Arman Pazouki Simulation-Based Engineering Laboratory Department of - PowerPoint PPT Presentation

MULTIPHYSICS SIMULATION USING GPU Arman Pazouki Simulation-Based Engineering Laboratory Department of Mechanical Engineering University of Wisconsin - Madison 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems

Acknowledgements • Prof. Dan Negrut • Dr. Radu Serban • Hammad Mazhar • Andrew Seidl • Colleagues from Simulation Based Engineering Laboratory, University of Wisconsin- Madison • National Science Foundation • NVIDIA 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 2

Motivation and Background 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 3

Motivation 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 4

Fluid Flexible Bodies Rigid Bodies 6/30/2014 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 5

Fluid Simulation: SPH • Continuity: • Momentum (Navier-Stokes): • Lagrangian Kinematics: W • Weakly Compressible model a r ab  h • XSPH b • Shepard Filtering S • J. Monaghan, Smoothed particle hydrodynamics, Reports on Progress in Physics 68 (1) (2005) 1703-1759. • M. Liu, G. Liu, Smoothed particle hydrodynamics (SPH): An overview and recent developments, Archives of Computational Methods in Engineering 17 (1) (2010) 25-76. 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 6

Neighbor Search and GPU programming Case Study: How GPU can affect an algorithmic design VS 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 7

Algorithm 1 Core property: saving contacts list Parallel threads: Bins Successful for rigid body dynamics: O(1e7) Advantages Failed for SPH: O(1e6) • One calculation per intersection • Possibility of re-using contacts list • Arbitrary shapes 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 8

Algorithm 2 Core property: find intersection whenever needed Advantage. • More process, less memory • Fixed size spheres Screen shot from particles demo, NVIDIA CUDA (CUDA Samples) Parallel threads: Particles • Reduces memory access • Improves cache hits. 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 9

Rigid and Flexible Bodies Dynamics • 3D rigid body dynamics • Gradient-deficient ANCF beam element Dynamics Dynamics   e g a Me + Q Q Q d V F  i i   T     T     d t M   e 11 Q = EA   dx+ EI   dx i   11     e e  ω d         J ω i  J T  g T Q = S g A dx i i i i dt s a T Kinematics Q = S ( x ) F a      0 Kinematics d X z y   t  i  V      0   i   L d i z x r        0   y x L   r d q 1 t  G  x i T   e = ; r ( x, e ) = S ( ) ; x e v ( x, e ) = S ( ) x e   ,   q q q q i i R d 2 y x w z r          G q q q q   R z w x y   r i     T q q 1 0 x   i q q q q     w z y x 3 12 S shapefunction matrix • E. Haug, Computer aided kinematics and dynamics of mechanical systems, Allyn and Bacon Boston, 1989. • A. Shabana, Dynamics of Multibody Systems, Cambridge University Press, 2005. A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 10 3/25/2015

Fluid-Solid Interaction Boundary Condition Enforcing (BCE) markers for no-slip condition • Rigidly attached to the solid body (hence their velocities are those of the corresponding material points on the solid) • Hydrodynamic properties from the fluid Rigid bodies/walls Flexible beams 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 11

Parallelization • thrust::reduce_by_key to reduce surface reaction forces and torques on to nodal values • Custom kernels to update solid objects • Fine grain parallelization • Position • Rotation • Velocity • Angular velocity • … 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 12

Example Simulations 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 13

Flow in porous media • Example applications • Oil Recovery • Biology • Diffusion of macro- molecules within tissues • Blood flow through muscles 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 14

Simulation of dense suspensions • “Dense” suspension • Finite size particles (rigid bodies) interaction • Drafting, Kissing, and Tumbling (DKT) • Short range interactions • Lubrication and collision • Flow characteristics • Particle Reynolds number ≤ 1.0 • Channel Reynolds number: 66 • Channel Dimension: (1.1, 1.0, 1.0) m • Volumetric concentration: 40% • Computational aspects • 23,000 rigid ellipsoids: (1.5, 1.5, 2.0) cm • 2,000,000 SPH markers. • Simulation performed on a single GPU, NVIDIA GTX 480 • 3.2 seconds of dynamics Animation shows the channel mid-section • 72 hrs to complete. 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 15

Interacting rigid and flexible objects in channel flow 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 16

Performance 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 17

Scaling analysis on NVIDIA GeForce GTX 680 • Rigid body dynamics • Flexible body dynamics • Fluid flow 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 18

Scaling analysis (all together, table) 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 19

Validation 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 20

Particle migration in 2D and 3D Poiseuille flow • Sphere in pipe flow, • Cylinder in channel flow • Transient Poiseuille flow • Effect of rigid body rotation • T. Inamuro, K. Maeba, F. Ogino, IJMF 26 (12) (2000) 1981-2004. • J. Morris, P. Fox, Y. Zhu, JCP 136 (1) (1997) 214-226. • J. Schonberg, E. Hinch, JFM 203 (1) (1989) 517. • D. Oliver, Nature (194) (1962) 1269-1271. • B. H. Yang, J. Wang, D. D. Joseph, H. H. Hu, T.-W. Pan, and R. Glowinski, JFM 540 (2005) 109. 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems APS- Pittsburgh 21

Radial distribution of particles in suspension [1/2]  • 192, 10 hour long simulation a R 0.07  a av l   • 14 seconds real time L ( )( )( ) [0,0.69]  R R • Bootstrapping method,95% confidence interval    Re 60, 0.027% Increasing distance from inlet • G. Segre, A. Silberberg, Nature (189) (1961) 2. 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 22

Applications 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 23

Hanging flexible beam in viscose fluid • Flexible cantilever in contained fluid • Track position of beam tip 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 24

Flow cytometry using microfluidic techniques • Fluorescence and laser-beam cell sorting   • Limited particle size, a 50 m • Unknown effect of external field on cell viability • Purification of 3D micro-tissues and cell aggregates • Finite size particles,   25..500 m a 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 25

Work in progress 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 26

Tha hank nk Yo You! Chrono::FSI (Project Chrono: https://github.com/projectchrono) Simulation Based Engineering Lab (SBEL) University of Wisconsin-Madison Email: pazouki@wisc.edu 3/25/2015 A Lagrangian-Lagrangian Approach For the Simulation of FSI Problems 27