using fieldAverage library to average postprocessing
 

Dear programming experts,

I am stuggelling with the following: I want to obtain the average of the fields already create in the time directories (i.e., during postprocessing).

During runtime, this can be done via the functions in the controlDict by adding libfieldFunctionObjects.so

No I want to do the same as a postprocessing step.

My starting point was the postChannel application, that loops over the fields and collaps the field data that is assemed to by homogeneous. This tools requires Umean to be present. I want to create the Umean fields by a similar postprocessing tool called for instance postAverage

This new tool (if it has been developped yet, please tell me) should read the controlDict dictionary, and than read the averaging information

I took out all the functionality of postChannel (the collapsing of the fields) and I changed the line

with the while runTime.loop construction, the controlDict is read and therefore the library is included.

However, if I run this I get the complaint From function Foam::fieldAverage::initialize()
in file fieldAverage/fieldAverage/fieldAverage.C at line 102.

Of course this makes sence, because I do not do an explicit creation of the fields.

Now the simple solution would be: explicity create the field U in the postAverage.

However, this would make the utility less generic. I want it automatically to create the field that are defined in controlDict function fieldAverage.

My idea was to include the fieldobject and reuse the code for creating fieldAverageItem. Unfortunately, I can not add or reuse the library in an external code, and therefore I can not reuse this functionality.

I am relatively new to C++, but I that the power of it lies in reusing code, therefore I am quite sure this must be possible (I moreover, is desirable)

Hopefully there is an OpenFoam programming expert out there who could give me a hint how to do tackle this problem.

Or, which would be even better: is there somebody of has written a postprocessing utility to do the averaging of the time steps over a certain time range?

Any hints very much appreciated!

Regards,

Eelco

well,

It seems there are not many users dealing with the same issue.

Let me refrase the my quest:

I want to write a program that loops of the time directories, reads the fieds, and than writes an average field in the last directory.

In this way, averaging can be done after the run as a postprocessing step, so that the first time steps can easily be left out if the start up is not stationary yet.

Ideally, I would like to use for instance the existing foamCalc utility and combine that the the fieldAverage library so that during the loop the controlDict is used to do the job. I dug into the code, but can not find how I should modify it. I am too inexperience in C++, so I leave this to the real programming experts maybe with the next release (I can imagine that I am not the only user who'd like to have this functionality)

Today I have spend my whole day digging into all the utilities to try to come up with my own solution. It is an ugly solution, in terms of proper C++ programming, but it is a start, and it works!. I am adding the code here. I would appreciate very much if somebody could comment on the most ugly construction I have used (see comments in the code). Probably there are better ways, but I just can not find how at the moment

It is all based on a combination of foamCalc and the vorticity utility. Hopefully somebody can make suggestions to get this a cleaner C++ code. However, it does exactly what I want!

for instance

foamTimeAverage Ux -time 300:

gives the time average Ux_mean field from the time steps 300 to the end.
Please free to use it or come with improvements

Regards

Eelco

If anybody is interested in post-processing field averaging (I can hardly imagine nobody ever needs it), here an improved version thanks to a hint of Bernard Gschaider I was able to force to read the fieldAverage library in the controlDict using the runTime.functionObjects().execute() statement. In this way, all the functionality of the library is available.

The code of postAverage.C looks like
and requires an include file createFields.H to read the field you want to average, for example
Using this routine allows you to read the given fields over a given time range and average them as a post-processing step, like

postAverage -time:300

The field averaging parameters should be given in the controlDict. The only important thing is that the outputControl is set on timeStep, e.g
I have one question, hopefully somebody is able to answer it. In the current code, the fields you want to average are given in createFields. Once you have compiled postAverage the code will always try to read all these fields, even if you only want to average say the velocity field. Much better, of course, would be if the fields to be averaged are read from the controlDict in advance, and the reading takes place based on the fields defined in the controlDict. I am sure that is possible by using the loop forAll(faItems_, fieldI) as can be found in the fieldAverage library. I have no idea, however, how to do this. I should probably define an instance of the class fieldAverage and use the method to read the dictionary, and then apply the loop to read the appropriate fields. I am unfortunately too inexperienced in C++, so I hope that somebody can a hint.

Thanks

Eelco


ps:
the code is attached below

I'm searching for this utility function. I'll try it firstly.

Hi Eelco,
thanks a lot for this very helpful tool

/Andreas

Hi guys

I want to do field averaging of Reynolds stress tensor (R) during running my code. It seems that the fieldaveraging library do the averaging just for U and P. How can I do the same thing for R?
I would be so appreciated if you can help me

Thanks

Dear eelcovv,

<If anybody is interested in post-processing field averaging (I can hardly imagine nobody ever needs it), here an improved version thanks to a hint of Bernard <Gschaider I was able to force to read the fieldAverage library in the controlDict using the runTime.functionObjects().execute() statement. In this way, all the <functionality of the library is available. >

Thank you very much for the very important code you supplied here. But when I execute the code postAverage (including the code i added in the system/controDict file), i realize that nothing changed. Can you give me a hint? Should a new variable be generated Umean or the averaged calculation is still saved as U?

thank you once more!
Alex

Sorry for the late reply, I only see this post now.

Please make sure that you have in you field average dictionary

outputControl outputTime;

and make sure that you write interval is equal to the write interval of the original simulation. If that is not the case, nothing is written indeed.

Good luck

Regards
eelco

thank you very much. You were very helpfull!!

Thank you!
 

This tool is very helpful indeed. Thank you very much, Eelcovv!!
:)

hi, the unility is very helpful, and I want ask what is UPrime2mean, the computational formula of it and transient k is the same, but do they the same in real physical meanings?

My utility just makes a call to the averaging libraries developed by Bernard sgnaider, so don't pin me on it. But I think uprime2mean does the following. The velocity in general can be writen as U=<U>+u', where <...> indicates the ensemble average and u' is the deviation on it. The purpose at the start of the calculation is to calculate <u'^2>=<(U-<U>)^2>, i.e. the <u'^2> mean which is indeed related to k. During the run, however, <U> is not known, because you still need to average it in time. Therefore, it does not give you <u'^2>, but <U^2>. Both of them are related according to <u'^2>=<(U-<U>)^2>=<U^2> - <2*U*<U>> + <U>^2 = <U^2>-<U>^2. In other words, you can obtain the <u'^2> by subtracting the Umean squared from the Uprime2mean

Thank you for the code . Yeah, I also believe that this feature is well needed and also wonder why there is not such a tool in OpenFoam.

Could you explain how it is determined where the averaged field is written to? The code iterates over all my time steps but the effect of setting outputInterval to　2, 3, 10, 100, 1000 does not yield my desired result. (which is: save in every 2nd time step folder, every 3rd, 10th, 100th, 1000th)

Basically, I just want to have the averaged field written into the last timestep folder.

Hi hanzo,

without testing it I would say play with

writeControl adjustableRuntTime;

writeInterval 100;

Where 100 now is not a interval in iterations but in time step and take you last time step only. The averaging is done by fieldAverage functionobject list, so the setting is control by this lib
good luck

Hi Eelco,


Thanks for the code. I tried running it many times with the instructions in this thread - but the code reads the fields at all the times and doesnt save the mean data anywhere. Am I missing something?

Here is my controldict file:

Can you please tell me what is wrong here? Is this declaration in Controldict all I need or do I need any other file too?
Thanks a lot. I have a lot of data to be averaged :confused:

Well, the code works, I can confirm that. But I also realised some strange (or not straightforward for me to understand) behaviour of how the output is written. At the moment I do the following:

Set
outputInterval 1;
or
outputInterval 2;

in fieldAverage1.
Then output is written to every / every second data folder. Then I copy the Mean files from the last timestep to some other place. The next step is to erase all *Mean files in the data folders.

When you have to do it only once or twice, I think it is okay to live with it.

execFlowFunctionObjects
 

Thank you very much for this tool, it works well for my purposes.

I have a question however. Can the same thing be done with the 'execFlowFunctionObjects' utility? Is there an advantage to this code, or do they work about the same?

If anyone knows the clarification would help. Thanks!

Hi eelcovv

Thank you very much for the code, which is really handy and useful.

As Hanzo mentioned, I just wonder whether it is possible for the code to write the Mean field at the very last time step? I am having a bunch of data (about 50,000 fileld files), it will take a lot space to write the Mean field out for every time step.

Thanks

jiejie

Hi eelcovv

Just got a quick question regarding to calculating <u'^2>. There are 6 values in the UPrime2Mean, which are xx,yy,zz,xy,yz,xz. Since the postAverage UPrime2Mean gives <U^2> instead of <U'^2>. Should I use the first 3 columns from postAverage's UPrime2Mean takes away UMean^2 to get the <U'^2>?

Assume the above was correct, I calculated the <U'^2>. However, I found it looks almost the same as UMean instead?

My other question is whether the UPrime2Mean is the variance of the velocity field? ( Please see this link: http://www.cfd-online.com/Forums/ope...n-etc-les.html)

Many thanks.

jiejie

If my calculations are right then UPrime2Mean should exactly be what it says:
U-prime, squared and averaged = <u'^2>. To check this, you can compute u' using U and <U> , calculate u'^2 and average it manually. I did this for a series of data sets and when I compared UPrime2Mean with manually generated <u'^2> they turned out to be the same.

Another hint is the magnitude. In my computations, the biggest components of UMean_X are around 0.97 and UPrime2Mean_XX of 0.0025.
UPrime2Mean cannot be U^2

Hi Hanzo

Thanks for your quick reply. I was not saying UPrime2Mean is U^2. I calculate the UPrime and found it is the same as sqrt(UPrime2Mean). I got confused as eelcovv hinted " you can obtain the <u'^2> by subtracting the Umean squared from the Uprime2mean". I am not trying to offend anyone but find the right meaning of UPrime2Mean.

Thanks

jiejie

I am sorry if my comment let you feel like this. I did not want to say that you are offending anyone. It's just my raw english :o I am also highly interested in the correct meaning of UPrime2Mean.

How did you compute UPrime? Did you take U - UMean and averaged this over several time steps? Or for a single realization only?
Averaging this would give <u'> (should be a value close to zero, right?). And then you compared to sqrt( UPrime2Mean ) (which is related to the rms value sqrt(<u'^2>) according to my understanding).
If you are right then UPrime2Mean gives <u'>^2.

Could you describe a little bit more in detail what you calculated?
I really hope that UPrime2Mean_XX corresponds to <u_x'^2>,
UPrime2Mean_YY to <u_y'^2> and so on. If not I have to recheck quite some of my results :D

Hi Hanzo

I thought I might upset eelcovv as I was confused with his hint :D. Anyway, I actually output the flow field for every single time step and use the postprocessing utility sample to probe the velocity at a few locations. Then, I calculate the the veocity RMS by its definition, which should be the UPrime^2. I did this a while ago, I think the RMS value is very close to the sqrt of UPrime2Mean. That's why I am thinking UPrime2Mean is giving the correct result.

jiejie

Hi jiejie,

You are correct, my hint in an ealier post is misleading. The uprime2suare is already correct for U^2.
So if you define U(t)=<U>+u'(t) then the uprime2square constains the in this order:

<u'^2> <uv> <uw> <v'^2> <vw> <w'^2>

no need to subtract anything. Forget my remark about that.

If you want to plot for instance the turbulent kinetic energy k over a line sampled from you uprima2square you can just do

plot 'sampledline_with_UPrime2Square' u 1:(0.5*($2+$5+$7))

cheers

Hi eelcovv,

Thanks for the utility, it is really helpful!
However I am not getting the results that I would expect.
I have two time directories (0 and 90) and I want to get the average of the U field. Running the application I get the Umean file but the results are not correct,
i.e. U in 0 : (0.1107 -0.0027 0.0006),
U in 90: (0.0036 -0.1216 -0.0004)

and the average I get is: (0.1017 -0.0126 0.0005).

Could you give me any hint of what it can be wrong?
Thanks!

My controlDict:

snappyHex,

I know this is an old thread, but I am trying to understand what's going on with the postAverage utility as well. Did you ever resolve the problem?

I tried to replicate your results, but I instead get
(0.075 -0.04233333333 0.0002666666667)
in 90/UMean. Strange.


I tried creating a single-cell mesh with a uniform velocity of (1 0 0) in timestep '0', and a uniform velocity of (2 0 0) in timestep '1'. Rather than a result of (1.5 0 0), I get (1.333333333 0 0)!

I have attached the case for reference (see README file).

Edit: This seems to be the same problem Yann was having:
http://www.cfd-online.com/Forums/ope...tml#post384964

I think the key lies in the 'uniform/time' files saved in each timestep directory; the information in these files (e.g. deltaT) determines how the averaging is performed! Removing all 'uniform' directories via 'rm -r */uniform' (be careful) changes the results for me. However, it's still not quite right.

For now, I am using ParaView's 'Temporal Statistics' filter and saving the resulting data to VTK (Save Data -> .vtm filetype).

postAverage 'bug' fix
 

All,

EDIT: the original code for postAverage seems to work fine on OpenFOAM/2.2.0/051613. I was having problems with OpenFOAM/2.1.x/071612. If in doubt, check your setup with the case I posted above or something similar.

I have found a solution, but so far it only works if your data is saved at consistent time intervals (1, 2, 3 etc., not 1, 3, 4, 4.5 5.1...).

The utility was counting values from the first timestep twice when computing the average. So, for instance, the average of '1' and '2' was (1 + 1 + 2)/3 = 1.3333 instead of (1 + 2) / 2 = 1.5. This may or may not be a big deal; if you are averaging a lot of data with a small standard deviation, you won't even notice. However, if you are trying to average only a few time directories, it can make a big difference.

I have fixed the problem by subtracting one from the startTime index 'timeI'. You should also clear or move your '(time)/uniform' directories, or else the '(time)/uniform/time' files will mess up the average. Perhaps changing the functionObject options in 'system/controlDict' would make this unnecessary and allow for averaging data saved at inconsistent time intervals, but I have not tried this yet.

Here is my change to postAverage (see bolded line):
This is giving me the results I expect. If someone else knows a better way to fix this problem, please chime in!

Good luck,
-Nuc

Hi everybody !

Thanks eelcovv for this very usefull tool !

However I am now trying to use it with a dynamic mesh case and I am facing some difficulties.
In order to make it work with a mesh changing over the time I've added #include "dynamicFvMesh.H" in the top of the postAverage.C file, I've switched #include "createMesh.H" to #include "createDynamicFvMesh.H" and I've added the line mesh.readUpdate(); in the time loop.
(see the file attached)

When I run the tool it seems to do the calculations but at the end it gives me completely absurd results.

Anybody knows how to make it work correctly with a dynamic mesh ?

Thanks in advance

Hello

Were you able to write the data only at last time step

unable to run the code
 

sir,
plz tell me how to run the given code and how to get time averaged pressure or velocity for the whole domain in a dat file.

The standard fieldAverage utility in OpenFOAM4.0 with the command "simpleFoam -postProcess" can also do this postAverage job.

time average over the all time steps stored in the case file
 

Dear friends
I want to calculate the average of a variable like a tau over the time steps saved in the case file, and finally, get a Tau_mean file that is the time-average of all stored time steps in the last time directory. How can I do this with the postAverage tool?

Extract Data from field average
 

Hi,

I have stored Tmean from 500 seconds to 1500 seconds using the field average function object. But, in the time directories, I have stored only last 50 time steps using the purge write option in the control dict.

Can I use this utility to extract field (Tmean) from 1000 seconds to 1500 seconds using post processing. Please note that I have time directories stored from 1450 to 1500.

Really appreaite if anyone can comment.