CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)

 bjcz April 22, 2002 11:16

Hi, friends,

for unsteady calculation, what difference is for the following two calculation procedures for results? ( in other words, which result is correct?):

1)

if i first compute a "steady" solution (e.g., 4000 time steps), then use the "steady" solution as initial field to compute the unsteady flow until cartain time ( e.g, time = 10,000dt )

2)

directly compute the unsteady flow using freestream value until certain time ( again, e.g., time = 10,000dt)

 sylvain April 22, 2002 11:53

Both procedure should be correct.

But since the steady solution is a solution of the Navier-Stokes equations - even if it is an unstable one -, it could be difficult to obtain a time dependante solution with the 1st procedure.

The second procedure should give directly an unsteady solution, may be unstable, but it will take a long time to get throught the viscous stage (creation of the first vortice behind the obstacle and convection of it outside of the computationnal domain).

To conclude, the first procedure will show the destabilization of a (forced) stable wake. The second one will show the establishment of the wake. Both will give at last the unsteady flow.

 Dimitri Nicolopoulos April 23, 2002 04:26

I agree, both solutions will eventually yield the correct unsteady flow. However, depending on the kind of flow and kind of code you use, you might decide for strategy 1 or 2. For instance, our code, RADIOSS-CFD is an explicit 3D Navier Stokes solver. This means that we only solve transient problems and that the time step is constrained by a CFL condition to a small value, independently of the flow velocity. Most simulations unfold in two phases, 1) reach an unsteady converged solution 2) carry on computation to study the flow, noise... With an explicit code, 1 can take as much time as 2 which sometime can be frustrating so we spend quite a lot of time trying to reduce phase 1.

If we try to simulate a reasonably high velocity flow, where we dont need a huge number a time step for a particle to go through the computational domain, we chose solution 1 and initialize the velocity flow to inlet velocity, possibly with a patch of velocity 0 in the wake of an obstacle to speed up the apparition of unsteadiness. Examples: flow passed a car side mirror.

If we try to simulate a much slower flow, then convergence can take forever. For example, it is extremely slow to have a big volume downstream of an automotive axial fan converge. In this case, it can be interesting to start from a RANS steady state simulation.

In any case, convergence toward the final unsteady flow has to be monitored thoroughly. The first order parameters to be checked include system kinetic energy, pressure evolution on given points. We also use second order convergence parameter by doing time/frequency analysis of the pressure fluctuations at given locations. As a matter of fact, there are spurious tones that remain present for a significant number of time steps after the convergence of first order params. Once the time/frequency analysis doesnt show any more spurious tones, we switch to 'recording' mode, that is study of the actual flow.

Good luck! Dimitri www.mcube.fr

 peter.zhao April 26, 2002 01:13

I think it depends on what kind of method you choose and what kind of problem you solve.If the problem is a unsteady one,you will not get a correct answer by using some particular techniques,such as local time stepping,multigrid and residual damping, which are very useful for steady flow calculations,but not for unsteady flow because the variables at different time have interacted each other; If you don't use these particular techniques, you can get a same answer as a unsteady flow calculation does.

 bjcz April 27, 2002 05:55

Hi,

Dr.sylvain, Dr. Nico., and Dr. Zhao,

Thank you very much for your excellent answers. your comments on unsteady calculation is very help to my work, and clarify my some misunderstandings.

thank you again, and best wishes to your research.

 Jim Clancy May 9, 2002 18:57