This PhD thesis is focused on the correlation between ground motion and the buildings response. In particular, we interested in reducing the uncertainties that occur in vulnerability methods in order to assess more precisely the seismic vulnerability of existing structures. In order to do this, this thesis is based on the use of data recorded within buildings. Records of ambient vibrations and low amplitude earthquakes within buildings are used to characterize the elastic behaviour of these structures. Linear models can then be defined in order to study the various components of the variability of the fragility curves, for the first damage level. Earthquake that have occurred since the 1970s have been recorded within Californian buildings. From these data, we focused on the relationship between the building response and parameters describing the noxiousness of ground motion. Depending on the indicator used to represent the noxiousness of earthquakes, the building response can be estimated more or less accurately. In particular, indicators involving the dynamic parameters of the buildings are more related with the response of structures, which is represented by its averaged inter-story drift. Grouping buildings by typologies (defined according to their main material of construction and their height) can improve the variability in the response of buildings. Indeed, by providing more information on the structure, we can reduce the epistemic component of variability. In addition, by combining noxiousness parameters, the accuracy in the prediction of the building response can be improved. A functional form is thus proposed to estimate the averaged inter-story drift within the structures, for several typologies of buildings. This functional form is then used to assess fragility curves and can also be used to get an estimate of damage after an earthquake, by comparing the values of inter-story drift from given by the functional form with reference values (FEMA, 2003). Finally, a hybrid method is proposed for the construction of fragility curves, involving a nonlinear model. The parameters of this model are defined so that the response of the model fits the earthquake recordings, which were made within buildings. This model is then used to evaluate the components of variability and to build fragility curves for all damage levels.
