Recherche de sgluons dans des états finals multitops avec le détecteur ATLAS auprès du LHC à $\sqrt{s}$ = 8 TeV

Despite its experimental confirmations, the Standard Model, which describes both the elementary particles and three of the fundamental interactions, has some theoretical and experimental limitations. As a consequence of these limitations, a Physics not described in the context of the Standard Model could exist : the New Physics. This work focuses on the search for scalar gluons, commonly dubbed $sgluons$, predicted in various New Physics theories. A phenomenological study considers various final states (containing between two and four top quarks) and signatures (with one or two charged leptons) that could arise from sgluon production. The sensitivity of an ATLAS-like experiment to this signal is estimated and the results indicate that one of the most promising final states contains four top quarks in a signature with two charged leptons (electron or muon) of the same electric charge. The search for sgluons within the ATLAS experiment uses the previous results. Thus, it only investigates the 4-top topology in a signature containing two same-sign charged leptons, $b$-jets and missing transverse energy. It represents the first search ever for sgluons in this final state. The main steps of the analysis (background estimation, selection optimisation and statistical interpretation) are detailed. The comparison with the data recorded in the first months of 2012 ($L = 14.3 fb^{1}$) shows no significant excess and is interpreted in terms of a limit on the sgluon mass. The latter is found to be $0.80^{+0.02}_{-0.03}$ TeV. The extension of this analysis to the full 2012 ATLAS dataset is also presented in this document and shows an increased expected sensitivity to sgluon masses up to $0.94 \pm 0.03$ TeV. Finally, this document also presents the analyses and results of performance studies. The first one is related to the high voltage system of the ATLAS Tile Calorimeter (TileCal). The main conclusion is that only 7 channels out of the 9852 TileCal ones are fond to be problematic. The second study focuses on the rejection of the pile-up effect by the means of a selection applied on the so-called JVF (Jet Vertex Fraction) variable. The efficiency of this selection is computed both in data and simulation and shows some differences that have to be corrected in the simulation. The derivation of the correction factors and their associated systematic uncertainties is detailed in this document.