Convergence parameter CFX Solver

chiragsvnit · March 16, 2014, 12:34

Hi,
Few Days back I have started a steady state simulation (Hexahedral mesh, node = 35 million) of a Penstock using ANSYS CFX Solver. Boundary condition are mass flow inlet (34000 kg/s) and pressure outlet (225 kPa absolute). Convergence criteria was 0.001 MAX. Residual for all
I have observed following points during the simulation.
(1) First I set automatic timescale, the solution was stabilized at 0.01 MAX Residual and automatic time scale was 0.7 s.
(2) Then I selected Physical time scale of 0.1 s the solution converged below than the automatic timescale residual value but not converged.
(3) Further I was continuously changed the physical timescale to 1E-9 s, then immediately the maximum residual reached to 0.001 and the solution converged.

The point is selected physical timescale was far below than the actual residence time in the domain. The residence time was 151 s (Checked in CFX Post streamline Timescale Min=0 and Max=151 s).

Mass imbalance <0.01%.

Is there something wrong with the solution since residence time (i.e 151 s)is far away from the selected timescale (i.e. 1E-9 s)?

ghorrocks · March 16, 2014, 18:00

Yes, I would be cautious of this result. When you need to use a microscopic time scale to get it to converge you can get into positions where the residuals appear tight but convergence has not been achieved.

If you have to use such a small time scale for convergence I would look at your simulation and improve the numerical stability - that means things like better mesh quality, double precision numerics or better initial condition.

chiragsvnit · March 17, 2014, 01:45

Thank You Glenn for suggestions.

(1) Quality of the hexahedral mesh is above 0.7 in ICEM CFD.
(2) For initial guess, I used result file which was obtained based on automatic timescale.
(3) It is always double precision, high resolution for all solving equations including turbulence numeric (k-epsilon).

More importantly, When I run this simulation as transient using time step of 0.001 s, it converges fast with just three coefficient loop iterations. I do have limitations of RAM I could not continue with transient simulation and switched to steady state.

(1) Will another turbulence model make difference in this case, i.e. k-omega
(2) Will I try another boundary condition, i.e. Pressure inlet and pressure outlet

March 16, 2014, 12:34	Convergence parameter CFX Solver	#1
chiragsvnit New Member Chirag Trivedi Join Date: Sep 2009 Location: Norway Posts: 26 Rep Power: 17	Hi, Few Days back I have started a steady state simulation (Hexahedral mesh, node = 35 million) of a Penstock using ANSYS CFX Solver. Boundary condition are mass flow inlet (34000 kg/s) and pressure outlet (225 kPa absolute). Convergence criteria was 0.001 MAX. Residual for all I have observed following points during the simulation. (1) First I set automatic timescale, the solution was stabilized at 0.01 MAX Residual and automatic time scale was 0.7 s. (2) Then I selected Physical time scale of 0.1 s the solution converged below than the automatic timescale residual value but not converged. (3) Further I was continuously changed the physical timescale to 1E-9 s, then immediately the maximum residual reached to 0.001 and the solution converged. The point is selected physical timescale was far below than the actual residence time in the domain. The residence time was 151 s (Checked in CFX Post streamline Timescale Min=0 and Max=151 s). Mass imbalance <0.01%. Is there something wrong with the solution since residence time (i.e 151 s)is far away from the selected timescale (i.e. 1E-9 s)?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
How to check the convergence of the CFX solver?	Amoul	CFX	13	June 17, 2024 19:55
Convergence	Centurion2011	FLUENT	48	June 15, 2022 00:29
Error in CFX Solver	Leuchte	CFX	5	November 6, 2010 07:12
Convergence Definition of Segregated Solver	Amir	FLUENT	0	April 18, 2006 07:21
Convergence with coupled implicit solver	Henrik Ström	FLUENT	1	October 29, 2005 04:57

March 16, 2014, 18:00		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,854 Rep Power: 144	Yes, I would be cautious of this result. When you need to use a microscopic time scale to get it to converge you can get into positions where the residuals appear tight but convergence has not been achieved. If you have to use such a small time scale for convergence I would look at your simulation and improve the numerical stability - that means things like better mesh quality, double precision numerics or better initial condition.

March 17, 2014, 01:45		#3
chiragsvnit New Member Chirag Trivedi Join Date: Sep 2009 Location: Norway Posts: 26 Rep Power: 17	Thank You Glenn for suggestions. (1) Quality of the hexahedral mesh is above 0.7 in ICEM CFD. (2) For initial guess, I used result file which was obtained based on automatic timescale. (3) It is always double precision, high resolution for all solving equations including turbulence numeric (k-epsilon). More importantly, When I run this simulation as transient using time step of 0.001 s, it converges fast with just three coefficient loop iterations. I do have limitations of RAM I could not continue with transient simulation and switched to steady state. (1) Will another turbulence model make difference in this case, i.e. k-omega (2) Will I try another boundary condition, i.e. Pressure inlet and pressure outlet