Using Bayesian networks to represent parameterised risk models for the UK railways
MetadataShow full item record
The techniques currently used to model risk and manage the safety of the UK railway network are not aligned to the mechanism by which catastrophic accidents occur in this industry. In this thesis, a new risk modelling method is proposed to resolve this problem. Catastrophic accidents can occur as the result of multiple failures occurring to all of the various defences put in place to prevent them. The UK railway industry is prone to this mechanism of accident occurrence, as many different technical, operational and organizational defences are used to prevent accidents. The railway network exists over a wide geographic area, with similar accidents possible at many different locations. The risk from these accidents is extremely variable and depends on the underlying conditions at each particular location, such as the state of assets or the speed of trains. When unfavourable conditions coincide the probability of multiple failures of planned defences increases and a 'risk hotspot' arises. Ideal requirements for modelling risk are proposed, taking account of the need to manage multiple defences of conceptually different type and the existence of risk hotspots. The requirements are not met by current risk modelling techniques although some of the requirements have been addressed experimentally, and in other industries and countries. It is proposed to meet these requirements using Bayesian Networks to supplement and extend fault and event tree analysis, the traditional techniques used for risk modelling in the UK railway industry. Application of the method is demonstrated using a case study: the building of a model of derailment risk on the UK railway network. The proposed method provides a means of better integrating industry wide analysis and risk modelling with the safety management tasks and safety related decisions that are undertaken by safety managers in the industry.
AuthorsBearfield, George Joseph
- Theses