Advanced Physics Models for Particle-to-Particle Interactions

Project Overview

High-speed particle transport and dust size particle-particle interactions are of significant interest to the DOE and DoD programs, multiphase flow sciences, and astrophysics flows. Current state-of-the-art macroscale (centimeters to meters) models use a point representation for particles. These point models represent the physics of transport, particle collisions, and material response at particle-scale (10s to 100s of microns). We have developed a multiscale computational approach based on data-driven physics models for time-dependent, particle-laden flows. The focus has been the development of particle-scale informed models and their evaluation in macroscale simulations. It is clear based on our investigation that using machine learning to predict particle force influence maps clearly demonstrated the viability of a data-driven approach for predictive modeling of particle-laden flows. Using Random Forest regression along with physics-based flow features, we were able to achieve highly predictive (R2 score: ~0.95) models for imposed aerodynamic loads due to the proximity of neighbors. Furthermore, our cross-validation experiments with particle-scale simulations at different Reynolds numbers indicates the ability to interpolate across different flow regimes. Two university collaborators have contributed to the goal of this effort: UC Davis contributed to the development of a physics-based material strength model required for particle-particle and particle-surface interactions and University of Florida contributed to the development of a procedure to construct force influence maps to predict proximity effects of neighbor particles at sonic speed.

Mission Impact

The project is directly aligned with specific national security applications in Weapons and Complex Integration (WCI) and Global Security (GS) Programs at Lawrence Livermore National Laboratory, and the effectiveness of the modeling approach will be demonstrated for these applications. The project is also aligned with the mission of the Computing Directorate by providing high-fidelity modeling and simulation capability.