Crashworthiness Workshop - Altair HTC 2012

HTC 2012 : Crashworthiness Training
Shan Bala
Program Manager
Innovation Intelligence® May 15, 2012

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

Agenda

1. TeamCenter-HyperMesh Integration for Assembly – Shan Bala
2. Forming Results Initialization – Subir Roy
3. Multidomain in Radioss – Jean-Pierre Bobineau


TeamCenter-HyperMesh Integration Overview

Teamcenter Side

PLM XML EXPORT

Import PLMXML
Batchmesh
Assemble
SIEMENS Connections
TEAMCENTER Export HM/decks
Part#:
Export PLMXML
YT29-89112081-9901002

PLM XML EXPORT

HyperWorks Side 3

Solution Elements

© 2009. Siemens Product Lifecycle Management Software Inc. All rights reserved.

Solution Elements
Configuration

Teamcenter PLMXML HYPERMESH
Data Model Transfer Mode Mapping Table
Process

Script
PLMXML PLMXML CAD PLMXML Post PLMXML
HM

HM

HM
TC

TC
Export Import Translation Export Operations Import
Environment

PLMXML Working
Teamcenter HYPERMESH
File Directory


Solution Elements - Details

Configuration
- The target is to use only one PLMXML
Transfer Mode for all use cases.
- The HYPERMESH Mapping Table is a
configuration file available in the
HYPERMESH directory. This file is common
to all sites and can only be modified by the
administrator.

Process Environment
- The user should be able to start manually - The Working Directory is the temporary
each process independently either from location where Teamcenter and
Teamcenter or from the working directory. HYPERMESH are exchanging the files.
- The same process is applied for all use - The PLMXML File exported by Teamcenter
cases. is overwritten by HYPERMESH at the
- The HYPERMESH-PLMXML Import and HYPERMESH-PLMXML Export process after
HYPERMESH-CAD Translation processes a copy of the original file has been created.
may be considered as one unique process


Solution Elements
Teamcenter – PLMXML Export
Configuration

Process

Script
HM

HM

HM
TC

TC
Environment

PLMXML Working
File Directory


Solution Elements
HyperMesh – CAD Translation
Configuration

Process

Script
HM

HM

HM
TC

TC
Environment

PLMXML Working
File Directory


Solution Elements
HyperMesh – PLMXML Export
Configuration

Process

Script
HM

HM

HM
TC

TC
Environment

PLMXML Working
File Directory


Solution Elements
Script – Post Operations
Configuration

Process

Script
HM

HM

HM
TC

TC
Environment

PLMXML Working
File Directory


Solution Elements
TeamCenter – PLMXML Import
Configuration

Process

Script
HM

HM

HM
TC

TC
Environment

PLMXML Working
File Directory




Teamcenter Side

PLM XML EXPORT

Import PLMXML
Batchmesh
Assemble
SIEMENS Connections
Part#:
Export PLMXML
YT29-89112081-9901002

PLM XML EXPORT

HyperWorks Side 13



Teamcenter Side

PLM XML EXPORT

Import PLMXML
Batchmesh
Assemble
SIEMENS Connections
Part#:
Export PLMXML
YT29-89112081-9901002

HyperWorks Side 14


Product Demo

Demo

Automated Forming Results Initialization for Full
Vehicle Crash Models

Subir Roy
Innovation Intelligence® Director – Industry Solutions
May 15, 2012


Outline

 Motivation

 Options

 Implementation

 Validation

 Application

 Conclusion


Motivation

1) Enable seamless incorporation of stamping effects into structural CAE to improve
accuracy

2) Leverage the effect of work hardening in stamping for possible weight reduction


Option 1

• Map incremental analysis results

• Use accurate incremental stamping analysis with adaptive mesh followed by
mapping of results to structural mesh

• Need stamping experts to define a feasible process which is difficult at the early
product feasibility phase

• Time consuming to run large number of parts

• Possibility of error from mapping between somewhat different geometries


Option 2

• Use inverse analysis directly on structural mesh

• Use One Step stamping analysis on the structural mesh and directly include
the stamping results with the structural model

• Addendum effect needs to be approximated with edge boundary condition

• Possible to run large number of parts within a short time

• Accurate enough for crash analysis


One Step vs Incremental (Numisheet 2005 part)

Incremental 1-step


Validation with Test Data (PSA, Altair EHTC 2010)

Test With stamping

Test


Validation with Test Data (PSA, Altair EHTC 2010)


PSA Peugeot Citroën:
HyperWorks Improves Development Process

Challenge: Define a new simulation process
development that includes accurate component data
(thickness, residual strains, etc.)

Solution:
• Use HyperForm and RADIOSS in product
development
• Evaluate the impact of the manufacturing process onto
the component formability
• Seamlessly include forming results into RADIOSS
crash simulations

Business Impact:
• Reduce overall development time
• Get a better correlation with test results
• Improve product quality and process robustness Different deformation pattens (Reference, Without and With Blankholder) at 0,008 Sec

“We are in the position to predict crash results via simulation much better than in the past. Building
on this approach will lead to sustainable time savings and better products for our customers.”
– Fabien Beda, Senior Engineer Crash & Fluid Simulations PSA Peugeot Citroën


Workflow


User Interface

I. GUI Mode

• User remains in the model setup environment (HyperMesh/HyperCrash)
• Stamping analysis conducted in background on user’s machine
• Results posted for review
• Include files created at the final step
• Solvers supported: RADIOSS and LSDYNA
• Demo

II. Batch Mode

• User provides the structural model and list of parts (from any pre-processor)
• Stamping analysis launched in batch mode on a cluster
• Include files created at the final step
• Solvers supported: RADIOSS and LSDYNA
• Demo


User Interface (HyperMesh)

Review results
Select parts to initialize


User Interface (HyperCrash)

Select initialization option Review parts initialized


Features

1) Material properties interpreted from crash model
2) Addendum effect via edge Blankholder force via following options:
a. Optimum FLC: Iteratively optimizes the blank holder force based on FLC to
minimize failure
b. Qualitative: Low, Medium, High, None
c. Optional: Cut off plastic strain to eliminate local hot spots
3) Automatic undercut check parts invalid for stamping
4) Automatic hole filling
5) Run in parallel mode


Materials Supported

RADIOSS • LSDYNA
• M2_PLAS_JOHNS_ZERIL
• MATL24
• M3_HYDPLA
• MATL33
• M4_HYD_JCOOK
• MATL36
• M22_DAMA
• MATL37
• MLAW23
• MATL39
• M27_PLAS_BRIT
• MATL81
• M32_HILL
• MATL98
• M36_PLAS_TAB
• MATL122
• M43_HILL_TAB
• MATL123
• M44_COWPER
• MATL133
• M48_ZHAO
• M57_BARLAT3


Case Study 1: Ford Taurus Frontal Crash Model

Plastic strains from forming


LSDYNA Frontal Impact Crash Results

Without stamping With stamping


Case Study 2: Dodge Neon Side Impact Model

Plastic strains from forming


RADIOSS Side Impact Crash Results

Without stamping With stamping


Conclusion

 Incorporating stamping data can improve the correlation of structural
CAE models for stiffness and deformation modes

 The magnitude of the influence of stamping on structural CAE
depends on the loading path

 Standardization of structural CAE processes with stamping data for full
vehicle models is now feasible

 Case studies to continue for NVH and Durability


Results Mapper 11.0

• General purpose mapping tool inside HyperCrash
• Map thickness, plastic strain, stresses, fiber orientation
• Read forming data from Radioss, Dyna, AutoForm
• Write mapped data to Radioss, Dyna, Abaqus input format
• Map results between solids, hydro-formed parts
• Handle symmetry
• Fill holes
• Batch process: Save and Replay
• Demo

Forming mesh Crash mesh


Thank you!

HyperForm 1-step solver based stamping results initialization
being used successfully at many OEMs and suppliers globally


Agenda

1. TeamCenter-HyperMesh Integration for Assembly – Shan Bala
2. Forming Results Initialization – Subir Roy
3. Multi-Domain in RADIOSS – Jean-Pierre Bobineau

Multi-Domain Approach to Parallel
Computations in Structure
Dynamics using RADIOSS
Innovation Intelligence® Jean-Pierre Bobineau
May 15, 2012


Multi-Domain approach to parallel computations

 Facts
• Time step discrepancy is a major bottleneck in explicit computations

• The smallest time step affects the performance of the whole model

 Method
The target of the Multi-Domain approach is to reduce the needed CPU time in order to
compute models with very different mesh sizes needing very different computation times.

The goal is to reduce the elapsed time without losing accuracy.
One of the challenges is to compute: cast, extruded and/or plastic parts (e.g. meshed with
Tetra10 solid elements) inside a “classical” full vehicle crash model in order to predict the
rupture mode of these specific detailed parts and their effects at global model scale.

The idea is to replace this global model by physically equivalent sub-domains, separating
parts with different minimal time step. Each sub domain is resolved as a distinct RADIOSS
model using its own time step, the force and momentum transfers between them being
calculated by a separate program handling stability constraints



Time step = dt_a Time step = dt_b Time step = dt_c



 Synchronization time



Radioss V10 – multiple starter input file (Classical approach)

• Each domain is built as a separate complete RADIOSS model using its own
complete input files.

• The RADIOSS runs are completely independent and do not communicate
directly with each other. Each one is using its own time step.

• The time step of each domain is arbitrary but to allow the best optimization
gain it should be significantly different from each other.

• All communication, data transfers, time step synchronization, equilibrium and
stability conditions on the domain frontiers are managed by the master program.

The main interest of this strategy is that each model is independent and may be replaced
by another one. This allows a modular approach for modeling, for example testing runs
with equivalent domains with different mesh size (multigrid approach). Reason why this
approach is still useful and is kept in Radioss V11.



Multi-Domain in
Version 10
P1_0000.rad P2_0000.rad P3_0000.rad

Radioss Radioss Radioss
starter 1 starter 2 starter 3

P1_0000.rst P2_0000.rst P3_0000.rst

input.dat

Rad2rad

Radioss Radioss Radioss
engine 1 engine 2 engine 3
dt_a dt_b dt_c



Radioss V11 – single starter input file format ( sub-domain approach)

• One drawback of the classical approach is that it implies additional work for
the user to manually build several independent input files.

• Furthermore it can become very long, difficult and a source of errors when the
small domains are extracted from large and complex models like full vehicle.

• The idea of the new approach in Radioss version 11 is to simplify the task of
the user by building sub-domains automatically :
• Only one starter input is now required including the entire model, like a
classical Radioss computation.
• The user only has to specify the parts of the model that he wants to place
in sub-domains.
• The starter automatically extracts the specified domains from the full
model and generates one restart file for each domain.



Multi-Domain in
Version 11

dt_a dt_b dt_c



New subdomain approach : definition of the subdomains

The subdomains are specified by parts

/SUBDOMAIN/subdomain_Id
subdomain_title
ID_part1 ID_part2 ... ID_partn

Where:
- ID_part are the Id of the parts that are in the sub-domain
- subdomain_Id is the domain identifier.
- subdomain_title is the sub-domain name

Radioss Engine input file
In Radioss V11 as in V10 an engine input file is required for each domain and in
order to activate the multi-domain coupling these files must contain the following
command line:

/RAD2RAD/ON



can be used to connect:
– Shell nodes to Shell nodes
– Shell nodes to Solid nodes
– Solid nodes to Solid nodes
– Beam nodes to Beam nodes


PSA head impact v. windshield wiper motor

Courtesy of PSA
 Model

Head impact :

PART1 : 13803 shell elements PART2 : 303648 shell elements
time step : 2.97e-7 sec [4.3%] [95.7%] time step : 8.00e-7 sec



 Results quality

Head acceleration vs time

---- MONO-DOMAIN
---- MULTI-DOMAIN

Note: In this case results quality is high also because there is small interaction between the 2 domains



 Head impact : scalability performances (elapsed time) :
1 CPU 16 CPUs (2 nodes) 32 CPUs (4 nodes) 64 CPus
Monodomain 63500 s 5420 s 3470 s 2790 s

Multidomain case 1 27260 s 2740 s 2100 s

Multidomain case 2 - // - 2584 s 1670 s 1390 s

Speed up case 1 2.33 2.00 1.65

Speed up case 2 2.30 2.09 2.08 2.01

Time step factor = (dt domain 1) / (dt domain2) = 2.6
Mono-domain = complete model with common time step
Speed UP = elapsed time Mono-domain / elapsed time Multi-domains

Case 1 : using synchronisation with total CPU number allocated to all processes
Case 2 : Case 1 + optimisation of MPI internal process communications (SPMD)


NEON Full Frontal with refined subframe

t = 00.00 ms t = 80.00 ms

Coarse Mesh in 1.Domain:
Timestep = 5.0e-007 [s]

Refined Mesh in 2,Domain:
Timestep = 1.0e-007 [s]


NEON Full Frontal with refined subframe

Application : Neon model
• Result
o The results are the same between the mono domain computed with a time step of 0.1ms
and the computation with sub-domain.

Mono domain : 71700 s
Multi domains : 35100 s

Rigid wall force

Speedup of ~2
compared to the
Mono-Domain with
same (small) Time step

Internal
energy



 Conclusions


Open for questions…

• Jean-Pierre Bobineau
• Phone: (248) 709 09 42
• E-Mail: jpb@altair.com

Crashworthiness Workshop - Altair HTC 2012

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Crashworthiness Workshop - Altair HTC 2012 (20)

More from Altair (20)

Recently uploaded (20)

Crashworthiness Workshop - Altair HTC 2012