Implicit Analysis Usimg LS-DYNA

DYNAmore Express
Implicit Analysis using LS-DYNA
Dr. T. Erhart, Dr. T. Borrvall

Dr. N. Karajan, Dr. M. Schenke
Dr. C. Schmied
July 2020
Copyright 2020 DYNAmore GmbH. All rights reserved.

Non-commercial usage is granted if credits are given clearly to DYNAmore GmbH and copyright remarks are not removed Slide 1 of 40
Why implicit ?
■ Prestressed, quasi statically loaded structures Nonlinearity
■ Long duration analysis > 500 ms
high
explicit ??
■ Different time scales in process
■ e.g. static loading followed by transient loading low
■ or transient loading followed by static loading
implicit
explicit & implicit
linear
■ Applications
■ metalforming, roof crush, door sag, short long (static) Duration
dummy seating, strength analysis, ...
How short is short duration?
Before supercomputers (prior to 1980) short < 5 milliseconds.
■ LS-DYNA provides explicit and implicit Hence problems mainly in ballistics!
solution schemes After supercomputers: short < 100 milliseconds.
Hence, problems in crashworthiness became feasible!
■ one code – one license – one data structure
■ one input / output

Explicit vs. Implicit (dynamics)
ρ 𝐮𝐮,tt = 𝛻𝛻 ⋅ 𝛔𝛔 + 𝐟𝐟
ext
𝐌𝐌𝐌𝐌n = 𝐟𝐟next − 𝐟𝐟nint 𝐌𝐌 ∆𝐚𝐚n+1 + 𝐊𝐊 ∆𝐮𝐮n+1 = 𝐟𝐟n+1 − 𝐟𝐟nint − 𝐌𝐌 𝐚𝐚n
Explicit scheme 𝐱𝐱 n+1 = 𝐟𝐟 𝐱𝐱 n , … Implicit scheme 𝐟𝐟 𝐱𝐱 n+1 , 𝐱𝐱 n , … = 𝟎𝟎
+ Solution: directly - Solition: iteratively

+ Decoupled: fast, efficient - Linearization necessary
- Many small time steps + Few large time/load steps

- Conditionally stable (Courant) + Unconditionally stable
Equilibrium ? Energy balance ! Equilibrium ! Convergence ?
Short time dynamics Structural dynamics

High frequency response, Low frequency response,
Wave propagation Vibration, Oscillation
Impact, crash, … Earthquake, machines, …

Explicit vs. Implicit
■ Explicit ■ Implicit
■ inevitably includes inertial effects and ■ can neglect inertial effects and
■ resolves high frequencies ■ the selected time step size determines
resolved frequency spectrum
"Explicit is handcraft"
■ Consequences for FE models

■ "cleaner" models in implicit for the sake of convergence,
e.g. no initial penetrations, smooth material curves, …
■ expensive features are not so expensive anymore
■ no resctriction on element size (time step size) in implicit
■ often more work to get "normal termination" in implicit
“Implicit is skill"

Troubleshooting convergence problems
Convergence behavior is depending

on the physics of the problem
General philosophy
Difference in physics ”Increased accuracy implies

better convergence”
Different method(s) for solving

convergence issues

Common reasons for convergence problems
■ Mesh
■ Coarse meshes may result in poor element geometry and bad contact behavior
■ Time/Load step size

■ The applied load/displacement etc. in a single step may be too large or small
■ Rigid body motions

■ Unconstrained d.o.f. due to missing BC/SPC, initial contact gaps, beams, …
■ Contact
■ Initial penetrations, too large step sizes, large forces, …
■ Material properties
■ rough data, softening properties, discontinuities in curves, incompressibility, …

Recommendations
■ Use the most recent LS-DYNA version possible (e.g. R11.1, R12.0)
■ Implicit functionality is rapidly improving
■ Use double precision ( _d_ in the name of the executable)

■ Required for accurate linear analysis
■ Improved convergence behavior in nonlinear analysis
■ Mandatory for current releases
■ Read Appendix P in the User’s manual and Chapter 37 in Theory Manual

■ Nice summary about LS-DYNA‘s Implicit Solver
■ The CPU penalty for out-of-core can be as high as 100 times the in-core simulation!
■ Use command line option "memory=" to run job in-core
■ Verify using LPRINT=1 on *CONTROL_IMPLICIT_SOLVER or "<ctrl-c> lprint".

Memory management up to R10 Memory
see: A Tutorial on How to Use Implicit LS-DYNA ®, R. Grimes, www.dynalook.com/...
OS,
MPI, …
■ Estimation for memory option
■ memory = 0.75 * ( available RAM ) / (number of MPI processes) Dynamic
Leave free RAM for dynamic memory
and system tasks, MPI, …
■ For a cluster node with 256 GB = 32000m of available RAM

memory = 1500m (16 MPI processes)
memory = 1000m (24 MPI processes)
Static
■ memory2 specified?
■ If not specified – memory2=memory is set internally
■ If specified – apply above recommendation to memory2
together with a possibly larger value of memory
up to R10 Version

Memory management after R10 Memory
OS, OS,
MPI, …
■ Implicit linear algebra converted from static to dynamic memory MPI, …
■ Huge dynamic and much less static memory needed Dynamic

■ Usage alert printed at start of simulation
■ Example for MPP version
Model with 2.1m nodes, 1.4m shells, 1.4m solids
Compute node with 256 GB using 24 MPI processes Example
Dynamic
usage of
================================================
available
== IMPLICIT USAGE ALERT ==
================================================ RAM
== Memory Management for Implicit has changed == Static
== after R10. Please use: ==
== memory= 155M memory2= 57M ==
================================================
■ This does not mean LS-DYNA requires less memory, Static

only the amount of static and dynamic memory changed
■ memory and memory2 are shrinking in importance up to R10 from R11 Version

Recommendations cont‘d
■ Element types
■ For solids use type 1, -1, -2, 13 or 16 elements for non-linear analysis
■ For shells use type 6 or 16 elements for non-linear analysis
■ Try to avoid pentahedral solid elements
■ Beware of free rotations when merging shells and beams to bricks
■ Contact
■ Try to avoid initial penetrations or try IGNORE=1
■ Use Mortar contact press-fit option (IGNORE=3 or 4) for intended initial penetrations
■ Switch (temporarily) to tied contact to identify problems
■ Use Mortar contacts or try IGAP=2 (on additional card C)
■ Try to decrease contact stiffness, observe penetrations
■ Contact often requires small time steps in implicit, too
■ Make sure that finer mesh is slave side
■ Turn off viscous damping with VDC=0
■ Better use separate contacts instead of only one "big“ contact

Mortar‘s IGAP – contact stiffness scaling
■ IGAP=1 (default)
■ contact stiffness is parabolic with respect to penetration up to a penetration
depth corresponding to half of the maximum penetration
■ IGAP > 1
■ functions as scaling factor on that contact stiffness
■ contact will stiffen for larger penetrations, in fact it will become cubic
Recommendations, cont‘d
■ General
■ Apply second order stress update by setting OSU=1, *CONTROL_ACCURACY
■ Try accuracy option IACC=1 on *CONTROL_ACCURACY (starts with version R9)
■ Try to model displacement driven simulation instead of force driven simulation
■ IGS=1 (not default) on *CONTROL_IMPLICIT_GENERAL may help in some cases (structures under tension)
■ Set DNORM=1 on *CONTROL_IMPLICIT_SOLUTION, DCTOL can often be increased then, e.g. DCTOL=0.005
■ Try ABSTOL=1.e-20 on *CONTROL_SOLUTION to improve accuracy
■ Sometimes Full Newton (ILIMIT=1) improves convergence
■ Often dynamic solution more robust than static solution
■ if static implicit fails to converge, try dynamic implicit first
■ Keep an eye on time step evolution, choose reasonable step size to avoid “yo-yo” effect
■ Try to avoid discontinuities, e.g. in material curves, geometry, ...
■ In problems where there is much rigid body motion the displacement tolerance DCTOL may be insufficient,
in some problems a tighter energy tolerance, e.g. ECTOL=0.001, may be advisable.
■ Be aware of causes and consequences of ill-conditioning

Try accuracy option IACC=1 on *CONTROL_ACCURACY
Variable OSU INN PIDOSU IACC
Implicit
1 2 0 0
default
■ Higher accuracy in selected material models (24, 123, …)

■ Fully iterative plasticity, tightened tolerances, smooth failure
■ Strong objectivity and consistency in selected tied contacts

■ Physical (only ties to degrees of freedoms that are ”real”)
■ Finite rotation
■ Strong objectivity in selected element types

■ Finite rotation support for hypoelasticity
■… see more in User’s Manual

Try accuracy option IACC=1 on *CONTROL_ACCURACY
■ Example: Plastic deformation of metal part
■ *MAT_024 with LCSS
■ DNORM=1 Stress in MPa
■ ENDTIM=0.014 IACC=0: brief overshoot
■ DTMAX=0.001
■ Only a few large steps in implicit analysis for plastic straining of 7 %
IACC=1: exactly
■ Smaller steps would also help, or other material models
on curve
LCSS curve
Plastic strain

Set DNORM=1 on *CONTROL_IMPLICIT_SOLUTION
■ Example: Compression of a foam block
■ *MAT_FU_CHANG_FOAM
■ ENDTIM=20.0, Force in kN
■ DTMAX=1.0,
■ DCTOL=0.005,
■ ELFORM=1, IHQ=6, QM=1.0
Displacement in mm

Keep an eye on time step evolution
■ Automatic step size control adjusts stepsize during simulation
■ Very persistent, reliable
■ After successful steps

■ compare iteration count to target value ITEOPT
■ increase/decrease size of next step if difference
exceeds window ITEWIN
■ After failed steps

■ decrease step size
■ back up, repeat failed step with new DT
■ Exponential algorithm for adjusting step size

■ Increase stepsize by 1/5 decade until DTMAX is reached
■ Decrease stepsize by 1/3 decade until DTMIN is reached
■ Error termination if convergence fails when DT=DTMIN
Keep an eye on time step evolution
… > glstat > time step
Bad strategy
Small initial time step
may help in order to Avoid time consuming
close contact “yo-yo” effect
Time step in ms
Good strategy
Simulation time in ms

Be aware of causes and consequences of ill-conditioning
■ Causes of ill-conditioning
■ Large stiffness differences (thin shells, varying stiffness moduli, etc.)
■ Elements of severe shape distortion or large aspect ratio
■ Fine meshes and mixing elements of different size
■ High Poisson's ratio (nearly incompressible materials) A numerical measure of the ill-conditioning
is the condition number of the system
■ Consequences of ill-conditioning accuracy digits lost ≈ log10 C 𝐊𝐊

■ Difficult (iterative) solution of linear equation system
■ Possible loss of accuracy
■ Possible bad convergence
■ If a set of equations is seriously ill-conditioned

■ usually better to rework the FE model than to make heroic
attempts to improve a poor solution by iteration

Nonlinear convergence problems
■ Output / Debugging
■ Activate print flags (LPRINT, NLPRINT) to get more information
■ Determine reason for termination (check d3hsp / messag files)
■ Set MINFO=1 on *CONTROL_OUTPUT for Mortar contact information like penetrations, release, …
■ In case of convergence problems, dump iteration states
residual forces in d3plot and d3iter via RESPLT=1 on *DATABASE_EXTENT_BINARY
■ Carefully inspect input deck and check …

■ if you use second order stress update (make sure you do)
■ smoothness on curves
■ material properties
■ contact penetrations, remove
■ magnitude of loads
■ contacts, make sure soft part is slave
■ elements, avoid small jacobians and distorted elements

Output of non-converged steps
■ With D3ITCTL ≥ 1 on *CONTROL_IMPLICIT_SOLUTION
■ search directions for the nonlinear implicit solution are written to the d3iter database
■ together with RESPLT=1 on *DATABASE_EXTENT_BINARY, residual values can be fringed
Deformation history Residual force evolution

d3plot during one implicit step
d3iter

Recommendations, cont‘d
■ For “typical” implicit analysis, the following keyword setting is a good start
*CONTROL_ACCURACY
$ osu inn pidosu iacc
1 4 1
*CONTROL_IMPLICIT_GENERAL
$ imflag dt0 imform nsbs igs
1 ... (1)
*CONTROL_IMPLICIT_SOLUTION
$ nsolvr ilimit maxref dctol ectol rctol lstol abstol
6 (1.e-20)
$ dnorm diverg istif nlprint nlnorm d3itctl
1 3 (4) (10)
$
$ lsmtd
(5)
*CONTROL_IMPLICIT_AUTO
$ iauto iteopt itewin dtmin dtmax
1 30 10 ...
*CONTROL_IMPLICIT_DYNAMICS
$ imass
(1)

Information, Help, Inspiration, …
www.dynasupport.com www.dynaexamples.com/implicit
→ HowTos → Implicit
Advanced and basic examples
Starter kit including guidelines

 Basic control card settings suitable for different implicit analysis
 types Accompanied by some basic examples.
 Purpose is to reduce the effort of getting started with implicit analysis
 also includes information about Implicit Mortar Contact Problems

T-joint component
*PART_INERTIA
v0= 5 m/s
*CONTACT_AUTOMATIC_SINGLE_SURFACE
overall contact
*MAT_024
DP 800
*MAT_COHESIVE_MIXED_MODE (_138)
adhesive bond with failure
5 mm mesh
for steel parts
*MAT_024
wooden blocks
*CONSTRAINED_RIGID_BODY
lower sheet and wooden block
run with LS-DYNA version R7.1.1 MPP, single and double precision

■ Dynamic explicit
velocity [0 - 10 m/s]
■ Process time = 5 ms
■ ~10,000 time steps
■ 52 cohesive elements fail
■ Low-frequency vibration and
high-frequency response
(wave propagation)
force in kN
displacement in mm
Now, we want do a static analysis of that process
Step by step
1. Start with explicit using a larger time period

“slow“ loading
2. Add implicit cards needed for dynamic implicit analysis

“fast“ and “slow“ loading
3. Remove dynamics and perform pure static analysis

no physical time – only process time

■ Static (??) explicit velocity [0 - 3 m/s]
■ Process time = 5 / 50 ms
■ Circa 10,000 / 100,000 time steps
■ No initial velocity, but prescribed motion
■ 52 cohesive elements fail
■ Response still dynamic
■ Damping … ??
5 ms – 10000 steps
50 ms – 100000 steps
Force in kN
Displacement in mm
■ Dynamic implicit (default)
■ Process time = 5 ms (“fast“) ■ 100 steps
■ *CONTROL_IMPLICIT_GENERAL ■ 2779 problem cycles
DT0 = 0.05 (100 steps) ■ 58 failed cohesives
■ *CONTROL_IMPLICIT_DYNAMICS
IMASS = 1
■ + Recommendations
■ *CONTROL_ACCURACY
OSU=1 ■ 51 steps
■ *CONTROL_IMPLICIT_SOLUTION ■ 1063 problem cycles
NSOLVR=12, ILIMIT=6, ■ 52 failed cohesives
DNORM=1 (DCTOL=0.005)
■ *CONTROL_IMPLICIT_AUTO
ITEOPT=30, ITEWIN=10, DTMAX=0.1

■ Dynamic implicit (default)
Force in kN
■ Process time = 5 ms (“fast“) explicit
■ *CONTROL_IMPLICIT_GENERAL implicit
DT0 = 0.05 (100 steps)
■ *CONTROL_IMPLICIT_DYNAMICS
IMASS = 1
Displacement in mm
■ + Recommendations
■ *CONTROL_ACCURACY
Force in kN
OSU=1 explicit
implicit
■ *CONTROL_IMPLICIT_SOLUTION
NSOLVR=12, ILIMIT=6,
DNORM=1 (DCTOL=0.005)
■ *CONTROL_IMPLICIT_AUTO
ITEOPT=30, ITEWIN=10, DTMAX=0.1
Displacement in mm
■ Dynamic implicit
■ What time step size is necessary to resolve the dynamic process?
■ User needs good knowledge about the problem at hand
■ User has to decide about the solution frequency
■ Contact dominated problems need small time steps
10000 explicit steps 10000 explicit steps

50 implicit steps 200 implicit steps
Force in kN
Force in kN
Displacement in mm Displacement in mm

■ Dynamic explicit ■ Dynamic implicit
■ Low- and high-frequency response ■ Low-frequency response

■ Dynamic implicit
■ Check influence of Newmark parameters
Number of iterations
GAMMA and BETA ∑ 1063 cycles
■ Default: GAMMA=0.5, BETA=0.25
■ Larger GAMMA and BETA values
introduce numerical damping
■ Often helps convergence ∑ 832 cycles
■ But: affects solution!
Time in ms
GAMMA=0.5, BETA=0.25
GAMMA=0.6, BETA=0.38
Force in kN
GAMMA and BETA can be used to

introduce some numerical damping
2
1 1 1
𝛾𝛾 ≥ , β ≥ 𝛾𝛾 +
2 4 2
Time in ms
■ Dynamic implicit 100000 explicit steps
■ Process time = 50 ms (“slow“) 50 implicit steps
■ Compare to “slow“ explicit run
Force in kN
Displacement in mm

■ Static implicit
■ Remove *CONTROL_IMPLICIT_DYNAMICS
■ No initial velocity, but prescr. motion no velocities!
■ “Time“ not physical anymore
■ Real static response
■ Statically defined !?!
explicit
Force in kN
implicit
Displacement in mm
■ Eigenvalue analysis
■ *CONTROL_IMPLICIT_EIGENVALUE
■ Reveals possible rigid body modes
■ Superelevated deformations in d3eigv database

■ Implicit contact
■ Contact is very important issue Explicit (“slow“)
Force in kN
(especially) in implicit analysis IGAP on
■ User should know about IGAP MORTAR
options (“sticky contact“)
and Mortar contact
too early
(continuous tangent)
with IGAP
■ Dynamic implicit shown here
Displacement in mm
IGAP on MORTAR

■ Static implicit with Mortar contact
first of 6 rigid
■ “Missing“ contact gap now reveals
body modes
6 rigid body modes (wooden block)
■ Additional action(s) needed
to allow for static analysis
■ Slight scaling of wooden block‘s
size causes initial contact penetration
to get statically determined system
■ +IGNORE=1 to avoid initial contact forces
only 1 rigid body

mode left, will be
kept by friction
(hopefully)

■ Static implicit with Mortar contact
■ More realistic results with Explicit (“slow“)
Mortar contact IGAP on
Force in kN
MORTAR
■ 5 different phases can be observed: (iv)
(i) no contact, (iii)
(ii) tipping, (v)
(iii) elastic bending, (ii)
(iv) adhesive softening and (i)
(v) glue failure Displacement in mm

■ Static implicit with Mortar contact IGAP on
Number of iterations
■ Convergence becomes MORTAR
more difficult
■ Reason(s) for difficulties can be
detected with special “iteration
plot database“ d3iter
■ Evolution of out-of-balance forces
during iteration process shows
critical areas
”Process time“
Troubles from damage

evolution in cohesive
material and contact to
impactor

■ Ideas for improvement
■ Perhaps Full Newton better
suited for this problem (ILIMIT=1)
■ Modify other implicit settings
(timestep size, tolerances, …)
or contact parameters
(IGAP, )
■ But maybe better to improve
the model itself:
■ Replacement for cohesive material
(MAT_186 with smooth curve?)
■ Mesh refinement in critical areas?
■ Dynamic implicit – very slow
■ …

Summary
■ Explicit analysis runs into its limits for long
duration processes or even real static load cases
■ Therefore, implicit analysis is often preferable.

Actually, computation time can be decreased in
many cases
■ But: more demanding to get a solution, especially

for large deformations, contact and nonlinear
material behavior Often greater effort is needed to
obtain a functional model in implicit
■ Users must be aware of crucial differences But also the feeling of success
between explicit (e.g. time step size) and implicit
is greater in the end
(e.g. “smooth” model)


Implicit Analysis Usimg LS-DYNA

Uploaded by

Copyright:

Available Formats

Implicit Analysis Usimg LS-DYNA

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Implicit Analysis Usimg LS-DYNA

Uploaded by

Copyright:

Available Formats

DYNAmore Express

Implicit Analysis using LS-DYNA

Dr. T. Erhart, Dr. T. Borrvall

Copyright 2020 DYNAmore GmbH. All rights reserved.

Copyright 2020 DYNAmore GmbH. All rights reserved.

Explicit scheme 𝐱𝐱 n+1 = 𝐟𝐟 𝐱𝐱 n , … Implicit scheme 𝐟𝐟 𝐱𝐱 n+1 , 𝐱𝐱 n , … = 𝟎𝟎

+ Solution: directly - Solition: iteratively

- Many small time steps + Few large time/load steps

Equilibrium ? Energy balance ! Equilibrium ! Convergence ?

Short time dynamics Structural dynamics

Copyright 2020 DYNAmore GmbH. All rights reserved.

■ Consequences for FE models

Copyright 2020 DYNAmore GmbH. All rights reserved.

Convergence behavior is depending

Difference in physics ”Increased accuracy implies

Different method(s) for solving

Copyright 2020 DYNAmore GmbH. All rights reserved.

■ Time/Load step size

■ Rigid body motions

Copyright 2020 DYNAmore GmbH. All rights reserved.

■ Use double precision ( _d_ in the name of the executable)

■ Read Appendix P in the User’s manual and Chapter 37 in Theory Manual

Copyright 2020 DYNAmore GmbH. All rights reserved.

■ For a cluster node with 256 GB = 32000m of available RAM

Copyright 2020 DYNAmore GmbH. All rights reserved.

■ Huge dynamic and much less static memory needed Dynamic

■ This does not mean LS-DYNA requires less memory, Static

Copyright 2020 DYNAmore GmbH. All rights reserved.

Copyright 2020 DYNAmore GmbH. All rights reserved.

Copyright 2020 DYNAmore GmbH. All rights reserved.

Variable OSU INN PIDOSU IACC

■ Higher accuracy in selected material models (24, 123, …)

■ Strong objectivity and consistency in selected tied contacts

■ Strong objectivity in selected element types

■… see more in User’s Manual

Copyright 2020 DYNAmore GmbH. All rights reserved.

Copyright 2020 DYNAmore GmbH. All rights reserved.

■ After successful steps

■ After failed steps

■ Exponential algorithm for adjusting step size

Copyright 2020 DYNAmore GmbH. All rights reserved.

■ Consequences of ill-conditioning accuracy digits lost ≈ log10 C 𝐊𝐊

■ If a set of equations is seriously ill-conditioned

Copyright 2020 DYNAmore GmbH. All rights reserved.

■ Carefully inspect input deck and check …

Copyright 2020 DYNAmore GmbH. All rights reserved.

Deformation history Residual force evolution

Copyright 2020 DYNAmore GmbH. All rights reserved.

Copyright 2020 DYNAmore GmbH. All rights reserved.

Advanced and basic examples

Starter kit including guidelines

Copyright 2020 DYNAmore GmbH. All rights reserved.

Copyright 2020 DYNAmore GmbH. All rights reserved.

1. Start with explicit using a larger time period

2. Add implicit cards needed for dynamic implicit analysis

3. Remove dynamics and perform pure static analysis

Copyright 2020 DYNAmore GmbH. All rights reserved.

Copyright 2020 DYNAmore GmbH. All rights reserved.

10000 explicit steps 10000 explicit steps