Non parametric continuous Bayesian belief nets in reliability of bridges under traffic load

Probabilistic design of infrastructure is based on estimates of design values. These are mostly done using classical statistical models. The statistical procedures used in structural reliability are restricted due to lack of data. The lack of data is most evident when the estimation is conditional on infrequent situations. Non Parametric Continuous Bayesian Belief Nets (NPCBBN) in combination with structured expert judgment could help approximate situations that are difficult to observe in the data. The advantage of using NPCBBNs is that data can be generated artificially once a model has been quantified. Additionally, the use of NPCBBN provides the user with the possibility to do fast updating on the joint distribution once evidence becomes available. This could be of great advantage for decision makers. In this paper first NPCBBN are described. As an example, with data from highway RW16 in the Netherlands for April 2008 a BBN for bridge design load is quantified. Variables included are axle load, number of axles per vehicle, velocity, total vehicular weight, and vehicular length. Possibilities to extend the model with structured expert judgment are discussed.