Bayesian networks with imprecise datasets : application to oscillating water column

Estrada-Lugo, H.D. and Patelli, E. and de Angelis, M. and Raj, Daniel D.; Haugen, Stein and Barros, Anne and van Gulijk, Coen and Kongsvik, Trond and Vinnem, Jan Erik, eds. (2018) Bayesian networks with imprecise datasets : application to oscillating water column. In: Safety and Reliability - Safe Societies in a Changing World. CRC Press/Balkema, NOR, pp. 2611-2618. ISBN 9780815386827 (https://doi.org/10.1201/9781351174664-328)

[thumbnail of Estrada-Lugo-etal-TF-2018-Bayesian-networks-with-imprecise-datasets]
Preview
Text. Filename: Estrada_Lugo_etal_TF_2018_Bayesian_networks_with_imprecise_datasets.pdf
Final Published Version
License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 logo

Download (1MB)| Preview

Abstract

The Bayesian Network approach is a probabilistic method with an increasing use in the risk assessment of complex systems. It has proven to be a reliable and powerful tool with the flexibility to include different types of data (from experimental data to expert judgement). The incorporation of system reliability methods allows traditional Bayesian networks to work with random variables with discrete and continuous distributions. On the other hand, probabilistic uncertainty comes from the complexity of reality that scientists try to reproduce by setting a controlled experiment, while imprecision is related to the quality of the specific instrument making the measurements. This imprecision or lack of data can be taken into account by the use of intervals and probability boxes as random variables in the network. The resolution of the system reliability problems to deal with these kinds of uncertainties has been carried out adopting Monte Carlo simulations. However, the latter method is computationally expensive preventing from producing a real-time analysis of the system represented by the network. In this work, the line sampling algorithm is used as an effective method to improve the efficiency of the reduction process from enhanced to traditional Bayesian networks. This allows to preserve all the advantages without increasing excessively the computational cost of the analysis. As an application example, a risk assessment of an oscillating water column is carried out using data obtained in the laboratory. The proposed method is run using the multipurpose software OpenCossan.