I am conducting a CFD analysis of a train component (pantograph). So far, I have tried dozens of different meshes, simulation setups, and various numerical parameters. The best result I obtained is in the project below:
I am attaching the residual, force plot, and drag/lift coefficient graphs here for easy visibility. The y+ values are also within the required range. I have also changed “relaxation type” to “automatic”, otherwise I have fluctuations in the force plot.
The problem is that, based on my research on CFD forums and other sources on the internet, it is said that a steady-state CFD solution should converge with residuals at the level of 1e-6 to claim convergence. My residuals are higher than this level.
Later, I checked SimScale’s public projects, and I didn’t see residuals at the 1e-6 level in almost any single project. This has further confused me.
Perhaps this 1e-6 residual level is necessary for a project where the error/divergence should be almost zero. In this project, I am open to deviations of 5% to 10%, indicating that I am not striving for absolute precision, and variations within this range are acceptable.
I am currently trying to create a mesh with hex-dominant parametric to achieve lower residual values. I haven’t succeeded yet; creating a mesh with hex-dominant parametric in such a complex and angular geometry is quite challenging.
I eagerly await your ideas, assistance, and especially your comments related to my situation, especially regarding residuals.
Hello @yuno76264 , and thanks for posting your question in the Forum!
I would say desired residual level will be different in each case. In a perfect scenario, we would of course want residuals to be very close to zero, however most of the time this is not possible due to model complexity, mesh quality due to complex geometries, etc.
I would recommend the following:
Make sure that you have performed a mesh independency study. This will allow you to chose the most optimum mesh configuration considering the computational demand and accuracy. The following knowledge base article may be helpful.
Residuals only may not be enough to check the convergence of a simulation. One can also interpret the convergence of simulations by looking at result control monitoring plots. Please make sure you take advantage of them. More information can be found here.
In case there are any, make sure you validate your design by comparing your results with experimental data, or other numerical data with higher fidelity.
Hello @kaany , thank you for your quick reply! I still have a quick question especially about your second point.
-I did indeed a mesh independence study. A finer mesh gives almost the exact values with this one, also that did not help in terms of residuals. So, a finer mesh is not an answer here I believe.
-In the link you shared on the second point; abpout result control items; I did not quite understand that actually. It says:
" In an iterative solution, the calculation process starts with initial estimations of quantities (pressure, velocity, temperature, etc.). In every iteration, quantities are updated with respect to the previous iteration results. In a converged steady state solution, one should expect to see that those quantities come to a level and don’t change anymore."
But does that also apply to high residual situation which we have let say residuals above 1e-4? I mean, on this simscale documentation page, pressure values come to a level, yes, and does not change anymore. But we do not know the residual values here, right? Lets say for this example about pressure covergence, even we have a flat pressure graph like here, but residuals are approximately 1e-3 or 1e-4, then can we still say that these values are ok and converged?
Overall, your intuitions and comments are correct.
There may be situations where the residuals cannot be reduced further due to imperfect mesh configurations or non-matching boundary conditions. For instance, defined values on the inlet and outlet may not perform well in terms of mass balance; however, with velocity inlet and pressure outlet coupling, this should not be much of a problem.
Let’s assume that the residuals could only be reduced to 1e-3 and remained constant. This error may be so small that overall results, such as forces acting on a body, moments, temperature distribution, etc., are not significantly affected, and the deviation is below my acceptable level. In that case, I would say, in this case, this is acceptable after a careful mesh independence study, of course.
I hope this clarification addresses any doubts. I also recommend referring to some validation cases and tutorials to gain an understanding of which residual levels and result deviations can be considered acceptable.
