Hi all,
I'm doing a transient simulation of flow past a rectangular cylinder close to the free surface. As far as I'm concerned, the common practice is to discard the initial stages of the transient solution up to some point that a so-called statistically stationary state is believed to have reached. Then the solution should proceed with a time resolution much finer than the vortex shedding dominant period in order to ensure accuracy. From this point on, flow properties are captured over an interval sufficiently long to realize statistical convergence of the results.
In the literature, the above has often been implemented in a subjective manner, mainly through visual examination of the results.
I'm wondering if there are more standard objective methods of determining the portion of the data to be discarded, and the appropriate frequency and duration of sampling.
I appreciate any comments.
Generally a good quantity to assess that the flow is well correlated is the monitor of the total kinetic energy in time. I suggest also to monitor separately its three components. You will see a final state in which the oscillations are around a steady averaged value.
Other than the good answer above, a common way to quantitatively check the solution convergence is to look at the cylinder forces. You can look at the the drag force over time and stop the solver when the instantaneous drag deviated less than 0.1% from the time averaged value. Cheers.
- Badges
- Science topic