Particle-based fluid simulations using GPUs
2 likes1,538 views
The document outlines the capabilities and features of Fluidyna GmbH's particle-based fluid simulation software utilizing smoothed particle hydrodynamics (SPH) and GPU acceleration. It details various applications, including powertrain oiling simulations, multi-phase flows, and complex geometries, while emphasizing the benefits of using this meshless CFD solver. A roadmap for future developments is also provided, indicating planned enhancements and features to be implemented in upcoming software versions.
![www.fluidyna.com
Introduction to SPH: walls
615.10.2015
S. Adami, X.Y. Hu, N.A. Adams (2012) A Generalized Wall Boundary Condition for Smoothed Particle Hydrodynamics,
J. Comput. Phys. 231(21): 7057-7075.
Sketch of wall boundary
general wall bc.
• no-slip condition
• arbitrary geometry
• Neumann condition
for pressure
“…The results of this paper show that while all methods give reasonable results they can be greatly improved
by a combination of (a) using three layers of fluid particles as boundary particles, (b) interpolating the pressure
and velocity from the fluid particles to the boundary particles in the manner described by Adami et al. [1] and
(c) using density diffusion as first suggested by Molteni and Colagrossi [19]…”
A study of solid wall models for weakly compressible SPH, Valizadeh & Monaghan (2014)](https://image.slidesharecdn.com/eatc2015gearboxsimulation-151113105326-lva1-app6891/85/Particle-based-fluid-simulations-using-GPUs-6-320.jpg)
Particle-based fluid simulations using GPUs
- 1. Particle-based fluid simulations using GPUs FluiDyna GmbH, Edisonstr. 3, 85716 Unterschleißheim b. München, Germany 115.10.2015
- 2. www.fluidyna.com • Introduction to • Introduction to SPH & • General capabilities of the code • Code strong points: powertrain oiling simulation • More capabilities • Roadmap Outline 215.10.2015
- 3. www.fluidyna.com • was founded in 2006 • Dr. Thomas Indinger, CEO • Located in Unterschleißheim (Munich), Germany • Specializing in CFD: • Consulting • In house codes: nanoFluidX, ultraFluidX, Culises • Hardware (Nvidia Preferred Solution Provider) • partnership: 2014 Introduction: the company 315.10.2015
- 4. www.fluidyna.com Introduction to SPH 415.10.2015 • Lagrangian discretization elements (“particles”) advect with the flow • Kernel smoothing with compact support • Particle-particle interactions model the fluid / solid: ∆t ∆t Schematic dambreak simulation with SPH 𝐴 𝑟 = 𝐴 𝑟′ 𝑊 𝑟 − 𝑟′, ℎ 𝐝 𝐫′ ≈ 𝑏 𝐴 𝑟𝑏 𝑉𝑏 𝑊 𝑟 − 𝑟𝑏, ℎ 𝛻 𝐴 𝑟 = − 𝐴 𝑟′ 𝛻𝑊 𝑟 − 𝑟′, ℎ 𝐝 𝐫′ ≈ − 𝑏 𝐴 𝑟𝑏 𝑉𝑏 𝛻𝑊 𝑟 − 𝑟𝑏, ℎ 𝑑 𝑣 𝑑𝑡 = − 1 ρ 𝛻𝑝 + ν𝛻2 𝑣 + 𝑔 𝑑𝑣 𝑎 𝑑𝑡 =. . . − 1 𝑚 𝑎 𝑏 𝑉𝑎 2 𝑝 𝑎 + 𝑉𝑏 2 𝑝 𝑏 𝛻𝑎 𝑊𝑎𝑏 ℎ . . . http://plaza.ufl.edu/jeffjtd/SitePics/kernelpic.jpg
- 5. www.fluidyna.com Introduction to SPH: walls 515.10.2015 Problem of the wall BCs: • Requirements: • Impermeability • No-slip condition Monaghan and Kajtar, SPH particle boundary forces for arbitrary boundaries, Comp. Phys. Comm. 180(10):1811-1820, 2009 Morris et al., Modeling Low Reynolds Number Incompressible Flows Using SPH, J. Comput. Phys. 136(1):214-226, 1997 + Coupling with FEM - Additional evolution equation - 3D formulation non-trivial (missing currently) + No-slip condition accurately imposed - Simple geometries - Multi-value problem + Straightforward implementation - Arbitrary geometries - Numerical parameter ? Ferrand et al., Unified semi-analytical wall boundary conditions for inviscid, laminar or turbulent flows in the meshless SPH method, Int. J. Num. Methods in Fluids 71:446-472, 2013
- 6. www.fluidyna.com Introduction to SPH: walls 615.10.2015 S. Adami, X.Y. Hu, N.A. Adams (2012) A Generalized Wall Boundary Condition for Smoothed Particle Hydrodynamics, J. Comput. Phys. 231(21): 7057-7075. Sketch of wall boundary general wall bc. • no-slip condition • arbitrary geometry • Neumann condition for pressure “…The results of this paper show that while all methods give reasonable results they can be greatly improved by a combination of (a) using three layers of fluid particles as boundary particles, (b) interpolating the pressure and velocity from the fluid particles to the boundary particles in the manner described by Adami et al. [1] and (c) using density diffusion as first suggested by Molteni and Colagrossi [19]…” A study of solid wall models for weakly compressible SPH, Valizadeh & Monaghan (2014)
- 7. www.fluidyna.com • is … • based on the Smoothed Particle Hydrodynamics method (SPH). • a meshless CFD solver. • most powerful for complex flows in arbitrary geometries. • using GPU-acceleration to minimize simulation times. Introduction to 715.10.2015 “No grid generation, nearly no limits.”
- 8. www.fluidyna.com Using a Lagrangian framework it is possible to… • Simulate free-surface flows • Simulate flows with moving rigid bodies General capabilities 815.10.2015 Fluid flow in an agitated boxDam break within oscillating tank
- 9. www.fluidyna.com Using a Lagrangian framework it is possible to… • Simulate multi-phase flows w/ high density and viscosity ratios General capabilities 915.10.2015 Water phase (ρ=1000 kg/m³) Air phase (ρ=1 kg/m³) Air entrappment in fluid phaseMulti-phase dam break simulation
- 10. www.fluidyna.com Using a Lagrangian framework it is possible to… • Simulate flows in/through complex geometries General capabilities 1015.10.2015 Water flow (red particles) through porous membrane (section of a fuel cell cathode) under gravity.
- 11. www.fluidyna.com Using a Lagrangian framework it is possible to… • Simulate flows with rotating rigid bodies General capabilities 1115.10.2015 Particle animation „Footprints“ Free-surface deformation Pathlines showing internal fluid motion
- 12. www.fluidyna.com Using a Lagrangian framework it is possible to… • Simulate filling of mixing tanks General capabilities 1215.10.2015 Volume-rendered fluid animation • Training case.
- 13. www.fluidyna.com Powertrain Oiling Simulation 1315.10.2015 as Powertrain Oiling simulation tool • Preprocessing with HyperMesh • Simulation setup (currently) with ASCII-file • High performance simulation using GPU • Postprocessing (currently) with and
- 14. www.fluidyna.com • Geometry discretization CAD input data (STEP, IGES, STL, …) is discretized using HyperMesh Powertrain Oiling Simulation 1415.10.2015 CAD input data Particle discretization for
- 15. www.fluidyna.com • Simulation using GPU Powertrain Oiling Simulation 1515.10.2015 NVIDIA GPGPU Example: Coarse resolution simulation result
- 16. www.fluidyna.com • Velocity vector plots • Pathlines • Flow fields on cut section Powertrain Oiling Simulation 1615.10.2015 • Postprocessing • Torque measurement • Flow rate measurement • Volume-rendered flow visualization
- 17. www.fluidyna.com Powertrain Oiling Simulation 1715.10.2015 • Generic example Rotating gears Hydrostatic settling
- 18. www.fluidyna.com • Gearbox case 1 • 5.8 million particles • 2000 rpm • 1 s of physical time (35.5 rotations) • 32.5 hrs on a 1 x K40 • Gearbox case 2 • 2.5 million particles • 5400 rpm • 0.3 s of physical time (27 rotations) • Pre-processing: 2 days • 3 days on a 1 x K40 • Possible speed up: 5 s of physical time in 3-4 days on 8 x K80 Some performance numbers 1815.10.2015
- 19. www.fluidyna.com • Nvidia Tesla K40 • Max Memory Bandwidth 288 (GB/sec) • Peak performance* 4.29 TFLOPS • Memory 12GB GDDR5 • Cores 2880 • Nvidia Tesla K80 • Max Memory Bandwidth 480 (GB/sec) • Peak performance* 5.6 TFLOPS • Memory 24GB GDDR5 • Cores 4992 • Initial investment: • 8 x K80 GPUs, 2 x 8-core CPU, 128GB RAM, 2 TB • Approximately 40,000 €/$ What do you need? 1915.10.2015 *performance for single precision
- 20. www.fluidyna.com • Heat transfer • Surface tension • Prescribed motion input for a geometry • One-way coupling with MotionSolve • Buoyancy and (rigid body motion) More features 2015.10.2015
- 21. www.fluidyna.com • Decoupled temperature equation. • Dirichlet boundary conditions: set constant temperature or allow it to evolve in time. • Heat transfer among all phases: fluids, walls and moving walls. Heat transfer 2115.10.2015
- 22. www.fluidyna.com • Two-phase surface tension model • Accurate and fast. • Planned extensions: free surface surface tension, multi-phase capability. Surface tension 2215.10.2015 „Tip-streaming“ (volume rendered) Drop in shear flow (Couette device)„Square droplet“ test case
- 23. www.fluidyna.com Prescribed motion 2315.10.2015 • Simulate any complex motion • Piston rod • Crankshaft • Planetary gearboxes t x y z η θ ζ 0 0 0 0 0 0 0 0.1 0.1 0 0.2 10 0 0 0.2 … … … … … … Translation & rotation as a f(t) Planetary motion of a cube in a pool of water.
- 24. www.fluidyna.com • Import geometry and its prescribed positions as a function of time. • One-way coupling with MotionSolve results. Prescribed motion 2415.10.2015
- 25. www.fluidyna.com • In development (testing phase). • Free translation and rotation of a rigid body as it interacts with the fluid. • Optional locking of motion (lock individual motion along/around axes). • Input: arbitrary definition of center of mass location, mass of the body, moment of inertia. • Measure forces and torques exerted on the body by the fluid. • For use in: • Naval industry • Hydro-turbines Buoyancy (rigid body motion) 2515.10.2015 *Public domain images.
- 26. www.fluidyna.com Roadmap (tentative) 2615.10.2015 2015 2016 2017 Q3 Q4 Q1 Q2 Q3 Q4 Q2Q1 Core version • Free-surface flows • Rotating/moving geometries • multi-GPU • multi-phase • Decoupled T equation • Two-phase surface tension • Dynamic load-balancing • Rigid body motion • Coupled energy equation (viscosity) • Inlet-outlet boundary conditions v1.0 v1.04 v2.04 • Non-Newtonian fluids • Single-phase surface tension • Steady-state coupling with FV/LBM codes • Dynamic coupling with FV/LBM codes • Von-Neumann temperature boundary conds. • Variable resolution • Multi-phase surface tension models v1.10
- 27. www.fluidyna.com THANK YOU FOR YOUR ATTENTION. 2715.10.2015 FOR MORE INFORMATION ABOUT VISIT www.nanofluidx.com