Mastering numerical simulations of the behaviour of railway stators remains an important challenge for designers. This allows both the understanding of some physical phenomena and the improvement of design in presence of different sources of uncertainties. The approach proposed in this work consists firstly on building and validating a numerical model of a typical design stator. By carrying out numerical-experimental confrontations and updating models, this first step allowed us to characterize the mean behavior of this heterogeneous assembling and mainly to establish generic modeling rules for other design stators. The second part of this work deals with the investigation of uncertainties affecting the structure or its model. In order to take into account all uncertainties types while performing a calculation of uncertainties propagation, a stochastic hybrid method, combining parametric and non-parametric models, was proposed. Because of the large sizes of finite element models of stators, the problem is treated in a component mode synthesis context. It amounts to carry out an approach reanalysis. In order to ensure a good compromise between reasonable calculation times and an acceptable precision, a generalized variant of the Combined Approximations method (VCA) has been introduced and adapted to component mode synthesis. The VCA method allows both a significant gain in computation time, comparing to an exact calculation, and a high robustness performance comparing to a standard reduction method or an improved method by static residual vectors.
