Hi,

I have a 3D model of a centrifugal pump which I am having problems getting to run. The model has no negative (Non-postive) volumes when I do a grid check prior to clicking "iterate". However, depending on the time step used, I get negative volumes either at the beginning of the 1st time step or the beginning of the second. The model stops running soon afterwards due to a problem in the AMG solver. The model is a segregated, sliding mesh simulation with a realizable k-e model. I am using the PISO scheme. I do not understand why Fluent is behaving in the manner. Any ideas?

Also, I am modeeling a slice of the pump assembly in 2D. I am trying to model cavitation, but my model stops running after a while due to error> greater than. The program is trying to limit the viscosity ratio to 500000 prior to this. Does anyone know how I might solve this problem? I am currently trying to see if the viscosity under relaxation parameter can have an affect on the problem. Thanks for your help.

Jack.

The sliding-mesh setup in Fluent is very buggy. You often have to tweak a lot of things before it runs well. Here are a few hints that has helped me:

Before defining each sliding interface do a grid/check. Also make sure that you have defined the rotational axis of the fluid-domains correctly before you define the sliding interface.

If running in parallel de-select "parition across zones" before partitioning.

Use quads on the sliding interfaces (seems to work better than tris).

Hi Jonas,

I have done a grid-check prior to defining my sliding interface. It makes no difference. I am aware that you have to tweak the model sometimes to get it to run. I learnt this from you several months back on this forum!! I have defined the rotational axis of the fluid domains correctly...That was the first thing I checked!! I am not running in parrallel as yet, because I want to simply get the model to run first, then I will learn about the parallel modelling and partioning. Thanks for your help...any other suggestions would be welcome. Any ideas for the 2D cavitation model? COntinuity converges very poorly in the cavitation mocel, and then diverges rapidly until the model stopes due to this. ????

I have the same problem with sliding mesh. After some time-steps the problems diverges. Even if I start a steady-state solution with initialisation=0 at this position with very low relaxations it doesn't start anymore. Seems to be a fluent-specific problem.

(1). What I am going to say is just a general suggestion. It is not in any way related to a specific code. (2). Create a simple 2-D blade (straight blade or circular arc) such that the mesh is nice and smooth. The blade must have a finite thickness with a round leading edge to avoid flow separation. Keep the blade short to avoid highly skewed mesh for testing purposes. (3). You can add an inlet so that the flow will make a 90 degree turn into the straight 2-D blade area. If necessary, add a smooth hub in the center to guide the flow on the hub side. In other word, you will have parallel hub and shroud surface (flat and 2-D). This is a typical 2-D configuration, although it is really 3-D. (4). Run two cases, one in stationary frame of reference, and the other in the rotating frame of reference. This can be done because you don't have the volute attached. By assuming that the volute has no effect on the blade passage flow field, you can simplify the problem. Just remove the volute and put the pump in the swimming pool, or something like that. (5). In this configuration, you can check out your mesh, algorithms, and the turbulence models or other physical models. We are assuming that the volute effect is secondary on the downstream side of the flow. (6). At this point, you should know whether you are going to have problems or not. (7). Once you have solved all of these problems, then you can move on to the fancy "sliding mesh" or something like that. I have never used that options, so there is not much I can say in that area. Especially, I have not touched the code for almost two years now. (8). In this way, at least you will be able to find out whether the code can handle the simple design or not. (9). Based on my experience, the mesh generated by one code does not automatically guarantee that the solver will have no problem. In other words, the solver in most cases is more sensitive to the mesh quality and skewness, not to mention the precision math issue. So, positive volume along is not really adequate. I had the similar problem with a solver, which tends to diverge after a while when used with a mesh which is legal in the mesh code. (10). So, I think, in most cases, it is the quality of the mesh which tends to create troubles in the solver stage. And by using simple blade configuration and boundary conditions, one can remove these un-necessary complications. (11). After these exercises, if the problem is still there, then just try something else. There is not much one can do with a black box.

As you ask for any other suggestions I did a lot of this simulation over the last years using StarCD and most of the time without stability problems (v31b with AMG solver works great).