Risk Analysis and Modelling of Autonomous Marine Systems

Abstract

Autonomous marine systems (AMSs) are of increasing interest for the marine and maritime industries. AMSs are engineered, computer-controlled systems that take (to some degree) decisions independent of their human operators. Different types of AMSs can be differentiated, for example, maritime autonomous surface ships (MASSs), autonomous underwater vehicles (AUVs), or unmanned surface vehicles (USVs). AMSs reduce the operational cost, the risk with respect to personnel, and the energy consumption in comparison to their conventional equivalents. AUVs are already in use and MASSs are expected to be in operation before 2020 (Kongsberg Maritime, 2017). To accept these systems, the public and authorities require that they are safe and do not have higher levels of risk than conventional systems (Danish Maritime Authority, 2018).

The objective of this thesis is to present risk analysis and risk modelling approaches for AMSs. These risk models and risk modelling approaches assist in demonstrating that AMSs are as safe as required and provide decision support during the design and operation of AMSs. This thesis addresses three issues: (i) Identification of risk-influencing factors for AMSs, (ii) presentation of risk analysis and risk modelling approaches for AMSs, and (iii) description of a risk monitoring approach for the operation of AMSs.

Risk assessments are used to analyse and evaluate the level of risk through risk models and suggest improvement measures to reduce the level of risk if necessary (Rausand, 2011). In this thesis, current risk models and approaches have been reviewed to evaluate their applicability for AMSs. AMSs have recently received more attention with respect to their development and design. Only a few risk modelling approaches exist for AMSs. It was found that software and the human operators are not considered in sufficient detail in current risk models for AMSs.

A process to incorporate the risk contribution from software into risk analysis is presented in this thesis. The process relies on the functional decomposition of software, identification of failure modes for the functions, and assessment of the effect of the failure modes on the software output through failure mode propagation. The functional level of software is defined. In addition, a functional failure mode taxonomy for software is developed from the literature. This is necessary since the current taxonomies are not coherent with respect to their level of system application, for example, the overall system level or functional level.

The identified effects on the software output are related to the effect on the external interfaces of the software, for example, human operators, other software systems, or actuators. These effects can be included in risk models, such as fault trees, event trees, or Bayesian belief networks (BBN).

This thesis also addresses the interaction between the human operators and the AMSs in risk analysis. First, the necessity to consider these interactions is highlighted in a risk management framework for AUVs. The framework identifies two phases of risk management where the human operators need to be considered; this is during risk analysis and during the identification of risk-mitigating measures.

Second, a risk model using a BBN for assessing human-autonomy collaboration (HAC) performance is presented. This BBN combines factors related to the human operators and AMSs that influence HAC performance. The most important factors are the human operators’ experience, human operators’ training, and workload. The influence of the human operators on the collaborative performance is mediated by the level of autonomy of the AMSs. Autonomous function reliability and the situational awareness capabilities of the AMSs are the most influential factors on HAC performance pertaining to AMSs.

This thesis also presents a process for developing safety indicators for the operation of AMSs. Safety indicators can be used to monitor the level of safety during the operation of AMSs. To prevent the occurrence of accidents, the proposed process allows the development of an indicator system that enables the human operators to assess whether the level of risk of operation is increasing. The indicators address subsystems and aspects of the organisation that allow the identification of organisational and technical weaknesses that may lead to an accident if not controlled.

Software governs the AMSs and controls most of the AMSs during operation. The software needs to be safe and reliable. The human operators have a supervisory role and need to act when AMSs are not capable of coping with the situation any longer. The risk modelling processes, approaches, and aspects that are described in this thesis address the need to ensure and demonstrate that AMSs are safe with respect to relevant human, technical, and organisational factors. Therefore, the implications from the risk-influencing factors identified for HAC are important for the design of human-machine interfaces and control systems to keep the human operators aware of the situation and enable them to act when required.