long hours simulation - how?
 

Hello,

I am simulating ventilated installlation. I am trying to do validation for a step increase in temperature at the inlet also I am performing an experiment for 3 hours and got 1600 row of data for 36 positions inside the room. In my CFX4.3 model the geometry is meshed to 60000 cells. I have a problem to do the transient iteration. previously I did 50 iterations per time step (5 seconds) and could not get the result because all the time the work station is out of memory. With the time step and the number of iterations I chose, it is not possible to perform simulation longer than 20 minutes. Do you think it is good (in the context of acuuracy)if I increase the time step? eg. to 60 seconds or so. I know some people are doing simulations for processes that are even longer than 24 hours. But I do not know how? just increasing the time step?

Thanks

CFX4.3 user

Re: long hours simulation - how?
 

How much RAM and Hard Disk space do you have on your machine ? 60,000 cells isn't particularly large. It sounds like you are running out of disk space and not RAM ? if so could you output your results every other time step, or why don't you output your results but just the variables you are interested in (say Temperature and velocity ?). These should reduce the sizes of the files that are output. Bob

Re: long hours simulation - how?
 

Bob has addressed the problem of memory. Note that before you do the long simulations, you are talking about, you will obviously need to have enough RAM and hard drive space.

Yes, you can get long simulations if you want. The long simulation is an interplay between the required time steps to obtain an accurate results and the set interation (inner and outer) per time step to achieve the required convergence. Therefore, if you reduce the time step, say by one order of magnitude (with constant interations), simulation time increase similarly.

You must be aware that with big steps you need more iterations per time step as compared to small time step in order to obtain a 'significant' convergence. Now, how big time steps should be is problem dependent. I have got no experiece with the type of model you mentioned, but that step appears big to me. In my case , I work with time steps in the range of 0.0001- 0.3 s.

regards Nyatoto