Hybrid Bayesian Networks for the Reliability Analysis of Systems with Continuous Variables

The standard way of dealing with continuous variables into reliability models is to discretize them, resulting in discrete state models. The present paper proposes an approach where continuous system variables can be directly exploited by resorting to Hybrid Bayesian Networks (HBN), where both continuous and discrete variables can be mixed in a general way. This allows one to: model the inter-dependencies between discrete state components or subsystems, model the inter-dependencies between continuous system variables, model the influence of contextual information on system variables and components, model the definition of specific system events or conditions given specific values of the system variables. We will show how the above issues can be captured in a principled way by the HBN formalism, by making the final analyses more grounded on the actual values of every system variable. We finally present a case study where the model of a granule storage tank system of a petrochemical plant is considered, and we present the results of specific analyses implemented as inference on the HBN model.