I would appreciate any comments on the following problem. I need to calculate mean fields (e.g. pressure, velocity and vorticity), for which I activate Trn Stats tab in CFX-Pre. However, what I get is not as expected; the calculated Trnavg variables differ from one time step to another, as opposed to being independent of simulation time. I'm really confused about what these Trnavg (Arithmetic average) variables actually represent and how the solver calculates them.
In order to get time-independent statistical averages you need to run CFX for a long enough time interval with e.g. 5000 timesteps or even more. You need to specify a start and end time for the statistical averages, allowing first the transient flow to get to a fully-developed state in order not to contaminate the statistical averages with the start-up phase flow data.
Regards,
Dr. Th. Frank.
The Trn is used when you are interest on variables that change over time. Each trn file is the current status of simulation at the specific time step stipulated in the CFX-Pre. Only after the simulation is converged, the variables should no longer change.
Dear Frank and Willian,
Thank you so much for your helpful comments.
Let me describe the problem in more details. I'm doing a transient simulation of the flow past a blunt body. I'm interested in instantaneous as well as mean fields, for which I set the Transient Statistics tab in Output Control in CFX-Pre. I don't understand exactly how this option works. Without specifying start and end iteration lists in the Tranient Statistics tab, the average fields such as 'Trnavg Pressure' apprer to vary from one time step to another. After specifing Start Iteration List = 5 and Stop Iteration List = 10, I got a constant 'Trnavg Pressure' field as expected. But once I probed these field values in a given cell, I noticed that the mean pressure does not match the instantaneous pressure averaged between time step #5 and time step #10. I've verified this fact through manual calculation. Now I wonder how CFX solver actually calculates the mean (arithmetic average) values.
Armin Hajighasem Kashani : This is not in correspondence with my experience.
A constant, i.e. statistically reliable, average of a flow property in a transiently varying flow can only be obtained by carrying out a sampling and statistical averaging over a large number of transient timesteps. Large means here something in the order of several thousands of timesteps and not just 5. And the data representation of CFX should not be forgotten here, if a probe() function is used. CFX is storing variables in cell vertices, not in the cell center. This might lead to the fact, that the probe() function already leads to some interpolation from cell vertices to cell centers. Unfortunately you have not specified, how large the observed differences are with respect to the local variations in the pressure field from cell to cell.
Using the statistical averaging without specifying start and end timesteps is just erroneous usage. And at the start time of statistical sampling the flow development time should have been accomplished already. I have used CFX's transient statistics function for dozens of LES-type simulations to obtain the statistically averaged flow data with great success and in very good comparison to experimental data in various V&V exercises. From that I can say, that it works to 100% reliable, if applied correctly.
Best regards,
Dr. Th. Frank.
Thank you Thomas Frank for your comment.
I totally agree with you about the inadequacy of my current data length to be considered statistically reliable; The simulation must continue for a much larger number of time steps to ensure a quasi-steady state, and from that point on, averaging must be based on a sufficient number of time steps.
However, I can't understand what CFX is doing mathematically behind the scenes. That's why I did some post-processing using Tecplot. As you suggested, I probed the vertex of a given cell, rather than its centroid. To better demonstrate this, I created an animation of the results, which you can see in the attachment.
As shown in the animation, for some unknown reason, the average value calculated by Trn Stat function in CFX-Pre is far off the range of the corresponding instantaneous values (Note I set Start Iteration List = 5 and Stop Iteration List = 10). I am wondering how CFX calculates this.
Also attached is the relevant part of the CCL code for your consideration.
I would really appreciate your help.
Armin Hajighasem Kashani What is the result of TrnAvg.Pressure in CFD-Post (plotted from the RES-File)?
This is not obvious to see from your provided information. It should be the sum of the 6 individual pressure values (end of timestep value) divided by 6.
Regards,
Dr. Th. Frank.
Thank you Thomas Frank
You are right. I seem to have misunderstood to what time step indices the start/stop iteration lists correspond. So based on what I verified (see attachment), I recommend the following clarification to anyone facing the same confusion.
In Tran Stats tab in Output Control, let's say, you select Pressure and Arithmetic Average. Once you prescribe:
Start Iteration List = i
and
Stop Iteration List = j,
what CFX solver actually does behind the scenes is that, it returns Pressure.Trnavg = instantaneous pressure, for any time step index < i. For i+1
- Badges
- Science topic