Dummy Positioning by Pre-Simulation

Innovation Intelligence®

Dummy Positioning by Pre-Simulation

Fabien Breda / Lionel Morançay

April 2013

Copyright © 2013 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved.

• Franck Le Gall

• Olivier Chertier

• Franck Njilie

• Upali Fonseka

• Erwan Mestres

Special Thanks


State of art

Current dummy positioner are very efficient for small rotation

For big rotation, we may have intersections inside the dummy

When it happens, users don’t know what to do and may think to the worst!!!


Possible solutions

Nothing

• Simulation may crash…

Remove manually intersections inside the self-contact of the dummy

• Long

• User-dependent

• Change the characteristics of the dummy

Make a pre-simulation

• Long (isolate a model, construct the subsytem, set the boundary

conditions, update the model…)

• Explicit solution is not so adapted for quasi-static solicitation

• Importing results of another code is too long


Focus on the pre-simulation solution

Hypercrash

Radioss

Make a pre-simulation

• Long (isolate a model, construct the subsytem, set the boundary

conditions, update the model…)

• Explicit solution is not so adapted for quasi-static solicitation

• Importing results of another code is too long


HyperCrash: Sub-system extraction

Joint N which leads

to intersections

Joint N-1

Joint N+1

• Select the minimum part to be taken into account

• Sub-system between Joint N+1, N-1

• All the bones

• Two new rigid bodies


HyperCrash: Sub-system characteristics

• Imposed displacement (quasi-static) : 0,314 rad in 500 ms around the axis of the joint (constant velocity for each

positioning : 0,628 rad/s or 35°/s)

• Two boundary conditions : one fixed the first rigid body, the second fixed except around the axis of

rotation/translation

• Second rigid body with ICOG = 3 and COG at the center of rotation of the joint

• Contact of the dummies

Copyright © 2012 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved. Copyright © 2012 Altair Engineering, Inc. Proprietary and Confidential. All rights reserved.

Radioss Solution

• Introducing Advanced Mass Scaling

• Possible Applications

• Roof crush

• Stamping process simulation


• Classical methods for raising time step

• Increase of mass & momentum

• All frequencies are affected

• Non diagonal mass matrix

• Assembling elementary matrices

• de large enough to obtain the target time step

• Added mass = zero / No change in total mass & momentum

• Low frequencies are almost not affected

Introducing Advanced Mass Scaling

3111

1311

1131

1113

12

ed

MM *


Crash, Buckling, High Non Linear Contact, …

Quasi Static , Non Linear

Dtx10 Dtx20

Speed up

• Decreases the number of cycles

• Cost of a cycle is more than doubled

• Solve using iterative method

• Cost of solving depends on Matrix conditioning

FM *

Introducing Advanced Mass Scaling


Roof Crush

• TAURUS model 850K elements

• Time step of 0.5ms raised to 10ms

16 cores

Standard 402044 cycles 33,4 hours

AMS 20146 cycles 4,2 hours x 7

Standard AMS

Impactor speed = 1m/s / total displacement = 200mm.


• Number of shell elements : 331246

• Dt ~ 0.2 ms (Element size < 2 mm) => more than 500 000 cycles

Stamping Process Simulation

No AMS

AMS x30


• Elapsed times / Number of cycles

• 48 CPUs, 48 SPMD domains

Intel(R) Xeon(R) CPU E5649 @ 2.53GHz (x86_64), 2533 MHz, 64449 MB RAM, 62403 MB swap

Elapsed

time

Number of

cycles Speed up Time step

No AMS 24177 sec 601260 -- 0,1979 µs

AMS x 30 4939 sec 17839 4,9 5,937µs

Stamping Process Simulation


Advanced Mass Scaling

• Well adapted to Quasi-static applications

• Energy content in high frequency domain is small

• Stamping, Roof crush, …

• May substitute to linear implicit solution if highly non linear

• Easy to setup

• No convergence issues

• Research is still ongoing to make it usable in Frontal Car Crash

• Performances factors 2 to 3 are usually obtained

• Dtx10 5 ms for a 1M element model

• w/ good energy balance

• Quality of the results to be improved

• To be coupled with RAD2RAD (MultiDomain)


Radioss: Sub-system performance

Original Model Original Model

Used of AMS

Density x1000

Use of AMS

Time step

(natural time step :

0,00036 ms)

0,01 ms (~X 30) New natural time step

: ~0,01 ms

AMS : 0,2 ms (X20)

Performance on

laptop (4 cores 2,3

Ghz with 16 GB

RAM).

Run with Radioss SP.

Estimation ~10000

secondes (2h40)

1700 secondes 30-45 secondes


Demonstration


Enhancements : planning and possibilities

• All parameters are under user control (available)

• Possibility to run the simulation on servers (target 12.110)

• Take into account initial constraint

• Based on deformed geometry (X-REF) (target Hypercrash 12.110)

• By importing initial stress in shells and solids (not yet scheduled)

• Propose a re-mesh after the pre-simulation (not yet scheduled)

• Propose a re-mesh and a mapping for initial states (not yet scheduled)

• Plug this tool for Ls-Dyna profile (target Hypercrash 12.110)

• Develop the same approach for seat deformer (target hypercrash 12.110)

• One set of parameter by dummy (not yet scheduled)

• Improve the interactivity

• Warning message when simulation crashes

• …

• Coupling with other Radioss enhancements (Rad2Rad…)

• …following YOUR FEEBACKS

Technology

Dummy Positioning by Pre-Simulation