Abstract: This PhD thesis tackles the general context of road safety, focussing on the safety of light vehicles (LV) in bends. A reliability engineering methodology is proposed to predict dangerous trajectories, based on the statistical processing and probabilistic modelling of actual trajectories in a bend. In the first part of this work, simple and robust probabilistic models are built to describe trajectories measured in an instrumented bend. The models are transforms of scalar normalized second order stochastic processes which are slightly stationary, ergodic and non-Gaussian. They offer a realistic description for the observed random variability of the Vehicle-Infrastructure-Driver system. They also inherently circumvent possible difficulties in the identification of the dominant parameters which control the system. The second part of this work is devoted to the development and implementation of a reliability engineering strategy intended to associate a risk level to each trajectory at a bend entry. Based on the joint use of probabilistic methods for modelling uncertainties, reliability engineering for assessing risk levels and statistics for classifying and processing the trajectories, this approach provides a realistic answer to the tackled problem. From its design and its possibilities, the proposed reliability engineering methodology constitutes a significant contribution to the development of warning procedures the deployment of which are expected to notably reduce the number of accidents in bends.
