FEA program opinions? 19

[MechElement]

Hello, I'm in the market for a license of SWx, but I would also like a good FEA program to accompany SWx. I've heard about NENastran & COSMOS running with SolidWorks, but it appears everyone has a different opinion about them. I'm a mechanical designer and working in the Middle East. I'm returning to the States for good in August to go back to school and finish my BSME.

Which FEA program works best with SWx?

 

[ekline]

mrmartins, I recently concluded an evaluation process for FEA and CFD software. The packages I evaluated for FEA included NEiNastran, Ansys/CFX, MSC/Nastran, Algor, Cosmosworks, Abaqus, Strand7, and Adina. Noran proposed MAYA's TMG Flow to go with NEiNastran, since both packages use FEMAP as a pre/post processor (NEiNastran also has it's own post processor, which is not bad at all and gives you essentially a free post processor for someone to use). Ansys has an excellent CFD package called CFX which has a lot of specialization for rotating systems. Abaqus has some partnership with Fluent which is also a very high end CFD package, but the total package is expensive. Algor and CosmosWorks have CFD capability, but I felt it was lower level, and the technical support did not impress me from either company. Anyhow, that list is how I eventually ranked the packages for our needs, based on capability and cost.

Noran was very responsive, and worked with me on price, found a good solution for the CFD requirement, and also worked out a payment plan that helped me get the purchase approved. The Ansys reseller (OhioCAE) was also very responsive, and had a superior CFD package, although I felt Ansys Workbench was a little weak (no beam elements) and the Ansys price break was still not a match for NEiNastran. I've used both MSC and NE/Nastran, both with a FEMAP front end, and I didn't really notice any difference between the two, other than better customer and technical support from Noran.

If you state your industry and types of problems you want to solve, you might get more detailed advice on these packages. You should definitely request a demo copy of the software to use, and have each vendor solve one of your problems.

 

[cbrn]

Hi everybody,

although there has been some flames here in, I read this thread with great interest. I was shocked by some opinions circulating about ANSYS, and I'll tell you why.

I personally did a lot of analyses with Cosmos/Works and Cosmos/M (for the non-linear sims), in the meantime trying to keep "up-to-date" with Ansys world. Now, in the company I work for, two FEA packages are treated as "reference" for all the FEA work (we also use in-house programs and a LOT of analytical spreadsheets). Both are considered to be "absolutely high-end".
These packages are ANSYS and ABAQUS. To be honest, Ansys is THE reference, and Abaqus is used by only one division mainly for "historical" reasons.

I personally do have the feeling of Ansys being a really high-end program, for a lot of reasons but mainly:

- you can not set up an analysis in Ansys without knowing exactly what you are doing: the program makes very little if not no assumptions at all, you MUST have complete control over the physicity of the problem you are trying to investigate: garbage in, garbage out... And it's not easy to give garbage to Ansys because your analysis would not run at all!!! I haven't been in these fora for a long time, but I think if you read Drej's (and others) posts in the Ansys forum you will have an idea of what I mean...

- you control every aspect of the program: an example has already been made in this thread and regards Contacts. Contact parameters in Ansys are the most complex and complete I've seen until now, Cosmos/Works is a toy in comparison...

- there is an ENORMOUS database of element types, each tuned for a particular application, and each with extremely strict and solid mathematical / theoretical basis underlying

- the calculated results have been benchmarked several times by my company against real prototypes, and the accuracy of the predictions always was extremely high: accurate at a point that the company has stated, in an internal prescription note, that for some critical components the FEA analysis made with Ansys can substitute entirely a prototype test: we know the error will be less than some 0.5% or so... Moreover, the same benchmarking made with Abaqus showed that, even Abaqus' results were in the same accuracy range, they tended to be a bit less conservative than Ansys' ones. So, being safety one of our major concerns, it's one more reason to rely on Ansys...

- although it's not "strong" in one particular field, Ansys is perhaps the most powerful general-purpose FEA I can figure out, nowadays. I'm dealing right now with a transient earthquake analysis of a complex structure (hydraulic turbine + generator assembly) full of intrinsic non-linearities (non-holonomic restraints, absorbers, etc...) and, though with a lot of difficulties (Drej knows something about it... ;-) ) I foresee that the analysis WILL be possible... I wonder how many FEA packages would give me the same impression...

So, definitely, you may consider me an Ansys "fanatic", but I love this program and I claim that it absolutely must be considered one of the high-end FEA kings...

Workbench is another thing (sigh...), but Ansys has claimed that the objective is to incorporate in WB all the Ansys Classical capabilities, so let's hope for the future... And, anyway, you can interoperate between WB and Classical.

Cheers to everybody!

 

[hansm2]

On this subject I would like to add my opinion. Please take this as just that, an opinion because I do not use Algor or Cosmos or Abaqus now though I did previously, I use Nastran. The variety of Nastran does not matter in this context because all the major nastrans handle dynamics loads, coordinate systems, symmetry, properties, etc. the same. IMHO, Nastran ís better because it is a well thought out program designed by 1000s of engineers and developers to handle aerospace applications 30 years ago. It is timeless and has evolved into one of the best general purpose FEA codes on the market. Much went into its planning and because of this it also handles automotive, medical, marine, etc. applications well. For example, how does Algor or Cosmos handle a pie slice model where the symmetry boundary is not orthogonal...they do it with stiff springs which is faster than a coordinate transformation but not as accurate and potentially dangerous. These codes do not output a true error measure as Nastran does. Without a true error measure there is no way of knowing if the results are accurate from the solution. Cosmos is at least a fast code for linear statics but I would not trust the results without either knowing what to expect or results from another FEA code to compare. There are many more examples but I spare them.

These codes are often collections of other programs developed by universities or internally from different disconnected groups and often slapped together with little understanding of how they will interact. Often the objective is just to have a check box in a product data sheet even though that feature may only be supported for one element type that no one uses. Having several Nastrans competing will only make it an even better product. You can say what you want about Algor and Cosmos, but there is no way you will convince me that these are superior when one considers all that went into nastran and still does every cycle. If the difference is only a few 1000 dollars then there is no reason not to get it. If the difference is 3-4x the cost then ask yourself what am I getting and is it a good investment and will it handle all the problems I may encounter in the future. Cost should be the last consideration for something so important as an FEA solver. Accuracy and robustness should be the first. I argue that these other codes lack this. Again, just my opinion if I can express this in this forum. I am not trying to offend anyone so if I did I appologize.

 

[GBor]

hansm2,

Many have expressed a like for Nastran and I am no different. I like the way the various Nastran codes handle different coordinate systems (something Algor recently included that Nastran has been doing for years) and the results are rarely questioned even though, as for all of these codes, they should be. But I must say that the global statements that you made regarding all Nastrans bring into question just how well you really know the Nastran market.

NENastran, as I understand it, generated their own code. They are proud of the developments that they have made and improvements over MSC Nastran (my particular favorite is 2-D plain stress). This is something that you've had to "fool" the MSC processor to do over the years. Since NENastran has emerged with this over the last decade (and I do believe you can get the right answers when this product is placed in the right hands), how could there have been 30 years of development? And some of the other codes that are named "Nastran" may not even have a true Nastran core? What exactly allows someone to call their product XXXXXXXXXX Nastran? Does it have to be based on the original NASA code? Cosmic Nastran? Or some more recent deviation such as NENastran?

If I develop a product from scratch, base it on some of the fundamental Newton-Raphson analytical methods and call it GENastran, will I immediately get the same high praise?

Admittedly Algor made some mistakes early on in their marketing...particularly in the Aerospace market, but they've been in business developing their software, including an in-house pre and post processor, for 25 years...I hope the learning curve starts to flatten out on the processors after a couple of decades, although I recognize the advancements made in computer technology, particularly in the area of graphics for pre and post processing. Incidentally, Algor has an AlgNastran product...is it Nastran? Does it get the benefit of your praise?

I would also say that NENastran and Algor are bitter competitors...NENastran, I don't think, is worried, nor should it be, about MSCNastran. In other words, it isn't competition between flavors of Nastran that are improving the product.

As for COSMOS...pretty cold blooded to shoot it down so readily. It's been used for a couple of decades in some pretty calculation intensive industries. I do think it is slowly falling behind and that the thing that will keep it alive is it's union with SolidWorks, but to suggest that a good analyst couldn't get the right answers from it for anything other than linear statics, and then that they better have another FE program result to back it up? OUCH! I've seen some pretty fancy non-linear analysis performed on COSMOS. I did duplicate it with NENastran and with Algor, not because I didn't trust it, but because the team with which I was working wanted substantial evidence to comfort our clients nerves about the project.

As for money sometimes driving decisions, not all of us have unlimited resources in our small businesses. We look for value (note: not price, but rather capability for the price). This is where, in my opinion, MSC has lost the edge. NENastran does at least as much, seemingly at least as well, and for a fraction of the price.

Don't mistake my wordiness for being offended...I REALLY like NENastran and Algor. Quite frankly, though, when it comes to "robustness" and "accuracy", you said nothing of Ansys or Abaqus. Your thoughts on these?

 

[hansm2]

Ansys and Abaqus are good codes but have their short comings as well. Abaqus requires a lot of training to use properly. Ansys can also be difficult to use and is expensive. Again, just my opinion.

ALG/Nastran is a simply Algor with a Nastran translator to convert the input deck to Algor's internal format and can not seriously be considered a Nastran. Their use of the name "Nastran" is simply a marketing ploy as they are very good at advertising and promoting their products. This is a fact and not my opinion.

If you like Algor and have some Nastran bulk data files to run it could be useful but should never be used as a replacement for Nastran. MSC, NX, and NE produce very similar results for complex models with lots of rigid elements, MPC equations, complex response loading, etc. All three have evolved and changed a lot from their initial offerings both in element formulation and solver technology.

I agree with your comments that NENastran is the best of all 3. Thanks for your response.

Hans

 

[GBor]

I'm sorry they erased the earlier response to your comments, but it was probably warranted...a little harsh.

Your comment about ALG/Nastran has to be your opinion because it is certainly NOT fact, although it is true that they created a Nastran bulk data file translator a few years ago (long before the discussions with NENastran regarding a teaming arrangement went REALLY Badly...now they are bitter rivals -- and I do mean BITTER!). The translator works pretty well, but I would want the ability to check the model in the Nastran pre-processor of choice before sending it to a client.

As for the evolution of MSC, NX, and NE: MSC has had time to evolve over the past 30 years. In my opinion, it hasn't evolved enough. NE, on the other end, has evolved rapidly, but it hasn't been in the market long enough for me to be comfortable with the rapid changes, but I don't use it everyday, so I'm not as familiar with its bugs and work-arounds.

I do agree with your comments about Ansys and Abaqus, but would suggest that ALL of these codes have their shortcomings...hence the need for a "toolbox"...not just a hammer! I actually think there is some good in being a little difficult to use...it means that not everyone will run out and design a bridge without having a clue what they are doing.

Anyway, I believe this comment closes out my thoughts on all of this...after 85 posts, any thread deserves a rest! Enjoy!

 

[dmacx]

GBor,

You seem to have a lot of interesting info on Algor. Do you or did you work there for a while? I was talking to a friend of mine that just dumped Algor and he says that the Nastran interface is just a simple translator to Algor. So I guess that confirms what Hans is saying. I do not blame Algor for coming up with this idea. MSC is doing it with Marc and UGS with Adina. It seems like a popular concept these days but I guess you never really know then what code is solving your problem.

 

[GBor]

I guess I'm violating my previously sworn silence...I'm just a long-time user that stays up to date. Never have and do not work there. Perhaps I'm just swallowing Algor propaganda on the Nastran product.