Using Simulation for Systems of Systems Hazard Analysis

For any safety-critical system, thorough and complete hazard analysis must be performed if the system is to be acceptably safe to operate. For the emerging class of systems known as Systems of Systems (SoS), however, performing hazard analysis is extremely difficult because of the complexity of SoS and the environments they inhabit. Traditional exploratory hazard analysis techniques commonly rely upon fixed models of component interaction, and have difficulties exploring the effects of multiple coincident failures. They therefore cannot be relied on to provide adequate hazard analysis of SoS. This thesis presents a hazard analysis approach that uses multi-agent modelling and simulation to explore the effects of deviant system behaviour within a SoS. A systematic process is defined for developing multi-agent models of SoS, starting from existing models in the MODAF architecture framework and proceeding to implemented simulation models. Throughout this process, a variety of cross-checks between model artefacts provide confidence that the model remains true to the original description and that it adequately describes the SoS being analysed. The exploratory simulations created by the process generate a substantial amount of data concerning the behaviour of the system under a variety of deviations. In order to identify the significant contributory causes of SoS accidents, a tool-supported analysis technique is presented that utilises both machine learning and agent behaviour tracing. The approach is evaluated against explicit requirements identified for SoS hazard analysis, and through application of the modelling and analysis process to case studies. The case studies demonstrate that the approach can reveal hazards that would be difficult to discover by existing manual hazard analysis techniques.