ANSYS Nonlinear

Convergence
g Best
Practices

Peter R. Barrett, P.E.

September 27, 2012

© 2008 CAE Associates

ANSYS Nonlinear Convergence Best Practices

 Nonlinearities Overview:
— Large Deflection
— Material Nonlinearities
— Contact

 Characterize Convergence Difficulty with Examples

— Relatively Straight Forward (easy) Problems
— Challenging (i.e. really hard) Problems

 Step-by-Step
p y p Convergence
g Procedure
1. Rigid body motion
2. Force balance not obtained
3. Material Instabilities and/or Element formulation error

 Q&A

2
All Products Exhibit Nonlinearities

 Always start with the simplest linear analysis if possible!

3
Large deflection affects stiffness

4
Material Properties Cannot Always Be
Assumed to Be Linear

5
Contact is the Most Common Source of
Nonlinearity and is Often the Most Difficult
tto S l !
Solve!

Status
changes,
friction,
pressure

When,
When where?
What is the pressure?

6
Characterize Convergence Difficulty
 Easier Problems
— Deformations
D f ti are relatively
l ti l smallll
— Nonlinear strains (plasticity, creep, swelling) are small
— Contact status does not oscillate
— M d l are smallll and
Models d simplified
i lifi d (2D
(2D, Axisymmetric)
A i ti )
— Symmetric boundary conditions are utilized
— Displacement based loads
— L d result
Loads lt iin ttensile
il memberb stresses
t
— Nonlinear buckling to the point of instability (Post buckling not needed)

409 Parts
967 Contact Pairs

7
Characterize Convergence Difficulty
 Harder Problems
— Very large
V l d
deformation
f ti
— Large strains with large distortion
— Contact chatter and/or loose fitting assemblies
— C t t sliding
Contact lidi with
ith hi
high
h ffriction
i ti coefficient
ffi i t
— Post buckling response
— Large 3D models with complex geometry
— N symmetry
No t b boundaries
d i
— Force based loads

8
Pin Insert Model
 Hard Solution
— Model the entire Pin / socket assembly
— Mesh fine enough to capture local
stress concentrations
— Use a force based analysis to model
pin insertion and removal
— Determine critical locations / load
steps
 Easier Solution
— Model a single axi-symmetric
Pin/Socket assembly
— Create mapped mesh with refinement
on contact surfaces and areas of high
stress
— Use a displacement controlled solution
— Use auto time stepping and smart
output controls since max stress might
not occur at the final solution step

9
Gasket Assembly

 Hard Problem
— Large model
— Complex loading sequence
— Multiple bolt loads
— Frictional Contact
— Nonlinear material response

10
Automatic Contact Reduces Modeling Time

11
Building Collapse Simulation

 Running an analysis dynamically can be used to obtain convergence

— Evaluating the thermal-structural response of the World Trade Center collapse

12
Elastomeric Bearing with Lead Plug

Force
Deflection of a
Single Truss
Element
Matches
Detailed 3d
Model

13
Clevis Pin Pullout

 Springs can be used to imposed loads and/or prevent rigid body motion

14
Surgical Staple – Displ. Controlled Example

15
ANSYS Nitinol Material Behavior
 Complex Material Response requires many substeps for
accuracy and
d convergence

16
Rubber Boot with Self-Contact
 Combined Geometric, Material and Contact Nonlinearities

17
Vena Cava Filter

 IVC filters are used in case of contraindication to anticoagulation

 I.e. – it captures clots!

18
FSI Example – Vena Cava Filter
 Nonlinear structural response coupled with fluid flow

19
FSI Example – Vena Cava Filter

Streamlines

Deformation

20
My Analysis Did Not converge. Now What?

 First step is to determine the cause:

1. Rigid body motion
2. Force balance not obtained
3
3. Material Instabilities
4. Element formulation error
5. Combination of items 1-4 above

 We will examine each in detail including:

— identifying the problem
— d t
determining
i i ththe cause
— providing solutions

21
Understanding the Solver Output

 Whether in WB or Mechanical APDL,

APDL the first step is to read the output file
to determine the origin of the non-convergence

22
Rigid Body Motion Error Messages
 DOF limit exceeded.
 Negative main diagonal.
 Small/Negative Pivot error.
 MAX DOF INC = “A very large number”
*** WARNING *** CP = 11.703 TIME= 16:15:15
Smallest negative equation solver pivot term encountered at UX DOF of
node 98. Check for an insufficiently constrained model.

23
Examples of Rigid Body Motion Cont.

24
Causes of Rigid Body Motion

— Insufficient supports
— Individual parts of an assembly are not supported. This is the most common
form that is found in a contact analysis where rigid body motion occurs
in the parts not associated with any supports.
— Insufficiently connected dissimilar element types (i.e. beams to solids, etc.)

25
Find the Rigid Body Motion

 Plot the unconverged or last converged displacement solution and verify

displacement scaling
— The example below is a converged solution where the rigid body motion is only
restrained by weak springs
— Note the unrealistic 10e-6 displacement scaling

26
Use Modal Analysis to Find Rigid Body Motion

 If it is not clear what constraint is required to eliminate rigid body motion

motion, a
modal analysis can be performed.
— A modal analysis determines the vibration modes including rigid body motion.
— Each rigid body mode is predicted as a zero frequency mode
— Animating the zero frequency mode shape illustrates the rigidly moving model.

27
Correct the Rigid Body Motion
— Take advantage of symmetries

— Axisymmetry
— Rotational
— Planar or reflective
— Repetitive or translational

28
Correct the Rigid Body Motion

 Isolate the error by creating a smaller simpler model – KISS!

— Make a 2d Sector model
— Delete/Suppress parts or fix DOF until you get an answer
 Add supports and/or use displacement controlled solution
— Adding a rigid region and pushing it with displacements will usually converge
better than a force loading.

29
Correct the Rigid Body Motion

 Use bonded contact for debugging

— If interfaces are all in compression, one might be able to leave as bonded
— Modify to standard contact one pair at a time
 Adjust the parts to all start in contact
— Can use adjust to touch ANSYS option, but be careful since the geometry will
be changed
 Increase pinball region
— Corrects interference fits
 Add friction

30
Correct the Rigid Body Motion
 Avoid mistakes from wrong assumptions
— Make sure to include large deflections when displacements are significant
• You can never get the wrong answer by adding large displacement effects
— Check Displacement Scaling
Large deflections included!

Li
Linear solution
l ti

31
Correct the Rigid Body Motion

 Use bonded contact to tie dissimilar

element types
— Example Shells or beams to solids
— Use MPC contact option to eliminate
iterations and penalty stiffness dependency
 Add/adjust weak spring stiffness
— Be sure to check reactions to make sure no
error is introduced

32
Correct the Rigid Body Motion
 Add Contact Stabilization Damping
 Rigid body motion often can occur in the beginning of a static analysis due
to the fact that the initial contact condition is not well established.

Fn
Ft
Pdn  d n u n
Contact
Pd1,d
Pd 1  d t u1
Target
Pdn
2
Pd 2  d t u 2

 Contact Stabilization introduces a viscous damping traction proportional to

b t opposite
but it to
t the
th relative
l ti pseudod velocities
l iti bbetween
t th
the ttwo surfaces
f
along contact normal and/or tangential directions.
 Where: d n = damping coefficient in normal direction
d t = damping coefficient in tangential direction
u = pseudo velocity
33
Contact Stabilization Damping

 Example: Consider a fixed pin interfacing with a hole in plate with initial
radial clearance and under a force based load
— Stabilization captures localized stress distribution more accurately because it
does not change the shape of the pin
Conventional Adjust to Touch Contact Stabilization Damping

34
Correct the Rigid Body Motion
 Buckling Response
— Nonlinear stabilization
• Local instabilities and global
instability.
• Used
U d ttogetherth with
ith liline search
h and
d
automatic time stepping
— Arc-length method
• Circumvent g global instability
y when
forces are applied.
• Simulate the negative slope portion
of a load-vs.-displacement curve.
— Running
R nning a static problem as a
"slow dynamic" analysis in ANSYS
— Running a static problem as a
"slow
slow dynamic"
dynamic analysis in
ANSYS/LS-DYNA

35
Connections that Prevent Rigid Body Motion

 Automation tools can save a lot of time in tying assemblies together

Constraint equations

Springs

Spot
Welds

Beam connections

36
Examples of Force Equilibrium not Obtained

 Definition:
— Convergence value is greater than criterion after min. load increment and max.
number of iterations are solved

37
Equilibrium Iterations

 A nonlinear structure is analyzed using an iterative series of linear

approximations, with corrections.

 ANSYS uses an iterative process called the Newton-Raphson

Method. Each iteration is known as an equilibrium iteration.
Load
F
4 A full Newton-Raphson
3
2 iterative analysis for one
increment of load. (Four
1 iterations are shown.)
u
Displacement
38
Convergence Procedure

 loads, {Fa} - {Fnr},

The difference between external and internal loads } is called
the residual. It is a measure of the force imbalance in the structure.
 The goal is to iterate until the residual becomes acceptably small; less
than the criterion, where the solution is then considered converged.
 When convergence is achieved, the solution is in equilibrium, within an
acceptable tolerance
tolerance.

Fa
{ Fnr
{Fa} {Fnr}

u
39
Newton Raphson Residuals

 Plot Newton Raphson Residuals to determine critical contact pair

— Note that these controls need to be activated prior to the solution

40
Causes of Force Balance not Obtained

 Contact Stiffness is too large

 Load is stepped too rapidly
 For small load increments, MINREF criterion exceeded
 Material instability
 Buckling

Definition of non-convergence,
sum of R (unbalanced forces)
never getst below
b l .5%
5% off th
the
sum of F (sum of external loads,
reactions, etc.))

41
Ideal Load Stepping

 Convergence after 3 to 5 iterations each substep

— Use more substeps to reduce iterations, use less if only one or two are
needed. (One exception: If contact without friction is the only nonlinearity,
sometimes one substepp with lots of iterations can be an efficient solution
method)

Load
Load Step 2
4
3 Load Step 1
2
equilibrium
iterations
1
Substeps

“Time
”
42
Causes of Force Balance not Obtained
 Contact Stiffness is too large
 Oscillation of contact status and force balance caused by large contact
stiffness.
— A typical range of ANSYS penalty stiffness is .01 to 10 times the ANSYS
i t
internal
l stiffness
tiff value
l which
hi h iis a ffunction
ti off material
t i l and
d mesh
hbbutt nott
geometry
F=KU

43
Causes of Force Balance not Obtained
 Load is stepped too rapidly
 U Mi
Use Min, MMax andd starting
t ti substep
b t control
t l tto iimprove convergence. Y
You
cannot get the wrong answer by adding too many substeps
Load

Incorrect
Time Strain Energy

tstart tmin tmax

Rule of Thumb:
The more nonlinearities, the
more substeps required

44
Overcome the Force Unbalance
 Nonlinear Solution Corrective Action
 ANSYS WB Mechanical offers a toolbox of options under the analysis
settings branch for achieving successful convergence.
— Step
p Control - Load steps
p and substeps
p
— Nonlinear Controls - N-R convergence criteria

— Contact Settings
g

45
Overcome the Force Unbalance

 Reduce Normal Stiffness Factor

 Increase the number of Substeps

— General Rule – more nonlinearities – use more substeps

46
Overcome the Force Unbalance
 Change contact stiffness update
• Recommend using iteration based adjustments

 Ramp on the interference fit

 Use Higher Order Elements

• Especially with curved surface contact

 Refine the mesh

• The more points in contact the better the convergence

47
Overcome the Force Unbalance
 Increase the MINREF criterion
• If the criterion is very small this will not effect the solution accuracy

 Extend the Stress-Strain curve

 Adjust the convergence tolerance

• If the unbalance force difference is very small this can increase solution speed
significantly
i ifi tl

48
Review Initial Contact Settings and Eliminate Gaps

 See the ANSYS output file or run CNCHECK before solving

49
Adjust Contact Pinball Region

 Default pinball region is different based on analysis type

type. (see below)
— Increase for large initial penetrations

Default Contact Contact Surface Initial Penetration Behavior

PINB* Classification Behavior KEYOPT(9)

3 x Depth** Rigid/Flex Standard Default (Include Everything)

4 x Depth Rigid or Flex Standard Include Everything/with Ramped effects
2 x Depth Flex/Flex Standard Default (Include Everything)
0.50x Depth Flex/Flex Bonded or No Separation Default (Include Everything)
0.75x Depth
p Rigid/Flex
g Bonded or No Separation
p Default ((Include Everything)
y g)
1.00x Depth Rigid/Flex Bonded or No Separation Include Everything/with Ramped effects

* Values are for NLGEOM,ON and are reduced by 50% for NLGEOM,OFF
**Depth
Depth = Underlying element depth (for solid elements)
**Depth = 4 x element thickness (for shells and beams)

50
Surface Projection Based Contact

 More accurate distribution of contact stresses.

stresses
 Satisfies moment equilibrium when an offset exists between contact and
target surfaces with friction.
 Can help with
ith contact con
convergence.
ergence
 Better handling of sliding contact.
 KEYOPT,<contact element type>,4 ,3

Default Contact Settings Surface Projection Contact

51
Examples of Material Instabilities

 Depending upon the size of the residual these can be caused by large
force unbalance or can be a result of incorrect material properties

52
Causes of Material Instabilities

 Force balance not obtained

— Material law cannot handle large load and/or time increment. If the force
unbalance is very large, the material model might not be the problem
— Element distorted shape results in negative volume calculation
calculation. Check the
mesh, but again with very large force unbalances the material model might not
be the problem

 Too large of plastic or creep strain increment

— Max. increment can be adjusted as part of cutback controls, but typically force
balance is the controlling criterion and thus this is rarely changed.

 Elements with mixed u-P constraints not satisfied

— Modifyy the Volumetric Constraint

53
Causes of Element Formulation Errors

 Element Shape distortion

 Excessive strain
 Volumetric locking (PLESOL,NL,HPRES)
 Hourglass modes
— Using Higher order elements or change to
enhanced strain to eliminate reduced
integration issues in lower order elements
 Buckling
— Reduce rate of loading
 V
Very llarge fforce unbalance
b l

54
Correct Element Formulation Errors

 Mechanical APDL commands

— Use CHECK command for overall verification including missing elastic
properties, unconstrained model, and element shape checks.
— MCHECK command can help you identify defects in the mesh such as holes or
cracks.

55
Correct Element Formulation Errors
 Make sure to use the correct material input
 Make sure to input True Stress vs. Log Strain
— Conversion from engineering data , :

 l  ln 1       1   
 Always extend your material law will beyond
expected
p strain levels

56
Element Selection

 Use Higher Order elements for curved surface contact for faster
faster, more
accurate results

20 node bricks

57
Solver Type and Tolerance

 Suggest using Direct (Sparse) solver unless PCG is significantly faster per
iteration

58
Correct Element Formulation Errors
 Add intermediate substeps
p
 Rezoning
— Requires a restart so generally only used for very large strain cases like
forming operations or seals
 Adjust the starting mesh shapes – this is the most common solution
 Change Element type / formulation
 Use shell or beam elements
— Faster and more efficient solutions where applicable
 Test with 1 Element model
— Change the HyperElastic material model
— Modify creep law or coefficients

Start mesh shape so that deformed

elements become more square

59
Element Type Selection

 Select the appropriate element type to maximize results efficiency and

quality
 Simple is almost always better and definitely easier to debug

Complex 3-D
geometries

Slender structures
(twisted pipe
Shell elements
model)

60
Remesh the Model Part Way Through the Analysis

 Rezoning
— Note that this requires a restart, but will map existing stresses and strains so
that the solution history is preserved

61
Change Material Models

62
Hyperelastic Material Models

 Yeoh is very robust for large strain problems such as seals

63
The Value of the One Element Test Case
 Testing nonlinear material models
— Make sure the material converges for all stress / strain levels expected with the
one element model before running full model
— Especially critical step for hyperelastic and creep analyses
 Testing of macros and user-defined routines
 Evaluation of the impact of large aspect ratios, skew angles or warped
elements

64
Element Shape

 Robust well shaped elements can improve solution convergence and time

65
Mesh Controls

 Numerous controls are available to locally modify the mesh to reduce

stress gradients through multiple elements
— Example showing the value of symmetry

66
Mesh Refinement

 Automated mesh refinement will increase the solution quality and

sometimes speed
— Automated mesh refinement in workbench is generally only used in linear
analyses

67
Element Formulation

 Solver Output records the element technology being activated based on

the element order chosen (midside nodes) and the material association.
Workbench uses higher order 2d elements by default
Elastic material or
metal plasticity with
higher order
elements
Default URI
2D Plane Stress
Elastic material or
Metal Plasticity with
lower order elements
Enhanced Strain
2D Plain Strain
Elastic material or
Metal Plasticity with
lower order elements Simplified Enhanced Strain
Fully incompressible
hyperelasticity with
h h or lower
higher l
order elements
B-Bar with Mixed u-P
68
You Have a Solution!

 How do you make the solution more efficient?

 Change the mesh
— Refine areas of steep gradient
— C
Coarsen areas where
h stresses
t are llow
— Adjust initial element shapes to create better deformed shapes
 Change the Substep settings
— Add substeps to reduce bisections
— Reduce substeps where convergence takes 1 or 2 iterations max.
 Include Material Nonlinearities?

69
Improve Performance in Subsequent Analyses

 The Force Convergence graph clearly indicates that starting with more
substeps would eliminate the 27 iterations performed before the first
bisection. Reducing the starting number of substeps might eliminate this
bisection

Force Residual vs. Iterations

Time vs. Cumulative Iteration

70
Thoroughly Investigate Your Results

 List and plot results to check to make sure you solved the problem
intended and that the results make sense

71
Check the Quality of Your Results - Forces

 First check should always be a free body diagram

Applied loads

Matching reaction forces

72
Check the Quality of Your Stress Results

 Comparing average vs.

vs un-average stresses can provide an estimate of
mesh error

73
Understand the Data Generated

 Displacements are based on original coordinate system

 Stresses and strain component data rotate with the elements

74
Need More Information?

 Additional Information on all of these topics is available in the ANSYS Help

or through taking training classes

75