This study proposes a computational framework for the high-fidelity analysis and damage assessment of concrete fragment fields under blast loading. The framework unifies physical simulation with machine learning, integrating an S-ALE-DEM-FEM coupled model with a novel Multi-level and Multi-scale HDBSCAN clustering algorithm. This enables a complete process from simulating blast loading and structural fracture to the intelligent identification of discrete fragments. Numerical simulations accurately reproduced the damage patterns and fragment cloud kinematics of a reinforced concrete slab under contact explosion, validating the model. Verification via near-field explosion tests on plain concrete blocks demonstrated that the proposed algorithm, through multi-level constraints including mechanical pre-screening and kinematic validation, accurately captures the critical fragment size-mass distribution and significantly outperforms traditional methods by overcoming parameter sensitivity and over-segmentation from mechanical interlocking. Furthermore, parallel algorithm optimization achieved high computational efficiency for systems with millions of particles. This integrated framework provides an efficient and reliable solution for generating quantitative fragment data, offering considerable potential for advancing engineering applications in explosion damage analysis. • A novel S-ALE-DEM-FEM framework exploits DEM's asymmetric strain-rate hardening for high-fidelity blast simulation. • Multi-level multi-scale HDBSCAN algorithm overcoming over-segmentation from mechanical interlocking. • Accurate fragment size-mass distribution prediction validated by dual-path explosion tests. • An efficient parallel architecture supports practical damage assessment on million-particle systems.
Li et al. (Fri,) studied this question.