[wjy-c]

in my transient simulation, the solve process stopped at the second loop of the first timestep. no error is pointed out, but the calculation does not move on to the next step even after a whole night.
I have done the simulation before, and i do not have this problem. this time, i changed my model structure and remeshed my grids which have greatly decreased the grid number. in addition, i added porous domain to consider more details.
could anyone tell me why my simulation stopped to move on without error. compared to my previous model, i did not see any clues to result in this problem.
please offer me some advice~ thank you very much!

[wjy-c]

now i think it is the "no enough free memory is currently available" problem. so i changed the Memory Alloc Factor to 0.9 as someone recommended. after that, the "floating point overflow" problem appears.
does that imply my grid number is not sufficient to solve this problem? my max rotate velocity is 1667r/min, and i set the timestep as 1e-4s. i do not think the timestep could be smaller than 1e-4s. so i could only remesh my grid. but in my experience, only increase the grid number does not figure out the "floating point overflow" problem. i think there could be some other factors affect my calculation.
could any one help, please...thank you..

[ghorrocks]

Please post your output file and an image of what you are modelling. Also an image of what you expect to see would be good.

[wjy-c]

ghorrocks, thank you for your reply.

I plan to model some gears rotating in a gear box and to see how water and air interact. my first model named box A was like this in picture 1. and i set the water and air as homogeneous model. it would work.

then i change the gear box more smaller named box B as shown in picture 2. this time, i set the water and air as inhomogeneous model and pick standard free surface model. it turns out that it could not work due to the floating point overflow problem.
in order to deal with this problem. i remeshed my gear box model which is a little smaller than the grid in gear box A. and set the timestep to 1e-6[s]. 1e-6[s] means that the gear with max velocity only rotate 1° per timestep, but it still could not work.

moreover, i changed the settings in gearbox A which worked correctly to inhomogeneous model, and it could not work neither.

after all that i have done. i begin to think there may be several reasons as below:

(1) since my total number of grid is about 2 million and maybe over 2 million. the number is always be my biggest concerning. so i wandering if my gird is still to coarse for my transient problem. but a much coarse grid quality in gearbox A could work, so i doubt if it is a problem.

(2) for gearbox A, the homogeneous model could work, but the inhomogeneous model with standard free surface model could not work. so is the inhomogeneous model need something with higher quality more than mesh quality which i ignore?


picture 1: picture 1.JPG

picture 2: picture 2.JPG

P.S. the green parts in the picture are the gears which going to rotate in the simulation.

the output file you asked is here.
output.txt

[ghorrocks]

You should determine mesh size and time step size by a sensitivity analysis. Then you have an objective measure of the mesh size required. And for the time step size if you use adaptive time stepping homing in on 3-5 coeff loops per iteration then no need for a sensitivity analysis as it will find its own time step size.

I would not refine your mesh until you know you need it. And I would do all the basic debugging of your simulation on a coarse mesh so it runs quickly.

[wjy-c]

hi. ghorrocks, thank you for your reply.

i talk to my co-worker about sensitivity analysis, i do not know if i understand you correctly, we both think sensitivity analysis means a huge statistical work for us. do you know some way to make the sensitivity analysis becoming easier and less work?

also, i did what you recommended and set the timestep as adaptive. after i re-checked my model carefully, the solve process still stopped since the floating point.
i'm really strucked with this problem and i do not know how to deal with it now...

[ghorrocks]

Yes, a sensitivity analysis is a lot of work. You can skip it if you don't mind your results being meaningless. You will still get pretty pictures if that is all you want.

For the crashes you need to simplify your model to find what is causing the problem. Introduce the complexity one step at a time to find the problem.

[wjy-c]

your advices means a lot to me. the model I analysised is a preparation for business work. so olny pretty picture will not satisfy my boss. it seems i have a lot of work to do to figure out this problem.
in addition. i have an other question. in cfx-post view, i would like to show the whole water surface. so i use the isosurface command and set the water volume fraction as 0.5, but in some cases, it turns out not what i wanted to plot. so is there any way to show the water as a whole part?

[ghorrocks]

Have a look at the CFX tutorials for examples of post processing, especially the free surface flow over a bump one which is very relevant to your case.

[wjy-c]

hi,ghorrocks.
i have read about the example in the cfx tutorials. in that example, it also pick the isosurface and set the water volume fraction as 0.5.
so i did the same thing. in my model, i try to simulate water flow to a dam. at the begin and the last timestep, the water surface seems to be what a wanted. but in the middle process, the water was splashed since the dam, and the water surface in not continuous.
i show the picture below to help you understand me better.
i was wandering the reslut truns out not to be what i expect, so if it is my setting wrong. i set the water and air inhomogeneous, and use standard free surface model, both material are continuous.

the water splashed:
3.jpg
the water after splashed:
2.jpg