[Red Ember]

Hi to all!

I have a suspicion that domain initialization can harm...
Sometimes it is useful for solution robustness, but I simulated a flow that is injected into the ring and goes out through numerous round holes and some other outlets.
I added cylindrical bodies to that holes to avoid reverse flows. Convergence was tolerable, but results proved that flow has no obvious simmetry. Flow was nearly absent inside some of cylinders - with very low velocities.
So, I got an average velocity meaning from CFX-Post and added this meaning to domains-cylinders as initialization option. New solution was finished successfully due to residuals and monitor points, but! Results proved to differ from previous ones. Flow velocities in cylinders are relatively even now, but massflow through other outlets that I have mentioned above is close to zero.

So, I added just one initialization option and get crucial difference that I couldn't predict looking at solver graphs. What is the purpose of initialization then?

Can anyone explain it to me or recommend something, please?

[ghorrocks]

The domain always needs to be initialised with something. If you don't specify what the initial condition is then CFX will make something up.

Some simulations bifurcate. That means that one initial condition will result in one flow condition and a different initial condition will result in another. For instance a thin fluid jet in a venturi expansion. If you initialise with the jet against the top venturi wall it will stay there. If you initial with the jet against the bottom venturi wall it will stay against that wall. So which wall the jet sticks to will depend on the initial condition.

[Red Ember]

Originally Posted by ghorrocks

The domain always needs to be initialised with something. If you don't specify what the initial condition is then CFX will make something up.

Some simulations bifurcate. That means that one initial condition will result in one flow condition and a different initial condition will result in another...

This doesn't bode well then... So should I input any initialization data if I suppose it close to true values? Cos I see no sense in this option in other way and it seems to me it'd be better to let CFX solver "make something up"...

Ghorrocks, could you add a picture concerning your example, please? Or may be link if such instance exists somewhere in the internet? I'd love to simulate such case and share it with junior colleagues.

[Opaque]

I assume you are running a steady state case, correct ? or is it transient ?

Not clear what you mean by averaging values. Are you referring to spatial averaging or time average ?

Are you solutions converged to tolerances where there is no doubt further iterations will not contribute to the solution any further ?

For steady state linear problems the well converged solution is independent of the initialization values. Initial values only provide a shortcut towards convergence.

For steady state non-linear problems, you may have multiple solutions. One of them may be the most stable, i.e. if you perturb the converged solution and iterate further it returns to the same solution. Other solutions may be moderately stable, i.e. after certain values of the perturbation, the solution converges towards one of the other solutions (if it does not diverge).

[ghorrocks]

There are many examples of flow bifurcation/multiple solution on the web. Just google it.

[Red Ember]

Originally Posted by Opaque

I assume you are running a steady state case, correct ? or is it transient ?

Not clear what you mean by averaging values. Are you referring to spatial averaging or time average ?

Are you solutions converged to tolerances where there is no doubt further iterations will not contribute to the solution any further ?

Steady case, correct
The averaging was spatial.
Yes, I have no doubt that convergence is very good for my mesh and RMS-level and monitor points won't change significantly if I continue

My ring has got one inlet (p* and T* as BC), 66 round and 24 other-shaped outlets.
As I mentioned above, I added additional domain - bodies of proper shape to neglect reverse flows influence.
To be honest, these flows exist at outlets, about 1% of area, depends on velocity values.
I united these bodies into one additional domain. The BC for all 24+66 outlets was massflow value.
So, I performed the solution with no initialization values. Using CFX-POST, I get mass-averaged velocity values for 66 round outlets and used it for additional domain. Velocity direction was accurate for cylinder bodies and deviated for other outlets.