hi,

I have discussed my research problem on this forum before. I have had good answers from this forum. I am facing a problem with the time it is taking me to get results from my model.

My experiment is 30 day long. With fluent runs in serial mode, it is taking weeks to get the results. The biggest time step size that my model is accepting without convergence problems is 1 sec. Is it true that natural convection problem can only be solved correctly with small time steps. A time step size say 30 sec or 60 sec is just not acceptable? Is it a problem with fluent or any other CFD code? I am wondering if I should go for some different software - May be the type of problem that I am trying to solve is not a typical CFD problem - Can some one suggest me any other alternatives ? Had I written my own code, would I have been able to solve the problem with bigger time step size?

Correct me if I am wrong, but isn't the timestep size is obtained from the smallest time scale of the problem. If your natural convection involves some physical process at a very small time scale, then maybe it is not possible to go for larger time steps. All the best.

Arun

We do a lot of work with natural convection of various types. With transients we have found that a small time step is always required initially, but we can usually increase the step size over time. For example, we might start with a step of one minute, and gradually increase it to 5 minutes as the solution progresses through time.

Our model is running well with small time step size (~0.5 sec). It is showing good convergence (all residuals less than 10^-4). Convergence is also verified by monitoring several other parameters such as avg. grain temperature, avg. headspace temperature, avg. co2 concentration in headspace etc. But we are not being able to get converged solution with time step size more than 1 sec. The biggest time step size possible is 1 sec. If time step size is increased above 1 sec, continuity residual oscillates around 0.5 to 0.9 range (which suggest incorrect solution).

SHOULD I JUST WAIT MORE AND THEN INCREASE THE TIME STEP SIZE ?

I waited a little longer and now my model running stable at 1.5 sec time step size. I guess I should just continue monitoring every day and increase time step size every day when solution stabilizes. Is there a way to automate this ?

there is an adaptive time stepping feature, I have only had a quick play with it so cant really help much but if you check it out in the user guide it may be of help.

i tried to do adaptive time stepping, but the time step size does not seem to rise. It is running at minimum step size of 1 sec. I have set max time step size of 5 sec. Truncation tolerance at 0.001

can you suggest me correct values of different parameters involved in adaptive time stepping ? Fluent manual does not have good discussion of this topic. I also could not find much information on adaptive time stepping on this forum after doing search for adaptive time stepping.

I was thinking may be I should write a UDF to control time step size - but i dont have much experience with that -- have anyone written a UDF for variable time step size ? -- In my current set up, 1 sec time step size is giving good results -- All the average values that I am monitoring are stabilizing very well in less that 15 iterations -- I am allowing 20 iterations per time step anyways ... -- Also all the residuals are reducing to less that 10^-4 or 10^ -5 thus I am quite confident about the results accuracy for time step size of 1 sec -- I let this model run for over 10 hours and then decided to increase the size to 2 sec -- but then at 2 sec, the model does not give good convergence ..

HOW DO I WRITE A UDF WHICH WILL INCREASE TIME STEP SIZE WHEN ALL RESIDUALS FALL BELOW 10^-5 AND WHICH DECREASE TIME STEP SIZE WHEN GOOD CONVERGENCE IS LOST.

PLEASE GIVE ME SOME ADVICE .

I just talked with fluent service engineer. It appears that the only option in my problem case is 1 sec time step. I just can not increase it. It appears that for my problem, the biggest time step size is 1 sec. It is hard to believe, but that is what Fluent engineer seems to think. Does anyone have a contradictory opinion ? This delays my project by several weeks. Since the grid size of my model is quite small, my problem is not suitable for a cluster run. Thus the real culprit in my case seems to be the long duration of experiments = almost 40 days ..

please let me know if you know any way of increasing the time step..