about convergence
 

Hi: I am modeling a minisize centrifugal pump.I set the boundary conditions as following :pressure inlet and outlet.I select segregated,RNGk-e model,standard wall function.Now I meet convergence problem.I think my mesh(unstructural )is good,the problem is that:when the residual reach 2e-3(the default is 1e-3),it never drops again.When I choose 1st order solution,it works well.But i think i shoud choose 2nd order and i don't want to low the under-relaxation( maybe I don't know how to change the under-relaxation appropriately) .Expecting your help!

regards

Re: about convergence
 

Why are you hesitating to give small the uner-relaxation as one of the possible approach ?

In general, residual can be decreased relatively easily with lower under-relaxation, even though not always so and even though it converges more slowly.

Sincerely, Jinwook

Re: about convergence
 

Hi Ling,

I also do think that when you switch onto the 2nd order one will have to go for lower under relaxation factors. After a considerable no. of iterations when you see that the residuals are coming down you can again gradually increase your relax params.

Another pt worth mentioning is: with 1st order you may not have much of a problem with residuals. You will not see a dramatic reduction of residuals with 2nd order. Moreover, residuals are not the only thing by which we can judge convergence. You can monitor some properties like pstat etc at some critical locns to see if there are changing substantially with iterns. Hope this helps.

Regards,

Sridhar

Re: about convergence
 

Hi Lingo,

Monitoring the mass imbalance would also be a good option in this case.Reducing urfs to some extent is fine.But giving an urf=0.01 will simply converge your solution which doesn't make sense.

sri

Re: about convergence
 

Thank you very much.I will follow your advice.Now I know I can monitor other properties like surfaces,forces and so on,but how can i judge convergence by them?Maybe my engish is too poor,so I want to know the meaning of "pstat" and "locns" in Sridhar's response ,"urfs" in sri' response.Expecting you help!

Re: about convergence
 

Hello lingo,

pstat: Static Pressure; locns: locations; urf's: Under Relaxation Factors.

For judging convergence while monitoring properties: Select some "critical" surfaces where you think certain properties may be varying/ are important from a physical point of a view, put monitors there. If you see that there is no variation (<1% or so) in those properties on the selected surfaces, you can more or less take that the solution is converged.

if you see that the residuals are widely oscillating then, more likely than not, either your Bound. conds. are not correct or your mesh quality might not be good enough.

Hope this helps. All the best.

Regards, Sridhar.