Seed-density and seed-undercooling models

dendritic solidification, eutectics, peritectics,....
mauvec
Posts: 19
Joined: Mon Mar 15, 2021 3:38 pm
anti_bot: 333

Re: Seed-density and seed-undercooling models

Post by mauvec » Wed Jun 30, 2021 12:19 pm

Dear Bernd,

Thanks for your prompt response. I have removed the line "ordered 2 1 2" from the input file and simulation is running. In the meantime, please find attached in a pm the input file.

Many thanks,

Mauro

Bernd
Posts: 1504
Joined: Mon Jun 23, 2008 9:29 pm

Re: Seed-density and seed-undercooling models

Post by Bernd » Wed Jun 30, 2021 6:18 pm

Dear Mauro,

I did not yet get the pm with the restart file - where did you send it?

Bernd

mauvec
Posts: 19
Joined: Mon Mar 15, 2021 3:38 pm
anti_bot: 333

Re: Seed-density and seed-undercooling models

Post by mauvec » Wed Jun 30, 2021 9:38 pm

Dear Bernd,

Thank you for your message. I was sure that I clicked 'submit' but it seems that the message did not go through. I have sent it now.

I have just checked one of the running simulations after removing 'ordered 2 1 2" but still getting,

--> Force automatic start values
initialisation failed, error = 2 interface FCC_L12/FCC_L12#2

Thermo-Calc error 1611 MICRESS error 20 phases 1/ 2

Many thanks,

Mauro

Bernd
Posts: 1504
Joined: Mon Jun 23, 2008 9:29 pm

Re: Seed-density and seed-undercooling models

Post by Bernd » Wed Jun 30, 2021 11:24 pm

Hi Mauro,

Just got your file. I will have a look tomorrow...

Bernd

Bernd
Posts: 1504
Joined: Mon Jun 23, 2008 9:29 pm

Re: Seed-density and seed-undercooling models

Post by Bernd » Thu Jul 01, 2021 7:35 pm

Dear Mauro,

I had a look on your input file, and the main problem seems to be the low resolution of Δx=0.1µm for rather typical SLM conditions. This leads to a very low effective interface mobility (through the action of the "mob_corr" option which avoids trapping of the pile-up in front of the dendrites by introducing effective kinetics), so that the dendrite tips get more and more undercooled. At the same time, they get completely diffuse due to the lack of resolution. I guess that especially the highly metastable system may lead to the problem that fcc gets ordered or γ' disordered...

I think what you need is at least a factor 4 of resolution (i.e. 400 x 2000 cells with Δx=0.025µ) to get things working reasonably. To get into a more comfortable situation I would advice you furthermore to increase the interface energy of the 0/1 interface (5.E-6 J/cm2 seems unrealistically small to me), which increases curvature undercooling and coarsens the microstructure (thus being equivalent to a further increase of resolution), and increasing the stabilisation parameter to 10 times the value of the interface energy (to avoid spreading of the interface due to the still low grid resolution).

On the other hand, you are using local updating of thermodynamic linearisation parameters, which is extremely time consuming in case of a 10-component superalloy. Essentially, in your current setup, you use 99% of the calculation time just for that. Using averaging instead ("global", "globalG", or "globalGF") avoids most of this effort, thus compensating greatly for the higher resolution which is absolutely required. Averaging linearisation data over the interface region is not optimal (especially in view of the high temperature gradients typical for additive manufacturing processes), but still the best choice available right now. We are already working on an intermediate option for more localized averaging, which will be available with the next release.

Bernd

mauvec
Posts: 19
Joined: Mon Mar 15, 2021 3:38 pm
anti_bot: 333

Re: Seed-density and seed-undercooling models

Post by mauvec » Wed Jul 14, 2021 8:16 pm

Dear Bernd,

Thank you for all your comments and feedback. I apologise in advance for spamming this forum with this long message but many comments and questions resulted from the previous discussion. Regarding the interface energy, as a first approach, I estimated this physical value through the property model calculator of Thermo-Calc, i.e. from 6.9E-06 to 7.8E-06 J/cm^2 within the temperature range from liquidus to solidus. When using individual values from this interval, I experienced issues concerning the stability of the interface, mainly when varying distance between predefined positions of gamma seeds and process conditions. Therefore, may I ask you about any previous experience when using this TC module? I could not find additional information on the forum.

The following captions correspond to a series of simulations when utilising higher values such as a) 1E-04, b) 7E-05, c) 5E-05, d) 3E-05 and e) 1E-05 J/cm^2 for the 0/1 interface with a grid size of 0.01 micrometres. It is worth to mention that same re-linearisation option (‘local’) and process conditions were implemented for all these.


a) 1E-04 J/cm^2
micress-1.png
a) 1E-04 J/cm^2
micress-1.png (96.75 KiB) Viewed 2049 times

b) 7E-05 J/cm^2
micress-2.png
b) 7E-05 J/cm^2
micress-2.png (85.26 KiB) Viewed 2049 times

c) 5E-05 J/cm^2
micress-3.png
c) 5E-05 J/cm^2
micress-3.png (88.97 KiB) Viewed 2049 times

d) 3E-05 J/cm^2
micress-4.png
d) 3E-05 J/cm^2
micress-4.png (84.39 KiB) Viewed 2049 times

e) 1E-05 J/cm^2
micress-5.png
1E-05 J/cm^2
micress-5.png (94.27 KiB) Viewed 2049 times

I read in the forum that ‘holes’ are caused by unstable interfaces and therefore, the interface energy shall be calibrated. Would be correct to say that after these trail-and-error simulations, 5E-05 J/cm^2 might be the most suitable value for the 0/1 interface?

I recently ran two additional simulations varying the re-linearisation options as a) 'globalF' and b) 'local'; but utilising a higher interface energy for the 0/1 interaction, i.e. 7E-05 J/cm^2 (and stabilisation parameter of 7E-04 J/cm^2). Also, I increased the resolution by a factor of 10 (i.e. 1000 x 5000 cells) and implemented ‘globalG’ for the solid/solid phase interactions. The following caption depicts the outcome of these simulations.
XZ-01.png
XZ-01
XZ-01.png (247.2 KiB) Viewed 2049 times
As it can be seen, no qualitative difference could be considered significant, however, an evident dissimilarity is observed when examining one of the solute profiles at the primary dendrite arm for both simulations (these curves correspond to the same time-step),
XZ-01-profile.png
XZ-01-profile
XZ-01-profile.png (71.28 KiB) Viewed 2049 times
Based on these quantitative results, I am wondering if it would be reasonable to keep implementing ‘local’ re-linearisation over the interface region to achieve 'higher' exactness?

In my current setup, I utilise 5E-06 J/cm^2 because this value renders a good qualitative agreement (see the following caption) on the experimentally measured primary arm spacing . However, I am still doubtful if this calibrated value, which is even smaller than those estimated via TC, would make sense to approach 'realistic' results.
XZ-02.png
XZ-02
XZ-02.png (124.75 KiB) Viewed 2049 times
Finally, as previously described, with the implementation of a higher resolution (1000 x 5000 cells), ‘globalF’ and ‘globalG’ accordingly; I am still generating error = 40200, but also additional messages and errors that I did not encounter previously. May I ask you about description of these and how to approach them?


A)

Fast moving fragment(s) relinearized LIQUID/FCC_L12 (7)

Fast moving fragment(s) relinearized LIQUID/FCC_L12#2 (18)

Fast moving fragment(s) relinearized LIQUID/FCC_L12#3 (11)

Fast moving interface(s) relinearized FCC_L12/FCC_L12#2 (3)

Fast moving interface(s) relinearized FCC_L12/FCC_L12#3 (3)



B)

Phase 2 disappeared at 1.2001872E-03 s


C)
Checking for nucleation Nucl.Ph: 2 error= 40200

Thermo-Calc error 1611 MICRESS error 20 phases 1/ 2


--> Force automatic start values

initialisation failed, error = 2 interface FCC_L12/FCC_L12#2



D)

--> Force automatic start values

# Error number 40200 in Interface 0/ 393

# time: 1.100000000000000E-003 s

# Serious error in linearisation!

trying hard phases 1 2 level: 4 zp= 99257 error= 30201

trying harder! Error = 30201

trying even harder! Error 30201


--> Force automatic start values

--> Force automatic start values

--> Force automatic start values

--> Force automatic start values

# Error number 105 in Interface 1/ 393

# time: 1.100000000000000E-003 s

# Serious error in linearisation!

Updating of diffusion data from database...

MICRESS error 24 phases 2/ 2



E)

Updating of diffusion data from database...

MICRESS error 22 phases 2/ 2

Diffusion data not updated in phase FCC_L12#2

MICRESS error 22 phases 2/ 2

Composition dependence temporarily disregarded in phaseFCC_L12#2 / Constituent-1

MICRESS error 22 phases 2/ 2

Composition dependence temporarily disregarded in phaseFCC_L12#2 / Constituent-2

MICRESS error 22 phases 2/ 2

Composition dependence temporarily disregarded in phaseFCC_L12#2 / Constituent-3

MICRESS error 22 phases 2/ 2

Composition dependence temporarily disregarded in phaseFCC_L12#2 / Constituent-4

MICRESS error 22 phases 2/ 2

Composition dependence temporarily disregarded in phaseFCC_L12#2 / Constituent-5

MICRESS error 22 phases 2/ 2

Composition dependence temporarily disregarded in phaseFCC_L12#2 / Constituent-6

MICRESS error 22 phases 2/ 2

Composition dependence temporarily disregarded in phaseFCC_L12#2 / Constituent-7

MICRESS error 22 phases 2/ 2

Composition dependence temporarily disregarded in phaseFCC_L12#2 / Constituent-8

MICRESS error 22 phases 2/ 2

Composition dependence temporarily disregarded in phaseFCC_L12#2 / Constituent-9

complete relinearisation!

trying hard phases 3 0 level: 3 zp= 101937 error= 104

trying hard phases 3 0 level: 3 zp= 100137 error= 104

Thermo-Calc error 1611 MICRESS error 14 phases 0/ 3

trying hard phases 2 0 level: 3 zp= 98255 error= 104

trying hard phases 2 1 level: 3 zp= 98255 error= 30201



F)

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Warning in spfAddPhInkr, t=0.1100002E-02

Wrong fractions at nTupelp= 1735

sum = 3.000000

x,y,z = 57 1 99

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!


!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Warning in spfAddPhInkr, t=0.1100002E-02

Wrong fractions at nTupelp= 1736

sum = 3.003789

x,y,z = 58 1 99

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!


!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Warning in spfAddPhInkr, t=0.1100002E-02

Wrong fractions at nTupelp= 1737

sum = 3.037011

x,y,z = 58 1 100

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!


!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Warning in spfAddPhInkr, t=0.1100002E-02

Wrong fractions at nTupelp= 1738

sum = 3.000000

x,y,z = 58 1 98

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!


!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Warning in spfAddPhInkr, t=0.1100002E-02

Wrong fractions at nTupelp= 1739

sum = 3.000000

x,y,z = 59 1 99

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!


Thanks for taking the time to read this.

Best regards,

Mauro

Bernd
Posts: 1504
Joined: Mon Jun 23, 2008 9:29 pm

Re: Seed-density and seed-undercooling models

Post by Bernd » Thu Jul 15, 2021 4:52 pm

Hi Mauro,

This indeed is a quite long series of questions, but don't worry about "spamming" - your questions and my answers to them hopefully provide useful information for other users, which can be searched systematically, and thus do not harm to anybody.

Let me try to answer in the order of your questions. The last part about the MICRESS Messages will soon be done in a separate new thread.

1.) Honestly, I also do not have experience with using interface energies σ from Thermo-Calc. Certainly, it is a great source of information if no other data is available. However, we should not expect too much from this type of model and rather consider it to represent a lower limit of the real value.

2.) You compared the effect of the interface energy σ on the morphology of dendrites in a cross-section model (similar to the Application Example A006_CMSX4_dri). This comparison is not really useful for "calibrating" interfacial energies, because
  • interface energies are physical constants. They should generally not be determined by numerical stability criteria - even if sometimes chosing a "convenient" value may be alluring...
  • the comparison is restricted as the primary distance is kept constant (by the initial positions of the dendrites) but should change with σ
  • with the interface energy σ the initial undercooling (resembling the tip undercooling of dendritic growth) should also change correspondingly, which certainly was not the case
  • the final shape of each dendrite cross section is practically fixed also, and the results you show are already close to this shape
You should rather try to calibrate σ by other means (see 4.)

3.) Using "global" or "globalF" for relinearisation of the liquid/fcc interfaces is only a crude approximation, if averaging is performed across strongly different concentrations or temperatures. Thus, this method comes to its limits in cases where a large interval of temperatures is simultaneously within the simulation domain and is connected to the same grain ("global") or phase ("globalG") interface. This is especially true in case of fast processes like SLM or welding in combination with alloys with large solidification range (e.g. Ni-base superalloys), if simulations are performed in longitudinal section. For me, this inaccuracy seems more easily visible in the shifted position of the dendrite tips than in the comparison of concentration profiles, but this may depend on how representative the line scan can be made.

Using the fragments option ("globalF" or "globalGF") sometimes may be helpful, if interfaces are interrupted by coalescence or precipitation of interdendritic phases. Then, the range of averaging is significantly reduced, and accuracy is improved.

Using "local" linearisations is only an alternative for simple (binary or ternary) alloy systems. For multicomponent systems like ni-base alloys this is typically not possible for performance reasons.

These limitations, especially for simulation of additive processes with high temperature gradients and ranges, lead us to the development of a series of new options for Version 7.1, including a distance option which limits the averaging range to a region with given radius.

4.) Calibration of σ should not be done using the primary spacing (PDAS), because determining the PDAS via phase field simulation is not trivial. The reasons are a large stability range, especially in 2D, the history dependency of the PDAS, as well as the problem of having sufficient individual dendrites for reliable statistics. Instead, using the secondary arm spacing (SDAS) is much more promising.

5.) The meaning of the MICRESS messages will be explained in a separate thread.

Best wishes
Bernd

Post Reply