Abaqus, ANSYS, Algor, Mechanica which is the best? 8

[mechengr1313]

I am at a company that has Algor and Mechanica but believe we are approaching the limits of our current software. I have used ANSYS in the past, about 3 years ago and know it is a fine program. Does anyone here have any experience with both Abaqus and ANSYS and could give me a comparison for the two programs. We build large assemblies on the order of heavy equipment and we want to look at how the entire structure behaves in a twisting enviroment. I have modeled parts of assembly in Algor but am having problems now getting the mesh to go though the solver with the large assembly. I am exporting the geometry from Pro/e and was going to have Algor do a midplane mesh of the parts due to their thin structure shap but that has failed due to the mesher not being able to provide a continuous mesh without holes. What I am wondering is, who in your opinion has the best mesher for thin sections and which one works well with large meshes? In my opinion, a large mesh is around 2,000,000 elements. One last thing, should we even mess with Mechanica or get something else? Thanks for your advise in advance.

Mark

 

[corus]

I would have thought that 2 million elements is too big a model for any solver. You should be able to reduce the size of a model either by considering forces on sub assemblies or by using a coarse global mesh and sub-modelling areas of interest.
There can be no advantage in using thin shell elements as you have 6 dofs per node whilst a brick model has 3 dofs per node, but twice as many nodes. For that reason you might be better just meshing it using bricks, wiht one elemnt through the thickness. That would solve the pronlem of mid-plane meshing.
Look at other posts for a comparison of software. For linear problems they'll all do the job. For non-linear/contact analyses Abaqus is probably better.

corus

 

[mloew]

Corus,

With only one (linear) element through the thickness, it is not possible to pick up the proper stress distribution from bending. At least three nodes through the thickness are required to model the parabolic nature of bending stresses.

Also, ANSYS has recently solved a 100 Million + DOF model on a single computer. There is no practical limit anymore for model complexity WRT solving. Meshing is, of course, another issue.

Best regards,

Matthew Ian Loew


Please see FAQ731-376 for tips on how to make the best use of Eng-Tips Fora.

 

[corus]

Matthew, you're quite correct when looking at a single plate, but when you're looking at an assembly (say a box girder) then there will be negligble bending across any individual plate. The bending will be across the full section and as such a single element through the thickness would be good enough.

In addition, there is a problem with using shell elements along the mid-plane in that you are left with gaps between each plate. This can involve either time consuming stitching together of the parts or having to stretch one or more plates to make them join at coincident nodes. This gives rise to errors at the intersection with too much material. This is why I'd consider using brick elements first if I were importing a CAD drawing.

corus

 

[mloew]

Corus,

I am not referring to the bending displacement, but the stress profile through the section. Less than three nodes through the thickness will not pick up the correct stress pattern even though the displacements might be accurate! As far as the midplane shells, most FEA packages handle these offsets just fine. MECHANICA and ANSYS (packages I am familiar with) handle the connections (welds, fasteners, etc.) very well indeed.

Best regards,

Matthew Ian Loew


Please see FAQ731-376 for tips on how to make the best use of Eng-Tips Fora.

 

[mechengr1313]

I appreciate the responses you three have given. I am sure that the number of elements in the Algor model can be reduced. The problem is, as Jason has said, it is very time consuming. Algor tech support said I could reduce the mesh size if I were to build the model inside superdraw. If you ever used superdraw you could see why a guy that uses Pro/E to make assemblies can't see too much super in superdraw. One of the main aspects I have seen with ABAQUS and I believe ANSYS is you can independantly mesh the parts and "tie" the parts together instead of having to make sure the nodes line up. Do any of you know about this? Also, Jason, you mentioned looking at other programs, do you have any suggestions?


Mark

 

[brad]

Re: tying meshes...
I know that ABAQUS has the facility to tie disparate meshes together. What it amounts to is a special form of contact, in which the "slave" surface's nodes are tied to the master (thus the displacements are locked to the interpolated displacement of the master surface). Thus there are effectively no artificially-induced stiffnesses (such as would come out of RBE2 linkages between the parts).
At the time it came out, only ABAQUS had this capability.

ANSYS didn't have this, but my information may be dated (Matt, do you know?).

---

Regarding the original question, there is no "best" FEA software for all users. The answer is user-dependent and is a mixture of several important criteria:
1) Problems to be solved
2) Amount you are willing to spend
3) Constraints within your environment
a) what data format is provided to you
b) what data format is expected from you.

1 and 2 are generally pretty straightforward--the more complex #1 is, generally the higher #2 is. There are a few exceptions, but this is a general rule.

ALGOR is fairly capable of linear problems, but as the problems get more complex its functionality becomes limited. Likewise Mechanica. Both have some nonlinear capabilities, but nobody doing detailed rubber analysis would likely use either of these codes.

ALGOR, however, is inexpensive. Mechanica is nicely embedded into CAD software, thus it has its obvious strenghths.

ABAQUS and ANSYS are both very capable codes in linear analyses, and are also both very strong for nonlinear analyses. However they both cost significantly more than ALGOR.

In a sense this is like asking "which is better, a Cadillac CTS or a Chevrolet Malibu?". Most would agree that the Cadillac is a "more capable" car, but that doesn't mean that most would buy the Cadillac (since the price has a big impact).

Regarding my point #3--a very critical though often overlooked set of requirements is what form your initial data takes (CAD data that you make? CAD data from your company or anohter company? FEA meshes from somewhere else?). Likewise, what form are you expected to deliver--in automotive many Tier 1 companies are expected to provide runnable models for the OEM's. Implicit in this is the fact that it must be in a form which OEM's can run themselves (thus negating the possibility of "non-OEM" FE packages). These "constraints" can dramatically impact the decision-making process for purchasing an FE package.

---
Finally...
Always be wary of people who advocate one code as being absolutely the best for everything. They are generally either selling the code, or else they don't have a wide enough experience to be able to seriously evaluate the various codes.

Brad

 

[gurmeet2003]

mechengr 1313,

I have been using ANSYS with a Mechanical desktop front end. The models are built as solid models with in the CAD package and transferred to ANSYS through a software called a plugin inside a new interface called workbench. I have worked with assemblies haveing a large number of parts. The transfer of geometry has always been flawless. I beleive that ANSYS will handle a problem of the size indicatedd by you ( 2 million elements). Run time will depend on the computer hardware you use.

Also workbench creates contacts automatically. It does auto meshing. But sufficient controls on mesh are provided to tailor the mesh to your need. I will recommend that you look at ANSYS workbench software. It is very good for modelling assemblies.

Thanks,
Gurmeet

 

[mechengr1313]

Thanks for all your responses, I really appreciate it.


Mark

 

[onix]

I enjoyed reading this post. I was looking for a more comparative analysis of tools, and while corus indicates there might be some links out there, this was the best I found actually.

What I'm looking for specifically is this:

o A good mesher, perferably one that does mixed meshing of volume elements. I need this as I often have to mesh suspended (ribbon) membranes using brick elements on complicated compliant constraints which can be meshed with tets.

o An FEM tool which allows mixed mesh elements types in one model.

o A CAD/FEM package like CosmosWorks which seamlessly integrates design geometry variations with FEM.

o Multiphysics to handle piezoelectrics and mixed orthotropic material types in one FE model.

o Geometric nonlinear and material non-linear problems.

I have been leaning towards ABAQUS, but I am worried about not having a good mesher front-end, and losing the geometry parameter variational study within CosmosWorks.

Any suggestions, or any links which compare various tools such as ANSYS, ABAQUS, I-DEAS, UniGraphics, NASTRAN, etc. would be appreciated.

Thanks!

 

[mechengr1313]

Onix,

I looked at Abaqus and ANSYS and we bought Abaqus this week. I think they are both very good programs for what we are doing, ie. very large assemblies. The reason we went with Abaqus was, their salespeople are engineers and actually knew how to run the software and not just a preplaned demo. I would call them and ask them to take your problem and show you how they would solve it using Webx so you could watch. Just have them prove it will work before you buy it. I would make any company show you their program will do what you want before buying it. After you buy it, salespeople tend to forget why you bought their program in the first place. Check this out thread727-67468 interesting advice. Hope that helps a little.

Mark

 

[DRDearth]

Regarding the better code. I have current copies of every commercially available FEA code on today's market, Algor, Ansys, Cosmos, DesignStar, CosmosWorks, Mechanica (old Rasna)... They all work and will all give solutions sufficient accurate for one to make an engineering decision on where to go with one's design problems --- provided the program software is used properly. However, for our work here we generally prefer Nastran. It is a more comprehensive code in my view. Even when our clients request say Algor for use in their in-house work, we eventually end up having to provide copies (translations) of the final FEA idealizations in Nastran. In fact over the years I've seen a steady migration of companies that experimented with other FEA codes changing to Nastran. FYI.. this is not a ad for Nastran.... Just my comments.


David R. Dearth, P.E.
Applied Analysis & Technology

 

[Jordonlaw]

I tend to agree with the last comments regarding Nastran and would like to extend this to MSC products in general. Yes, they are expensive and maybe don't have the tie in with CAD to the degree of say Cosmos BUT the important thing should not be speed of meshing but accuracy of results. I've used Nastran for linear and Marc for non-linear and can recommend the latter as a competitor for Ansys, Abaqus et al.

I've now in the process of defining a non-linear package for my new employer and would totally agree with the comment to get the suppliers to prove their product on your product. I'm pitching Ansys against Marc - any comments for or against ?

 

[fkmeyers]

Since we all seem to be recommending our favorite FEA products I would like to put my vote in for NEiNastran. I have be able to run huge models over 2 million DOF in nonlinear analysis and have had great success with their support department. ABAQUS is a good FEA product but difficult to learn and very specialized and expensive. NASTRAN is universally recognized. MSC is expensive and I found the NEiNastran support to be more responsive and the product more capable than MSC's nastran.

 

[GBor]

I know this thread is somewhat outdated, now, but I'm new to Engineering Tips and am searching some of my favorite topics. I am an independent consultant that makes my living using various FEA packages and let me say that just because one of us hasn't done something, it doesn't mean it can't be done. You can mesh individual parts in Algor and the mesh will line up, or you can mesh the entire model and the mesh will line up...if you do it right. With that said...

Brad made some vary good points and, fortunately, recognized his own limitations in certain areas. I have used Algor, NEiNastran, COSMOS, and many specialty packages. Algor continues to enhance their user interface, but their engine is sound (including non-linear). "Bang for the buck" Algor is hard to beat. NEiNastran is good competition as is COSMOS. Each have their strengths.

I am generally leary of anyone that says "Stay clear of ..." This is a competitive industry. If something didn't work, and wasn't getting supported technically, the package would have long since been out of business. NEiNastran is based on the Cosmic Nastran kernel. A good package, but it is still maturing. Teaming with FEMAP was smart because that gives it CAD capabilities and the ability to translate to a variety of packages. COSMOS has a very sound engine. Now that it is owned by the same company that owns SolidWorks, we will likely see continued growth in the interface beyond the current COSMOSWorks that is available. Algor has remained fairly independent. The greatest advantage, perhaps, is that there is one person to contact for all your needs, and they are very responsive. As for the question about highly non-linear materials, all of these packages (which I would consider mid-grade) have Odgen strain models in addition to the general hyperelastic models. The analysis code is well-understood and well implemented.

For a model your size, the p-element convergence codes (like Mechanica) are generally very effective. They reduce the number of nodes (and, therefore, calculation overhead) by orders of magnitude.

The higher end codes, like Mechanica, Ansys, and Abaqus, are excellent if you need to afford them. They are great for fracture mechanics and specialty applications, which it doesn't sound like yours qualifies. As for limitations on any of these or the other packages, most of the limitation is in your hardware, not the software.

I'm all for gaining additional capability, but remember, I'm a consultant that has to interface with many different people using a variety of packages...if you have the tools, learn how to use them, don't just assume you need new ones.

 

[ParabolicTet]

Mechanica is great for its optimization and running hundreds of "what-if" types of analysis in less than a week. I use it to filter out concepts and try things completely uncoventional. Once I have a good design, then I will build a more detailed mesh in another software tool. Mechanica is not good for predicting absolute stress magnitudes near welds. BUt relative it's fine.

http://web.peoriadesignweb.com/calculator

- Online Engineering Spreadsheet Caclulators

 

[GBor]

Correct me if I'm wrong, but I always thought that Mechanica and other p-element convergence codes were good for complex brick models. With the polynomial convergence, you can use much larger bricks and still get high-quality resolution. Runs times are much faster and results are still acceptable...

Garland E. Borowski
Borowski Engineering & Analytical Services, Inc.

 

[johnhors]

Gbor

With "bricks" don't you mean tetrahedral elements?
Bricks are commonly used to refer to hexahedral shaped elements.

 

[mechengr1313]

You are correct in saying Mech uses less elements for a specific job but just like everything else there is a trade off. If the problem becomes more complex, the polynomial order of the elements increases. I have ran one model in mech and had it ran in 14 min, the same model in Abaqus ran in less than 2 min, both indicating the same overall stresses. The reason we went with Abaqus was, Abaqus has a "Tie" constrant that allows you to tie different surfaces together regardless of individual mesh densities so the nodes from one part don't have to line up with the nodes from the connecting part. Algor is a great program on its ease of use and straight forward approach. Mechanica works inside Pro/E so it has a great "pre possessor" All the programs have their strengths. We were just trying to connect over 300 parts together and Abaqus had a feature to make it easier for that application.

Mark

 

[ParabolicTet]

With Mehcanica you can use Single pass adapative (SPA) solution to converge much faster than the Multi-pass adaptive. SPA does not increase order of all elements in the model. ALso in mehcanica you can use a rigid connection to tie together different parts of the model. I am not sure if it works the same as abaqus tie, but it works like RBE2 element in Nastran i think.

http://web.peoriadesignweb.com/calculator

- Online Engineering Spreadsheet Caclulators