The Safety Engineering Group (SEG) at ERA has been involved in a number of projects using Bayesian Networks to assist with the assessment of risk. Bayesian Networks are a powerful probabilistic modelling tool that enables predictive reasoning.

A Bayesian Network is a causal graph supported by probability tables that model the conditional relationships between the variables in the network. The causal structure and conditional relationships encoded in the model enable information entered via input nodes to propagate through the model and modify the parameter values of the output nodes. The model can therefore be used for both predictive and diagnostic reasoning, and hence provide decision support.

Similar to a Fault Tree, Bayesian Networks enable prediction and “what-if” analysis. Unlike a Fault Tree however, a Bayesian Network can utilise a range of information types within the same model (i.e. not just reliability estimates), thereby widening the scope of application. Additionally, the relationships between variables in a Bayesian Network are probabilistic rather than deterministic. Deterministic relationships between data are a feature of Fault Trees and many other risk management tools available. Probabilistic relationships between data elements enable uncertainty to be encoded in the model, which is important as it assists with the representation of an uncertain world and is representative of how we reason about the world.

ERA has been involved in a number of complex risk assessments using Bayesian Networks. The following case studies illustrate some applications:
Condition Assessment Modelling
Production of a condition assessment model based on a Bayesian Network for a Middle Eastern Gas Processing Company. The model predicts corrosion rates of plant vessels in order to rank the vessels according to risk and recommends inspection intervals. One of the key benefits of using a Bayesian Network for this application was the ability to combine a range of data types within the one model.

Railway Safety Arguments
	The Rail Safety and Standards Board commissioned ERA to assess the use of Bayesian Networks and Goal Structuring Notation (GSN) for improving the efficiency of safety case submissions and approval for railways. Whilst this application would benefit from further research, it was found that a Bayesian Network could be used for parts of a safety argument where the quantification of safety goals is considered to be important.

SERENE - SafEty and Risk Evaluation using bayesian NEts
ERA developed the SERENE (SafEty and Risk Evaluation using bayesian NEts) method under the ESPRIT programme. The project’s aim was to develop a methodology that could be applied to the development of arguments for the different aspects that relate to functional safety of safety critical Programmable Electronic Systems, e.g. failure rate prediction. The project developed a methodology for constructing software safety arguments using Bayesian Networks and adapted an existing Bayesian Network
tool to support the method, evaluating the application of the method and tool through practical trials.