GPU-based simulation of dense suspensions: coupling the DEM and the LBM in standard C++

Diplôme Master GPU-based simulation of dense suspensions: coupling the DEM and the LBM in standard C++ eng 2023-08-16 author Maggio-Aprile Raphaël Anthony 0009-0002-1497-0174 906614 Département d'informatique (DINFO) director Latt Jonas 512884 Département d'informatique (DINFO) director Coreixas Christophe Guy 899040 Département d'informatique (DINFO) 88 Granular suspensions simulations are essential for comprehending the behavior of particle-laden fluids. This research project focuses on enhancing simulations of granular suspensions using GPUs. The composition of these suspensions typically involves a high volume fraction of particles, presenting challenges for accurate and efficient simulations. Traditionally, simulations employ the Discrete Element Method (DEM) for particles on the CPU, while fluid dynamics simulation is handled on the GPU using the Lattice Boltzmann Method (LBM). However, the communication between the CPU and GPU in this approach hampers simulation speed. Existing DEM frameworks, like LIGGGHTS, are effective for particulate systems but lack efficient GPU deployment. To overcome this limitation, our approach involves adopting a data-oriented strategy, creating a portable, GPU-optimized code in standard C++ with parallel algorithms from the C++17 standard. This enables us to fully leverage the computational power and parallel processing capabilities of GPUs, enhancing performance and efficiency in granular suspension simulations. The code is implemented in standard C++ for platform versatility and can be seamlessly deployed on GPUs using compilers such as nvc++ from Nvidia. Our goal is not immediate superiority but laying a robust foundation for future advancements. Representing particles as simple spheres with interactions described by a spring-dashpot model, we simulate fluid using LBM with the Two Relaxation Time (TRT) model, coupling based on the partially saturated bounce-back method. This exploration aims to gain insights into GPU parallel computing complexities and develop specific approaches for effective granular suspension simulations using this powerful hardware. By harnessing GPU computational power and adhering to standard C++ practices, our aim is to create an innovative framework for more efficient and comprehensive simulations of granular suspensions. Through the development and validation of our open-source C++ framework, we contribute to advancing simulations in this domain, opening new possibilities for understanding particle-fluid dynamics on a larger scale. Granular suspensions GPUs Discrete Element Method (DEM) Lattice Boltzmann Method (LBM) Parallel computing Data-oriented approach Computational efficiency Particle-laden fluids High volume fraction Particle interactions Spring-dashpot model Two Relaxation Time (TRT) model Bounce-back method Open-source framework C++ programming GPU optimization Computational power Fluid dynamics Nvc++ Stlbm Partially saturated method (PSM) C++17 STL parallel algorithms LIGGGHTS Dewey 025.063 Master in Computer Science Prix Arditi 2023 en Informatique 32 Département d'informatique 18 false 728 Scientific and Parallel Computing 49486762-f339-463e-9299-583c09bb8549 Master thesis Public Travail_Master_Raphael_Maggio_April.pdf 1269541 application/pdf 8a7644ca8dcbe925025bf2d32700e9951b6a6c46c792a75923337be12f6c04cf Vignette Public Screenshot from 2024-01-24 13-48-43.png 582081 image/png openAccess Faculté des sciences / Département d'informatique

Title	Published in	Access level	OA Policy	Year	Views	Downloads

GPU-based simulation of dense suspensions: coupling the DEM and the LBM in standard C++				2023	426	1,013

Archive ouverte UNIGE

Maggio-Aprile, Raphaël

account_balance Affiliation entities

Research groups

Affiliation entities