Premium
Bayesian Networks‐Based Probabilistic Safety Analysis for Railway Lines
Author(s) -
Castillo Enrique,
Grande Zacarías,
Calviño Aida
Publication year - 2016
Publication title -
computer‐aided civil and infrastructure engineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 2.773
H-Index - 82
eISSN - 1467-8667
pISSN - 1093-9687
DOI - 10.1111/mice.12195
Subject(s) - bayesian network , probabilistic logic , computer science , bayesian probability , conditional probability , variable (mathematics) , inference , sensitivity (control systems) , bayesian inference , process (computing) , data mining , algorithm , machine learning , artificial intelligence , engineering , mathematics , statistics , mathematical analysis , electronic engineering , operating system
A Bayesian network model is developed, in which all the items or elements encountered when travelling a railway line, such as terrain, infrastructure, light signals, speed limit signs, curves, switches, tunnels, viaducts, rolling stock, and any other element related to its safety are reproduced. Due to the importance of human error in safety, especial attention is given to modeling the driver behavior variables and their time evolution. The sets of conditional probabilities of variables given their parents, which permits quantifying the Bayesian network joint probability, are given by means of closed formulas, which allow us to identify the particular contribution of each variable and facilitate a sensitivity analysis. The probabilities of incidents affecting safety are calculated so that a probabilistic safety assessment of the line can be done and its most critical elements can be identified and sorted by importance. This permits improving the line safety and saving time and money in the maintenance program by concentrating on the most critical elements. To reduce the complexity of the problem, an original method is given that permits dividing the Bayesian network in to small parts such that the complexity of the problem becomes linear in the number of items and subnetworks. This is crucial to deal with real lines in which the number of variables can be measured in thousands. In addition, when an accident occurs the Bayesian network allows us to identify its causes by means of a backward inference process. The case of the real Palencia–Santander line is commented on and some examples of how the model works are discussed.