Prise en compte des données expérimentales dans les modèles probabilistes pour la prévision de la durée de vie des structures

Abstract

Probabilistic engineering mechanics addresses problems in which the parameters of a mechanical model are considered as uncertain and thus modelled in a probabilistic framework. Various methods of uncertainty propagation and sensitivity analysis are available to study the properties of the random response of a model. Many industrial applications have been carried out over the last 20 years and show that probabilistic methods can be useful.

In some cases, the weak point of the computational process is the construction of the probabilistic model of the input parameters, often by absence or lack of data collected on these parameters. Moreover, it may happen that the available experimental data is not directly related to the model input parameters of interest. In this context, the major objective of the thesis is to develop a general formalism of identification of the probabilistic models starting from the available experimental data. In this context, we propose a formulation to estimate the random vector of the input parameters of a model in two cases:

• at the design stage of the structure: when the input parameters cannot be measured whereas measurements of response quantities exist, a probabilistic inverse problem may be posed and specific procedures have been developed to solve it;

• during the service lifetime of a structure: a Bayesian framework is suited to update the prior probabilistic model, using the experimental information collected in time on the structure of interest.

The first part of the thesis deals with the identification of input random variables using response measurements, viewed as results of experimental investigations carried out at the design stage of a structure. Starting from measurement responses, the proposed method allows one to solve the probabilistic inverse problem using a semi-parametric representation of the unknown input random variables. The second part of the thesis deals with the assessment of existing engineering structures. The purpose is to justify that the operation of a particular structure can be carried on by making use of experimental information available on it. Two approaches are proposed: the first one is based on well-known reliability methods, and more specifically on an inverse FORM approach. The second is directly derived from the Bayesian framework: the input random variables are updated starting from a prior probabilistic model and experimental measurements of the mechanical model response obtained at different time instants on the same structure. The two proposed Bayesian updating schemes introduced are applied to the long-term prediction of creep strains in concrete containment vessels used in nuclear power plants.

The methods pertaining to each of the two main categories of problem are also applied to fatigue problems: on one hand for the high-cycle thermal fatigue assessment of pipes in the nuclear industry and, on the other hand, to predict crack propagation curves.

Keywords

Structural reliability - Bayesian updating methods - stochastic inverse problems - Thermal fatigue

BibTeX cite

@PHDTHESIS{PerrinThesis,
  author = {Perrin, F.},
  title = {Prise en compte des données expérimentales dans les modèles probabilistes pour la prévision de la durée de vie des structures},
  school = {Universit\'e Blaise Pascal, Clermont-Ferrand, France},
  year = {2008}
}