Éléments finis stochastiques étendus pour le calcul en fatigue de joints soudés avec géométries aléatoires
Auteur / Autrice : | Olivier Pasqualini |
Direction : | Franck Schoefs, Mathilde Chevreuil |
Type : | Thèse de doctorat |
Discipline(s) : | Sciences de l’Ingénieur, Mécanique |
Date : | Soutenance en 2013 |
Etablissement(s) : | Nantes |
Ecole(s) doctorale(s) : | École doctorale Sciences pour l'ingénieur, Géosciences, Architecture (Nantes) |
Partenaire(s) de recherche : | autre partenaire : Université de Nantes. Faculté des sciences et des techniques |
Jury : | Président / Présidente : Anthony Nouy |
Examinateurs / Examinatrices : Anthony Nouy, Alaa Chateauneuf | |
Rapporteurs / Rapporteuses : Alaa Chateauneuf |
Mots clés
Mots clés contrôlés
Résumé
Welded joints are essential components for the construction of various fixed or floating structures. These elements are so important that we need to fully understand the fatigue process in order to foresee the structural behaviour under cyclic loading. The fatigue lifetime computation of a welded joint depends of various parameters such as the geometry of the structure. The stress concentration factor computation Kt is an efficient key parameter to model the fatigue lifetime. It has the advantage to link theoretical stress with the maximum value of local stresses and so with the fatigue lifetime thanks to S-N curves. In order to compute the Kt coefficient from real data and their uncertainties, some measurements along welded joints were realized by using a Non-Destructive Control device with laser process measurement. A statistical analysis of these measures were carried out to model the geometrical parameters by random variables and to identify their probability distribution. Kt-computation were performed by using eXtended Stochastic Finite Element Method; this computation method combines the Stochastic Finite Element Method, efficient to solve problems governed by random physical inputs, and the XFEM, efficient to implicitly define the domain geometry by using level-sets. In particular, we use a non-intrusive method of least-square computation to carry out, with a few numbers of random values, a Kt formulation defined on Polynomial Chaos base. From these results, an original semi-probabilistic model is suggested which introduces the geometrical parameters.