This paper is focused on the probabilistic updating of the local parameters of a structural model by Bayesian networks. The approach uses their inference property to update the probability function parameters (mean and standard deviation) related to the variables from the measurements of the structural response given by load tests. To make satisfying the computational effort, the choice of the number of random variables is discussed and relies on several sensitivity indicators given by a structural reliability analysis. The study of a prestressed beam highlights the different corpus required for applying this updating approach: the different structural mechanics theories for providing a pertinent mechanical model, the reliability theory to analyze the sensitivity of the different variables, the Bayesian network theory to update the a priori probability functions of the input variables. The comparison with the experimental data shows that the relative error between the average experimental and numerical response can be significantly reduced with the help of the proposed updating process.
