Nonlinear Analysis:

Modeling Considerations

Arturo Valentin, PhD

and the staff at ADINA R&D, Inc.

March 8, 2016

© ADINA R&D, Inc., 2016 1

 Chief Concerns
• Achieving convergence
• Ensuring accurate solutions
– Inelastic models (plasticity, viscoelasticity)
– Frictional contact
– Dynamic analyses
• Minimizing expense

Our task is to meet these (sometimes competing) goals by

• Monitoring convergence
• Selecting appropriate convergence tolerances
• Identifying convergence barriers
• Overcoming those barriers appropriately
• (Temporary) problem simplification

© ADINA R&D, Inc., 2016 2

 Diagnostics
First recommendation: always use
detailed diagnostics. This provides
crucial information!
 Stiffness matrix conditioning
 Plasticity status
 Contact information
 Oscillatory behaviour

Causes the .out file to be larger

(but not the .por file).

Also offers “Symptoms” and

“probable causes and remedies”
when convergence is not achieved

© ADINA R&D, Inc., 2016 3

 Interpreting the .out file
Quadratically converging model:

OUT-OF- NORM OF CONVERGENCE RATIOS CONVERGENCE RATIOS OUT-OF-BALANCE LOAD

BALANCE OUT-OF-BALANCE NORM OF INCREMENTAL FOR OUT-OF-BALANCE FOR INCREMENTAL VECTOR CALCULATION
ENERGY FORCE MOMENT DISP. ROTN. CFORCE ENERGY FORCE DISP. CFORCE BETA RATIO
NODE-DOF NODE-DOF NODE-DOF NODE-DOF CFNORM MOMENT ROTN. (ITERNS)
MAX VALUE MAX VALUE MAX VALUE MAX VALUE
COMPARE WITH COMPARE WITH
ETOL RTOL DTOL RCTOL
1.00E-03 (NOT USED)(NOT USED)(NOT USED)

ITE= 0 1.00E-05 8.53E-07 0.00E+00 1.47E-02 0.00E+00 0.00E+00 1.00E+00 1.53E-04 1.20E-01 0.00E+00
3 0 4 0 0.00E+00 0.00E+00 0.00E+00
-6.08E-07 0.00E+00 -9.09E-03 0.00E+00

ITE= 1 5.77E+01 5.11E+05 0.00E+00 1.13E-04 0.00E+00 0.00E+00 6.91E-03 1.00E+01 9.21E-04 0.00E+00 1.00E+00 9.00E-01
3-Y 0-F 3-Y 0-F 0.00E+00 0.00E+00 0.00E+00 ( 1)
1.81E+05 0.00E+00 3.99E-05 0.00E+00

ITE= 2 2.23E-08 1.00E+01 0.00E+00 2.22E-09 0.00E+00 0.00E+00 2.67E-12 1.97E-04 1.81E-08 0.00E+00 1.00E+00 -1.97E-05
3-Y 0-F 3-Y 0-F 0.00E+00 0.00E+00 0.00E+00 ( 1)
-3.55E+00 0.00E+00 -7.85E-10 0.00E+00

Using ETOL as a criterion, but not RTOL, DTOL,

or RCTOL (non contact problem)

© ADINA R&D, Inc., 2016 4

 Failed Convergence
Diagnostics:
______________________________________________________________________________________________________
** FAILED CONVERGENCE ANALYSIS **

Trend: No trend detected from iteration history

Actions: Iterations terminated early due to mesh distortion

Symptoms:
- Contactor nodes still changing status between contact and gap
- Big increase in out-of-balance forces at iteration 1
- Incremental displacements large from iteration 0 until end ( 2.615E+00 % of model size)
- Integration points still changing status between elastic and plastic
- Contactor nodes still going in and out of target surface boundaries

Most probable causes and remedies:

- Big displacements at beginning of time step (compared to model size)

* Apply loads more gradually, or use smaller time step

- Contactor nodes exceeding target surface boundaries

* Increase 'Contact Surface Extension Factor'
* Reverse contactor and target surfaces
* Switch to 'Small Displacement Contact'

- Plasticity nonlinearity
* Apply loads more gradually, or use smaller time step
* Toggle 'Plasticity Type' algorithm
* Toggle 'Line Search' feature

© ADINA R&D, Inc., 2016 5

 Remedies
• If the solution was converging,
increase MAX-ITER
• Enable line search
• Change plasticity algorithm
• Use smaller load increments
– Manually reduce time step size
– Enable ATS (perhaps with Low-
Speed Dynamics)
• Improve conditioning!

© ADINA R&D, Inc., 2016 6

 Stiffness Stabilization
MASTER STABILIZE
Stabilizes static problems
with rigid body modes (e.g.,
improperly supported bodies)

Rigid body modes – singular

stiffness matrix (zero pivots);
causes ADINA to abort

Scales diagonal stiffness

terms without affecting the
right-hand-side vector.

Will slow or even prevent convergence, but will not affect the accuracy of
the converged solution in nonlinear problems (within specified tolerances).

© ADINA R&D, Inc., 2016 7

 Stiffness Stabilization
There are three stiffness
stabilization options available:
• STABILIZE=NO
• STABILIZE=YES
• STABILIZE=AUTOMATIC
(activated if needed).

The defaults are STABILIZE=NO

and STABFACTOR=1e-10.

By default, STABILIZE=NO as it
affects the solution in linear analysis
(but not in nonlinear analyses).

The stabilization constant is non-dimensional; it should always be small.

© ADINA R&D, Inc., 2016 8

 Weak Springs
Another widely-used method to prevent zero pivots is to manually add
weak springs to the parts with rigid body modes. However:
• determining the number, location and stiffness of the springs requires
more user intervention
• springs generate internal forces which affect the final solution
• difficult to assess how much the springs have affected the solution

Physically, stiffness stabilization applies weak springs to all displacement

DOFs in the model, but in such a way not to affect the accuracy of the
solution in nonlinear analyses.

Effect of stabilization factor STABFACTOR

• Small STABFACTOR leads to no stabilization
• Large STABFACTOR leads to slow convergence or no convergence
• Does not affect nonlinear solution

© ADINA R&D, Inc., 2016 9

 Condition Number

A high condition number, as indicated by the message:

Model may be unstable, ratio of diagonals 1.E11, please check your input data.

Indicates instability (e.g., zero stiffness) or because a zero pivot is

hit before reading glue mesh equations.

• Can be physical (e.g., Euler buckling) or due to poor modeling

practices
• Will affect the convergence rate or prevent convergence
altogether
• User should determine what is causing the ill-conditioning and
resolve the cause, if possible

© ADINA R&D, Inc., 2016 10

 Stiffness Matrix Conditioning

In some (quasi-)static problems (e.g., forming) there are

unsuppressed rigid body modes before contact is established.

• Stabilize the stiffness matrix using:

– Use stiffness matrix stabilization
– Use Low-Speed Dynamics
– Apply contact damping
• Prevent large motions as surfaces come into contact
– Use small initial steps until contact is established
– Prescribe displacements

© ADINA R&D, Inc., 2016 11

 Damping
Nonlinear static analyses can be amongst the most
challenging problems because they do not include inertial
and rarely any damping effects!

Adding damping (dissipative) may help in some problems.

There are different ways that damping can be included:
• Low speed dynamic analysis
• Rayleigh/Contact damping
• Physical damping
– Plasticity/viscoelasticity
– Friction

© ADINA R&D, Inc., 2016 12

 Full Dynamic Analysis
Of course, one could run a dynamic analysis. It is more expensive,
but the effects of inertia are then included.

As time steps get smaller, inertial effects dominate and tend to

stabilize the problem

• For implicit dynamic analyses, the Bathe method should be used

• ‘Cuts out’ spurious high frequency response but accurately

captures low frequency modes (provided the mesh can resolve
those modes)

• No need to adjust parameters – time step size dictated by the

physics of the problem

• Stable and accurate – large time step can be used

© ADINA R&D, Inc., 2016 13

 Low-Speed Dynamics
ATS includes a special “Low-
Speed Dynamics” option for
static analyses. “Inertia relief.”
When enabled, low-speed
dynamic includes mass and
damping effects in the (otherwise)
static analysis.

• Can be applied to whole model

or specific element groups
• Can automatically activate if no
convergence
• Can be disabled again after a
specified time interval

© ADINA R&D, Inc., 2016 14

 Low-Speed Dynamics
Especially useful for problems involving:
• Collapse
• Wrinkling
• “Snap through”
• Limit point instabilities
• Rigid body modes before contact is established

Unlike stiffness stabilization, low-speed dynamics will affect the

solution! Constant  can be interpreted as a decay time (time
constant of exponential decay). Increasing  increases the
decay time.

It is recommended that the user:

• Apply large  to achieve convergence, then reduce
• Run problem with different values of  to determine its
effects on the solution parameters of interest.

© ADINA R&D, Inc., 2016 15

 Benchmark: Rubber Mount
1 2

3 4

High-deflection rubber mount

• Obtain force-displacement
response 5 6
• Axial compression
• Using an axisymmetric
model

© ADINA R&D, Inc., 2016 16

 Benchmark: Rubber Mount

High-deflection rubber mount

• Obtain force-displacement
response
• Axial compression
• Using an axisymmetric model

© ADINA R&D, Inc., 2016 17

 Benchmark: Rubber Mount

© ADINA R&D, Inc., 2016 18

 Low-Speed Dynamics

© ADINA R&D, Inc., 2016 19

 Low-Speed Dynamics

© ADINA R&D, Inc., 2016 20

 Low-Speed Dynamics

Parametric study
LSD “always on”

Smaller  :
Increasing accuracy,
Decreasing stability

© ADINA R&D, Inc., 2016 21

 Low-Speed Dynamics

• Can be applied to whole model

or specific element groups
• Can automatically activate if no
convergence
• Can be disabled again after a
specified time interval

This feature allows the user to take

advantage of stabilizing effects
without worrying about excessive
loss of accuracy

© ADINA R&D, Inc., 2016 22

 Low-Speed Dynamics

© ADINA R&D, Inc., 2016 23

 Low-Speed Dynamics
Stiffness stabilization is good for
removing rigid body modes from static
models, but it cannot help with
instabilities associated with:
• Buckling
• Wrinkling
• Contact

Low-speed dynamics also removes

rigid body modes and it can also
stabilize many kinds of instabilities.
Can be applied selectively:
• When no convergence
• Aids convergence without
excessively affecting solution, even
with a larger damping factor .

© ADINA R&D, Inc., 2016 24

 Contact Damping
CGROUP DAMPING
• Useful for stabilizing static problems and removing high
frequency oscillations in dynamic problems
• Adds normal and tangential viscous dampers to all contactor
and target nodes
• Can act only in the first time step (DAMPING=INITIAL), in
which case, contact damping will not affect the solution! Can
safely use a very large damping factor to achieve convergence
in the first time step.
• Constant damping throughout the analysis
(DAMPING=CONSTANT)
• If it remains constant, the energy associated with damping is
written in .out file

© ADINA R&D, Inc., 2016 25

 Rayleigh Damping
C=M+K
Where ,  are the
Rayleigh damping
coefficients.
• Mass-proportional
coefficient  resists
rigid body motions
• Stiffness-proportional
coefficient  resists
distortions

Mass-proportional term damps the low frequency modes

(including the rigid-body motions), and the stiffness-
proportional term damps the high frequency modes.

© ADINA R&D, Inc., 2016 26

 Rayleigh Damping
Reasons for (and ways of) using Rayleigh Damping:
To model physical damping in the real structure: Both mass-proportional and
stiffness-proportional damping are applied. The Rayleigh damping factors are
determined by fitting the Rayleigh relationship to two modes where the damping
ratio is known from experiment.

To stabilise the model: Only mass-proportional damping is applied. [e.g., to

stabilise problems that are improperly supported before contact is established]

To smooth the response: Only stiffness-proportional damping is applied.

Stiffness-proportional damping removes high frequency oscillations. [The high
frequency oscillations might be physical oscillations or spurious oscillations such
as chatter or ringing. Spurious high frequency oscillations are most pronounced in
the contact forces, accelerations, and stresses.]

© ADINA R&D, Inc., 2016 27

 Energy View
• Rayleigh damping
dissipates energy
• Dissipated energy can be
computed and output by
ADINA’s energy view
feature
• Monitor this energy to
assess if the contact
compliance factor is too
high

© ADINA R&D, Inc., 2016 28

 Energy View
• Rayleigh damping
dissipates energy
• Dissipated energy can be
computed and output by
ADINA’s energy view
feature
• Monitor this energy to
assess if the contact
compliance factor is too
high

If Rayleigh damping is applied

to stabilise the problem or
smooth the response, the
damping energy should be
much smaller than the other
energies.

© ADINA R&D, Inc., 2016 29

 Accuracy Indicators
Solution accuracy indicators in the .out file after each
converged solution
Show if the forces due to the various stabilization and
conditioning effects are excessive

If the accuracy indicators are all small (say, less than 1% of the external
force indicator), then the solution may be reasonably accurate.

© ADINA R&D, Inc., 2016 30

 Iteration Tolerances
• Energy
– Based on out-of-
balance energy norm
• Force/Moment
– Based on norm of
residual forces and
moments
• Displacement/Rotation
– Based on norm of
incremental
displacements and
rotations
• Contact

© ADINA R&D, Inc., 2016 31

 Achieving Convergence
The necessary conditions for
quadratic convergence are strict.

• Fastest type of convergence

• Truly tangent stiffness matrix

• Close to accurate solution –

within the “radius of
convergence”

Observing quadratic convergence

gives a high degree of confidence
that the solution has accurately
converged.
 force norm ITE   force norm ITE 1
2

© ADINA R&D, Inc., 2016 32

 Convergence Criteria
For an accurately converged
solution, we expect to see:

• The maximum incremental

displacement should be
Unreliable
small as compared to the
solution!
initial iteration

• The norm of the out-of-

balance force should be
small as compared to the
initial iteration
The solution norms should not
be oscillating wildly!

© ADINA R&D, Inc., 2016 33

 Setting Appropriate Tolerances
If tolerances are too loose:

• Lost energy and lost momentum, especially if impact is involved.

• Inaccurate impact forces as convergence can be very slow for

small time step sizes; see contact impact.

• Inaccurately converged plasticity solution, which can eventually

lead to divergence.

• Instabilities such as chattering (or ringing), especially when contact

is involved.

• A diverging solution such that the solution “blows up”.

In nonlinear analyses, implicit solutions can diverge if the

convergence tolerances are too loose

© ADINA R&D, Inc., 2016 34

 Basic Guidelines
By default ADINA uses the energy criterion with loose tolerances.
For highly nonlinear problems, it is a good idea to:

• Use the energy and force convergence criteria: The force

criterion is more conservative (stricter) than energy; better
measure to ensure the equations have tightly converged

• Tighten the energy convergence tolerance: Setting

ETOL=1.0e-6 is a good starting point. When energy and force
criteria are used, the energy norm is typically < 1.0e-10 at
convergence

• Tighten the contact force tolerance: Setting RCTOL=0.005

(0.5%) is a good starting point; RCONSM should be three or four
orders less than CFNORM at convergence.

© ADINA R&D, Inc., 2016 35

 Limiting Displacements
TOLERANCES MAXDISP
Useful for problems where one or more iterations can produce
excessive incremental displacements
• if a load is applied to contacting bodies before contact is properly
established
• during the first unloading steps after a material has undergone
plastic deformation.

The excessive displacements lead the program away from the

converged solution, making the return to the proper solution difficult.

MAXDISP does not stabilize the stiffness matrix so in many cases it

may be necessary to use MAXDISP together with stiffness
stabilization and/or contact damping/LSD.

© ADINA R&D, Inc., 2016 36

 Solution Parameters
Quadratic convergence is only possible when
we are sufficiently close to the solution – within
the radius of convergence.
Line search
Line searches can be used to expand the radius
of convergence:

• Contact initiation/separation.
• Onset of metal plasticity, or elastic unload
after plastic deformation.
• Rubber problems with wrinkling and folding
instabilities.
• Buckling and collapse problems.
• Large displacement problems involving
beams and shells.
• Problems with element birth/death, and
element death upon rupture.

© ADINA R&D, Inc., 2016 37

 Solution Parameters
MESH TOO DISTORTED (JACOBIAN DETERMINANT NOT POSITIVE)

ITE=2

Diagnostics: -------------------------------------------------
Elements switching from elastic to plastic.
Plasticity Algorithm
--------------------------------------------------------------

ITE=3

Diagnostics: -------------------------------------------------
Elements switching from plastic to elastic.
--------------------------------------------------------------

Type 2 plasticity algorithm

• can sometimes obtain convergence
when elements invert (Jacobian
determinant becomes non-positive)
• Allows large time steps to be used.
• Can prevent oscillation between elastic
and plastic deformation during the
equilibrium iterations

© ADINA R&D, Inc., 2016 38

 Analysis Restart
• Sometimes, it can be useful to analyze a part of a problem
statically and another part dynamically or change some
solution/model parameters
• Or perhaps implicit analysis fails to converge – must use
explicit for part of the analysis
• ADINA’s restart feature (MODEX=RESTART) allows the user to
continue an analysis from a previous run

MASTER IRINT specifies the frequency of writing analysis results

to the restart file
= 0 Default. Restart file is overwritten every step
> 0 Restart file is overwritten every IRINT steps.
< 0 Restart file is appended every IRINT steps.

© ADINA R&D, Inc., 2016 39

 Analysis Restart

Some other supported changes:

• Loads and/or boundary conditions
• Convergence tolerances
• Contact parameters (but not between friction and
frictionless)

© ADINA R&D, Inc., 2016 40

 Analysis Switch
Switches analysis type automatically:

• Switching between static/implicit dynamic/explicit dynamic

formulations
• Switching to a frequency analysis after a static or dynamic
analysis
• Useful for determining natural modes in a loaded
(deformed) configuration

© ADINA R&D, Inc., 2016 41

 Temporary Simplifications
Reducing complexity improves “signal to noise ratio” in .out file and
facilitates solving one convergence issue at a time.
• Material nonlinearities
– Extend plasticity curves (prevent rupture)
– Specify a positive strain hardening modulus
• Loading
– Specify deformation independent loading
– Apply loads more gradually and/or prescribe displacements
• Contact
– Remove friction and/or use friction delay
– Add contact compliance
– Add contact damping

© ADINA R&D, Inc., 2016 42

 Basic Strategies

• Add boundary conditions

• Use stiffness stabilization
• Use Low-Speed Dynamics
• Use damping (contact or Rayleigh)
• Limit maximum displacements
• Use line search/toggle plasticity algorithm
• Use prescribed displacements instead of forces where
possible
• Run an implicit dynamic analysis using the Bathe method

© ADINA R&D, Inc., 2016 43