Problem with convergence in heat transfer problem

[maheshh]

What are the probable reasons for a heat transfer problem (using only SOLID70 elements) not converging?

I have created a very detailed structure of semiconductor die (resolution: few micron level details included), thermal interface material and lid (dimension: @37.5 mm square). I have heat flux BC on the die and convection on top of lid surface (trying to keep it simple).

Solution:
Steady State Analysis
The solver starts off with "Nonlinear Solution"
I am using Sparse direct solver
Newton-Raphson option: FULL

ANTYPE,STATIC ! Steady-state analysis type
EQSLVE,SPARSE
NROPT,FULL ! Program-chosen Newton-Raphson
TUNIF,300 ! Uniform starting temperature
AUTOTS,ON ! Automatic time stepping
NSUBST,2 ! Number of substeps
KBC,0 ! Ramped loading (default)
SOLVE
FINISH



I have tried upto 50 substeps. But the solution does not converge. I am using the default convergence criterion.

I keep on getting a message:
"There are 'n' small equation solver pivot terms. May be a transitional effect caused by the N-R logic."

I know this warning can be ignored. So not worried about that.

But why does the solution not converge?

Could this be because there is a huge variation in terms of element size as I go from die details to lid tessellation? I thought this would not be a problem because I am solving for only TEMP dof.
OR
Is it because of the huge difference between thermal conductivities of different materials?

Any hints?

Regards

Mahesh Hardikar

 

[Drej]

Why is the analysis non-linear? Do you have temperature dependent material properties for example? If not, remove these items - the items telling ANSYS that you wish to perform a non-linear analysis - from your input deck and just have:

ANTYPE,STATIC ! Steady-state
TUNIF,300 ! Uniform starting temperature
SOLVE
SAVE
FINISH

Best of luck,

-- drej --

 

[maheshh]

Yes, properties are temperature dependent.
Apparently, if I use a very small element size, then solution has a hard time converging. I doubled the element size, and the solution converged in 2 substeps.

Can anybody think of any particular reason for this?

Regards,

Mahesh.

 

[slava74]

Thermal analysis are almost always non-linear. IMHO, double sizing just eliminates non-linear effects, especially if you are dealing with radiation problems.