[dzaky.naufal]

I’m performing a transient simulation of a 3D solar chimney system with 24 hours time range (starting from midnight until the next midnight) for my thesis with ANSYS Fluent to see the natural convection phenomenon. Right now, I’m only simulating the fluid domain (the solid domain isn’t simulated) and the fluid that I'm simulating is air. I'm using pressure-based solver, PISO coupling scheme, and Boussinesq density function. The computational domain and boundary conditions are shown below.




The result variables that I want to see are temperature, velocity, and volumetric flow rate. I wanted to make sure that the mesh size and time step size are small enough and the number of iterations per time step is big enough to give me valid results, so I did 'convergence tests' for each of them separately. Unfortunately, I failed to reach convergencies. Here are the details:

1) Convergence test of mesh size: 270 mm, 90 mm, 70 mm, 50 mm (time step size = 300 s, number of iterations per time step = 15)
2) Convergence test of time step size: 300 s, 200 s, 100 s (mesh size = 50 mm, number of iterations per time step = 15)
3) Convergence test of the number of iterations per time step: 5, 10, 15, 20, 40 (mesh size = 50 mm, time step size = 300 s)

Here are the results of the volume-averaged temperature and volume-averaged velocity for each of the convergency tests above.








I’m not so sure about the rest but aren’t the temperature plot for both time step and iteration convergence test show that those plots diverge?

Did I make mistakes in the convergence tests? Or is it possible that I went wrong on other parameters, such as density function, coupling scheme, or even in the meshing process? Is it also possible that the cause of the problem is the absence of the solid domain?

I'm relatively new to CFD, so I rely heavily on trial-and-error method but it seemed ineffective and time-consuming, so I'm really looking forward to your help.

If you need more details, please let me know.
Thank you.

[dzaky.naufal]

It seems that the images aren't appearing in my post above and I don't know what's wrong. So here I am reuploading the images.

The computational domain
https://drive.google.com/file/d/1zCT...usp=share_link


The Boundary conditions
https://drive.google.com/file/d/1csE...usp=share_link


Results of mesh size convergence test
https://drive.google.com/file/d/1hDI...usp=share_link

https://drive.google.com/file/d/1nN6...usp=share_link


Results of time step size convergence test
https://drive.google.com/file/d/1ugj...usp=share_link

https://drive.google.com/file/d/1_Hd...usp=share_link


Results of iteration convergence test
https://drive.google.com/file/d/1lX2...usp=share_link

https://drive.google.com/file/d/1U89...usp=share_link

[AlexanderZ]

initialize case with small time step to get solution with good convergence for several time steps, then increase timestep

in perfect case solution should be independent on mesh and time discretization (mesh size and time step size)

to prove (if you need it) independence you are decreasing mesh size and time step until your solution is not significantly different anymore (you are defining appropriate level of error by yourself)

[dzaky.naufal]

Thank you for your answer.

"initialize case with small time step to get solution with good convergence for several time steps, then increase timestep"
I'm not sure if I understand that correctly. So in my case, should I use small time step size for, let's say, 10 time steps only? How small is the time step to be considered small? Don't I have to find the right value with the time step independence test?

And regarding iteration per time step, some references I found said that 15 to 20 iterations per time step is usually used as a rule of thumb. Can I use that directly without doing an independence test similar to that of mesh and time step size? Or should I decide the number of iterations based on the residual errors?