Non-Linear Convergence

[Nick7805]

I'm trying to model a silo with a hopper (shell elements) resting against the ground (solid elements) using Inventor Nastran. In my current model, the ground has significantly lower stiffness than the hopper base and there is a separation contact defined between them. There is also various offset bonded contacts set between components of the silo, however the ground/base separation contact by-far produces the most contact elements.

I'm having a lot of trouble getting convergence, even at very low load levels (0.01% total load). I've attached a picture of the model below for reference.

The model will run and produce good results as long as the ground isn't in the model. Only when I add the ground to the model, I have issues getting convergence. So far I've tried the following:
- Reducing the number of contacts in the model (reducing the activation distance for base/ground contact)
- Modifying the convergence criteria to not include the load
- Modifying the convergence criteria to start at a low load level and increasing the number of increments
- Reducing contact stiffness to 0.5
- Changing the contact type from separation to separation with no sliding, as very little sliding is expected
- Testing as a bonded contact -- the model runs and produces results as expected if the contact is set to bonded

Typically the convergence issue is with load (and to a lesser extent, work); displacement always has converged so far. Depending on the settings, the solution either diverges or reaches the maximum number of bisections permitted. I'm fairly inexperienced using FEA and I'm not sure what to do in order to diagnose and fix the problem. Does anyone have any suggestions?

 

[Nick7805]

@rb1957
I don't disagree with your sentiments about Inventor Nastran. I think it's designed to be so user friendly that it may often lead to misuse among beginner to intermediate FEM analysts. It also has some major limitations to its usability since more advanced features are either hidden or non-existent, probably to make it more user friendly.

 

[ThomasH]

If you have lifting due the loading it can also cause an instability. The entire structure may be overturning even if I don't think that is the problem you have here .

I had a look in your other tread that I assume relates to the same project. Are the problems with convergence related to the problems you previously had with modelling the soft soil?
It also seems that you structure is fairly weak at the base. That is another thing that can cause convergence issues.

You also mentioned in a previous post that there were almost no change in the frequencies when you added the ground. Based on the other thread, doesn't that seem odd? If you reduce the support stiffness, shouldn't the frequencies drop?

I hope that you have have developed this model in steps so that you can return to a situation where the model worked and move forward from that point. Since you can't share the model it is difficult to do much more than guess regarding what is going on. But based on that you have problems starting at a very small load, my best quess is that there is an instability causing the problems. I would add a few strategically located soft springs to ensure stability. They don't have to be realistic, this is just to ensure stability.

The fairly high frequencies does not indicate a stability problem. But on the other hand, the analysis for modes probably doesn't include the contact boundary conditions.

Unfortunately, I currently have more questions and answers regarding your modelling approach .

Thomas

 

[Nick7805]

@ThomasH
I am also sure the structure isn't overturning. I don't really see why you say the structure is weak at the base nor why that would cause my convergence problems, considering the problems have been occurring at a load level of <1%.

Perhaps the frequencies should drop; but it's hard for me to say that given the ground is a solid mass instead of a thin shell element. It is also much smaller than the rest of the structure. And yes, contact is turned into bonded when you run modal analysis.

The model is developed and tested in steps; that's how I got this far with it. However, it worked in every single step up until this point with no errors and no singularities. It also works in non-linear analysis if the ground/base contact is set as bonded. This leads me think that it is simply the separation contact causing my problem. Maybe you're right, it's not stable... but I'm confused about why since I have contact stabalization turned on, and there's no instabilities in my model if the ground is no included. I will try add some soft springs if the current simulation I'm running fails -- so far it seems promising.

I have two thoughts:
1) The ground/base contact creates 70,000 contact segments... the large number of contacts could cause convergence issues. It is 60-70x more segments than created by all the other contacts combined.
2) Maybe I should implement the other loads (especially the wind load) in a 'subcase' after an initial subcase with only gravity -- maybe this could allow the contacts to activate and work properly?

 

[ThomasH]

I don't think that the main issue here is the solver. I worked with NEiNastran for several years and I have also used MSC.Nastran and currently use NX Nastran (or Simcenter Nastran to be up to date in the names ). My experience is that the difference are in the details. However, I don't know how Autodesk has developed NEiNastran since they bought it. When I got the information that their plan was to integrate it into Inventor I dropped it because I don't want that type of tool for the pre- and postprocessing.

But I have a question for Nick7805 and I don't mean to be rude, how much experience do you have in this type of analysis, non-linear FEM-analysis? You mentioned if the other thread that you work under the supervision of a senior engineer, how much experience does that person have with this type of analysis? I am just trying to understand what type of help you have from somebody with who you can share the model.

Thomas

 

[Nick7805]

@ThomasH
I agree with you, I didn't mean to imply it was a solver issue with what I said.
I don't have much experience and not very much help.

 

[ThomasH]

I was writing a post so I did not see your until now.

The reason I thing it is weak is because there is no bracings at the base. You mention contact stabilazing is on, what does that mean?

Regarding your thoughts:
All the contact elements will take time, I don't know if they will cause stability issues. But I thing there should be a parameter where you can set a search distance to reduce that number.

Regarding the loading, there you may have the problem. To apply self weight and wind in a single subcase, I would normally not do that. First the self weight (stabilizing) to get all the contact settings to work properly, then the wind (de-stabilizing) to start to test the stability. Also, to apply self weight and wind in a single subcase might produce incorrect results. In reality the self weigth comes first, and the wind after. And for a non-linear analysis superposition does not apply so the order of how you apply the loads will probably effect the results.

I would not be very surprised if that approch solves the problem .

Thomas

 

[Nick7805]

@ThomasH

About contact stabalization:
"Surface contact solution stabilization option. When set to ON, will generate stabilization spring stiffness via the model parameters NLKDIAGSET, NLKDIAGAFACT, and NLKDIAGMINAFACT on the contact boundary."

Yes I agree! I hope it will solve my problem. Thank you for your input .

 

[ThomasH]

@Nick7805
I appreciate you honesty about you inexperience. You situation is unfortunately not unique. You probably needed help "in-house" that were not there.

Regarding the contact stabilizing, I honestly can't remember those parameters from when I worked with NEiNastran so I can't comment on that.

But hopefully the change in loading approach will solve the problems you have. If not, I will try to keep an eye on the tread.

On a more cheerfull matter, Have a really nice Christmas and try not to think to much about contact stabilizing.

Thomas