Elmer Discussion Forum

Hello all!

This is a very general question, but still.

I am trying to benchmark a simple 2D Heat transfer model in Elmer against a commercial FEM Solver. If I impose constant boundary conditions and material properties (temperature independent), the results are very accurate (below 1%). But if boundary conditions are temperature dependent (Heat transfer coefficient, Density and Heat conduction), the results start to differ a lot (almost 30%).

So my question is: What is the best way (if possible) to increase accuracy of a thermal model?

I was trying to use different methods, both iterative(CG, BiCGstabl and ILU2) and direct (Banded). I also tried to increase convergence tolerances, but no success. Could higher mesh density improve the result? And structured mesh grid (I an currently using a mesh with triangles, not quadrangles). The error is very big so I am probably missing something.

Thanks in advance, any answer is appreciated.

Hi

Assuming that you have converged already alternative linear solver strategies make no difference.

Look at nonlinear iteration parameters: number of iterations, tolerance etc. This would probably explain differences resulting from nonlinear material parameters.

For discretization error you could try p-elements (Element = p:2), or increase mesh density.

-Peter

Hello Peter. Thanks for the reply.

I diminished the "Nonlinear System Relaxation Factor" from 1 to 0.5 as suggested in another threads and it works well now.

Hi,

i'm new in elmer so, i don't khnow a lot but i need this information quickly for big meeting
i want introduce equation in .sif ( heat flow is not constant) but my equation includes theta modulo 2pi for boundary condition
how i put it?????
thanks all

Hi,

see viewtopic.php?f=3&t=3237&sid=8495b41984 ... 663#p18391

You don't need to post the same question twice...

Matthias