Generalizing event trees using Bayesian networks

Abstract

Bayesian networks comprise a probabilistic modelling technique representing influences between uncertain variables. Event trees are a popular technique for modelling the probability of occurrence of accidents in system safety analyses. In this paper it is shown how event trees can be generalized using Bayesian networks. This approach is applied to generalize an existing analysis of train derailment accidents at different track locations on a commuter railway. A number of separate event trees were used in the original analysis; it is shown how these can be replaced with a single generalized model. The generalized derailment model allows new locations to be analysed by selecting the state of the influencing factors appropriate to the location. Moreover, these factors are explicit in the generalized model, whereas the original event trees only included event probabilities that varied by location, with no explicit representation of the causes of these variations. The behaviour of the factors as parameters in an accident model is described, allowing more flexible system-wide risk analysis.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 19 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.