While working with a CFD problem, Iíve followed two different approaches in specifying a Pressure Boundary condition. In the first case Iíve set the Boundary pressure to 0 and the reference pressure to the Boundary absolute pressure (= 1 atm.) and in the second case Iíve done it the other way around.
For the purpose of having a mere observation of the convergence trend in both the cases, Iíve set a very loose residual target of 5.E-03 R.M.S. In the first case, the solution converged very quickly within two minutes and in 5 iterations. In the second case the solution wasnít converging even after 300 iterations so I stopped the run and by the time the run was stopped, the residuals were around 1.8E-02.
The convergence history in the first case was zigzag but in the second case it is almost a straight line. Attached images can throw a better illustration.
The following are the questions Iíve in relation to this description:
1. In spite of having high residuals, at around 1.8E-02 the post processed
results of the second case are found to be more accurate than those
of the first case with low residuals at around 5.E-03. what can be the
reason for this?
2. In specifying a pressure boundary condition what is the recommended
approach to follow, the first (or) the second please clarify upon this.
re: pressure being set in two different ways.
As far as I can tell these are identical setups so ought to give identical results?
How are you certain that your solution has converged? Just because you have met the residual target you have set does not mean that the solution has converged. For example, wall shear at a certain point might still be changing rapidly even though your RMS residuals are "converged". You have to check whether certain monitor point values are close to constant to get an idea as to whether your solution has converged or not.
In the second case where you can't get your residuals to do anything other than flat line, have you tried altering the time step of the simulation?
I didn't try such steps because I was working upon a electronics cooling problem which is indeed one of the ANSYS work shops so I already have the correct result for temperatures in the simulation from the presentation. I was only trying to do some simple experiments to understand the trends of convergence and more over I was saying that I ran the case 1 with residual target of 5.E-3 but later on when I ran it for a residual target of 1.E-04 everything went well and I got the results that exactly matched with those given in the presentation. My problem is not that I don't have accurate results. And in the second case with a flat trend of convergence, I also checked the domain imbalances and found them to be convincing . I was only asking that if it is reasonable to stop the run if a flat trend as seen in the second case is encountered.
Then if you say that both the approaches for specifying a pressure boundary condition do the same job then why is it that the convergence histories are that very different. When I ran the case 1 for a residual target of 1.E-4 I didn't see any flat trend it monotonically converged. I know that both the approaches serve the same purpose of setting up the desired absolute pressure at the boundary. I was asking a suggestion for a better approach so that there can be an ease in getting a good trend of convergence and in most of the tutorials I found that the first approach, i.e.., the approach of setting up a 0 pressure at the boundary and the Boundary absolute pressure as the reference pressure was recommended. I asked the question only to confirm upon that.
This is very basic numerical methods. Simply when you set the reference pressure correctly (your first case) then the pressure is resolved more accurately by the numerical precision of the solver. For instance 100001 Pa abs and 100002 Pa abs is resolved as 1Pa rel and 2Pa rel, a factor of 2 different.
In the second case you are loosing precision because the difference between 100001 and 100002Pa is about 0.001% and this is challenge for any numerical system to accurately resolve.
Hence case 1 converges easily and case 2 converges slowly, if it converges at all.
If case 2 is more accurate than case 1 then that is just luck. You obviously have large errors in your analysis (even case 1) and should do a lot of work to improve accuracy.
|All times are GMT -4. The time now is 13:40.|