Fraunhofer and Sidact Software

• FEMZIP – Compression of simulation results

• SDMZIP – Compression of sets of simulation results

• DIFFCRASH – Robust Design: Identification of areas in geometry

causing scatter of simulation results
CHALLENGE

• In order to improve engineering design …

• more simulations are performed
• larger, more detailied Models are used
» Large amounts of data are generated!
(several PetaByte per year)
The data has to be …
• analyzed
• exchanged
• archived
» Network connections and storage space can become
bottlenecks!
SOLUTION
Data Compression
• Two fundamentally different compression approaches:

• Lossless Data Compression

The original data can be restored identically from the
compressed data

• Lossy Data Compression

The original data cannot be restored identically from the
compressed data

» With lossy data compression schemes a much stronger

reduction can be achieved!
Data Compression
• Floating-point data cannot be efficiently compressed losslessly:
45.00
40.00
» A compression factor of only 1.2 is
obtained
35.00
30.00
» The solution is FEMZIP
25.00
20.00
15.00
10.00 Airflow simulation around a car:
5.00 6 variables, 43 million elements,
21 time steps
0.00
compressed (gzip)
FEMZIP
Advantages
Reduced Archive Size
Storage and backup capacities can hold more
simulation results

Shorter Data Transfer Times

Simulation results can be transfered significantly
faster

Quicker Data Loading

Compressed data can be loaded quicker into post
processors
Lossy Data Compression
• Quantization
Floating point data is rounded to a given precision and mapped into the integer domain.
• Prediction
A bijective transformation is performed to prepare those integers for encoding purposes
(reduce entropy).
• Encoding
Afterwards an entropy encoding method is used which removes redundant bits.

32/64-Bit float

13534,11289527977912437

011010110110
» FEMZIP achieves compression factors of about 10!
Parameter File
FEMZIP Standard Configuration File (mm, s, Tons, N)

Number of extra values per shell: 5

Number of extra values per solid: 6
Number of extra values per thick shell: 5

Node values: precision

coordinates : 0.10000000
velocities : 10.0000000
accelerations : 10000000.0
Shell values: precision
sigma : 1.00000000
epsilon : 0.00100000
bending_moment : 1000.00000
shear_resultant : 10.0000000
normal_resultant : 10.0000000
thickness : 0.00100000
internal_energy : 1.00000000
Thick shell values: precision
sigma : 1.00000000
Solid values: precision
sigma : 1.00000000
1D-element values: precision
axial_force : 10.0000000
s_shear_resultant : 10.0000000
t_shear_resultant : 10.0000000
s_bending_moment : 1000.00000
t_bending_moment : 1000.00000
torsional_resultant : 1000.00000
Lossy Data Compression
• The visual
appearance of the
original and the
compressed results
is shown. While a
compression factor
of 8.8 was achieved
no difference is
noticeable.

Source: topcrunch.org
SDMZIP:
Compressing sets of simulation results
Compression of simulation results

Standard solution for industry applications

• Lossy compression of a single simulation result

• Versions for several data formats
• Continuing improvement
• Integrated decompression in several postprocessors
• Compression rates of 15 to 30
Example: Chevrolet Silverado
simulated using LS-DYNA

Timesteps 38/152
Nodes 942,749
Elements 929,181
Parts 679
Variables 14

The displayed model has been developed by the National Crash Analysis Center (NCAC) of the George
Washington University under a contract with the FHWA and NHTSA of the US DOT.
The crash simulation was done by SIDACT with help of Fraunhofer Institut SCAI, Sankt Augustin Germany,
and are their respective property
Compression results
Single simulation
• Original size: 1,525.1 MB
• FEMZIP size: 55.9 MB
• Compression rate: 27.28

Additional simulation
• Original size : 1,525.1 MB
• FEMZIP size : 56.25 MB
• Combined compression rate: 27.19
Comparison: Two simulations,
first and last timestep

• Are the simulations disparate?

• Can the commonalities be exploited?

The displayed model has been developed by the National Crash Analysis Center (NCAC) of the George
Washington University under a contract with the FHWA and NHTSA of the US DOT.
The crash simulation was done by SIDACT with help of Fraunhofer Institut SCAI, Sankt Augustin Germany,
and are their respective property
Recently developed:
New software for compression of sets of
simulation results

• Lossy compression
• Information is processed part-based
• Modular storage concept
• Commonalities of similar simulations is
aggregated
Compressing sets of
simulation results
Simulation 1
Sim_1.efz
Simulation 2
Sim_2.efz

… …

Sim_S.efz
Simulation S
Database.fdb
Compression results
30 Simulations, with 38 timesteps:
• Original size: 45,752.58 MB
• FEMZIP size : 1,684.54 MB
• SDMZIP size : 379.23 MB
 Database : 110.17 MB
 *.efz ca. : 8.98 MB
Incremental compression

Simulation 1 Simulation 2 … Simulation S

Sim_1.efz Sim_2.efz … Sim_S.efz

Database.fd
Update_1. Update_K.
b …
ufdb ufdb
Database development
for 30 simulations
Size of databases in MB for incremental
compression.

Size of *.efz ca. : 19.01 MB

Results vary depending on:

• Similarity of simulations
• Time resolution
Robustness analysis –
process to analyze multiple simulation runs
Process

FE Model Simulation Sim.Result

Verification

Car Crash Test Result

"The model has been developed by The National Crash Analysis Center (NCAC) of The George Washington
University under a contract with the FHWA and NHTSA of the US DOT“
http://www.ncac.gwu.edu/vml/models.html
Process

FE Model Simulation Sim.Result

Emulate m.-
Verification
tolerances

Car Crash Test Result

Manufacturing
tolerances
"The model has been developed by The National Crash Analysis Center (NCAC) of The George Washington
University under a contract with the FHWA and NHTSA of the US DOT“
http://www.ncac.gwu.edu/vml/models.html
Process

FE Model Simulation Sim.Result

Emulate m.-
tolerances

"The model has been developed by The National Crash Analysis Center (NCAC) of The George Washington
University under a contract with the FHWA and NHTSA of the US DOT“
http://www.ncac.gwu.edu/vml/models.html
 Process

FE Model Simulation Sim.Result

Simulation n

Emulate m.-tolerances

"The model has been developed by The National Crash Analysis Center (NCAC) of The George Washington
University under a contract with the FHWA and NHTSA of the US DOT“
http://www.ncac.gwu.edu/vml/models.html
Process

FE Model Simulation Sim.Result

Emulate m.- Simulation n

tolerances
Simulation n
Methods
Variation Highlighting

Impact Quantification Extreme Runs

Simulation n

Root Cause
Dominating Effects
Identification
Methods
Variation
Highlighting

Impact Quantification Extreme Runs

Simulation n

Root Cause
Dominating Effects
Identification
Methods
Variation Highlighting

Impact Quantification Extreme Runs

Simulation i

Simulation j
Simulation n

Root cause
Dominating Effects
identification
Methods
Variation Highlighting

Impact Quantification Extreme Runs

Simulation n

Root Cause Dominating

Identification Effects
Methods
Variation Highlighting

Impact Quantification Extreme Runs

Simulation i

Simulation n

Root Cause Dominating Effects

Identification
Methods
Variation Highlighting

Impact Extreme Runs

Quantification

Simulation n

Root Cause
Dominating Effects
Identification
Example: Toyota Yaris
Toyota Yaris
Model Toyota Yaris
Year 2010
Number of Parts 917
Finite-Elements 1,514,068

"The model has been developed by The National Crash Analysis Center (NCAC) of The George Washington
University under a contract with the FHWA and NHTSA of the US DOT“
http://www.ncac.gwu.edu/vml/models.html
Example: Toyota Yaris

Maximum variation of node position in [mm]

© SIDACT GmbH
Example: Chrysler Neon
Chrysler Neon
Model Chevrolet
Silverado
Year 2006
Number of Parts 712
Finite-Elements 1062140

"The model has been developed by The National Crash Analysis Center (NCAC) of The George Washington
University under a contract with the FHWA and NHTSA of the US DOT“
http://www.ncac.gwu.edu/vml/models.html
Example: Chrysler Neon

Maximum variation of node position in [mm]

© SIDACT GmbH
Summary
• Production tolerances can have a big impact on
simulation results

• Easy emulation of thickness variation triggers model

instabilities

• More robust design

© SIDACT GmbH