How Does Mesh Size Affect Heat Transfer Modeling in COMSOL?

[GT1]

I'm a student with no much experience with FEA and COMSOL. I'm trying to model the heat transfer from a heated cylinder. The cylinder has diameter of about 0.18 m and height of 0.36 m. The cylinder is hollow with holes from one side to the other side.
I'm interesting in finding the temperature of the cylinder on steady state.
my questions are:
1. How much elements the mesh should have? I just want to know the average temperature of the cylinder, I'm not interesting in some boundary effects or some thing like this. The mesh I have now has 170,000 elements. Is it enough for this kind of problem? When I try to model something, how do I know what is a reasonable size for the mesh?
2.Does the size of the mesh effects the accuracy of the whole solution? If a problem solved once with fine mesh and once with coarse mesh, would the average temperature of the whole object will be different?

 

[Mech_Engineer]

1. How much elements the mesh should have? I just want to know the average temperature of the cylinder, I'm not interesting in some boundary effects or some thing like this. The mesh I have now has 170,000 elements. Is it enough for this kind of problem? When I try to model something, how do I know what is a reasonable size for the mesh?

170,000 elements is too many if you're just looking for a simple tempertuare profile and aren't conncerned about boudnary effects. Without a picture of your geometry and mesh, I would think you could have a mesh of around 10,000 and very easily accomplish what you're wanting. Of course, if you want an average tempertaure you could just solve for it analytically, it sounds like a fairly straightforward problem.

As for how to know if the mesh is fine enough, the user has to excersise some engineering discretion. You decide based on what you know about the problem, where the areas of interest are, and what values you are expecting based on some simple analytical modeling or empirical testing. Usually experienced users know if a mesh is fine enough just by looking at it and knowing what is being solved for and how accurate the result must be.

2.Does the size of the mesh effects the accuracy of the whole solution? If a problem solved once with fine mesh and once with coarse mesh, would the average temperature of the whole object will be different?

It depends on the geometry, the problem description, and what you're solving for. Generally there is a method called a "mesh convergence study" where the mesh is incrementally refined and results are compared to see if their change is greater than a certain percent value. If the mesh is refined and nothing changes (or the changes are below some threshhold, say 5%) the mesh has "converged."

For a problem such as you're describing it, I seriously doubt a mesh convergence study will be required.

 

[GT1]

170,000 elements is too many if you're just looking for a simple tempertuare profile and aren't conncerned about boudnary effects. Without a picture of your geometry and mesh, I would think you could have a mesh of around 10,000 and very easily accomplish what you're wanting. Of course, if you want an average tempertaure you could just solve for it analytically, it sounds like a fairly straightforward problem.

Thanks Mech_Engineer!
170,000 elements is the minimum size for which the meshing succeed. I imported a CAD model (IGES) to COMSOL. The model is not a perfect cylinder, it has holes, small faces, short edges etc, and every time I tried to mesh it with coarser mesh, it failed.
Maybe I don't know how to mesh it right. Do you have meshing tips?

 

[Mech_Engineer]

Thanks Mech_Engineer!
170,000 elements is the minimum size for which the meshing succeed. I imported a CAD model (IGES) to COMSOL. The model is not a perfect cylinder, it has holes, small faces, short edges etc, and every time I tried to mesh it with coarser mesh, it failed.
Maybe I don't know how to mesh it right. Do you have meshing tips?

Hmm, well if the part has complex geometry like little holes and stuff, you'll have to keep the mesh where its at. It's important to keep in mind that if you don't really care about precise temperature distributions around certain geometry elements, you can help make the mesh simpler by removing those elements. Also, removing things like countersinks, chamfers, and radiused edges will help simplify the mesh dramatically.

A mesh with 170,000 elements may be too large for a "regular" computer to solve, it depends on how much ram and hard drive space you have available. Let me also stress again that you'll probably want to have some analytical calculations so you have a ballpark estimate of what the FEA answer will be.

 

[minger]

I agree with Mech, get yourself an analytic solution to start with. The solution to the heat equation for a cylinder is a Bessel function IIRC. You should be able to easily find an exact equation.

 

[Claws]

One of the things that characterizes FEA, is the need to simplify your model according to what you want to know. If the little holes have no influence on your final solution, simply leave them out. If you are interested in a simple cylinder, model the cylinder and leave everything else out :)

 

[Urmi Roy]

Can I just squeeze in a little question here?

In the 'extremely coarse' mesh in comsol, what is the mesh size? I need the mesh size details, and they don't seem to be mentioned in the free mesh parameters...please suggest.