CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums

what is under relaxation factor?

Register Blogs Members List Search Today's Posts Mark Forums Read

Rate this Entry

what is under relaxation factor?

Posted August 27, 2012 at 09:23 by kira

something to keep in mind:
Quote:
Originally Posted by Phil
;140717
Each of the variables(mass,mom,density) represents an equation the solver is trying to solve. Each iteration the values obtained for the variables should get closer and closer together - converge. FOR SIMPLE PROBLEMS ESPECIALLY COLD FLOWS WITHOUT COMBUSTION YOU SHOULD SIMPLY KEEP THE RELAXATION FACTORS AT DEFAULT.

Sometimes for many many reasons the solution can become unstable so a relaxation factor is used - takes part of value from previous iteration to dampen solution and cut out steep oscillations.

If you are having convergence trouble start the solution on default then when it starts going to shit(becomes unstable) put pressure 0.2 momentum 0.5 turbulence KE 0.5 turbulence DR 0.5 - this should be in the manual. This should sort out most issues much worse and you need a better mesh or something isn't correct.

RELAXATION FACTORS MAKE SOLUTION TAKE ALOT LONGER TO CONVERGE SO ONLY USE WHEN YOU REALLY NEED TO.

Generally start off without RF's then when solution becomes unstable later on bring them in where needed - whichever equations (residual graph) are unstable, meaning not a nice smooth line but up and down rapidly. Start off moving down from 1 to 0.8 or 0.8 to 0.6 etc. For energy equation start off with 0.9 it takes hundreds or thousands of iterations to converge with relaxation factors in.

ALWAYS REMEMBER--------- ALWAYS START WITH DEFAULTS. ONLY WHEN SOLUTION BECOMES UNSTABLE DO YOU RELUCTANTLY LOOK AT THE RESIDUALS AND BRING IN APPROPRIATE RF's.

hope this helps
Posted in Uncategorized
Views 577 Comments 1 Edit Tags Email Blog Entry
« Prev     Main     Next »
Total Comments 1

Comments

  1. Old Comment
    In my experience if you have to reduce RF's to very low values then it indicates meshing issues or the wrong model (i.e. turbulence model for laminar flow etc.). I agree that the normal procedure is to start the solution using the default values and allow the solution to evolve without trying to force convergence which can produce meaningless results.

    However I have heard a user state that for very large models such as 10 million or 20 million fluid cells that you need to lower the URF's to very low values, but I totally disagree with this. The convergence should not depend on the model size, but for a given number of iterations required to converge with default URF's it will take more CPU time for a larger model, correct?

    William
    permalink
    Posted January 11, 2013 at 17:27 by wxmoore wxmoore is offline
 

All times are GMT -4. The time now is 14:32.