Bridge condition modelling and prediction using dynamic Bayesian belief networks

The development of a condition-based deterioration modelling methodology at bridge group level using Bayesian belief network (BBN) is presented in this paper. BBN is an efficient tool to handle complex interdependencies within elements of engineering systems, by means of conditional probabilities specified on a fixed model structure. The advantages and limitations of the BBN for such applications are reviewed by analysing a sample group of masonry bridges on the UK railway infrastructure network. The proposed methodology is then extended to develop a time dependent deterioration model using a dynamic Bayesian network. The condition of elements within the selected sample of bridges and a set of conditional probabilities for static and time dependent variables, based on inspection experience, are used as input to the models to yield, in probabilistic terms, overall condition-based deterioration profiles for bridge groups. Sensitivity towards various input parameters, as well as underlying assumptions, on the point-in-time performance and the deterioration profile of the group are investigated. Together with results from ‘what if’ scenarios, the potential of the developed methodology is demonstrated in relation to the specification of structural health monitoring requirements and the prioritisation of maintenance intervention activities.