[lehoanganh07]

Hi, I am doing unsteady calculation. I read in the Fluent theory guid that the time step size should be set so that the residual should reduce 3 order. But I dont understand what it mean?
Also I am trying to do one test and the residual become like in the picture below. It seem not good.

From Fluent, time step can be estimate as: dt=typical cell size/velocity

So please help me:
1. How to set time step in order the residual reduce 3 order magnitude !!!
2. How to check the residua converse in Unsteady solution !!!
3. What is the typical cell size !!!

[beer]

Hi

It looks perfect for me.
What happens in your picture is that each step converges to 1e-3 and then the next step starts. Obviously some stuff change and the residuals go up at the beginning of each time step again. That's why saw tooth tread design of the residuals.
If you want to converge it further than that click on "Monitor" in the left panel, choose the residual monitor and change the settings in the window which pops up. You can set scaled and normalized residuals (make sure you know what each means before changing it), set the convergence criteria or make the iteration number independent from convergence completely, which means that fluent will perform the chosen number of steps regardless of the residual.
To question one: That is only needed for normalized residuals which, I repeat myself here, are not very handy.
2. should be clear from the text.
3. The cell size depends on the geometry and the physics used. If you showed a geometry I could tell you more.

[lehoanganh07]

Originally Posted by beer

Hi

It looks perfect for me.
What happens in your picture is that each step converges to 1e-3 and then the next step starts. Obviously some stuff change and the residuals go up at the beginning of each time step again. That's why saw tooth tread design of the residuals.
If you want to converge it further than that click on "Monitor" in the left panel, choose the residual monitor and change the settings in the window which pops up. You can set scaled and normalized residuals (make sure you know what each means before changing it), set the convergence criteria or make the iteration number independent from convergence completely, which means that fluent will perform the chosen number of steps regardless of the residual.
To question one: That is only needed for normalized residuals which, I repeat myself here, are not very handy.
2. should be clear from the text.
3. The cell size depends on the geometry and the physics used. If you showed a geometry I could tell you more.

Thank you for your quick answer.
I don’t know the meaning of scaled and normalized. Could you explant them?
Also, in my calculation, I check the mass flow rate at inlet and outlet, the different is around 1%. So it seems OK.
But in monitor residual, I set the critical is 1e-8. So from this result, the solution is not conversed yet and also it increases after each time step.
I post my domain as picture below.
Please help me

[beer]

Scaled residual are the sum of the absolute values of the residual devided by the sum of ap*phi, which is basically the error in percent in some sense. The normalized residual is this residual devided by the one in the first iteration (or 2nd, 3rd). The normalized one shows you how much better the solution becomes, not how good it is. For example if you start with the perfect solution, the normalized will never be less than 1, the scaled residual is around 1e-13 due to machine precision. That is why I say stick to the scaled residual for the beginning. It is just more straight forward. To your changing solution: It is reasonable that the solution changes. It is what you want to see in a transient simulation. From what you write it sound a little bit like you are more interested in a steady state solution, I could be wrong though. But what you should do is start with a steady state solution and than switch to transient.With that you get rid of simulating the whole process from the beginning, if you don't need if of course. The options in your panel look ok, 1e-8 is very small though, you will calculate a long time for that. 1e-4 for momentum is usually ok, less for energy, I go for at least 1e-6

[lehoanganh07]

Originally Posted by beer

Scaled residual are the sum of the absolute values of the residual devided by the sum of ap*phi, which is basically the error in percent in some sense. The normalized residual is this residual devided by the one in the first iteration (or 2nd, 3rd). The normalized one shows you how much better the solution becomes, not how good it is. For example if you start with the perfect solution, the normalized will never be less than 1, the scaled residual is around 1e-13 due to machine precision. That is why I say stick to the scaled residual for the beginning. It is just more straight forward. To your changing solution: It is reasonable that the solution changes. It is what you want to see in a transient simulation. From what you write it sound a little bit like you are more interested in a steady state solution, I could be wrong though. But what you should do is start with a steady state solution and than switch to transient.With that you get rid of simulating the whole process from the beginning, if you don't need if of course. The options in your panel look ok, 1e-8 is very small though, you will calculate a long time for that. 1e-4 for momentum is usually ok, less for energy, I go for at least 1e-6

Thank you,
So It better to run with steady calculation first to get convergent, then switch to transient solution !

[beer]

Yep. As long as you are not interested in the process from the beginning on.