Enhancing multi-sphere clump informed by particles geometry via correcting dynamic properties and mechanical interactions in DEM

Multi-sphere clumps are favored to approximate irregular particles in discrete element model (DEM) due to algorithmic simplicity and efficiency, which, however, leads to approximation errors in dynamical properties and contact forces for motion integration. The latter becomes substantial when cost-effective DEMs are pursued with an acceptable minimum number of subspheres per clump (SPC). This study endeavors to eliminate these errors while maintaining DEM accuracy for low-SPC clumps. Investigations are conducted on a new database where diversely shaped particles are triangulated and stored with local geometries of each vertex. Surface error is defined to quantify the deviation between each boundary subsphere of clump and its surrounding mesh vertices. Errors in dynamic properties are eliminated by optimizing subsphere density distribution via an unconstrained linear system. Contact force errors are alleviated by correcting stiffness via an average curvature radius weighted by local characteristic areas of surrounding vertices, and by correcting penetration depth via a local shape-weighted surface error as an offset approximation. A vertex-subsphere information mapping is established for real-time corrections in DEM. Results show that the enhanced clumps significantly improve DEM accuracy at low SPC. Once SPC $\ge$ 100, the predicted compression-rebound responses during dynamic collisions and stress–strain-strength behaviors from triaxial tests closely match the converged predictions at SPC = 300. Detailed analysis reveals that the correction of dynamic properties, surpassing the common voxel-grid approximation, achieves machine precisions and is crucial for updating particles motion/orientation in collision dynamics, while the correction of contact forces is more critical for quasi-static simulation by predicting more realistic microscopic force chains. Our findings suggest that the enhanced clumps at SPC = 100 can yield sufficiently high-accurate and cost-effective DEM, being promising for modern large-scale computations.