Fault diagnosis is an essential task in ensuring the smooth operation of complex dynamic systems. The consequences of faults can be serious, leading to loss of life, harmful emissions to the environment, high repair costs and economic losses caused by unplanned production line stoppages. The work developed in this paper concerns the modeling and diagnosis of faults (sensor faults, system faults, actuator faults) in hybrid dynamic systems using our multi-model approach (which combines two sub-models, one continuous and the other discrete). The aim is to integrate three well-known tools in the literature: the Bond Graph, the Observer and the Timed Automata, to design a global diagnostic model. The hybrid dynamic system is modeled by connecting the tools for the continuous part, i.e. the bond graph and the observer, to the timed automata for the discrete part. The resulting model is used for fault diagnosis in two stages: The first is fault detection by analyzing the residuals generated by the system output and that of the observer. The second step involves fault localization, which results from analysis of the signature matrix and temporal identification of the system. The proposed method combines the advantages of these tools to obtain the best performance, particularly in the fault location phase. The simulation results prove the effectiveness of the proposed model for the hybrid dynamic system. Moreover, these results also evaluate the performance of the proposed diagnostic approach while reducing non-detections, detection delays and false alarms.