Earthquake-Induced Building Pounding (EIBP) is a solicitation whose consequences are difficult to quantify, both with experimental set-ups and numerical models considering the important number of parameters involved and multiple possible configurations (non-exhaustively, modal behaviour of the structures, impact location, elastic or inelastic behaviour, separation distance, number of stories, etc.). Through the increasing world urban densification, this phenomenon spotted non-exhaustively in Mexico in 2017 for instance, could become more frequent. Furthermore, the continual updates of the regulatory seismic hazard maps could result in future structural checks. Nowadays, EIBP is scarcely addressed in building codes, and the recommendations focus on avoiding the contact, which is not realistically applicable in dense urbanized areas. Building Pounding (BP) is then likely to keep occurring in the future, the first chapter of the document detailing this matter. An important part of the literature and engineering methods focus on the parametrization of Kelvin-Voigt type contact-impact models to better treat the collision. Although yielding satisfying results, these models use two to three parameters whose calibration is subject to discussion as presented in the second chapter. Also, literature has only begun to achieve performance-based risk analysis (introduced in the third chapter) on this phenomenon in the 2010's. The objective of the present work is to contribute to both aspects. Chapters 4 to 6 show and detail the relevant outcomes of the present work. To the knowledge of the author, the use of the Non-Smooth Contact Dynamics (NSCD) method is rare in civil engineering field, and more particularly in the building pounding analysis. This approach is numerically unconditionally stable (so large time steps can be chosen), and it only uses one scalar parameter called the coefficient of restitution e defined between 0 and 1. It represents the loss of energy of the system upon contact. First, test cases, whose analytical response are known, are applied to validate the method. Thereafter in the fourth chapter, e is assessed in the framework of BP, by comparing numerical and experimental data of 2.5m and 5m high steel-frame structures, either "launched" against one another or subjected to seismic solicitation, and pounding on their reinforced concrete slabs. These experimental data were made available by the EMSI laboratory at CEA Saclay in the frame of national project ANR SINAPS. e=0.6 yielded good and consistent results, both in terms of kinematics and spectra analysis. Then, after the experimental calibration of the e value, risk analyzes are conducted. In engineering projects, risk analysis helps the decision makers by giving, considering the method used, either probabilities or occurrence of events of Engineering Demand Parameter (EDP) of interest (e.g. the interstory drift or base shear), reaching a critical value. These risk analyzes usually need to use Intensity Measures (IM), which are indicators of the "importance" and destructiveness of a seismic event, e.g. the Peak Ground Acceleration (PGA), or the Pseudo-Acceleration (Sa) at the fundamental period T of the structure (Sa(T))), they are studied on linear single-storey structures in chapter five. IMs have properties called Efficiency and Sufficiency, that represent respectively their prediction capabilities and their unconditional use regarding ground motions parameters such as magnitude and source-to-site distance. First, commonly-used IMs properties are studied both with and without considering EIBP, to rank them and select the most appropriate ones. PGA, Sa($T$), and mainly the average pseudo-spectral acceleration (Savg($T$)) present the best characteristics, although it is Sa($T$) that is used for further risk analyzes due to the deeper understanding and experience scientists have towards it. Then, based on Sa($T$), intensity-based and risk-based analyzes are conducted, which are the Conditional Scenario Spectra (CSS), and the Incremental Dynamic Analysis (IDA). These methods have both their respective advantages and drawbacks. For instance, both use very large, potentially non-realistic, scaling factors, and the IDA alone represents a considerable cost in computation runtimes. Drift values, maximum and accumulated impact energies, and number of impacts are the EDPs of interest studied throughout chapters five and six, respectively for linear single-storey and two-storey structures, and for nonlinear single-storey and multi-storey structures. It showed and confirmed with the literature body that, considering what is analysed, pounding can either be beneficial by strongly reducing the interstorey-drifts of the heavier structure, or damaging by increasing slightly these drifts and inducing impact energies and peaks of velocities and accelerations in floor response spectra. These elements are displayed by the means of Fragility Curves (FC) and Structural-Response Hazard Curves (SRHC) that can serve as a basis for future calculations using the NSCD method. When structures behave linearly and elastically, it showed that impact might indeed increase drift values et probabilities of reaching a damage state for the stiffest structure, but not as significantly as it reducing the drift for the more flexible one. CSS and IDA yield mostly similar results, especially in terms of interstorey-drifts and impact energies. In such configuration, preference would then go to CSS because averagely 10 times faster than the IDA. Thereafter, a 1D Kinematic-Isotropic Hardening (KIH) law on a SDOF model was used and compared with its homologous linear version. Unless reaching the highest damage states, nonlinearities do not have significant effects on fragility curves and structural responses. It showed that if residual displacements are output parameters of interest, IDA would this time suggested as main methodology, as only 87/178 ground motions of the CSS database lead structures to developing yielding. Overall, nonlinearities did not affect much the global structural response, either being beneficial as drifts decrease, or destructive only for high damage states (drift greater than 1.5\%) and only for the stiffest structure. Afterwards, the numerical study of a 1/3rd scale multi-storey reinforced concrete structure named CAMUS1 (whose mass and stiffness were calibrated through experimental data) ends the report, by accounting for traction and compression damage of concrete, and steel yielding. using respectively the La Borderie and Menegotto-Pinto constitutive laws. A Multiple Stripes Analysis (MSA) in applied on CAMUS1 and an adjacent linear elastic structure. It showed that building pounding is more damaging for this type of structure, bringing it faster to damage states of drifts, but especially to crack opening, and increasing significantly the high frequencies of floor-response spectra.
