Best FEA software? LetÆs find out!

[bjpdwp]

This question seems to be asked quite often in this forum. The result is often a discussion of what each person prefers and/or has been exposed to resulting in no clear best FEA software. There are many reasons for this inconclusive outcome such as:
1) certain software are better at solving certain types of solution, so it depends on what type of analysis and how good of results are needed
2) continuos software upgrades changes their abilities quickly
3) lack of knowledge or experience of users resulting in an non-encompassing outcome
4) money

I propose a new method to answer the question. How about all that want to participate that we try an experiment and try a couple of different types of analysis. I have always wondered what would happen if a group of FE users, from beginners to experienced, solved a common everyday problem using different pre/post processors and solvers. What kind of results would we get? What techniques would people use to solve the problem? What would I and the rest of the group learn from each other to make us all better at FEA.

I think everyone would agree that when it comes to a straightforward linear analysis, all commercial FE programs are capable of getting good results. So, I would like to start with a more complex analysis such as contact, non-linear material, or vibration and find out which software does net the best results in each category.

For those interested, let’s start a discussion on what and how we should do this. What problem should we solve? How do we make this more interactive such as download/upload files to compare results? How do we correlate the results? What criteria do we need from each analysis, i.e. computer platform, pre/post processors, solvers, assumption…

Anyone game? Let's have some fun.

BJP

 

[Drej]

> This question seems to be asked quite often in this forum. The result is often a discussion of what each person prefers and/or has been exposed to resulting in no clear best FEA software. There are many reasons for this inconclusive outcome...

From what you've said it's an interesting issue, but I believe it's also a non-starter, which is why people's previous comments have been so fragmented or "inconclusive" as you've stated.


> There are many reasons for this inconclusive outcome such as:
1) certain software are better at solving certain types of solution, so it depends on what type of analysis and how good of results are needed
>2) continuos software upgrades changes their abilities quickly
>3) lack of knowledge or experience of users resulting in an non-encompassing outcome
>4) money

Agree with you to a certain extent on the fact that some codes are "better" than others at certain types of analyses, but "better suited" would be more appropriate I believe. I disagree with your point regarding how "good results are needed" however. Every FE software I know and have used (ABAQUS, ANSYS, NASTRAN, PAFEC, ELFEN, DIANA...) commercially sold will give you a mixture of exact solutions (closed form solutions) and approximate (open form) solutions dependent on the system being modelled and of course the experience of the analyst/engineer. The quality of the result at the end of the day is not down to the software *per se*, but with the experience of the engineer pushing the buttons and the information available to him/her; and it is this I believe that you cannot reasonably quantify under the challenge you offer. What you would probably achieve from this study is identify a group of "poor" analysts/engineers from a group of "good" analysts/engineers, and I use those poor/good terms very loosely.


> I propose a new method to answer the question. How about all that want to participate that we try an experiment and try a couple of different types of analysis. I have always wondered what would happen if a group of FE users, from beginners to experienced, solved a common everyday problem using different pre/post processors and solvers. What kind of results would we get? What techniques would people use to solve the problem? What would I and the rest of the group learn from each other to make us all better at FEA.

Not sure what you mean by a "common everyday problem", but as I mentioned above, in its closed form the results will be *exactly the same* for each code (usual disclaimer here on capability of the engineer/analyst pushing the buttons). All commercial codes worth their salt are benchmarked using this very same methodology. If you propose a series of closed form problems and then ask people to produce models of these, you would indeed be reproducing these benchmarks (which are invariably included with the codes anyway).

> I think everyone would agree that when it comes to a straightforward linear analysis, all commercial FE programs are capable of getting good results.

No. And yes. Assuming the person pushing the buttons knows what he/she is doing, as well as other disclaimers of software verification and having clearly definable problems to start with.

> So, I would like to start with a more complex analysis such as contact, non-linear material, or vibration and find out which software does net the best results in each category.

The complexities of doing this are enormous. You would just get a garbled mixture of results. Also, you mention "better" a lot, without clearly defining exactly what "better" or "best" means, which leaves it open to interpretation by the user of each software. What defines the "best" result? That which matches closest to someone elses model? Someone's PhD thesis? Experimental evidence/data? Of course, the latter, always. How about access to such data?

I should apologise for my cynicism but, for the reasons given above, I'll say again that this is a non-starter.

Cheers,

-- drej --

 

[corus]

I'd agree with Drej in that it'd be a mistake to see which program gave the best results.
In comparisons of household goods they usually have a set of categories for which each product is rated. If you want to compare FE products then this is probably the better way. Mostly the categories might be a matter of opinion though some could be more factual, such as does the FE code allow non-linear materials?
Here's a list of a few things which I'd consider:
1. Ease of use, ie. friendliness/GUI interface
2. Import of CAD geometry
3. Flexibility, ie. can the program be adapted to your specific needs
4. Capabilites, ie. Non-linear materials, contact, thermal radiation

etc...

corus

 

[bjpdwp]

drej and Corus,

Thanks for your comments. I am hoping to convert your pessimism and persuade you in helping me. I agree with some of your comments but disagree with others. Here goes.

True, incorrect solutions are often the result of bad methods employed by FE engineers. This is often a result of laziness, inexperience, and/or lack of mentoring. We won’t be able to change the former but I look at this “challenge” as an opportunity for experienced FE analysts such as yourself, Corus, and others to help boost the abilities of those that lack the experience and mentors. Also, show the capabilities of what truly can be achieved using FEA.

I have found and heard that solving more complex problems, software algorithm can and do have an effect on the results. There is a good possibility that each software will have different results. Some conservative and some not so. Most of the time it is not worth noting but if you don’t know what your solver’s capabilities are how would you know without testing? Even testing may not help. A simple case would be contact analysis. Mechanica uses a different contact algorithm than Abacus, and even MSC/Nastran used a different method of creating contact. All three will result in different answers but which is correct?

Common every day problems I had pictured are things like:
1) A bolt torqued to a proper value or better yet the proper clamp load. What are the results at the bolt head interface due to the contact. How many ways can this solution be solved. Axi-symetric, solid hex, solid tets, using gap elements, using contact algorithms… What will the different solvers predict? How do we correlate? Not sure.
2) Some type of modal analysis. Find a common part that can be purchased anywhere and can be easily modeled. Maybe a bell, stamped oil pan, metal serving plate … Something that can be easily tested by anyone that wants to test it for correlation purpose.
3) Take number two and do a vibration analysis and test using constant acceleration. Maybe do a transmissibility analysis and test.
4) Buy a hacksaw blade and put a strain gage on it, measure the deflection, do a non-linear large displacement analysis.
5) Use 4 and do a dynamic analysis.
6) Do something with non-linear material.
I am open to ideas.

As for bench marks, I’ve only seen those from the software companies who have a particular agenda and are usually small examples.

As for it being a huge task, let’s find a way to make it manageable. Maybe we all use a common meshed model, or develop a well-defined problem before we start.

Corus,

Is there a list of which solver is better for one type of solution such as the “household goods” example you suggested? This is really what drove me to post this thread. If I now that I will use FEA for linear analysis 60 % of the time, 20% for modal, 10 % for vibrations, 5% for contact, 2% for thermal and the remainder is for others I hope that I never have to do. Do I need to pay $70,000 for the high end software or will the $5,000 software work for me?

My list of preference would be different from yours only for the first two.
A good GUI is nice but I would rather put my money in the solver. I can get by with most pre-processors as long as I or a good CAD jockie has a good handle on the CAD software used to create the original geometry.

BJP

 

[GregLocock]

That's called a round-robin test in the UK. I've seen the one for experimental modal analysis - the results of which were quite educational. To be honest I think you are testing the user more than the software.



Cheers

Greg Locock

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

 

[gwolf]

bjpdwp,

I have used the following FE packages over 15 years:

ABAQUS,NASTRAN,SAMCEF,DYNA,PAFEC,LUSAS,BEASY,BERSAFE,FEMSOL,VISAGE,LRSTRAND,MARC,ANSYS,SC03.

ABAQUS is by far the best in every sense, don't bother messing about with anything else if you have a choice.

 

[johnhors]

Gwolf, those are all basically solvers (although some have limited pre and post capabilities bundled with them and that includes Abaqus/CAE!), what pre/post software have you used and what would you recommend?

Greg - "To be honest I think you are testing the user more than the software." couldn't agree more.

 

[gwolf]

Pre/Post Processors used:

FEMGEN,FEMVIEW,PATRAN,SC03,FEMAP,ANSYS

Well, it used to be said that the best combo was PATRAN/ABAQUS/FEMVIEW which I have used. I now use only PATRAN and ABAQUS when given a choice. The PATRAN to ABAQUS translator does not support all ABAQUS functions; but does generate most of the bulk data, especially contact.

ABAQUS/CAE is too weak - HKS entered the pre/post processor market very late and never really caught up - hence their eventual purchase by Dassault/CATIA.

I would say use PATRAN/ABAQUS still but keep an eye on ABAQUS/CATIA in the near future, it could be outstanding.

 

[apldmch]

What if we were to develop a method to identify the "best" automobile? We could define a course, say off-road, highway, city, etc., define different loads, like towing, hauling materials, hauling people, etc., and define different criteria, like cost, gas mileage, acceleration, emissions, styling, etc. We could then take several vehicles, ranging from small compact cars to big SUV's and put them through the paces to find the "best" automobile ...

Of course, there is no "right" answer, as there is no "right" answer to the choice of FE code. It all depends on what features are important to the user. I would therefore argue that rather than trying to solve a problem or group of problems, you should first define the criteria, e.g. speed, interface to CAD, meshing, linear analysis, non-linear analysis, maintenance and support, cost, etc. You would then evaluate the codes against these criteria (using a QFD-like process) and rank-order them against the criteria. That way, casual users or new users who are in the process of selecting an FE code can assign a weighting to the criteria and select the code(s) that best fits his needs. QED.

 

[rb1957]

as a follow-up to the previous post, can we kill this thread ? (please)

 

[Drej]

I second rb1957's suggestion on the grounds of not wanting to waste people's time and effort.

Cheers,

-- drej --


------------
See faq569-1083 for details on how to make best use of Eng-Tips.com

 

[ThomasH]

I have to agree with rb1957 and drej.

This kind of test suggested will, as Greg said, probably just test the users and not the software. If you look at how the major softwares are tested before the release of a new version you realize that to choose "a king" based on a few examples simply can't be done.

A good analyst can get good results even when not working with "the best" software. And a bad user will probably have unreliable results no matter how good the software is.

I hope I haven't offended anyone.

Regards

Thomas

 

[ramshree]

This question is like asking what is the best car out there? There are soo many variables such as cost to power ratio, overall technical support and on and on and on...

All these softwares have their own advantages and disadvantages...

And further more I do not think we can define a "typical user" first of all.

The needs for a bio-medical engineer might be totally different as compared to some one who is doing gear design!

I cant agree more with Thomas: A good analyst can get good results even when not working with "the best" software. And a bad user will probably have unreliable results no matter how good the software is.