Estimation of uncertainties and validation of computational models for structural concrete.

摘要

Over the last three decades, a large number and variety of computational tools have been developed for predicting the full non-linear response of concrete structures to imposed loadings. Many of these tools attempt to capture complex effects including the development, opening, and slip along cracks, bond degradation, dowel action, confinement, compression softening, tension stiffening, and the degrading effects of cyclic actions and corrosion. While these tools have been used in specialized applications, designers have been reluctant to use these tools in common practice for a number of reasons including the inadequacy of model validation procedures for quantifying the predictive accuracy of these tools. This thesis is concerned with establishing rigorous procedures for model validation that are well founded in reliability theory.;This study develops a statistical framework to assess the accuracy of a computational model, addresses issues specific to cracked structural concrete, and illustrates the proposed methodology in case study applications. The suggested framework is able to utilize hierarchical data structures as well as a normal single level data structure, and the accuracy of the computational model is given as a function of specimen parameters so as to account for any systematic bias of the model. Definitions are proposed for key aspects of the proposed model validation methodology to ensure unambiguous assessments of computation tools with the use of experimental test data.;Three case studies are presented to illustrate the proposed framework of model validation. The first involves an assessment of the accuracy of the shear strength of reinforced concrete beams, within which the fundamental aspects of the proposed framework are shown. The second case study involves assessing the accuracy of a non-linear finite element analysis program for predicting the full response of reinforced concrete panels subjected to in-plane shear and membrane stresses. The single-level probabilistic model is used to statistically analyze shear strength prediction, while the hierarchical probabilistic model is adopted to analyze the shear stress---shear strain response. For the final case study, the finite element analyses of prestressed bulb tee girders are used for the strength prediction and shear stress---shear strain response. In the second and third case studies, two programs, VecTor2 and ATENA are used to predict shear strength and shear stress---shear strain response.;These case studies suggest that the proposed framework for model validation is able to appropriately estimate the confidence interval of experimental observation, identify important sources of bias of computational models, and evaluate the model safety factor.