Dubois Book

Crashworthiness Engineering
with LS-DYNA

P.A. Du Bois
H.E.N.V.
2000

Table of Contents
Crashworthiness Engineering with LS-DYNA iii

Contents

I. Introduction and Applications overview.................................................................. I.1

1. Timestep Control .......................................................................................................1.1

2. Aspects of Shell Element Technology.......................................................................2.1
2.1 Introduction........................................................................................................2.2
2.2 Common Features of Shell Elements in LS-DYNA..........................................2.3
2.3 Implemented Shell Theories: Determination of Fiber Directions ...................2.11
2.4 In-Plane Element Integration...........................................................................2.26
2.5 Element Integration Through-the-Thickness...................................................2.44
2.6 Large Rotation Stress Update ..........................................................................2.60
2.7 Triangular Elements.........................................................................................2.72
2.8 Hourglass Prevention.......................................................................................2.74
2.9 Modeling for Large Displacements .................................................................2.86
2.10 Shell Element Classification............................................................................2.88

3. Numerical Treatment for Contact Problems ..........................................................3.1
3.1 One-Sided Contacts for Impact Analysis: Master Slave Contacts ....................3.5
3.2 Contact Algorithms for Automotive Crash: Single Surface Contacts .............3.19
3.3 Connections of Thin Structural Sheets ............................................................3.45

4. Material Modeling .....................................................................................................4.1
4.1 Classification of Materials for Numerical Simulation.......................................4.2
4.2 General Notions .................................................................................................4.3
4.2.1 Measures of Stress in Solids .......................................................................4.3
4.2.2 Measures of Strain ......................................................................................4.9
4.2.3 Isotropic Hypoelastic Material..................................................................4.20
4.2.4 Elasto-Plasticity for Metals.......................................................................4.22
4.2.5 Elasto-Plasticity with Strain Hardening....................................................4.29
4.3 Material Laws for 2D Elements in LS-DYNA................................................4.35
4.3.1 Simulation of Mild Steel in LS-DYNA....................................................4.35
4.3.2 Simulation of Cast Iron, Al, and Mg.........................................................4.65
4.3.3 Simulation of Woodstock and Lignotok...................................................4.65
4.3.4 Simulation of Thermoplastics in LS-DYNA............................................4.67
4.4 Material Models in LS-DYNA for 3D Elements.............................................4.79
4.4.1 Material Models for Recoverable Foams..................................................4.79
4.4.2 Material Models for Brick Elements.......................................................4.116
Table of Contents
iv Crashworthiness Engineering with LS-DYNA
4.4.3 Development of a Material User-Subroutine..........................................4.139

5. Occupant and Restraint Systems Simulations ........................................................5.1
5.1 Numerical Models of Occupant Dummies ........................................................5.2
5.2 Head Impact Simulations for MVSS201.........................................................5.14
5.3 Numerical Simulation of Airbag Deployment.................................................5.17

6. Component Models in LS-DYNA.............................................................................6.1

7. Quality Assurance of Numerical Models .................................................................7.1
7.1 QA of Numerical Models Before Analysis .......................................................7.2
7.2 Post-Processing: QA After Analysis .................................................................7.4

8. Modeling of Deformable Barriers ............................................................................8.1

9. Optimization of Structural Crashworthiness..........................................................9.1

10. Numerical Treatment of Shockwaves in Solids...................................................10.1
10.1 Hydrodynamic Materials ...............................................................................10.7
10.2 Elasto-Plastic Waves ...................................................................................10.19
10.3 Rate and Temperature Dependency.............................................................10.28

11. Quasistatic Simulations with LS-DYNA..............................................................11.1

Introduction and Applications
Crashworthiness Engineering with LS-DYNA I.1
Our Experience: 1987-1999

Automotive customers:

Ford Motor Cie FoB
FoG
Dunton
Merkenich
GMI Opel
SAAB
Ruesselsheim
Trollhattan
Mercedes-Benz Sindelfingen
Fiat Research Center Orbassano
Volvo Car Corporation Goeteborg
PSA La Garenne
Renault Technocentre Guyancourt
GM Technical Center Warren
Ford Motor Cie Dearborn
Chrysler Technical Center Auburn Hills
Nissan Technical Center Atsugi
Honda Tochigi R&D
Asaka R&D
Utsunomya
Asaka
Hyundai Motor Cie Namyang
Kia Motors Kwangmyong

I.2 Crashworthiness Engineering with LS-DYNA
Joint research projects:

FAT Side impact dummies Opel
Ford/Germany
Mercedes-Benz
VW
Audi
Porsche
BMW
others
FAT Foam materials Opel
Mercedes-Benz
VW
Audi
Porsche
others
EUCAR Pedestrian
impactors
Opel
VW
Volvo
Ford
Rover
Jaguar

Industrial Applications of Numerical Simulation
for Crashworthiness Engineering

A Short Survey: 1985-2000

History of Numerical Simulations for Crashworthiness

Explicit FE-codes were developed in the 60s and 70s at the DoE Labs in the USA

All essential algorithms were available: explicit integration, shell element, contact, radial
return

First full vehicle car crash models built and analyzed in the mid 80s, industrial application
rendered possible by the introduction of supercomputers (Cray)

In Germany, a research project was sponsored by FAT to investigate the feasibility of car
crash simulations, 2 models were selected: VW-Polo and BMW-300

Rapid development in the next 10 years, today FE-crash simulation is a fully integrated tool
in vehicle design, no competitive development is possible without intensive CAE support

Impact of CAE in Crashworthiness Engineering

Evolution of CAE in crashworthiness:

1985: pure research activity
1 legal safety criterion must be met
about 100 prototypes needed per newly developed carline
1995: CAE part of design process
4 legal criteria + many in-house criteria must be met
design cycles shortened
average number of prototypes per newly developed carline
still around 100

Potential of CAE in crashworthiness:

Avoid building prototypes that fail by virtually optimizing the structure using trend
predictions
Quickly find the structural fix if problems occur in a test
Reduce the number of development prototypes
Allow for more verification & certification testing
Certification needs to be done by testing hardware because of :
Legal requirements
Limited capacity of numerical models for absolute predictions

Example of CAE Implementation

Dedicated FE-models and engineers for all important load cases at Mercedes-Benz (structural
crashworthiness):

carline frontal side rear
C
E
S
A
SLK
M
V

Additional group for occupant simulation

Over 30 vehicle models of over 200000 finite elements each, updated to the actual design
state on a daily basis to monitor crash performance

Complementary roles of testing and simulation

Cost and Gain of CAE in Crashworthiness Engineering:
Example of Ford-Mondeo

Potential savings in testing and prototype hardware:

About 150 prototypes crashed in Europe and USA

Development cycle of 5 years: 30 prototypes/year

Average prototype cost at least $250,000


By conservative estimates, 30% of prototypes can be saved through the use of simulations

Roughly 10 prototypes or $2.5M per year and per carline

If we need about 20 simulations to eliminate a single protoype, then 200 simulations are
needed per year and per carline

At a rate of 50 CPU hours per simulation, a total of 10000 supercomputer CPU hours are
required per carline and per year

This is equivalent to occupying a single processor year round or a 2 processor machine half
time

The yearly cost is certainly covered by savings in hardware

In any case, CAE is the only way since numerous regulations create a demand beyond the
capacity of the safety testing labs

How Could Industrial CAE Grow So Fast?

Need and opportunity

Need for simulations was regulation driven

Opportunity to deliver results was created by rapid development in hardware and software
since 1985, and by buildup of hardware resources in the automotive companies

Software development was inevitably performance driven

The number of loadcases has increased dramatically:

Approximate evolution of computer time needed to run 100 milliseconds of crash simulation
on a very small vehicle model (8000 elements)

1983 VAX-750
Single scalar processor
2000 hours
1984 Cray-1
Single scalar processing
100 hours
1985 Cray-1
Single vectorized processing
25 hours
1986 Cray-1
Single vectorized processing,
vectorized gather-scatter
10-12 hours
1998 Cray-T90
Single processor
30 minutes

Supercomputing made the difference between research and industrial application

Car companies have built up considerable computing power, motivated for over 60% by
crash analysis:

Fords supercomputing resources grew by a factor of 500 in about 10 years (1988-1998)

Are We Losing the Battle?

The need for higher reliability in the simulation work resulted in a continuous increase of
size and complexity of the numerical models

Consequently the need for CPU resources has increased dramatically

Review from Mercedes-Benz:

year Size
(elements)
CPU-time
(hours)
Cray
1988 8-10000 5-10 XMP
1990 15-20000 10-20 YMP
1992 30-40000 20-30 YMP
1994 60-80000 30-40 C90
1998 160-180000 60-80 T90
2000 400-500000 120-160 SX-4/5

Overnight runs are still possible but require parallel calculations

The potential of parallel computing is remarkable, time needed to run 120 milliseconds of
crash analysis on a 500000 element vehicle model:

8 processor
Intel
MPP
8 processor
Origin
MPP
4 processor
SX-4
SMP 24 hours
4 processor COMPAQ SMP 48 hours
16 processor
COMPAQ
MPP 12 hours
32 processor
COMPAQ
MPP 6 hours

Crashworthiness Engineering:
State-of-the-Art in Vehicle Modeling

The Problem of Reliability

Difference between test and calculation results have many different causes; such as in the
material description of body panels:

Theory and numerical implementation (radial return) of plasticity for mild steel and other
metals

Hardening and rate dependency parameters for mild steel and other metals

Initial stresses are difficult to account for when using explicit integration

Material Properties for Steel Sheet

Variability is introduced by:
Initial anisotropy
Forming process (stamping, hydroforming...)
Thermal treatment (?)

Material properties exhibit:
Non-homogeneous thickness
Initial stresses
Initial plastic strain, thus non-homogeneous yield stress

Example:

In LS-DYNA v960, these effects can partially be taken into account:

Mesh-independent projection of the results of forming analysis upon the corresponding panel
in the crashworthiness model

A first possibility is to consider the influence of the stamping operation only:

Stamping simulation
explicit adaptive mesh
1 panel DYNAIN file
input & mesh
deformed geometry
t/shell
stress/IP
plastic strain/IP
Crashworthiness simulation
explicit
Many panels (PART=PID)
Coarser mesh
*INCLUDE_STAMPED_PART
PID
Reference node positions
DYNAIN
Project and plastic strain
Remarks:

Simple manipulation if the stamping simulation results are available

Quality of projection will decrease if differences in mesh density increase

No stresses are projected, so no problems with initial equilibrium in the crash analysis

Geometry after stamping may deviate from final geometry due to springback

No further changes necessary to crashworthiness input deck

A second possibility is to consider the influence of the stamping and the springback:

*INCLUDE_STAMPED_PART
PID
Reference node positions
DYNAIN
Project t, plastic strain, stress & backstress
Crashworthiness simulation
explicit
Many panels (PART=PID)
Coarser mesh
DYNAIN file
input & mesh
deformed geometry
t/shell
stress/IP
plastic strain/IP

Sprinback simulation
implicit
very fine mesh
1 panel
F1 shells
NIP/t > 7
Anisotropic ML
Remarks:

Simple manipulation if the springback simulation results are available

Quality of projection will decrease if differences in mesh density increase

Stresses are projected, so there is a need to ensure initial equilibrium in the crash analysis

Geometry after springback should be final geometry in the crashworthiness model

Crashworthiness input deck should become more similar to the springback deck:

crash springback
NIP/shell 1 4
NIP/t 2-5 >7
ML Von Mises

Isotropic hardening
Hill (steel)
Barlat (alu.)

Isotropic+
kinematic hardening
Mesh size 5-10 mm ?

Considering the effects of manufacturing on the material properties may become a necessity
if fat hardening steels (DP500, DP600, TRIP...) are considered

yield TRIP
Mild steel
Plastic
strain

Difference between test and calculation results have many different causes, some are easily
solved by mesh refinement:

Numerical methods:
Use of underintegrated shell elements
And penalty based contact algorithms

Too coarse meshes allow only low curvature buckling modes

Required Mesh Density for the Simulation of Buckling Problems

Convergence studies were performed very early on

A good reference is:

Mats Larsson, 1989
An assessment of four different shell elements implemented in DYNA3D, a FEM-code used for
transient nonlinear dynamic analysis of three-dimensional structures.

A convergence study was performed on the axial buckling of a straight rail with typical hat
section:

M
V0
Flanges and spotwelds were neglected leading to a quarter-symmetric model

In fact a symmetric geometry does not guarantee symmetric buckling
The buckling of a quadratic beam was examined using 6 to 40 shell elements per side



Mesh convergence can be checked by refining the mesh and comparing results or by judging
the smoothness of the deformed mesh

The study shows 16 elements per side to be adequate in describing the physical behavior with
sufficient accuracy; 12 elements per side gives an approximate solution

The Problem of Reliability (Continued)

Difference between test and calculation results have many different causes:

Functioning (initial stresses) and rupture of connections:
Bolts
Screws
Spotwelds
Glue
etc...

Behavior of non-steel components

Realistic friction models are missing

Different Ways of Modeling Spotwelds

For crashworthiness applications, we always model each individual flange with real flange
thickness, each spotweld is taken into account as an element

Node-to-node connections:
- Nodal rigid body constraints
- Welded node sets
- Spotweld elements
- Discrete beams

Element-to-element connections:
- Beam element type 9, material type 100
- Sliding interface type 7 (shell_edge_tied_to shell_surface)

Remarks:

Individual weld modeling is necessary to allow simulation of spreading in the flanges before
buckling

In a node-to-node connection, the weld must be orthogonal to the flanges, this requires
corresponding meshes in both flanges

A skewed weld will weaken the structure by rigid body rotation:

Spotweld beams have a typical height of 1.mm (sum of half sheet thicknesses) and a diameter of
6.mm (corresponding to weld nugget), so they are disks and not beams:

1

This requires a special-purpose element (type 9)

With steel properties the timestep is prohibitive:

1
0.2
5000 /
l mm
t s
c mm ms
= = =

Material type 100 allows enforcement of the timestep by mass scaling, added mass is
reported in d3hsp and should be checked

6
1
Comparison of both models:

Node-to-node Element-to-element
Spotweld element has no spatial dimension Spotweld element has a finite size
Allows spreading of flanges Allows spreading of flanges
Transmits forces and moments between
flanges
Transmits forces and moments between
flanges
Rotational stiffness is mostly infinite Material rotational stiffness
Failure possible Failure possible
No free arrangement of spotwelds, meshes
on flanges must match
Free arrangement of spotwelds, flanges can
be meshed independently
May promote hourglassing Does not promote hourglassing
For certain options will not influence the
timestep
Beam elements usually influence the
timestep
2 elements in the flange allow central
positioning of the weld
3 elements in the flange needed to position
the weld centrally
Element size influence? Element size should be about equal to the
weld nugget diameter

Possible meshless weld definitions:

Non-Symmetric
Force on 4 nodes
Symmetric
Force on 2 nodes

Beware of bad weld definitions:

This can be caused by inaccurate CAD data and automatic weld generation

More information is in:

A.K. Volz
Spot weld modeling for crash computations, requirements and new functionalities in LS-DYNA,
version 940
5
th
international LS-DYNA Users Conference,
Southfield, 1998h

Improved spotweld modeling:

Contact type s_7 to consider torsional stiffness in the weld

Modeling by brick elements (v960)

Symmetric
Force on 4 nodes
5. 7 mm

Modeling of Bolts and Rubber Bushings

Subframe and engine mounts should no longer be modeled by single spring or joint elements
since this can lead to considerable error in relative rotations between the connected parts

Concentric cylinders with contact or brick elements modeling the rubber are far more
accurate

Consider a typical subframe mount with rubber bushing

Old way of modeling could be with a discrete beam element and 2 rigid bodies:

Rotational behavior from test could not be matched

Force-displacement and Moment-angle characteristics in all 6 DOF had to be guessed, initial
slopes could be obtained from NVH studies:

A good model of the rubber bushing is more important for low velocity impact

Displacement in mount characteristic involves closing the air gap and some compression in
the rubber, cannot easily be used to determine rubber material properties

f
d
This can affect the accelerations at B-pillar level:

Engine mounts as 4 springs
Engine mounts modeled
The Problem of Reliability (Continued)

Difference between test and calculation results have many different causes:

Behavior of non-steel components

Realistic friction models are missing

Mass Distribution in Vehicle Models

Assume a typical small vehicle:

Total vehicle mass 1200.kg
Car body 200.kg 100000 elts
100000 nodes
2 gram/node
powertrain 200.kg meshed
Chassis & wheels 200.kg meshed
components 600.kg Added mass
For 600 masspoints:
1.kg/node

The non-structural components are modeled with a precision that is 2 orders of magnitude lower
then the car body itself.

Additional considerations are excentricity (rotational inertia) of the components and the stiffness
of the connection with the car body.

It is important to carefully consider the components (other then powertrain) that move relative to
the car body:

Exhaust system
Fuel tank
Seats
Dummies
etc...

Consider a frontal crash simulation performed to obtain the acceleration signal in the airbag
sensor:

Airbag sensor

Passenger compartment
Car body mass = 200 kg
Component mass = 600 kg

The passenger compartment (floorpan and tunnel in particular) remains elastic during the
first 20 milliseconds of the impact

During the first 20 milliseconds the impact force is a function of the plastic deformation in
the front assembly, the deceleration is:

( )
800
( )
200 800
f
simulation
f
x
f m
real life
active mass
< <
&&

Fixing components to the car body will lead to an underestimation of acceleration peaks early in
the analysis, this can be critical for the investigation of airbag sensor response but usually not for
determining occupant kinematics.

Stiffness of certain components (radiator...) may also play a part.


Major stumbling block for predictive simulations today is in the material modeling:

Structural use of non-steel materials:

Lack of test data

Lack of suitable material models
(plastics, foams...)

Discontinuous cell structures must be approximated as continuous materials
(foam, honeycomb...)

Inhomogeneous composites must be approximated as continuous materials
(woodstock, lignotok...)

Lack of models to predict (brittle) rupture (aluminum, magnesium, fiber
reinforced composites) and crack propagation, this is a major problem in
castings (engine mounts)

Trends in Crashworthiness Simulations

The decision to build a prototype is often taken on the basis of a simulation result

Reliability then takes priority over runtime

Reliability is improved by:

** Finer meshes
** Better quality meshes
** Improved numerical algorithms

Size of Numerical Models

Full vehicle state-of-the-art model size grew from 10000 to 50000 elements in about 5 years
and to 150000 elements in the next 5 years

The desire for more reliability and accuracy is due to:

**Finer meshes that allow capturing higher curvature buckling modes
**More components that can be modeled in detail, including their connection to the car body
**Hourglass and contact energy problems that are avoided by increasing mesh density

Uniform meshes require less assumptions about the response (buckling mode) of the
structure

The tendency is towards a single homogeneous model for all 3 major load cases (front, rear
& side)

Less restrictions on model size has advantages from a mesh generation point of view:
automatic meshing and easy integration of component models because of mesh compatibility

A car body contains 20-25 square meters of metal sheet, consequently for a uniform mesh:

Mesh size Number of
Shell elements

10*10 200000
Weldnugget size 5*5 800000
l=t
(ductile failure)
1*1 20000000

The ultimate future for car body crashworthiness studies may see:
Thick shell elements
Quadratic interpolation (continuous surfaces)
Kinematic constraint contacts on intrados and extrados

Quality of Numerical Models

Car body geometry (CAD) is smoothed to allow an element size of about 5mm, this way
reasonable timestep values are obtained initially without mass scaling

Mesh density must allow smooth representation of the deformed geometry

Mesh density must allow for sufficient resolution in terms of integration points: static stress
analysis mesh is usually too coarse

Limit number of triangular elements

Regular meshes in order to represent the wave propagation problem with minimal dispersion

Limit warping in BT shell and avoid zones meshed predominantly with warped elements

Mesh every sheet in its own neutral plane, do not distort flanges and model spotwelds with
special purpose elements

CAD Smoothing example:

Wavefront dispersion in irregular mesh:

FE
CAD
Avoid initial penetrations at all cost: induced stresses may exceed the yield stress

Example: Penetration caused by non-homogeneous mesh, distance between slave node and
master segment is less than the sum of the half sheet thicknesses:

In thin steel sheets, initial penetrations of 0.1mm or more can cause stresses that exceed the yield
stress: a plastic hinge will result.

Contours of plastic strain should be checked after 0.5 of 1 millisecond, before the structure
impacts.

Automatic offset of slave nodes will not be failsafe if more then 2 sheets are involved
Contact force
time
Contact force
Avoid mesh tangling at all cost: instabilities may be generated

If tangling exceeds the contact thickness, no warning will be printed to the d3hsp file

Non-physical connections are generated, entire panel assemblies could be suspended upon very
few nodes and high nodal contact forces will occur

Tangling can occur during the simulation if edge-to-edge penetrations are not prevented

Impact of a slave node on the rear side of a master segment is the most frequent cause of
numerical problems due to extremely high nodal contact forces

Causes can be:

Tangled mesh (see above)

Edge-to-edge penetrations (see next)

Deep penetrations through the midplane of the master segment with release:



Conclusion: the differences with static stress analysis for fatigue evaluations:

Statics model Crashworthiness model

Geometrical detail Smoothed CAD
Undeformed geometry Deformed geometry
Fully integrated shells Underintegrated shells
Irregular mesh
(stress peaks)
Regular mesh
(wave propagation)
Welds as common nodes Welds as elements
Higher order shells Bilinear shells
Penetrated meshes No initial penetrations nor intersections nor
perforations

Major Advances in Software

Element Technology:

The traditional BT element has the merit of computational efficiency and robustness,
deficiencies are:

Underintegration (possible occurrence of zero-energy or hourglass modes)
Corotational formulation (limited to small shear deformations)
Plate element formulation: no coupling of membrane and bending strains

Affordable shell elements are now available that correct several deficiencies of the BT element

Example of a side-impact study on an empty car body:



Contact algorithms:

Classical contact algorithms are node-to-segment and do not check for edge penetrations.
Contact type 26 allows treatment of contact between beams and free shell element edges:

Currently, only free edges are treated.

Material Models:

The number of materials used in cars seems infinite (rubbers, foams, plastics, fiber reinforced
plastics...)

Much more validation testing is needed

Development of material models becomes more and more the responsibility of the
application analyst

Development of user subroutines becomes more frequent

Conclusions

Between 1985 and 2000:

Number of test scenarios rose from 1 to nearly 20

Design cycles have shortened

Prototypes remain expensive

Testing capacity is limited

Cpu-time decreases in price

Thus:

CAE needs to deliver fast and reliable results

To achieve this:

Model size increases

Model quality increases

Approximate solutions in software become less acceptable

Databases of standardized component models are built

Large scale simulations are performed by automotive companies and suppliers

Need for computer resources increases, MPP solutions become more important

Timestep Control in LS-DYNA
Crashworthiness Engineering with LS-DYNA 1.1

1. Timestep Control in LS-DYNA

1.2 Crashworthiness Engineering with LS-DYNA
Acoustic Wave Propagation

Consider a 1D linear elastic (small displacement) stress wave:

In the small deformation case, the wavespeed c can be considered constant and consequently
the wavefront does not deform as it propagates: this is an acoustic signal

E
c
=

The numerical model of this simple 1D impact can consist of a high number of truss
elements, we make them of equal length and apply a force load:

In the explicit integration, the numerical stress wave will always propagate one element per
timestep, thus:
n
l
c
t
=

We necessarily have optimum accuracy for:

n
l
c c t
c
= =

p
t
p
t
1
A stable solution (with some numerical dispersion) will be guaranteed as long as:

l
t
c

Conditional Stability

The timestep of an explicit analysis is determined as the minimum stable timestep in any
deformable finite element in the mesh

In general this is determined by the so-called CFL-condition (Courant-Friedrichs-Lewy) that
determines the stable timestep in an element as characteristic length divided by the acoustic
wavespeed:
c
l
t
c
=

The CFL condition thus requires the numerical timestep to be smaller than the time needed
by the physical wave to cross the element.

The physical stresswave propagates with the speed of sound c

The numerical stresswave propagates one element per timestep

Consider a one-dimensional pressure wave propagation in a rod, using a constant spatial
increment:

t
t(n)
t(n-1)
1 t
x c

x(n-1)
x(n)
x(n+1)
x
A stability proof of explicit integration methods that mathematically leads to the CFL
condition is only possible for linear problems

The acoustic wavespeed for a 1-dimensional (truss) element is trivially computed:

For a truss with a unit section Newtons second law leads to the familiar wave equation:

2
2
0
0
0
2 2
2 2
0 0
xx xx
xx
xx
x
dx dx dx
t x x
u x x
u
x
x u
E E
t x x

=

=

= =

2 2
2 2
x E x
t x
E
c

=

=

The previous derivation assumes a free transversal deformation (Poisson effect) in both
directions for the truss element

A shell element can be viewed as a number of truss elements placed next to each other, this
geometry will confine the Poisson effect in the plane of the shell during longitudinal
compression, thus increasing the longitudinal stiffness as well as the wavespeed which
becomes:

( )
2
1
E
c
v
=

dx
x

The critical timestep for shells is determined by the wave propagation speed for membrane
deformations (rotational inertia is adjusted in order to sufficiently lower the bending
frequencies).

If the material Poisson coefficient is 0.3, the critical timestep in a shell will be 0.954 times
the critical timestep in a truss with the same characteristic length.

In brick elements, one can consider the Poisson effect to be confined in both transversal
directions during longitudinal compression, consequently the membrane stiffness and the
acoustic wavespeed will increase more dramatically:

The acoustic wavespeed in 3D media becomes:

( )
( )( )
4
1
3
1 1 2
K G
E v
c
v v
+

= =
+

For a material with a Poisson coefficient of 0.3, a brick element will show a critical timestep
that is 0.862 times the critical timestep in a rod with the same characteristic length
Clearly the critical timestep for a brick element is zero if the material is incompressible

( )
0.5 0
3 1 2
E
v K t
v
= = =

The characteristic length of trusses and Hughes-Liu beams is estimated as the element length:

c
l L =

The characteristic length of brick elements is estimated as element volume over the maximal
side area for 8-node hexagonals:

( )
1 2 3 4 5 6
max , , , , ,
c
V
l
A A A A A A
=

The characteristic length of a shell element is estimated by default as the area divided by the
maximal side length:

( )
( )
( )
( )
1 2 3 4
1 2 3
4
max , , ,
2
3
max , ,
c
c
A
l node
L L L L
A
l node
L L L
=
=

Alternatively the characteristic length can be estimated in a more conservative way as area
divided by maximal diagonal by setting ISDO=1:
(CONTROL_TIMESTEP)
( )
( )
( )
( )
1 2
1 2
4
max ,
2
3
max ,
c
c
A
l node
D D
A
l node
D D
=
=

Alternatively the characteristic length can be estimated to optimize performance as the
maximum of area divided by maximal sidelength and minimum sidelength, by setting
ISDO=2:

( )
( ) ( )
( )
( ) ( )
1 2 3 4
1 2 3 4
1 2 3
1 2 3
max , min , , , 4
max , , ,
2
max , min , , 3
max , ,
c
c
A
l L L L L node
L L L L
A
l L L L node
L L L
(
=
(

(
=
(

This will give a large timestep, in particular for shallow triangles, but can lead to instabilities
since the characteristic length should be interpreted as the shortest path through the element
(from any node to the opposing side) and in shallow triangles this does not correspond to the
minimum sidelength:

Consequently it is dangerous to increase the timestep for badly conditioned triangular
elements by setting ISDO=2

The altitude of the triangle should be the basis of the calculation

If the shortest side is used, the results of the simulation must be carefully checked for local
instabilities

If this works well or not, is dependent upon the dimensions of the neighboring elements

ISDO=2 option OK:

N1
L1
N2
L2
lc
L3
N3
ISDO=2 option not OK:

Spring elements require a slightly different treatment since they can be defined on nodes with
identical coordinates

The equivalent of the CFL condition for discretized continua formulates the critical timestep
in function of the spring stiffness and the nodal masses

For a spring with longitudinal stiffness k connecting 2 masses m1 and m2 we obtain:

( )
1 2
1 2
4m m
t
k m m
=
+

In the code the nodal spring masses are approximated as half the actual nodal masses

For a spring with longitudinal stiffness k and two equal nodal masses m we obtain:

2m
t
k
=

This can be easily seen to be the equivalent of the CFL condition applied to the equivalent
truss element if the spring length is non-zero:

Indeed:

2
2
2
c
Al
m
EA
k
l
m l l
t
k E c
=
=
= = =

m
k
m
A
l
The zero-length spring element has a finite timestep because it also has a finite stiffness, this is
not the case for a (physical) truss

Stiffness/length and timestep/length diagrams for trusses and constant stiffness springs:

l=l0
k
k0
spring
truss
l
l=l0
truss
dt
dt0
spring
l
The acoustic wavespeed for the 2 most common automotive metals is:
Steel: 5240m/s
Aluminum: 5328m/s

In both materials, an element characteristic length of 5.mm will lead to a timestep of 1
microsecond, frequently set as a standard in the industry

Since a reasonable timestep leads to a minimum element sidelength of about 5.mm, car body
geometries cannot usually be respected completely by the FE mesh:

This is the most serious limitation of vehicle models today: the required minimum element
size often prevents geometrically detailed and/or sufficiently dense meshing

LS-DYNA allows improving the timestep for any mesh by adding mass to the critical
elements, this is done through mass scaling by specifying a minimum timestep for any of two
parameters on the CONTROL_TIMESTEP card:

MS1ST: mass scaling during first cycle only
DT2MS: mass scaling repeated every timestep

The amount of added mass is monitored, (on the d3hsp file), however this should be avoided
since due to the nature of car body geometries structural errors can be introduced into the
model that will change the dynamic behavior of the structure

1 or 2mm ridge cannot be
represented by 1 element
* Global mass error:

M/M =
total mass added by scaling / initial model mass

Mass scaling procedure is performed for every element individually:

2
min
min
min
?
c
c
t l
E
yes ok
t t
t
no t l
t E

| |
= =
|
\ .

Consider the example of a simplified wheelhouse with a 1.mm ridge that was modeled using
shell elements: (the corresponding unscaled timestep would be 0.2e-6 seconds)

In order to achieve a standard 1e-6 seconds timestep we would mass scale as follows:

25
1. 500 1. 24 12000 96
0.096
0.00006
1500.
96
0.09
1200
s
m gram
m
M
m
M

=
= =

Thus small global mass changes may be caused by important and structural local mass
changes

The mass scaling option can however be extremely useful to prevent the timestep from
dropping during the analysis (option DT2MS)

Indeed a badly deforming shell element used to model a mild steel sheet usually corresponds
to a local instability or an hourglass mode, this means the simulation is locally already highly
inaccurate and the element would bring the simulation to a virtual standstill if nothing is done

During the crash of a vehicle made primarily out of thin mild steel sheets, no finite
membrane compression strains occur, rather the metal folds causing high curvatures but
small deformations and almost no change in (characteristic) length

Crashworthiness analysis is thus a large rotation but small deformation problem

Undeformed thin sheet:

dt
dtmin
no scaling masses
initial mass scaling
t
Deformed configuration: (small compressive strains)

If the mesh is capable of reproducing this behavior, then the analysis timestep will
necessarily be constant; this will be the case if the mesh is fine enough to allow a smooth
representation of the deformed vehicle geometry (no high curvatures within a single element)

A reliable crashworthiness analysis can be recognized by either of the following 2
statements:

The mesh is capable to smoothly represent the deformed structure

The timestep in the shell elements that represent the sheet metal parts is constant without
mass scaling

To account for nonlinear effects upon structural frequencies the CFL determined timestep is
multiplied by a safety factor (default value 0.9) in LS-DYNA, the default can be changed by
setting the TSSFAC variable on the CONTROL_TIMESTEP card

Particularly if brick elements are used with highly nonlinear material laws, TSSFAC=0.66
often improves the stability of the simulation

In fact, for shell elements and elasto-plastic material the default timestep scale factor should
always suffice:

E
Et<<E
An elastic and a plastic stress wave are generated:

t
e p
e p
E E
c c
l l
t
c c

= >>
= <<

The non-linear stress field could thus be treated with a larger timestep (the elastic unloading
prevents us from doing so)

Things are different if the material characteristic is stiffening as in the densification phase of
certain material laws for foams:

Then:

max ,
l
t
E

| |
|
\ .

In the densification phase TSSFAC<0.66 may be required because the timestep reduces
drastically:

t

E
In general, stiffening materials will require a reduction of TSSFAC:

2
2
0
>

* TSSFAC<0.66

Workshop on Acoustic Wave Propagation

A 10.m long lead bar is loaded by a triangular pressure pulse of 2.ms duration applied on one
side

x
p
2. ms t
Input files:

deltat1000.k: 1000 brick elements 10.mm each
deltat0600.k: 500 elements 10.mm + 100 elements 50.mm

Check the input deck for the base load case (deltat1000.k) and verify that:

3
10. 6 /
7.
16.
E kg mm
G GPa
K GPa
=
=
=

This are properties for a soft, heavy metal like lead

The acoustic wavespeed is thus:

28 4
16
3 3
1600. /
1. 5
K G
c m s
E
+ +
= =

Consequently the time needed for the stress wave to travel to the other end of the bar is:

10
0.00625 6.25
1600 /
m
s ms
m s
= =

The duration (period) of the pressure pulse is 2.ms, the spatial extension (wavelength) of
the pressure pulse is then:

2. *1600. / 3200 3.2 Tc ms mm ms mm m = = =

The analysis timestep is the same in all cases and based on the smallest element in the mesh
(10.mm):
10
0.9 0.9 5.33 3
1600
c
l
t E ms
c
= = =

TO DO:

For the base load case (deltat1000.k):

Plot contours of xx-stress for a range 0.02 to 0.02 to visualize the stress wave, verify:
Transition time of compression wave (wavespeed)
Reflection of tensile wave
Wavelength of the load

Plot time histories of xx-stress in selected elements and x-velocity in selected nodal points to
verify wave propagation and reflection
Free surfaces stress?
Free surface velocity?
Particle velocity?

Fill out the following table based on the coarse part of the mesh:

element
analysis
t
t

load
element
l

Deltat1000

1. 320.
Deltat600

5. 64.
Deltat510

50. 6
Deltat505

100. 3

When does the solution degenerate completely?

Why?

This example shows that the meshsize imposes a limit on the frequencies that can be
represented for a given wavespeed

Aspects of Shell Element Technology

2. Aspects of Shell Element Technology

Shell elements are the main tools for crashworthiness and metal stamping applications.

Review of terminology and some basic features of finite element shell theory

The features that distinguish the different shell element formulations in LS-DYNA from each
other

Classification of shell elements

Importance of the different shell element features for applications involving small
displacements

Importance of the different shell element features for applications involving large
displacements


2.1 Introduction

Current shell element formulations in LS-DYNA:
(version 940)

Type Name Nodes Interpolation
Order
cpu
1 Hughes-Liu 4 2 3.5
11 Fast
Hughes-Liu
4 2
6 SRI Hughes-Liu 4 2 20.
7 Fast
SRI Hughes-Liu
4 2 10.
2 Belytschko-
Lin-Tsay
4 2 1.
10 Belytschko-
Wong-Chiang
4 2 1.1
8 Belytschko-
Leviathan
4 2 1.3
3 Belytschko-
Machertas
3 3
4 Belytschko-
Kennedy
3 2
5 membrane 4 2
9 FI
membrane
4 2
16 FI-ANS
(Bathe-Dvorkin)
4 2 3.5

The Hughes-Liu element family was the first shell elements to be implemented in LS-DYNA.

The table shows the remarkable efficiency of the Belytschko-Tsay element compared to fully
integrated shells.

An additional advantage of this element for crashworthiness analysis is the numerical robustness
of the formulation (warp angles of 180 degrees usually do not cause coredumps).

Consequently the BT element is the workhorse of all crashworthiness analysis.
2.2 Common Features of Shell Elements in LS-DYNA

Almost all shell formulations in LS-DYNA are lower order elements using bilinear interpolation
functions to define the element surface from the nodal coordinates. (Type 3 is the only
exception)

User Input Finite Element Surface

4 Nodal Points:

N1 (x1,y1,z1)
N2 (x2,y2,z2)
N3 (x3,y3,z3)
N4 (x4,y4,z4)

Bilinear interpolation creates an element surface where all 4 edges are straight lines, the element
surface is flat if the 4 nodes are coplanar.

These lower order elements are preferred because of:
Simplicity of coding
Numerical robustness
Higher order elements lead to reduced timestep values in explicit codes

Shell Element Formulations with Bilinear Interpolation

The real curved CAD-surface is approximated by a finite number of element surfaces (the FE-
surface) interpolated bilinearly between the nodal point locations: (with flat elements we obtain a
polygonal surface):

Available elements are:
* 3-node shells (type 4)
* Flat & warped 4-node shells (types 1-2-6-7-8-10-11-16)

The element surfaces represent the midplane of the shell, except if an offset is defined (only
types 1-6-7-11).
N2
N3
N1
N4
We approximate a curved CAD-surface by a polygonal FE-surface, even if all nodal points are
on the CAD-surface.

Smooth curved surfaces are thus approximated by surfaces with kinks (element angles).

A smooth undeformed finite element surface will develop kinks during bending deformation.

Mesh refinement is the only way to reduce kinks and chordal deviation!

Element angle
Chordal deviation
Alternative ways of modeling radii with bilinear shells:

4-5.mm
4-5.mm
4-5.mm
Bilinear Interpolation

We approximate the surface of the shell element by calculating the coordinates of any point
in the shell as a bilinear function of the coordinates at the 4 (or 3) nodal points:

3 1 2 4
1 1 2 2 3 3 4 4
1 2 4 3
x x x x x
y N y N y N y N y
z z z z z
| | | | | | | | | |
| | | | |
= + + +
| | | | |
| | | | |
\ . \ . \ . \ . \ .

1 1 1 1 1
2 2 2 2 2
3 3 3 3 3
4 4 4 4 4
N a b x c y d xy
N a b x c y d xy
N a b x c y d xy
N a b x c y d xy
= + + +
= + + +
= + + +
= + + +

The midsurface of the shell is thus approximated over the element by an as flat as possible
surface

The interpolation functions have identical values in a physical point of the element,
independently of the coordinate system

The 16 coefficients can be determined from expressing the trivial condition that each
interpolation function has a value of 1 in the corresponding nodal point and zero in the three
others:

N1 N2 N3 N4
x1-y1 1 0 0 0
x2-y2 0 1 0 0
x3-y3 0 0 1 0
x4-y4 0 0 0 1

For a general (trapezoidal, warped) element, this is not trivial and usually the interpolation
functions are determined in the isoparametric coordinate system since the interpolation formulas
must hold in any coordinate system:

In the isoparametric system the interpolation functions are trivially determined in their more
familiar form:
( ) ( )
( ) ( )
( ) ( )
( ) ( )
1
2
3
4
1
1 1
4
1
1 1
4
1
1 1
4
1
1 1
4
N
N
N
N

=
=
=
=

In the element center we thus have:

1 2 3 4
1 1
2 2
3 3
4 4
0
1
4
1 1

4 4
1 1

4 4
1 1

4 4
1

4
N N N N
N N
N N
N N
N N

= =
= = = =

= =

= =

= =

= =

1
4

N4 (-1,1)
N1 (-1,-1)
N3 (1,1)
N2 (1,-1)
Local Coordinate Systems

All operations needed to treat the shell element are performed in a local coordinate system
that is called local, or corotational system for MR and CBR shells

Strains: strain rates and stresses are typically only expressed in this system

In a small displacement context, this system is determined once at the first cycle and
corresponds to the undeformed configuration of the shell element

In a large displacement context (LS-DYNA) the local system must be recomputed every
cycle and is based on the current geometry of the shell element

This local system is assumed to be stationary in space at every cycle, it is not convected and
has no velocity

The default version of the local system in LS-DYNA is dependent upon the node numbering
of the element: it follows the global rotation of the element side N1N2, this is irrelevant only
if the element shear deformations remain small

The local z-axis is determined as the vector product of the 2 element diagonals:

As seen the origin for the local system is chosen in the first element node N1

The local z-axis is determined as:

1 3 2 4
1 3 2 4
z
N N N N
e
N N N N
uuuuuur uuuuuuur
r
uuuuuur uuuuuuur

z
N1
N2
N4
N3
The local x-axis is then positioned as well as possible along the element side N1-N2, the
match will only be exact if the element is flat:

The local x- and y-axes are then determined as:

( )
( )
1 2 2 4
1 2 1 2
z z
x
z z
N N N N e e
e
N N N N e e

=

uuuuuur uuuuuuur
r r
r
uuuuuur uuuuuur
r r

y z x
e e e =
r r r

This simple procedure defines an orthonormal local coordinate system for every
configuration of the shell element

All vectorial quantities (velocities, coordinates) will be transformed in the local system in
order to calculate strain rates and stresses in the shell

This is done using the orthogonal transformation matrix

( )

x y z
Q e e e =
t
r r r

1
T
x
T
T y
T
z
e
Q e Q
e
| |
|
= =
|
|
\ .
r
t t
r
r

z
N1
N2
x
The transformations are trivial and correspond to the projection of global vectors on local
base vectors:
T
T
T
x Q x
x Q x
Q
=
=
=
t
r r
t
r r
& &
r r t
& &

Note: in these notes we usually do not use the hat-notation since all equations are expressed
in the local element system

The definition of the local reference system is somewhat arbitrary since the x-axis is placed
along the N1N2 side, this means there is a preferred direction in the element that depends
upon the nodal numbering

This leads to problems if shear deformations are large

In LS940 and higher, a different, node invariant formulation of the local system can be
selected that will considerably improve numerical stability if in-plane shear deformations
occur in the shell

This formulation prevents non-physical rotations of the stress tensor and thus improves
overall stability and prevents hourglass modes from developing

This is selected by setting INN=2 on the CONTROL_ACCURACY card

This option is available for all shell formulations that use a local reference system

All shell elements use a local reference system, except types 1 and 6 (original Hughes-Liu)

2.3 Implemented Shell Theories: Determination of
Fiber Directions in Shell Elements

The geometrical description of the shell is fixed by the 3 coordinates of the 4 (3) nodal points

Additionally a definition of the fiber direction in every point of the shell element surface is
required to fully define the shell element

The fiber in a point is the straight line of material particles that is orthogonal to the midplane in
that point in undeformed configuration

Fiber directions are determined by LS-DYNA and depend upon the element formulation

2 basic approaches exist:

MR (Mindlin-Reissner) plate theory
CB (Continuum Based) and CBR (Continuum based resultant) shell theories

The fiber direction allows determination of the upper and lower surface of the shell by offsetting
the midplane by +/- half the thickness in every node.

The MR-formulation

For every element, determine 1 single fiber direction for the entire element as the vector cross
product of the 2 element diagonals:

1 3 2 4
1 3 2 4
N N N N
n
N N N N
uuuuuur uuuuuuur
r
uuuuuur uuuuuuur

This assumes the normal direction is the same in every point of the element surface.

Thus the shell element is effectively reduced to a plate element.

If the element is warped, the normal cannot be orthogonal to the shell surface in all 4 nodes.

This formulation is obtained with following element formulations:

Type 1-6-7-11 (HL) elements with control variable IRNXX=0 (CONTROL_SHELL card)
Type 2 element (BT) with control variable BWC=0 on the CONTROL_SHELL card
Type 16 element

Clearly the local z-axis and the normal to the element surface do not coincide in all points if
the shell element is warped, this is no problem for real shell elements but plate elements do
not properly account for this warpage

Even if the element undeformed configuration is flat, a good functioning requires a plate
element to remain nearly flat at all times

This can be expressed by restricting the deviation of the 4 nodal fiber directions from the
local z-axis:
( )
1 0.025
z i j i k
e N N N N
uuuuur uuuuur
r

The CBR and CB Formulations

For every element, determine the fiber direction for every nodal point as the vector cross product
of the 2 element sides connected to that node:

i j i k
i
i j i k
N N N N
n
N N N N
uuuuur uuuuur
r
uuuuur uuuuur

1
1
j i
k i
=
= +

Nj
Ni
Nk
The fiber direction in any point of the curved shell surface can be obtained by linear interpolation
from the 4 nodal fibers

Calculation of normal to the element surface can be done by bilinear interpolation:

( )
( )
4
1
4
1
,
,
p i i
i
p i i
i
x N s t x
n N s t n
=
=
=
=
r r
r r

This allows correct treatment of warped shells, the fiber directions remain orthogonal to the
current element surface in all 4 nodes

However if kinks are present in the geometry, a single nodal point may have different fiber
directions in different elements

This is often referred to as the 5-DOF approach on the element level

Corresponds to the MR formulation if the elements are flat

This formulation is obtained with following formulations:

Type 8 (BL) -10 (BW)
Type 2 element (BT) with control variable BWC=1 on the CONTROL_SHELL card
Type 1,11,6 and 7 (HL) with control variable IRNXX=-2 on the CONTROL_SHELL card
and if the FE-surface is smooth
Type 16 by adding *HOURGLASS with IHG=8

N1
N2 N3
P(s,t)
N4
Drill Projection

The component of the rotational velocity normal to the element surface (drill rotation) cannot be
resisted by shell elements. Drill rotations must be zero-energy modes and are therefore
eliminated from the rotational velocity in every node before the shell element is treated:

( )
_ : _
0
T T
i i i i i
T
T
x i
x i
T T T
y i i i y i
T
T
z i
z i
i i
T
x i
T
y i
Q n n
i node number
e
e n
e n e n
e n
e
if n e
e
e

=
=
| |
| |
|
|
|
=
|
|
|
|
| \ .
\ .
=
| |
|
|
=
|
|
|
\ .
r r r t
r r & & &
r
r &
r r
r r r
r r r r & & &
r r r
r &
r r
r
r &
r r
r & &

Thus for a flat element the drill projection just means we ignore the local z-component of the
rotational velocity

Warped Elements in Different Formulations

Consider dynamic bending of a cantilever beam under a load kept constant in time:

f
f
t
Using a pulse-type load the dynamic amplification is expected to be close to 2.

In the linear elastic case, the displacement under the load point is an oscillatory function around
a static equilibrium value of the deflection:

Beam dimensions are:

l=100.mm
b=30.mm
t=1.2mm

The static deflection can be estimated as (for a flat elastic beam loaded with 1. N):

3
1.1000000
0.39
3 3.200000.2.5
fl
D mm
EI
= =

A first series involving a flat beam shows equivalent results for elements of type 2, 4, 1 and 8.

In particular the bending stiffness of the C
0
-triangle element (type 4) seems equivalent to the
bending stiffness of the 4-Node elements.

Twisted Beams

The previous test is repeated for a twisted beam.

The section rotates gradually from horizontal at the clamped end to vertical at the loaded end.

d
ds
t

Side View

Top view

This is a continuous geometry, normals to a node in all adjacent elements are identical. MR
elements are not expected to perform well since all elements in the mesh are warped.

However the warpage of the individual elements is not high: about 10% for a 10 by 3 mesh.

The test is performed for 4 elastic and 4 elasto-plastic beams using elements of type 1, 2 4 and 10
in each.



Comparison of the deflection time histories for elements type 2, 4, 8 and 1 (MR-3N-CBR-CB)
show that the result diverges very quickly for the BT element (type 2).

The dramatic divergence is disturbing.

Similar problems have never been seen in full vehicle crash simulations based on the BT
element.

As a result the twisted beam problem with a higher load and elasto-plastic material occurs.

A single plastic hinge is formed at the clamped end of the beam, the structure rotates around this
point and is subsequently loaded by membrane forces.

All element types now yield similar global results.

The deformed shapes for the BT element contain some hourglassing.

The problem shows that BT elements are not prohibitive for the large-scale simulation of small-
deformation/large-rotation problems with catastrophic loads.

However, warped BT elements may create initial weak spots in the structure and initialize
buckling where it physically does not occur.

Therefore it should be avoided to have zones in the model where the number of warped elements
dominates the number of flat elements.

Flat surfaces should be strictly modeled as such and not disturbed in order to model connections
with other parts.

In particular, spotwelds between flanges should be modeled with springs, rigid bodies, massless
beams, spotweld elements or other methods, all avoiding the use of common nodes that would
deform the flange geometry:

plane flanges

Meshing of CAD panels should take the Gaussian curvature into account:

CAD PANELS
R=0
CAD panels
R>0 or R<0
>80% of shells are warped

Type 2 shell
Warp angle < 2
robust
Warp angle < 15
Type 2 shell
Too weak, robust

Type 4 shell
Too stiff, robust

Type 2 shell
BWC=1
Stiffness OK
Global loss of robustness

Type 10 shell
BWC=0
Stiffness OK
Local loss of robustness

distorted flanges
Behavior of plates versus shells:

accuracy
2 15 30
shell
plate
180
w
2.4 In-Plane Element Integration

Displacements, velocities and accelerations are evaluated at the nodal points

Stresses, strains and stress resultants are evaluated in the element integration points

The number of integration points and their location on the element surface depends on the
element type

Several choices exist:

FI: full integration: 4 Gauss points in the plane of the shell for all strain components

SRI: selective reduced integration: 4 Gauss points in the plane for membrane and bending
strains, 1 integration point in the element center for the out-of-plane shear strains

ANS: assumed natural coordinate strain element: 4 Gauss points in the plane of the shell for
all strain components, a special interpolation is used for the out-of-plane shear strains

RI: reduced integration: a single integration point in the center of the element is used for all
strain components

Shear Locking

FI of bilinear elements leads to shear locking: the out-of-plane shear strains (and stresses) do
not uniformly disappear in the case of pure bending and this results in an overstiff response
of the structure

The problem is easily illustrated by considering pure bending of a bilinear beam element:

l
z
N1
N2
x
The nodal rotations, deflections and interpolation functions are:

( )
( )
( )
( )
2
2
1
2
1
1
1 0
2 0
y
y
N
N
w N
w N
l x
N
l
l x
N
l

=
=
=
=
=

Which allows to evaluate the out-of-plane shear strain:

( )
1 2 2 2
2 2
2
0 _
2
/ 2 0
zx y
zx y y
zx y y
zx y
zx
w
x
w bilinear
N N
l x x
l l
x l
l
x l

= +
= +
= +
=
=

The out-of-plane shear strain is thus non-zero everywhere except in the center of the element,
evaluation in any other integration point will lead to a prohibitively stiff element

This problem can be avoided by using cubic interpolation for the deflection so that non-zero
deflection gradients cancel the rotation values
zx

x=l/2
x
No FI elements can be used unless some special techniques are used in order to prevent the
locking problem.

The obvious way is to use RI elements and restrict the strain calculation to the element
center, this also leads to computationally very efficient algorithms

Underintegrated Shell Elements

Element strains are evaluated at the element center only

This avoids shear locking but allows the existence of zero-energy or hourglass modes

Consider the membrane deformation of a 2D element where all4 nodes have a single DOF
(x-direction):

The 4 nodal velocities in x-direction then contain all the information necessary to fully
describe the incremental deformation of the element

The element velocity (displacement) vector can be seen as a vector in 4D space and decomposed
along 4 physically meaningful base vectors: rigid body translation, tension-compression, simple
shear and hourglassing (or in-plane bending):

1
2
3
4
1 1 1 1
1 1 1 1
1 1 1 1
1 1 1 1
rb tc ss hx
x
x
x x x x
x
x
| | | | | | | | | |
| | | | |

| | | | |
= + + +
| | | | |
| | | | |

\ . \ . \ . \ . \ .
&
&
& & & &
&
&

y
N3
N1
N4
N2
x
The last base vector is the hourglass mode:

The hourglass base vector is problematic since a zero strain value will result in the element
center, this is easily seen in the case of a rectangular element:

( ) ( )
( )
1 2 3 4
1 2 3 4
3 1 3 2 3 4
2 3
3
3
1 2 3 4
2 3
1 3
2 4
1
_ _ : / 2
2
_ _ :
0
xx
xx
xx
xx
x N N N N
x x x x
x x x x x
y y x y y x yx yx
x y
element center y y
y
x x x x
x y
hourglass mode
x x
x x

= = + + +

= + + (

=
= + +
=
=
=
&
& & & & &
& & & & &
& & & & &
& &
& &
&

The expression for the strain rate in x-direction is easily interpreted geometrically as the
gradient of the x-velocity component over the element or as the incremental change in chord
length of a linepiece parallel to the x-axis through the element center

Obviously the length of this linepiece (from one element border to the other) does not change
if the element deforms in a pure hourglass mode:

N2
x
N4
y
N3
N1
y
N4
N1
N3
IP
N2
x
Clearly this problem does not occur in a FI element since fiber elongation (positive strain and
tensile stress) is calculated in the lower integration points and fiber shortening (negative strain
and compressive stress) is calculated in the upper integration points:

In a RI shell element there are 5 hourglass modes: one for every nodal DOF:

- out-of-plane mode
- w-mode
- can be exited whenever out-of-plane loads are applied to 1 or 3 element nodes

y
N4
N1
N3
N2 x
x-membrane
y-membrane

- rotational hourglass modes are not visible in the deformed mesh

In-Plane Integration of the Shells in LS-DYNA

RI FI FI-SRI FI-ANS
1-11
2
8-10
5
9 6-7 16

Modeling Rules Using RI Elements:
RI-Elements and In-Plane Bending

Underintegrated shell elements are unable to carry in-plane bending loads:

Consider a single shell element representing a beam under in-plane bending loads:

rotational- x
rotational-y
x
IP
compression
zero strain
tension
No strain is calculated in the element center: the element will not resist the deformation

Obviously mesh refinement can fix this problem:

To quantify this problem a simple test case of in-plane beam bending can be used

We have simulated the in-plane bending of a beam clamped at both ends under a load in the
center that is ramped up and kept constant in time:

The beam material is linear elastic and the load is designed to cause only small deflections,
we are thus in a typical structural analysis case

We calculate the deflection under the load point as a function of time
IP2
IP3
IP4
compression IP1
tension
f
f
t
The analysis has been performed using 4 different meshes: 10 elements are used over the
length of the beam and 1, 2, 3 and 4 elements over the height

Elements of type 1, 2 and 9 are used



Inspection of the results for the type 2 element clearly shows that:

**Solution converges for meshes with 10*3 and 10*4 elements
**10*2 mesh shows a deflection that is almost 20% too high, this is a considerable error
**10*1 mesh diverges since only membrane stress is calculated in the underintegrated element
All underintegrated elements will fail this test if a single element is used over the structures
width

Some further results are summarized in the table:

element Cpu on R4000 Mesh 10*1 result
2 - BT 2 minutes +1000%
1 - HL 7 minutes +1000%
7 - HLCSR 21 minutes ok
6-9-16 ok

element Cpu on R4000 Mesh 10*2 result
2 - BT 2 minutes +20%
1 - HL 7 minutes +20%
7 - HLCSR 21 minutes ok
6-9-16 ok

All FI integrated elements pass the test with a mesh of 10*1 elements although 2
underintegrated elements still behave about 20% too weak, this is due to the location of the
Gauss points in the element plane

The need to accurately represent in-plane bending stiffness of structures thus leads to the
requirement of using a minimum of 3 underintegrated elements between any 2 events such as
kinks, loads, boundary conditions, connections etc

In particular a minimum of 3 underintegrated elements should be used per side of any open
or closed section

Due to its computational efficiency, the BT element remains the preferred element of all
modeling for crash simulations

As noted before the CFL timestep condition represents the main problem here since using 3
elements over the width of narrow metal strips will always cause a severe reduction in
timestep
Typical structures where this problem might occur are:

** Kneepolster brackets
** Window guiderails in doors
** Ondulated structures

Local use of (expensive) SRI or FI-ANS elements may actually reduce total computation
time considerably in these cases

Example of kneepolster bracket:

Convergence behavior of shell elements for inplane bending test:

Cut:

8mm 8mm
exact
d
RI-4N
FI-4N
3N
N
RI Elements and Out-of-Plane Bending

We have simulated the out-of-plane bending of a beam simply supported at both ends under a
load in the center that is ramped up and kept constant in time:



The analysis has been performed using 4 different meshes: 10 elements are used over the
length of the beam and 1, 2, 3 and 4 elements over the width

We use underintegrated BT elements

The load is distributed equally over the nodes in the beam midsection (2-3-4 or 5 nodes)

f
f
t

All results converge except the 20*2 element mesh which develops a w-hourglass mode

This can be prevented by using mass-proportional loads for all loads:

It is the alternating acceleration pattern around some average value that causes the w-
hourglass mode to develop, as soon as 3 or more elements are used, this pattern is
automatically broken:

f/3 f/3
f/3
f/4
f/2
f/4
m
2m
m m
2m m
f/6m
f/4m
f/3m
Accelerations:
f/6m
f/4m
f/3m
f/4 f/4 f/4 f/4
m 2m 2m m
Accelerations:
The conclusion is again that 3 elements per side of any open or closed section represents a
minimum condition for meshing since 2 elements will be rather sensitive to hourglass modes
under arbitrary loading

This requirement is non-trivial for vehicle models since it applies also to zones of small
deformations (non-impacted sides)

2.5 Element Integration Through-the-Thickness

( )
( )
( )
( )
( )
( )
2
2
2
2
2
2
2
2
2
2
2
2
1
1
1
1
1
1
2
2
2
2
2
2
t
t
t
t
t
t
t
t
t
t
t
t
n
xx xx i xx i
i
n
yy yy i yy i
i
n
xy xy i xy i
i
n
xx xx i xx i
i
n
yy yy i yy i
i
n
xy xy i xy i
i
zx zx
yz yz
t
f dz w z
t
f dz w z
t
f dz w z
t
m dz w z
t
m dz w z
t
f dz w z
f t
f t

=
= =
= =
= =
= =
= =
= =
=
=

After calculating stresses from strain rates using the material law (elasto-plastic in general),
stress resultants that act in the element midplane are calculated by integration of the stress
through the element thickness

The integration coefficients depend upon the location and the relative weights of the
individual integration points
Trapezoidal or Gauss integration rules can be selected:

NIP Zi Wi
Trapezoidal
Zi
Gauss
Wi
Gauss
1 0. 2. 0. 2.
2 +1.
-1.
1.0
1.0
0.5774
-0.5774
1.
1.
3 +1.
0.
-1.
0.5
1.0
0.5
0.7746
0.
-0.7746
5/9
8/9
5/9
4 0.8611
0.3399
-0.3399
-.8611
0.6521
0.3478
0.3478
0.6521
5 +1.
+0.5
0.
-0.5
-1.
0.25
0.50
0.50
0.50
0.25
0.9062

0.

-0.9062

Up to 10 Gauss points or trapezoidal integration can be selected for all shell elements in LS-
DYNA, this will correctly integrate a polynomial of order 9 (2*5-1).

Any other desired combination (equal thickness layers...) can be inputted as a user-defined
integration rule

Equal-thickness layers with constant stress are often used as user-defined integration rules to
model composites

Equal-thickness layers with linear stress distribution are sometimes used to model continua

Example of a user defined integration rule for a 4-layered composite:

Assume constant stress layers:

*INTEGRATION_SHELL
ID 4
-0.75 0.5 1
-0.25 0.5 1
0.25 0.5 1
0.75 0.5 1
zi wi PID

A sufficient number of Gauss points allows to integrate any continuous stress profile that can
be represented as a polynomial function exactly

n Gauss points allow exact integration of a polynomial of order (2n-1)

Thus 1 Gauss point allows to integrate a linear (elastic) stress profile and calculate the
normal force:

( ) ( ) / 2
m
N z dz t t Et = = =

z
z = 1
z = -1
x
stress
-t/2 t/2
z
Only 2 Gauss points are necessary to also integrate the bending moment (second order
polynomial) correctly

( ) M z z dz =

This clearly shows the superiority of Gauss integration in the linear elastic region, however
the elasto-plastic stress distribution is not a smooth function in z and can theoretically only
be represented by a polynomial of order infinity:

stress
-t/2
z
t/2
stress
-t/2
t/2
z
Simpson and Gauss-Lobatto rules are also good alternatives:

NIP Zi Wi
Simpson
Zi
Gauss-L
Wi
Gauss-L
1 0. 2. 0. 2.
2 +1.
-1.
1.
1.
0.5774
-0.5774
1.
1.
3 +1.
0.
-1.
0.33333
1.33333
0.33333
1.
0.
-1.
0.3333
1.3333
0.3333
4 1.
0.4472
-0.4472
-1.
0.16666
0.83333
0.83333
0.16666
5 +1.
+0.5
0.
-0.5
-1.
0.16666
0.66666
0.33333
0.66666
0.16666
1.
0.65465
0.
-0.65465
-1.
0.1
0.544444
0.71111
0.54444
0.1

These integration rules allow evaluation of the maximum bending stresses in the top and bottom
integration point.

Out-of-Plane Bending of Shells:
Elastic Bending of Moderately Thick Shells

We have simulated the out-of-plane bending of beams simply supported at both ends under a
pulse load

f
f
t


All beams are 100mm long and are modeled with 10 elements over their length and 1
element over the width (acceptable since only pure out-of-plane loads occur)

The first 8 beams have a thickness of 5.mm, the last 4 have a thickness of 10.mm, the shell
elements constituting these beams are thus perfect bricks

For elastic problems the linear stress distribution over the beam thickness is exactly
integrated with 2 Gauss points, these problems were run with 2,3,4,and 5 Gauss points,
obtaining identical results

We thus run the following cases:

Element Aspect Ratio l/t Integration points
2 2-3-4-5 Gauss
2 2-3-4-5 Trapezoidal
1 2-3-4-5 Gauss

Example of elastic beam bending:

10 elements over the length so element side length=10.mm

For thickness 10.mm we have the following ratios:

0.1
1.0
structure
structure
t
shell like
l
t
l
(
=
(

(
=
(

The shell elements are perfect cubes!

100.mm


The midpoint deflection can easily be estimated as:

3 3
3
48 4
fl fl
d
EI Ebt
= =

All numerical results are consistent, thus the element length to thickness ratio does not affect
the capability of the element to predict bending stiffness (equal displacements are obtained
by applying loads proportional to the thickness cubed)

Indeed the condition is for the Mindlin-Reissner assumption to be valid (straight fibers) and
this is a function of the slenderness and behavior of the structure, not the individual element

Consequently, sheet thickness will usually not impose a limit on the mesh refinement

Elasto-Plastic Bending of Shells

We have simulated the out-of-plane bending of beams simply supported at both ends under a
dynamic load in the center:

The beam material is elasto-plastic and the load is designed to cause bending stresses that
will exceed the yield strength in the top and bottom fibers of the beam only


f
f
t
Evolution of stress field in an elasto-plastic beam under a pure bending load:

All beams are 100mm long and are modeled with 10 elements over their length and 1
element over the width (acceptable since only pure out-of-plane loads occur)

The beams are 5.mm or 10.mm thick and 10.mm wide as in the elastic bending test

t/2
-t/2
z
elastic transition plastic hinge
x
We simulate the bending problem with 2 to 5 integration points through-the-thickness


The results with Gauss points show that:

**Using 2 integration points erroneously predicts elastic behavior

**Using only 3 integration points, the entire section plastifies somewhat too early, the deflection
is too high and no elastic rebound occurs, a visible deformation is present in the beam

**Using 4-5 integration points, the section shows partial plastification, the deflection remains
much lower and up to 60% elastic rebound occurs, no visible permanent deformation remains
in the beam

This shows the importance of using 5 integration points for thick parts of the structure that
may remain at least partly elastic during the loading phase

Under a much higher load, full plastic hinges develop rapidly in both cases and the results
become coincident (a horizontal constraint must be added under this high load)

Stress profile 3 IP 5 IP
For thin metal sheets under extreme loading (crash), this can be assumed to be mainly the
case

Bumper analysis frequently requires 5 integration points through-the-thickness of the plastic
covers

Steel sheet with thicknesses over 1.5 mm may need a number of integration points that exceeds
default (2) in order to avoid instabilities and non-physical behavior:

Results differ for 3 and 4 integration points (Simpson points were used)


4 and 5 integration points is converging towards the same solution (3 buckles)

2.6 Large Rotation Stress Update

Objectivity requires the stress tensor to rotate with the material, consequently a rotation will
change the stress tensor components in a fixed reference system without altering the state-of-
stress:

,0
0
,0
0
0 0 0 0 0
_ 0 0 0 0 0
0 0 0 0 0 0
_
_
0 1 0
_ 1 0 0
0 0 1
xx
yy
xx yy
T
R R
R

| | | |
| |
= =
| |
| |
\ . \ .
=
=
| |
|
=
|
|
\ .
t t
t t
t t
t

The update of the stress components thus requires the knowledge of the material rigid body
rotation matrix

Objectivity requires the stress tensor to rotate with the material, consequently a rotation will
not change the stress tensor components in a corotational reference system that follows the
material motion:

Y
Y
X X
Y
X
Y
X
,0
0
0 0 0 0
_ 0 0 0 0 0 0
0 0 0 0 0 0
xx xx
r

| | | |
| |
= =
| |
| |
\ . \ .
t t

The update of the stress components thus becomes very simple

However in a deformable body, the corotational system will be different in every material
point

First consider the cae of a shell element formulated in global coordinates:

In a large displacement FE-code, the stress components must thus be updated every cycle to
account for the incremental rotation as well as for the incremental deformation:

( )
1
_
T
n n
R R t
+
= +
t t
t t t t
&

Here dR is the incremental rotation of the material between timepoints t(n) and t(n+1)

To determine dR is trivial in a number of cases such as pure rigid body rotational motion,
pure tension or compression but becomes very difficult when the motion has a rotational and
a shear component

In LS-DYNA, dR is estimated according to Jaumann for the Hughes-Liu family of shell
elements (and most brick elements) (elements type 1,11,6 and 7)

This involves a rather lengthy calculation to be performed in every integration point to
estimate the instantaneous rotation field from the skew symmetric part of the velocity
gradient (Hughes-Winget):

1
1 2
1 2 1 2
1 2
1 2 1 2
1 2
1 2
2
1
2
1
2
n n
n
T
n n
n
n n
n
n
x x
x
x x
x x
t
R I
I t

+
+
+ +
+
+ +
+
+
+
=
| |
| |
|
=
|
|
|
\ .
\ .

= +

r r
r
r r
& &
t
r r
t
t t
t t

The formulation yields physically acceptable results for shear deformations of up to 50%
(never attained in metal sheets during a crash event)
Now consider a shell in local, corotational coordinates:

The stress components are constant in a reference system that follows the material rotation!
Indeed:

0
0 0
_ _ :
_
_ _ :
_
T
r T T T
if
R R
then
R R R R R R

=
= = =
t t
t t
t t t t t t
t t t t

This observation is the basis of the corotational element formulation, the stress is expressed
in a moving reference system that coincides with the motion if the shell undergoes a rigid
body rotation (=objectivity requirement)

In the Belytschko family of elements the assumption is made that the material rotation in all
element integration points equals the rotation of the element local system

The incremental rotation of the stress in the local system at tn is then easily obtained by
expressing the element base vectors at t(n+1) in the previous local system at t(n)

( )
( )
, , ,
1 , 1 , 1 , 1
1
1
1
1 1 1

n x n y n z n
n x n y n z n
n T
n n
n T T
n n n n
n T T n
n n n n
R e e e
R e e e
R R R
R R R R R R
R R RR R R R
+ + + +
+
+
+
+ + +
=
=
=
= =
= = =
t
r r r
t
r r r
t t t
t t t t t t
t t t t t t t

The stress update then becomes:

( )
( )
( )
1 1 1 1 1
1 1/ 2
1 1 1 1 1 1
1 1/ 2
1 1
1 1/ 2
_
_
_
n n n n T n
n n n
n n n T n n n T n
n n n
n n n
n n n
R R t
R R R R t
t

+ + + + +
+ +
+ + + + + +
+ +
+ +
+ +
= +
= +
= +
t t
t t t t
&
t t t t
t t t t
&
t t t t
&

The simplification arises because the material rotation and the rotation of the local system r,
cancel each other out between the timepoints t(n) and t(n+1): the stress components in the
corotational system are constant if no deformation occurs
We thus make 2 important assumptions:

** First the rotation is the same all over the element, thus the deformations must be small, in
particular shear deformations are limited to less then 5% for this formulation to be valid

** Second the stress tensor rotation follows the rotation of the element local system, for the BT
(type 2) element, this means the stress rotates with the element side 1-2, this can cause rather
arbitrary rotations of the stress tensor and lead to instabilities

The second problem is solved by using the invariant local system

Obviously both formulations yield identical results if no shear deformations are present

Moderately Large Membrane Deformations

Large in- or out-of-plane shear deformations are the most obvious cases where the rotation of
the material at the shell integration points and the rotation of the local system are no longer
the same

Since the element local system follows the side N1-N2 of the element, a dependency of the
stress values upon the element nodal numbering is introduced when using corotational
elements such as the BT

Illustration using simple shear load on 2 identical elements, numbered clockwise and anti-
clockwise resp.:

1 0
0 1
R I
| |
= =
|
\ .
t t

Yo=Yn
N4
N1
N3
Xo=Xn
N2

cos sin
sin cos
R I

| |
=
|
\ .
t t

A test case is set up to investigate the influence of element nodal numbering and large
rotation stress update on the energy absorption of a single element subjected to an in-plane
shear load:

Element: clockwise Anti-clockwise
Type 6 1 2
Type 7 3 4

N2
Xo
N1
Yn
N3
N4
Xn
Yo


The shear stresses that develop are independent of nodal numbering for the type 6 element
(Jaumann), but large differences develop once the strains become large in the type 7 element
(corotational). No other differences in the element formulations explain this phenomenon.

A somewhat more complicated test was performed using an elasto-plastic material law with
strain hardening:


Few explicit codes offer a Jaumann rate based shell element

The importance of this is in the simulation of weak materials such as tissues in seatbelts and
airbags, the use of Jaumann rate based shells allowed to avoid numerous instabilities (as does
the invariant local system)

Example of tensile tests on strips of belt tissue using type 2, 1 and 4 shell elements:



2.7 Triangular Elements

Triangles allows easy automatic meshing

First triangles in LS-DYNA were degenerated quads where one node gets double mass and
stiffness, this yields results that are dependent upon the nodal numbering

Today, the C0 triangular element (equivalent to BT) is used,

C
0
triangles require definition of separate parts specifying element type 4 OR automatic
sorting by setting ITRIST=1 on the CONTROL_SHELL card

Bending stiffness is equivalent to flat quadrilaterals

No hourglass modes

Results are usually too stiff

Conversion of an entire frontal car assembly from 4-node into 3-node elements caused all
results (rigid wall force, energy absorption, accelerations...) to rise by roughly 30%, this is
not acceptable

The element has low convergence: a very fine mesh is required

The use of triangles is unavoidable, they are needed for mesh transitions and in areas with
high double curvature. However we try to keep their number limited with respect to the total
number of elements in the mesh.
N3 N4
N1 N2
ok
n-ok
N3=N4
N1
N2
Too high stiffness is due to the in-plane shear behavior and depends upon mesh topology and
loading

The problem is unlikely to be solved unless higher order triangular elements are used
4-node: ok
3-node:
100% stiffer
f=kd
f=2k*(d/2)
2.8 Hourglass Prevention

If underintegrated shell elements are used, the hourglass modes may develop infinite
amplitudes unless they are prevented from doing so, nodal velocities will become unbounded
and cause the code to coredump

The classical methodology of hourglass prevention is referred to as perturbation hourglass: a
penalty is defined by the user to constrain the hourglass modes

Perturbation hourglass is implemented in element types 2 (BT), 1 and 11 (HL) and 10 (BW)

Alternatively, physical hourglass formulations require no user-input and take the actual
material law of the shell structure into account. Physical hourglass prevention is implemented
in element type 8 (EW-BL). The practical advantage does not seem very big.

The availability of fast FI elements (type 16) has also strongly reduced the importance of
elements based on physical hourglass prevention.

Perturbation Hourglass Prevention

Consider again the example of a 4-noded shell with a single DOF per node:

Assume that projection of the nodal velocities on the hourglass base vector reveals a non-
zero hourglass velocity component:

( )
1 2 3 4 1 2 3 4
1
1
0
1
1
hx
x x x x x x x x x
| |
|
|
= + =
|
|
\ .
& & & & & & & & &

N4
N1
N3
N2
x
It seems logical at this time to introduce a nodal force field that opposes the hourglass
component of the nodal velocities:

Mathematically, these forces must thus be opposed to the hourglass base vector, it also seems
logical to let them grow with the hourglass velocity component:

1
2
3
4
1
1
1
1
m hx
f
f
h ax
f
f
| | | |
| |
| |
=
| |
| |
\ . \ .
&

Here hm is the dimensionless, user-defined penalty (the membrane hourglass viscosity), the
default value is 0.1

The factor a has the dimension of a viscosity (Ns/m) and is often given as:

( )
2/ 3
4
c
a tA
=

Above we have generated a viscous hourglass force-field, alternatively stiffness hourglass
forces can be used:
1
2
3
4
1
1
1 8
1
m hx
f
f
Et
h x dt
f
f
| | | |
| |
| |
=
| |
| |
\ . \ .
&

When a stiffness formulation is employed, the hourglass factor hm is usually reduced to 0.01
or 0.02

Viscous hourglassing only stops the hourglass mode from developing further, stiffness
hourglassing will push the element back towards its undeformed configuration

Stiffness hourglass formulations seem to considerably stiffen the behavior of solid elements

f4
f1
N4
N1
N3 f3
N2 f2
x
LSDYNA parameter choices:

*HOURGLASS
HGID IHQ QM

For Shells:
IHQ=0=1=2=3=viscous hourglassing
IHQ=4=5=stiffness hourglassing
IHQ=8 only for type 16 element

For Bricks:
IHQ=0=1=viscous hourglassing, not recommended
IHQ=2=3=viscous hourglassing, recommended
IHQ=4=5=stiffness hourglassing (to be avoided)
IHQ=6 =stiffness hourglassing with correct bending stiffness (better then viscous in low
velocity impact)

QM: hourglass coefficient
QM=0.1 is default, ok for all viscous hourglassing

In shells, viscous and stiffness formulations seem largely equivalent if penalties are adapted

The hourglass energy is the work done by the hourglass forces:

0
1 , 1 1/ 2
t
h
nodes
n n h n n
nodes
he f xdt
he he f x t
+ + +
=
= +
r
r
&
r
r
&

To store the hourglass energy on the GLSTAT and MATSUM files set HGEN=2 on the
CONTROL_ENERGY card

The hourglass forces correspond to a non-physical external force field that will tend to reduce
the kinetic energy in a crashworthiness model

In a crash model the kinetic energy is reduced by the combined action of internal forces and
hourglass forces:

Stability requirement: ke+ie+he=ke(t=0)

Accuracy requirement: he<<ie

Since the hourglass forces cannot be expected to compensate exactly for the missing element
stiffness due to underintegration, an accurate result requires low hourglass energy compared
to deformation energy

This condition should ideally be fulfilled in every PID (check in the MATSUM file as well as
in the GLSTAT file):
0.1
he
ie

Nodal force development for an underintegrated element in a combined
symmetric/asymmetric mode:

M,v0 ke
he
ie: elastoplastic
+

min ,
2
8
i h
i y
h m
f f f
Et
f d t
Et
f h d
= +
| |
=
|
\ .
=

Hourglass Prevention and Rigid Body Rotation

If the element shape is not flat and rectangular, the perturbation hourglass forces may prevent
rigid body rotation of the element

Consider the rotation around the z-axis of a trapezoidal element in the xy-plane:

f
fi
fh
d
y
N4
N1
N3
x
N2
The nodal rotational velocities are easily obtained from the rigid body motion:

x y
y x
=
=
&
&

Thus:

1 1 1 1
2 2 2 2
3 3 3 3
4 4 4 4

x y y x
x y y x
x y y x
x y y x

| | | | | | | |
| | | |
| | | |
= =
| | | |
| | | |
\ . \ . \ . \ .
&
&
&
&

The hourglass velocity components are:

( )
( ) ( )
1 2 3 4
1 2 3 4 3 2
0
0
h
h
x y y y y
y x x x x x x

= + =
= + =
&
&

For non-rectangular elements, the calculated hourglass velocity components are non-zero and
this will lead to an erroneous prevention of the element rigid body rotation

Belytschko and Flanagan have shown how to prevent this

A similar problem occurs with warped elements, an adapted perturbation hourglass
formulation was proposed by Belytschko and Tsay

The following table summarizes the situation:

Flat rectangular Flat trapezoidal warped
Kosloff-Frazier
(1978)
Base vectors
ok n-ok n-ok
Belytschko-
Flanagan (1981)
Shape vectors
ok ok n-ok
Belytschko-Tsay
(1983)
ok ok ok

The importance of using an objective perturbation hourglass formulation cannot be
overstated for crashworthiness simulations

In version 930 (and previous) Belytschko-Flanagan hourglassing was used for all shell
elements of types 1, 2, 10 and 11, this leads to a non-objective behavior for warped shells

The objective Belytschko-Tsay formulation is default in LS-DYNA version 940, the older
Belytschko-Flanagan can be selected if necessary to obtain comparable results with previous
analyses

LS-DYNA implementation summary:

LS-DYNA
shells
LS-DYNA
bricks
Kosloff-Frazier
(1978)
Not
implemented
IHQ=0,1
Belytschko-
Flanagan (1981)
V936
Or in v940
*CONTROL_HOURGLASS_93
6
IHQ=2,3,4,5
Belytschko-Tsay
(1983)
V940 and higher -

On the CONTROL_HOURGLASS card, IHQ=1,23 gives a viscous formulation, IHQ=4,5 a
stiffness formulation for shells

A Pendulum simulation under gravity load shows that no interference with rigid body rotation
modes occurs for shell elements in LS-DYNA 940:




If we repeat the test with the following option:

*CONTROL_HOURGLASS_936


Then the pendulae with warped BT elements lag in the rigid body rotational motion, except for
element type 8 where physical hourglassing is used

Practical examples are abundant:

** Engine rotation during frontal impact
** Dummy spine behavior during side impact
** Seat back collapse in rear impact

Often, hourglass modes are best avoided by mesh refinement:

2.9 Modeling for Large Displacements

When folding and buckling of sheets must be simulated, factors other than element structural
stiffness become important

The mesh must be fine enough to give a smooth representation of the deformed structure

Therefore the number of elements required between any two events may go up from 3 to 6
(requirement due to the use of underintegrated elements) or more depending on the
curvatures in the deformed structure

Since enough DOF are required to smoothly represent the deformed structure, the element
type (integration points...) does not affect the need for mesh density

To obtain a simulation of good numerical quality, the mesh outlay should ideally be designed
for the deformed rather then for the undeformed structure

A parametric study performed by Mats Larsson at SAAB in 1989 gives the influence of the
mesh density on the energy absorption in a rectangular tube during axial collapse. The tube
dimensions were similar to a typical automotive longitudinal member

The actual number of elements required to converge depends upon the section shape, section
size, thickness, material properties and tube geometry

A smooth deformed mesh is the best indicator for mesh convergence

Current practice uses 6 to 8 elements per side in automotive front rails rather then 12

This partly explains the constant growth of state-of-the-art FE models for crashworthiness
simulations

Meshes should be regular and consist of lines parallel and orthogonal to the sides of the
component (and to the loading direction if possible)
E
6 8 10 12 14 elts/side
Mesh convergence
Then physical wave propagation can be simulated with minimal dispersion

Consider the example of a bar impacting a rigid wall and compare numerical and physical
wavefront:

Good quality meshes are usually esthetically pleasing!

2.10 Shell Element Classification (LS-DYNA v940)

Type Stress update type Integration
points
Hourglass
control
1 Jaumann Plate 1 Pert.
11 corotational plate 1 Pert.
6 Jaumann plate 4 -
7 corotational plate 4 -
2
bwc=0
corotational plate 1 Pert.
2
bwc=1
corotational shell 1 Pert.
10 corotational shell 1 Pert.
8 corotational shell 1 Physical
3 corotational triangle 1 -
4 corotational triangle 1 -
5 corotational membrane 1 Pert.
9 coroational membrane 4 -
16 corotational shell 4 -

In the above table assume the default value for the control parameter IRNXX.

By plate element we mean that the element surface curvature is not accounted for and bending
and membrane strains and stresses are fully uncoupled.

Workshop on Shell Elements in LS-DYNA

Repeat the class problems:

datafile problem
twist.inf twisted beam
inplane.inf in-plane bending
outelas.inf elastic out-of-plane bending
outplas.inf Elasto-plastic out-of-plane bending
memb.inf tension tests on belt tissue
hshell.inf pendulum simulation with shell elements
hbrick.inf pendulum simulation with brick elements

The following variants should be examined:

Datafile variants
twist.inf CONTROL_SHELL
IRNXX=0
BWC=1
use element types 1,2, 16 and 3
check element type 16 with hourglass type 8
inplane.inf use element types 5, 16 and 6
outelas.inf obtain convergence in all
cases
implement a Gauss-Lobatto rule
outplas.inf multiply load by 10. and
obtain convergence in all cases
outplas.inf obtain convergence in all cases
implement a Gauss-Lobatto
rule
memb.inf CONTROL_ACCURACY
INN=2
hshell.inf Check _v936 option
hbrick.inf Ensure objectivity in all pendulae


Numerical Treatment of Contact Problems

3. Numerical Treatment of Contact
Problems

The main contact definitions used to treat crashworthiness and impact engineering problems in
LS-DYNA (status version ls960).

Classical Contact Definitions

Principle: To prevent slave nodes from penetrating master segments

A segment corresponds to a 4-node shell, a 3-node shell or a side of a brick element

There is no notion of connectivity on the slave side in a contact algorithm

tied contacts are really constraints, the similarity with contacts is limited to the input format

We will consider contact algorithms for crash and impact analysis only:

One-sided
Contacts

Master/slave
Two-sided contacts

Single Surface (slave)
Nodal search
(old contacts)
orientation
required
3-5-10
surface_to_surface
nodes_to_surface
one_way_
surface_to_surface
4
single_surface
Segment search
(new contacts)
no orientation
required
a5-a3-a10 13-26-a13
automatic_single_surface
automatic_general
airbag_single_surface

Two-sided contacts always require a thickness to be defined since the impacting slavenode
can come from both sides

One-sided contacts also function without thickness offset since it is sufficient to check the
position of the slave node with respect to the normal of the master segment

Segment orientation is required for all old contacts

Contact type 26 introduces the new notion of edge-to-edge contacts

Overview of old contact implementations:

Type 5: N_2_S:

Type 10: 1_WAY_S_2_S:

Type 3: S_2_S:

SS

MS
SS

MS
SS

MS
Type 4: S_S:

Slave surface can be defined by segments, element ids, part ids, boxes...

SS
SS
MS
3.1 One-Sided Contacts for Impact Analysis:
Master-Slave Contacts

Nodal Based Search: Old Contacts

Principle:

For each slave node:

1. Global search: find the closest master node or neighbor node (using a bucket sort algorithm
in type 4 or incremental searching in types 3-5-10)

2. Local search: identify the segments connected to the closest masternode, project the slave
node upon the plane of each of these elements and determine which element contains the
projection of the slave node

3. Check if the slave node penetrates this element

4. If yes, apply contact forces to the slave node and the nodes on the closest master segment in
order to reduce the penetration

This methodology was first developed by John Hallquist in 1976 for public domain DYNA3D
(oldest 3D FE contact)

Global Search

For any slave node, identify the closest (neighbor) node on the master surface

Methodologies:

_ _ : _ *
_ _ : _
brute force cpu NSN NMN
incremental search cpu NSN

Incremental search is the only practical possibility, one complete search is done in the
initialization to provide initial neighbor nodes:


Overshooting is very unlikely:

If d is the distance a slave node can move during a single timestep and l is the smallest
characteristic element length in the mesh then overshooting would require:

d v t l c t v c = > = >

Whereas for normal crash conditions:

15 /
5000 /
0.015
v m s
c m s
d mm
Sn
Sn+1
Mn+1
Mn
Search Area
Sn
d
Sn+1
Mn
Mn+1
Search Area
Local Search

Determine a single neighbor segment for every slave node:

Identify all segments connected to neighbor node:

Define a plane for each segment defined by the 2 edges connected to the neighbor node

Orthogonal projection of slave node on this plane for each segment, check if the projection
point falls inside the segment:

If only one projection point is inside the corresponding segment, this segment becomes the
neighbor segment
SN
MN
SN
P1
MN
P2
This method will fail if the master surface is not smooth and/or not convex:

Penetration Check

For a flat master segment, the orthogonal projection of the slave node on the neighbor
segment is already known from the local search and the penetration check is trivial:

Using the unit length segment normal (oriented):

0 0
0
p
n PSN
p n PSN
=

< =
uuuur
r
uuuur
r

For warped elements this part of the problem requires solving a nonlinear equation

n

MN
SN
P
Contact Forces

A linear tension-only spring is added to the model if a penetration is detected:

The spring element force resultant is equilibrated by nodal forces, in the slave node:

c
SNP
f kp
SNP
=
uuuur
r
uuuur

The reaction force must be distributed over the 3 or 4 nodes of the neighbor segment:

( )
4
1
ci i c ci c
i
f N P f f f
=
= =
r r r r

MN
n
P
f=kp>0
SN
p
The closer the projection of the slave node to a master node, the larger the contribution of
this master node in the total reaction force:

Penalty Definitions

The contact spring stiffness is the product of the user-defined penalty factor (dimensionless)
and a stiffness determined from the master segment, or the slave node of the minimum of
both (default, see CONTROL_CONTACT):

( )
( )
( )
1 2
2
_
max ,
_
A
k sK shells
D D
A
k sK bricks
V
=
=

0.25
0.25

P
P
P
0.25
0.5
0.5
0
1
0.25
0
0
0
0
Ideally the contact spring stiffness should be very high in order to prevent penetrations and
add realism to the simulation

In reality the value of k is limited by the stability condition for the contact spring (not
checked by the code)

The default value of 0.1 ( = slfac on CONTROL_CONTACT) for the penalty optimizes the
stiffness of the contact spring for impact of 2 sheets with the same material and elements of
equal sizes

Limitation of contact spring stiffness:

The timestep of the analysis is determined by LS-DYNA from the elements of the FE-mesh
without considering the contact forces:

_
c
all elements
l
t Min
c
| |
=
|
\ .

Contact springs have a stiffness and can be attributed for part of the mass of slave node and
master segment:

( )
4
s s
c
s m
m m
t
k m m
=
+

During the initialization, an estimate of the contact timestep value is printed for each contact
surface on the d3hsp file (...timestep should not exceed...)

To guarantee stability the user must ensure that:

( )
_ all contacts c
Min t t

Consider the impact of 2 bodies with equal brick elements (perfect cubes):

Then:

( )
( )
3
3
_ _
2 2
1
4
3
2 / 2
2
1
1 0.5
2
c
V l
m surface node
l
t
K
K G
m l l
t
k s Kl K
s
s

= =
=
+
= =

The penalty s is set to 0.1 by default in order to keep contact forces small with respect to
numerical stability

How deep are the allowed penetrations if the contact spring is exactly on the stability limit
(s=1/2)?

Equalize kinetic energy of the slave node and potential energy of the contact spring:

2 2
2 2
mx kp m
p x
k
p x t
= =
=
&
&
&

With default values of s (s=0.1), penetrations can grow during about 5 timesteps and thus:

p=5*10mm/ms*0.001ms=0.05mm<0.1mm

Make sure all elements in contact with each other are of nearly equal size!!!!

Sliding Interface Energy

During contact kinetic energy of the impacting body is typically converted in deformation
energy of both colliding objects

This energy is numerically buffered in the linear contact springs that have potential energy:

2
2
springs springs
kp
sie fdp = =

This energy is non-physical and should ideally remain very small compared to the physical
energies in the system

f
f=kp
p
Numerically, the sliding interface energy is evaluated incrementally, the following is
summed over all springs:

( )
( )
1
1
1 1

0
1 1 1

1

2
n
t
n n
n n n n
contact springs
n n n n n
contact springs
n
contact springs
f f
sie fdp sie p p
sie sie f p p
sie f p
+
+
+ +
+ + +
+
+
| |
= = +
|
\ .
+
=

This shows clearly that the sliding interface energy increases with increasing penetration and
vice versa (tensile contact force is always positive)

For constant penetration, the sliding interface energy is also constant

Impact of a single node on a rigid element with contact algorithm corresponds to a perfectly
elastic impact:

Energy balance:

V0
-V0
e
sie
ke
t
Impact of a single node on a rigid wall corresponds to a perfectly plastic impact:

The sliding interface energy can also be interpreted as the incremental (external) work of the
contact forces done upon the contact springs (= - the work done upon the structure)

This is illustrated for the 1-D case:

1

, 1 2 , 1 2
1 , 1 , 1
, 1 , 1/ 2 , 1 , 1/ 2
, 1 , 1/ 2 , 1 , 1/ 2
3 _ :
n
contact springs
sn n mn n
n sn n mn n
sn n sn n mn n mn n
slavenodes masternodes
sn n sn n mn n mn n
mastern
sie f p
p x t x t
f f f
sie f x t f x t
D
sie f x t f x t
+
+ +
+ + +
+ + + +
+ + + +
=
=
= =
= +
= +

& &
& &
r r
r r
& &
slavenodes odes

V0
e
ke
t
MN
SN
x
Sliding interface energy is properly accounted for on the GLSTAT file by setting
SLNTEN=2 on the CONTROL_ENERGY card

Save the SLEOUT file with a reasonable time-interval

Nodal contact forces can be stored on NCFORCE

Sudden changes in master element thickness (or stiffness) can also lead to negative contact
energies:

e
SN1
SN4
SN2
SN3
ke
sie
e
t
Negative contact energy always corresponds to a numerical generation of energy in the
system

Initial penetrations will always lead to negative contact energies:

SN
pi
p
sie
t
t
Contact Friction

Sliding interface energy on the GLSTAT file contains the work done by normal and surface
contact forces

Frictional (surface) energy is physical, dissipated and monotonically increasing

Work of normal contact forces is numerical and restituted, should remain small

Set FRCENG=1 on CONTROL_CONTACT to account for frictional forces separately

Bookkeeping is done in the interface force file: DATABASE_BINARY_INTFOR ( s=...)

Contact Quality Checks

Requirements:

( )
1 0
0
0
_ : _
_ : _
i
c
c
e
p
sie
f
f t C C
t
p t crash
sie ie
p elastic
=

<
<<

<
r
r

3.2 Contact Algorithms for Automotive Crash: Single
Surface Contacts

Overview:

4 13

26
SINGLE_SURFACE AUTOMATIC_
SINGLE_SURFACE

AUTOMATIC_GENERAL

AUTOMATIC_GENERAL_I
NTERIOR

NODAL
SEARCH
SEGMENT SEARCH

SEGMENT SEARCH
PENMAX=0.4

PENMAX=0.4 PENMAX=200
DEPTH=1
(fixed)

DEPTH=2
(variable)
DEPTH=3
(variable)
NODAL NORMALS

CYLINDRICAL EDGES SEGMENT NORMALS
(SMP)

CYLINDRICAL EDGES
(MPP)

IGNORE=0

IGNORE=0/1 IGNORE=0 (SMP)
IGNORE=0/1
(MPP)

SHELL
BRICK
SHELL
BRICK
SHELL+BEAM
BRICK

Problems with Axial Folding in Automotive Members

Incremental search works fine for contact between smooth and convex surfaces

Incremental search may fail to find the closest Masternode in accordion folding problems:

In general incremental search may fail on meshes that are not simply connected or have high
curvature (axial folding)

Full searching becomes necessary at least every N cycles

Bucket sorting (first introduced in 1985 for type 4 contact) is done every 10-100 cycles

Incremental search still performed every cycle

Search frequency controlled by parameter NSBCS on the CONTROL_CONTACT card, or
BSORT on optional contact card A, default depends upon the contact type

Bucket sort is the expensive part of the contact

Bucket Sorting

A 3D fixed grid is spanned over the structure and at every cycle all nodes are sorted according to
their current bucket position:


Distance to the nodes in the slave node bucket and the 27 neighbor buckets are calculated, the
cost is proportional to: (N is the averaged number of nodes per bucket):

N*(27N-1)

As opposed to:

N*(N-1)

for the brute force approach .

Problems with Axial Folding in Automotive Members

A 2-sided contact becomes necessary since the metal sheet contacts itself on the top and
bottom side:

A contact thickness must be defined in order to check for penetrations:

0
c
p n PS t
r
r

Forbidden zone
Contact
thickness
Contact thickness should ideally be the sum of the half thicknesses of the sheets in contact:

The contact of slave node to master segment should occur simultaneously with the physical
contact between the two sheets

Note that contact thickness is not assigned to a master segment but to a contact pair or impact
event:

Contact pair: MS+SN1: tc=0.5*(tm+ts1)

Contact pair: MS+SN2: tc=0.5*(tm+ts2)

Too small contact thicknesses will allow for penetrations since:
- The slave node must be detected before crossing the midsurface of the master segment
- Only full search guarantees detection

20*15 / *0.001 0.3 1.0
c
Nx t t t
mm ms ms mm mm

=
&

ts
tm
(tm+ts)/2
SN1
SN2
MS
Practical experience shows that the contact thickness in crashworthiness problems should be
equal to or larger than 0.5 mm

In a two-sided contact the penetration must be detected before the slave node penetrates
through the midplane of the impacted master segment

2
s m
c
t t
t
+
=

Once the contact force direction is determined for a slave node, tracking allows deep
penetrations and switching neighbor segments (normal directions of master segments are
compared)

2
s m
c
t t
t
+
=

SN1
SN2
tc
tc
SN1
If penetration is detected after crossing the midplane, the contact force will be zero because
of the shooting node logic: SNLOG=0 (default)

Correct structural stiffness may require contact thicknesses equal to physical thicknesses in
order to prevent voids:

SNLOG=1
SN2
tc
SNLOG=0
SN2
tc
tc
2 2
s m s m
c
t t t t
t void tc
+ +
< =

Thus for tm=ts=2.mm and tsc=tmc=0.6mm a void is created between the flanges of 1.4mm

Contact Thickness in LS-DYNA

General principle for single surface and master/slave contacts:

2
c
MST SST
t
+
=

Here MST is the thickness of the master segment and SST is the thickness of the slave node
(=maximum thickness of all connected slave segments)

Mostly the MST and SST are user-defined on contact card 3, the contact thickness is then
uniform and initial penetrations can be avoided in complex meshes

Default contact thickness for shells in single surface contacts (4-13-26) is a function of the
parameter SSTHK on the CONTROL_CONTACT card:

SSTHK=0 SSTHK=1
tc=min(t,lmin) tc=t

Thickness scaling can be done on the CONTACT card as well as per part on the
PART_CONTACT card (typically use scale factors on MST and SST of 0.8 to 0.9)

Recommendation for contact thickness definition:

*Mesh on the midplane geometry

*Use a uniform mesh size

*CONTROL_CONTACT
SSTHK=1

*CONTACT (Card 3)
Possibility 1: SST=MST>0.6mm SFST=SFMT=1.
Easier, less initial penetrations
Possibility 2: SST=MST=0 0.9>SFST=SFMT>0.8
More physical, more difficult to avoid initial penetrations

*Introduce PART_CONTACT cards to define contact thicknesses different from physical
thickness in order to allow parametric studies on sheet thickness without remeshing

For airbag contact, a variable contact thickness in time may be defined on optional contact
card A (LCIDAB) allowing to gradually correct initial penetrations

Automatic elimination of initial penetrations (v960):

CONTROL_CONTACT : ignore=1

The penetration is reduced by the previous minimal penetration that occured:

( )
( )
, 1
min ,
n n in
in n i n
f k p p
p p p

=
=

Example:

This should be used to eliminate SMALL initial penetrations:

0.5
i
tc p mm

Modeling recommendations for contact definition:

*Mesh on the midplane geometry

*Use a uniform mesh size

* Guarantee penetration free model for tc=0.6mm

*Introduce PART_CONTACT cards to define contact thicknesses different from physical
thickness in order to allow parametric studies on sheet thickness without remeshing
p
p = 0
pi
tc
t
Recommended input settings for crashworthiness contact definition:

*CONTROL_CONTACT
SSTHK=1
IGNORE=1

*CONTACT (Card 3)
SFST=1.
SFMT=1.
SST=0.
SMT=0.

*PART_CONTACT
t (does not change in parametric study)

*SECTION_SHELL
t (changes in parametric studies)

Ensure the deck is penetration free for a testrun using SST=SMT=0.6 mm

Problems with Nodal Based Search

Cost reduction depends on number of nodes per bucket N , this number increases towards
the end of the simulation when the structure is crushed

Sometimes the closest node does not contain the nearest master segment:

Slave node
Closest masternode
If sharp kinks are present in the mesh, the selection of the closest master segment may not be
clear and lead to penetrations:

Segment Search Solves Most Problems

Implemented in all new (LS-DYNA) contact algorithms

Loop is performed over all master segments

Multiple neighbor segments are stored for every slave node

Search is very fast

For each segment, a subset of buckets is determined that contains the segment.

Orthogonal distance of all nodes in the subset to the segment is determined.

For every node the DEPTH closest segments are stored.

Release of Nodes that Penetrate Too Deeply

The original motivation was to prevent large contact forces when false penetrations are
detected by old one-sided contacts (3-5-10):

This can be prevented using the new contact algorithms (a3-a5-a10 or single surface)

Or by activating the small penetration check by setting PENCHK=1 or 2 on the second
CONTACT card, or define PENMAX on optional contact card B

General release conditions are controlled by parameters XPENE (CONTROL_CONTACT)
and PENMAX (optional contact card B)

Note that by default nodes are released in contacts of type 4, a3, a5, a10 and 13 before they
penetrate through the midsurface of the master segment:

d=0.4*(SST+MST)

Note that by default nodes are released in contacts of type 26 long after they penetrate
through the midsurface of the master segment:

d=200.*(SST+MST)

In general, the release criterion depends on element type (shell-solid) and contact type

In a two-sided contact the penetration must be detected before the slave node is released:

( ) 0.4
r s m
t t t = +

For ts=tm=0.6 mm this is only 0.45 mm

It is preferable to release nodes when they penetrate deeply (= through the midplane) in order
to avoid impact on neighboring segments from behind

The node is currently not eliminated from the slave set, just from the current contact pair, so
impact on another segment later remains possible

Deep penetrations followed by release should be avoided

SN1
SN2
tr
SN1
tr
tr
Treatment of Edges

Normals must be defined on the contact master surface in order to determine the direction of the
contact force and to create a contact surface by offsetting the segment surface by the contact
thickness:

Thickness offset is optional for contacts 3-5-10 and activates new contact algorithms
(SHLTHK=1/2)

Thickness is always included for contacts 4-13-a5-a3-a10 and 26

Old contacts (type 4, type 3-5-10 without thickness offset) use nodal projection: (slow, consistent
orientation required)

New contacts (types 13, a3,a5,a10 and 3-5-10 with thickness offset) use element projection:
(faster, no consistent orientation required)

Problems with Segment Based Projection

The direction of the contact force (normal to master segment) is not normal to the modeled
curved surface and can change abruptly when a slave node changes master segment

Consequently small changes in slave node coordinates as may occur when using different
platforms can lead to a different path after impact:

Example:

platform A
platform B
node-on-edge impact
t
v

Problems with Segment Based Projection

Negative sliding energy can be generated when a node penetrates between 2 segments at the
edge, extensions to the master segment must be defined in order to avoid penetrations

The discontinuous rise in the value of the penetration results in a burst of potential energy for
the impacting slave node, this is converted in kinetic energy as the node exists normally to
the master segment, resulting in a (permanent) negative sliding interface energy

The incremental calculation of the sliding interface energy fails to capture the discontinuous
growth which is why negative energies appear in the energy balance:

Indeed for the sudden penetration:

1 1
, 1/ 2 , 1/ 2 1/ 2
n n n
sn n mn n n
p p p
x x p
t t
+ +
+ + +
<<< = =

& & &

these node-to-edge penetrations cause the energy in the system to rise since the potential
energy is created in the contact spring without a corresponding loss of kinetic energy for the
impacting node
p
sie
true
calculated
t
Contact segments are extended beyond the element size to prevent energy jumps and
penetrations:

The size of the extension is given by the MAXPAR parameter on the second optional
CONTACT card

max
l dl
par
l
+
=

In contact type 13 (AUTOMATIC_SINGLE_SURFACE) the impact of slave nodes on
element edges is treated ( contact surface is continuous):

dl
This can lead to unexpected initial penetrations in structures with thick ribs:

Cylindrical extension should be optional on free edges

Contact with Parts of Different Stiffness

It is difficult to optimize the penalty value if the stiffness, density and mesh size vary a lot

A penalty that is too high will result in instabilities, therefore in foam materials a separate
modulus is defined that determines contact stiffness and timestep

Automatic optimization of the contact spring stiffness in every contact point is obtained by using
the soft constraint option where contact spring stiffness is a function of slave node mass and
current timestep:
2
m
k s
t
=

This is obtained setting SOFT=1 on the optional contact card A, the penalty factor s must be
smaller then 1 and is obtained from SOFSCL on optional contact card A as

s=SOFSCL

Practically s must be between 0.1 and 0.5

If a soft constraint contact allows too deep penetrations, then the spring stiffness can only be
increased by scaling the timestep value

General Contact Conditions

The idea is to convert the kinetic energy of the impacting body in deformation energy of the
impacted body

Intermediate storage of all energy in the contact springs is necessary, good functioning of the
contact means a relatively small allowable maximum penetration pmax and requires to fulfill the
following condition:
2 2
max
2 2
allsprings
mv kp
<<

There are 3 possible ways to fulfill this condition:

Allow larger penetrations and loose accuracy
Increase k by decreasing the timestep
Increase the number of springs by refining the mesh
m
v
Mesh refinement can solve most contact problems

Example: metal sphere impact on soft foam block

The coarse mesh leads to an unacceptable energy balance:

Refining the mesh solves this problem:


The energy balance is now physical:

Edge-to-Edge Contact

Null beams with diameter equal to the thickness of the shell are automatically added to free
edges in type 26 contacts

Null beams with diameter equal to the thickness of the shell should be manually added to
non-free edges that represent kinks in the mesh

Contact forces are orthogonal to the plane of the penetrating edges:

Segment Based Contact

Activated by setting SOFT=2

Available for contacts type 3,10,4,a3,a10,13 and a13

Penetration of all nodes of a segments is checked with the plane of a penetrated master
segment

n1
n2
( )
1 2
f pk n n =
r
r r

Penetrating segments are identified checking nodal penetration of the side planes

Reaction forces on the master segment are equilibrated with respect to the center of force on
the slave segment

More cpu intensive then standard contacts

Initial penetrations ignored

No problems in corners (on generation of energy)

Penalty is of soft constraint type based on initial timestep

If DEPTH=3, contact forces are evaluated at the edge contact points rather than at the nodes
(forces are still orthogonal to segment surface)

Limited edge-to-edge capability if EDGE=1 on Contact Card A (v970)

Solves airbag deployment problems without time-dependent contact thickness definition
(SOFT=2 recommended for a13 type contact)

Modeling for Contact

Avoid initial penetrations at all cost

Avoid tangled meshes at all cost

The previous require adequate offset of the metal sheets, careful thickness definition and
regular meshing

Use old contacts only for smooth, convex, if possible closed master surfaces

Type 13 general contact is very reliable, due to segment searching 1 large contact zone is not
more expensive then several overlapping small ones

Undeformable or very stiff parts may still require fine meshes if their kinematics depend on
the contact pressure distribution, contact force must then be distributed over many nodes in
order to obtain realistic contact resultants (example: EUROSID shoulder clavicle)

Use contact thickness values that are close to real sheet thicknesses in order to avoid creating
unrealistic gaps that decrease the structural stiffness of the model:

This becomes particularly important if aluminum structures are considered because of the
relatively high sheet thickness

tc
t
tc t
3.3 Connections of Thin Structural Sheets

The following connection types may be important in vehicle bodies made out of steel,
aluminum, or composites:

Continuous welds (MIG-TIG...)
Spotwelds
Rivets
Adhesive bonding

Several issues must be solved for all:

Flexibility of the connection
Rupture

In a modern vehicle model we would like to respect one principle:

FREE MESHING OF ALL FLANGES (meshing must be possible before
spotweld locations are decided upon)

If flanges are to be independently meshed, the use of tied contacts becomes unavoidable, in
LS-DYNA we dispose of the following:

Type Name
6 tied_nodes_to_surface
o6 offset
7 tied_shell_edge_to_surface
o7 offset
2 tied_surface_to_surface
o2 offset
8 tiebreak_nodes_to_surface
9 tiebreak_surface_to-surface

The following tied contacts are equivalent except for the way we define the slave surface:

Slave surface as nodes Slave surface as segments
8 9
6 2
o6 o2

The main issue with tied contacts is the transmission (or not) of bending moments, consider a
simple test:

The transmission of rotations makes type 7 tieds suitable for the simulation of spotwelds

Other remarks:

Tied contacts of type 2/6/7/8/9 only function well if the slave nodes are on the master surface

Tied contacts 2/6/7 will reposition the slave nodes on the master surface if an offset exists,
this may create warped elements

Tied contacts of type 8/9 do not perform this repositioning and require the slave nodes to be
on the master surface

Tied contacts of type 2/6/7/8/9 are kinematic constraints, they are incompatible with rigid
bodies, boundary conditions and other tied contacts

Tied Contacts 6/o6/8 Tied contacts 7/o7
Tied contacts o2/o6/o7 will function properly with offset (penalty formulation) and are
compatible with rigid bodies, boundary conditions etc...

Continuous Welding

Usually we assume no failure in the weld

However the HAZ around the weld (usually taken 1 element wide) has reduced yield stress
and reduced rupture strain (4-8%?)

Nodal rigid body constraints or common nodes or tied contact type o7 can be used to
simulate the actual weld

Material properties for the HAZ must be determined experimentally

Rupture can be considered in the weld using
CONSTRAINED_GENERALIZED_WELD_OPTION

Rivets and Spotwelds

Type 9 beams with material type 100 will include flexibility of the weld element

CONSTRAINED-SPOTWELD and CONSTRAINED-RIVET are undeformable up to
rupture and incompatible with other kinematic constraints

o7 tied contact for spotwelds, o6 tied contact for rivets (or also type 7 or 6 if beam elements
are used)

Rupture characteristics must be determined experimentally in terms of tensile and shear
forces

Minimum 3 elements in the flange between individual welds or rivets to allow spreading
HAZ
Preferably 3 elements over the width of the flange to allow central positioning and obtain
element size corresponding to the spotweld or rivet diameter

Automatic contact to be added in order to avoid penetration of the flanges

Adhesive Bonding

This is a continuous connection involving the entire flange area

Flange surfaces could be directly tied if the flexibility of the adhesive can be neglected:

Simplest method: o2 tied contact, problem: no rupture and no flexibility in the bonding

Alternative: type 9 contact, problem: no offset allowed, no flexibility in the bonding

To consider the flexibility in the bonding, a single layer of brick elements can be used:

The brick element should be type 9 (small strain) to avoid timestep problems, material
properties (elasticity) and failure stress/strain must be determined experimentally.

Brick element surfaces can be connected to the flanges using tied contacts of type 9 (if no
offset, allows to consider failure in the tied contact rather then in the brick material model) or
tied contacts of type o2 if offsets cannot be avoided in the model.

Shell t1
Adhesive ta
Shell t2
Brick: ta+t1/2+t2/2
Adhesive Bonding + Spotwelds or Rivets

The method above can be combined in principle if the welds/rivets use only o6 or o7 contacts

The adhesive will have melted away in spotweld locations

Material Testing for Adhesives

Typical tests are:

Crack propagation:

Shear:

Tension:

Final Remark

if tied contacts are used, check the d3hsp file carefully for warnings indicating that the slave
node is too far away from the mastersurface, this will result in a missing connection

Typically a distance < 0.6*(tm+ts) is required


Material Modeling

4. Material Modeling

Material Modeling
4.1 Classification of Materials for Numerical
Simulation

Main decisions:

Is the structure 2D or 3D?
Is the material reversible or irreversible?

Reversible Irreversible
(permanent
deformations)
Elastic ML
Visco-elastic ML

Elasto-plastic ML
2D (Shell) Tissue (Bag-Belt)
Plastic (Polymers)
Mild steel
High strength steel
Aluminum
Magnesium
Plastic (Polymers)
3D (Solid) Rubber
(Elastomers)
Soft foams
Confor foam
EPF (EPP,EPS...)
EA-PU
Honeycomb
Aluminum foam
Crushable PU

Sometimes the choice is hard since part of the deformation is reversible and part is permanent.
(Plastics!)
Material Modeling
4.2 General Notions

4.2.1 Measures of Stress in Solids

We consider Cauchy stresses (force over current area)

We will write the stress tensor as:

_
xx xy xz
T
yx yy yz
zx zy zz

| |
|
= =
|
|
\ .
t t

Since the stress tensor is symmetric, we can always find a proper orthogonal matrix (=rigid body
rotation) that diagonalizes it:

1
2
3
0 0
0 0
0 0
T
R R
| |
|
=
|
|
\ .
t t
t

The diagonal components are the principal stresses and allow a 3D representation of the state of
stress in a point of a continuum

Invariants

The state of stress in a point is determined by the invariants of the stress tensor, the three
invariants are:
( ) ( ) ( )
( )
2
2 2
2 2 2
_
3
3
_ 2 2 2
2
_
xx yy zz
vm xx yy zz xy yz xz
p
p p p

+ +
=
= + + + + + + + +
t

The first and second invariant are called pressure and von Mises stress resp.

By invariance we mean invariance under a proper orthogonal transformation:

0
0
0
_
_
_
T
vm vm
R R
p p

=
=
=
t t
t t

Material Modeling
Objectivity of the Stress Tensor

Example of a bar in compression with rotation of the reference system:

*
0 0 0 0 0
0 0 0 * 0 0
0 0 0 0 0 0
*
3
vm vm
p p
| | | |
| |
= =
| |
| |
\ . \ .
= =
= =
t t

Invariant Space

For an isotropic material, the plasticity law will depend upon the stress invariants only, if we
further restrict to the first 2 invariants, we can represent the material law in a 2D plane:

y x
x y
3
1
p
_
vm

Material Modeling
Here:

- the horizontal axis corresponds to a hydrostatic loading path
- the vertical axis is pure shear
- the line with tangent 1/3 is uniaxial compression
- the line with tangent -1/3 is uniaxial tension
- a line of constant von Mises stress is horizontal
- etc...

Uniaxial State-of-Stress

0 0
0 0 0
0 0 0
xx
| |
|
=
|
|
\ .
t

( )
( )
2 2
3
3 _
3 2 3 _
2 0
2 3 3
xx
vm
xx xx vm
vm xx
p
p compression
p tension
`
| |
= | | | |

= + =
|
| |
|

\ . \ .
\ .
)

X
Material Modeling
Deviatoric Stresses

The pressure or first invariant is related to the change in volume of the solid, the deviation from
a hydrostatic state of stress is linked to the change in shape.

The stress deviator is defined as:

_
0 0
_ 0 0
0 0
xx xy xz xx xy xz
yx yy yz yx yy yz
zx zy zz zx zy zz
s pI
s s s p
s s s p
s s s p

= +
| | | |
| |
| |
|
= +
| |
|
|
| |
\ .
\ . \ .
t
t t

The second invariant in terms of stress deviators becomes:

( )
2 2 2 2 2 2
3
_ 2 2 2
2
vm xx yy zz xy xz yz
s s s s s s = + + + + +

A surface of constant von Mises stress in deviatoric space (9D) or in principal deviatoric space
(3D) is thus a sphere.

Representation in the Invariant Plane

p
_
vm

( ) , ,
vm
p =
t

( ) , 0 pI p =
t

Material Modeling

Remarks

A hydrostatic state of stress is the kind that can exist in a liquid or a gas. It has such a high
degree of symmetry that the corresponding deformation can only be a change of size, without
change of shape (cube remains cube, sphere remains sphere etc...), as long as the material is
isotropic.

Similarly the deviatoric stresses can produce only shear deformation changing the angle between
2 faces of a block or between 2 of its diagonal planes. Indeed, the deviatoric stresses have a zero
trace (volumetric component) and can thus be represented as the superposition of 5 simple shear-
stress systems:

0
_ :
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0
_ 0 0 0 0
0 0 0 0 0
xx yy zz
xy xz
xy yz
yz xz
xx
xx zz
zz
s s s
thus
S S
s S S
S S
S
S S
S
+ +
| |
| | | |
|
| |
= + + +
|
| |
| |
|
\ . \ .
\ .
| | | |
| |
+
| |
| |
\ . \ .
t

Material Modeling

Y
X
0
xy yx
=
Y
X
0
yy xx
= <
Material Modeling
Concept of Material Law in Hydrocodes

4.2.2 Measures of Strain

All measures of strain are a function of the velocity gradient and/or the deformation gradient:

0 0 0
0 0 0 0
0 0 0
x x x
x y z
x y y y
L
x x y z
z z z
x y z
x x x
x y z
x y y y
F
x x y z
z z z
x y z

| |
|
|
|
= =
|
|
|
|
\ .
| |
|
|
|
= =
|
|
|
|
\ .
& & &
r
&
t
& & &
r
& & &
r
t
r

Material law

Constitutive Law
Shear
Change in shape

s

Equation of state
Pressure
Change in volume

p
Material Modeling
Evaluate the First Component of the Velocity Gradient

Evaluation is in an integration point P of a finite element:

The velocity field must be evaluated in two points on an axis parallel to x through P

2 1
2 1
x x x
x x x

& & &

The most familiar strain measure is the small strain tensor:

0
0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
1 1
2 2
1 1
2 2
1 1
2 2
u
u x x v
w
u u v u w
x y x y z
u v u w v
E
y x y y z
u w u w w
y z y z z
| |
|
= =
|
|
\ .
| | | | | |
+ +
| | |

\ . \ .
|
|
| | | |
|
= + +
| |
|
\ . \ .
|
| | | |
|
+ +
| |
|

\ . \ . \ .
r r r
t

z
P
2
1
x
y
Material Modeling
The small strain tensor is the symmetric part of the displacement gradient:

( ) ( )
( )
0 0
0 0
1
2
1
2
2
T
T
x x x x
E
x x
E F F I

| |
| |
| = +
|
|
\ .
\ .
= +
r r r r
t
r r
t t t t

The small strain tensor is the classical total strain measure. It is however useful only for linear
small displacement analysis since it does not vanish under a rigid body rotation.

Indeed:

( ) ( )
0
0
_ _ :
_
_
1 1
_ 2 2 0
2 2
T T
if
x Rx
x
F R
x
E F F I R R I
=
= =
= + = +
t
r r
r
t t
r
t t t t t t t t

If this strain measure would be used in a large rotation context, a rigid body motion would cause
strains, thus stresses and the body would start to expand or compress.

Consequently the small strain tensor is not used at all in LS-DYNA which is designed to treat
large displacement problems.

small strain for a 2D rotation:

y
1
x
2
Material Modeling
1
2
1
2
2 1
2 1
0
cos
0 0
0
cos 1
0 0
xx
u
u l l
x
x l
u u
E
x x
=
=
=
=
= =

Most formulations in LS-DYNA are incremental, rather then total and are based upon the strain
rate or rate of deformation tensor that is defined as the symmetric part of the velocity gradient:

( )
1
2
T
L L = +
t t
t
&

1 1
2 2
1 1
2 2
1 1
2 2
x x y x z
x y x z x
x y y z y
y x y y z
x z y z z
z x z y z
| | | |
| |
+ +
| | |

\ .
\ .
|
|
| | | |
| = + +
| |

| \ . \ .
|
| |
| |
|
+ +
| |
|

\ .
\ .
\ .
& & & & &
& & & & & r
&
& & & & &

The strain rate components have a dimension (1/s)

Strain rates clearly vanish in the case of rigid body rotation:

( ) ( ) ( )
_ _ :
_
_ _ :
_
1 1 1
_ 0
2 2 2
T T
if
x x
then
x
L
x
L L
=
= =
= + = + =
t
r r
&
r
& t t
r
t t t t t t t
t
&

The strain rate is thus an objective, incremental measure of strain.

Material Modeling
Strain rate for a 2D rotation:

1
2
1
2
2 1
2 1
0
0
0
0
xx
x
x
x
x l
x x
E
x x
=
=
=
=
= =
&
&
& &

It is important to realize that the time derivative of the small strain tensor and the strain rate are
not equivalent. Actually they are the same only if both rotations and deformations remain small.
(e.g. civil engineering problems).

The latter can be illustrated for the uniaxial case:

0 0
0
xx xx
xx xx
u x x
E
t x x x
x x E

| |
= = =
|
\ .

& &
&
&
&
&

The physical analog for the different strain measures can be obtained from the change in length
of an elementary truss:

y
1
x
2
1
l0
2
Material Modeling
0 0 0
0 0 0 0
0
2 1 2 1
2 1
2 1
ln
2 1
xx
xx xx
u x x x x l l
E
x x x l
x x l dl l
dt dt dt
x x l l l

+
= =
| |
= = = =
|
\ .

&
& &
&

The logarithmic strain tends towards minus infinity if the length of the truss approaches zero:

Strain Rate Invariants

We define the first invariant as the volumetric strain rate:

v xx yy zz
v
x y z
x y z

= + +
= + +
& & & &
& & &
&

For a small material element this can be written as follows:

y
x z
v
x y z
x y z x y z x y z
v
x y z
v
l
l l
l l l
l l l l l l l l l
l l l
V
V
= + +
+ +
=
=
&
& &
&
& & &
&
&
&

l0
l
Material Modeling
The volumetric strain rate thus equals the rate of change of (material) volume over the actual
volume.

Deviatoric Strain Rates

Deviatoric strain rates are defined based upon the volumetric strain:

1
3
1 0 0
1
0 1 0
3
0 0 1
d v
dxx dxy dxz xx xy xz
dyx dyy dyz yx yy yz v
dzx dzy dzz zx zy zz
I

=
| | | |
| |
| |
|
=
| |
|
|
| |
\ .
\ . \ .
t
t t
& &
&
& & & & & &
& & & & & & &
& & & & & &

Clearly the trace (sum of diagonal components) is zero for the deviatoric strain rates as it was for
the stress:

( )
( )
0
xx yy zz v
d dxx dyy dzz
tr
tr

= + + =
= + + =
t
&
& & & &
t
&
& & &

The deviatoric strain rate thus has no volumetric component.

Second Invariant of Strain Rate

The effective (or equivalent) strain rate is defined as:

2
:
3
eff d d
=
t t
& &
&

Or:

( )
2 2 2 2 2 2
2
2 2 2 _
3
eff dxx dyy dzz dxy dyz dzx
= + + + + + & & & & & & &

Material Modeling
Note that under uniaxial loading: (in x-direction):

( )
( )
( )
0 0
0 0
0 0
1 2
2 0 0
1 0 1 0
3
0 0 1
2
1
3
xx
xx
xx
v xx
xx
d
eff xx
V
V
v
V
V

| |
|
=
|
|
\ .
=
| |
|
= +
|
|
\ .
= +
&
t
&
&
&
& &
& t
&
& &

Effective Strain Rate Under Uniaxial Loading

Consider an incompressible elastic material:

0.5
eff xx
V = = & &

The consider a cyclic uniaxial loadcase:

Strain rate and effective strain rate are state variables: they can be determined from the
current configuration

,
eff xx
& &
t
Material Modeling
Effective Strain Under Uniaxial Loading

Consider the effective strain in the previous case:

0 0
t t
eff eff xx
dt dt = =

& &

Then consider a cyclic uniaxial loadcase:

The effective strain is a history variable: computation requires knowledge about the entire
loading history

In elasto-plasticity, strain rates will be split in elastic and plastic parts. Each part can then again
be split in volumetric and deviatoric components:

ed
e
ev
pd
p
pv
I
I
+
=
=
+
=
t
&
t
&
t
&
t
&
t
&
t
&
t
&

Metals will usually exhibit a zero volumetric plastic strain component: plastic deformation in
metals occurs at constant volume, this is not the case however in foams.

The second and third strain rate invariants will not often be used, more commonly the second
invariant of the plastic strain rate tensor is defined as follows:

t
,
eff xx

Material Modeling
( )
2 2 2 2 2 2
2
2 2 2 _
3
p e
p pdxx pdyy pdzz pdxy pdyz pdzx
p p
dt

=
= + + + + +
=
t t t
& & &
& & & & & & &
&

The last two equations are the definitions for the equivalent (or effective) plastic strain rate and
equivalent plastic strain respectively.

Note that this definition is valid, also if the plastic strains are not deviatoric.

Clearly the equivalent plastic strain must always be a monotonically increasing function of time.

Elastic and Plastic Poisson Coefficient

The Poisson coefficient relates to elastic deformations only and gives the relationship between
longitudinal and transversal deformation under uniaxial loading:

yye yy yyp
xxe xx xxp

= =
& & &

& & &

Similarly we can introduce a plastic Poisson coefficient based upon plastic deformations only:

yyp
p
xxp
=
&
&

And a total Poisson coefficient based on total deformations (the only one we can measure):

yy
t
xx
=
&
&

Values of effective strain rate and equivalent plastic strain rate under uniaxial loading conditions:

For elastic metals we have that:
0.866
0.3
0.
eff xx
t
p
v v

=
= =
=
& &
&

For plastic metals we have that:

,
0.866 0.3
0.5
xx eff xx
p t p p xx
v
v v

=
= =
& & &
& &

Material Modeling
For plastic metals we also have that:

,
, ,
0.3
0.5
eff xx
p t p p xx xx
xx p xx e xx
v
v v
since

=
= =
>>

So whereas in the plastic regime effective strain and equivalent plastic strain are always nearly
equal, the same is not true for effective strain rate and equivalent plastic strain rate due to the
oscillatory nature of the elastic strain rates.

For foams we have that:

,
2
3
0
2
3
eff xx
p t
p p xx

=
= =
=
& &
& &

And after time integration:

,
2
3
0
2 2
3 3
eff xx
p t
xx
p p xx xx
E

=
= =
| |
= =
|
\ .

Remarks:

There is some arbitrariness in the definition of the invariants:

- the first stress invariant is positive in compression rather than in tension

- the second stress invariant (von Mises stress) was chosen since it corresponds to the plasticity
limit in uniaxial tension for steel and most metals

- the second invariant of the plastic strain rate is defined so that its product with the von Mises
stress yields a rate of plastic deformation energy

p vm p
e = & &

Material Modeling
4.2.3 Isotropic Hypoelastic Material

Most explicit codes implement a hypoelastic law as a linear relationship between stress rate and
strain rate, expressed in some corotational reference system.

The deviatoric stress rate will depend upon deviatoric deformation rates only and similarly,
volumetric strain rate determines the pressure:

0 0
2
ln ln
d
v
s G
V
p K K
V
V
p K K
V
=
= =
= =
t t
& &
&
& &

Hypoelasticity is equivalent to linear elasticity if both the rotations AND the strains are small

Hypoelasticity in terms of Youngs modulus and Poissons ratio becomes:

( ) ( )
_
1 1 2
v
E
I

(
= +
(
+

t
t t
& &
&

With:
( )
( )
2 1
3 1 2
E
G
E
K
=
+
=

If the Poisson ratio equals zero we obtain:

_ E =
t t
& &

and every stress component is proportional to the corresponding strain component: the individual
stress components are uncoupled.

Foams usually exhibit very small or zero Poisson effects in compression.
Material Modeling
Material law 1 in LS-DYNA (MAT_ELASTIC) corresponds to this hypoelastic formulation

Numerically the algorithm is trivial and central thus second order accurate:

New stresses must be calculated from old stresses and strain rates at the half timepoint

Numerical Integration of Hypoelastic Law

First split in deviatoric and volumetric components:

( )
, , ,
, 1/ 2 , 1/ 2 , 1/ 2 , 1/ 2
1/ 2
, 1/ 2 1/ 2 , 1/ 2
1
3
1
3
n xx n yy n zz n
n
n n n
v n xx n yy n zz n
n
d n n v n
p
s p I
I

+ + + +
+
+ + +

= + +
= +
= + +
t
t
t t
& & & &
t
&
t
t t
& &
&

Then integrate the material law:
1
, 1/ 2
1
v
n n
v n
n n v
p K
p p
K
t
p p K
+
+
+
=
=
& &
&

And for the deviatoric part:
1
, 1/ 2
1
2
2
2
d
n n
d n
n n d
s G
s s
G
t
s s G
+
+
+
=
= +
t t
& &
t t
t
&
t t t

And assemble the new stresses:

1 1 1 n n n
s p I
+ + +
=
t
t t

t(n) t(n+1/2)
t(n+1)
n
t

1/ 2 n
+
t
&
1 n
+
t

t
Material Modeling
Note that:

3
:
2
2
: 3
3
2
vm
eff d d vm eff
d
s s
E E E GE
s GE
= =
`
)
t t
t t
t
t

Under uniaxial loading:

4.2.4 Elasto-Plasticity for Metals

We will treat the case of perfect plasticity: the von Mises condition then is:

0
vm y


Svm
3G
Svm
E
eff

E
E
xx

Material Modeling
Such a material law will be sensitive to strain localization

A full description of isotropic elasto-plastic behavior is given by:

1. Material law based on introduction of plastic strain rates and hypoelasticity:

( ) ( )
2
d dp v vp
G K I + = +
t
t t t
& & &
& &

taking symmetry into account, this is a system of 6 equations with 12 unknowns: 6 stresses and 6
plastic strain rates.

2. A flow rule that makes an assumption about the plastic strain rates: a good description of
metals is obtained if we assume the plastic strain rate to be proportional to the deviatoric
stress:
( )
2 2 2 2 2 2
3
2
2
:
3
2
2 2 2
3
p p
vm
p dp dp
p dpxx dpyyx dpzz dpxy dpyzx dpzx
s

=
=
= + + + + +
t t
&
&
t t
& &
&
& & & & & & &

This reduces the number of unknowns in the plastic strains from 6 to 1; knowledge of the
equivalent plastic strain rate will now be sufficient to solve the problem.

Since metallic plastic strains are deviatoric, the equivalent plastic strain rate is best interpreted as
the magnitude of the plastic strain rate vector.

Consequences of the Flow Rule Choice

The principal directions of plastic strain rate, deviatoric stresses and stresses all coincide.

Since plastic strain rate and deviatoric stress are proportional, the trace of the plastic strain rate is
zero and the plastic strain rate is equal to the deviatoric plastic strain rate:

0
p
pxx pyy pzz xx yy zz
p pd
s
s s s
+ + + + =
t t
&
& & &
t t
& &

Consequently the equivalent plastic strain rate can be based upon the total plastic strains, and the
volumetric plastic strain rate is zero:

Material Modeling
0.
2
:
3
vp
p p p
=
&
t t
& &
&

Plastic deformation does occur at constant volume (confirmed experimentally by P.`Bridgman).

Reformulation of the Material Law

The flow rule allows rewriting the material law as follows:

( ) ( )
( )
2 2
_ :
2
2 2
3
d dp d p
v vp
d p
vm
v
s G G
p K
becomes
s G G s
p K

= =
=
=
=
t t t t t
& & & & &
& & &
t t t
& &
&
& &

Thus:

The pressure only depends upon the volumetric strain (density) and is not affected by the
plasticity. Thus a nonlinear EOS relating pressure and density can be used without adding
complexity.

The deviatoric stresses are obtained from an elastic increment minus a stress reduction that is
applied parallel to the deviatoric stress vector. To determine the stress reduction we need to
know the equivalent plastic strain.

3. Loading and unloading conditions:

Define a yield surface described by a function f of the stress components and some material
parameters, in the case of a von Mises yield condition:

( )
, 0.
y vm y
f =
t

Material Modeling

The state of stress is elastic iff:
The state of stress is below the yield surface
The equivalent plastic strain remains constant
Thus the equivalent plastic strain rate is zero

The state of stress is plastic iff:
The state of stress is on the yield surface
The equivalent plastic strain increases
Equivalent plastic strain rate is > zero and can be determined from the equation f=0

Other Yield Surfaces

The Drucker-Prager criterion can correspond to a parabola or an ellipse in the invariant plane:

( )
2
2
0 1 2
, 0.
3
vm
i
f A A A p A p
=
t

p
vonMises
_
vm

_
vm

Drucker-Prager
p
Material Modeling
These formulations correspond to material laws 5 and 75 in LS-DYNA

Solving the Elasto-Plastic Problem Numerically

Numerically, the elasto-plastic problem can be stated as follows: find the stresses at timepoint
(n+1) from the known stresses at timepoint (n) and the known total strain rates at (n+1/2).

Thus:

1 1
2
_ ,
n n
n

+
+
t t t
&

In order to achieve this, we will also have to calculate the new values of the hardening
parameters (yield stress) and the equivalent plastic strain from the old values:

( ) ( 1) 1
, ,
yn pn p n y n

+ +

The usual way to solve this problem in explicit codes is by using the radial return algorithm.

This extremely simple and accurate algorithm (Wilkins, 1964) is one of the main reasons for the
efficiency and succes of ecplicit codes.

The Radial Return Algorithm

STEP 1: Calculate elastic trial stresses:

Define the incremental strain:

1
,
2
1
,
2
, 1 ,
d
d n
v
v n
v n v n v
t
t

+
+
+
=
=
= +
t t
&
&

Assume the total strain increment to be elastic:

( )
, 1
, 1
, 1 , 1 , 1
2
_ :
e n n d
e n n v
e n e n v n
s s G
p p K
or
p p
+
+
+ + +
= +
=
=
t t t

(EOS)

Material Modeling
STEP 2: Check if the elastic trial stress fulfills the yield condition:

( )
, 1
, , 1 , 1 , 1
, , 1 ,
, 0.
3
_ :
2
_ 0?
e n yn
vm e n e n e n
vm e n y n
f
s s

+
+ + +
+
=

t
t t

If this condition is fulfilled, the process is elastic and the numerical treatment ends here.

If not, the stress must be returned to the yield surface.

STEP 3: if the process is plastic, return stress to the yield surface:

The radial return algorithm consists in discretizing the material law for the deviatoric stresses
and using the trial elastic stress to estimate the direction of the plastic strain rate:

, 1 1
1 , 1
,
, , 1 2
1 , 1
, , 1
3
2 2
2
3
2
2
3
1 2
2
d p
vm
e n n n n
e n
p n
e vm n
n e n p
vm e n
s G G s
s s s s
G s
t t
s s G

+ +
+
+
+
+ +
+
=

=

| |
=
|
|
\ .
t t t
& &
&
t t t t
t
&
t t

We thus obtain that the new elasto-plastic stress is obtained by multiplying the elastic trial stress
with a scalar function of the equivalent plastic strain increment.

Since we do not alter the pressure we obtain:

1 , 1
1 , 1
, 1 , , 1
_
_
_
_ 1.
n e n
n e n
vm n vm e n
p p
s ks
k
k

+ +
+ +
+ +
=
=
=
t t

Material Modeling
We determine k directly from the yield condition at time n+1 if the material is perfectly plastic:

, 1 , 1
, , 1 ,
,
, , 1
_ 0
_ 0
_
vm n y n
vm e n y n
y n
vm e n
k
k

+ +
+
+
=
=
=

The deviatoric stresses are scaled by the yield stress over the new trial elastic von Mises stress.

The radial return is radial in deviatoric space:

The resulting stress return path is parallel to the deviatoric stress vector as suggested by the
constitutive law. This intuitively explains the high accuracy of the radial return algorithm.

A representation in the invariant plane clearly shows how the pressure remains unaffected by the
plasticity:

Sn+1
s2
Se,n+1
s3
, , 1 vm e n
R
+
=
, 1 y n
R
+
=
s1
_
vm

, , 1 vm e n
+

, 1 vm n
+

, 1 e n
+
t

1 n
+
t

p
1 , 1 n e n
p p
+ +
=
Material Modeling
The stress projection depends upon the flow rule only and is independent of the choice of yield
surface. Material laws 10, 3, 12, 24 and many others are all based on the same metallic flow
rule.

Algorithmic Setup

The algorithmic setup illustrates the simplicity of radial return:

( )
, 1 1 , 1/ 2
1 1 , 1
?
, , 1 , 1 , 1
1 , 1 1
1 , 1 1
, , 1
2
3
:
2
_ _ :
_ _ :
e n n d n
n n v n
vm e n e n e n y
n e n n
y
n e n n
vm e n
s s G t
p p
s s
IF YES
s p I
IF NO
s p I
+ + +
+ + +
+ + +
+ + +
+ + +
+
= +
=
=
=
=
t t t
&
t t
t
t t
t
t t

4.2.5 Elasto-Plasticity with Strain Hardening

The von Mises condition becomes:

( )
0
vm y p


Consider linear hardening first

Such a material law will be much less sensitive to strain localization

Material Modeling
To describe the material behavior we need additionally:

4. A hardening rule allowing calculation of the evolution in time of the hardening parameters,
Perfect plasticity is obtained if the hardening parameters do not depend upon the plastic strain.

For metals, with a von Mises yield condition however the yield stress is a function of the
equivalent plastic strain and a hardening rule is necessary:

( )
, 1 , , 1/ 2
_
_
_
y y p
y
y p p
p
y n y n p n
H
H t

+ +
=
= =
= +
& & &
&

If the hardening rule is assumed linear in the equivalent plastic strain rate, all numerical
algorithms simplify considerably.

Simple elasto-plastic material laws with linear hardening are available in LS-DYNA as:

MAT_PLASTIC_KINEMATIC (law 3)
For brick and shell elements, kinematic hardening (Bauschinger effect) is optional

MAT_ISOTROPIC_ELASTIC_PLASTIC (law 12)
For brick elements only, no plane stress condition is enforced for shells

The radial return algorithm applies as before.

The same scaling as for the individual deviatoric stress components must obviously apply for the
von Mises stress:

, 1 1 1
2
, 1 , 1 , 1
, , 1
, 1 , , 1
, , 1
, 1 , , 1
3
:
2
3 3
1 2 :
2 2
3
1 2
2
3
vm n n n
vm n p e n e n
vm e n
vm n vm e n p
vm e n
vm n vm e n p
s s
G s s
G
G

+ + +
+ + +
+
+ +
+
+ +
=
| |
=
|
|
\ .
| |
=
|
|
\ .
=
t t
t t

If hardening is present in the metal, the stress scale factor must be calculated from the increment
in equivalent plastic strain.
Material Modeling
The equivalent plastic strain increment is solved from the yield condition at time n+1, for linear
hardening this can be done in a single step:

, 1 , 1
, , 1 ,
, , *1
, , 1 ,
, , 1 ,
_ 0
3
_ 1 0
_ 3 0
_
3
vm n y n
vm e n p y n p
vm e n
vm e n p y n p
vm e n y n
p
G
H
G H
G H

+ +
+
+
+
=
| |
=
|
|
\ .
=
=
+

The last formula allows to determine the incremental plastic strain, which in turn allows to
calculate the stress scale factor k from the yield stress value at t(n+1).

Clearly nothing changes in principle, the stress scale factor must be based on the updated value
of the yields stress:

, 1
, , 1
_
y n
vm e n
k

+
+
=

Material Modeling
Algorithmic Setup

The algorithmic setup illustrates the simplicity of radial return:

( )
, 1 1 , 1/ 2
1 1 , 1
, , 1 , 1 , 1 ,
1 , 1 1
, , 1 ,
, 1 ,
, 1
1 , 1 1
, , 1
2
3
: ?
2
_ _ :
_ _ :
3
e n n d n
n n v n
vm e n e n e n y n
n e n n
vm e n y n
p
y n y n p
y n
n e n n
vm e n
s s G t
p p
s s
IF YES
s p I
IF NO
G H
H
s p I
+ + +
+ + +
+ + +
+ + +
+
+
+
+ + +
+
= +
=
=
=
=
+
= +
=
t t t
&
t t
t
t t
t
t t

Plastic Flow in Shells and Thickness Change

The plastic flow corresponds to a discontinuity in the Poisson effect of the material (0.3 in the
elastic region and 0.5 in the plastic region).

Plastic deformation occurs at constant volume.

Plane stress plasticity must be solved for iteratively since the through-the-thickness strain
component is not known from the shell kinematics:

Material Modeling
A first elastic estimate for incremental strains is made:

( )
1
xx xx
yy yy
xy xy
yz yz
zx zx
zz xx yy
t
t
t
t
t

=
=
=
=
=
| |
= +
|
\ .
&
&
&
&
&

The apply radial return and check the stress component in the through-the-thickness direction:

, 1
1
, , 1 , 1 , 1
, 1
1 , 1
, , 1
, 1
, 1 ,
, , 1
2
3
:
2
2 0?
e d e n n
v n n
vm e n e n e n
y n
n e n
vm e n
y n
zz n zz d zz v
vm e n
s G s s
p K p p
s s
s s
G K
+
+
+ + +
+
+ +
+
+
+
+
= =
= =
=
=
= = + =
t t t t
t t
t t

If the state-of-stress is plastic, the plane stress condition will not be fulfilled and we calculate a
second point with:

( )
zz xx yy
= +

(Assume a fully plastic increment).

The true z-strain increment is then determined by secant iteration:

1
1 1 1
1
i i
i i i zz zz
zz zz zz i i
zz zz

=

i-1 i+1 i
Material Modeling
3 options are possible:

Radial return
3 iterations
Full iterative plasticity

(set on CONTROL_SHELL)

This allows correct thickness updates of the shell if required on the CONTROL_SHELL card.
(ISTUPD=1)

/ 2
/ 2
t
zz
t
t dt

Influence on element membrane and bending stiffness is usually negligible in crashworthiness
applications. If this option is activated, bulk viscosity should be added to shell elements in order
to ensure numerical stability.

Material Modeling
4.3 Material Laws for 2D Elements in LS-DYNA

4.3.1 Simulation of Mild Steel in LS-DYNA

Numerical simulation of thin sheets that can be:

Stamped
Fabricated
Extruded
Hydroformed

These sheets can be made out of:

Mild steel
High strength steel
Fast hardening steel (Trip, DP, Boronsteel...)
Aluminum
(Magnesium)

For stamped, extruded and hydroformed parts the manufacturing process influences the material
data, this becomes more important as more hardening occurs (strain hardening + bake hardening)

Features needed for the simulation of thin steel sheets in crashworthiness applications:

Small deformations
Elasto-plastic material
Isotropic hardening
Plane stress (iterative plasticity)
Ductile rupture
Strain rate effects

Other common features of steel behavior are not treated in these notes on crashworthiness:

Kinematic hardening (Krieg & Key)
Anisotropic behavior (Hill, Barlat...)
Temperature dependency (Johnson-Cook...)

Material Modeling
Small deformation elasto-plasticity with isotropic hardening is used in practice:

1
2=3
4=5
6
p

5
4
2
3
6
1
xx

xx

Material Modeling
Following material laws are available in principle: (exclude temperature effects and anisotropy)

law hardening Rate effect notes
3 Plastic
Kinematic/
isotropic
linear CS
12 Isotropic
Elastic/plastic
linear CS Bricks only
18 Power law
plasticity
Power law CS
19 Rate dependent
plasticity
linear 3 Load curves For
Plastic or
glass
24 Piecewise linear
Isotropic
plasticity
Load curve
linear
CS
Tabulated
curves
Very efficient
123

Plasticity with rupture As 24
rupture
81 Plasticity with
damage
As 24
damage
103 Visco-plasticity Sum of
Exponential
terms
VP
104 As 103
rupture
105 Visco-plasticity with
damage

As 103
damage

Determination of Elasto-Plastic Material Properties

Perform a tensile test:

l0
A0
f
f
Material Modeling
And convert the force-displacement curve in an engineering stress/strain curve:

The engineering stress/strain curve must then be converted in a relationship between Cauchy
stress and true strain:
( )
( )
0
0 0
0
0 0
0 0
0 0
ln ln 1
ln 1
1
_ : _
l l l
l l
f f l
A A l
since Al A l
| | | |
= = +
| |
\ . \ .
= +

`
= + = =

=
)

In deriving this we used the fact that plastic flow happens at constant volume and elastic
deformations are very small compared to total deformations

The resulting curve has a different shape:

f
d=l-l0
f/A0
(l-l0)/l0=d/l0
xx

xx

Material Modeling
To derive the material property we split the strain in elastic and plastic parts, the elastic strain is
the part that recovers upon unloading:

xx pxx exx
xx
pxx xx
E

= +
=

And we add the loading and unloading conditions of plasticity in the uniaxial case:

0
0
pxx xx y
pxx xx y

> =
= <
&
&

This allows plotting the yield stress as a function of plastic strain

Shift and shrink for the elastic strains:

A 3D generalization is formulated in terms of von Mises stress and equivalent plastic strain:

2
3
3
2
p dpij dpij pxx
uniaxial
vm ij ij xx y
uniaxial
dt dt
s s

=
=
=
=
& & &

xx

xx

Material Modeling
This results in the hardening curve that gives yield stress as a function of equivalent plastic
strain:

The hardening curve can be inputted in LS-DYNA in different ways

Piecewise linear (24-81-103-105-123)

Typical properties of hardening curves for steel in input data:

Monotonically increasing yield stress
Monotonically decreasing tangent
Constant plastic strain increment
Sufficient number of datapoints (deps=0.01)
Smooth stress curve
Continuous derivative

Bilinear with tangent modulus (or plastic hardening modulus) (3-12-19-28-81-24)

y

p
y

p

Material Modeling
( )
0
/
_ : _
_ : _
y y
p
p y
y p p y
t
p
t
E
E
if E
EE
then E
E E

= =

= +
=

Power law (15-18-):
( )
( )
1/
,
1
1 ln
p
n
eff
y e eff p
eff n
y p
ref
k
c
a b c
| |
| |
= + + |
|
|
\ .
\ .
| | | |
= + + | |
| |
\ .
\ .
&
&
&

In combination with strain rate effects, law 24 gives the more general input capability, this law
has also been optimized performance wise

Example of a hardening curve for a mild steel derived from a true stress-true strain curve by
means of Excel:

Irregularities should be smoothed in practice
y

p

Et
E
y

Ep
Material Modeling
Necking and Ductile Rupture

The hardening curve that was determined in this way is reliable only as long as the state of stress
in the test specimen was homogeneous.

This condition is no longer fulfilled as soon as necking occurs. Then the section decreases locally
and strain tends to increase in that region.

If a hardening curve needs to be determined beyond the necking point, this can only be done
iteratively.

First we determine the necking point:

0
0
0
f
d

= = =

The negative slope in the engineering stress is due to a reduction in cross section since the true
stress curve is monotonically increasing.

1 2

2 1
>>
0
0
0

Material Modeling
Determine the necking point on the true strain/true stress curve:

( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( )
( ) ( )
[ ]( )
0 0
0
0 0 0 0 0
0 0 0
0 0
0
0
0
0 0 0
0
0 0
1
ln 1
1 1 1
ln 1 1 ln 1
1
1
1
1 1
1
_ :
1
necking

= +
= +
+ + +
= =
+ + +
+
= +
+
(

= + + +
(
+

= + =

The latter equation is easily plotted:

Stress-strain data beyond this point can be extrapolated (e.g. constant slope of the hardening
curve), the tensile test must be simulated using a sufficiently fine mesh until one obtains the
measured force-displacement values in the necking region

_

necking
, p n

Material Modeling
Hardening curves should be cut at the plastic necking strain, beyond we only know a lower and
upper limit:

y
p
n

, y n

, p n

Material Modeling

Extrapolated hardening data:
Improved estimate
of material data
FE-Model and tensile test simulation
Compare test & simulation results
f
d
Cross section def.
Fixed vel.
Material Modeling
Workshop

Example file: steel.k

3 quasistatic tensile tests on mild steel coupons are run with type 2 and type 4 elements, yield
stress is 180Mpa, use material type 3 with linear hardening

Material Modeling

Hardening is set at 2000Mpa, 2.MPa and 0. MPa, clearly strain localization (necking) occurs
for the lower modulae

Material Modeling
For low values of the tangential modulus, the force drops due to decreasing section, stresses
in the necked zone do not decrease:

Necking is seen to occur immediately: plastic strain at necking is zero

Exercise: What is the influence of the parameters ISTUPD and NITER on the
CONTROL_SHELL cards?
(needed to find force curves from test back)

Material Modeling
Exercise: vary the tangential modulus of the material, from which value on does immediate
localization no longer occur?

Following results were obtained using a single value of the tangent modulus:

Material Modeling
Stresses show hardening in all elements, force is constant since hardening exactly
compensates the reduction in section:

Material Modeling
Generating Material Data for Structural Sheets:
Coarse Meshes: l > 10.mm

Determine stress/strain data up to necking point from test and keep stress value constant for
plastic strains beyond necking:

Define a plastic rupture strain higher then 100% (just to delete very distorted elements that
could go numerically unstable)

Do not consider thinning of the shells (ISTUPD=0)

Check fringe plots of equivalent plastic strain in LSPOST using a range as follows:

,
0
p p n

All red zones in the fringe plots will localize deformation and must be checked for mesh
behavior and structural integrity

y

necking
p

Material Modeling
MAT_3: Necking Behavior

In the plastic region we have:

( )
y y t y
E = +

No immediate necking means that:

n y

The strain at necking follows from:

( )
n n y t n y t
E E
= = + = =

And thus:

180.
t y
n y
t
n y t y
E
E
E MPa
= +
=

Et
y

y

Material Modeling
Typical Hardening Curves

Steel:

Aluminum:

t y
E
t y
E
Material Modeling
Polymers:

Generating Material Data for Structural Sheets:
Fine Meshes: l < 6.mm

Determine stress/strain data iteratively up to rupture point from test, a Krupkowsky-type
exponential extrapolation beyond the necking point is usually a good starting point:

The hardening curve will still be monotonically increasing

t y
E
2
2
0
<

2
2
0
>

y

Necking Rupture
p

Material Modeling

Determine a plastic rupture strain from equivalent plastic strain fringe plots on the simulation
results of the tensile test, check the necked area towards the end of the simulation

This value is still mesh-dependent so the tensile test must be simulated with a mesh similar in
size to the actual model

A rupture strain value obtained in this way will be much larger then the global value
supplied traditionally since these are based on a strain measure for the entire coupon rather
then just the necked region
Activate thinning of the shells (ISTUPD=1) and add bulk viscosity for shells for numerical
stability

For a given mesh density, this model could predictively simulate ductile rupture, with
material law 24 only equivalent plastic strain can be used as rupture criterion, so rupture
occurs in tension and compression

CONTROL_SHELL istupd=1
CONTROL_BULK_VISCOSITY type=-1

lr=l0+dr
0
ln
r
pr
l
l
>
Material Modeling
Rupture Criteria for Thin Sheets

Using material law 123 allows using the maximum principal strain as a rupture criterion,
compare however with a real rupture criterion such as FLD for metal forming:

The validity of FLD as a rupture criterion is however also limited to metal forming type
deformation histories

Rupture will now occur in tension only

To compare both criteria calculate equivalent plastic strain as a function of principal strains
under biaxial loading with a tensile major principal strain:

( )
( )
1 1,
2 2, 1,
3, 1,
2 2
1,
1
4
1
3
p
p p
p p
p p
a
a
a a

=
= +
= + +

Remark that under uniaxial and biaxial loading:

1 2 2, 1, 1,
1 2 1,
0.5
2
p p p p
p p

> = =
= =

Thinning can also be used as a rupture criterion

1 2
>
2

Material Modeling
Rupture then occurs gradually through the shell thickness: stresses are set to zero if the
rupture strain is reached in the integration point, the element is eliminated after all integration
points have failed

Strain Rate Effects

Rate effects can be accounted for using the Cowper-Symonds formula (most laws), or using a
power law (64) or a function for the yield stress as function of the strain rate (19).

The Cowper-Symonds model calculates a dynamic yield stress by scaling the static value:

1/
1
p
eff
y ys
c

| |
| |
= + |
|
|
\ .
\ .
&

C is a reference effective strain rate for which the yield stress doubles.

c p
steel 40/s 5
aluminum 6500/s 4

This approach usually causes spurious noise in the numerical solution.

The effective strain rate is oscillatory as a function of time due to the elastic component:

,
eff p
& &
t
Material Modeling
The difference between effective strain and equivalent plastic strain is small:

A consistent visco-plastic approach prevents spurious noise.

Remember:

2
:
3
2 2
: :
3 3
eff d d
p dp dp p p

=
= =
t t
& &
&
t t t t
& & & &
&

The effective strain rate is based on the total deviatoric strain rates and easily calculated,
however it is the elastic part of this quantity that will cause spurious oscillations if used in rate
dependent plasticity algorithms.

The equivalent plastic strain rate is a proper parameter for visco-plasticity, but this must be
solved for in a consistent way. The first implementation of such an algorithm was in the
(anisotropic) law 103. (see Berstad/Hopperstad)

All strain rate dependent materials in LS-DYNA are based on:

, 1/ 2
p
p n
t
&

if the parameter VP is set to 1 in the material cards.

,
eff p

t
Material Modeling
Use the linear version of material law 19 as an example:

( )
y yo efff p p
y yo eff p p
E
E

= +
= + +
&
&

The radial return algorithm gives:

, , 1 ,
, 1 ,
, 1 , 1
3
vm e n y n
p
p
p n p n p
y n yo eff p p n
G E
E

+
+
+ +
=
+
= +
= + + &

The corresponding visco-plastic formulation is:

( )
y yo p p p
y yo p p p
E
E

= +
= + +
&
&

The radial return algorithm becomes:

( )
, 1 , , 1
, 1 ,
, , 1 ,
, 1 ,
, 1/ 2
, 1 , 1/ 2 , 1
3
3
vm n vm e n p
p
y n yo p p n p
vm e n y n
p
p
p n p n p
p
p n
y n yo p n p p n
G
E
t
G E
t
t
E

+ +
+
+
+
+
+ + +
=
= + + +
=
+ +
= +
= + +
&
&

This will avoid all spurious oscillations.

The best generality is obtained by inputting a table of hardening curves valid for different strain
rates (24-81).

Stress-strain curves are defined for each strain rate. (DEFINE_TABLE). Table lookup is very
fast.

Material Modeling
Consistent visco-plastic treatment again prevents spurious oscillations.

( )
( ) ( )
, 1 ,
,
y y p p
y ys p yd p
ys p
y n ys n p yd
p
t

+
=
= +

| |
= + +
|

\ .
&
&

(solved iteratively)

Tests can be performed as follows:

Rate range methodology
0.-1. Hydraulic
10-500 Impact (EMI)
>1000 Hopkinson bar

y

p

Material Modeling

EMI Setup:

All tests at high strain rates are performed at a constant nominal strain rate (=constant impact
velocity) and not a constant strain rate proper.

In the small strain region (below 30%) this is not a very big problem:

0
x x
l l
=
& &
&

projectile
fixation
probe
strain gauge
rigid structure
Material Modeling
Workshop

Example steelrate.k

3 dynamic tensile tests on mild steel coupons at an average strain rate of 100/s or 0.1/ms

Material Modeling
Hardening data are such that no necking occurs:

Identical material data are used, second and third coupon have VP=1 and first coupon has VP=0

Compare the following Cowper-Symonds coefficients with VP=1 and VP=0:

p C
steelrate 1 0.2
steelrate1 1 0.02
steelrate2 5 0.04

Material Modeling

With P=5, slight oscillations are generated

With P=1, (= linear rate effect) and c=0.02 oscillations lead to divergence:

Oscillations in the effective stress time history are prevented by setting VP=1

Note that p=1 and c=0.2 adds 50% to the yield stress for a strain rate of 100/s (=0.1/ms)

Material Modeling
Why is Steel Easy?

Necking comes at relatively high plastic strain, no early localization

Constant volume during plastic deformation

Full 3D response can be extrapolated from the uniaxial tensile test result

Plastic stresses propagate much slower then elastic (linear) stresses

Why is Steel Difficult?

Anisotropic

Kinematic/isotropic hardening

Influence of the forming/thermal treatment process

Nonlinear elastic property (high strength steel)

Rupture criteria

4.3.2 Simulation of Cast Iron, Al, and Mg

The main issue is the simulation of castings (important for engine mounts and other structural
parts)

Although usually modeled by brick elements we treat all metallic parts here

Added feature compared to simulation of mild steel is brittle rupture.

Non-localized (smeared) failure algorithms must be used, material law 96 could be used but
almost no experience is available

4.3.3 Simulation of Woodstock and Lignotok

These materials are best modeled as a composite of short wood fibers in a matrix of glue.

Currently material 55 offers the best possibilities.

Lignotok shows a linear force-displacement behavior in tension as well as in bending followed
by rupture.
Material Modeling

The specific problem is that the modulus of elasticity determined from the bending test can be
much larger then the modulus determined from the tensile test:

3
3 3
192 16
t
t
b b
b
b t
N E tb
k
d l
T E I E t b
k
d l l
E E
= =
= = =
>>

For a test piece of length l, width b and thickness t.

This behavior is not consistent with the theory of an elastic continuum (or any continuum) but
for the application both tensile and bending resistance are important and must be modeled.

One possibility is to model the trimboard by a shell with the correct thickness and the tensile
elastic modulus.

The user-defined integration rule can then be used, positioning integration points outside the
shell upper and lower surface to account for the larger bending stiffness.

The user-defined integration points must be positioned as in a shell with the following thickness:

3
b b
t
t E
t E
=

In order to obtain the correct bending stiffness. The attributed areas must obviously sum up to
the real thickness of the shell.

A reduction in timestep will usually be necessary to allow a stable treatment of shells with
artificially increased bending stiffness.

Additionally the material model 55 allows definition of the failure stresses in tension and
compression that can be determined from the corresponding tests as follows:

N T
d
tensile test
d
bending test
Material Modeling
1 2
2 2
t
t t
c
c c
N
bt
N
bt

= =
= =

The material is thus isotropic in the sense that the behavior is the same in both principal
directions but tensile and compressive failure may and usually will happen at different stress
levels.

Finally, failure values must be inputted for the in-plane shear stress. This is the hard part since
usually no shear test is available. (They are practically very difficult to perform).

4.3.4 Simulation of Thermoplastics in LS-DYNA

Thermoplastics gain importance in crash analysis:
Bumper facia
Trimpanels
Rib reinforcements
Ventilation ducts
etc...

All are 2D (plane stress) components, dynamically loaded up to rupture

Main material variants:
PP
PC
ABS
PC/ABS blend
etc..

Polymers do not behave like metals:

E is not constant but
E is a function of strain rate (viscosity)
E is a function of plastic strain (damage)
Volumetric plastic strain is non-zero (plastic flow is non-associated)
Necking occurs very early
Necked zone tends to stabilize and redistribute straining (strain softening followed by
strain hardening)
Plastic hardening is different in compression, tension and shear
Yield curve and rupture strain can be very rate and temperature dependent

Currently no material law available that allows to simulate all these phenomena
Successive approximations with LS-DYNA are offered:
Material Modeling
MAT_24
MAT_19
MAT_81
MAT_89
MAT_101

Simulation of Thermoplastics as Elasto-plastic Materials

The classical way

MAT_PIECEWISE_LINEAR_PLASTICITY (MAT_24)

Based on results of tensile tests (quasistatic and maybe dynamic)

Assumes plastic flow at constant volume

No damage

No viscosity in the elastic region

Isotropic hardening: same in shear and compression as in tension

The engineering stress-strain curve must be converted in true stress-true strain:

0

2
2
0
>

0

Material Modeling
This should be done iteratively due to the occurence of necking at low strain

The increase of the slope in the stress-strain curve at high strain causes the necked region to
stabilize:

The necked region does not stay localized but spreads over a high number of element rows

Rate Dependency of the E-modulus

For plastics with low rupture strain, the viscosity in the elastic region dominates the behavior

A first approximation is offered with MAT_19 which considers a rate-dependent E:



No damage

viscosity in the elastic region approximated by rate dependent E


Material Modeling
Linear hardening: no stabilisation in the necked region (OK for small rupture strain)

Typical behavior with strain rate:
0
0
0
0
y
t
r
E
E
>
>
>
>
&
&
&
&

Rate Dependency of the Rupture Strain

Plastics do not show a clear yield point and a highly rate dependent rupture strain

These features are accomodated in MAT_89



No damage



Hardening curve internally derived from true stress-true strain curves

Material Modeling

0
max , E E

| |
=
|
\ .

Strain rate values are prefiltered

Can be considered a generalisation of MAT_19

The Ultimate Material Law for Plastics

Required are:

Visco-elasticity
Damage
Temperature dependency
Visco-plasticity
Non-associated flow
Anisotropic hardening

Some features seem contained in the proprietary MAT_101 from GE-Plastics

( )
1
ln
r
f = &
e p
= +
Material Modeling
Damage Leading Up to Rupture

MAT_81



Damage considered



Difficult to determine true stress-true strain curve taking damage AND necking into
account

Recommended if strain softening occurs in the true stress-true strain curve

Simulation of Polypropylene Panels

Certain trim panel plastics (PP) show the following qualitative tensile response:

Initial elastic and elasto-plastic behavior with hardening
Softening occurs very early
Rupture at high strain (6100%)
Reduction of the modulus upon unloading
Rupture strain reduces under cyclic loading
Hysteresis indicates viscosity

Material Modeling
From version ls950 on, material law 81 is a combined elasto-plastic model with isotropic
damage, additionally material 105 offers the isotropic damage model according to Lemaitre

Damage models can simulate some of the phenomena observed in PP-plastics

Damage models are useful iff:

The material ruptures
Strain softening occurs in terms of true stress
No visible necking
Elastic modulus is reduced with deformation

In modeling of plastics the viscous effects are still neglected

Also the plasticity assumes constant volume flow which is not the case for plastics,
combination of damage with a generalized plasticity (non-associated) may be necessary in
the future

Principle of Damaged Models

The elasto-plastic stress is assumed to work on a section that was reduced by microcracks
and results in an apparent (damaged) stress working on the full section:

( )
1
1
0 1
d
d
f
A
f
d
A d
d
=
`

=

)
<

Consequently the apparent Youngs modulus is reduced with respect to the Youngs modulus
of the undamaged material in the same way:

When the section has ruptured (d=1) the apparent stress and modulus become zero

A
A(1-d)
Material Modeling
The test result in terms of damage stress/strain could look like this:

Failure strain is where softening is first observed (d=0), and rupture strain is where the
damage has consumed the entire section (d=1)

We can only measure damage stress and total strain

One possibility to simulate rupture is to assume that a macrocrack occurs for a critical value
of the damage parameter smaller then 1:

In the default model the damage is assumed to be a linear function of effective plastic strain

d

f

r

d
Dc
pf

pr

p

Material Modeling
An elasto-plastic hardening rule (yield stress as a function of equivalent plastic strain)
applied to the reduced section in the damaged area, allows to recover the measured
(damaged) stress-strain curve

Data Preparation for LS-DYNA Material 81

The reverse problem must be solved: from the measured curve giving damage stress as a
function of total strain, derive the true stress/strain curve of the material

The hardening rule must be derived by solving for the stress and the plastic strain from
measured pairs of strain and apparent stress:

d
p

| | | |

| |
\ . \ .

Damage evolution can be assumed linear or tabulated in the effective plastic strain

In the general case of nonlinear damage this problem cannot be solved by hand

Here we assume a linear damage evolution

In every point of the stress-strain curve we have:

( )
1
1
p
d d
p pf
c
pr pf
E
d
D

=
= =

( ) 1
d
d =
r

f

Material Modeling
At failure we can calculate the plastic failure strain:

f df
f
pf f
E

=
=

(All variables on the right hand side can be measured)

At rupture we calculate the plastic rupture strain based on our estimate of Dc, the damage at
rupture:

1
dr
r
c
r
pr r
D
E
=

The true stress/equivalent plastic strain pairs can then be obtained by solving a quadratic
equation for the true stress in every point:

( ) ( )
2
0
c
pr pf c c pf d pr pf
p
D
D D
E
E

+ + =
=

This results in a curve for yield stress as a function of plastic strain:

No more softening should occur; reduction in section causes the apparent softening

y

pf

pr

p

Material Modeling
Nonlinear Damage Evolution

A user-defined curve for d as a function of equivalent plastic strain can be defined in material
81

Lemaitre damage is used in material 105

This complicates the determination of the true hardening curve considerably, LSOPT has
been used to achieve optimal correlation with experimental data

Comparison of Linear and Lemaitre Damage Models

The linear model is simply:

p pf
c
pr pf
c
p
pr pf
d D
D
d

&
&

And the Lemaitre model gives damage evolution as:

The idea behind Lemaitre damage evolution is to make damage proportional to the elastic
energy, illustrate this for the simple case of a zero Poisson coefficient:

d
1
Dc
pf

pr

p

Material Modeling
( )
( )
( )
( )
2
2
2
2
2 2
2
(1 )
:
1
:
2 1
2
1
: :
2 1
1 2
3
3
2 1
e
e e
vm
d E
E
Y
Y E
E d
Y s s p I I
E d
Y p
E d

=
`
=

)
= +
| |
= +
|
\ .
t t
t t
t t
t t
t t

Which is easily generalized

Material Modeling
4.4 Material Models in LS-DYNA for 3D Elements

4.4.1 Material Models for Recoverable Foams

Non-linear elastic material laws
Zero Poisson effect
Rate dependency
Rupture in tension & shear
Large deformations

Foam Materials in the Automotive Industry:
General Description of Foams

Foams are not continua, they are open- or closed cell structures

Mechanical properties depend upon:

Geometric structure of the foam (size and shape of the cells)
Intrinsic properties of the cell wall material

Deformation mechanisms include:

Cell wall bending
Elastic buckling
Plastification and rupture
Effect of the contained fluid (closed cell)
Cell wall collapse and contact
Microscopic behavior seems extremely complex

Response can vary from glassy (thermoset) to soft-flexible (elastomer) depending upon
loading rate and temperature

Conclusion

Physical description of foam material seems out of range at this time

The chosen approach is to consider the foam as a continuum and develop material laws able
to reproduce the foams macroscopic behavior

Geometric aspects will be related to the foams density, this is often expressed in terms of
relative density (density of the foam divided by the density of the solid it is made of) which
varies typically between 0.002 and 0.05.
Material Modeling

Consequently volumetric strain and volumetric strain rate are intuitively seen to be important
parameters for determining the foams properties

Application of Foams

Foams are used for energy-absorbers and comfort enhancers in automotive applications
because of their:

Lightness
Low stiffness
Low strength
Large compressive strains

Examples are abundant:

European side-impact barrier was made of crushable PU foam
Door and pillar paddings
Dummy components (ex.: confor foam on EUROSID ribs)
Seat cushions
Bumpers
Covers of penduli simulating pedestrian impact
etc...

Seat foams

Padding and Bumper foams

Structural foams

Density in g/l

20-60

40-200

200-1000

Base
Material

PU

PU
confor
Polymers
(EPF)

PU

PU
Polymers
Metals (Al,Mg...)

Unloading
Behavior

recoverable

recoverable

crushable

crushable

Material Modeling

Foam material modeling

Low density <200g/l
Poisson Ratio=0

High density >200g/l
Poisson Ratio > 0

Elastic

Crushable

Crushable
Elastic

Rate-
dependent-
elastic

Visco-
elastic
Isotropic Anisotropic

Trans-versely
Anisotropic

57-83

73-62

53-63-75 26-126-142
(v960-1106)
53-63-75
Seatfoams

Confor-
foams

Bumper
foams

Padding
foams
(PU)

Strandfoam
S1=3s2

Aluminum
Honeycomb
S1>20s2

Structural foams

Alufoam

PU-based
Epoxy-based

Quasistatic Macro-Mechanical Behavior
of Foams Compared to Metals

1. Uniaxial Compression

stress
metal
foam
strain
Material Modeling
Metals show an elasto-plastic behavior, in the plastic phase the metal deforms at constant
volume (Poisson coefficient = 0.5)

Foams show elastic, plastic and densification phases in compression with a Poisson
coeffcient close to 0.

Summarize experimental observations on the foams I uniaxial compression:

Stress-Strain response:

High porosity causes low Poisson coefficient (small or even negative)

The absence of lateral motion during uniaxial compression leads to:

0
0 0 0 0
ln ln ln
v xx
A A
V lA l
V l A l

=

= = = =

It is thus unclear at this point if the material law should be formulated in terms of volumetric or
linear (longitudinal) strain.

Two formulations are equivalent in the uniaxial case:

_
elastic
phase
plastic
phase
Densification
phase
Material Modeling
( )
( )
_ _
_ _
xx xx xx
xx xx v

=
=

Additional information is needed to determine the physically correct choice.

2. Uniaxial Tension


Foams show an elastic deformation phase with a (usually) non-zero Poisson coeffcient,
followed by rupture

Tensile rupture may occur between 5% and 50% of deformation (higher rupture strain for
soft foams)

3. Pure/Simple Shear


stress
metal
foam
strain
stress
metal
foam
strain
Material Modeling
Foams show an elastic deformation phase, followed by rupture

Shear and tensile response are often similar since the major component of the shear response
is the tensile stress in the first principal direction

Material Modeling
Principal Stresses for Pure Shear Loading

In foam materials, we will usually have:

3 1
<<

So the main component of the shear resistance is the tensile stress along the first principal
diagonal

4. Hydrostatic Compression

1
2
3
3 1
0
0
0

>
=
<
=

stress
metal
foam
strain
Material Modeling

Metals show an elastic behavior, for many practical applications they can be considered
incompressible

Foams show an elastic, plastic and densification phase in hydrostatic compression and are
highly compressible (Pressure Yield)

Example of a hydrostatic test device:

5. Uniaxial Versus Hydrostatic Compression

uniaxial foam
volumetric strain
hydrost. foam
Metal
(uniaxial=hydrost.)
pressure
Material Modeling
A metal exhibits the identical volumetric stress (pressure) independently of the loading
condition, the pressure depends upon the volumetric strain only
The hydrostatic pressure in a foam can be 2 to 3 times higher then the uniaxial pressure at the
same volumetric strain: the pressure depends upon the volumetric strain AND upon the state-
of-stress

Example of dytherm:

In a cell structure uniaxial compression causes bending and buckling of the cell walls,
hydrostatic compression will induce higher membrane stresses:

Material Modeling
The onset of the densification phase occurs at similar volumetric strain values in the uniaxial and
in the hydrostatic tests:

This indicates that foam material laws should be formulated in terms of volumetric strain

Remark:

If a volumetric and a uniaxial compression test are both performed up to 80% volumetric strain,
then we have:

In the uniaxial test:

( )
0
ln ln 0.2 1.6
xx v
l
l

= = =

xx v
=
uniaxial test: _
xx

uniaxial test: p
p
p
hydrostatic test: p
3
v xx

Material Modeling
In the volumetric test:

( )
3
3
0 0
0 0
ln ln ln 0.2 1.6
1.6
ln 0.58
3.
v
xx
V l
V l
l l
l l

= = =

= =

Thus only 42% compression is needed in all 3 directions to reach densification.

6. Combined Load Cases

Also due to the very different behavior of foams in compression and shear, many questions
remain open:

Does shear strength increase with compression in a combined compression/shear test?

Does the longitudinal strength of a foam increase if the foam was previously compressed in
lateral direction during a biaxial or triaxial test?

Not enough experimental evidence has been gathered today to answer these questions.

7. Unloading

Unloading can be crushable (elasto-plastic) or recoverable/hysteretic (visco-elastic):

Material Modeling
8. Rate Dependency

Rate dependency during the loading phase can be of the elasto-viscoplastic or of the viscous
type:

Minimum Requirement

The crucial question is how many experimental results must be matched by a material model
before it can be considered to reliably simulate the foams macroscopic behavior.

A minimum requirement was formulated by prof. Kikuchi (Univ. of Michigan):

Uniaxial quasistatic compression
Uniaxial quasistatic tension
Simple shear
Hydrostatic compression

Material Modeling
Uniaxial dynamic compression (2 load speeds)

Classification

For numerical simulation, 4 classes of recoverable foam materials can be identified, these
should be modeled with some non-linear elastic material law:

Soft Polyurethanes
Confor Foam
Reversible Energy-absorbing Polyurethanes

Expanded Particle Foams

Soft Polyurethanes

Main automotive applications:

Seat cushions & backs
Head supports
Certain dummy parts

Main raw material suppliers:

Dow
ARCO
ICI
Recticel
...

Production method:

Usually the reaction takes place in a mold where the part is directly shaped under a temperature
between 30 and 60 degrees.

Rarely the part is cut out of large foam blocks.

Remark: seat cushions usually contain softer central and harder lateral parts!

Main mechanical characteristics:

99% open cell (crushed)
Density between 30 and 60g/l
Reversible
Hysteretic, no important delay
Material Modeling
Skin is built during forming
Initially isotropic
Initially anisotropic if cut from a block (rare)

Numerical simulation:

Material 57: nonlinear elastic hysteretic material law with uncoupled principal stresses, if time-
effects not so important.

Material 83: if important rate effects

Uncoupled Non-linear Elastic Material Law:
MAT_LOW_DENSITY_FOAM

Assumptions for material law 57:

* Principal stresses are uncoupled
* Principal directions of stress and strain coincide
* Linear tensile behavior
* Non-linear compressive behavior

Nominal compressive loading and unloading stress-strain curves as well as Youngs modulus are
measured:

tension
0
0
f
A
+ =
compression
E
0
0
0
l l
l

=
Material Modeling
The algorithm is then as follows:

1. Calculate (objective) strain values:

0
0
_ _
V I
x
VR F
x
V left stretch tensor
+ =
+ = =
+ =
t t
t
r
tt t
r

Note that this is just the small strain tensor if the rotations are small:

( )
0
0
1
2
2
T
T
R I V F F F
V I
F F I
= = =
=
= +
t t t t t t
t t
t
t t t
t

2. Calculate principal strains
01
02 0
03
0
0 0
0 0
0 0
1
T
i
i
oi
l
l

+ =

=
t t
t

3. Calculate nominal principal stresses by using the available test data in the principal strain
directions
( )
0 0
0 0
_ : _
_ : _
i i
oi i i
tension E
compression

=
=

Thus the principal stresses are fully uncoupled.

4. Calculate principal Cauchy stresses:

( )( )
0 0
0
0 0
1 1
i i
i oj k
j k j k
l l
l l

+ = =
+ +

Indeed a transition must be made from nominal to physical stresses if the state of stress is not
uniaxial:

Material Modeling
Example of hydrostatic compression:

0 0
0
0
_
f f A A
A A A A
= = =

5. Calculate Cauchy stresses:

1
2
3
0 0
0 0
0 0
T

+ =

t t
t

Remarks about law 57:

* Shear stresses are not uncoupled, they depend upon the tensile-compressive behavior

* Nominal approach causes Cauchy stress values under triaxial (e.g. hydrostatic) conditions that
are too low initially because the longitudinal strain is used rather then the volumetric strain and
the decreasing section area

* The approach assumes a zero Poisson coeffcient (at least in the compressive region)

* Inspired by G. Storakers (RIT-Stockholm)

Hysteretic unloading is achieved by defining loading/unloading for 3D conditions based on the
relationship between the maximum energy density and current energy density:

( ) ( )
0
max
_ 1 _
m
i oi oi
e
e

= +

Here m is the unloading shape factor and ksi is the hysteresis unloading factor. These constants
must be fit to experimental data.

nominal
physical
stress
(l-l0)/l0
Material Modeling
Example of droptower simulation:

Material Modeling

Material Modeling

Material Modeling

Material Modeling
Damping in Foam Material Laws

Depends upon parameters KCON (E-modulus for contact) and DAMP (viscous damping)

A viscous damper is defined in parallel to the hyperelastic foam law:

c
Material Modeling

Material Modeling
Maximum Energy Absorption in a Foam Component

Physics:

Numerically the potential energy in the foam block may be too small and the impactor will fly
through, inverting the solid elements

max 2
0 0 0
0
max
2
1.0
mv
ke V d
=
<

Extrapolated
from
measurment
Trues
stress/strain
curve
0

0.8 1.0
Material Modeling
Data Generation for the Densification Phase

Densification phase is often missing in test results:

0.81 0.99
0.01
i

=

Missing data points can be generated using a higher order hyperbolic function:

0.80
1
1.0 0.8
1.0
n
i i
i i
i

+

=

= +

Usually a value for n between 2 and 3 is chosen

The curves can be cut as soon as a stress is reached of the same order as the yield stress in the
supporting structure

0.8 1.0
0

Material Modeling

Material Modeling

Material Modeling

Material Modeling
Data-setup for Material Law Type 57

Use stress-strain functions with constant strain interval
Use enough points to allow a smooth definition of the stress-strain relationship: no sudden
changes in the tangential modulus
Make sure that the potential energy of the foam block is high enough to absorb the kinetic
energy of the impacting body
If necessary, extrapolate the stress-strain curve beyond the experimental values, linear
extrapolation as provided automatically in LS-DYNA may not be sufficient, we have had
good experience with extrapolating along a third order hyperbolic function
Use a viscosity that is low enough not to affect the stress values in the plastic phase and high
enough to ensure stability in the densification phase
Use viscous orthogonal hourglass only

CONFOR foam


Wheel chair & aircraft seats
Helmets (F1-Indy...)
Certain dummy parts (EUROSID)


EAR Specialty Composites

Production method:

CONFOR is a polyurethane foam.


Open cell
Medium density: 6pcf or 60-70 g/l
Reversible with slow recovery
Highly damped (relaxation)
High rate effects
High thermal sensitivity
Initially isotropic


Material 62: Developed by Ove Arup for the simulation of CONFOR foams, allows simulation
of high viscosity with densification.

Material Modeling
Material 83: Modern way for simulating highly rate dependent reversible foams

EA-Polyurethanes


Bumpers
Knee and head polster
Side impact padding


Bayer (Bayfill)
Elastogran

Production method:

Usually the reaction takes place in a mold where the part is directly shaped under a temperature
between 30 and 60 degrees.

Alternatively the part is foamed directly in position.

Different crush strengths can be obtained for a given density.


Medium open cell, up to 95% closed cell
Density between 50 and 110 g/l
Reversible with slow recovery & damping
Permanent deformation possible
Crush strength not uniquely related to density
Plastic part of stress-strain curve in compression is nearly horizontal: good energy absorber
Rate effects
Temperature effects (less then EPF)
Initially isotropic


Reversible EA Polyurethanes can be simulated like EPF (see next page)

Irreversible EA Polyurethanes can be simulated with material laws 63, 53, 26, 75, 5, 10...

Material Modeling
Expanded Particle Foams


Bumpers
Child seats
Side impact padding

Main raw materials: (EPP-EPS):

Napoleen: Polypropylen, 20-85 g/l (BASF)
Arpro: Polypropylen, 20-200 g/l (JSP)
Styroterm: Polyphenylen-ether (BASF) (Copolymer Beads)
Caryl: Polystyrol (Shell)
Dytherm: Polystyren , 2.5-5 pcf
Dylite: Polystyren, 1.0-4.5 pcf (ARCO)
Cellular Polyethylene
...

Production methods:

Foam particles are obtained by expansion of soft plastics and have similar properties: they
become viscous at high temperatures. Particles are welded with steam under 2-4 bar pressure into
any shape. Weld is stronger than the particles.


Closed cell
Density between 20 and 200 g/l
Reversible with slow recovery & damping
Crush strength determined by density
Crush strength higher then PU at same density
Plastic part of stress-strain curve in compression has a slope, EA more difficult to control
Complex parts since:
Any shape can be welded
Holes necessary to obtain constant force levels
Rate effects
Temperature effects
Initially isotropic


Material law 83: input of a family of stress-strain curves at different strain rates to account for
high rate dependency.

Material Modeling
Simulation of Rate Dependent Reversible Foams
Using Material Law 83 in LS-DYNA: MAT_FOAM_FU_CHANG

Principle:

This is a generalization of law 57 where a family of rate-dependent stress-strain curves can be
defined in the tensile and the compressive region

Remarks:

Rate-dependent tensile curves allow to simulate rate dependency under shear loads

Unloading always follows the curve with the lowest strain rate

Optional input of a hydrostatic test result allows considering the influence of the increasing
density for biaxial and triaxial loadcases

MAT_FU_CHANG_FOAM is currently the generally accepted way to simulate recoverable
foams

Droptower tests are required to obtain dynamic stress-strain curves in the strain rate region of
interest

tension
compression
0
0
f
A
+ =
0
0
0
l l
l

=
Material Modeling
Consider true strain rate and nominal or engineering strain rate in a droptower test:

True strain rate can only be kept constant in hydraulic machines for rate up to 1/s

In droptower tests, only the nominal strain rate can be kept constant for a large part of the test by
using large dropmasses, this enforces a nearly constant dropvelocity

At high strain values, the difference between true strain rate and engineering strain rate can be
very important:

Drop tests can be performed for strain rates of several hundred per second at constant nominal
strain rate

l0
x
M
0
0
x
l
x
l
=
=
&
&
&
&

Strain
rate
&
0
&
strain
0.9
Material Modeling
Consequence:
Input Must Allow Curves at Constant Nominal Strain Rate

In version 950 of LS-DYNA, it is possible to specify input of strain-stress curves at constant
engineering strain rate

Velocity of the dropmass must be delivered as part of the test result in order to allow cutting of
the stress-strain curves as the velocity drops:

stress
Strain
Dropmass
velocity
Strain
Material Modeling
Crossover of stress-strain curves are often caused by drop of the velocity in the densification
range

Cutting and extrapolating curves will eliminate crossovers that are likely to cause numerical
problems

stress
Dropmass
velocity
Strain
Strain
Material Modeling
Input Data for Material Law 83

Get the tensile modulus and tension cutoff stress from a quasistatic tensile test, dynamic
tensile data can also be used since version 950 if available (TFLAG=1)

Get the stress-strain curves from a quasistatic compression tests and droptower tests, if
possible at constant engineering strain rate (SFLAG=1)

Use the unloading path of the quasistatic compression test as a stress-strain curve for zero
strain rate (=unloading path)

Define stress-strain curves with 100 equidistant points (strain increment 0.01 or 1%), ensure
smoothness and extrapolate if necessary

Use viscous hourglassing

Contact modulus also determines timestep and hourglass viscosity, adequate values ensure
numerical stability

Remaining Problems with Material 83

Confor foams can show unloading paths at high strain rates that are above loading paths at
lower strain rates, currently material law 83 cannot simulate such a behavior since a unique
unloading path is used

Dependency of test results upon sample size and shape cannot be simulated, this dependency
is important for very low density foams (seat foams) and results from the outflow of air
during compression, simulation requires implementation of the theory of porous media

Representation of the skin layer typical of cold formed foam parts is possible with shell
elements, material tests should be done on cut samples without skin to determine the material
property and on samples with skin to determine the skin property

One important effect of skin can be to prevent air outflow, in this case internal air pressure
should be added to foam parts with skin

Modeling of Reversible Foam Parts

The use of type 10 tetrahedron elements seems to give acceptable accuracy and allows for
fast and automatic modeling

Degenerated hexa elements have been shown to perform very badly

Good quality tetra elements must be generated
Material Modeling

Material Modeling

Material Modeling
4.4.2 Material Models for Brick Elements Representing
Energy-Absorbing Materials

Elasto-plastic material laws
Zero or non-zero Poisson effect
Rate dependency
Failure in tension & shear
Large compressive deformations

Applications:

Crushable PU
High density foams
Aluminum honeycomb
Aluminum foam

Summary

Simulation of plastic deformation with plastic volumetric strain is required.

This can be done in a number of ways:

Use of a suitable EOS with classical plasticity (material 5 & 10) this is the old way

Use of non-associated flow rules (material 75) very suitable for simulation of aluminum
foams

Yield functions for uncoupled stress components (material 26 & 126) best way to simulate
honeycomb because of the orthotropic features

Yield functions for uncoupled principal stresses (material 53 and 63) original way to simulate
crushable PU

Material Modeling
Simulation of Crushable Foams Using Hydrodynamic Elasto-plastic
Material: (Material Law 10 in LS-DYNA)

This is a material law combining metals-type elasto-plasticity with a tabulated equation of state.

The pressure term is treated separately and contains a tabulated loading term and a hypoelastic
unloading term:

0
0
ln
ln
u
V
p p
V
V
p K
V
| |
=
|
\ .
=

The deviatoric stresses are calculated according to classical elasto-plasticity.

This material law was originally developed to treat metals under very high pressure, where
compaction (=permanent volumetric deformation) occurs and the metal becomes compressible.

Material data are calculated from a uniaxial compression test as follows:

Remember that under uniaxial compression:

( )
( )
1 2
2
1
3
v xx t
p xxp p

=
= +
& &
& &

If the elastic strains are small compared to the plastic strains, this simplifies to:

( )
( )
1 2
2
1
3
_ : _ 0.
v xx p
p xx p
p t
p
v
v
v v
foam v

=
= +
& &
& &

x
Material Modeling
The load-deflection characteristic of the foam under uniaxial compression is easily reproduced
by translating the measured force-deflection curve into a yield stress-plastic strain curve:

The correct volumetric behavior is then obtained by adding compressibility: a non-linear
equation of state is used to obtain the correct volume at all times:

y
f
A
=
0
2
ln
3
p
l
l
| |
=
|
\ .

0
d l l =
f
p=f/3A
0
ln
v
l
l
| |
=
|
\ .

Material Modeling
This must be combined with suitable elastic constants determined as follows:

0
0
2
3
xx
xx
E
E
G
E
K
| |
=
|
\ .
=
=
=

The material model will then have a foam-like uniaxial behavior.

In the densification phase the elastic strains are no longer negligible and deviations between test
and simulation will arise. They can be largely overcome by setting:

0 0
2
3
2
ln
3
y
p xx
p
E
l f
l EA
| |
=
|
\ .
| |
| |
=
|
|
\ .
\ .

Remarks

* The predicted limit shear stress in this model will be:

1
3
xy y
=

This may or may not be realistic for a foam

* Pressure depends upon the density only: for similar volumetric strain, identical pressures will
be obtained independently of the state of stress. This is not realistic for most foams.

* The uniaxial behavior is perfectly well predicted by this model

* Strain rates are not considered, so a uniaxial test should be performed at a velocity similar to
real world loading

* The numerical stability of this model is remarkable

* This model can be generalized for non-zero Poissonss ratio
Material Modeling
Simulation of Crushable Foams Using Drucker-Prager
Elasto-Plastic Material: Material law 5 in LS-DYNA

A Drucker-Prager elasto-plastic material is usually implemented with non-associated (von
Mises) flow rule:

0 1
0
1
0.
vm
vm
f B B p
g
B
q
B
q
=
=
| |
=
|
\ .
=
r
r
&

Sometimes the yield function is formulated as:

2
2
0 1 2
3
vm
f A A p A p
=

Since the flow rule used is still of the von Mises type, the pressure will depend upon the density
only and this can be combined with any non-linear EOS:

( )
v
u v
p p
p K
=
=

The numerical treatment of this material is simpler then the previous one since we have a
perfectly plastic behavior: the yield surface depends upon the (known) pressure only.

The yield condition at timepoint (n+1) gives the consistency parameter immediately:

, 1 0 1 1
1 1 0 1 1
, 1 0 1 1
, 1
0 1 1
, 1 , 1
0.
3
: 0.
2
3
1 2 0.
2
3
1 2
2
vm n n
n n n
evm n n
evm n
n
evm n evm n
B B p
s s B B p
G t B B p
B B p
G t

+ +
+ + +
+ +
+
+
+ +
+ =
=
| |
+ =
|
|
\ .
| |
+
=
|
|
\ .
t t

Material Modeling
This leads to the new stresses immediately:

0 1 1
1 , 1
, 1
n
n e n
evm n
B B p
s s
+
+ +
+
| |
+
=
|
|
\ .
t t

Use of Drucker-Prager Elasto-Plastic Material to Simulate Foams:

In this case the equation of state is determined from the uniaxial compression test results:

Exactly the same EOS is thus defined as in the elasto-plastic hydrodynamic case.

The elastic constants are also defined in an identical way.

The yield surface follows from:

0
1
0 1
2
3 0
0
3
0
3
vm
p
B
B
A A
A
=
=
=
= =
=

In the invariant plane, the yield surface thus coincides with the uniaxial line:

f
p=f/3A
0
ln
v
l
l
| |
=
|
\ .

0
d l l =
3
vm
p =
Material Modeling
Remarks

* The shear strength grows with compression, in particular the shear strength is initially zero: the
uncompressed material will behave like water, highly compressed material becomes overly stiff

* Again the pressure depends upon the density only and not on the state of stress

* Uniaxial behavior of foam can be well simulated

* No strain rates considered

* Numerical problems are likely because of the initial zero deviatoric strength

Simulation of Honeycomb for Small Shear Deformations
MAT_HONEYCOMB
Material Laws 26 in LS-DYNA

In the honeycomb model, each stress component has a proper elastic modulus and yield
function, the yield function is a function of volumetric strain only:

( )
,
_
_
ij ij ij
ij y ij v
E

=
& &

Hereby the stress components are expressed in a local (corotational) element system, this
ensures objectivity

This model was originally motivated for modeling of the NHTSA side impact barrier by
Skolnikow

The numerical algorithm is trivial and uncoupled for every stress component:

( ) ( )
1 1/ 2
1 1 , 1
min ,
n n n
n n y v n
E t

+ +
+ + +
= +
=
&

Material law 26 is inherently orthotropic

Isotropic behavior can be obtained in the elastic phase by setting:

2 2 2
xx yy zz xy yz zx
E E E G G G = = = = =

In the non-linear region, this material always behaves orthotropically. (If we load under 45,
the stresses along both coordinate axes tend towards the yield value independently)

Material Modeling
For a given relative volume, the material is considered compacted and behaves elasto-
plastically, the properties should then correspond to aluminum

Moduli increase linearly as a function of volumetric strain between the initial and compacted
values, estimate of the timestep is based on the (maximum) compacted value:

( )
0
1
ij ij ij v
v
c
E E E E
V
V
l
t
E
= +
=
=

This material model can give a good representation of the aluminum honeycomb in the case
of small shear deformations, i.e. no tear-off may occur and the spatial deformation gradients
must remain small at all times: the honeycomb must be compressed rather homogeneously

The corotational formulation in the brick element strictly spoken already limits the use of this
material model to small shear deformations

Data Preparation for MAT_HONEYCOMB

Reasonable values for the fully compacted material:

0.1
70000.
200.
y
VOLf
E MPa
MPa
=
=
=

For volumetric strains below 90% we need modulae and yield curves from experiments

Material Modeling
Tensile and compressive tests give yield stress as a function of volumetric strain in
longitudinal and transversal directions (4 tests necessary):

For the initial modulae, the maximum tangent to the measured curve (before compaction) is
usually adequate:

max
ii
ii
v
E

Maximum shear is given as a function of volumetric strain, this means that no tear-off can be
considered since the shear resistance is considered to remain constant with increasing shear
strain

Typically we use a yield curve for all 3 shear components that is equal to the longitudinal
yield curve scaled by a factor comprised between 0.3 and 0.6

The factor can be estimated form a shear test in uncompressed configuration:

Shear strength for small shear strains is then assumed to increase with compression like the
longitudinal strength

1-V/V0
Material Modeling
Simulation of Honeycomb with Tear-off:
MAT_MODIFIED_HONEYCOMB
Material Laws 126 in LS-DYNA

A variant is material law 126 where the yield functions are expressed in terms of the
corresponding linear strain components:

( )
,
_
_ _
ij ij ij
ij y ij ij
E

=
& &

This allows to control the nonlinear shear resistance

The 6 local stress components are now fully uncoupled

This model was originally motivated for the modeling of the European ODB

The difference between material laws 26 and 126 is perhaps most clear under a longitudinal load,
followed by a transversal load.

The volumetric strain dependent yield functions harden in all directions if one uses material 26,
this is not the case with material law 126 where the stress components are fully uncoupled.

Consider the example using identical bilinear yield curves in all directions:

This material model aims at simulating large deformation gradients in the honeycomb
(puncturing) with tear-off or shear failure of the elements

t
( ) ( ) 26 126 26
xx yy
= =
( ) 126
yy

Material Modeling
Element deformations may be extremely high and negative volumes become almost
unavoidable, therefore, small strain elements were developed that allow continuation of the
calculation if the element volume goes negative

Possible element formulations are:
Type 1: traditional large strain
Type 0: non-objective small strain, only for very specific applications such as ODB
barrier, no full compaction will occur
Type 9: objective (corotational) small strain

Data Preparation for MAT_MODIFIED_HONEYCOMB:

The longitudinal axial compressive and tensile strength must be determined from uniaxial
tests, the tensile strength is higher then the compressive strength

In the transversal directions the compressive yield stress is very low and should probably be
raised somewhat for numerical stability

Note that the strain definition depends upon the element formulation but is always positive in
compression:

Type Strain

0,
1
i
ii
i
l
l
=

0,
ln
i
ii
i
l
l
| |
=
|
|
\ .

2
t c

t c

c
Tension
Yield stress
compression
100
c
c

Material Modeling
Remark: in the absence of test-data, yield curves can be determined by a simulation of small
honeycomb test cubes subjected to tension, compression and simple shear loads, modeling
the honeycomb structure realistically with 4 elements per side

The shear strength for the 2 strong directions (xy and zx if x is longitudinal) is usually given
as a fraction of the compressive yield strength and includes failure:

Elements that failed in shear can be eliminated specifying a maximum shear strain

The shear behavior in the weak direction (yz-plane) may be very complex, the initial strength
(up to about 20% shear deformation) is very low but as the cells align this may change
dramatically

0.6
s c

Shear
tension
Material Modeling
Simulation of Crushable Foams Material Laws with Uncoupled
Principal Stresses:
MAT_CRUSHABLE_FOAM
Material Laws 63-53 in LS-DYNA:

Crushable PU-foams are popular as door-paddings for side-impact protection and other
safety applications, a typical example is Bayfill from BAYER

Typical properties are:
No or little rate dependency
Failure in tension and shear at low strain
Relatively high compressive strength
Permanent deformation
Influence from confined air if skin is present (cold formed parts)
No influence of confined air if no skin is present (cut parts)

For the skeleton stresses in the Sandia model, (law 53) and the crushable foam model (law
63) a single elastic modulus and a single yield function are used for all principal stress
components:

( ) ( )
1
2
3
1
2
3
_
0 0
_ 0 0
0 0
_ min , _
0 0
_ 0 0
0 0
ij ij
T
i i y v
T
E
=
| |
|
=
|
|
\ .
=
| |
|
=
|
|
\ .
& &
t t
t
t t
t

Since the stress components are uncoupled, no Poisson effect will develop. These materials
behave isotropically. The same load can be sustainded in all directions.

We compare the behavior of material laws 26 and 63 where all stress components (26) and all
principal stresses (63) have been given the same yield values as a function of volumetric strain

Material Modeling
Loading is under 45 degrees in the material xz-plane:

Material Modeling
The stress components in x- and z-direction reach the yield value independently for material 26,
only the principal stress (under 45 degrees) reaches the yield value for material 63;

Material Modeling

This results in a higher principal stress in material 26 (element 1), under loading along the x- and
z-axes, such principal values could not develop in this element

In a model with uncoupled stress components the pressure depends upon the state of stress but in
a way that the user cannot control.

Consider an isotropic material with uncoupled stress components and an elasto-perfectly plastic
behavior:

Consider uniaxial and hydrostatic compression up to a given volumetric strain in the elastic
region, the pressures are:
1
3
1
3
3 3
u v
v
h u
p E
p E p
=
= =

stress
strain
Material Modeling

Showing that in the small strain elastic region, pressure still depends upon density only,
if we work in the plastic region however:

1
3
1
3 3
3
u y
h y u
p
p p
=
= =

The hydrostatic pressure is 3 times the uniaxial pressure at the same volumetric strain.

Obviously this factor depends upon the shape of the yield function.

This will overestimate pressures in foams since values between 2 and 2.5 are measured.

In the Sandia model (M57) the influence of the contained air can be considered as follows:

( ) ( )
0 0
0
0
0
1 /
1
polymer polymer
v
v polymer
v
p V V p V V
p p
V V
V
V
=
=

=

The air pressure exceeding atmospheric pressure is added to the skeleton stresses as a
hydrostatic component and will cause a Poisson effect

Simulation of High Density Foams Elasto-plasticity
with a Generalized Flow Rule:
MAT_BILKHU_DUBOIS Material Law 75 in LS-DYNA

High density foams (200-700g/l) are used to enhance the stiffness (NVH) as well as the
strength (safety) of tubular structures

These foams can be polymeric (Terocore, Betacore, Essex...) as well as metallic (aluminum,
magnesium...)

Typical properties are:
Permanent deformation
Non-zero Poisson coefficient
High compressive strength
Even higher confined compression strength
Failure in tension and shear
Material Modeling
A Generalized Flow Rule

To describe foams we will use a family of flow rules based upon a stress invariant defined as
follows:

2 2
_
f vm
p = +

This is obviously invariant since the components, the von Mises stress and the pressure are
invariant.

We will consider the surfaces where this variable has a constant value to be flow surfaces, the
plastic strain rate must then be normal to the flow surface or:

f
g =

By differentiation we obtain:

1
2 2
2
vm
p vm
f
p
p

| |
= +
|
\ .
t
& &
t t

To evaluate the remaining terms we set:

1
3
xx xy xz
yx yy yz
zx zy zz
p p p
p p p p
I
p p p

| |
|
|
|
= = |
|
|
|
|
\ .
t
t

Similar but somewhat more complicated algebra shows that the stress derivative of the von
Mises stress is proportional to the stress deviator:

3
2
vm
vm
s

=
t
t

f
p
=
t
& &
t
Material Modeling
If we insert these 2 expressions into the flow rule we obtain:

Thus the plastic strain rate is proportional to the stress deviator plus a fraction of the pressure.

The deviatoric and volumetric components of the plastic strain rate are easily determined:

In the general case, the volumetric plastic strain rate is thus non-zero and proportional to the
pressure.

The parameter is called dilatancy. The larger the dilatancy, the more volumetric plastic strain
will occur.

The dilatancy must be comprised between 0. and 4.5 for physical reasons.

9
0.
2

In the lower limit we obtain the flow rule used in classical metals plasticity:

f vm
g = =

3 3
2 2
p dp
f vm
s s

= = =
t t t t
& & & &

Here:

the flow surface is a sphere in deviatoric space

the flow surface is a cylinder in stress space

the plastic strain rate is proportional to the stress deviator

consequently the plastic strain rate is deviatoric

the volumetric plastic strain rate is zero: the plastic deformation occurs at constant volume
3 2
2 9
p
f
s pI
| |
=
|
\ .
t
t t
& &

( )
( )
3
1 3
3 2
vp p
f f
dp p vp
f
p p
tr tr I
s

= = =
= =
t
t
& & &
&
t t t
& & &
&

Material Modeling
In the upper limit for the dilatancy we obtain:

( )
2 2
9
2
3 3
2 2
f vm
p
f f
g p
s pI

= = +
= =
t
t t t
& & &

Here the plastic strain rate is proportional to the stress tensor.

In this case:

volumetric plastic strain rates are non-zero: the material is compressible during plastic
deformation

the flow surface is a sphere in stress space, indeed in 3D principal stress space:

( ) ( ) ( )
( ) ( )
( )
2 2 2
2 2 2
2 2
1 2 3
2 2 2 2 2 2
1 2 3
2 2 2 2
1 2 3
9
2
9 3
2 2
9 3
3 2 3
2 2
3
2
f vm
f
f
f
p
p p p p
p p p p

= +
(
= + + + + + +

= + + + + +
= + +

Magnitude of Plastic Strain Rate

In general plasticity, the parameter
&
is called the consistency parameter. It can still be
interpreted as the magnitude of the plastic strain rate:

2
2 2 2
2
:
3
_ :
vm
p dp dp
f
vp
f
vp
p vm
f
p
thus
p

= =
=
+ = + =
t t
& & &
&
&
&
&
&
&
&

Material Modeling
Note that there is no problem in the incompressible limit:

0 0
vp p
= = =
&
& &

Measurement of the Dilatancy

We will consider a triaxial experiment where a cilindrical probe is first hydrostatically loaded in
a tank of water, from this initial condition it is then further compressed axially under constant
lateral pressure:

We obtain a triaxial state of stress:

_ 0.
_
2
_ 0.
3
_ 0.
yy zz
xx yy zz
xx yy
vm xx yy
p

= <
< =
+
= >
= >

Remember the plastic Poisson coefficient:

yyp
p
xxp
=
&
&

The volumetric plastic strain rate is then:

( )
1 2
vp p xxp
= & & (*)

Expressions for the longitudinal and volumetric plastic strain rate can also be obtained by
directly applying the flow rule:
x
Material Modeling
3 2
2 9
xxp xx
f
vp
f
s p
p
| |
=
|
\ .
=
&
&
&
&

Inserting these values in (*) we obtain:

( )
3 2
1 2
2 9
p xx
p s p

| |
=
|
\ .

This can be solved for the dilatancy.

We first evaluate the stress deviator:

( )
2
3
2
3
xx xx
xx xx yy
xx vm
s p
s
s
= +
=
=

If we replace and solve for the dilatancy:

( )
( )
( )
1 2
3
1 2
3
_
2 1
p vm
p
vm
p
p
p
p
| |
=
|
\ .
=
+

In the case of uniaxial compression (zero lateral pressure) we obtain:

( )
( )
_ 3
1 2
9
_
2 1
vm
p
p
p
=
+

This test is sufficient if the dilatancy does not depend upon the state of stress.

We can try to estimate the dilatancy from a uniaxial compression test by assuming that:

the plastic Poisson coefficient is constant during the loading phase

the plastic strain rates differ little from the total strain rates
Material Modeling
Then:

0
0 0
ln ln
ln 2 ln
yy p xx
yy p xx
y
x
p
yo x
x
p
x
dt dt
l
l
l l
A l
A l

& &
& &

This allows estimation of the plastic Poisson coefficient and thus the dilatancy from the
measurement of the test piece cross section during a uniaxial compression test.

Since small values of the plastic (or total) Poisson coefficient are measured, the metals type flow
rule cannot apply to simulate foams.

The results of the uniaxial test illustrate the limit values of the dilatancy:

0.5 0.
9
0.0
2
p
p

= =
= =

Thus:

* Dilatancy zero means plastic deformation at constant volume

* No lateral motion under uniaxial compression means the dilatancy is 4.5

* A dilatancy higher then 4.5 would lead to shrinking of the probe during compression, this is
actually possible in real foams but was not considered in our models

Clearly a single uniaxial test is sufficient to determine the dilatancy if the latter is a constant. For
most foams this will not be the case and multiple triaxial experiments are needed to allow
understanding of the variation of the dilatancy with the state of stress.

Material Modeling
Numerical Treatment

We will illustrate the numerical treatment for an elasto-plastic material law with a generalized
flow rule:

( ) ( )
_ 2
d dp v vp
G K I =
t
t t t
& & &
& &

Radial Return Algorithm

Step 1: Calculate elastic trial stress:

, 1/ 2
, 1/ 2
, 1
, 1
_
_
_ 2
_
d d n
v v n
e n n d
e n n v
t
t
s s G
p p K

+
+
+
+
=
=
= +
=
t t
&
&
t t t

Step 2: Check the yield condition:

( )
( )
, 1 , 1
, , 1 , 1 , 1
, , 1 , 1
1
_
3
3
_ :
2
_ , ,... 0_?
e n e n
e vm n e n e n
vm e n e n
p tr
s s
f p
+ +
+ + +
+ +
=
=
t
t t

Step 3: Perform radial return if necessary:

( )
, 1 1
, 1
, , 1
1 , 1
, , 1
_ :
2
3
2 2
3
2
2
3
1
d dp
d
f
e n n n n
e n
f e n
n e n
f e n
deviatoric
s G
s G G s
s
s s s s
G s
t t
G
s s

+ +
+
+
+ +
+
=
=

=

| |
=
|
|
\ .
t t t
& & &
t t t
& & &
t t t t
t
&
t t

3
_ 2
2 3
e
f f
G s K pI

=
t
t t t
& & & &

Material Modeling
( )
, 1 1
, 1
, , 1
1 , 1
, , 1
1 , 1
, , 1
_ :
3 2
3
2 9
1
3
1
3
v vp
v
f
e n n n n
e n
f e n
n e n
f e n
n e n
f e n
volumetric
p K
p K K p
p p p p
K p
t t
K
p p
G K
p p
G

+ +
+
+
+ +
+
+ +
+
=
=

=

| |
=
|
|
\ .
| |
=
|
|
\ .
& & &
&
& &
&

Identical scaling for pressure and deviatoric stresses is performed if:

( )
( )
( ) ( )
( ) ( )
( )
( )
1
3
9 1 2
_ : _
2 1
_ : _ 1
2 1 2 1
3
3 9 1 2 3 3 1 2
p
p
p
K
G
v
if v v
v
K
then
E v v K
G
G v E v
=
=
+

`
=
+ +

= =

)

In this case the scaling is radial in stress space:

1 , 1
, 1 , , 1
1 , 1
, , 1
_
_
_
3
_ 1 1
n e n
vm n vm e n
n e n
f e n
k
k
p kp
G
k

+ +
+ +
+ +
+
=
=
=
=
t t

The stress scale factor is determined from the requirement that we fulfill the yield condition
exactly at time n+1, often a Drucker-Prager type yield condition is used:

( )
( )
, 1 1
, , 1 , 1
2 2
, , 1 2 2
0 1 , 1 2 , 1
, 0
, 0
0
3
vm n n
vm e n e n
vm e n
e n e n
f p
f k kp
k
A Akp A k p
+ +
+ +
+
+ +
=

Material Modeling
The last is a second order equation for k from which the positive root must be determined, this is
easily seen in invariant space:

The penalty for scaling radially, independently of the value of the dilatancy is that a metallic
flow rule cannot be considered. This would require an elastic Poisson ratio of 0.5 and thus 100%
incompressible material.

As in metals type plasticity, this algorithm respects the physics of the problem very well for a
foam.

Indeed a radial return is performed in stress space, thus giving a stress return path parallel to
the vector of principal stresses, as indicated by the material law

k>0
k<0
p
1
_
n
+
t

, 1
_
e n
+
t

Material Modeling
Representation in the Invariant Plane

The direction of the plastic strain rate vector in the invariant plane varies between a vertical line
and a line through the origin:

The pressure is modified by the plasticity algorithm since the volumetric plastic strain (rate) is
non-zero.

The pressure in this case depends on the density but also on the state-of-stress.

This type of material law is implemented in LS-DYNA in material law 75.

Simulation of Crushable Foams Using
the Bilkhu-Du Bois Model

Features of LS-DYNA material law 75:

Generalized flow rule with the following restriction:

0 4.5
0 0.5
p
v v
<
= <

Radial return in stress space

Elliptical (second order) yield surface in invariant space

Perfect plasticity: the foam does not harden and the yield surface depends upon the total
volumetric strain only

vm
+
vm
cte + =
p(n+1) pe(n+1) p
Material Modeling
Yield Surface

The results of uniaxial compression test, uniaxial tension tests and hydrostatic compression test
are expressed as yield stress in function of volumetric strain:

Yield stress in uniaxial tension and compression:

Pressure yield in hydrostatic compression:

Bayer performs hydrostatic testing up to over 200 bar.

0
ln
v
V
V
| |
=
|
\ .

0
ln
v
V
V
| |
=
|
\ .

Material Modeling
In the invariant plane this provides 3 points of the yield ellips that varies with volumetric strain:

If no tensile test is available a third point can be arbitrarily defined as:

10.
c
t
p
p =

An ellipse is obtained only if:
3
vmu
c
p

Variation of the ellips with positive and negative volumetric strain:

,
,
,
3
vm u
vm u
| |
|
\ .

( ) , 0
c
p
( ) , 0
t
p
p
V0
V<V0
Material Modeling

No hardening takes place in this foam model, a comparison of theoretical unloading curves
shows this to be somewhat more plausible then a hardening model where the plastic deformation
would govern the variation of the yield surface:

An ellipse is fitted to obtain a continuous yield surface

V0
V>V0
-ln(V/V0)
stress
Current model
Isotropic hardening
Material Modeling
( )
( )
2
2
0
0
2
0
2
2
0
1 0.
2
3
vm
c t
c
vmu
vmu
p x
f
a b
p p
x
a p x
a
b
a x
= +
+
=
=
=
| |

|
\ .

This yield condition is easily brought into the standard Drucker-Prager form:

2
2
0 1 2
2
0
0
0
1
2
0.
3
3 3
2
3
3
vm
A A p A p
b bx
A
a
bx
A
a
b
A
a

=
=
=

Additional (triaxial) testing is necessary to obtain more realistic yield conditions

Use of the Bilkhu-Du Bois Material Law

Data consist of uniaxial and hydrostatic tests:

( )
( )
0
xx
xx
vmu vmu v
c c v
E
p p
| |
=
|
\ .
=
=

If a Poisson effect can be measured, the uniaxial test results must be converted to true strain and
true stress.

Material Modeling
Test result:

M75 input:

The conversion can be done easily by fortran or using EXCEL:

( )
0
0
0 0
0
0
ln
ln
ln ln 2
ln 1 2
ln
x
xx
xx
x
y
yy xx xx
y
z
xx
zz xx
z
l
l
l
l
l
A
v v
l A
V
l
v
v
V
l

=
=
= = =
`

=
= =

)

0
0
f
A
=
0
0
0
l l
l

=
f
A
=
0
ln
v
V
V
=
Material Modeling
Thus:
( ) ( ) ( ) ( )
0
0
0
2
2
0
0 0 0
0
1 2 ln 1 1 2 ln 1
xx
v
l l
l
A l
e
A l

| |
| |
= =
| |
\ . \ .
| |
= = =
|
\ .

Good validation of such data require verification using for example a confined uniaxial
compression test

Continued Development

Strain rate effects can be considered by scaling as a function of volumetric strain rate:

( ) ( )
( ) ( )
( ) ( )
*
*
*
uysc uysc v v
uyst uyst v v
c c v v
af
af
p p af

=
=
=
&
&
&

Test results of uniaxial compression at different impact velocities were obtained by DLR-
Stuttgart, EMI-Freiburg and other laboratories

Determination of the yield surface from uniaxial and triaxial data:

( )
,
3
,
2
3
vmu
vmu
x vmx
vmx xx yy
xx yy
x
p
p

| |
|
\ .
=
+
=

Material Modeling
Remarks

* Elasto-plastic models were based upon uniaxial data only and could predict erroneous response
under triaxial loading conditions

* Bilkhu-Du Bois model is based on uniaxial AND hydrostatic data

* No experimental information about shear is used

* No viscous effects are included

* Model is very simple and allows efficient and stable numerical treatment

A Comparison of Crushable Material Models

In the example, single brick element tests are performed for uniaxial tension, uniaxial
compression, simple shear and hydrostatic compression

Material models 10, 26 and 75 are compared, foam-like data are set up to give nearly identical
results under uniaxial compression

Material Modeling
All problems are given displacement:

Under uniaxial compression the results are similar independently of the chosen material law:

Material Modeling
Under uniaxial tension, material law 26 and 10 give similar results, material law 75 is weaker

In uniaxial shear, the results of model 75 depend upon the intersection of the yield ellipse with
the vertical von Mises axis in the invariant plane:

Material Modeling
In hydrostatic compression, results differ by up to 200%:

Compare uniaxial and hydrostatic pressure values:

Material Modeling
Workshop on Material Models

mat.inf compare material laws
26, 10 and 75
m26631.inf compare material laws
26 and 63
usr41b.inf implement user material law 41
pull11.inf
bend11.inf
user-defined integration rule

Run the following variants:

mat.inf compare material laws
63, 5, 126 in the same way
m26631.inf check that 26 and 63 give identical results
when loaded along x-axis
usr41b.inf make yield stress dependent upon
volumetric strain
usr41b.inf make yield stress dependent upon
hardening parameter

Material Modeling
4.4.3 Development of a Material User-Subroutine: Material
Law 41-44 in LS-DYNA

As an example we develop an elasto-plastic material
Law with the following features:

Zero elastic Poisson coefficient
Generalized flow rule
Von Mises type yield surface
Perfect plasticity

Or in terms of equations:

2 2
3 9
0
.
p
p
f
vm y
y
E E
s pI
cte

=

=

=
t t t
& & &
t
t t
& &

The radial return algorithm contains following steps:

( )
, 1
, 1 , 1
, 1 , 1 , 1
, , 1 , 1 , 1
, , 1
1
3
3
:
2
0?
e n n
e n e n
e n e n e n
vm e n e n e n
vm e n y
E
p tr
s p I
s s

+
+ +
+ + +
+ + +
+
= +
=
= +
=

t t t
t
t
t t
t t

If the yield condition is not fulfilled, the state of stress is plastic and the elastic stresses must be
returned to the yield surface, for the deviatoric stresses this means:

1 , 1
, , 1
, , 1
1 , 1
, , 1
1
y
n e n
vm e n
vm e n y
n e n
vm e n
s s
s s
+ +
+
+
+ +
+
=

=

t t
t t

Material Modeling
The return path of the pressure term depends upon the dilatancy (and varies linearly with the
dilatancy)

, , 1
1 , 1
, , 1
, , 1
1 , 1
, , 1
1
0 0
4.5 1
:
2
9
2
1
9
vm e n y
n e n
vm e n
vm e n y
n e n
vm e n
p p
thus
p p

+
+ +
+
+
+ +
+

=

= =
= =
=

=

After reduction the final stresses are obtained as;

1 1 1 n n n
s p I
+ + +
=
t
t t

vm

vm y
=
, 1 e n
+
t

p
Material Modeling
The programming is very simple:

c
character*(*) etype
dimension cm(*),eps(*),sig(*),hisv(*),s(3)
c
c material constants
c
e=cm(1)
sy=cm(2)
alpha=cm(3)
c
c elastic trial stress
c
do 100 i=1,6
sig(i)=sig(i)+e*eps(i)
100 continue
c
c invariants and deviatoric stress
c
p=(-1/3.)*(sig(1)+sig(2)+sig(3))
do 101 i=1,3
s(i)=sig(i)+p
101 continue
svm=s(1)**2+s(2)**2+s(3)**2+2*sig(4)**2+2*sig(5)**2+2*sig(6)**2
svm=sqrt((3./2.)*svm)
c
c check plasticity
c
scale=0.
IF (SVM.NE.0.) scale=(svm-sy)/svm
scale=max(0.,scale)
c
c reduce stress
c
do 102 i=4,6
sig(i)=sig(i)*(1-scale)
102 continue
do 103 i=1,3
sig(i)=s(i)*(1-scale)-p*(1-scale*2.*alpha/9.)
103 continue
c
return
end

Material Modeling
Implementation Procedure

1: Develop user-subroutine in fortran

2: Replace for umat41.f in standard set dyn21.f

3: Compile and link by executing the Makefile:
make

4: Run the newly created executable:
./ls... I=

5: Always run a verification testcase

Test Case

A simple test case can be made by uniaxially compressing 2 brick elements

The input required is 3 constants per material

Type 41:

, ,
y
E

We create a foam-like and a metals-like brick as follows;

Element alpha
1 4.5
2 0.

Brick 1 will compress without lateral bulging, whereas Brick 2 will deform at constant volume
Material Modeling
Test data including user-defined material are as follows: (test case usr41b.inf):

*KEYWORD
*TITLE

*CONTROL_TERMINATION
8.00000-3 0 0.0000000 0 0.0000000
*CONTROL_TIMESTEP
0.0000000 0.0000000 0 0.0000000 0.0000000 0 0
0
*CONTROL_SHELL
0.0000000 0 0 0 0 0 0
*CONTROL_DAMPING
0 0.0000000 0.0000000 0.0000000 0.0000000 0 0.0000000
0
*CONTROL_CONTACT
0.0000000 0.0000000 0 0 0 0 0
0 0 0 0 0.0000000 0 0
0
*CONTROL_OUTPUT
0 0 0 0 0.0000000 0 0
*CONTROL_BULK_VISCOSITY
0.00001 0.00001
*CONTROL_ENERGY
1 2 1 1
*DATABASE_BINARY_D3PLOT
1.00000-5 0
*DATABASE_BINARY_D3THDT
0.1000000
*DATABASE_EXTENT_BINARY
0 0 0 0 0 0 0
0
0 0 0 0 0 0
*CONTROL_CPU
0.0000000
*MAT_USER_DEFINED_MATERIAL_MODELS
1, 1.e-9, 41,5,,,5,4

10.,1.,0.,5.,3.333
*MAT_USER_DEFINED_MATERIAL_MODELS
2, 1.e-9, 41,5,,,5,4

10.,1.,4.5,5.,3.333
*SECTION_SOLID
1 0
*SECTION_SOLID
2 0
*PART

1 1 1 0 0 0 0
0
*PART

2 2 2 0 0 0 0
0
Material Modeling
Results and datasetup for usr41b.inf:

Material Modeling

Material Modeling
Generalizations

1. Make the yield stress a function of the hardening parameter (metals-like):

0 y y
H
dt

= +
=
&

2. Make the yield stress a function of the total volumetric strain (foam-like)

0 y y v
v v
H
dt

= +
=
&

This requires storing history variables in the vector called hisv(*) for the yield stress value
(hardening parameter) or the volumetric/equivalent plastic strains

Hint:

For the first exercise, derive that;

, , 1 ,
, 1
, , 1
2
, , 1 , , 1 , 1
_
3
2
_
vm e n y n
vm en
f e n
f e n vm e n e n
H E
p

+
+
+
+ + +
=
+
= +

When is the hardening parameter equal to the equivalent plastic strain?

Confirm the increase in yield strength as a function of lambda for this exercise theoretically.

Material Modeling

Occupant Dummies and Restraint Systems

5. Occupant Dummies and Restraint Systems

5.1 Numerical Models of Occupant Dummies

Frontal & Rear Impact
Side Impact
Hybrid-3
SID
Eurosid

Historical Development of Crash Dummies

1960 ARL 5
th
, 95
th
6 & 3 year child
1973 ARL-GM HYBRID-2
6 year child (1975)
1979 UMTRI DOTSID
1985 GM HYBRID-3
5
th
, 95
th
(1988-1989)
6 & 3 year child (1994)
1990 GM
EEVC
BIOSID
Eurosid

Dummies and Legal Requirements

Situation in 1994:

FMVSS208 Hybrid-2 (mandated)
Hybrid-3 (optional)
FMVSS213 3 year old ARL (mandated)
6 year old H-2 (mandated)
FMVSS214 DOTSID (mandated)
BIOSID (considered)
ECE48 Eurosid (proposed)
Both child dummies will be replaced by the Hybrid-3 counterparts

New FMVSS208 will mandate Hydrid-3 50
th
and 5
th

Comparison of Side Impact Dummies

DOTSID Eurosid BIOSID
head 1000 HPC 1000 HIC
chest 85g TTI 1.0 VC
2.0 42.mm
85g TTI
abdomen 2.5 kN
pelvic 130 g 10. kN 130 g
channels 9 22 >70

Eurosid History

Development started in 1982 by EEVC as a European project (induced by poor biofidelity of
DOTSID)

Several institutes contributed based on their previous experience:
- INRETS: pelvic
- TNO: abdomen
- TRL: thorax
- APR: neck
- BASt: validation testing

1986: EUROSID-0 preproduction prototype

1990: development frozen as Eurosid-1

Eurosid Rationale

Since late 1970s, ECE had a biomechanics program combining accident research and testing
on corpses, some findings are:

During side impact, the head can be seriously injured when hitting cantrail, B-post or other
objects, the head motion must thus be correctly predicted, the head motion is very complex
and depends on the deformability of the neck

Pelvic fractures are rare during side impact and happen mostly in the pubic area, thus pubic
symphysis force should be measured

Abdomen is endangered since inner bleedings can be caused by intrusion of armrest or other
objects

In the thoraic area, serious injuries (rupture of arteries) are mainly caused by intrusion, rib
fracture is not so dangerous unless multiple ribs are concerned and the victim can no longer
be transported, thus the model needs multiple ribs in order to distinguish between local and
distributed force loads and intrusions should be measured, no doubt measuring accelerations
is also useful

The arm can reduce thoraic injuries if it gets trapped between thorax and impacting object,
thus Eurosid has a moveable arm, however this contributes a lot to variability
Eurosid Reliability

Testing by Volkswagen has shown non negligible scatter in the results obtained from simple
sled tests using Eurosid and a rigid barrier

Overall scatter of 78% was noted when measuring the VC criterion testing 4 dummies
multiple times under the same conditions:

Similar scatter for rib deflections and other criteria

Before use in side impact testing, the full assembled dummy should be tested rather then the
currently required component testing of the rib module only

This is also an argument for numerical models: the dummy model is more then the sum of
the individual components: full scale validation is required for many reasons (example:
eigenstresses induced by assembly of the components)

FEM Versus LPM Models

LPM models will always be useful for the simulation of belted dummies during frontal
impact since the behavior of dummy and structure can be uncoupled

This is not true in a side impact: door velocities and intrusions can differ 50% or more due to
the presence of the dummy, thus dummy and vehicle must be simulated using a single model

The Integrated Dummy-Vehicle Model

Contact definitions must represent the force transmission between dummy parts and vehicle
parts (example: thorax and door inner trim) that will both deform: the stiffnesses are of the
same order

Contact timing, contact location and contact area must be respected in the simulation if the
simulation is to give reliable information about the dummy behavior

This means for the dummy model:

Initial geometry of the dummy is needed with a degree of precision similar to the vehicle
model

Deformability of the dummy must allow the ability to account for a variable contact area
during loading

Local intrusions and loading may be important to predict injury criteria, their effect can only
be handled by a sufficiently fine mesh
Consequently FEM models are certainly better suited for integrated side impact simulations

Current Dummy Models

First FE dummy models appeared 1988-1989

Currently, dummy models are used ranging in size from 2000 to over 60000 finite elements,
their characteristics can be summarized by:

- Geometry: based on drawings or CAD?
- Validation: based on material testing, component testing, sled test(s) or full scale
test(s)?

Main problems with dummy models are:

- Foam and rubber material behavior is largely unknown and certainly under non-
uniaxial loading, surprises are possible
- Bulky parts ideally require extremely fine meshing that is out-of-range today,
certainly for full vehicle analysis
- Component validation does not seem to be sufficient to also characterize the fully
assembled dummy
- Real-life dummies show rather high variability

Example of a SID Model

Dozens of models exist worldwide for SID today

Even a very simple model can often do the job provided that it is numerically stable and
reproduces the dummy kinematics (gives a qualitatively correct answer)

Under these conditions, a trend prediction or A/B comparison may be possible

This model often used to examine restraint systems

A Simple FE-Model of SID

Built in 1988-1989

Coarse modeling with 332 brick elements and 1114 shell elements

Most parts are rigid such as head, legs and pelvic

Ribs are steel (elastic) and foams are simulated using material law 10 for numerical stability
Timestep 1.7 microseconds

Carefully respected mass distribution

A Simple FE Model of SID: Validation Tests

Validation was done by simulating:

4 drop weight tests on the damper, this leads to a reliable damper characteristic
Thorax impact test at 4.27m/s
Pelvic impact test at 4.27m/s

Even with this simple model, a visual inspection of the dummy kinematics during the thorax
impact allows to verify the dummy behavior.

The pelvic impact illustrates the good behavior for the brick elements simulating the pelvic foam
under the impact of the rigid pendulum

A Simple FE Model of SID: Conclusions

A certain discrepancy remains between the filtered measured acceleration data and the
simulation results

Better correlation (curve fitting) is possible with any model but in this case will not
contribute to the reliability of the model under real loading conditions

With this type of coarse mesh, the contact forces are spread over few nodes, any
improvement means mesh refinement first

This is an example of a dummy model with:

- Low development cost
- Low running cost
- High reliability (used in several 100 side impact simulations)
- Mostly interpretable results

Dummy Models with Extended Validation

More detailed models, based on CAD-data have been developed, partly in joint research
projects, examples are:
- ETA SID and Eurosid models
- Ove Arup SID and Eurosid models
- FAT SID and Eurosid models
FAT AK27 (Berechnung und Simulation) created FE models for side impact dummies (SID
and Eurosid) in Germany, the consortium members were amongst others:

- Mercedes-Benz
- Volkswagen
- Audi
- BMW
- Opel
- Ford of Germany
- Porsche

This project started in 1992

Development of FAT-Project

In phase 1 of the project (1993-1997), models were built of 12000 finite elements for SID
and 22000 finite elements for Eurosid

This is not exaggerated since:

- Deformable parts are modeled by solid elements, many element layers are necessary
- Even rigid parts cannot be modeled coarsely since a good distribution of contact
forces must be obtained
- Complex dummy geometry is respected to optimize contact modeling with the
structure

In phase 2 of the project (1997-1999), a further validation sometimes combined with mesh
refinement of the models is performed, the number of elements grows to 28000 for Eurosid
and 20000 for SID

Creation of a Phase-1 Crash Dummy Model

1: Procurement of drawings

2: Procurement of CAD-data: drawings will be insufficient for areas such as abdomen and
pelvic and must be complemented by digitizing the actual dummy

3: Meshing

4: Component testing: the dummy manual tests must be repeated but are not sufficient, for
example the neck pendulum test must be performed at minimum 2 speeds if any information
about the viscous rubber material properties is to be obtained

5: Component validation simulations

6: Performance of validation sled tests: the fully assembled dummy must be tested in a
simple sled or barrier test designed to reproduce the kinematics of the dummy during vehicle
testing while contacting only non-deformable bodies, for FAT in phase 1 a total of 5 different
barrier shapes were used at 2 different speeds for each dummy

7: Sled test validation simulations: all 10 sled tests must be simulated until correlation is
reached, since the dummy has over 10000 elements and 60ms must be simulated, 1 series of
simulations required about 40 hours of CPU on a single C90 processor

Creation of the Phase-1 SID Model

Geometry based on drawings

CAD-data were digitized for head, pelvic and legs (Porsche)

Component testing for validation (Audi):
- Head impact
- Neck and lumbar spine pendulum tests at 2 speeds
- Damper drop weight tests
- Drop weight test on single rib
- Drop weight test on thorax module (no arm)
- Thorax pendulum impact
- Pelvic pendulum impact

Sled tests on complete dummy for validation (Mercedes-Benz)
- Flat barrier
- Flat skewed barrier
- Barrier with armrest
- Barrier with bumper
- All sledtests performed at 19kmh and 25kmh

Creation of the Phase-1 SID Model

Detailed modeling of all parts

Calculation of resulting mass and inertia values, comparison to experimental data made
available by TNO

Simulation of all component tests

Simulation of all sled tests, eigenmodes of the barrier were important in determining the
pelvic acceleration levels
Distribution of the model to all members for testing in their proper full car environment

Creation of the Phase-1 Eurosid Model

Sled tests were performed by Opel.

The rib deflections are very dependent upon:

Arm kinematics, determined by the contact between clavicle and shoulder plates, the arm
motion would converge only for a relatively fine mesh of these very stiff components

Arm deformation: simulation needs careful modeling of all arm components (including inner
plastic plate), foam material is a factor of uncertainty

Sliding of the piston in the cylinder, this is a friction dependent rather chaotic phenomenon


Dummy Models Based on Real Physics

Material testing must be done in a controlled way in order to obtain constant stress, strain and
strain rate in the testpiece

No material testing was performed during phase 1 of the FAT side impact dummy project

Rather material data were generated to match component test data using a given mesh density
(component validation)

Component test
FE-model
Use material
data to fit
The goal of phase 2 was to base material data on real material testing (over 600 material tests
were performed) and create a dummy model based on the requirement of mesh convergence,
such a model then represents the real physics of the component

Component test
FE-model
Use mesh convergence
to fit
material testing
The current status can be found in:

Franz, Walz, Graf
Enhancements to the FAT FE dummies using specific features of LS-DYNA
Second European LS-DYNA users conference, 1999
Gothenburg, Sweden

Franz, Graf
Accurate and detailed LS-DYNA FE models of the US- and Eurosid: a review of the German
FAT project
LS-DYNA users conference, april 2000
Southfield, Michigan

Numerical Modeling of Frontal Dummies

A model of Hybrid-3 50
th
was developed by ERAB and the university of Linkoping (L.
Frederikson)

Important is the ability to change mesh density and switch from rigid to deformable parts in
different applications

A first validation was at component level (head impact, neck pendulum test and thorax
impact test)

Further validations were performed by users performing a variety of sled tests

Numerical Modeling of Frontal Dummies

Today a full suite of dummies is offered by Ove Arup and FTSS, Hybrid-3 50
th
, 95
th
and 5
th

5.2 Head Impact Simulations for MVSS201

Impactor modeling
Vehicle modeling

Hybrid-3 Head Models

Finite element model of Hybrid-3 headform:

Model Specifications:

- 4 brick elements through the thickness of the skin
- Aspect ratio below 1.5 in the impacted zones
- Separate modeling of all physical parts
- Element size ratio between skin and skull below 2
- Brick element modeling of headform parts (in order to obtain correct mass and inertia values)

Model developed by FTSS:

- Model size:
- 7 material properties
- 21673 nodal points
- 15890 brick elements
- 4.6kg headform mass
- Type 5 contact between skin and skull
- Accelerometer definition for output of accelerations in local axes
- Rigid body connection between skin and skull at the edges

Headform Model Validation

Mercedes-Benz performed a total of 20 validation tests:

Standard headform droptest

Horizontal droptest on a flat steelplate at 10, 20, 25 kmh

45 degree droptest on a 45 degree steelplate at 10, 20, 25 kmh

Horizontal droptest on a 20mm steelridge at 10, 20, 25 kmh

Horizontal droptest on a 20 by 20mm steel dice at 10, 20, 25 kmh

24 degree droptest at 24kmh on caryl foamblocks of 40, 60 and 80 g/l

24 degree droptest at 24kmh on neopoleen foamblocks of 60, 70, 80 and 90 g/l

Validation Simulation Results

Excellent correlation of horizontal and vertical acceleration components in all cases

The 20*20 dice is clearly the smallest obstacle that can still be impacted with this mesh and
yield realistic results

Modeling of the ridge and dice obstacles was necessary in addition to a rigid wall
representation

Contact type 13 with soft constraint gave good treatment of the contact between stiff bricks
(rubber skin) and weak bricks (foam) in all cases

Material 57 was used for all foams, stress-strain curves were available for different densities
and impact velocities, missing values could be interpolated linearly

MVSS01 Vehicle Modeling

Start from a full vehicle model, mostly this model can be cut and fixed at the waistline

Mesh with homogeneous element size of 6mm in all impacted zones, models will be large
but with a model of 100000 elements and 1 microsecond timestep, cpu-time will be
reasonable since simulation time is 15 milliseconds or less

Crush space between car body and trim is all important and very scarce, geometrical detail is
all important and the mesh should adhere very strictly to the CAD surfaces (chordal deviation
< 0.5mm)

Contacts must be defined using real part thickness rather then a uniform gap value since part
thickness cannot be neglected against the crush space

Model should contain glass parts (windshield, side windows), trim parts (with ribs) and
padding, with mesh sizes comparable to car body parts

Foam padding parts are best modeled using the type 10 tetra solid element

Modeling of buckling in plastic ribs may require the use of type 3 (thick shell) elements

Trim panels and ribs can be meshed independently and connected using null beam elements
and tied interfaces, the option TIED_SHELL_EDGE_TO_SURFACE is very convenient,
null beams on the rib edge are slave surface and the plastic trim panel is defined as master
surface

Model the null beams close to the shell surface (distance smaller then half the thickness of
the shells) and use the OFFSET option in order to avoid mesh distortion during the
initialization phase in LS-DYNA

Material models: 81 with damage for plastic trim panels, 83 for reversible foams and 63 or
75 for crushable foams (see part on material laws)

5.3 Numerical Simulation of Airbag Deployment

In-position
Out-of-position
Driver bag
Passenger bag
Side bag

The Engineering Problem of Airbag Development

Equivalent to saving the life of a person falling face down from the third floor using an inflatable
cushion that is detonated about 60cm before impact.

_ _ :
50 15 /
impact velocity
x kmh m s = &

2 2
2
_ _ :
9.81 / 10. /
1.5 _( _ )
11 3
2 2
equivalent drop
g m s m s
x gt t s final velocity
gt xt
d m stories
=
=
= =
&
&

The energy needed can be of the same order as the entire vehicle kinetic energy:

2
2
_ _ _ :
2
2000*15
225
2
k
k
vehicle kinetic energy
mx
e
e kJ
=

&

_ _ _ :
_ 30 :
0.03*1000*600 18
_110 :
0.15*1000*800 120
p i
total airbag enthalpy
h mC T
driver l
h kJ
passenger l
h kJ
=

Only a relatively small part of the airbag energy is used to create an overpressure in the bag
(=the safety aspect)

An analysis of the use of the airbag energy is given in [1]

Reference

[1]: Energy and Entropy in Airbag Deployment
The Effect on an Out-Of-Position Occupant
Nusholtz, Wu, Wang, Wylie
SAE 1999-01-1071

Goals of Airbag Simulations

Develop reliable predictive simulation tools

Develop an analytical capability to generate a restraint system for a given car body

Generate impact system specifications for sensors, inflators and modules

Verify the response of the OOP dummy

Reference:

John Cooper
Breed Technologies
European development Center
Coventry UK

Aspects of Airbag Modeling

Generator characteristics
Tissue properties
Geometry and folding
Contact and unfolding
Aerodynamics
Mass losses
Validation

Experiments are necessary:

Tissue tensile stiffness in different directions
Tissue porosity
Generator tank test
Static airbag inflation
Static & dynamic pendulum tests

Airbag Model Type 5 (Wang-Nefske)

Pressure is assumed constant inside the control volume.

Conservation of energy in the airbag can be derived from the general expression of energy
conservation for an open unsteady system (neglecting kinetic energy):

_ _ _ _
_
_
_
_ _ _ _ _ _
i i o o
i
o
e mh m h q pV
e internal energy in the tank
m incoming mass
m outflowing mass
q heat loss
pV rate of work done by expanding gas
=
=
=
=
=
=
&
& & & &
&


Inflator Modeling

Many gas generators still use nitrogen gas (N2)

For a diatomic gas, specific heats are not very dependent upon the temperature and can be
considered constant in a first approximation

Mass flow and inflator temperature are determined in a tank test: an isochoric experiment

Typically a 60 liter or 120 liter tank is used, originally filled with air at atmospheric pressure

Gas
generator
p V
venting
porosity
In an isolated tank of constant volume conservation of energy is reduced as follows:

0
0
_ _ _ _
_ _ _ _ _
i i
i
i p i
v
q V m
e mh
h specific enthalpy of incoming gas
e energy of gas in the tank
h C T
e C Tm
= = =
=
=
=
=
=
&
& &
& &

In general (gas mixtures), this can be a complicated expression, but if we assume that the specific
heats of the incoming gas and the gas originally in the tank are the same, then:

0
0
v v v i p i
t
mC T mC T mC T mC T
m m mdt
+ + =
= +
& &
& &
&

Assume that:

Air in tank and injected nitrogen have same specific heat
Specific heat does not depend on the temperature

Then:

Perfect gas EOS in the tank:

pV mRT
pV mR
pV mRT mRT T T
mR mR
=
= + =
& &
& &
& &

Conservation of energy:

( ) 1
i i p i
i p v v
i p v v
v
i p
i p i
mh mC T e
mTC mTC mTC
pV m
mTC mTC mC T
mR m
p VC
m
T RC
pV pV
m
RT C T
= =
= +
| |
= +
|
\ .
=
= =
& & &

&
& &
& &
& &
&
&
& &
&

The flow rate can thus de deduced from the measurement of the inlet temperature and the tank
pressure.

Practical problems:

Temperature measurements are very difficult due to the high temperature rates
Pressure measurement in a tank test has a relatively high spread even for inflators from the
same batch

A mass averaged inlet temperature is estimated from the total injected mass as follows:

,
_
i ave
m injected mass mdt
V p
T
R m
=
&

Such a constant inlet temperature is often used in simulations. The total injected mass is easily
determined from the inflator weight before and after the tank test.

Better precision can be obtained by performing a numerical simulation of the tank test.

Chambered bag must be used.

The use of a 2-chambered bag model allows consideration

The tank initially filled with air at room temperature, simulates the injection of hot nitrogen or
other gases:

Input data are:

Gas characteristics
Temperature of incoming gas
Mass flow of incoming gas

Check:

Tank pressure (chamber 1 = chamber 2)
Total injected mass

Temperature and mass flow curves can be adapted to generate the correct tank pressure while
respecting the total amount of injected mass.

A mixture of gases with temperature dependent specific heats is used:

( )
2
1
_ _ _ _ _ :
p
v p
n
i i i
i
C a bT
C C R
mixture of n gas fractions
e a T bT RT m
=
= +
=
= +

Chamber 2
Cp(air)
Chamber 1 (V1<<V2)
Cp(N2)
Typical values for nitrogen (N2) are:

Temperature
(K)
Cp
(Nm/kgK)
Gamma
273 1025.85 1.405
400 1049.56 1.393
600 1086.89 1.374
800 1124.20 1.357
1000 1161.51 1.342
1200 1198.80 1.328

Typical values for air are:

Temperature
(K)
Cp
(Nm/kgK)
Cv
(Nm/kgK)
273 1004,64 715,81
400 1008,83 719,99
600 1021,38 732,55
800 1042,31 753,48
1000 1059,06 770,22
1200 1084,17 795,34

Typical values for water vapor are:

Temperature
(K)
Cp
(Nm/kgK)
Cv
(Nm/kgK)
273 1854,40 1393,94
400 1866,96 1406,50
600 1908,82 1448,36
800 1971,61 1511,15
1000 2030,21 1569,75
1200 2101,37 1640,91

Clearly the effect of mixing becomes important if water vapor or other polytropic gases are
released

Tissue Modeling

Tissue has some bending stiffness, mostly this is neglected in the simulation
Membrane elements (type 9 to avoid hourglass) are used consequently
Alternatively C0-triangular elements (type 4) with 1 IP through the thickness to ensure
membrane behavior, this will double the number of elements
Tissue behavior is nonlinear and anisotropic, approximated by using material law 34: linear
orthotropic. Too high difference in moduli in the warp and weft directions will lead to
instabilities
Wrinkling is simulated by suppressing the compressive stresses, an elastic liner allows to
keep some compressive stiffness if desired (coating layer?)
Shear moduli must be input and are best validated against a 45 degree tensile test
Poisson modulus is estimated between 0.2 and 0.4, this may actually influence simulation
results
Density is determined as weight of the tissue divided by the volume (surface*thickness) to
implicitly take the seems into account

Bag Geometry: Folded Bags Versus Reference Geometry

Unfolded airbags can be modeled with a few 100 elements

Realistically folded bags need between 4000 and 8000 elements for a driver bag, between
12000 and 24000 elements for a passenger bag, this is based on an element sidelength of
between 10. And 20.mm

Contact definitions must include tissue-to-tissue as well as tissue-to-environment

Folding has been shown to greatly influence the interaction with the pendulum and/or
dummy, particularly if contact occurs in the early phases of the deployment

Realistic folding involves a large number of folds and a distance between the tissue layers of
the order of the fabric thickness (0.3 to 0.4mm), only this way can the airbag be fitted in the
model of the containment and cover assembly

The single most important issue in performing an out-of-position airbag simulation is to
obtain a realistic undeformed geometry:
- Correct airbag geometry
- Correct folding pattern and fold thicknesses
- Correct containment and cover
- No initial penetrations

This task is best accomplished with LS-INGRID

Airbag Folding with LS-INGRID

Interactive folding in-plane and out-of-plane possible

Mesh lines should be positioned on the fold lines, if possible 2 parallel mesh lines at equal
distances left and right of the fold line

Try to simultaneously avoid tangling, minimize initial penetrations and minimize distortion

With the classical thin fold logic, a certain amount of initial penetration relative to the tissue
thickness cannot be avoided:

Here l is the element size (about 10mm) and t is the layer thickness (about 0.3mm),

Consequently, d is of the order of 0.015mm

Increasing t by moving the inner tissue node inward will reduce penetration by increasing d,
but additional distortion will be generated

t is defined directly in the fdef command and t is obtained in the fdef command by scaling t

t=t
t
t
d<t
t=t
l
t=t
t
t
d<t
t=t
l
t is the distance of the newly folded layer to the nearest layer and is defined directly in the
pfold command

Try to limit ALL initial penetrations to p<9t/10, then an initial contact thickness of t/10 (about
0.03mm) will result in no initial penetrations

Use the area check (tmass) of the airbag tissue to minimize distortion: error on tissue area
should certainly not exceed 5%

Large folds should be used as soon as the vertical side of the fold contains a non-negligible
part of the tissue area

Airbag Deployment and Contact Problems

Airbag unfolding has 4 characteristics that make the contact problem particularly
challenging:

- High nodal velocities (of the order of 100m/s)
- Initial penetrations
- Thin master segments (about 0.3mm)
- Layer thickness smaller then element size

With a classical contact algorithm, the slave node can penetrate several master segments
between 2 global searches:

10*100 / *0.001 1. d Nx t mm ms ms mm = = = &

(This could correspond to 3 tissue layers)

Contact type a13 solves these problems although cpu cost can be rather high, option SOFT=2
(pinball contact) adds reliability

Segment based search necessary

Contact type a13 has no friction although tissue-to-tissue friction may absorb some energy
during unfolding

Initial penetrations are avoided by using a time dependent contact thickness (user-defined),

This allows to adjust the mesh in the fold corners gradually during the first milliseconds of
the simulation,

The corresponding deformation in the tissue can be recovered

Aerodynamic Effects

During the initial deployment, the assumption of a constant pressure in the bag is wrong

3 empirical features can be used to consider aerodynamic effects in the bag:

- Jetting
- Pressure delay
- Chambered bags

These effects can be important in the simulation of folded bags, where fully coupled
Lagrangian-Eulerian simulations are not yet realistic

Jetting

Jetting is an empirical way to take the kinetic energy of the injected gas into account

Conservation of energy (including kinetic energy) can be written for steady flow along a
streamline in terms of specific enthalpy and kinetic energy as follows:

tc
t
t/10
0.
2.ms
time
1
2
( )
1
1 1
1
1 1 1 1 1 1
2
2
1 2
2
v s v p
p
h e
h C T pV C R T C T
x
h h
= +
= + = + =
= +
&

If we compare the 2 terms in the last equation for a typical 60 liter driver side airbag:

0.030
30000. 1000
1300. 1000 /
300 /
k p
m kg
h Nm T K
e Nm C Nm kgK
x m s
=
= =

`
= =

=
)
&

The kinetic energy of the incoming gas is small compared to the enthalpy

The virtual origin can be a point (driver bag) or a line (passenger bag):

Cone shaped jet:

z
Wedge shaped jet

In the jetting model, a supplementary dynamic pressure is applied to those tissue elements of
the bag model whose centerpoint is in the line of sight to the virtual origin of the inflator:

( )
2
s d
gas tissue
p p p
mRT
p x x
V

= +
= + & &

Warning: beta (jet intensity factor) is user-defined and high values may severely overestimate
the kinetic energy in the bag

Assumptions are made about the velocity profile of the gas in the code or through user input,
for a given distance z to the origin of the inflator, the gas velocity can then be estimated from
the total massflow and this velocity profile

Consider the example of a cone shaped jet where we assume a Gaussian velocity profile over
each circular cross section of the cone

z
r
v
z

( ) ( )
2
,
r
ztg
v r z v z e

| |
|
\ .
=

Conservation of mass gives the maximum velocity:

( )
( )
( ) ( )
2
0
1 2 2 2 2
2
1
1.58
1
r
ztg
m rv z e dr
m m
v z
e z tg z tg

| |
|
\ .
=
=
&
& &

See also [Nuscholtz, 1995]

The gas velocity is seen to decrease with the square of the distance to the virtual origin
(inflator)

Reasonable jetting parameters can only influence the early phases of the unfolding

Pressure Delay

The arrival of the gas in any point of the bag is estimated as a function of the distance to the
inflator and the average velocity of the gas, pressure is applied after arrival of the gas

For a sonic gas velocity, the delay in a typical driver bag would not exceed 1.ms

V=300m/s
R=300mm
T=R/V=1ms

In practice, delays of 5-6ms can be deduced from video recordings and the average velocity
could be 50-60m/s (due to losses in the folds?)

D=600mm

Chambered Airbags

The chambered bag technology can be applied using shell elements with a null material to
create separate bags for every fold

The result is a discrete pressure delay, however the effort in data preparation is very high if
many folds are present

Venting must be defined in the chamber where the physical venthole is located

Mass Loss in the Airbag

Venting is easily defined by giving the area of the venthole or modeling the actual venthole
as a separate PID, in the latter case the time dependent size of the venthole due to tissue
stretching is automatically taken into account, in the other case venting area can be given as a
function of time

If the bag is not coated, porosity must be taken into account additionally, 2 factors determine
the mass loss through porosity:

- Porosity of the tissue (function of overpressure)
- Amount of uncovered tissue

LS-DYNA calculates the amount of uncovered tissue through which mass losses can take
place

The porosity must be inputted as a function of bag pressure, experimental data are needed

Chamber 2
Chamber 3
Vent
Chamber 1
Gas Generator
Validation Testing

As an example we consider a series of pendulum tests conducted at Millbrook Proving
Grounds (UK):

- Unfolded bag with static and dynamic pendulum
- Folded bag with static and dynamic pendulum
- Folded bag with cover with static and dynamic pendulum

The idea is to validate airbag model parameters by consecutive testing

The static and dynamic tests show that:

- Accelerations are much higher in unfolded tests
- Cover delays the acceleration peak
- Addition of the fold requires significant amount of inflator energy

Additional testing may include impact(s) on a static dummy, sled tests and full vehicle tests

The Reference Metric Methodology

The first application of a reference metric technology to airbag simulation was probably due
to Hoffmann (1989).

Stress in the tissue is calculated by a total Lagrangean formulation with respect to a reference
state X, take the example of a truss element (1D):

1
1
1
dx
tension
dx
dX
E
dx dX
compression
dX
>
| |
=

|
\ .

<

In a tissue material compressive stresses are set to zero and if the elements in the initial state
are all in compression with respect to the reference state, no stresses are generated

The simulation then requires a reference geometry for the airbag tissue corresponding to the
real size of the bag in any (example: unfolded) configuration

The initial geometry is arbitrary and must only fulfill the following conditions:

- Same topology as reference geometry
- Small enough to fit the containment
- Nodes connected to the containment (or housing) must be on their correct coordinates
- All elements in compression with respect to the reference geometry (otherwise
undesirable damping is required)
- No tangling
- No initial penetrations
- No negative areas

Such an airbag model will unfold without generating stresses in the tissue until tensile
stresses are generated once the elements reach their reference size

The tissue model is projected with the correct kinetic energy since the reference bag mass is
used, this also guarantees a reasonable timestep value for the explicit integration

The pressure and volume will therefore develop much like in an unfolded bag model
Initial bag geometry can be generated in many ways from the reference geometry:

- Scaling
- Projecting in the plane of the housing
- etc...

Validation Simulations

Carefully simulate the pendulum test conditions:

Pendulum is best modeled in LS-DYNA as a flexible body with correct mass, inertia and
swingperiod, model consists of shell elements (pendulum) and beam elements (arm)

Modal analysis can be done implicitly, then compare primary mode shapes and frequency
results with results of a hammer blow test on the pendulum

This allows to account for the eigenfrequencies of the pendulum in the simulation results if
necessary

Positioning of the airbag must be very carefully done in the simulation model (relative to the
pendulum)

Then use the pendulum test simulation to perform parametric studies on those input parameters
that cannot be measured:

- Poisson ratio of the tissue
- Jetting and pressure delay parameters (typically will determine the time of first contact)
- Tissue resistance against wrinkling (liner thickness)
- System damping (kept at low values)
- (Venting and porosity)
- Contact stiffness, gap etc...

Compare pendulum velocities/accelerations to test results and if possible results animation to
video recording of the actual test.

Examples of Validation Studies for Airbag Deployment

Application examples from Mercedes-Benz AG (Sindelfingen, Germany).

References:

Hirth, Du Bois
OOP Airbagsimulation fuer ein MB-Fahrerairbagmodul
LS-DYNA users conference, Goettingen, 1995

Hirth, Du Bois
OOP Airbagsimulation fuer ein MB-Beifahrerairbagmodul
LS-DYNA user conference, Fulda, 1997


Component Models in LS-DYNA

6. Component Models in LS-DYNA

Example of NHTSA side impact
Application to new MVSS208
Input sections concerned:

- Interface definitions for component analysis (section 11)
*INTERFACE_COMPONENT_NODE
- Nodes moved via section 11 interface file (section 34)
*INTERFACE_LINKING_DISCRETE_NODE_NODE

Example of NHTSA Side Impact Simulation

Simulation is on Opel Omega and was performed in 1994 for Adam Opel AG and Breed
Technologies

Total model size is about 100000 finite elements, following components are contained:

- Car body with typical element size 10mm in side structure
- Front and rear doors on impacted side
- Front and rear door trim
- Door hinges and locks, modeled with spring elements and rigid bodies
- Wheels, suspension system modeled with spring and beam elements
- Rigid wall representing the laboratory floor
- Gravity activated at t=0.
- Front seat steel structure
- Front seat cushion, back & covers
- NHTSA honeycomb deformable barrier
- Front and rear SID (coarse FE models)
- Single type 13 (AUTOMATIC_SINGLE_SURFACE) contact definition with soft
constraint for all impacts between barrier, vehicle and occupants


Specific Problems of Side Impact Modeling

Barrier positioning with respect to vehicle:

The barrier lower edge must be 300mm above the floor, this does not specify the position
with respect to the car that must be obtained from measurement in the lab.

Representing the lab floor is necessary to avoid exaggerated rotation of the car as the barrier
slips over the sill into the passenger compartment

exaggerated roll must be prevented:

Barrier
d
door
Applying a gravity field at t=0 leads to a dynamic amplification on the gravity load, spurious
oscillations are induced

Positioning of the dummies with respect to the vehicle:

- Positioning of the undeformed dummy in the H-point on the undeformed seat cushions
will create important intersections in the model

- One possibility is to do a pre-run to position the dummy at a fixed velocity and deform
the cushions, a typical value is (for dummy initial position 80mm above the H-point):

- This is not an acceptable solution since high kinetic energy of the impact will deform the
seat during the consecutive side impact simulation

- Very slow positioning is excluded due to cpu

- One solution is to dump deformed geometry after the pre-run using LS-TAURUS and use
this geometry as the undeformed geometry for a new side impact simulation with
correctly positioned SID and accordingly deformed cushions (neglecting the initial stress
in the cushion under the dummy weight)

- At least the material law of the foam in the cushions must be shifted in order to reach
densification for the correct remaining cushion thickness
g
large ramptime?
t
Dummy positioning prestresses the cushion, timing to densification should be respected:

H

Prepositioning
D=80mm
T=10ms
V=8m/s
Initial
stress
initial geometry
Reference geometry
shifted stress/strain

This model cannot be considered state-of-the-art today for a number of reasons:

- More detailed modeling of hinges and locks
- More detailed modeling of inner door components such as window guiderails, electrical
engine etc...
- Use of more sophisticated dummy models
- Use of reference metric in prestressed foam parts such as the cushions
- Implicit initialization under gravity (v960)

Simulation results:

- Correlation is good for longitudinal and transversal velocities of car body cog and barrier
- B-pillar and lower door velocities also correlate well
- Peak velocities in upper parts of the front door arrive about 4 milliseconds late in the
simulation
- Dummy accelerations are too late (again about 4 to 5 milliseconds) and up to 100% too
high

In order for this model to be useful, the 4 millisecond delay in upper door velocities must be
explained and corrected!

Matching the longitudinal barrier velocity requires a good functioning of the contact in the C-
pillar area:




Component Models

In a full model simulation, a component section is defined as a set of nodes whose x-y-z
velocities as a function of time are stored on a binary file

The component section is defined as an interface for component section definition (manual
section 11) and velocities are stored using a time interval given on control card 18
(CONTROL_OUTPUT) as parameter OPIFS

In a component model simulation, the section nodes are given fixed velocities that are read
from the previously created binary file and constitute a boundary condition for the
component model that is equivalent to the presence of the full model beyond the section

Component model for side impact:

a component model was built containing:

- front SID
- NHTSA barrier
- Front door and trim
- Front seat and cushions

the component section where boundary conditions are given as fixed velocities consists of all
nodes at ;

- front floor
- rocker
- A- and B-pillar
- Roofrail
- Rear of barrier
The component section contains a total of over 20000 nodes

So 60000 time histories were stored every 0.1 milliseconds in the full vehicle run, a total of
36000000 words or about 286Mb of information were automatically stored by the component
section definitions in LS-DYNA

Use of the Side Impact Component Model

The numerical results are equivalent to the full vehicle simulation if no model changes are
made

cpu time was down from 75 hours to 25 hours on a J90 Cray system

Component model is valid as long as non-structural countermeasures are investigated:
padding, sidebag, trim changes... anything that does not influence structural behavior beyond
the component section

The full vehicle car body is used as a boundary condition, so this approach can be considered
a numerical side impact sled

The number of parametric studies performed in a given time could be substantially increased

Final results of parametric studies on a component model should alaways be verified with a
full vehicle simulation

Results of Parametric Study on Component Model

4 milliseconds delay in upper door velocities could be attributed to poor modeling of inner
door components and corrected accordingly


Application of Component Models to New MVSS208

The new MVSS208 frontal impact proposal comprises an important number of load cases:

50%H3 50%H3 5%H3 5%H3
belted unbelted belted unbelted
-30degree
+30degree

The total number of load cases is thus 4 times the chosen number of impact angles

If a single full vehicle simulation is performed for each impact angle, the 3 remaining cases
can be investigated using a component model containing dummy, seat, restraint system,
steering column/wheel, dash panel and kneepolsters

CPU cost of such a component model can be an order of magnitude below the full vehicle
model


Quality Assurance of Numerical Models

7. Quality Assurance of Numerical Models

7.1 QA of Numerical Models Before Analysis

A summary of modeling rules based on elements type 2&4

Mesh:

Have a minimum of 3 elements per side of any section in the entire vehicle (or 1 fully
integrated element)
Have a minimum of 6 elements per buckle in the energy absorbing part of the structure
Have a timestep of 1.0 microseconds without mass scaling initially
Limit the number of triangular elements to 10% of total
Limit the warpage of any element to 15
Identify zones (PIDs) with a majority (>80%) of warped (w>2) elements
Have mesh lines parallel and orthogonal to sheet borders (regular meshing)
Use uniform meshing in the energy-absorbing part of the structure

Contacts:

Avoid edge-to-edge penetrations (intersections)
Avoid all initial penetrations for a contact thickness of minimum 0.6mm
Work with contact thickness equal to the real sheet thickness (SSTHK=1 on
CONTROL_CONTACT)
Set ignore=1 on CONTROL_CONTACT
Have sufficient mesh density to ensure functioning of the contact with sufficiently small
penetrations
Have sufficient mesh density to correctly simulate contact pressure distribution
Avoid premature release (contact thickness) (PENMAX): make sure maximum penetration in
all contact definitions exceeds 0.45mm (80% of 0.6)
Use a uniform mesh in the contact zone (sidelength ratio below 2)

Connections:

Proper modeling of bolts and bushings
Founded choice of spotweld representation (connect nodes or elements by tied interfaces?)
Avoid discontinuities in spring force-displacement curves

Numerical integration:

Use 2 integration points through the thickness by default
5 integration points if part thickness exceeds 1.5mm (Gauss-Lobatto rule?)
Use the node invariant local system
*CONTROL_ACCURACY
INN=2
Use underintegrated (type 2) elements by default
Use ANS element (type 16) for narrow brackets in zones that do not deform very much

Use type 16 element with hourglass type 8 to represent structures that remain mainly elastic
Use hourglass formulation that does not prevent rigid body modes (Belytschko-Tsay) in
shells (Default)
Use hourglass formulation that does not prevent rigid body modes (Belytschko-Flanagan) in
bricks (IHG=2,3 viscous or 6 stiffness)
Do not add warping stiffness (BWC=1) to the type 2 element, (this applies to the entire
model) rather use type 10 elements for parts with a majority of warped elements (use full
projection scheme, PROJ=1)
If shell thickness update for crash problems is used, activate bulk viscosity for shells

Material values:

Base the data on test results corresponding to the actual material!!!!
For metals, use material type 24 with table-type input
Recoverable foams are best modeled with materials 57 (non rate-dependent) or 83 (rate-
dependent)
Avoid discontinuities in the derivative of the stress-strain curves
Ensure sufficient energy can be absorbed in any foam part for a deformation that can still be
represented by the mesh
Use proper visco-plasticity to simulate rate-dependency in metals (v950, VP=1)
Use proper damage models to simulate materials that exhibit strain-softening (material 81 in
v950)
Do not expect predictive simulation of ductile and/or brittle rupture unless a very fine mesh
is used in conjunction with well-understood 3D failure criteria (not available?), this is most
important for the brittle rupture of engine mount castings

7.2 Post-Processing: QA After Analysis

1. Check numerical stability:

Normal termination message (d3hsp)
Visual inspection of the deformed shape (shooting nodes) (d3plot)
Energy balance: total energy must be constant if no energy is added to the system (glstat),
numerical integration must not destroy nor generate energy
If all this is ok, Courants criterion was respected during the numerical integration

2. Check numerical accuracy:

The ratio of numerical energy over physical energy should be small in every subset of the
model:

_( _ _ )
_( )
he
for every PID
ie
he sie
global
ie

The check per PID may be particularly useful to trace down local instabilities that may have
globally affect results

Check the mass error global (d3hsp, glstat) and locally (matsum, per PART)

3. Check if the model behaved the way you assumed (=was the model suitable for the physical
problem):

Does the mesh give a sufficiently smooth representation of the deformed structure?
Did penetrations occur? (input error, release conditions, edge-to-edge...)
Are equivalent plastic strains in all metallic parts reasonable (below 30% or 0.30, necking)?
Is the kinetic energy smaller then 1% of the internal energy in the case of quasistatic loading?

4. Is the solution physically plausible?

Check the consistency of all outputted results:

Cross check of nodal accelerations with section forces
Cross check of buckling mode with section forces
cross check of plastic strain levels with energy absorption
cross check of buckling mode with energy absorption

Example:

Then:
sec
:
tion wall
mass
wall mass
structure wall mass
average mass
f f
x
f mx m
t
ie f ds
and
f t m x
&
&&
&

A note on aliasing:

In LS-DYNA the velocity is integrated from the acceleration every timestep:

1/ 2 1/ 2 n n
n
x x x
x x
t t
+

= =

& & &
&& &&

Accelerations and velocities of nodal points, stored in the NODOUT files, are typically sampled
every 40 to 100 timesteps (0.1 to 0.04ms sampling interval)

The high frequency contents of the acceleration signal in particular can lead to aliasing of the
stored signal:

Rigid wall
structure
section definition
rigid mass
Initial velocity
a
t
Aliasing does not conserve the integral and thus for the stored time histories one finds that:

x
x
t
&
&&

How to counter aliasing:

Smaller sampling intervals:

Fighting the symptoms and not the disease, the problem is caused by numerically high
frequency noise

Consider the time derivative of the velocity:

The velocity is a smoother function and aliases less, this works only in the global inertial
system though and NOT for accelerometers where velocities and accelerations are stored in a
local frame

Prefiltering before storing the data

?

Avoiding numerical noise is best done by modeling the geometrical extension (as a rigid
body) and the mass (10 to 50 gram) of the accelerometer, this will reduce the local frequency
contents of the acceleration signal considerably

Acceleration signal frequency contents:

The highest frequency in the system can be estimated from the timestep:

2 2
1 300000
2
t s Hz

= =

This can be considered valid for a node with a typical nodal mass in the BIW structure, about 0.5
gram for a BIW of 200 kg and 400000 nodal points

The mass of the accelerometer (with fixture) is 25 to 100 times larger and the frequency thus 5 to
10 times smaller, the noise will be cut between 30000Hz and 60000Hz

Little prefiltering should be necessary with a sampling interval of 0.04ms (25000Hz)

If aliasing remains obvious with the above procedure, it becomes a debugging tool to detect
spurious noise caused by numerical oscillations

5. Is the solution reliable?

How dependent is the result on friction?
How dependent is the result on rupture phenomena in spotwelds, metal sheets or other
components?
How dependent is the result on other phenomena that are difficult to simulate predictively?
(example: coarsely or equivalently modeled components and parts)
The answer may require a few parametric iterations and/or additional validation testing

Present methodology is iterative:

Load case and performance
Criterion definition
Prototype design choice
Numerical model & analysis
Result Plausible?
no
no
yes
yes
yes
Result Reliable?
Optimum response?
Prototype & test
no
6. Is the solution predictive?

Compare to previous analysis and test results performed on similar vehicles (differences
between model and actual car?)

Comparison of Test and Simulation Results

Test Results
Laboratory testing:
- Boundary conditions?
- Recordings?
Car
Vehicle description:
- Complete?
- Up-to-date?

Car Concept:
- CAD-data
- Package description
- Material testing
- Component testing
- ...
Data preparation
-Correct geometry?
-Suitable mat. law?
Analysis
- Stable?
- Mesh convergent?
Simulation results
FE-Model
Modeling of Deformable Barriers

8. Modeling of Deformable Barriers


Overview of Deformable Barriers in Automotive Testing

barrier Legal requirement

ODB ECE R94 (56kmh)
European NCAP (64kmh)
MDB
950kg
ECE R95 (50kmh)
MDB
1300kg
MVSS214 (35mph)

Currently the deformable parts of these barriers are made of aluminum honeycomb structures

MAT_HONEYCOMB is used for the moderately deforming side impact barriers (see chapter
4)

MAT_MODIFIED_HONEYCOMB is used to model the severely deforming ODB (see
chapter 4)

Different levels of approximation can be considered to model the ODB in particular
Cross sectional view of the ODB barrier:

The numerical modeling of the ODB shows many problems:

Bumper blocks tend to disconnect from the body because of failure in the adhesive

Slipping of the cladding sheet over the body block

Cladding sheet has a rupture strain that seems of the order of 7%, punch through occurs

Very high deformation gradients, tear-off shear type failure in the honeycomb body block

Deformation seems chaotic in the experiment

The problem is certainly not limited to a material model for the honeycomb

A fully predictive model seems impossible due to the many different failure mechanisms

Aluminum
Honeycomb
adhesive
Body Block
Aluminum cladding sheet
3 * bumper block +
aluminum cover sheet
Different Models of ODB

3 levels of approximation can be considered:

Spring model
Hybrid model
Brick element model

Spring Model for ODB Barrier

All nodes fixed 011-111
Aluminum sheets modeled with shell elements
Aluminum honeycomb modeled with springs representing longitudinal compressive stiffness
of the blocks

Problems with the spring barrier model:

No detachment of bumper nor cladding sheets
High membrane loads in alu sheets
Too stiff versus local intrusions, tunneling cannot occur

Body
Block
Cladding sheet
Hybrid Model for ODB Barrier

Aluminum sheets modeled with shell elements
Aluminum honeycomb body block modeled with springs representing longitudinal
compressive stiffness of the blocks and null shells on the surface
Aluminum honeycomb bumper block modeled with brick elements (MAT_HONEYCOMB)

The model is connected as follows:

Tied interface between bumper blocks and their cover sheets

Sliding and voids interface between cladding sheet and null shells on the body block

Tied interface with failure representing the adhesive between cladding sheet and bumper
blocks

No rupture is considered in the cladding sheet

Null shells
springs
BC 011-111
This is a compromise model with very moderate development effort

Main limitation is still the BC on the nodes of the null shells that may block the transversal
motion of the vehicle if deep penetration in the barrier occurs

Brick Element Model of the ODB Barrier

No more BC on nodes representing the barrier

Brick element representation of the body block (MAT_MODIFIED_HONEYCOMB), see
chapter 4 for shear failure

Further model features:

Failure in cladding sheet

Representation of the adhesive

Over 20000 brick elements in a coarse model of the body block

High development effort

Model developed for FAT by GNS

Very extensive validation testing





Optimization of Structural Crashworthiness

9. The Optimization of Structural Crashworthiness
by Massive Use of Numerical Simulation

Optimization of structural crashworthiness:

Should ideally be coupled to optimization of weight, noise and fatigue behavior of the
vehicle
May easily require hundreds of numerical simulations
Requires a reliable numerical model where numerical sensitivity is negligeable with
respect to physical sensitivity

Often applied in industry is the use of optimization techniques to fix a specific problem of the
vehicle structure with respect to its crashworthiness

In this case, a Taguchi approach can be sufficient while cost and time effective

Formal Description of the Taguchi Process

Choose a performance criterion: examples are selected dummy injury values, rocker base
acceleration, floor panel intrusion, door intrusion velocity,... structural criteria should be
prefered to dummy related numbers since there simulation shows less sensitivity

Select a total of typically 7 structural parameters that are felt to influence the performance
criterion, most common is to consider the thicknesses of selected structural members

Run a baseline simulation to asses the current structural performance

Now assign reasonable minimum and maximum values to the 7 selected parameters, if sheet
thicknesses are used a zero means the part is non-existant

Then run a total of 8 simulations as follows:

P1 P2 P3 P4 P5 P6 P7

T1 T1 T1 T1 T1 T1 T1 S1
T1 T1 T1 T2 T2 T2 T2 S2
T1 T2 T2 T1 T1 T2 T2 S3
T1 T2 T2 T2 T2 T1 T1 S4
T2 T1 T2 T1 T2 T1 T2 S5
T2 T1 T2 T2 T1 T2 T1 S6
T2 T2 T1 T1 T2 T2 T1 S7
T2 T2 T1 T2 T1 T1 T2 S8

From the 8 simulation results, choose the optimum values for all 7 parameters, with a
Taguchi setup this is easily done by hand

Example for P1:

If the intention is to minimize the performance criterion then the choice is made as follows:

( ) ( )
5 6 7 8 1 2 3 4
1 4 2 1 4 2 1
0 1
0 2
opt
opt
S S S S S S S S S
P t t t t
S
t t
t
S
t t
t
+ + + + + +
=

> =
< =

This must be done for all 7 parts

Then construct a verification case with the selected thciknesses for each part and evaluate
the improvement in structural behavior versus the baseline.

If sufficient improvement is not reached, start over choosing different parameters starting
around a new baseline or selecting new parameters.

S
dS/dP1
S5
S6
S7
S8
t t2
t1
S1
S2
S3
S4
Limitations of this Approach

Difficult to prevent conflict with other design variables then the performance criterion
selected

Functions best when the performance for a certain loadcase is fairly well known and must be
improved

Since a linear variation of the performance criterion with each of the 7 parameters is
assumed, the range in a single Taguchi iteration should not be too large and multiple
iterations should be used:

Background and Rationale

The basic assumption is that the performance criterion is a linear function of the 7 parameters
(part thicknesses):

0 1* ( 1) 2* ( 2) 3* ( 3)
4* ( 4) 5* ( 5) 6* ( 6) 7* ( 7)
S A A t P A t P A t P
A t P A t P A t P A t P
= + + + +
+ + +

In this case 8 experiments are needed to determine the 8 coefficients (gradients) AI:

S
t11 t12=t21 t22=t31 t32 t
1 1 11 12 13 14 15 16 17
2 1 21 22 23 24 25 26 27
3 1 31 32 33 34 35 36 37
4 1 41 42 43 44 45 46 47
5 1 51 52 53 54 55 56 57
6 1 61 62 63 64 65 66 67
7 1 71 72 73 74 75 76 77
8 1 8
S t t t t t t t
S t t t t t t t
S t t t t t t t
S t t t t t t t
S t t t t t t t
S t t t t t t t
S t t t t t t t
S t

=

0
1
2
3
4
5
6
1 82 83 84 85 86 87 7
A
A
A
A
A
A
A
t t t t t t A

For n parameters, we need (n+1) experiments in general

The Taguchi approach corresponds to a specific choice of experiments:

1 1 11 12 13 14 15 16 17
2 1 11 12 13 24 25 26 27
3 1 11 22 23 14 15 26 27
4 1 11 22 23 24 25 16 17
5 1 21 12 23 14 25 16 27
6 1 21 12 23 24 15 26 17
7 1 21 22 13 14 25 26 17
8 1 2
S t t t t t t t
S t t t t t t t
S t t t t t t t
S t t t t t t t
S t t t t t t t
S t t t t t t t
S t t t t t t t
S t

=

0
1
2
3
4
5
6
1 22 13 24 15 16 27 7
A
A
A
A
A
A
A
t t t t t t A

For each parameter we consider a minimum and a maximum value,

The experiments are chosen to have an equal number of tests with minimum and maximum
value for each parameter

In general, the number of experiments (n+1) should correspond to a power of 2

So we can choose 1, 3, 7, 15... parameters

The Taguchi matrix of experiments is orthogonal as is seen easily by transforming thew
thicknesses into a domain between 1 and 1:

( ) ( )
*
2 1 2 / 2 1
*( 2 1) / 2
ti ti ti ti ti ti
Bi Ai ti ti
=
=

1 1 1 1 1 1 1 1 1 0
2 1 1 1 1 1 1 1 1 1
3 1 1 1 1 1 1 1 1 2
4 1 1 1 1 1 1 1 1 3
5 1 1 1 1 1 1 1 1 4
6 1 1 1 1 1 1 1 1 5
7 1 1 1 1 1 1 1 1 6
8 1 1 1 1 1 1 1 1 7
S B
S B
S B
S B
S B
S B
S B
S B

=

And can thus be trivially inverted to solve for the gradients, for example:

( )
8 7 6 5 4 3 2 1
1
8
8 7 6 5 4 3 2 1
1
1 4 2 1
S S S S S S S S
B
S S S S S S S S S
A
P t t
+ + +
=
+ + +
= =

As we derived previously by common sense

Workshop

Material parameter identification using LS-OPT

Numerical Treatment of Shockwaves in Solids

10. Numerical Treatment of Shockwaves in
Solids: Introduction to High- and Hypervelocity
Impact

Material Testing at High Strain Rates

Uniaxial state of stress is used for strain rates roughly between 500/s and 5000/s, different
methodologies can be used depending on material a. o. factors:

Hopkinson bar
Inverted Taylor bar
Etc...

Reference from EMI-Freiburg:

Ermittlung dynamischer Materialkennwerte mit dem Taylor-Verfahren (H. Nahme, R.
Tham)

Example of inverted Taylor bar setup generating a uniaxial state of stress in the bar:

G.I. Taylor, 1948
Wilkins & Guinan, 1973

Uniaxial State of Stress

Acoustic waves are usually studied in a state of uniaxial stress:

0 0
0 0
0 0
xx
xx
xx
v
v
| |
|
=
|
|
\ .
&
t
&
&
&

0 0
0 0 0
0 0 0
xx
| |
|
=
|
|
\ .
t

Such a state of stress exists in a long rod if all lateral motion is unconstrained.

For higher rates a uniaxial state of strain is generated in a flyer plate test or by using high
explosive charges ignited by a plane wave lens

For strain rates between 10000/s and 1000000/s, a uniaxial state of deformation is generated
in a flyer plate test:

References are from EMI-Freiburg:

Planar-Platten Impakt-Untersuchungen als Mittel zur Bestimmung von Werkstoffdaten bei
extremdynamischen Belastungen (H. Nahme)

Uniaxial State of Deformation

Shockwaves are usually studied in a state of uniaxial deformation:

0 0
0 0 0
0 0 0
| |
|
=
|
|
\ .
&
t
&

0 0
0 0
1
0 0
1
xx
xx
xx
v
v
v
v
| |
|
|
|
=
|
|
|
\ .
t

Such a state of strain exists in a long rod if all lateral motion is confined

Evaluate velocity gradient and deformation gradient:

v xx
x i V
x l V
| |
= = = = =
|
\ .
&
& &
&

0
0 0 0
x l V
x l V
| |
= = =
|
\ .
&

The kinematical relations above are completed (in brackets) with the Lagrangean expression for
conservation of mass.

Penetration and Perforation of Solids

The most common application generating approximately a uniaxial state of strain is the impact of
a high-speed projectile on a solid plate

Consider a discontinuous (shock) wave:

wavefront

1
1
1
1 s
x
p
e
&

0
0
0
0 s
x
p
e
&

Here e(s) stands for the specific energy (per unit mass) of the material.

In many penetration problems at high impact speed we have the following phenomena:

Localized impact: during the penetration the stress waves do not travel very far laterally due
to the short duration of the impact, in front of the impactor the state of deformation is nearly
uniaxial

Hydrodynamic material behavior: impactor and target completely pulverize: the pressure will
exceed the yield stress by 2 or more orders of magnitude, the material behaves like a fluid

Applications are in the following fields:

Reactor safety
p
x
wavefront
Defense (ballistics)

These were historically the first applications of explicit codes (hydrocodes)

A Note on Units

Compare typical units for low velocity impact (=crash) and high velocity impact:

low High

time ms microseconds
length mm cm
mass kg g

force kN Mkgf
stress GPa Mbar=100GPa

Typical orders of magnitude:

E (steel) 2. Mbar
Sigy (steel) .002 Mbar
Typical pressures in high velocity impact .02 20. Mbar

Structure of Workshop

10.1: Hydrodynamic materials
MAT_NULL (type 9)

10.2: Elasto-plastic hydrodynamic materials
MAT_ELASTIC_PLASTIC_HYDRO (type 10)

10.3: Influence of strain rate and temperature
MAT_STEINBERG (type 11)
MAT_STEINBERG_LUND (type 11)
MAT_JOHNSON_COOK (type 15)
MAT_MODIFIED_ZERILLI_ARMSTRONG (type 65)
MAT_MTS (type 88)

10.1 Hydrodynamic Materials

We will restrict ourselves to materials that behave hydrodynamically: the deviatoric stresses are
much smaller than the pressure and can be neglected:

vm
xx yy zz
p
p

<<

In this case the material law is fully defined by the equation of state that gives pressure as a
function of density and energy.

Cases of hydrodynamic material response:

hydrodynamic loading (difficult for metals)
materials with very low shear stiffness (fluids)
very high compressive loading of metals: the yield stress limits the deviatoric response,
pressure exceeds yield stress by 2 orders of magnitude or more

EOS for Waves with Infinitesimal Deformations

For very small compressions, the temperature changes can be neglected and a barotropic
EOS can be sufficient to describe the hydrodynamic material

Linear elasticity gives a law linear in volume:

0 0
0
_ _ _ : _ 1
V V
linear in volume p K K
V
| | | |
= =
| |
\ . \ .

Soundspeed follows from:

( ) ( )
2 0
2
2 0
0 0 0 2
0
dp K
c
d
dp K K
c
d

= =
= = = = =

This EOS is barotropic, consequently energy is uncoupled and must not be considered further

Barotropic EOS

To obtain a more physical description of solids under high compressive strains, we must go away
from the law linear in density, some alternatives:

0 0
0
_ _ _ : _ 1
V V
linear in volume p K K
V
| | | |
= =
| |
\ . \ .

0
0
_ : _______ ln ln
V
logarithmic p K K
V
| | | |
= =
| |
\ . \ .

0 0
0
_ _ _ : _ 1
V V
linear in density p K K
V
| | | |
= =
| |
\ . \ .

2
0 0
_ _ _ : _ 1 1 quadratic in density p K K

| | | |
= +
| |
\ . \ .

All 4 EOS behave the same if the volumetric strain remains small enough.
The differences between these EOS become clear when we calculate the corresponding speed of
sound (propagation speed for small amplitude pressure waves):

2 0
2
_ _ _ : _
p K
linear in volume c

= =

2
_ : _______
p K
logarithmic c

= =

2
0
_ _ _ : _
p K
linear in density c

= =

2
0 0
_ _ _ : _ 2 1
p K
quadratic in density c

| |
= =
|
\ .

Note that the acoustic (small disturbance) wavespeed is always the same.

Such EOS describe a barotropic fluid or a metal under loads where the pressure exceeds the yield
strength many times, however without generating heat effects.

For a barotropic EOS, the normality conditions (Bethe-Weyl) just specify that the EOS has a
negative slope and is concave upward:

2
2
0
0
p
V
p
V
<
>

Most real materials satisfy these conditions, such an EOS is called stiffening: the pressure grows
faster then linear in volume

The law linear in volume does not satisfy the second Bethe-Weyl condition (is not stiffening)

Realistic EOS must be at least linear in density since the acoustic wavespeed cannot decrease in
compression.

p
V
Overview of Barotropic EOS

Linear in volume EOS_
GRUNEISEN
C=K/rho0
S1=s2=s3=gama0=a=E0=0, V0=1
Linear in density EOS_LINEAR_POLYNO
MIAL
C1=K
C0=C2=C3=C4=C5=C6=0
Quadratic in density EOS_LINEAR_POLYNO
MIAL
C1=C2=K
C0=C3=C4=C5=C6=0

0
0,2
0,4
0,6
0,8
1
1,2
1,4
1,6
0 0,2 0,4 0,6 0,8 1 1,2
Reihe1
Reihe2
Reihe3

Pressure Wave Propagation

In a material with EOS linear in volume, a pressure wave is steady state with respect to an
observer that travels with the wavefront: the wavefront does not deform.

The pressure signal thus always has the character of an acoustic (small deformation) wave

In a real, normal (stiffening) material, a continuous compressive pressure signal cannot be steady
state: the wavefront steepens until a shockwave is created that is then steady state

The shockfront is a discontinuity in density, pressure, velocity and specific energy, the thickness
of the shockfront is of the order of a few micrometer.

p p
t
t
p
p
t t
Properties of Shockwaves

The speed of the shockfront U is always supersonic with respect to the unshocked region and
can be determined for any EOS, some examples:

0
_ _ _ : _ linear in volume U c =
1 1 0
0 1
1 0
_ _ _ : _ linear in density U c x

= +
&
0 1
_ _ _ : _ quadratic in density U c x = + &

In a shock wave each element acts to compress the next element, support is from the rear,
pressure and speed decrease

Steady Shock Waves in a Linear Solid

Shock loading goes from the reference state A along the Rayleigh line 2 to point B on the
EOS 1 that is hyperbolic in volume
Unloading follows the isentrope (=EOS) 1 from point B back to point A
All energy between loading and unloading paths was dissipated by the shock
The pressure difference between lines 1 and 2 is the viscous pressure q in every point

P
1
B
2
A
V
Artificial Bulk Viscosity

In the very thin shock layer velocity gradients are very high and viscous pressures exist that
dissipate energy

In the shocklayer, the pressure evolves on a straight (Rayleigh) line in the pV diagram that
lies above the EOS

Numerically the shockfront cannot be represented by any mesh,

Therefore John von Neumann (1949) proposed a numerical diffusive pressure term q that
allows numerical simulation of shockfronts by meeting the following conditions:

Thickness of the shock layer is of the same order as the computational grid,
independently of the material and the strength of the shock
Effect of q is negligible outside the shocklayer
Mass, energy and momentum are exactly conserved across the shockfront

A method for the numerical calculation of hydrodynamic shocks, von Neumann, Richtmyer,
1949

The implementation of a quadratic bulk viscosity in LSDYNA is described in Noh (1976)
and Hallquist (1998) (*):

( )
( )
0 0
1
1
3
1 1
0
2
2 1
1/ 2 0 1 , 1/ 2
1 1 1 1/ 2
1.5
2
n
n
n n
n n
n n v n
n n n n
V
V
l V
C
q C l
p p V q
+
+
+ +
+
+ + +
+ + + +
=
=
=
+
| |
=
|
\ .
= +
&

(*) LSDYNA, theoretical manual, may 1998

Workshop

Example problem is shock00.lsd

A 10m long rod is placed along the x-axis and modeled with 1000 brick elements, 10mm long
each:

All nodes are fixed in y and z so the rod is laterally confined and we generate a 1D state of strain

A triangular 2.ms pressure pulse is applied on the left end of the rod:

The duration of the pressure load is much smaller then the time needed by the pressure wave to
travel to the other end of the rod

Pressure is saved as a function of time every 100 elements (=every 1.m along the rod), this
allows to visualize the wave propagation and formation of shockwaves if applicable

The material of the rod is type 9: MAT_NULL and thus has no deviatoric strength

The EOS is polynomial (type 1) with the following input parameters:

C1=20.GPa
C0=C2=C3=C4=C5=C6=0.

x
p
2.ms t
In the first case we apply a pressure with amplitude of 2.GPa, this generates an acoustic wave,
the wavefront does not deform:

The pressure/volumetric strain curve in any element shows identical loading and unloading paths
that are approximately linear in volume although the EOS is linear in density:

In the second case we apply a pressure with amplitude of 20.GPa, this generates a shockwave,
the wavefront steepens and the peak decreases as energy is dissipated into heat, also clearly the
signal moves faster

The pressure/volumetric strain diagram in a shocked element shows the hyperbolic EOS as
unloading path as well as the Rayleigh line loading path:

Exercise: use a barotropic EOS quadratic in density with C1=C2=20.GPa and compare the
results to the EOS above linear in density (shockwavespeed etc...)

Try to run with an EOS linear in volume

10.2 Elasto-plastic Waves: High Velocity Impact Test
Cases without Temperature Effects

For these test cases we use material law type 10 in LSDYNA combined with a polynomial EOS:

*MAT_ELASTIC_PLASTIC_HYDRODYNAMIC
*EOS_LINEAR_POLYNOMIAL

Here the elastic and plastic wavespeeds under uniaxial state of strain can estimated as follows
for small deformations:

4 / 3
p
K G
c
K
c
+
=
=

Two separate wavefronts (elastic and plastic) will thus develop

This numerical approach has typical applications are in reactor safety and defense for the
following class of problems:

Medium high impact velocity: 100m/s to 800m/s

Airplane impact on NPP

Impact of shells & bombs on concrete and soil structures, concrete dippers, a.o.

Problems may contain shock waves, spalling e.o.

No thermal effects yet, temperature change is too small to effect material properties

For this type of applications we are interested in materials with a wide range of densities, some
examples:

material density (g/cm**3)

Cp (m/s)
magnesium 1.74 4500
aluminum 2.7 5380
Niobium carbide NbC (ceramic) 7.58
iron 7.85 3580
Stainless steel 7.89 4570
copper 8.93 3910
lead 11.34 2030
tantalum 16.66
Uranium 18.93 2510
W-sintered alloy D17
Powder of 6%C, 88%W and 6%Co,
compacted at 700C and sintered at 1600C
achieves hardness near diamant
17.00

For our examples we choose lead and D17

The model consists of a laterally confined 10m long metallic bar modeled with 1000 brick
elements, the bar is impacted by an equal rigid mass

The EOS is quadratic in density in all cases and we use the following deviatoric data (units are
GPa):

Shock01.lsd
Lead Pb95 E=16Gpa,
nu=0.44
Shock02.lsd
D17
MAT
G 5.7 127.
SIGY 0.05 0.6
EH 0.2 1.0
PC 0. -2.5
FS 0. 1.0
SPALL - 2.

EOS
C1 43. 275.
C2 43. 275.

c 2100. m/s 5110 m/s
cp 1950. m/s 4020 m/s

In soft metals such as lead, a single shockfront develops at almost all impact speeds: the soft
metal behaves like a fluid, the velocities of elastic and plastic shockwaves are nearly equal.

In hard metals, such as D17, 2 separate stresswaves develop, we have repeated the analysis
for the following impact speeds:
7 m/s
70 m/s
700 m/s
1400 m/s

For impact speeds of the order of 7 m/s we see a single elastic wavefront.

For an impact speed 70 m/s we see an elastic wavefront with a stress up to the HEL (Hugoniot
Elastic Limit), followed by a plastic shockwave

The HEL can be determined from simple von Mises theory and elastic stress values at first yield
as follows:

0
0
0
0
0
0
4
3
2
2
3
HEL xx
vm xx yy y
HEL y
G V V
K
V
V V
G
V
V V
K
V

| |
| |
= = +
| |
\ .
\ .
| |
= = =
|
\ .
| |
= +
|
\ .

In the D17 we find:

0.6
1.05
y
HEL
=
=

The value of 1.Gpa is easily recognized in the simulation results.

Due to the stiffening behavior of the metal in compression, the plastic stress waves will overtake
each other leading to the formation of shocks:

The velocity of the plastic shockwaves can be derived from the impactor velocity and the
compression by combining the Hugoniot and the EOS quadratic in density:

In the above we have neglected the plastic hardening modulus with respect to the tangent bulk
modulus and based the plastic wavespeed upon the latter

0
1
=
1 0
U x c = + &
1 0
/ x c = &
1
0
x
c
&

1
For an impact at 700 m/s, strong shocks are seen but still 2 separate wavefronts exist:

The following diagram illustrates the phenomena:

Compression up to A happens with 2 separate wavefronts, in B a single wavefront is created and
in C the single shockfront travels faster then the elastic wave

Note that the elastic pressure is linear in volume

In impact problems the surface (or impact) velocity is related to the material velocity of the
particles in the bar during the steady state as follows:

1
2
s
x
x =
&
&

For D17 we thus expect a single shockfront for an impact speed of:

2*( ) 2200 /
s p
x c c m s = = &

In fact, at these impact speeds other physical phenomena such as phase changes and rate effects
will become non-negligible.

C
B
A

HEL

0
0
v
V V
V

=
For impact speeds above 1400 m/s, the plastic shockwaves have overtaken the elastic precursor
and a single shockfront is created

Spalling can be seen at the rear end of the bar at high impact velocities (for 700 m/s and
1400m/s), the location of the spall plane(s) depends upon the tensile wave reflections:

t=0. ms

t = 2.6 ms

t = 4.0 ms

A historical reference on spalling is:

Experiments with nthe Hopkinson pressure bar
Landon, Quinney
1923

10.3 Rate and Temperature Dependency: Testcases
in the Hypervelocity Range with Consideration of
Temperature Effects

Typical Applications:

Application Impact speed Material law Physics

Ordnance,
SFF,
Shaped charges,
Schardin-Misznay
fragments,
High velocity
1-3km/s
MAT_JOHNSON_COOK

MAT_ZERRILI_ARMST
RONG
Temperature reduces
yield, rate effects
increase yield,
phase changes from bcc
to fcc change modulae

Meteorite impact on
space vessels
Hypervelocity
20-40km/s
MAT_STEINBERG_GUI
NAN
Phase changes, metal
liquifies and evaporates

A general EOS gives pressure as a function of density and energy:

( ) ,
s
p p e =

An extra normality condition (Bethe-Weyl) specifies that the hotter isentropes are further from
the origin:

0
0
s
V
p
V
p
s
<
>

Most real materials satisfy these conditions

High velocity impact simulation on a steel bar using the Johnson Cook law:

Room temperature: 297K
Melting temperature: 1539C = 1836K

Specific heat: 0.114 kcal/kgC = 477.4 Nm/kgK
1.0 kcal = 426.9 kgm

p
V
References:

G. Johnson, W. Cook
A constitutive model and data for metals subjected to large strains, high strain rates and high
temperatures
7
th
international symposium on ballistics, The Hague,
The Netherlands, April 1983

Landau, Lifschitz
Effets hydrodynamiques et modeles mathematiques
Editions MIR, Moscou 1970

Reference State

The configuration (state) of unidimensional continuum in a material point is defined by velocity,
specific energy, density, and pressure (the state variables).

A reference state will be refered to as:
0
0
0
0
0
0 0
s
x
p
e
e
V
&

Derived quantities will be added such as: temperature, specific volume and energy per unit of
undeformed volume and sound speed:

For solids, the reference or cold state will often be conveniently chosen as:

0 0 0
0
s
x p e = = = &

Corresponding to a stress-free reference state.

In the case of an explosive, the energy will at least have to take the value of the chemical energy
in the solid explosive or propellant.

For gases, the reference pressure cannot be zero since this
would implie an infinite volume.

Polynomial Equations of State

A polynomial equation of state is often used in LS-DYNA:

2 3 2
0 1 2 3 4 5 6
0
0
1
s
p C C C C C E C E C E
E e

= + + + + + +
=
=

This EOS:

Is linear in energy
Does not need a zero-energy reference state, so gases can be modeled, the value of the
reference energy E0 must be provided in order to calculate reference pressure
Not barotropic: pressure change causes energy change and thus temperature change
One-to-one, loading and unloading path are identical if all volumetric changes are reversible

E is the internal energy per unit undeformed volume,the choice is convenient since E has the
dimension of pressure.

The polynomial EOS does not assume a zero-energy state of reference, so gases can be modeled

In the reference state:

0
0 0 0
0 0 4 0
0
s
E e
p C C E
=
=
= +

This allows great flexibility in modeling

Stiffened Gas Equation of State

Obtained from the polynomial EOS as follows:

0 2 3
1
4 5
6
0 0
0
1
0
1
C C C
C K
C C
C
p K E

= = =
=
= = =
=
| |
= +
|
\ .

Alternatively:
( ) ( )
0
0
1
s
K
p e
= +

This EOS suitably describes gases and metals if the departures from normal density are slight.
(Harlow & Amsden, Los Alamos, 1971).

For K=0 we obtain the familiar EOS of a polytropic gas.

( )
( ) ( )
( )
1
1 1
1
s
s
v
p e
pV me e
pV mC T mRT

=
= =
= =

If Gamma=0 or thus gamma=1, we recover the EOS for the linear solid:

0
1 p K K

| |
= =
|
\ .

Gamma, the Gruneisen coefficient is a measure of reversible heat exchange during deformation

For gases, gamma lies between 1.3 and 1.7, for metals between 2 and 4, rarely below 2. A
reasonable value for Gamma for steel is 2.2 (gamma=3.2).

Thermal Expansion

Assume a constant coefficient of (small) thermal expansion

( )
0 0 0
0
0 0
1 1
V V
T T
V

| |
= = =
|
\ .

And calculate the energy per unit undeformed volume:

( ) ( )
0 0 0 0 0
0
0 0
0
1
s s v
v
E E e e C T T
C
E E

= =
| |
=
|
\ .

And combine this with the EOS of a stiffened gas, the pressure increase is zero for pure thermal
expansion:

( )
0 0
0 0
0 0
0 0
1 0
1 1 0
v
p p K E E
C
p p K

| |
= + =
|
\ .
| | | |
= =
| |
\ . \ .

And thus:

0
1
v
K
C
= =

Isentropic Conditions

Isentropic conservation of energy and EOS can be combined for a stiffened gas:

0
1
s
s
e pV
p K e
| |
= +
|
\ .
&
&

We rewrite the EOS as follows:

( )
( )
0
0
0
1
s
s
s
pV K e
pV K V V e
e pV pV KV

| |
= +
|
\ .
= +
= + +
& &
& &

Eliminating specific energy material time derivative yields:

( ) 1
V V
p p K
V V
p p K

= +
= +
& &
&
& &
&

This differential equation governs isentropic compression and expansion of the stiffened gas.


Solving gives the isentrope:

0 0
0
0
dp p K
dV V
p K
d dp
d dV
V
p K V
p K V
K K
p p
+
=
= +
=
=
| | +
=
|
+
\ .
| | | |
= +
| |
\ .\ .

The sound speed can be derived easily:

2
dp p K
c
d

+
= =

The solution gives the isentropic relationship between pressure and density:

0
0
K K
p p

| | | |
= +
| |
\ .\ .

Following are interesting special cases:

0
0
0
0
0
_ _ : _ 0
1
_ _ : _ 1
1
_ _ : _ 0
general solid p
K
p
linear solid
p K
perfect gas K
p p
=
(
| |
= (
|
( \ .

=
(
=
(

=
| |
=
|
\ .

The Hugoniot

Similarly we can combine the energy jump condition with the EOS in order to obtain a reference
path for steady shocks:

( )
( )
[ ] ( )
0
0 0
0
0
0 0 0
2
1
1
s s
s
s
s
p p
e e V V
pV K V V e
e pV K V V
e p V
+
=
= +

Eliminating the energy gives the Hugoniot:

( ) ( )
( )
0 0 0 0
0
2
2
p V V V K V V
p
V V V
| |
+ +
|
\ .
=

The Hugoniot line gives all the configurations in which a stiffened gas can be shocked from a
given reference state.

The Hugoniot can be reformulated as a function of relative density:

0
0
1
V V
V
= =

And thus:

0
1
2
1
2
p K
p

| |
+ + +
|
\ .
=

The case of the solid is obtained by setting the reference pressure equal to zero, the polytropic
gas case follows from the condition K=0.

In any case there is a limiting compression (infinite pressure values) at:

0
1 0
2
2 1
1
=
+ +
= =

In the case of a solid we can approximate the Hugoniot:

( )
2
2
0
2
1
2
_ 2
K K
p K
if
p K K
= +

+ =

A pure quadratic law in density!

Steady Shock Waves in a Stiffened Gas

Shock loading goes from the reference state A along the Rayleigh line to point B on the
Hugoniot 1
Reversible loading would follow the isentrope 2
Shock loading results in a higher pressure for the same final volume compared to isentropic
loading
Unloading follows the isentrope 3 from point B to point C
Unloading after shock loading results in a higher pressure compared to the reference
configuration

2 P
3
1
B
C
A
V
Continued Workshop

Use MAT_HYDRODYNAMIC
Introduce a Gruneisen coefficient 2 in the EOS:
C4=C5=2
Apply a pressure of 20.Mpa

The result again shows a shockwave formation:

Compare the pressure/volumetric strain curves in element 1 (=left end of rod) and element 500
(=center of rod)

In element 1 we load and unload along the isentrope

In element 500 loading follows the Rayleigh line and unloading the isentrope corresponding to
the increased temperature, the unloading path crosses the loadpath as the material has expanded
after the shock

Exercise: use a barotropic EOS quadratic in density with C1=C2=20.Gpa and compare the
results to the above stiffened gas EOS

Equations of State Linear in Energy

The most general EOS that is linear in energy is:

( ) ( )
1 s
p p e = +

Where p1 has the dimension of pressure and the Gruneisen function is dimensionless.

The Mie-Gruneisen EOS assumes a constant (dimensionless) Gruneisen coefficient and a linear
relationship between the variations of pressure and energy from a given reference path:

( )
H s sH
p p e e =

In practice the Hugoniot from the cold reference state is taken as a reference path. If we assume
pressure and energy to be zero in the reference state we have:

0 0 0
0
0 0
2 2
H H
sH s sH
p p p
e e e

| | | |
= = =
| |
\ . \ .

Thus the Mie-Gruneisen EOS can be written as:

1
2
H s
p p e
| |
= +
|
\ .

And is fully determined if the Hugoniot pressure can be determined as a function of density by
performing a number of shock experiments from a cold reference state.

If Gamma is taken as a constant then this equation corresponds exactly to the stiffened gas EOS!

Usually Gamma is taken as a simple function of density assuming a constant Gamma per unit
undeformed volume:
0
0
0
1
2
H
E
p p
| |
= +
|
\ .

Combined this gives for the Mie-Gruneisen EOS:

0
0
1
2 1
H
p p E
| |
= +
|
+
\ .

Linear Us-Up Laws

A better approximation for solids is obtained by the following Hugoniot:

( )
[ ] ( )
0
0
2
0
0
0
0
2
1
1
1
1 1
H
H
V V
K
V
p
V V
s
V
V V
V
K
p
s
=
| |
|
\ .
=
+
+
=

This gives a limiting compression:

1
1
1
s
s
= >

The linear Us-Up law is derived from the experimental observation that the speed of the
shockfront and the particle speed behind the shock can be linearly fitted as follows:

0 1
U c sx = + &

This is easily derived from the jump conditions for zero pressure and velocity in front of the
shock:

1 0 1
0
1
1
p Ux
x U U

=
| |
= =
|
+
\ .
&
&

Then:

[ ]
[ ] ( )
[ ] ( )
0 1
2
2 2
0 1 0 0 0 2
1 1
1 1
1 1
1 1 1
1 1 1
1 1
. . .
H
s c x
s s
U U
K s
p Ux U c
s
q e d

= = =
+ +
+ | |
= = = =
|
+ + +
\ .
&
&

The full Mie-Gruneisen EOS for linear US-UP Hugoniot is then:

( )
( )
0
0 2
1
1
2 1
1 ( 1)
K
p E
s

+ | |
= +
|
+

\ .

The lower limit for this type of law (s=1) clearly gives a quadratic Hugoniot in pressure:

( )
0 1
1
H
p K U c x = + = + &

A stiffened gas and a linear UsUp EOS will have the same limiting compression if:

0
0
2
2 1
2 2
1
s
s
s
+
=
+ +
= =
`

)

Another interesting comparison if for low pressures where:

2
E
<<

Then equating the stiffened gas EOS and the linear UsUp EOS leads to:

( )
( )
0
2
0 0
1
1 1
2 1
1 ( 1)
1
1 / 2 2
2 2
s
s

+ | |
=
|
+

\ .
+
+
= =

This leads us to say that the stiffened gas EOS and the linear Us-Up EOS are in agreement at low
pressures if:

1
2
s
+
=

A typical value of s for metals is 1.5. A detailed description of solids requires even more
complicated EOS.

Exercise: Repeat the problem shock00.lsd with a pressure applied of 20.Mpa in order to
generate a strong shock.

Use a type 4, Mie-Gruneisen EOS with the following parameters:

*EOS_GRUNEISEN

c=1414 m/s
s1=1, s1=1.5, s1=2.0
s2=s3=0
a=0
E0=0
V0=1.

Verify that this corresponds to a linear Us-Up EOS and compare the results to the polynomial
EOS


Quasistatic Simulations with LS-DYNA

11. Quasistatic Simulations with LS-DYNA

Interest for vehicle analysis:

prestressing of vehicle and occupant under gravity (particularly for side impact)
roof crush
static door intrusion
seat belt pull

Implicit solution may fail if:

model is very large
loading contains very high frequencies and small timesteps are required for accuracy
highly nonlinear material
large rigid body rotations
contact
post buckling and softening

Implicit initialisation will work if the initial stresses remain largely elastic, then initialise the
explicit dynamic load simulation using the DYNAIN file generated by:

*INTERFACE_SPRINGBACK_DYNA3D

In most cases a slow dynamic simulation must be performed using ramp curves to apply the
load:

Linear or better, sinusoidal ramping can be applied, the results can be considered quasistatic
iff:

so the first eigenfrequency of the structure must be determined, speedup of loading with
respect to test may or may not be possible
min
1
2
T

f
fs
T
t
In the GLSTAT file, check that:

ke ie <<

2
T cpu
T cpu
n n
cpu
n cpu
n

= =
= =

Decrease in computation time can be obtained by speedup of the loading and/or mass scaling,
in the following ratio:

So mass scaling is not very efficient

The speedup always introduces higher amplitude oscillations in the results and consequently
deviations from the quasistatic response

Example: Gravity Load on a Dome Structure

The structure is a truncated icosahedron with 15 m diameter, made out of 65 mm thick
wooden orthotropic unipanels

Modeled using material law 2 (orhtotropic elastic)

We apply gravity as a body load with ramp time of 50.ms and plot the tip vertical
displacement:

The eigenperiod is seen to be 100. ms and thus the lowest eigenfrequency of the strcuture is
10 Hz, the structure behaves pretty much like a SDOF system

Using a longer ramp time (300.ms) allows to identify the static deflection:

Here the undamped solution using a ramp time of 300.ms is compared to a damped solution
using a ramp time of 50.ms

Damping

Viscous damping (proportional or Rayleigh damping) can be introduced by:

*DAMPING_GLOBAL
*DAMPING_PART_MASS
*DAMPING_PART_STIFFNESS

Global system damping damps all motion including rigid body motions

we try to critically damp the lowest system eigenfrequency

Critical damping in a SDOF example:

2 2 2
e
mx dmx kx f
k
m
c km m dm d

+ + =
=
= = = =
&& &

Mass weighed damping damps all motion including rigid body motions

stiffness proportional damping is orthogonal to rigid body motions

In a SDOF example we can calculate the values needed to critically damp a frequency:

, 1/ 2
, 1/ 2
2
2
i n n n
i n n n
f kx k x
f kx mx

= + =
= + =
&
&

After time integration mass proportional damping leads to:

( )
1/ 2 1/ 2
1
n n n
x t x x t
+
= + & & &&

the implementation thus just invokes a relaxation factor

In the dome example we compare the undamped and critically damped solution for a ramp
time of 50.ms:

The critical damping factor is:

4 4 10 120/ d s = =

At least one eigenperiod is needed by the damping to find the static solution!

Influence of Non-Linearity

Beware of elastic buckling problems:

Compare the axial and bending frequency of a bar:

axial bend
>>

If we base the loading velocity on the eigenperiod for axial loading, the mode suddenly
changes to a bending mode as we reach the critical Euler load

At that point we have:

1 1
2 2
axial bend
T

<<

And the simulation suddenly becomes dynamic, the oscillations in the response then
completely hide the quasistatic postbuckling behavior

The previous can be illustrated by applying a high concentrated nodal force load at the tip of
the dome untill a snap-through effect occurs:

This problem is geometrically nonlinear

First apply a load of 720.kN with ramp times of 180. Ms and 720.ms and compare to the
damped solution:

A snap-through occurs in all cases, to determine the exact static deflection is not trivial

as the elastic instability occurs:
the structural stiffness decreases
the eigenfrequency decreases
the problem becomes suddenly dynamic



This means that:
Much higher ramp times are needed then originally estimated from the linear
structural behavior
The system may suddenly become overdamped

Illustrate for a load of 100.kN:

The damped model fails to find a snap-through

Dubois Book

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Dubois Book

Hochgeladen von

Copyright:

Verfügbare Formate

Crashworthiness Engineering

& & &

& & &

Das könnte Ihnen auch gefallen