Agent-based Approach to Retrospective Analysis of Aviation Accidents

Abstract

Agent based modelling is a relatively new paradigm allowing to model the various components of the complex socio-technical air transportation system as well as the interactions between these components. Because situation awareness plays a key role in agent-based modelling, researchers at NLR had developed approaches in modelling Multi-Agent Situation Awareness (MASA) and MASA differences, including methods for their use in prospective safety risk analysis, i.e. to predict safety risk well ahead of an accident. The aim of the current research was to investigate how the combination of MASA and agent-based modelling can be used for retrospective safety analysis, i.e. to conduct an in-hindsight analysis of an accident occurrence. More specifically, the objective of this research was: To create a structured approach for retrospective modelling and analysis of an air transportation accident through a combined use of Multi-Agent Situation Awareness and agent-based modelling. The research showed that contributing factors to an accident can often be specified as MASA differences. This finding formed the basis for the development of a formal agent based approach to retrospective accident analysis. The working of this formal agent-based retrospective analysis approach has subsequently been demonstrated for three specific accidents from aviation history. Though information gathered from the occurrence database Skybrary, all relevant agents have been identified and their states and perceived states have been modelled at points where MASA differences are likely to occur. Finally the causation and propagation of MASA differences in the agent-based model are analysed. Based on the insights obtained from these analyses, it is concluded that a MASA and agent-based retrospective analysis of accidents yields new insights that are not obtained using a classical retrospective safety analysis.