This paper proposes a novel and efficient mesostructure generation framework integrating stochastic geometry with physically based packing. First, a random radius field (RRF) method is developed, utilizing multi-scale noise superposition and topology optimization to generate 3D aggregates with realistic and controllable morphologies. Second, a packing strategy based on Rigid Body Dynamics (RBD) is developed to simulate the physical casting process including gravity falling and vibration, achieving high-density aggregate skeletons. The framework is validated through the generation of a multi-phase mesostructure and the fracture simulation of recycled aggregate concrete (RAC). The simulation results successfully reproduced the crack propagation patterns and damage evolution paths associated with different aggregate shapes. These findings confirm the capacity and effectiveness of the proposed framework as a robust tool for the mesoscopic modeling of heterogeneous concrete materials.