
[Sponsors] 
August 28, 2009, 06:41 
Under relaxation factor for external coupling

#1 
New Member
Join Date: Jul 2009
Posts: 16
Rep Power: 7 
Hi all,
In the solver control of external coupling, there should be a number for under relaxation factor, its default value is 0.75. Also the variables can be selected from the drop down list. But, what is the meaning of this under relaxation factor? Does it mean that the variables from ansys multibled by this under relazation factor are transferred to CFX? What about with a value 1.0 for this under relazation factor? Thank you for your attention! 

August 29, 2009, 11:01 

#2  
Senior Member
Join Date: Jul 2009
Posts: 210
Rep Power: 8 
Quote:


September 4, 2009, 20:59 

#3 
Senior Member
Join Date: Apr 2009
Posts: 511
Rep Power: 12 
This is a relaxation factor for the loads (force, displacement, heat flow or temperature) passed between ANSYS and CFX. The default means that 75% of the change in the load will be passed. A value of 1 means no underrelaxation  this is what I usually use.


September 7, 2009, 07:37 

#4  
New Member
Join Date: Mar 2009
Posts: 13
Rep Power: 8 
Quote:
When we leave the relaxation set to 75%, only 75% of the load will be transfered from CFX to ANSYS and viceversa, PER STAGGERED ITERATION? I mean, if I set lower relaxation factor, i have smaller load transfers, but at the end, after some staggered iterations, 100% of my load is transfered? Am I right? Is it ok to think that? Bigger under relaxation needs more staggered iteractions? If I use the relaxation parameter set to 1 (100%) I make my simulation more instable. And sometimes it crashes easier then setting underrelaxation, because the full load is transfered in the same iteration. Setting under relaxation, the load transfer happens gradually. Otherwise, if the relaxation limits my load to a percentage of the real load, the result of a simulation will never fit to reality. Thank you. 

May 14, 2010, 11:10 

#5 
New Member
Daniel Paukner
Join Date: Apr 2010
Posts: 17
Rep Power: 7 
Hi,
i am trying to perform a steady state aerothermoelastic simulation in a hypersonic flowfield. So far so good, BUT: In order to keep the FSI from crashing when commencing the second stagger iteration, i need to use very low under relaxation factors (down to 0.1 or 0.05) Now my problem is, that the results obtained don't seem right. My deformation with two way fsi is significantly smaller than with one way fsi (cht > structural) also, if i take the pressure and temperature distribution from a converged FSI and apply them as loads to a structural solution, i get more reasonable results, as the calculated deformation is bigger than with the chtstructural. So could someone confirm what the under relaxation factor does? Do i only transmit this percentage of the load itself or does it just affect the CHANGE in loads? I hope you can help me, Daniel 

May 14, 2010, 13:09 

#6 
Senior Member
Join Date: Apr 2009
Posts: 511
Rep Power: 12 
smagmon's assumption was correct. With 0.75, only 75% of the CHANGE in load is included PER STAGGER. So as you iterate within a timestep you get the full load. If you assume the load is constant within a timestep, then after 3 staggers you get:
1  (1  0.75)^3 = 98.4% of the change in the load. As you reduce the under relaxation factor you need to increase the number of staggers to get the correct solution. With a value of 0.1, even after 20 staggers you only get about 88% of the true change in load. Rather than using under relaxation, there's a much better way increase stability by using source term coefficients. It's a little involved, but Ansys support or their FSI training class would cover this. 

May 14, 2010, 23:36 

#7 
New Member
Daniel Paukner
Join Date: Apr 2010
Posts: 17
Rep Power: 7 
Thanks for the clarification stumpy.
Since i am transmitting forces and heat fluxes, i guess i would have to add source terms to the energy equation as well, right? I guess i'll just run with the long convergence for now, as i only have 30 days left and the simulations is very likely converge within this timeframe. But if i end up with some time left, i'll have a look at the source terms. At least it will make for a good proposition for further investigations Last edited by Pocket; May 15, 2010 at 02:53. 

May 17, 2010, 17:13 

#8 
Senior Member
Join Date: Apr 2009
Posts: 511
Rep Power: 12 
That's correct, you'd need source term *coefficients* (not actual source terms) for energy if the heat fluxes were giving convergence problems. Note that if heat fluxes were fairly stable and forces were not, then just increase the under relaxation factor for the heat fluxes only, then use a small relaxation factor (or source term coefficients) for forces.


August 29, 2012, 13:06 
How to tackle error in FSI convergence

#9 
New Member
abdul khader
Join Date: Dec 2010
Posts: 19
Rep Power: 6 
Dear stummpy, thanks a lot for making more clear clarifications in load transfer and relaxation factor.
I have a new prob. even after using timestep of 0.005 for total time of 0.8sec, and under relaxation factor of 0.2 with max. stagger iterations of 4 (load transfer comes to around 0.9984). When solu reaches around 0.72 sec, i get an error saying "Preconditional conjugate gradient solver error level 1, possibly the model is unconstrained or additional iterations may be needed. try running setting the multipier MULT on the EQSLC command to greater than 1(but less than 3.0)". Please suggest how to tackle such problems. 

September 13, 2012, 11:42 
Please help:

#10 
New Member
abdul khader
Join Date: Dec 2010
Posts: 19
Rep Power: 6 
Dear Stumpy,
I have been following many of the replies from your end with regard to applying source term @ interface to achieve convergence rather than using underrelaxation factor. When I couldnt achieve convergence using underlaxation factor, so i tried using mass flux pressure coefficient of 1e8. I used stagger iterations of 10 and mesh displacment coefficient of 10 and max coefficient of fluid flow as 5 and settling relaxation factor to 1. I got the complete fsi transient run possible after running steady state. I used this condition to a concentric stenosis. But when i use the same BC for eccentric type of stenosis, inspite of using several trial and error values of stagger iteration, mass flux cofficient etc. I am unable to get the convergance for steady state itself. Please help me out, @ the earliest. also suggest is it advisable to first set the stagger iterations and then slowy vary the mass flux coefficient or how this funda works. 

Thread Tools  
Display Modes  


Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
What is weakstrong coupling in FSI problems  hajo  OpenFOAM Running, Solving & CFD  5  May 15, 2008 01:45 
one/two way coupling of DPM  Angela  FLUENT  3  April 28, 2008 09:29 
Relaxation and convergence  sammi  Phoenics  0  March 20, 2008 04:32 
relaxation parameter  rvndr  Main CFD Forum  0  May 5, 2004 12:34 
Question on adjusting relaxation factor  CFD Rookie  Main CFD Forum  3  January 26, 2004 15:37 