29
IMPETUS AFEA | SOLVER® 1 GTC May 2012 Fluid-Structure-Interaction Using SPH and GPGPU Technology IMPETUS AFEA | SOLVER® Jérôme Limido Jean Luc Lacome Wayne L. Mindle

IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

  • Upload
    others

  • View
    6

  • Download
    0

Embed Size (px)

Citation preview

Page 1: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

1

IMPETUS AFEA | SOLVER® 1

GTC May 2012

Fluid-Structure-Interaction Using SPH and GPGPU Technology

IMPETUS AFEA | SOLVER®

Jérôme Limido

Jean Luc Lacome

Wayne L. Mindle

Page 2: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

2

IMPETUS AFEA | SOLVER® 2

2D Sloshing Water in Tank

2

Page 3: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

3

IMPETUS AFEA | SOLVER® 3

History

Development started in 2007

Languages

Fortran

Cuda

C++

Structure Source Code % of Lines

Finite Element 40%

Smoothed Particles 10%

Discrete Particles 10%

GPU computing 25%

Misc(output,util,lic) 15%

Number of lines of code: 110,000

IMPETUS Afea | SOLVER ®

3

Page 4: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

4

IMPETUS AFEA | SOLVER® 4

Particle methods are very well suited for the GPU

- All particle methods run on the GPU

- Peak acceleration can reach 100x on simple (non industrial) cases

Finite element method is more complex to port to the GPU - Most time consuming parts are treated on GPU

- Classic FEM could be easier to port on GPU

Contact algorithm is very complex to optimize on the GPU - Balance between computation time and uniqueness of contact definition

- Current state : 50% CPU time spent in contact algorithm

Porting to GPU

4

Page 5: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

5

IMPETUS AFEA | SOLVER® 5

Meshless Methods : GPU vs CPU

Sort – Neighbor list Particle Interaction System update

0.0% 50.0% 100.0%

Sort

Entities Interaction

System Update

Sort Entities Interaction System Update

GPU 6% 92.5% 1.5%

CPU 0.7% 99% 0.3%

System Update

Particle Interaction

Sorting

5

Page 6: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

6

IMPETUS AFEA | SOLVER® 6

Gas

High Explosives

Soil

Fluids

DP (Discrete Particles)

SPH (Smoothed-Particle Hydrodynamics)

Particle Methods

6

Page 7: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

7

IMPETUS AFEA | SOLVER® 7

Kinetic molecular theory for gases (modified to handle high explosives)

Discrete grains in contact. Grain size distribution, friction, damping and contact stiffness are adapted to match a given EOS.

DP (Discrete Particles)

Buried Mine Blast

7

Page 8: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

8

IMPETUS AFEA | SOLVER® 8

4M Sand Particles

60K HE Particles

Run Time: 1 hour

8

DP (Discrete Particles)

Page 9: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

9

IMPETUS AFEA | SOLVER® 9

t = 0 t = 0.1ms t = 0.2ms

Penetration of a Sand Filled Profile

9

DP (Discrete Particles)

Page 10: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

10

IMPETUS AFEA | SOLVER® 10

• Unique SPH formulation

• Compute accurate and stable pressure fields

• Multi-fluid applications

• Air, Water and Gas

• Efficient SPH to Finite Element contact algorithm

SPH (Smoothed-Particle Hydrodynamics)

10

Page 11: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

11

IMPETUS AFEA | SOLVER® 11 11

Hydraulic RAM – Projectile Impacting a fluid filled structure

Page 12: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

12

IMPETUS AFEA | SOLVER® 12

Hydraulic RAM – Projectile Impacting a fluid filled structure

12

Page 13: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

13

IMPETUS AFEA | SOLVER® 13 13

Hydraulic RAM – Projectile Impacting a fluid filled structure

Page 14: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

14

IMPETUS AFEA | SOLVER® 14

DAM Break Run Time: 1 hour

14

Page 15: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

15

IMPETUS AFEA | SOLVER® 15

Run Time: 1 hour DAM Break

15

Page 16: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

16

IMPETUS AFEA | SOLVER® 16

Water Filled Balloon

Half filled Balloon Dropped followed by Base Excitation

848,947 SPH Particles

Run Time: 24.7 hours

16

Page 17: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

17

IMPETUS AFEA | SOLVER® 17

Water Filled Balloon

Half filled Balloon Dropped followed by Base Excitation

848,947 SPH Particles

Run Time: 24.7 hours

17

Page 18: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

18

IMPETUS AFEA | SOLVER® 18

Heart Valve 468,931 SPH Particles

Run Time: 12.5 hours

18

Page 19: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

19

IMPETUS AFEA | SOLVER® 19

Heart Valve

19

Page 20: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

20

IMPETUS AFEA | SOLVER® 20 20

Heart Valve

Page 21: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

21

IMPETUS AFEA | SOLVER® 21

Speed up allows development of new algorithms to solve

more complex problems and still provide an accurate solution.

Speed up is a way to increase the model size and still provide

an accurate solution.

What does increased speed mean?

Performance of Real World Problems

Full scale blast simulation (Discrete Particles + FE) : ~10x

Full scale FSI simulation (SPH + FE) : ~ 40x

21

Page 22: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

22

IMPETUS AFEA | SOLVER® 22

Development Goals

• Accuracy

• Robust Algorithms

• Simplified Input

22

Page 23: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

23

IMPETUS AFEA | SOLVER® 23

GPU Technology

23

Page 24: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

24

IMPETUS AFEA | SOLVER® 24

Accurate and Robust

Unique accurate higher order solid elements that can handle extremely

large deformations (good for plasticity and bending, fully integrated, no

hourglassing and no element inversion).

Cubic 64-node hexahedron Cubic 20-node

Tetrahedron

Cubic 40-node

Pentahedron

Linear and Quadric: Hex, Tet and Penta Elements too.

24

Page 25: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

25

IMPETUS AFEA | SOLVER® 25

• Unique higher order element technology is well suited for processes involving extreme

deformations.

• Elements are accurate, even at large aspect ratios, and they behave well in dynamics.

Accurate and Robust

25

Page 26: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

26

IMPETUS AFEA | SOLVER® 26

Performance

Standard Explicit Dynamic FE Codes

Solid Elements based upon 1 point integration

1. Because Higher Order elements are too computational intensive

2. Require At least 3 elements through the thickness to get good

bending behavior

3. Require Hourglass Control (artificial stiffness) to numerically stabilize

the elements

IMPETUS AFEA Solver

Good performance because of GPU

Technology

Up to 64 integration points per solid element

26

Page 27: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

27

IMPETUS AFEA | SOLVER® 27

Methods and Algorithms

Designed with as few numerical parameters to tune as possible.

User Input is kept to a Minimum:

• Material models

• Contact Definition

• Failure Criteria

What Does this means to the Analyst?

More time to focus on the physics of the problem instead of troubleshooting input data.

Simplified Input

27

Page 28: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

28

IMPETUS AFEA | SOLVER® 28

Demands on Simulation Software:

Larger and Larger Models multiple GPUs

• Visualization

• Complex mixing, phase transition multi-phasic

• Composites: crash and impact multi-scale approach

Future Development

28

Page 29: IMPETUS AFEA | SOLVER®...Title: Fluid-Structure-Interaction Using SPH and GPGPU Technology Author: Wayne L. Mindle,Jérôme Limido, Jean Luc Lacome Subject: We present the use of

29

IMPETUS AFEA | SOLVER® 29

Website: www.impetus-afea.com

email: [email protected]

Contact Info

29