I have recently observed that the CFD results shows significant variations with different initialization techniques (mainly standard, fmg and hybrid). What could be the technical reason for this ?
I believe, the results should not change irrespective of the values they are initialized with.
The mesh and all other solver settings remains the same.
Thanks in advance.
Rajesh.
Hello Rajesh:
Without more details it is difficult to give you specific comments.
On steady simulations, the initialization should not impact the final converged solution. It only affects the time (number of iterations) to reach it.
Even in unsteady simulations, the results should not vary significantly.
On the other hand, what do you mean by significant variations? 100%? 10%? 1%? If the criteria to stop the simulations (to reach convergence) is not small enough you might expect some deviations but I would say in the order of few percentage points.
Now, if we talk about all significant digits...yes, I'm not surprised you get deviations. Back in my early days learning CFD, I found out in several tests I performed that even the computer/processor one uses affect the results one gets. But again, those deviations are minimal and insignificant when you consider truncation or discretization errors.
Hope it helps.
Larry
Two options: i) the flow is not steady; and ii) the solver do not reach the steady state.
Hello Rajesh,
The 1st answer is a possibility, but there are other ones. If we think that the system has a solution, and its solution is unique at steady state, the solution will be independent of initial conditions, and the hypothesis written by João Henriques are probably the source of your case.
But, Navier-Stokes equations can easily acquire a chaotic behavior. These systems are dependents on the initial conditions Little modifications in initial conditions let the solution to diverge of a expected behavior after some iterations or time steps. Weather previsions are a common case. We have some confidence of few days due the dynamic behavior of the system and the intrinsic error in the measures on each station as temperature, wind velocity, moisture and pressure.
Nevertheless, an average behavior can be expected and sometimes predicted.
Other case is the system to have a periodic behavior, statistically stationary, but with predictable instantaneous variations as a sin function, for example.
Dear Rajesh Alayil,
I am commenting only on "I believe, the results should not change irrespective of the values they are initialized with."
Nowadays, I am also working on INITIALIZATION step. As Dr. João C Henriques mentioned above. Steady sate cases normally independent of initialization values. We just notify the bigging sate like volume fraction of gas in the liquid tank then CFD starts the steady state conditions.
For unsteady state case, As Dr. Ricardo Vicente de Paula Rezende explains... We write the initial values in the INITIALIZATION step. CFD Code use these values for its 1st time step then 2nd time step use 1st time time step value. We can say, rth time step use (r-1)th time step values. So, If the initial values are changed, 1st time step is changed then all following values of time step are change progressively..
I do not agree with Ricardo Resende because you are solving the Reynolds Averaged Navier- Stokes equations (RANS) with a turbulent model using a high dissipative solver - FLUENT. It is very, very unlikely to find chaotic solutions within this framework.
@João C Henriques, thanks for the response. I will check it out.
Thanks & Regards,
Rajesh.
@Ricardo Vicente de Paula Rezende , thanks for the response.
Thanks & Regards,
Rajesh.
@Muhammad Furqan Ali, thanks for the esponse.
Thanks & Regards,
Rajesh.
@Abdul aziz Jaafar, i am investigating a jet flow near to a wall. So the velocity field is the one i am interested in after convergence.
Thanks & Regards,
Rajesh.
It depend on your problem, even in steady states problem such as natural convection with Nu number bigger than one and high Ra number; your initialization can change the results for example the velocity streamlines vector direction can change, and it is possible to have multiple results for a steady state problem.
@Hadi Kafil, thanks for the response. Could give me a little deeper explanation. I am confused, how can a steady state problem have multiple solutions.
Thanks & regards,
Rajesh.
For example in natural convection within the closed domain, The direction of yourvelocity streamlines vectors can change by your initialization.
Could you please tell me: what is your problem? and what kind of differences do you observe?
- Badges
- Science topic
- Similar topics
- Engineering
- Mechanical Engineering