GPU-accelerated SPH modeling of flow-driven sediment erosion with different rheological models and yield criteria

Flow-driven sediment erosion exhibits multiphase characteristics and involves reversible solid-liquid mass exchange, wherein various rheological models and yield criteria are available, but their cooperative performance remains unclear. Here, six combinations of popular rheological models and yield criteria with varying physics are implemented in the Smoothed Particle Hydrodynamics (SPH) to reproduce four classic erosion experiments. We propose a GPU-accelerated Effective Interface Searching Technique (EIST) that reduces computational complexity from O(N²) to O(1). Our results suggest that the combination between Herschel-Bulkley-Papanastasiou (HBP) model and Mohr-Coulomb (MC) criterion best captures the erosion and waterfront, whilst the Drucker-Prager (DP) criterion appears to over-predict the erosion. The HBP-MC model reveals that velocity decreases exponentially and sediment concentration increases linearly before saturation concentration along with the depth in the movable sediment layer. This work provides a robust GPU-based EIST method and a useful reference for identifying appropriate adoptions of rheological models and yield criteria that vary in physics.