[hkw]

Hi everyone, i am currently simulating a 2D air ventilation inside a room where the hot air is coming out from a wall-mounted unit (heating mode). Here are some details for your reference:

a) Pressure-based analysis, time transient, planar.

b) Multiphase,radiation, heat exchanger species,discrete phase and the rest are off. Only energy is on and i am using Realizable K-epsilon.

c) The hot air coming out from the wall mounted unit is set as inlet with air of 8m/s and 316K, the outlet is pressure outlet with backflow total temperature of 293K (i am not sure is it correct though)

d) The scheme is Coupled with Courant number of 20; grad=least square;pressure;2nd order;momentum;2nd order;turbulent kinet.enrgy;2nd order,turbulent dissipation rate; 1st order, 1st order implicit

e)convergence criteria for continuity continuity and velocity are 1e-3, and energy is 1e-6

f) Time step size: 0.0015s, 20000 of time step, 60iterations.

The problems that i encountered are:
1) The residual plots were steady after some iterations but it became fluctuate again, but when the iterations proceeded, the fluctuation parts became smooth, i am dont understand why it occurs. (ie at 4500 iterations)

2) After 10k of iterations, the residual plots seems to increase to above the convergence criteria and the fluctuations occurred as well.

[LuckyTran]

In a transient simulation, this is desirable.

The residuals should decrease for the 60 iterations that are withing a time-step. Then on the next time-step the residuals will jump and then decrease again. This ends up looking like a saw-tooth pattern.

Actually, you should investigate why your residuals did not "oscillate" during the first 4000 or so iterations. The results you obtain in a transient simulation with smooth residuals are likely inaccurate and I would not trust those results.

[hkw]

Thank you for your reply. For your information, the residuals did oscillate in the 1st 4000 iterations, but when the iterations proceeds the oscillation seems to be less obvious. I was wondering the end of the plot is oscillating and the value is greater than the convergence criteria.

[LuckyTran]

You need to make your convergence criteria more strict, 1/2 iterations per time-step is not enough. At least try to get 3 or 4, I generally do 10 to make sure the simulation is converged at every time-step. Notice how you entered a time-step on iteration 711 and it was converged at 713. How can you determine convergence of any sequence with only 2 values? For example if you have the following sequence, can you tell me if it is converged or not?
1 1 ...

Over a large number of time-steps there will be a lot of accumulated errors and the solution will be less and less accurate.

[hkw]

ya..that is the problem I was wondering..at the beginning it needs about 50+ iterations to converge but as It proceeds it only takes around 1,2 or 3 iterations although I use 60iterations per timestep. meaning to say the criteria of e^-3 is not stricy enough right?