16 Décembre – Thesis defense - Wafaa Abdallah
14 h Amphi G - Building A29 (University of Bordeaux / Talence campus)
Contribution of probabilistic approaches for the non-destructive evaluation of concrete by multi-physics inversion for integration into mechanical models.
Maintaining the performance level of civil engineering structures involves their inspection and auscultation, using destructive and non-destructive methods. Non-destructive evaluation (NDE) methods allow the assessment of physical (porosity, degree of saturation,..) and/or mechanical (strength, modulus of elasticity) concrete properties based on conversion models linking NDE measurements and concrete properties measured on samples (destructive measurement). These models are built by the implementation of an experimental campaign combining NDE tests with a number of destructive tests. This non-destructive evaluation strategy faces many challenges due to the sensitivity of the measured parameters to endogenous and/or exogenous factors, in addition to the heterogeneity of concrete and its spatial variability. Taking this spatial variability into account is a main interest to make the NDT methodology more reliable. It can be determined, for example, from the estimation of the spatial variability of the NDE measurements obtained during the diagnosis of reinforced concrete structures, thus making it possible to reduce the number of cores. In addition, the calibration and inversion procedures of the conversion models must be consolidated by controlling the influencing factors such as the number and the position of samples (cores), the control or even the reduction of measurement uncertainties and perhaps the combination of NDE techniques.
In the framework of this thesis, the indicators porosity Ф and degree of saturation Sr as well as their variabilities, including their spatial correlation lengths, were estimated simultaneously by three complementary NDT techniques: ultrasound, electrical resistivity and radar. These two indicators are essential in order to estimate the service life and to predict the evolution of physicochemical degradation in reinforced concrete structures. Three multi-physical conversion models of different shapes were considered. Several real data sets were considered for calibration by minimizing the root mean square error (RMSE). Spatial variability was generated using the “Circulant Embedding” spatial discretization method and an isotropic exponential autocorrelation function. Numerical approaches for the calibration and the inversion of these three conversion models were then proposed. The effect of measurement uncertainty, the combination of two or three NDT techniques and the number of samples on the NDE quality of Ф and Sr, in terms of RMSE and risk curves, were also evaluated.
At last, spatial variability has been integrated, as part of a probabilistic approach, in mechanical models to study its effect on the behavior at failure of a reinforced concrete shear wall.2D numerical simulations were carried out considering the Fichant mechanical damage model for concrete. The effect of the variability of Young's modulus E and tensile strength f_t, considered as input parameters, on the behavior law, damage fields and crack opening fields were analyzed. The results showed the usefulness of taking into account the spatial correlation in the localization of the damage.