constraints application

[selopez]

xyconstraints2.jpeg (95.16 KiB) 6840 mal betrachtet

Some weeks ago we discussed about the application of constraints. As it was explained (or at less as I understood it), all directions must be constrained to avoid the structure to be statically undetermined. No necessarily applying the three directions constraints to the same nodes, but accomplishing the condition of that all degrees of freedom were locked. Even, as I observed in other studies, the constraints on each direction can be in different areas of the model.

At the example of the image, only X and Y constraints were applied at the base and the cap of the barrel. There's no Z constraint at any other place. At the same parts (the base and the cap) was applied a pressure. Also other test was made applying forces instead of pressure (Z positive at the base and Z negative at the cap), with analogous outputs. I find that the results are coherent, both in stress and deformation, but... not sure if there's not something conceptually wrong, or if this case is enough representative. This last because, actually, the loads are applied in Z direction, that is the missing constraint.

Notes: Pardiso couldn't end the task (I think that not because memory or system limitations). SORCG did it in a few seconds. The mesh is T16. At the image, the magnification of the deflection is 10x.

[selopez]

After posting the previous comments I did another essay with no constraints at all. Just pressure at the planar faces of the barrel. The FEA results (through SORCG) also seemed to be consistent. Then I tried to apply the non constraints concept to other piece, considering that the applied system of loads would provide the needed statica determination, but... (as always) the thing didn't work.

At this point I need to say that I'm not "playing" with the idea of eliminate constraints. For the most of the parts I design, and for which a reliable FEA is very important, there's not absolute blockages at any place. It always is the case of contact loads that can be displaced along its application. Certainly I don't pretend a full realistic representation of these cases, but I think that some kind of approach to them would be good.

[wehmann]

Dear selopez,

to ensure the results being correct, a statically determined model should be used. You can for example lock the middle plane between cap and base in z-direction in order to get a fully statically determined model.
If your model isn't statically determined, the direct solvers (pardiso or cholesky) will fail, because they are calculating exactly (and the system of equations is singular/not solvable). Depending on the model characteristics, the iterative solvers (siccg or sorcg) are able to find (correct) solutions for statically undetermined models, but you can't trust them in general, if the model is statically undetermined.

Best regards,

Christoph Wehmann

[selopez]

Prismas05.jpeg (213.06 KiB) 6703 mal betrachtet

Estimated Mr. Christoph Wehmann

Thank you very much for your answer, helpful as usual. But, in this case, I beg you to let me continue the discussion.

See please the attached images. Naturally all them show the same model, of the same material, mesh, etc. The case A is constrained in all directions at the cap and the base. An equal pressure is applied at the centre of each face of the triangular prism (through a circular pattern). In the case B the system of loads is the same but, there're no constraints at all.

The stress results shown by both FEAs are practically the same, which suggests that the lack of constraints at B wasn't an impediment to achieving a good result (even some unjustified stress spots can be seen at case A). However the deformations show an important difference. If the real case is that of a piece loaded "on air", I mean that its position is determined only by the applied loads (which is the normal for the products with I work), the case B would be much more realistic than the A. As it can be seen in case B, deformations in all directions were produced, what is what to be expected (I imagine that Poisson would be proud of AURORA ). On the other hand, I can't guess how the same deformation results could be obtained applying constraints.

In my opinion, Static Determination is more a matter of forces composition, than an issue of locking all Cartesian axes. Actually, a locked position is the geometrical resultant of a neutral composition of forces. As I see them, position constraints are a kind of dangerous wildcards that must be used with care. I accept that there're cases in which they can be real structural conditions, but there're other ones in which positional locking can produce wrong stresses and deformations.

Concerning the solvers (thanks again for your explanations), and if I understood well, it doesn't bring me peace of mind that the solver "correct" the missing of constraints. It would be better to hear that if it finds a coherent system of loads, it will perform its task normally, that is what I believe that is actually occurring in the cases that I exposed.

I imagine that in these days, all of you are overwhelmed with the launching of AURORA V2. When you consider it appropriated, I would like to expose you another case (a real piece) in where there's not a real place for constraints and in which the no positional blockage concept doesn't work.

Best regards, and thanks for your time.

[Tim]

Hi Selopez.
Just some comments re your findings. I'm not 'support', but just another occasional user.
I agree with you, constraints should be considered very carefully. It takes a thorough understanding of applied mechanics to get accurate reliable results. Just considering 'Beam' problems for a moment ( ha ha ), cases like: 'simple supports' , 'Fixed end / ends','uniform loads' etc. are all significantly different in calculation formulea to determine stress,delection etc.
I met a principle engineer from Leitz Metrology ( 3D CNC coordinate measuring machines) a couple of years ago who told me that when they started using FEA extensively, their engineers on the same problem were initially getting very different results!
A fully constrained face is a very unlikely condition in the real world and its application i would
be concerned with - quick and easy to 'click apply' though!
Yes, i too think ' Support' is busy on V2. It must take alot of effort.
Looking forward to Z88 evolving over time. I cant afford ' commercial FEA' licences.
Cheers.

Tim.

[mz15]

Hello

I want to clarify the fact of the missing of contraints again clear. For a clear and reliable calculations are always required certain static determined models, even if iterative solvers provide a result. Only Forces can be lead to a result, even a technically reproducible and logical, but Static determination is NOT a matter of force composition. It is wrong that you just take a clever composition of forces, the model must be determined and always static, both models above do not. You can use the virtual fixture to get a static determined model. Especially in the FEA, it is the challenge of the correct boundary conditions non static determined model, is not a smart apply of boundary condition but just plain wrong. That is an mathematical fact, because the system of equations is singular and so not solvable.
It is NOT so that the iterative solver "correct" the missing of constraints, the iterative solver are only able to find a solutions for statically undetermined models,this is something completely different, because there are infinitely many solutions.
If the system of equations can not be successfully calculated using a direct solver, than the somthing is wrong. Even the most expensive FEA programs will produce a result which is just as right and wrong as Z88Aurora supplies. This experience, I have done.

Best regards,
Markus Zimmermann

[selopez]

Mr. Zimmermann,

I don't understand why you say that both models aren't statically determined, while the A case is constrained in all direction in the two triangular faces, as it's explained in the description of the cases.

Anyway, you can be sure that I know well that for structural calculation the static determination is something essential, and that within this, FEA has its own rules. What I'm trying to find with my observations, is help about how to establish the more accurate and realistic boundary conditions possible. I'm not trying to demonstrate any new theory.

The exposed case is for me a paradigm of several problems with which I've crossed making FEA of real pieces. For example, these days I need to resolve a part that's basically an open ring to be pushed through a conical hole (say a kind of a specialized Seeger ring). In fact it's a redesign of an existing piece and, for experience, I know that it will close itself along the cone maintaining its circular shape and distributing its stress homogeneously within its elastic limit. I wonder then where should I allocate the constraints to obtain a realistic FEA of this so simple case. The logic suggests that that constraint would be at the centre of the ring, that doesn't physically exist. I tried to apply them at several zones of the existing surfaces, but always with not realistic deflections and stresses. Cases like this moved me to guess that a virtual constraining, based on load balance, can be the solution. That's all.

On the other hand, I'd like to point that wasn't me that said the iterative solvers "correct" the missing of constraints.

Regards and thank you very much for your concern.

[Tim]

I will have to acquire a good book for Application of Force / Pressure and Boundary Constraints for successful FEA modeling. ANSYs workbench has many typical examples - edit. ANYS Workbench 13 - well, maybe not. it's
obviously written for that software product.

In my mind, it is not for 'support' to teach FEA application but only to answer questions on Z88 - if they did that it would be a very frustrating experiance for them and i don't think they have the time either. Also, of course, the best explanation would be in their native German, so easy for language interptation errors to creep in.

Youtube has many FEA videos - but personally, and immediately now, i'm buying a good 'worked examples' book.

If anyone can suggest a good introduction starting book on 'Application of Force / Pressure and Boundary Constraints for successful FEA modeling' i'd be gratefull.This book has good reviews 'Building Better Products with Finite Element Analysis' by Vince Adams and Abraham Askenazi (1998) . At 14 years old, a copy is
expensive at approx £70!
This is the 'level' i'm at - can anyone else suggest another similar book but at a lower cost?

I'm convinced Z88 Aurora Team are doing their very best.

Looking forward to V2 alot!

Tim.

[selopez]

Tim:

I also appreciate very much the Aurora Team effort, mostly for putting in our hands a so professional FEA software. Neither is my intention to be teached here about BC application. But... what's a forum if not a place for share experiences, for make questions, receive answers, or even for make observations beyond any dogmatism? All this through serious, kindly and respectful interventions. On the other hand, all what I exposed was made within AURORA and with its tools, so I believe they have been pertinent to this forum. Anyway, thank you very much for your interest in this discussion.

[Tim]

Well, yes a forum is for exchange of ideas / suggestions / help on the product or just plain HELP, my
pants are on fire! etc.
I'm the type of person that will contribute /help if i can. So until we chat again, i've some
reading up to do. Please see my edited previous post. Any other good book suggestions?

Cheers.

Tim.

[mz15]

Hello,

sorry if my post has come negative, but I just wanted to clarify that calculation with static undetermined models do not yield reliable results. For a secure and reliable computing a static determined model is necessary.

I don't understand why you say that both models aren't statically determined, while the A case is constrained in all direction in the two triangular faces, as it's explained in the description of the cases.

By that I mean your modell in the first post und case B.

Best regards,
Markus Zimmermann

[selopez]

Good day Mr. Zimmermann

Being the case "A" the right and orthodox resolution, it would be of professional rigor to asking ourselves why the stress output of the case "B", the wrong solution, is exactly the same, and being achieved after the same number of iterations. It's to regret that in this change of opinions we cannot go beyond the mere disqualification of a solution, just because static determination is not achieved through positional constraints. Limitation this last that also prevent from improve the management of BCs when a better approach to real deformation is needed. But don't worry, I guess that these days you are too busy and, in a way, this one is becoming an abstract discussion, as its subject is not the importance of the static determination (that no one disputes), but how to resolve certain practical cases.

Best regards.

[selopez]

Estimated Tim

Sorry, I don't know the book you mentioned, nor any other that I can recommend. If you get it, I would be very interested in know your opinion about it.

Regards

[auroraIco]

Hello,

at the moment we have plans to translate our book "FEA for engineers" into English. Discussions with the publishing house are in progress. Within this book there are many examples for right solver choice, iteration criteria, convergence and so on. We will keep you informed...

Greetings, aurora_ico

[Tim]

Auroro_icd
Well, thats promising! - You could be about to sell ALOT of books!
But, the timeframe for 'FEA for Engineers' book could be ... a month ( ?) or so, i guess it depends on the
publishing technology used.
Perhaps as a possibility, the book publishing could coincide with V2 release.
Can you confirm or indicate a time frame for publication.

Thank you Z88 aurora team.

Regards.

Tim.

Selopez,
Well, i made the wrong decision and bought another application book:
Applied Finite Element Analysis 2e G. Ramamurty for £ 17 from ebay.

Well, why wrong? Because,i just couldn't spend £70 on the aforementioned book as it was a well used secondhand copy all the way from the US -thats a poor decision as it's a rated book!

However, with the announcement of the book 'FEA for Engineers' by the Aurura team as above- thats got to be very good news and the ultimate information resource for FEA + Z88 Aurora usage or so i hope .

If you can't wait... then, i have done a little further leg work and this has come up:

'Practical Stress Analysis with Finite Elements (2nd Edition) (Paperback)
by Bryan J Mac Donald (Author)' from Amazon. First pub 2007, second 2011.Good reviews for the novice FEA beginner - undergraduate / post grad level. Cost ~£ 30 upwards. Looks promising in my view - but i admit to knowing less and less nowadays

Edit: Bought the above as well- i have to invest in knowledge. I will let you know my opinion after i've chewed
and digested it.
Regards.
Tim.