification to the representation of accident as additional evidence becomes available. The overall objective is to explore any possible factors that have contributed to the occurrence of unwanted events (accidents and incidents), in order to promote prevention actions (reactive approach). It is thus desirable to represent not only the events/actions that were directly involved in the accident, but also other Performance Shaping Factors that affected the workload and perception of operators. Apart from the preparation and reporting phases, the investigation activity includes two main phases that themselves involve several detailed activities (see also AI part in D 5.2 & D 5.3):

the reconstruction of the facts to establish a timeline for the accident; this is progressively done on the basis of the data obtained from interviewing witnesses, available documentation (procedures, checklists, working documents etc.), information on past performance of the relevant part of the system.
the search of the causes. This can be done with the support of a variety of methods such root-cause analysis, Systematic Causation Analysis Technique (SCAT), Integrated Systemic Approach for Accident Causation (ISAAC) (Cacciabue, 2004), Cognitive Reliability and Error Analysis Method (CREAM) (Hollnagel, 1998) [VIRTHUALIS has adopted the CREAM method].

Applications

The activities to be supported by VR – together with and in complement to already existing tools are presented in the following:

Application 1: supporting the reconstruction of the facts and the research of cause by supporting a stronger HF focus

For the reconstruction of the facts, end-users need to collect, delimit and organize different kind of data, from various actors and of different nature. The facts’ reconstruction phase is a crucial one that can be seen as different preparation phases before leading the research of the causes. At this stage, it has been proposed to implement more HF data to support a systemic approach in order to improve the quality of the reconstruction and then the analysis. These data must concern technical aspects, operator’s activity aspects and contextual/organisational aspects concerned by the accident. Moreover, this work needs an agreement among the members of the team with the necessity that all understand and share a same representation of theses different kind of information. For the search of the causes, the chosen method is primarily CREAM (Hollnagel, 1998) as it covers the triad Man-Technology-Organization with clear differences between effects (phenotypes) and causes (genotypes), and it is bidirectional (can be used both for prospective and retrospective analysis). Furthermore, it contains a clear stop rule (see D 5.3.). A computer- based CREAM hypertext can be already used to guide the analysis of the possible HF-related causes. Thus the idea is not to re-implement this already existing tool, but rather to provide a way to integrate within the Accident Model conclusions and identified causal paths that resulted from the use of CREAM. However, it should be made clear for any other HF method.

Application 2: VR as a support to the analysis dissemination

One of the main objectives of accident investigation is to provide information that would help the organisation to learn from the failures and hence improve the safety of the system. Therefore one of the main assumptions of the accident investigation team is that the events that occurred in the past are something that potentially could repeat in the future. One of the targets of the analysis is therefore to communicate the results of the investigation to the involved actors to make them aware of the event. However, one of the main limitations of the current dissemination is that it is done mostly through a report or through an abstract of the report where the results of the analysis and the main consequences and causes of the accident are known at the beginning of the report. This format results in an increase of the hindsight bias which in turn increase the invulnerability bias, i.e. the reader has the feeling that what happen in this event would never occur to him/her. When other formats of reports are used such as a narrative one where the reader is virtually projected at the place of the involved actor, the invulnerability bias tends to decrease, hence increasing the efficiency of the report.

How VR could improve the accident investigation

The main axes of using VR with a HF focus have been developed up to now on the basis of the needs iteratively built from WP1 to WP5. These ideas are partially deeply detailed and decomposed in the functional specifications documents (WP6)4. The idea (covering a large amount of the required functions of AI in terms of VR-related ones) is “using VR as a visualization tool to display information that will facilitate the elaboration of a shared and objective model of the situation among all the participants within a team, at each step of the analysis of the events/accident” to: a. Reconstruct the facts related to the accident (by the way, this requirement is almost common to AI activity whatever the HF method used, i.e. CREAM or ISAAC); b. Explore and check as many as possible of the probable causes that can be hypothesized, in a systematic and documented fashion, especially those related to human and organizational factors; b. Exploit the results of the analysis to disseminate efficiently the causal patterns and the proposed prevention means among sites; The VR innovative and functional added-value to foster efficiency within the Accident Team has been clearly identified regarding the parts of the analysis with a focus on Human-Factors (probable) causes, like e.g. providing a (mostly) visual interactive support to:

Have a common understanding/review of the process, environment and multiple operators’ task elements involved in the situation
Help investigators in checking and maintaining consistency between process dynamics, spatial organization, physical constraints and human actions timeline;
Help in identifying possible deviations at operation and organizational levels in relation to the process and spatial model of the work environment.