CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   FLUENT (http://www.cfd-online.com/Forums/fluent/)
-   -   Help---Gambit/Bounadry Layer (http://www.cfd-online.com/Forums/fluent/28110-help-gambit-bounadry-layer.html)

Jackson December 19, 2000 12:18

Help---Gambit/Bounadry Layer
 
I have troubles to generate a 3D grid with boundary layers attached on some surfaces and/or edges.

1. Geometry: It is a 3D turbine stator. The meshing domain is made of 7 surfaces: blade, hub, tip, inlet, outlet, and two periodic surfaces. The blade surface is splitted into two surfaces with a leading edge and a trailing edge.

2. Uniform mesh---OK. Gambit works fine if there is no any non-uniform edge and boundary layer mesh. But it is no good for viscous flow simulation because there is no enough grid points to resolve the boundary layers.

3. Cooper Scheme---not work. The two periodic surfaces are connected. The grid points on the blade contour edge in the tip surface are clustered in the vicinity of the leading and trailing edge points. A boundary layer mesh is attached on the blade contour edge in the tip surface. The tip surface is meshed with triangles or quads. The other surfaces (except hub surface) are meshed with Map Scheme with a few clustered grid lines nearby the tip surface and the hub surface. Gambit works with Cooper Scheme to generate a volume scheme. But the grid quality is very poor in the trailing edge-hub surface region. Also, there are some elements with negative volume in that region!!

4. Tet/Hex Scheme---not work with boundary layers. Surface boundary layers are attached on the tip and the hub. Gambit stops at initilization step and asks me check the surface grid. Surface meshes are just perfect!

I appreciate any suggestions!

John C. Chien December 19, 2000 13:54

Re: Help---Gambit/Bounadry Layer
 
(1). You can keep trying. Or take a look at the Turbogrid and TASCflow. (2). I have used the software for a couple of years, for 3-D turbine passage flow (stator, rotor, or stage). The grid generation and the post-processing are designed to handle the turbomachinery component flows. The boundary layer mesh is fairly easy to create. (3). You can also try the ICEMCFD products for mesh generation. The mesh can be imported into Fluent. (4). 3-D blade passage geometry for viscous flow is highly complex (you also need to use two-layer or low Re model to handle flow vortices and separations with very fine mesh). Based on my experience, Turbogrid and Tascflow is a better choice in terms of mesh generation and the speed of convergence. (it also has a lot of difficulties in generating very fine mesh for my applications.) (5).ICEM/HEXA also can handle complex geometry with boundary layer mesh, but it is a general code and requires a lot of work on your side. (6). If you use coarse mesh in 3-D turbine passage flow calculation with wall function, then you are going to miss some important features in the flow field.

Jackson December 19, 2000 15:04

Re: Help---Gambit/Bounadry Layer
 
Thanks. I work on CFD/turbomachinery for many years. I have my own CFD codes. Recently I have a chance to play with Fluent. Unfortunately I am disappointed with it. Maybe I have not invested enough time on it yet. Anyway I appreciate any comments and suggestions about Gambit with boundary layer mesh.

John C. Chien December 19, 2000 17:16

Re: Help---Gambit/Bounadry Layer
 
(1). Why not try to get a sample tutorial from the support engineer, to see if they have something similar to what you are looking for? (2). 3-D blade passage geometry is fairly typical, and the vendor must have looked at it in the past.

Jackson December 19, 2000 18:02

Re: Help---Gambit/Bounadry Layer
 
(1) I have tried the samples they have. But there is no any applications in turbomachinery. There is one example about the grid generation with boundary layer mesh inside a tube. But it is much simple than our case. (2) That is what I am saying either!

John C. Chien December 20, 2000 00:36

Re: Help---Gambit/Bounadry Layer
 
(1). I am not currently using the Turbogrid. (2). But if you just take a look at it, in terms of how the various multi-block domains are constructed to create the mesh, you might be able to use the concept in the mesh code you are using now. (3). Well, the general guideline here is to cut the computational domain into many smaller blocks, so that you can control the mesh in each block.

Dan Williams December 20, 2000 01:50

Re: Help---Gambit/Bounadry Layer
 
I'll say sorry ahead of time for not being able to offer any constructive advice on creating boundary layer grids in Gambit. I've got to agree with John here, for once ;-), that you should really dump Fluent for turbomachinery applications. TurboGrid-TASCflow, possibly with ICEM HEXA, are really the way to go. Fluent's turbomachinery cababilities are really weak.

Another product you might also look at, depending on what you need to do, is BladeGen+, also put out by CFX. It's more along the lines of a "quick" design tool for blades, but the geometries you come up with can be gridded in ICEM and brought into TASCflow for more detailed analysis.

CFX-TASCflow and Bladgen+ also use coupled solvers which work, so you get answers pretty damn quick. I'd hate to see you spend all this time in grid generation and then have Fluent's solver completely crater on you.

Maybe none if this helps you though. Your probably already commited to using Fluent.

Dan.

Frank December 20, 2000 02:50

Re: Help---Gambit/Bounadry Layer
 
I'm still a beginner in cfd for turbomachinery but I don't see gambit so terrible. ok, in the current state it is not very usable for turbomachinery-problems, but if you have the time you can try to cut the problem into several volumes (also long and narrow volumes at the blades to make the boundary-clustering). I make in the middle of the channel another volume with large hex and between the boundaries and the middlechannel a transition with tets. The great advantage I see with fluent is the possibility to make unstructured grids. Why care about skewness and aspect-ratio of structured hex-cells, why think about where which block to put and if the cell-numbers at one side plus the cell-numbers of the other block....

What advantages does the tascflowsolver have ? Is there a difference from incompressible to compressible ?

Just what I said: I'm a beginner with fluent ...

Jonas Larsson December 20, 2000 04:08

Re: Help---Gambit/Bounadry Layer
 
The boundary-layer handling in Gambit is still not robust enough to handle well-resolved boundary layers over complex 3D surfaces - typically what you have in turbomachinery blading applications. We've had very few successes with Gambit in this type of applications. One thing which has worked sometimes is to divide the domain into several smaller blocks and then mesh the boundary layers by grading the edge-meshes instead of using the buildt-in boundary-layer functionality in Gambit. A colleague of mine meshed a quite twisted and nasty compressor blade like this. I have managed to mesh a simple turbine with the buildt-in boundary-layer functionality in Gambit - it was very tricky though and I had to play with the default settings to finally get it to work. If the blades had been more complex (the blading was not very twisted and fairly 2D) I doubt if it would have been possible. Next version of Gambit will be better Fluent says...

Giovanni December 20, 2000 05:09

Re: Help---Gambit/Bounadry Layer
 
I do not know if your problem will be resolved, but you could try to use TGrid for boundary layer.

By Gambit you can obtain a boundary mesh, and by TGrid the boundary layer mesh and then the rest of the volume. See the tutorial of car in Tgrid user's guide.

I've used this procedure for a ship hull and I succeeded. You must work hard.

Giovanni

Jackson December 20, 2000 11:32

Re: Help---Gambit/Bounadry Layer
 
Thank you guys.

Dan Williams December 20, 2000 23:19

Re: Help---Gambit/Bounadry Layer
 
BladeGen+ uses unstructured grids. If you want more control though it sounds like you could do things in CFX-5. It can generate unstructured meshes tetrahedral and tet/prism grids for blade geometries quite easily. Although you might still have to be careful with the tip clearance region on the shroud.

One of the main advatages to the TASCflow and CFX-5 solvers are that they solve the hydrodyamics system coupled. What this ends up meaning is that they are much faster, more robust, and converge in far fewer iterations than Fluent. I've seen cases where CFX-5 is 10 times faster than Fluent. CFX-5 also has a very accurate and robust advection discretisation which works very well on unstructured meshes.

Dan.

Frank December 21, 2000 03:28

Re: Help---Gambit/Bounadry Layer
 
Gambit should also mesh with boundary (prism) and unstructured with tets, but in more complicated geometries...Did you see the successful meshing in cfx-5 ?

Fluent also has a coupled solver. Did you compare the tascflow-solver with the coupled-solver in fluent or only with the segregated ?


John C. Chien December 21, 2000 03:56

Re: Help---Gambit/Bounadry Layer
 
(1). In the last several years, I have used Fluent and Tascflow families extensively. (2). I don't want to get into the area as to which code is better. But What I can say is: Tascflow has done home works in the turbomachinery area. As a result, it is easier to use in grid generation , post-processing, and the solver is faster. (3). My understanding is Tascflow is based on low speed pressure-based formulation which can be used in low supersonic speed. While Fluent used to have a pressure-based formulation for low speed flow, and a density-based formulation for compressible flow. The compressible flow code is much , much slower than its incompressible low speed code. (4). Another difficulty in answering the question is: some users in research field are not too concern about the time involved to create the mesh or the speed of the solver. On the other hand, in industries, time is money. Therefore, users in industries are extremely sensitive to the time required to generate a mesh, and the speed of the solver to get the converged solution. (5). Here, I am assuming that the code involved will generate results with acceptable accuracy. But in many cases, this is a much harder issue, because of the turbulence model and the computer capacity or the mesh size. So, the user's constraints are also very important in rating the code.

Craig Hornsby December 21, 2000 11:36

Re: Help---Gambit/Bounadry Layer
 
The upcoming version of Gambit will certainly address many of the issues raised in these discussions.

Over the past year we've been developing an add-on product, G/Turbo, precisely to address the difficulties many of our turbomachinery users are facing currently.

This turbomachinery focussed pre-processing tool will allow such meshes to be created much faster and easier, whilst retaining the flexibility of general purpose Gambit should additional, innovative features wish to be included in the geometry. Regular Gambit will also benefit from many enhanced features employed in G/Turbo - boundary layers being a particular focus.

Rest assured these issues are being addressed.

Craig Hornsby

G/Turbo Product Manager

Philip Zwart December 22, 2000 09:16

Re: Help---Gambit/Bounadry Layer
 
Fluent may say they have a coupled solver, but you don't have to read through too many messages in this forum to see that it really tends not to work. So in practice most people use their segregated solver.

With TASCflow and CFX-5, on the other hand, there is no choice but to use the coupled solver. This results in reliable, robust, and fast convergence, with no need to mess with all kinds of underrelaxation factors. CFX has many years of experience with the solver for both structured and unstructured grids... for more technical details see the following papers:

B.R.Hutchinson and G.D.Raithby, "A multigrid method based on the Additive Correction Strategy", Numerical Heat Transfer, Vol. 9, 511-537, 1986.

M.Raw, "Robustness of Coupled Algebraic Multigrid for the Navier-Stokes Equations", AIAA-0297, 1996.

Regarding the mesher, CFX-5 has an automatic hybrid mesher with a bunch of useful features such as curvature-sensitive and proximity-sensitive mesh spacing. See the web page for more details.

Shyam Kishor December 22, 2000 16:56

Re: Help---Gambit/Bounadry Layer
 
You should not have any problem with Blayers & unstructured mesh (Prism + Tet). Please make sure that the boundary layers have "internal continuity" option on.

For Hex elements & Blayers: You would need to divide the domain in multiple blocks. For larger twists, you should use MAPPER in individual block, which can handle the curvature in a better way.

As mentioned by Craig, G/Turbo will have it all automated.

Dan Williams December 24, 2000 01:01

Re: Help---Gambit/Bounadry Layer
 
Yes, CFX-5 can successfully mesh many complex geometries with boundary-layer grids, or grids with prismatic layers on walls to be more specific. I'd say it does the job as well as many other tools out there, maybe with the exception of ICEM. There are nice meshing features in CFX-5 such as proximity detection, angular resolution, line/surface/point mesh controls which make things alot easier also.

Yes, Fluent has a coupled solver. However, in practice people tend to revert back to the segregated solver becuase the coupled solver has an extremely sparse feature matrix and it tends to not work well. Other postings in this newsgroup show that this is the case.

There could be many reasons for this, but I'd venture to say that it's because Fluent does not have much experience with this sort of technology. It's one thing to say you have a particular feature, whether it works or not is a completly different issue.

In CFX-5/TASCflow there is no segregated solver and there will probably never be one. Hydrodynamics is always solved coupled. Multizone grid interfaces are always fully coupled. This results in a fast, robust, and very reliable solver.

Dan.

Amadou Sowe January 2, 2001 11:55

Re: Help---Gambit/Bounadry Layer
 
I have used Fluent since 1994 and used CFX 4 enough to make an objective comparison between the two. Both have strengths and weaknesses. The superiority of CFX over Fluent that you consistently pupport in this forum simply does not exist.

It seems to me, that having a density based solver for fully compressible flows is a good idea. On the other hand, it is also a good idea to have the pressure based solver (segregated) for midly compressible to incompressible type flow problems.

It is true that the coupled solver is more difficult to use but I have had some good results using it.

Finally, I personally would rather see this forum continue to be a plat-form for exchange of ideas on CFD as it pertains to Fluent products than be an advertising forum for other CFD vendors.

Dan Williams January 2, 2001 17:41

Re: Help---Gambit/Bounadry Layer
 
You obviously have not used CFX-5 or CFX-TASCflow at all then. The coupled solvers used by these packages are far superior to what is avaliable in Fluent. There are many other areas in which CFX-5 and CFX-TASCflow are very superior to Fluent, the main ones being robustness and accuracy. In addition there are usability issues like not having to recompile user source code to run a transient boundary condition for example -- there are likely others as well.

I have colleagues who are very experienced users of Fluent and have had limited success with their coupled solver. In some cases it might work (yourself being an example), and in many others it just doesn't and you have to fiddle around forever to get it working at all. When you finally do, it is slow and probably doesn't converge. So, along with many other complaints about the coupled solver in this forum, I've personally drawn the conclusion that it is not working all that well. Since Fluent provides an extremely limited feature set for use with the coupled solver, I've suggested what I felt is simply a better alternative. By no means am I categorigally suggesting that someone should switch from Fluent. You've gotta use what you need.

There is no sound technical argument why a pressure based solver cannot work as well as a density based solver for compressible flows. People do it all the time. I'm not sure what you are trying to say.

I'm not sure what you mean by "advertising for CFD vendors", I've simply been pointing out an alternative when it seems people are having a lot of trouble with the most basic things. Maybe I've been a little blunt when pointing things out, but I've just been suprised with some of the things I've read.

If discussing the alternatives to Fluent in a Fluent forum does not fall into the category of "an exchange of ideas on CFD as it pertains to Fluent products", then what the heck does? You should feel free to point out the weaknesses of CFX products on the CFX forum if you so desire.

Later, Dan.


All times are GMT -4. The time now is 23:26.