MSC, NX or NEiNastran 5

[ThomasH]

Hi all

I have what might be an odd question. Let's say that you were given a choice between the three different Nastran "flavours" mentioned.

There are no financial restraints, only choose the "best"?
Primarily for use in structural applications, linear and nonlinear as well as dynamics.

Are they "the same" or does it differ?

Regards

Thomas

 

[GBor]

Interesting question...I'm curious to see the responses. NX is owned by UGS. UGS also owns FEMAP, the pre- and post-processor for many FEA applications. UGS just announced a termination of their agreement with MSC to support MSC Nastran. NX, in turn, announced that they will pick up ANY MSC agreements at no charge. I don't know how long NE Nastran's OEM is open with UGS, but their pre- and post-processor is not fully developed outside of FEMAP. Noran's development pace is pretty fast, but if UGS cuts them off, I don't know what Noran's pre- and post-processing will look like. I suspect they have some screen shots on their website (

http://www.nenastran.com).

I haven't used NX, but it is generally the pre- and post-processor that people talk about. Most of the actual processor either produce acceptable results or parish. Those that like FEMAP for the interface may only have NX as an option in the future...

 

[fkmeyers]

I doubt UGS will cut off Noran any time soon. There were specific reasons why MSC was cut off (lack of support, no new FEMAP sales, etc) that is not the situation with Noran and it was certainly more of a mutual thing than is being made out to seem. Noran has assured us that their relationship with UGS remains very strong. As a user of both MSC and NEi Nastran (and by default NX since it is just MSC 2001) I can tell you NEi is the best for a lot of reasons. I have posted these already on this forum but to sum it up their surface contact is better than what MARC provides (basic MSC Nastran for Windows had no surface contact, just gaps and slide line which is very limited), their Editor is fantastic and has doubled my productivity, and their quad and hex elements are way more accurate. They have better composites support (more failure theories like NASA LaRC02, stability indexes for sandwich materials, etc.). Interpolation for CFD and heat transfer to structural models, etc. I would just go to their website as they have detailed most of this there. I am very biased against MSC and do not see NX as much of a step in the right direction. NEi just seems to be more innovative. Just my opinion for what it is worth. They obviously have a lot of companies switching over from MSC to NEi based on their customer lists so I would assume these companies did it for similar reasons…better product and a better value.

 

[jakeline6]

We have MSC and are looking at NX and NEi Nastran. Right now we are favoring NEi based on some initial testing and a side by side comparison with NX and MSC N4W. They seem to have everything we need and are offering an attractive deal to switch. Our MSCN4W licenses are expiring and we do not want to learn Patran (we know FEMAP and want to stick with what works for us). We had a recent meeting with MSC and they seem clueless as to how they will handle UGS pulling their FEMAP.

 

[ThomasH]

I started this tread as a result of the discontinued cooperation between MSC and UGS regarding MSC.Nastran for Windows (N4W). I didn't like the news but wasn't really concerned. My initial idea was to take a look at NEiNastran since so many good things have been said about it, for example in this forum.

I'm a beginner at NEiNastran, MSC.Nastran I have used for some 10 years but NE is new so there might be some setting to play with but so far I'm seriously disappointed. Primarily I have looked at two things accuracy and speed.

Accuracy first: Running the same file on MSC and NE typically the same results. Small differences, not all digits are the same (of cuorse) but the "result" is the same. One exception that I will get back to.

Speed: My first impression was that the solver was slow. Since there is a lot of testing on this at Norans site I decided to be a bit more systematic. Create model in Femap and export MSC file and NE file, even running MSC's file in NE solver and the results are always the same. NEiNastran is slower, sometimes much slower.
Then I read the "fineprint" in Norans testing. A lot of tests have been made with NE's version 8.3, todays version is 8.4 so 8.3 is probably one or two years old. The MSC versions used are v70.05 and v2002. v 2002 came in 2002 (or 2003) and I'm not sure about v70.05 but there was a 70.5 before v2001 that came in 2001 if I remember correctly.
What I tested was NEi ver 8.4 against MSC version 2005r2. That is as far as I know the most recent of both and NE came second, every time.

Now my evaluation version has a 300 node limit and to make performance (speed) tests based om 300 nodes can be argued to not be serious. On the other hand you can increase the load increments, timesteps or modes extracted and 3 seconds is faster than 5 seconds.

Now, I don't mean to bash och critizise NEiNastran. I've heard many good things about it so this came as a huge surprise and I could be doing something wrong. My original idea was to replace N4W with NEiNastran but now I will have to at least a look att NX Nastran.

I said I would get back to the accuracy issue. I tested something very simple. A flat plate thickness 10mm and 0.2mx0.2m with a load att the center and pinned edges. Model 1/4 and use 10x10 elements. Using symmetri and now apply the load 1000N at the corner of the model that is 0.1 x0.1 m2.
I ran it and got about 62 MPa stress. Now I say 50 MPa yield stress and make a nonlinear static analysis. In NE as well as MSC this means an decrease in deformation due to membrane effect added to the bending. In MSC I also get a maximun stress of 50 MPa and a yield zone, NE gives 59 MPa and a surprised user.

Any comments or ideas would be appreciated. Like I said I'm a bit surpriced and reserve the right to completely change my mind after further testing.

/Thomas

 

[GBor]

Material? (I'm just curious because I'm trying to run you simple model myself)

 

[ThomasH]

+++++

 

[ThomasH]

Ignore the post just before this one. I tested because right now I have a problem to post anything. Just typical.

The material is steel E=210 GPa or 210000000000 Pa. von mises ans isotropic hardening with H=1.0.

I was hoping for this reaction from somebody who knew NEiNastran.

Thanks

Thomas

 

[fkmeyers]

I do not see how you can do a serious performance comparison with a 300 node demo. Having both products I would say that for shell models MSC tends to be faster depending on the model. We have seen upwards of 2 times faster for MSC for linear statics in some cases. Not all models are faster in MSC as we have many that run faster in NEiNastran including shell models. For solid models NEi tends to be faster and in this case it was more like 5 times faster for NEiNastran. For nonlinear NEi seems to be faster but for dynamics MSC was faster but the differences were not that great. I think that overall for shell structures MSC will be faster and for solids NEi will be faster. The differences are not that great so it was not a big issue for us.

 

[NoranTech007]

Hi Thomas,

We ran your model in NEiNastran and think we know what the problem is. Use the EQUIVALENT STRESS and not the VON MISES STRESS. The difference between the two is mainly with material nonlinear analysis. NEiNastran computes the equivalent stress at the integration points and report this value as EQUIVALENT STRESS. This is what is tracked on a stress-strain curve and in your case should not exceed 50MPA. When we ran your model in V8.4 and plotted EQUIVALENT STRESS it was 50MPA peak (no surprises). But, the VON MISES was 60MPA and the reason is this measure is calculated using direct stresses. MSC reports the equivalent stress as von Mises. NEi does not and that can be confusing to an MSC user.

As for performance we noted this job took 4 seconds to run on a P4 PC. I am not sure what your time was but for these small models it is difficult to compare performance due to other factors that result in overhead which is constant regardless of model type (i.e., licensing, etc.).

We pride ourselves on the accuracy of our solver and suggest you visit our knowledge base for comparisons to MSC and other solvers. In all fairness I do not think that one example defines the accuracy of a product.

Noran Tech Support

 

[ThomasH]

You can't make serious performance tests based on 300 nodes, I agree to some extent. But I think you should compare relevant versions and I'm not sure Noran did that. It seems, from your (fkmeyers) tests that sometimes Noran is faster and sometimes MSC. Thats what I would expect. I don't think I have seen any test from Noran where MSC is faster (but let's not argue about that). Obviously before I make any final judgement I will test the full version. Lika I said, so far the result is just a surprise.

As for the yield problem. I understand what you are doing but if I give a yield stess of 50 MPa I want the von Mises plot to result i max 50 MPa, not 60 MPa. Maybe it's just me. And you can't define a problem with one example, I won't argue, but I tested five or ten with different yield conditions. It was always the same result comparing MSC to NE, that why I asked.

Thank you anyway.

Thomas

 

[GBor]

NoranTech007,

What is the analytical basis of "equivalent stress"? Is it more of a Maximum Principal Stress? If that is the case, then I could understand why Von Mises would report out higher than the yield. Is it possible to designate which failure criterion tracks with the stress-strain curve (so you can set Von Mises yield criterion as the one tracking)?

 

[ThomasH]

I read my own post again and relized that it could easily be missunderstood.

Concerning yield: What I mean by saying that "there is always the same result comparing MSC to NE" is that MSC always reports the yield stress in my input as maximum von mises stress. NE never does that, instead the basis for the yield is "equivalent stress".

I'm not sure I like the approach but that another story.

Thomas

 

[NoranTech007]

Some other differences which you may encounter later. MSC.Nastran does not use corner stress output for nonlinear results. NEiNastran takes plastic strain and equivalent stress at integration points and extrapolates these to corner values to better predict stress and strain with coarse meshes. For shells, layers are variable from 2 to what ever you want but the stress/strain values are taken at the center of each layer. An extrapolation is taken from the center of the 1st and last ply to the element surfaces which is what is reported, again to enhance accuracy.

With regard to using von mises as a result output useful for plasticity analysis there is a distinct difference between the yield definition for a material with any of the 4 available yield criteria including von mises and the formula for the calculation of von mises stress. I can assure you that MSC may call it von mises stress in their output in FEMAP but if you look in the .f06 file you will find that this is actually the equivalent stress and is labeled as such. The two are not the same but are very similar. You can email us and we will send you the 2 formulas and more on the theory. I will say this can be confusing but by strict definition you should be looking at equivalent stress which is the stress which corresponds to a uniaxial stress/strain test of the material being used and will match your stress/strain curve or plasticity modulus.

 

[ThomasH]

Hi

Now I have ordered the full 30-day version. It will be interesting to see if my initial surprise and conclusion regarding performance was wrong.

One question on this test. I oftet run MSC.Nastran in "batch-mode".
Basically: full_path\nastran filename1.dat scr=no

Can this be done with NEiNastran?

I've tried: full_path\nastran filename1.nas scr=no

And that didn't work. I've noticed that the default installation of NEiNastran includes spaces in the path. Could that be the problem? If so I will reinstall but since it is the default setting I wonder if NE isn't supposed to be run in batchmode. Or is there another way to do this?

Regards

Thomas