La connaissance du domaine de variation des reponses d'une structure dues aux incertitudes qui decoulent des tolerances de fabrication, des conditions limites et des excitations est essentielle dans un processus de conception globale. Nous proposons dans cet article une alternative au calcul de la probabilite de defaillance afin de caracteriser la fiabilite lorsque l'information sur les incertitudes parametriques est limitee. Une approche non-probabiliste de fiabilite basee sur une modelisation convexe des incertitudes ainsi que sur la fonction robustesse permet de relier les trois principaux composants de l'analyse d'incertitude : le modele de la structure, les incertitudes et le critere de performance. Dans cet article, nous nous interessons au probleme inverse qui consiste a determiner le domaine maximal d'incertitude tel que les critere de performance soit respecte. La methodologie proposee est appliquee a deux applications dans le domaine de l'elastodynamique.
A knowledge of the variability in structural responses due to uncertainties in manufacturing tolerances, boundary conditions and excitations is essential for design in order to ensure reliable performance. We are interested in making a reliable decision when knowledge of uncertainties is quite limited and when the use of a probabilistic law is unjustified. A non-probabilistic concept of reliability based on the info-gap robustness function allows us to define the three main components of the uncertainty analysis: structural model, uncertainty model represented by info-gap models, and a performance criterion. This article concentrates on the inverse problem which consists in determining the maximal volume in the domain of uncertainty within which the performance criterion is guaranteed. The proposed methodology is illustrated based on two applications in the field of linear elastodynamics.