Fighting Heat conduction convergence with multiple metal type
Conductive heat transfer containing two or more different metals (Aluminum and Steel) with tetrahedral cells (equangle skewness less than 0.7) caused severe convergence problem . I could not get that case converged and the heat transfer rate (Report --> Flux) did not balance (100% difference).
1) For Heat Conduction calculation for two or more metal taype (no convection) with tetrahedral cells :
Double precision is a must.
Double precision +Node based = reasonable convergence (1% imbalance)
Double precision + W-cycle multigrid = Perfect convergence (0% imbalance)
Double precision + W-cycle + Node-based = Perfect convergence (0% imbalance)
2) With hex cells, one may achieve reasonable convergence with 1% heat transfer imbalance without using W-cycle multigrid and/or Node based solver option. For better convergence, one must use node-based option or w-cycle multigrid
Details of the tests for this case are given below.
The I tried double precision solver with 1st order upwind differencing scheme and this certainly somewhat helped to achieve heat transfer rate balance. However, the residual values went up to 10e-4 for energy and heat flux balance did not reach to an acceptable value (50% difference (0.95 to 0.56 ).
Then I turned on the Node Based option on the Define --> Model --> Solver panel. This dramatically lowered the residual values. This led to a good heat transfer balance (from 0.56 to 0.06).
Now for this case with node based and 3ddp selections, I switched to W-Multigrid (from V). This lowered the residual values even further (10e-12) and got a perfect heat transfer balance (10e-8)
I have also found that if you use double precision with W cycle multigrid (with out node based option turned on), the convergence is also perfect with very good heat transfer rate balance. Node based option is not essential in this situation.
I tried 3d single precision solver with W cycle (no Node based option) and the convergence is slightly better but the heat transfer imbalance is significant (dropped from 0.95 to 0.76-75%)
Then I tried, 3d solver (single precision) with W-cycle and Node based option. Again, this did not help the convergence.
Hence, you must use double precision solver.