Sensitivity analysis and uncertainty propagation in multiscale modeling of concrete

Electricité de France operates numerous large facilities and thus, is to warrant their safety and their service-ability. Industrial plants endure aging induced by the environment, the plant usage or unexpected pathologies. Thus, the properties of the materials are evolving with time. In addition, in the prospect of building new facilities, the responsibility of the future owner is to prescribe the exposure class and the required properties of a set of concrete depending on their location in the plant. In both cases, the assessment of the structural integrity requires to handle monitoring, in some cases, additional non destructive evaluation and structural computation in combination with the knowledge or the estimation of the history of mechanical, thermal or hydraulic loadings. It must be kept in mind that when decade-old structures are considered, concrete prescription at the time of construction can not be expected to provide an explicit knowledge of the relevant parameters required for aging analysis. Due to the high variability of the initial compounds (cement oxides composition, water to binder ratio, aggregate porosity...), it is to be expected that effective transport or mechanical properties will exhibit some strong variability. Two questions must be raised: is it possible to estimate some concrete properties taking advantage of the knowledge of the initial mixture? What is the impact of the variabilities and uncertainties of the initial compound on the overall properties? As a first attempt to answer these two questions a numerical approach combining homogenization and uncertainties propagation theories is proposed.