Accessing OpenFOAM shared memory

jbcourt · June 17, 2016, 06:42

I'm working to develop a two-way co-simulation between OF and another program, and I think that I've identified the use of shared memory as the best (most economical) means of communication between the two at run time. The other program simulates a system (with losses) that takes force and position output from the OF simulation and applies a user-identified control strategy to establish the resistance that the external system applies to any motion - this is then fed back to OF for the next time step. EDIT: The OF simulation is a numerical wave tank, with motion being calculated from wave-structure interaction.

What I wanted to ask is what would be the best way of implementing this from the OpenFOAM side? Should I be looking to customise a solver to include interaction with the external model; using the shared memory to provide information on the OF simulation to the external program and then using the output of that to update the OF simulation parameters (it would be a restraint in dynamicMeshDict that it would edit at each time step); or something else?

I've already managed to get a non-continuous version of this working, at least in principle, by using OF's system.call run-time function object to trigger execution of the external model - reading the necessary OF run-time postProcessing output - with a duration of the time step length and then using the output to write a restraint for #include-ing in dynamicMeshDict, but there are transient effects in the external model that are left unaccounted for between time steps by this approach.

This is my first post on here, so sorry if it's at all vague or confusing. I should point out that I'm still learning cpp and am relatively new to OF, so it could well be the case that there's something obvious that I'm missing.
Thanks in advance for your advice!

gkarlsen · June 19, 2016, 04:51

There is new functionality for co-simulation in OpenFoam 3.0. It sounds like that would be suitable for your case?

http://openfoam.com/version-v3.0+/so...-co-simulation

Unfortunately, I have never used it myself... but maybe you will find some usefull pointers in the tutorial they refer to?

jbcourt · June 20, 2016, 06:43

Thanks Geir, I've had a bit of a look at the externalCoupled boundary condition mentioned on that link: it seems to me that it would work well for passing temperatures at a boundary/patch (as done in the tutorial that is used there). Similarly that would work well if I wanted the other model to take, as an input, the pressure or a velocity component at a patch. However, I'm looking to use the force on a body within the simulation as the information transferred between the two - since OF already calculates the force on that body due to the pressure difference between two patches I'm loath to get my external model to replicate that postprocessing itself.

That being said, one very good thing that I've noticed about the externalCoupled BC is that it gets OF to write a lock file while it's calculating, so I could use that functionality to make the other model wait for OF to finish doing its bit before running. That would at least allow me to get a form of synchronised interaction going, with the other application reading the forces.dat postprocessing file that OF produces to sort out the input that it needs.
Thanks again for your suggestion.

June 17, 2016, 06:42	Accessing OpenFOAM shared memory	#1
jbcourt New Member James Bridgwater Court Join Date: Jan 2016 Posts: 14 Rep Power: 10	I'm working to develop a two-way co-simulation between OF and another program, and I think that I've identified the use of shared memory as the best (most economical) means of communication between the two at run time. The other program simulates a system (with losses) that takes force and position output from the OF simulation and applies a user-identified control strategy to establish the resistance that the external system applies to any motion - this is then fed back to OF for the next time step. EDIT: The OF simulation is a numerical wave tank, with motion being calculated from wave-structure interaction. What I wanted to ask is what would be the best way of implementing this from the OpenFOAM side? Should I be looking to customise a solver to include interaction with the external model; using the shared memory to provide information on the OF simulation to the external program and then using the output of that to update the OF simulation parameters (it would be a restraint in dynamicMeshDict that it would edit at each time step); or something else? I've already managed to get a non-continuous version of this working, at least in principle, by using OF's system.call run-time function object to trigger execution of the external model - reading the necessary OF run-time postProcessing output - with a duration of the time step length and then using the output to write a restraint for #include-ing in dynamicMeshDict, but there are transient effects in the external model that are left unaccounted for between time steps by this approach. This is my first post on here, so sorry if it's at all vague or confusing. I should point out that I'm still learning cpp and am relatively new to OF, so it could well be the case that there's something obvious that I'm missing. Thanks in advance for your advice! Last edited by jbcourt; June 17, 2016 at 12:02.

June 19, 2016, 04:51		#2
gkarlsen Member Geir Karlsen Join Date: Nov 2013 Location: Norway Posts: 59 Rep Power: 14	There is new functionality for co-simulation in OpenFoam 3.0. It sounds like that would be suitable for your case? http://openfoam.com/version-v3.0+/so...-co-simulation Unfortunately, I have never used it myself... but maybe you will find some usefull pointers in the tutorial they refer to? elvis likes this.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Getting Started with OpenFOAM	wyldckat	OpenFOAM	26	June 21, 2024 07:54
Map of the OpenFOAM Forum - Understanding where to post your questions!	wyldckat	OpenFOAM	10	September 2, 2021 06:29
OpenFOAM v3.0+ ??	SBusch	OpenFOAM	22	December 26, 2016 15:24
twoPhaseEulerFoam error: "new cannot satisfy memory request."	basiratshamsa	OpenFOAM Running, Solving & CFD	1	November 7, 2015 15:52
OpenFOAM virtual memory error	spoon	OpenFOAM Running, Solving & CFD	1	July 12, 2013 08:24

June 20, 2016, 06:43		#3
jbcourt New Member James Bridgwater Court Join Date: Jan 2016 Posts: 14 Rep Power: 10	Thanks Geir, I've had a bit of a look at the externalCoupled boundary condition mentioned on that link: it seems to me that it would work well for passing temperatures at a boundary/patch (as done in the tutorial that is used there). Similarly that would work well if I wanted the other model to take, as an input, the pressure or a velocity component at a patch. However, I'm looking to use the force on a body within the simulation as the information transferred between the two - since OF already calculates the force on that body due to the pressure difference between two patches I'm loath to get my external model to replicate that postprocessing itself. That being said, one very good thing that I've noticed about the externalCoupled BC is that it gets OF to write a lock file while it's calculating, so I could use that functionality to make the other model wait for OF to finish doing its bit before running. That would at least allow me to get a form of synchronised interaction going, with the other application reading the forces.dat postprocessing file that OF produces to sort out the input that it needs. Thanks again for your suggestion.