liquid to gamma-alpha in a fast transition

[omid]

Dear Bernd,

I am trying to simulate the phase transition from the liquid to gamma+alpha.
the process is so fast (5000 K/s) and I do have 2000 K/cm TG. I start from high temperature which I do have 50% liquid at top and 50% solid at the bottom at first and then go to 1767 K and pause at this temperature for 6 ms, 114 ms and 234 ms. and investigate the fraction of Delta phase at the end. The hypothesis is that the holding time would give more time to carbon to diffuse, so it will make more chance for the Delta to nucleate! I came up with these problems:
- in the simulation with 6 ms pause time, I should not get Delta phase at the end or at least I should get very few amount of it because the process is so fast and the holding time is too few, I do get around 1 % delta during the pause time but afterward I reach to around 9 % which is obviously wrong!
- in the simulations with longer holding time I get up to 5 % delta phase fraction but in dissolves afterward!
Would you please have a look at the set up to find out why it happens? Your help is highly appreciated in advance!

[Bernd]

Dear Omid,

Can you please explain a bit more why you use such a simulation setup? What is the meaning of the strange initial grain distribution? Why do you start with FCC+LIQUID? What is the real process behind your simulation?
And can you send some pictures of your resulting microstructure? It takes quite a long time to get there.
By the way, the interface mobility between FCC and BCC is so high that the simulation must crash as soon as both phases get in contact...

Bernd

[omid]

Dear Bernd,

The reason behind choosing such a domain is the actual experiment condition. We have a domain almost 50% liquid at the top and the rest are solid at the bottom and at the interface there should be carbon movement. (It's just hypothesis!). I have uploaded the video of process progress for 6 ms but I changed the further nucleation temperature range manually so that the phase fraction is a bit more reasonable but as I said they are manipulated and forced to be so :/

Regards,
Omid.

[Bernd]

Dear Omid,

I found several issues with your setup:

1.) You cannot use periodic boundary conditions at bottom/top when you define a temperature gradient. This is inconsistent!

2.) The interface mobilities are not set correctly (too low for 0/1, 0/2, too high for 1/2). I strongly advice to use mob_corr instead.

3.) The nucleation setup is not optimal. The shield parameters which you use (0.1 s, 20 µm) completely suppress most of further nucleation. My advice is to use the seed_density model for bulk nucleation (which removes the necessity to control density by shields. Remember using different random numbers for the two types), and/or to use different shield groups (second parameter after shield time, allows decoupling shielding between different seed types for the same phase). Then you are more flexible to find suitable shield parameters. Once you are closer to local equilibrium (using mob_corr, see above), you can also adjust critical undercooling in order to prevent excessive nucleation. This is more physical than using shield parameters.

4.) Local relinearisation is extremely time intensive and probably "overkill". Instead I would propose "globalG" or "globalGF" (MICRESS 6.4) or "global/globalF".

Bernd

[omid]

Dear Bernd,

Many thanks for your helps and very constructive comments. I have tried to apply all of your points :
Boundary condition is revised.
mob_corr is considered along with global database for interfaces with "none" relinearisation.
for the further nucleation part I have switched off the interface nucleation and just have two bulk one but I am not sure whether have I defined it correctly or not?! Actually I have no idea about How many classes shall be chosen for the critical radius?!! Would you pls have a look below to see how does it look and pls make your comment on Shield parameters. I got totally confused why it is still running such a slowly! (the whole dri file is attached)

MFG,
Omid.


# Type of 'position' of the seeds?
# Options: bulk region interface triple quadruple [restrictive]
bulk
# Phase of new grains (integer) [unresolved|add_to_grain]?
1
# Reference phase (integer) [min. and max. fraction (real)]?
0
# Which nucleation model shall be used?
# Options: seed_undercooling seed_density
seed_density
# Integer for randomization?
546132
# How many classes shall be chosen for the critical radius?
0
# Determination of nuclei orientations?
random
# Shield effect:
# Shield time [s] [shield phase or group number] ?
1
# Shield distance [micrometers] [ nucleation distance [micrometers] ]?
5
# Nucleation range
# min. nucleation temperature for seed type 3 [K]
1700.000
# max. nucleation temperature for seed type 3 [K]
1768.000
# Time between checks for nucleation? [s]
# Options: constant from_file
constant
# Time interval [s]
5.00000E-04
# Shall random noise be applied?
# Options: nucleation_noise no_nucleation_noise
no_nucleation_noise

[Bernd]

Hi Omid,

Sorry for the late reply. I have been on a business trip...

In principle, it is up to you to choose the number of seed classes. However, it must be at least one (which is maybe also sufficient for your purpose), otherwise, you will get no seed at all! You can check in the screen output or the log file, how much potential seeds will be created depending on the density value which you give for each class.

If you use "mob_corr" for automatic diffusion limited choice of the interface mobility, please specify a high enough value in the phase interaction data (e.g. 1.0 for all phase interactions). If you already chose a too small value, it will not be increased by mob_corr but only reduced if it too high.

I cannot tell you without trying why it is still slow. Please check the .TabP output to see where most time is going to. I could imagine it is the too frequent calculation of diffusion coefficients from database - 1.E-4 as intervall is overkill for updating diffusion data.

Bernd

[omid]

Dear Bernd,

Many thanks for the reply. I wish you had a good trip.
I have been revised the number of classes already and get some seeds.Kindly have a look below. The speed is somehow acceptable right now but the problem is I get no nuclei growth!! When the simulation reaches to 1200 Celsius and all the domain must be solid, I get no further solids besides the initial grains! Do you have any idea why it is going so? I also have revised the phase interaction data based on your advice!

Thanks in advance,
Omid.

# Input for seed type 1:
# ----------------------
# Type of 'position' of the seeds?
# Options: bulk region interface triple quadruple [restrictive]
bulk
# Phase of new grains (integer) [unresolved|add_to_grain]?
1
# Reference phase (integer) [min. and max. fraction (real)]?
0
# Which nucleation model shall be used?
# Options: seed_undercooling seed_density
seed_density
# Integer for randomization?
546132
# How many classes shall be chosen for the critical radius?
4
# Specify radius [micrometers] and seed density [cm**-3] for class 1
0.0045 1e5
# Specify radius [micrometers] and seed density [cm**-3] for class 2
0.003 2e5
# Specify radius [micrometers] and seed density [cm**-3] for class 3
0.0025 5e5
# Specify radius [micrometers] and seed density [cm**-3] for class 4
0.0018 1e6
# Class 1: 1 seed(s), 3.7500E-03 < radii < 5.2500E-03 [micrometers]
# Class 2: 0 seed(s), 2.7500E-03 < radii < 3.7500E-03 [micrometers]
# Class 3: 1 seed(s), 2.1500E-03 < radii < 2.7500E-03 [micrometers]
# Class 4: 1 seed(s), 1.0000E-08 < radii < 2.1500E-03 [micrometers]
# Determination of nuclei orientations?
# Options: random randomZ fix range parent_relation
random
# Shield effect:
# Shield time [s] [shield phase or group number] ?
1 4
# Shield distance [micrometers] [ nucleation distance [micrometers] ]?
0

[Bernd]

Dear Omid,

You have specified now 4 classes of seeds for each seed type. However, the radii for the classes are very small (in the nm range) which leads to a very high corresponding critical undercooling of above 1000K. This is the reason why you do not get any nuclei.

Furthermore, the density (in particles per cm³) is so low, that you normally could not expect to have even a single seed particle in your domain. However, when converting the density to a 2D-density, MICRESS projects all particles of the third dimension which are within a thickness of the average particle distance into the plane. This is the reason why you get at least 1 particle in most of the classes...
So, you should definitively increase the seed density for the individual classes.

Bernd

[omid]

Dear Bernd,

I am almost close to the final step but I get a very bothering problem! I see the nuclei and phase fraction of phase 2 during the pause time but the problem is that I do not get the phase growth of phase 1! at the initiation I am at the phase 1 region of phase diagram so of course it should be the predominant phase! Regarding the experimental observation, the phase fraction of phase 2 should not be much higher than the generated amount during the pause time! So phase 1 must be nucleating and growing much faster than phase 2! Which I can't get! Would you please have a look at the Dri file attached here?

Regards,
Omid.

[Bernd]

Dear Omid,

The main problem which I found is that you restrict MICRESS to use quite big time steps of 1E-4 s (at the end of Time Input Data). For maintaining numerical stability, MICRESS reduces the interface mobility in those grid cells which would require a smaller time step value. This is a good method to increase performance, but it should take effect only on few grid cells. Otherwise, like in your case, the whole interface is slowed down, and the grains cannot grow anymore. Output files for control of this behaviour are the .TabT text output and the .mueS (mobility) graphical output.

Please also note that for the seed density model, the classes need to be chosen with decreasing critical radius. Otherwise, the range of radii is chosen wrongly (which you can see in the screen output).

Bernd