Simulating human thermal plume convergence problem

[rglsj]

Hello everyone. I'm trying to simulate human thermal plumes inside a room. I have a simple room with dimensions L×W×H = 10.8×5.4×3.5 m3. Inside that room, four human models are standing (two shorter, and two taller modes). There is no inlets and outlets in my domain. In the next pictures, I'll show my mesh where geometry can also be observed.

[Rob]

Not sure why you're including radiation, but otherwise the set-up looks sensible. If you plot temperature & velocity contours on a plane through a couple of the bodies and compare results over (say) 100 iterations how do they look? Hint, if you deselect "auto range" you can fix the range so it's easier to compare.

[rglsj]

Rob, thanks for your answer.
I'm including radiation because I suppose it's more realistic that way? Actually, I have an interesting comparison of the radiation effect which is shown in the next picture (the radiation included is on the left):
However, that's not the important part of my discussion. As you suggested, here are the velocity and temperature contours for every 100th iteration:
Based on these results I guess that my results don't change that much in the sense that the thermal plume gets bigger and smaller/hotter and colder, but that it goes from left to right? I'm actually surprised that the results are similar if we take into account that my monitored velocity (at the point little above the model is ranging from 0.19 to 0.26 m/s). Is the correct approach to use data sampling for steady statistics (i.e. to take the average values from x iterations?

[Rob]

This is where the engineering comes into CFD - and why you're spending the time working towards a degree of some sort.
The buoyant plume goes up, easy to see. However, the amount of inertia is low so you'll see the length vary very slightly, and it'll wobble based on other flow features. In an experiment you'll see some noise, in CFD in a transient run you'd see slight changes in the plume and understand why. In steady state we're seeing a pseudo transient effect (NOT the same as the pseudo transient solver option - that was a poor choice of label when it was implemented) where the solver drifts between an infinite number of stable solutions. Taking iteration averaged data is probably the best solution, but you also need to check that doesn't lose a jet region if the jet wobbles a lot: here it doesn't. You use the Data Sampling once the solution is fairly converged, too early and you'll be including the "unconverged" part of the run before the solver settled down.

[rglsj]

Thanks Rob, that helped a lot. I will try to do my simulations that way.