[saisanthoshm88]

I have a query regarding transient simulations.

In case a transient simulation produces a steady state solution after progressing for a certain time, how is it possible to confirm such trend.

I mean is this done generally by looking at some monitor points.

And if a steady solution exists to such problems why is it not possible to achieve it by a steady state simulation directly instead of running a transient simulation for a certain time.

Could someone please provide a clarification to this

[ghorrocks]

The main reason is because if the flow has transient behaviour then a steady state simulation will never converge because it is not steady. These flow transients are very common in fluid flows - they occur as vortex streets off bluff bodies, at separations jiggling about and at laminar to turbulent transition with laminar separation bubbles which jiggle about.

[evcelica]

You very possibly could achieve the same result using a steady simulation, but your residuals won't converge as Glenn says. Instead you would monitor the result of interest and your imbalances to determine convergence, instead of the residuals.

[JuPa]

A faster way of getting the solution is to run a steady state solution and get as close to a steady state solution as you can, and then switch to transient simulation. You effectively reduce how long you spend in transient mode and shorten the overall combined run time.

[yylaw]

Originally Posted by ghorrocks

The main reason is because if the flow has transient behaviour then a steady state simulation will never converge because it is not steady. These flow transients are very common in fluid flows - they occur as vortex streets off bluff bodies, at separations jiggling about and at laminar to turbulent transition with laminar separation bubbles which jiggle about.

Is it possible to track the 'steady state' situation from transient simulation ?

[ghorrocks]

What do you mean by "track" it?

If you run a transient simulation of a steady state flow it will eventually settle down into the steady solution and not change with time (which is the definition of a steady state flow).

[yylaw]

Please forgive my confusion. "tracking" refers to the detection of steady-state behavior from transient flow simulation. My question is how can I observe or know when and which region has reached steady state and will not change with time, for example, from a velocity plot, single or multiple point value of a contour plot, or NS equations ?

My case is multiphase flow simulation. It can only converge in a very small time step (<1e-10) when running with steady state setting. So it is impossible for me to run in the steady state.

Now, I am running transient simulation with higher time step, but the time-varying flow condition has brought me trouble in my later works.

From my current observations, this flow will never reach a state, but it may reach a "periodic flow 'in a certain time range, rather than a" fixed flow' in the whole time. So my idea is to get 'periodic flow', some sort of 'pseudo steady state'.?

Thank you for any advice.

[ghorrocks]

Do not use tiny timesteps (unless your geometry is tiny and requires it). It will make the residuals decrease, but it is not real convergence. It is just the tiny increments from step to step are not handled well by the residuals algorithm.

The best way to see if a transient flow is actually steady state is just to march it out in time and see if it settles down to steady state.

If your simulation is periodic in time then you will have to run it long enough until the periodic flow is established well enough that you are confident about it.

[yylaw]

Originally Posted by ghorrocks

Do not use tiny timesteps (unless your geometry is tiny and requires it). It will make the residuals decrease, but it is not real convergence. It is just the tiny increments from step to step are not handled well by the residuals algorithm.

The best way to see if a transient flow is actually steady state is just to march it out in time and see if it settles down to steady state.

If your simulation is periodic in time then you will have to run it long enough until the periodic flow is established well enough that you are confident about it.

Thank you for the reply. Even I said "transient simulation with higher time step" , but actually time step size I am using in the case will never converge unless i set the time step smaller than 1.25e-6. I am not sure whether the value consider "tiny time step" as you mentioned ?

After several trials, I set the transient simulation as follow:
For "Fixed" type
Number of time steps: 2e5
Time step size: 1.25e-6s
Max iterations per time step: 25

At the same time I run "Adaptive" type :
Number of time steps: 2e5
Initial time step size: 1.25e-6s
Number of fixed time step: 1s
Minimum time step size: 1e-8s
Maximum time step size: 1s
Max iterations per time step: 100

Adaptive type failed to converge but Fixed type yes.

From Adaptive type, I observed that each time step will converge at around 25 iterations no matter how I set the "Max iterations per time step" .

Hence, based on my condition, how can I observed that each time step has reached "real convergence" as you mentioned above ?
--------------------------------------------------------------------------------
Besides, since the case is too complicated and time steps size cannot be bigger than 1.25e-6s, the solution process take long period to run until "periodic steady state" can be observed.

I have 2 questions here:
1. How can I verify the "periodic steady state" is confident enough to explain the flow field, by means of ?
2. Is there any other way to speed up the calculation process (eg: set time step size much bigger than 1.25e-6, set higher tolerance? but able to converge)?

Thank you again for any replies.

--------------------------------------------------------------------------------
All methods mentioned above is with RNG-k-epsilon turbulence model, 2nd upwind, PISO scheme

[ghorrocks]

PISO scheme? CFX does not have the PISO scheme. Are you using CFX?

[yylaw]

Originally Posted by ghorrocks

PISO scheme? CFX does not have the PISO scheme. Are you using CFX?

Nope, I am using Ansys Fluid Flow (Fluent)

[ghorrocks]

Then try the Fluent forum. This is the CFX forum.

[yylaw]

Originally Posted by ghorrocks

Then try the Fluent forum. This is the CFX forum.

Thank you for the remind. I have post a new thread at Fluent forum.
"steady/repeat periodic flow" from transient simulation for multiphase flow ?

But I think it doesnt really limit to whether CFX or Fluent I am using. My main task is to find the "steady/repeat patterns from transient simulation".It is more mechanical issue rather than software issue.

If more information required lemme know !

Open for any suggestion. Thank you so much !

[ghorrocks]

We will not be able to give you any advice on your simulation set up or which options to try in that case. But general comments:

"How can I verify the "periodic steady state" is confident enough to explain the flow field, by means of ?" - Once you have a validated verified simulation then you run it long enough to get a few cycles of the periodic flow and if you are happy it is periodic then it is done.

" Is there any other way to speed up the calculation process (eg: set time step size much bigger than 1.25e-6, set higher tolerance? but able to converge)?" - The validation and verification process will tell you the fastest your simulation will run. If that is too slow then you need to purchase more computing power to go faster.