ANSYS Users Group

Discussions

Promotions

20,000 members

Jobs

Members

Search

How to tackle Convergence issues without compromising

accuracy in any static structural non linear analysis using
ANSYS Workbench 13.0 ?
Follow Venkat

Member

Venkat Anumula
Manager-CFD with automotive engineering background

Like Comment (15) Share Follow Reply Privately August 30, 2011
Comments

Top Contributors in this Group

Guillaume V.
Research engineer chez SDTools
Follow Guillaume
See all members

Your group contribution level

Start by commenting in a discussion. Group
participants get 4x the number of profile views.

15 comments Jump to most recent comment

Thomas M. Hermann
Aerospace Engineer, Odonata Research LLC
Thomas
M.

There is no cookbook method for assuring convergence. You need to look at the results prior to
divergence and try to identify what is causing the issue. I check the following things in this order:
(1) boundary conditions; (2) increasing the number of substeps; (3) decreasing the size of the
load step and add more load steps; and (4) explicitly specifying the solver and solver options.
Finally, it could be a result of your material properties or your geometry/mesh.Good luck,Tom
Like Reply privately Flag as inappropriate August 30, 2011
Rick Kerner
Lead Project Engineer at CNH Benson

Rick

In Classic ANSYS, one can check the element quality and either correct error elements or delete
them in order to prevent convergence issues. In WB 13, I do not have enough seat time to
understand how element quality is maintained. Also, see if you can view the max and min
coefficients of the stiffness matrix: if the difference between the min and max is over 10^9, you
may experience convergence issues. You should be able to see this in the Solution Information
window as it is similar to the Output window in Classic.

Getting Started

Latest Activity
Tuomas Eerola likes a discussion in
ANSYS Users Group. Rainer
W. ANSYS and public clouds
Read how MacGregor was
able to scale out business
critical ANSYS simulations from privete
hosted cloud to public clouds. The
whole case study report is freely
available from: http://www.... more

Like Reply privately Flag as inappropriate August 30, 2011

Ajay Kardak
Co-op at ANSYS, Inc.
Ajay

Hi, I would follow what Thomas has indicated above. I would also check for any singularities
present. Now that you are talking of convergence and accuracy issues, a good account of these
is discussed in a report published at LSU ME department. The link to report is given below.... I
hope you find it useful... If you do have any questions about singularities send me a pm...
http://appl003.lsu.edu/mech/mechweb.nsf/$Content/MS+Technical+Reports/$file/ME_MS1_08.pd
f
Like Reply privately Flag as inappropriate August 30, 2011
Benjamin Franklin
Manager at Ashok Leyland

Benjamin

If your model consists of contact elements, chances of divergence would be more. In that case,
try to reduce the FKN value. If FKN value is decreased , consider increasing penetration
tolerance. Next step will be changing the pin ball region or other contact settings as described in
manuals. This link may help you to resolve your issue.
http://www.ansys.com/staticassets/ANSYS/staticassets/resourcelibrary/confpaper/2008-Int-

ANSYS In Cloud Scaling Up

Computational Capacity
Techila and MacGregor study how to
optimize the TCO of ANSYS simulation
with cloud and usage based licensing.
On-premises, hosted, Amazon, Azure,
or Google?
13m ago

Arash Pourbagheri commented on a

discussion in ANSYS Users Group.
best element type in nonlinear models
what do you think is the best solid
element type in a 3-d model
,containing all the three major types of
nonlinearities which are geometrical ,
contact and material nonlinearities ?
19m ago

Xavier Chauffleur commented on a

discussion in ANSYS Users Group.
Anthony V. How can I apply
different thermal loads on
different elements in APDL?
Hi everyone, I was planning

to use APDL to make a static thermal

analysis, and apply a different heat
generation load on each elements of
my model. By different I mean: different
value. I have read... more
1h ago

ANSYS-Conf-guidelines-contact-convergence.pdf
Like Reply privately Flag as inappropriate August 31, 2011
Venkat Anumula
Manager-CFD with automotive engineering background
Venkat

Thanks everyone. The links were highly useful. Please continue to comment if any. My model
has contact elements i.e. A meshed spur gear pair as shown in ANSYS Conference ppt. I use
contact tool prior running the solver and check the same after evaluation of results. I solved the
problem. Often these questions arise in my mind:
1. Does an interface treatment solve contact issues ?
2. Does changing advanced timesteps options would help in resolving contact problems ? Like
"Predict for impact or Use Impact constraints" Is it related to Time options in Analysis or is it
irrespective of the settings (be it a time or substep) I couldn't quite get the required info from
Help.
Initially i did a 2D analysis and got the results. Then i proceeded with 3D analysis for final
solution.
Like Reply privately Flag as inappropriate August 31, 2011
Alex Kagan
Principal Scientist at General Electric Global Research

Alex

Venkat, in addition to what Thomas and Benjamin have recomended, in my models I found weak
springs adjustment makes huge difference.
Like Reply privately Flag as inappropriate September 1, 2011
Venkat Anumula
Manager-CFD with automotive engineering background

Venkat

Alex, Do you have any best practice measure on adjusting weak spring stiffnesses ? or Do you
turn the Weak springs ON?
Like Reply privately Flag as inappropriate September 2, 2011
Venkat Anumula
Manager-CFD with automotive engineering background
Which one is better program controlled or explicit solver settings in most of the cases ?

Venkat

Like Reply privately Flag as inappropriate September 2, 2011

Alex Kagan
Principal Scientist at General Electric Global Research

Alex

Vencat, in WB analysis settings I set weak springs control to ON and change weak spring factor
to say 5 to start. In APDL, to obtain accurate solution, I kill weak springs elements at the last
load step. I do not know, how weak springs are handled in WB environment, it is question to
ANSYS developers.
Like Reply privately Flag as inappropriate September 2, 2011
George W. Weilbacher, PE
Analysis Group Manager, Compression Engineering Center of Excellence at Exterran
Vecat,

George
W.

I agree with all the comments above. Everyone makes a lot of good points. With respect to Ben
Franklin's point on the normal contact stiffness (FKN), sometimes it is the tangential stiffness
(FKT) that allows the modeled components to slip past one another. To Tom Hermann's point, if
you can get an unsolved solution you can look at the displacement results. To do this you may
need to let it run over a weekend or such. This will give you an ideal on where the rigid body
motion occurs and may give insight into what boudary condition changes you need to make.
As for weak springs in WB, the program will autogenerate some weak springs. These are not
always sufficient in their locations. The program may write them in the normal direction when it is
the tangential direction that is the problem.
To deal with these type of issues in WB I have created constraint equations and my own weak
spring definitions. The constraint equations require remote point definitions.
The springs I have written using the cursor in a point select mode.
When writing the springs, I typically use 5 or 10 pounds per inch. You may want to use

See all activity

About

Feedback

Privacy & Terms

LinkedIn Corp. 2014

something more or less depending on the physics of your particular model. I found in my own
experiences that it is best for these springs to be symmetric so that they do not create a moment
on the part before contact is detected. Thus, I typically create paris in a given direction.
To help detection, Tom makes the suggestion of increased load steps. I agree with this absolutly.
Not knowing the specifics of your problem, I have had to do this in the past myself. Start by
creating load steps with very low loads or displacements and then ramping them up in the
following load steps. Starting with these lower loads or displacements should help detection. Just
remember, however, you need to think through your complete problem as you cannot add or take
away boundary conditions in the following load steps. You can however take initial conditions as
zero and ramp them up or take specified conditions and take them to zero in the following load
steps.
Goodluck! and keep the group posted.
George
Like Reply privately Flag as inappropriate September 2, 2011
Venkat Anumula
Manager-CFD with automotive engineering background
George,
Venkat
Thank you for your detailed explanation. I would like to clarify the following from your statements:
1. Do you mean to say that even turning ON weak springs result in normal direction than
tangential direction or a program controlled weak spring locate in normal direction?
2. "You can however take initial conditions as zero and ramp them up or take specified conditions
and take them to zero in the following load steps. " - Could you please elaborate ?
Like Reply privately Flag as inappropriate September 2, 2011
George W. Weilbacher, PE
Analysis Group Manager, Compression Engineering Center of Excellence at Exterran
Venkat,
George
W.

It has been my experience that while WB will create contact elements where it decides based on
its programing, I may not want contact elements at some of the locations it has created them.
Likewise, I may desire contact elements at locations where WB did not create them. Under these
conditions, I typically create my own. In fact, I typically delete all the contact sets created by WB
and define my own.
Letting WB create its own weak springs, I have experienced the convergence issues you are
experiencing. This is not to say that your particular convergence issues are the same. More than
once, I have let a model run to its unconverged solution. Only then was I able to determine the
issue as the Solution output file and error file were of little use. For the uncoverged solutions, I
was able to plot displacement results. WB will not provide stresses for the unconverged solution or at least it did not in my models.
From the results provided I was able to discern that weak springs the program were creating were
insufficient to provide the contraint desired. The program was creating them in a direction normal
to the contact surfaces. In these particular models, my parts were sliding tangentially to the
contact surfaces.
To overcome this issue, I created my own springs in the normal and tangential direction. In the
next model constructed I tried to short cut the time spent defining the springs by creating only
one spring in each direction. After several hours of running, I checked the Solution output file and
could see I was no where close to solution. I stopped the model, create spring sets that would
prevent rotation and launched the analysis again. Within minutes it solved the first substep.
As for the load steps, if I have Node A and Node B and I want in the first load step to apply only
a load to Node A, I must still define a load on Node B in the first load step. This I do by assigning
a zero load. In the next load step, I may ramp this load as 0.1* Max. Then 0.25* Max, then 0.75*
Max and finally Max load.
In the gear problem you are attempting, you may reduce your initial load to 5 or 10 percent of
what you think it should be. In the following load steps increment it up until you achieve your final
desired load.

Best regards,
George

Like Reply privately Flag as inappropriate September 2, 2011

Venkat Anumula
Manager-CFD with automotive engineering background
George,
Venkat
It sounds interesting ! Your insights are very much helpful and could be treated as best practice
guideline. To be honest, none of the ANSYS Tech support engineers in my region were able to
give me proper reasoning. Hope you had great experience in resolving contact non linearities. I
ran a static structural non linear analysis as i've used plastic material models for gears and
metals (SS) for gear mounting shafts. It took almost a month to resolve the case on a 12 GB
Ram computer as i experienced the same. I couldn't explore beyond a given timeframe as i had
to deliver a client project with better results using an ANSYS evaluation license. I'm in a process
of procuring an FEA software license. Several iterations were carried out to narrow down the
issue with the use of solver output file. In case, if i need to run a transient state non linear
analysis (to be more realistic) i.e. i have moments applied at various time intervals (from
experiment or client's requirement), then i believe that ramping of loadsteps from lesser to final
max load is not required as the data itself comprises varied loads at varied timesteps. Any
suggestions in transient direction ?
Like Reply privately Flag as inappropriate September 3, 2011
Daniel Sadono
Attended Institut Teknologi Bandung (ITB)
Hello, i have the same problem about unconverged solution
Daniel
I'm about using stamping sheet metal by ANSYS Transient structural. The part consist of blank
Search
groups...

Profile
Connections
Jobs
Interests
holder, die, puncherHome
and
sheet metal
(then i assembled
it).
The problem is when i run the simulation, it always told about unconverged solution. let me
explain about what i already done.

Feedback

1. boundary condition, i already created first frictionless to make it simple one (ideal condition)
and it failed. i already tried frictional etc, but it simply didn't work. so i just back to the frictionless
condition to all. in the assembled part, between blank holder and puncher, there's a gap about 1
mm, puncher and sheet metal about 0.1 mm, blank to sheet metal is 0.1 mm. for die and sheet
metal there's no gap. for sheet blank holder, puncher to die, gap is 1.1 mm.
then i also tried the solution based on this site
http://www.padtinc.com/blog/the-focus/overcoming-convergence-difficulties-in-ansys-workbenchmechanical-part-i-using-newton-raphson-residual-information. and the result said the same error.
1. then i tried based on the Thomas's comment
the result is "internal solution magnitude limit was exceeded" and i searched it, it said about weak
spring. so i tried the solution based on those comments again and the error said "an internal
solution magnitude limit was exceeded. please check your environment for inappropriate load
values or insufficient supports. also check that your mesh has more than 1 element in at leas 2
direction if solid brick elements are present." the load that i used is about velocity 30 mm/s and
i'm just using 1 step with the end time 4 seconds
Could you like to give opinion / comment related to my problems ?
Thank's
Like Reply privately Flag as inappropriate 2 months ago
Predrag Pudar
Mechanical Engineer
Predrag

When using weak springs make sure that you insert force reaction probes to
confirm these are 0 or close to 0 (Total each direction)
If such is not the case you may have an inaccurate analysis
Also make sure that you enable Newton-Raphson residuals before starting the analysis
If analysis does not converge it will help you identify the culprit contacts
If you enforce zero penetration by using Normal Lagrange formulation chances of non
convergence/chatter are much higher
Augmented Lagrange will always converge better if you can live with a certain amount of
penetration
Check the penetration after the analysis and see if it is acceptable
If you use mid-surface shells even with checking shell thickness effect on chances are you are
going to have more trouble

Business Serv

Help Center

Like Reply privately Flag as inappropriate 2 months ago

Add a comment...
Send me an email for each new comment.

Add Comment

Suggested discussion
How can I apply different
thermal loads on different
elements in APDL?
Advertising Talent Solutions Small Business

About

Press

LinkedIn Corporation 2014

Blog

Careers

User Agreement

Privacy Policy

Community Guidelines

Cookie Policy

Mobile

Developers

Copyright Policy

Language
Send Feedback

Upgrade Your Account