Writing an expression based on residual values?

Trev0r · April 15, 2015, 21:58

I'm modelling a complicated system with various pressure feedback loops at the outlets:

Pressure@Outletx=Constant*MassFlow@Outletx

It also contains a porous domain with a directional pressure loss model.

I am really struggling to get a fast convergence, using the autotime scale option results in no convergence. I need to initially apply a very small physical timescale ~10^-5 or a local timescale factor ~0.01 to get a stable decreasing residual with each iteration. Unfortuneately these methods are taking up to 4000 iterations (24 hours) to complete!!!!!!

My model contains over 2 million elements and has a very well refined mesh.

The interesting part for me was that I noticed that as the pressure residual became progressively smaller, these timescales could be increased, rapidly decreasing solution time. By manually sitting at my computer and upping the timescales over a two hour period as the pressure residual decreased I brought the total number of iterations needed for solution down from 4000 to 800!

So the next logical step would be to write an expression in CEL that defines the timescale as a function of the pressure residual. I have previously written timescale functions based on the iteration number (citern) however this is inefficient in the case of restarts and requires an assumption for how the pressure residual will decrease over time (I generally have no idea).

MyTimescale=( $\frac{Constant}{Pressure Residual From Previous Iteration}^{n}$ )

Is there anyway to do this or can anybody think of another way to lower the number of iterations I need to get a converged solution. As of now solution is 'achieved' when all RMS residuals are at 10^-6.

ghorrocks · April 16, 2015, 03:55

it does not surprise me that feedback loops like this are unstable and require fine time steps, but as convergence is achieved the time step can be made larger.

You can access the residuals as CEL expressions but I am not sure it will give you the last time step value. To hold the last time step value you might need to use a user fortran routine.

SGCam · May 30, 2018, 16:12

On that note (and being late to the party) how does one access the current residuals via CEL? I have yet to find a variable or monitor probe that works.

ghorrocks · May 30, 2018, 20:05

You can add the residuals to the results file in the results file tab in CFX-Pre. Then the residuals field is available in CFD-Post and you can do operations on it there.

I recall you can access it via a CEL variable but cannot remember exactly how. Something like u vel.residual or something like that.

April 15, 2015, 21:58	Writing an expression based on residual values?	#1
Trev0r New Member Andrew Norfolk Join Date: Mar 2015 Location: Sheffield Posts: 5 Rep Power: 11	I'm modelling a complicated system with various pressure feedback loops at the outlets: Pressure@Outletx=Constant*MassFlow@Outletx It also contains a porous domain with a directional pressure loss model. I am really struggling to get a fast convergence, using the autotime scale option results in no convergence. I need to initially apply a very small physical timescale ~10^-5 or a local timescale factor ~0.01 to get a stable decreasing residual with each iteration. Unfortuneately these methods are taking up to 4000 iterations (24 hours) to complete!!!!!! My model contains over 2 million elements and has a very well refined mesh. The interesting part for me was that I noticed that as the pressure residual became progressively smaller, these timescales could be increased, rapidly decreasing solution time. By manually sitting at my computer and upping the timescales over a two hour period as the pressure residual decreased I brought the total number of iterations needed for solution down from 4000 to 800! So the next logical step would be to write an expression in CEL that defines the timescale as a function of the pressure residual. I have previously written timescale functions based on the iteration number (citern) however this is inefficient in the case of restarts and requires an assumption for how the pressure residual will decrease over time (I generally have no idea). MyTimescale=( $\frac{Constant}{Pressure Residual From Previous Iteration}^{n}$ ) Is there anyway to do this or can anybody think of another way to lower the number of iterations I need to get a converged solution. As of now solution is 'achieved' when all RMS residuals are at 10^-6.

May 30, 2018, 20:05		#4
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	You can add the residuals to the results file in the results file tab in CFX-Pre. Then the residuals field is available in CFD-Post and you can do operations on it there. I recall you can access it via a CEL variable but cannot remember exactly how. Something like u vel.residual or something like that. __________________ Note: I do not answer CFD questions by PM. CFD questions should be posted on the forum.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
buoyantSimpleFoam and watertank	Tobi	OpenFOAM Running, Solving & CFD	100	December 18, 2022 08:15
Extrusion with OpenFoam problem No. Iterations 0	Lord Kelvin	OpenFOAM Running, Solving & CFD	8	March 28, 2016 11:08
Wrong fluctuation of pressure in transient simulation	caitao	OpenFOAM Running, Solving & CFD	2	March 5, 2015 21:33
Simulation seems to converge but crashes suddenly	xxxx	OpenFOAM	16	September 12, 2014 08:07
SLTS+rhoPisoFoam: what is rDeltaT???	nileshjrane	OpenFOAM Running, Solving & CFD	4	February 25, 2013 04:13

April 16, 2015, 03:55		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	it does not surprise me that feedback loops like this are unstable and require fine time steps, but as convergence is achieved the time step can be made larger. You can access the residuals as CEL expressions but I am not sure it will give you the last time step value. To hold the last time step value you might need to use a user fortran routine.

May 30, 2018, 16:12		#3
SGCam New Member Cam Demler Join Date: May 2018 Location: Ohio, USA Posts: 1 Rep Power: 0	On that note (and being late to the party) how does one access the current residuals via CEL? I have yet to find a variable or monitor probe that works.