CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > FLUENT

Welcome!

Register Blogs Members List Search Today's Posts Mark Forums Read

Reply
 
LinkBack Thread Tools Display Modes
Old   November 22, 1999, 03:14
Default Welcome!
  #1
Jonas Larsson
Guest
 
Posts: n/a
Welcome to this brand new Fluent User Forum. The purpose of this forum is to offer Fluent users worldwide an undisturbed place to exchange ideas and discuss common problems.

All topics related to software packages sold by Fluent Inc. are welcome. More general CFD questions still belong in the main forum. I hope that we can get some discussions going here.

Fluent Inc. are aware of this new forum and will probably keep an eye on it also. However, the Forum is independant and not in any way supported by Fluent.
  Reply With Quote

Old   November 22, 1999, 10:02
Default Re: Welcome!
  #2
reyman
Guest
 
Posts: n/a
Jonas, I feel very, very fortunate to have these forums as CFD resources and I wish to thank you for your hard work and dedication. The short time I've spent learning CFD- and its many components- has been enriched by CFD-Online and I know that my understanding of the details is more complete because of the forum. Thank you.

And on to the questions and comments...

I spend most of my CFD time in Gambit building meshes for air sampling devices. Some of the difficulties I've encountered so far are as follows: (1) forgetting to make sure I have plenty of disk space available for large meshes because Gambit will crash during a meshing or saving operation if the disk is nearly full; (2) failing to get an *even* number of nodes on edges when I desire tetrahedrals to grow from quad surface meshes; and (3) the tendency to mesh more than I really need to.

I imagine these three are items that other Gambit beginners have experienced also. I practice almost everyday, but since there are only beginner Gambit users in our facility, its hard to know what's *good* versus what's *acceptable* and we can only speculate -with guidance from technical support at Fluent- if we are doing the right thing. It will take more practice and experience as well as input from other experts, and thus I'm encouraged by the creation of this Fluent forum and I look forward to reading comments and sharing ideas.

Thanks!
  Reply With Quote

Old   November 22, 1999, 23:12
Default Re: Welcome!
  #3
Scott Gilmore
Guest
 
Posts: n/a
reyman@lycosmail.com writes about Gambit:
> Some of the difficulties I've encountered so far
> are as follows:
> (1) forgetting to make sure I have plenty of disk space
> available for large meshes because Gambit will crash
> during a meshing or saving operation if the disk is
> nearly full;

We have definitely fixed the problem with saving to a full disk. The fix will be available in Gambit 1.2, due very soon. Your local support engineer(s) will be glad to let you know when Gambit 1.2 is available for download if you will please let them know you are awaiting this fix.

I haven't personally seen crashes while meshing with a full disk, but we'll check it out. Thanks for mentioning it. Please don't hesitate to report problems to your local support engineers(s) as soon as you encounter them.

> (2) failing to get an *even* number of nodes on edges
> when I desire tetrahedrals to grow from quad surface
> meshes

As far as I know, it is mathematically impossible to create an all-quad mesh with an odd number of intervals on the outer edge loop. If this is too inconvenient for you, you may want to consider a quad-dominant mesh (mostly quads, with a few triangles), or even an all-triangle mesh. Many users are achieving good solutions with triangle surface meshes and prism (wedge) boundary layers.

> (3) the tendency to mesh more than I really need to.

I don't understand what you mean here. Perhaps you could elaborate, or contact your local Fluent office for advice.

Good luck!
  Reply With Quote

Old   November 23, 1999, 05:24
Default Re: Welcome!
  #4
Jonas Larsson
Guest
 
Posts: n/a
Thanks for those nice words about CFD Online. I'm glad that you like it.

About Gambit. I also use Gambit now and then. I recognize some of your comments. Gambit can be a wonderful product. I used it to mesh a very complex 2D domain recently (9 cavities+solids inside the high-pressure region of a jet-engine). It worked perfectly and saved me a lot of time. However, for more complex 3D cases Gambit is not very useful for us ... we have this saying here that "someone gets the Gambit disease" - the symptoms are that after having spent 14 days struggling to mesh a complex 3D case in Gambit you come to the coffee table talking about "if I just could get this last face/boundary layer to mesh then I'll be finished, hope to have it working within a few hours". Two weeks later you come back again and you still say the same thing - you've got the Gambit disease. One of my colleagues compared it with a woman who can't leave her abusive husband - You dream about how wonderful it could be, but deep inside you know that the abuse is just going to go on and on. Still you can't do anything about it.

  Reply With Quote

Old   November 23, 1999, 05:30
Default Re: Welcome!
  #5
Jonas Larsson
Guest
 
Posts: n/a
About the 1.2 release. When do you expect to have it finished and available to your customers?
  Reply With Quote

Old   November 23, 1999, 11:50
Default Re: Welcome!
  #6
Althea
Guest
 
Posts: n/a
That's it exactly Jonas,

(I mean about Gambit - I thankfully have no direct experience of abusive husbands)

All the time your friends, family and colleagues think you're mad and that there must be a better alternative, but you are always afraid the alternative will be just as bad and require all that time on the learning curve again.

Thanks for the new forum. Sometimes I have been reluctant to send postings to the general forum because they are Fluent specific.

All the best

Althea
  Reply With Quote

Old   November 23, 1999, 12:45
Default Re: Welcome!
  #7
jurek
Guest
 
Posts: n/a
There are two main problems in GAMBIT: 1.) You have big geometry, which isn't closed at one corner. And now you try and try and try ... 2.) The boundary-layers are a good idea, but especially in complicated geometries it does not work. I think, in this cases the whole idea (selecting face, and there it is) of those boundary layers is not good. There should be something like in I-DEAS, where you can set each node by hand and click the element after this. Perhaps this sounds archaic, but so you always (!!!) get a mesh even in very very complicated corners. jurek
  Reply With Quote

Old   November 23, 1999, 17:11
Default Re: Welcome!
  #8
Scott Gilmore
Guest
 
Posts: n/a
Jonas writes:
> About the 1.2 release. When do you expect to have
> it finished and available to your customers?

Gambit 1.2 will probably be available for download beginning next week.
  Reply With Quote

Old   November 23, 1999, 17:14
Default Re: Welcome!
  #9
Scott Gilmore
Guest
 
Posts: n/a
Jonas, I'm sorry you've been so frustrated by Gambit. I think you'll agree, however, that each release has been a great improvement. I'm confident you'll find Gambit 1.2 to be another big step and a reliable tool.
  Reply With Quote

Old   November 23, 1999, 17:26
Default Re: Welcome!
  #10
Scott Gilmore
Guest
 
Posts: n/a
"jurek" writes:
> The boundary-layers are a good idea, but especially in
> complicated geometries it does not work. I think, in
> this cases the whole idea (selecting face, and there it
> is) of those boundary layers is not good.

What you see displayed while defining layers is an approximation of the final layers. Still, these can take a long time to compute on NURBS surfaces and can slow down the responsiveness. You can disable the display of these temporary boundary layers and improve responsiveness by unchecking the "Show" option at the top of the Create Boundary Layer form. You can also speed your display by using a coarse edge meshing before defining the layers, and then applying a finer spacing afterward (before meshing the face/volume).

> There should
> be something like in I-DEAS, where you can set each node
> by hand and click the element after this. Perhaps this
> sounds archaic, but so you always (!!!) get a mesh even
> in very very complicated corners.

There are several default settings which can affect the quality of boundary layers. For cases with sharp corners, normal and offset smoothing (newly supported for all boundary layer types in Gambit 1.2, due to be released very soon) can greatly improve quality. Such smoothing is off by default because it requires significantly more computation time.

  Reply With Quote

Old   November 24, 1999, 02:44
Default Re: Welcome!
  #11
Jonas Larsson
Guest
 
Posts: n/a
Yep, every new release has been a major step forward. I'm anxious to test 1.2 next week. I'm sure it will be a lot better than 1.1.3.

However, I have to say that I think that you released Gambit far too early. You've had your customers doing beta-testing for you for more than a year now.
  Reply With Quote

Old   November 24, 1999, 11:12
Default Re: Welcome!
  #12
Scott Gilmore
Guest
 
Posts: n/a
Jonas, it is true that there have been problems with Gambit in some areas, and our initial testing should have caught many of them. However, I think it is unfair to say that we have been using customers as beta-testers. We have hundreds of customers using Gambit 1.1 extensively and successfully on a wide range of applications.

Certainly, some problems still remain. We have made major investments in our testing resources, and Gambit 1.1 was the first outcome of that expanded testing. Most customer responses to Gambit 1.1 have been very positive. Gambit 1.2 will be even better, as we have further refined and expanded our testing process. (By the way, 1.1.3 is an internal development version that has not been passed through our full testing process. Please make sure you upgrade to Gambit 1.2 as soon as possible.)

Let's not degenerate into a "mud-slinging contest", but instead focus on a constructive discussion of where Gambit needs to improve.
  Reply With Quote

Old   November 24, 1999, 14:32
Default Re: Welcome!
  #13
Jonas Larsson
Guest
 
Posts: n/a
Scott, I'm sorry if you think that my comments are "mud-slinging". That is not my intention at all, in fact I'm one of your customers who do use Gambit successfully on a number of applications. And for these applications we are also very happy with Gambit - it is a wonderful product that can save you a lot of time. However, fact is that for many cases we still have big problems with Gambit. We have been using Gambit for more than a year now. I'm glad to hear that you have improved your testing process - sounds good for the future.

Does anyone else here successfully use Gambit to mesh large (> 1 million cells) and complex 3D CAD-based (IGES) geometries with resolved boundary layers? We try but often fail...

You asked about where Gambit needs to be improved. Here are a few points that come to my mind:

1. Boundary layers on curved surfaces in 3D - very unstable today.

2. Stability. Gambit still crashes quite frequently, especially if you have large and complex models.

3. CAD features - still lacking many basic functions, one example - you can't find intersections between curves.

4. Virtual geometry handling - often very inconvenient to use since there are so few geometry tools once you go to "virtual". Then when you need to do some modification everything "falls apart".

5. Speed and Memory Needs - We run Gambit on high-end workstations with 2Gigs of memory and still frequently run into speed and memory problems. If you work on large models the speed can really be a big problem - display takes very long time to update after minor changes etc.

6. Robustness - If you run into problems with Gambit it is often very difficult to take the "long and safe way" instead. A pcube style fallback would be very nice.

7. Parametric studies - If you want to mesh a number of very similar CAD geometries (very common when you are doing a new design) you have to go through the whole meshing process again for each small modification in the CAD file. A mesh-model "independent of CAD-numberings" or so would be very useful. Or perhaps a function in the journal file that would allow you to "select CAD objects" to be treated - that way you could write journal files that are general and not dependent on the CAD file numberings.

Once you get these fixed I'll be a *very happy customer* ;-)
  Reply With Quote

Old   November 29, 1999, 09:16
Default Re: Welcome!
  #14
reyman
Guest
 
Posts: n/a
We have meshed >one million cells and gotten nice solutions on a Sun 6000, but the meshes were not complex.

I've been working on the mesh of an aerosol test chamber for several months -maybe I have the Gambit disease - and the chamber contains an air sampling device with great detail. The decomposed volumes are very accurate, but meshing is very time consuming and extremely fine.

I've taken away unneccesary portions of the model to reduce numbers of cells, but the flow in the test chamber is so low that grid independent solutions are hard to come by because the values of the velocity and pressure are so low there isn't much of a delta to make comparison. Since the flow is so low, its quite possible I would be concluding grid convergence when there isn't. I'm focusing on LES and full N.S. solutions and while the Sun is a parallel computer, we have only two licenses, so there isn't a real benefit by going parallel ...

So sometimes I don't know what to do except just put one foot in front of the other and keep trying....hmmm...it is -as fluent inc. has taught us- a matter of smart volume decomposition while trying to keep cell numbers down, but this so easy to write or say and much more difficult to do. Anyone have a magic wand? Maybe just another cup of coffee for now...eh?
  Reply With Quote

Old   December 2, 1999, 22:42
Default Re: Welcome!
  #15
Scott Gilmore
Guest
 
Posts: n/a
Jonas, thanks for the feedback. I've made some comments below.

> 1. Boundary layers on curved surfaces in 3D

This will be a major focus area of the next gambit release (1.3).

> 2. Stability.

Should be much better in 1.2.

> 3. CAD features - still lacking many basic functions,
> one example - you can't find intersections between curves.

Gambit 1.2 includes edge/edge intersections, even for cases where they do not exactly intersect. We've also added projections of edges onto faces, face primitives (circles, rectangles, etc.) and more.

> 4. Virtual geometry handling - often very inconvenient
> to use since there are so few geometry tools once you
> go to "virtual".

Our plan is to provide all boolean and split operations on virtual entities. Some of these may be available as early as Gambit 1.3.

> 5. Speed and Memory Needs...

We're steadily improving speed. For example, tet meshing is 2-3 times faster in Gambit 1.2 than prior releases.

> 6. Robustness... pcube style fallback would be very nice.

We're working on a virtual unite capability that will provide the same type of "overlay" faces that you create manually in P-Cube. However, Gambit will shape (rubberband) and associate them automatically, so it should be much more convenient.

> 7. Parametric studies... you could write journal files
> that are general and not dependent on the CAD file > numberings.

We're working on approaches to make label changes less common. In the meantime, you may find it helpful to provide labels when creating entities (instead of accepting the default labels--- volume.1, etc.).
  Reply With Quote

Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On



All times are GMT -4. The time now is 09:30.