Analysis just stops ...

[346gnu]

Hi,

I've had this before and have never got to the bottom of it, your input greatfully received.

Occassionally, we have a model with a large number of elements >200k, usually an assembly with several 10's possibly 100-200 contacts.

Using SPA, the model meshes fine and gets to Pass 1, shows that the max edge order is 3 in the summary file and then the msengine process vanishes. No errors or warnings. This is consistent.

Our unscientific experience is that things seem 'twitchy' when going beyond 200k element figure and seems associated with the presence of contacts where the mesh is refined on one side of the contact pair. (previous post re-disappearing contacts)

Thanks

Further background ...

In this case we have an assembly and we are interested in getting good, converged stresses from one component within the assembly for the purposes of a fatigue study.

The assembly meshes fine and runs fine with no mesh refinements, about 60k elements and 3 hours, 101 contacts, 53 springs, 6 rigid connections, 2 free connections.

Apply mesh refinements to the component to be studied - 260k elements - msengine dies. (we have large amounts of hard drive space and a bucket of RAM and it's unlikely to be a permissions related issue)

It is not possible to carry out a single component analysis as the there is no sensible way to determine the force system to input due to the way components move against each other. Quite simply, single component analysis answers would be wrong.

One work around is to 'wander around' the model with mesh refinements, sequential analyses where modest areas of the component being studied are refined. This is the route being taken. Each run being about 4 hours. This becomes a administration rather than an engineering exercise, down stream time and effort increases etc. as does potential for error.

The approach in other s/w would be submodelling.

This thread is inactive and closed by the PTC Community Management Team. If you would like to provide a reply and re-open this thread, please notify the moderator and reference the thread. You may also use "Start a topic" button to ask a new question. Please be sure to include what version of the PTC product you are using so another community member knowledgeable about your version may be able to assist.

[danderson]

Charles,

When you mesh the assembly do you get any Autogem warnings?

I have run into it where I have a round that blends to a sharp point. Bad transition at the end of the round. When I crank down on the mesh refindment in that area Autogem has trouble interpeting the geomerty. Looking at the autogem status dialog, geom checks or stopping the analysis run just before msengine fails it will sometimes point to the suspect geometry.

Hope this help,
Don A

[346gnu]

Hi Don,

Thanks,

I modelled over troublesome geometry.

Only messages are those warning me I have parts with no material properties.

Contacing line search has finished and the element calculations have begun.

[TadDoxsee]

Hi Charles,

The msengine process should never vanish without first wrting an error or warning message to the summary file. Would you please file an SPR with technical support so we can investigate this problem?

Thanks,

Tad Doxsee

PTC

[346gnu]

Hi Tad,

I will file an SPR though the content has to be nebulous as I cannot release any models.

I know TS require something a bit more deterministic; hence my initial question at the forum.

Thanks

[346gnu]

Tad,

The call number is ... 11478679.

I will update wth as much info as possible.

Regards

[TadDoxsee]

Hi Charles,

I'm afraid that without a specific model that reproduces the problem, we probably won't be able to isolate it. If the problem reproduces for you on a model that you can share with us, that will help us tremendously. PTC has special procedures for handling NDA and ITAR models, if that makes any difference.

Tad Doxsee

PTC

[346gnu]

I know.

This would be ITAR and the hoops took time to jump through.

and then I thin there are export licensing hoops.

I shall investigate.

Anything that is problematic that I can share with yourselves I will.

Thanks

[JonathanHodgson]

Hi Charles,

I've certainly had instances of suddenly realising that msengine.exe just isn't running any more, with no error messages.

I think that these would have been in linear analyses, although probably ones with fairly complex geometry and hence mesh. I would suggest that it may not be down to the presence of contacts, beyond their contibution to general complexity.

Nothing more concrete, I'm afraid.

[integrisbrad]

I agree with Jonathan, and I remember them being linear analyses as well, solid elements using SPA convergence. I think it can be a hardware or config issue. More than once, I received a support call with this issue, had them share their model with me, and it ran on my machine to completion.

Don't remember that we ever isolated a cause for the behavior, but they were large models, they may or may not have been using inertia relief, and they were using a significant number of load cases (for subsequent scaling and superposition) and many weighted links.

[346gnu]

Yes, I kind of agree with you both, nothing concrete to nail down. The usual process of elimination technique is not an option as it takes 40+mins to fall over.

We only seem to encounter this issue when much above 150k elements and so have a general rule of keeping below this 'glass ceiling' if possible.

We have batch run large models before (making use of the long bank holiday weekends), each run being for a different load case. Some have run and others have not.

Associating it with contacts could of course be a red herring. I cannot fully dissocciate them from the problem though as these bigger models tend to have contacts.

For this exercise, I nudged the element count up to 150k, this ran. 163k elements fell over. Again, not methodically applied mesh refinements for the purposes of debugging.

In this particular instance we have some imported geometry, but earlier projects did not ...

I will try the model that fell over on another machine (no other changes) over the w/e to see what happens.

[SylvainA.]

Could it be a RAM size and/or SOLRAM setting issue ?