gpu parallelization is the perfect match with the discrete ...€¦ · gpu parallelization is the...
TRANSCRIPT
GPU Parallelization is the Perfect Match with the
Discrete Particle Method for Blast Analysis
Wayne L. Mindle, Ph.D. 1
Lars Olovsson, Ph.D. 2
1 CertaSIM, LLC and 2 IMPETUS Afea AB
GTC 2015
March 17-20, 2015
San Jose, CA
Blast Analysis
What is Blast Analysis?
IEDs (Improvised Explosive Device)
Buried
Surface Laid
Land Mines Conventional Devices
Air Blast
Impact on Buildings
Modeling Soil
Soil Properties
Type of soil
• Rock, Sand, Clay
• Combination of all 3
Moisture Content
Density
Constitutive Model
Equation of State (EOS)
Photos from http://freebigpictures.com
Conventional Method for Modeling Blast Scenario ALE Solver
Eulerian Solver for Soil and HE
Langrangian Solver (Finite Element) for the structure
Detonation
Point
( )RV t
Molecules Particles
Cylinder Test to Calibrate the HE
Kinetic Molecular Theory for Gas
Average distance between Particles is
large compared to the particle size*
Equilibrium exists*
Molecules obey Newton’s Law
Molecular collision is perfectly elastic
(*Not valid for HE requires modification)
One Particle represents ~ 1015 – 1020 molecules
Discrete Particle Method (DPM) - HE & Air
Air modeled with same approach as HE
Discrete Particle Method (DPM) - Soil
Normal
Spring Constant Normal Damper
Tangential
Spring Constant
L L
HE
Soil
Soil Unit Cell Soil Domain
Soil Model
Rigid particles
Inter particle contact for both friction and damping
Grain size distribution, friction, damping and contact stiffness
are adapted to match a given EOS
Soil is packed using a unit cell with periodic boundaries which
means there are no gaps between cells
Soil Calibration
Soil Density*
Friction Coefficient between soil and the structure*
Spring Stiffness between soil particles
Friction Coefficient between soil particles
Damping Coefficient between soil particles
Soil Particle Radius
*Determined with standard soil tests
Remaining parameters are calibrated for a specific soil
by simulating a blast test and optimizing the values to
match the measured Maximum Impulse.
Discrete Particle Method (DPM)
Soil is automatically filled around any object
placed in the soil domain.
Soil and HE must be able to accurately
follow the contours of the structure that it
impacts in order to correctly account for the
dynamic loading.
Discrete Particle Method (DPM) - Example
Influence of an Embedded Object in the Soil
Discrete Particle Method (DPM) - Example
Discrete Particle Method (DPM) - Example Particle Interaction with an Embedded Object
Experimental Results for Mine Blast Loading
Flat Plate
V180 120 Degrees
V120 90 Degrees
V90
Mine Blast Loading: Experiments and Simulations* C.E. Anderson, Jr., T. Behner, C.E. Weiss, S. Chocron, R.P. Bigger
SwRI® Report 18.12544/011
April 2010
Prepared for US Army RDECOM-TARDEC
*Simulation results in this study used the Eulerian Solver CTH with runtimes of 96 hours
Discrete Particle Method (DPM) - Results
Discrete Particle Method (DPM) - Results
Test ID
Soil
Moisture
Content
IMPETUS Afea Results
Maximum Velocity
(m/sec)
Relative Error
from Average
Target Maximum
Velocity
Within
Test
Scatter
Runtime
(mins)
V180-07-30 7% 5.52 1.3% YES 19
V180-07-20 7% 6.52 -12% YES 20
V120-07-25 7% 4.03 5.7% YES 20
V090-07-25 7% 2.56 -2.7% YES 20
V180-14-20 14% 7.03 -2.1% YES 18
V180-21-20 21% 7.62 -9.0% YES 17
Discrete Particle Method (DPM) – GPU Performance
Millions of Particles
Run t
ime in M
inute
s
Flat Plate Buried Mine
Test ID V180-20-7
The Generic Vehicle Hull is a real structure that has
been blast tested.
Tests include physical Hybrid III Dummies, similar to
what is used in car crash tests, to understand the
potential affect of the blast load and to evaluate blast
mitigation technologies.
RDECOM-TARDEC Generic Vehicle Hull
RDECOM-TARDEC Generic Vehicle Hull
RDECOM-TARDEC Generic Vehicle Hull
RDECOM-TARDEC Generic Vehicle Hull Simulation includes the IMPETUS-Afea Hybrid III 50th Percentile Dummy Model
Many unclassified studies from past researchers have utilized
fictitious vehicle geometry due to the non-availability of realistic
information. Due to the sensitive nature of the work performed
by the Department of Defense, data generated from testing
military vehicles is usually classified, making it difficult to share
data in the public domain. In order to increase the operational
relevance of studies performed by the wider scientific
community, the US Army Tank Automotive Research,
Development and Engineering Center (RDECOM-TARDEC) has
fabricated a generic vehicle hull to help evaluate blast mitigation
technologies, and also shared an FEA model of the same for
purposes of this research.
Acknowledgment
The simulations were performed using the IMPETUS Afea
Solver which is the only commercially available Nonlinear
Explicit Transient Dynamic Solver that takes full advantage of
GPU Technology for parallel processing.
4 million discrete particles were used to model the Soil and HE.
Runtime on a workstation using one K40 GPU
12 hours for a 20 ms simulation.
RDECOM-TARDEC Generic Vehicle Hull
GPU Functionality
Discrete Particle (Soil and Air Blast) Runs on both CPU and GPU
GPU acceleration about 10x
At this point in time using more than 1 GPU for a single
job has no benefit but allows for larger models to be
run.
The Tesla K40 GPU with 12 GB of memory
Eliminates the need to use more than 1 GPU per analysis for
current problems that have been run.
Contact Info
Wayne L. Mindle, Ph.D CertaSIM, LLC
www.certasim.com
Lars Olovsson, Ph.D. IMPETUS Afea AB
www.impetus-afea.com