Determining Ideal Mesh Size -- CFD Online Discussion Forums

RaiderDoctor

Senior Member

Join Date: Dec 2016

Posts: 152

Rep Power:

10

Rep Power:

Short answer, yes, mesh size is really important and there are a few equations (like in turublence, the calculation of the y+ value, and the Flow Courant Number) that you can use depending on your flow conditions to maintain a good equation. However, these will only give an approximation and cannot be used to give the perfect mesh the first time. Iteration must be used to determine the ideal mesh.

Long answer, let's say that you mesh your fluid domain with a really coarse mesh and run your simulation. You get your results, and they don't really match up to your hand-calculated values (if you can hand-calculate), or they just don't make sense. The next step would be to make your elements smaller, AKA increase your mesh resolution. You run your simulation again, and you get in the ball park of where you're supposed to be. Congrats, you have a decent solution. Now, this does not mean this is right.

There are a lot of numerical errors that are introduced in CFD due to mesh size. Sometimes there are a lot of elements where nothing particularily interesting is happening in the fluid, and not enough where there's this awesome vortex going on. And obvious solution would be to make a super-fine mesh (>1 billion elements) that can capture everything in the domain perfectly. However, this is not computationally efficient, nor is it even possible. Furthermore, there's a relationship between the size of your mesh and how nicely your solution converges (coarse mesh usually leads to better convergence, fine mesh usually leads to divergence). The goal is to fine the happy medium.

So, back to your decent solution. How exactly can engineers trust these results? Easy, we prove that the mesh size is adequate to give our desired solution, while not being too computationally expensive. To do this, we run a mesh sensitivity analysis. To do this, we start out with a mesh size that we think is good, and then refine it. Then, we either run the simulation again or, if the simulation takes a long time, we run a steady-state simulation at the most chaotic time point/conditions. We then export a global value (like velocity along a set line in each domain, or pressure along the exterior of the surface, etc.) and compare it to the first mesh. Remember that numerical error I mentioned earlier? Well, this will prove its existence, as the two values will not be the same. Then, you take a coarser mesh and run it again, and compare the values. Is the percent difference between the coarse mesh and the original mesh under what you think is acceptable (< 5%)? What about from the original to the fine mesh? Can you justify the length it took to run the fine mesh over the original mesh?

For extra references, check out (

Hey, welcome to the world of CFD.Short answer, yes, mesh size is really important and there are a few equations (like in turublence, the calculation of the y+ value, and the Flow Courant Number) that you can use depending on your flow conditions to maintain a good equation.these will only give an approximation and cannot be used to give the perfect mesh the first time. Iteration must be used to determine the ideal mesh.Long answer, let's say that you mesh your fluid domain with a really coarse mesh and run your simulation. You get your results, and they don't really match up to your hand-calculated values (if you can hand-calculate), or they just don't make sense. The next step would be to make your elements smaller, AKA increase your mesh resolution. You run your simulation again, and you get in the ball park of where you're supposed to be. Congrats, you have a decent solution. Now, this does not mean this is right.There are a lot of numerical errors that are introduced in CFD due to mesh size. Sometimes there are a lot of elements where nothing particularily interesting is happening in the fluid, and not enough where there's this awesome vortex going on. And obvious solution would be to make a super-fine mesh (>1 billion elements) that can capture everything in the domain perfectly. However, this is not computationally efficient, nor is it even possible. Furthermore, there's a relationship between the size of your mesh and how nicely your solution converges (coarse mesh usually leads to better convergence, fine mesh usually leads to divergence). The goal is to fine the happy medium.So, back to your decent solution. How exactly can engineers trust these results? Easy, we prove that the mesh size is adequate to give our desired solution, while not being too computationally expensive. To do this, we run aTo do this, we start out with a mesh size that we think is good, and then refine it. Then, we either run the simulation again or, if the simulation takes a long time, we run a steady-state simulation at the most chaotic time point/conditions. We then export a global value (like velocity along a set line in each domain, or pressure along the exterior of the surface, etc.) and compare it to the first mesh. Remember that numerical error I mentioned earlier? Well, this will prove its existence, as the two values. Then, you take a coarser mesh and run it again, and compare the values. Is the percent difference between the coarse mesh and the original mesh under what you think is acceptable (< 5%)? What about from the original to the fine mesh? Can you justify the length it took to run the fine mesh over the original mesh?For extra references, check out ( https://caeai.com/blog/how-do-i-know...sh-good-enough ), ( Grid sensitivity analysis ), or just google "mesh sensitivity analysis". As for the quantities I mentioned earlier, feel free to look them up if you like, but I'm not sure they would benefit you at the moment.

chek321, Sai Krishna, fedenr and 2 others like this.