Analyse et interprétation statistique des mesures d'auscultation de piézométrie à l'interface béton-rocher d'un barrage voûte

Large concrete arch dams are frequently subjected to the phenomenon of the aperture of their rock-concrete interface, which involves strongly non-linear features. While in engineering practices, monitoring data are classically analysed using multi linear regression (MLR) models, the present work proposes to interpret monitored piezometric levels using first Artificial Neural Networks (ANNs), in order to draw conclusions on the structural behaviour of the dam, and more specifically on the temporal evolution of the aperture at the heel of the dam. A specific methodology is proposed regarding the way to process the available data in order to avoid extrapolation, and the results are compared to those of the classical MLR model. This comparison shows that when it comes to studying non linear phenomena, ANNs greatly enriches the analysis and provides a quality physical interpretation, compared to MLR. Second, a non linear regression model based on a phenomenological law is used, that is currently still under development. Finally, perspectives are presented, that will be investigated afterwards during this PhD thesis.