Thermite materials, consisting of a metallic fuel and a metal oxide, have attracted significant interest due to their high energy density and wide range of potential applications. Predicting their combustion through numerical simulations is a major challenge for their integration into industrial systems. However, this remains difficult because thermite combustion involves strongly coupled multi-physics and multi-scale phenomena, including transport processes, phase changes, and homogeneous and heterogeneous chemical reactions occurring over spatial scales ranging from nanometers to millimeters and temporal scales from milliseconds to seconds. Most studies on thermites have focused on experimental characterization of their combustion properties. In contrast, only a limited number of predictive models have been developed, and existing approaches generally rely on simplifying assumptions and global reaction kinetics that restrict their range of validity. Recent advances in high-performance computing (HPC) and numerical methods now enable direct numerical simulations (DNS) capable of accurately resolving combustion processes. In this context, the objective of this work is to improve the understanding of combustion mechanisms in aluminum-based systems, from a single particle to an Al/CuO thermite, in order to identify the dominant mechanisms and provide macroscopic models with quantitative information on mass, momentum, and heat transfer. The work is organized around three main research directions. - First, a three-dimensional DNS model of an aluminum particle burning in a reactive flow was developed. The model accounts for homogeneous gas-phase reactions, heterogeneous interfacial reactions, and their coupling with the surrounding flow. After validation against experimental data, a series of simulations was performed to investigate the influence of flow conditions and oxygen concentration on both local fields and global combustion characteristics. - Second, DNS results were used to improve macroscopic modeling. Mass, momentum, and heat transfer around burning particles were characterized in detail. The classical Spalding model was extended to aluminum combustion through the development of the Al-SM model, which highlights a robust dimensionless parameter capable of unifying the description of inert, evaporating, and reacting particles. Improved correlations for the drag coefficient and the Nusselt number were subsequently proposed. - Finally, the model was extended to an Al/CuO thermite system. A periodic medium composed of a random arrangement of aluminum and copper oxide particles was generated and simulated under different operating conditions. The influence of particle spatial organization was quantified using statistical indicators ensuring result reproducibility. Analysis of local fields and comparison with classical correlations demonstrated the need for improved modeling of interphase transfer processes in thermite systems.