International audience ; A probabilistic model is developed to investigate the crack growth development in welded details of orthotropic bridge decks. Bridge decks may contain many of these vulnerable details and bridge reliability cannot always be guaranteed upon the attainment of a critical crack. Therefore insight into the crack growth development is crucial in guaranteeing bridge reliability and scheduling efficient maintenance schemes. The probabilistic nature of the crack growth development model and the dependence of this model on many interdependent random variables results in significant uncertainties regarding model outcome. To reduce some of these uncertainties the probabilistic model is combined with a monitoring system installed on a part of the bridge. In addition, a Bayesian network is used to determine the dependence structure between the different details (monitored and non-monitored) of the bridge. This dependence structure enables us to make more accurate crack growth predictions for all details of the bridge while monitoring only a limited number of those details and updating the remaining uncertainties.