Fractures are ubiquitous geological structures controlling both the hydraulic and mechanical rock properties. Numerical modelling of fracture networks is therefore an important step for the simulation of physical processes in fractured rock mass, for many industrial applications such as nuclear waste disposal. Most of the time, a direct observation of the fractured volume is not feasible, and fractures cannot be modelled deterministically. Hence, the modelling must be stochastic, which makes it possible to generate three-dimensional networks, statistically equivalent to measurements and observations, but neglecting the spatial correlations resulting from the chronological fracturing process. On the other hand, purely mechanical models require too much numerical resources to model such dense networks. This thesis aims to develop genetic models, making it possible to model multiscale, dense fracture networks using simplified mechanical rules. The fracturing process can be divided in three simplified stages: nucleation, propagation, and fracture arrest. In this work, we show that the spatial organization and scaling properties of such generated fracture networks result from these processes. We quantify these correlations using mathematical tools from the fractal theory and quantify their impact on the connectivity properties of generated networks. Finally, a refined study of fractures mechanical properties such as friction, and remote stress boundary conditions responsible for fracture development, showed how much they need to be considered into the genetic modelling framework.