MICRESS Forum

Hello,

I am simulating peritectic solidification in Fe-C alloys. I have a serious problem which is formation of melt pool (I think they are numerical errors and unphysical) in the middle of growing delta-ferrite dendrites. these melt pool changes peritectic transformation rate considerably. I tried to eliminate them by using finer meshes and changing interface mobility value but it did not work. Would you please help me with that and tell me what is the problem?

Thanks

Dear Ghavam,

Formation of liquid inclusions during dendritic growth is a typical sign that the interface gets numerically unstable, so that broad diffuse parts are formed which afterwards lead to entrapped liquid. This can be due to poor grid resolution, but also to unsuitable values of other parameters like interface energy, anisotropy, interface mobility, etc., or a combination of those. It would be much easier if you could share your input file, perhaps there is an "out of the range" parameter which can easily be detected. Otherwise, you should try to exclude potential reasons one by one by:

• using "mob_corr" together with a sufficiently high value of the interface mobility for assuring correct diffusion-limited growth kinetics

• using "metallic" anisotropy with small coefficients (<0.3) for kinetic and static anisotropy

• checking that the δ/liq-interface energy has a reasonable value in the order of 1.E-5 J/cm2

• avoiding large differences between the melt diffusion coefficients of different elements as well as the use of the full diffusion matrix

• avoiding too high initial undercooling

Bernd

Thanks Bernd,

I try to solve the issue using your instructions. At the same time would you please take a look at the input file I attached?

Thanks again.

Dear Ghavam,

The interface mobility of all interfaces is obviously too large, if you do not use mobility correction (which automatically reduces it to a value which corresponds to diffusion limited growth for the given numerical interface thickness). This problem leads to spreading of the interfaces and subsequent trapping of liquid phase. However, it should work fine if you just keep the mobility values and use mobility correction:

• append "redistribution_control" in the line after the keyword "phase_interaction" for each interface between different phases (phase interactions 0/1, 0/2, 1/2 in phase interaction data)

• insert the line
  
  normal mob_corr
  
  at the end of each phase interaction in "Phase diagram input"

This should do the trick.

Bernd

Thanks Bernd,

I almost resolved the issue. But have a little bit more problem. Just a quick question. Does have interface thickness any effects on liquid inclusion formation? if yes how does it affect it? it is better to have more grids in interface or less?

Thanks,

Hi,

the bigger the interface in absolute values, i.e. the coarser the resolution and the bigger the number of cells in the interface, the more probable is a liquid inclusion. This numerical effect is most frequently observed between the secondary branches (in 2D-simulations).

You should not go below 2.5 cells for the interface thickness.

Bernd

Thanks Bernd,
Your advices were really helpful and I resolved the issue.

Just a general question. The mobility between phases (with unit of cm⁴/JS) and Anisotropy of interfacial mobility are materials property or numerical parameter?

How you know the values used for them are exact/correct. what is the logical way to determine them?

Thanks a lot

Hi,

For the interface mobility the logical assumption for all solid-liquid interfaces and also for many solid-solid interfaces is "diffusion control". That means that the physical interface mobility is so high that it does not matter anymore. By using "mob_corr" on a arbitrary high physical mobility, a numerical value is created automatically which fits to that condition (see .mueS output).
If "diffusion control" cannot be assumed (e.g. for highly ordered intermetallic phases where reordering inside the interface is slow), you need to know a reasonable value either from atomistic simulations or from experiments. It makes sense still to use "mob_corr" to include the numerical effects of the finite interface thickness. I don't know any other rules or methods to make estimates.

For anisotropy values things are even more difficult. In principle, they also can be obtained from atomistic simulations or experiments, but thin interface correction ("mob_corr") leads to some intermixing between the kinetic and the static anisotropy which creates additional uncertainties. I don't know any theory for that...

Bernd