Hi Bernd,
Thank you for your reply & consideration!
Now, I have already started the first run with and without the line 4.
The cause of the discontinuity will be clear soon.
I have done the Scheil simulation with Thermo-Calc considering M23C6 formation with FEV6.
Attached file includes the figures of the results and the command list. The pass is the same as previous one.
The figure doesn't show M23C6 formation. Does this mean that M23C6 formation isn't needed in MICRESS simulation?
Your advice about time step is very helpful!
Our simulations spent long time, typically 4-5 days per a solidification simulation! I'm sure to use higher values!
I'd like to consult you about the .mueS output of the new simulation.
Regards,
Taka
temperature dependent mobilities
-
ilovemicress
- Posts: 26
- Joined: Tue Sep 20, 2011 8:41 am
- anti_bot: 333
Re: temperature dependent mobilities
- Attachments
-
- Thermo-Calc_Scheil.zip
- (12.77 KiB) Downloaded 282 times
Re: temperature dependent mobilities
Hi Taka,
you are right, carbides seem not to form in this alloy - the carbon level is too low, and no strong carbide forming elements are present. I was not checking that...
The alloy composition of your Scheil calculation is not exactly the same as in the MICRESS simulation, are you considering different alloys?
4-5 days is too much, for me (after a "quick" optimisation) it took 55000s!
Bernd
you are right, carbides seem not to form in this alloy - the carbon level is too low, and no strong carbide forming elements are present. I was not checking that...
The alloy composition of your Scheil calculation is not exactly the same as in the MICRESS simulation, are you considering different alloys?
4-5 days is too much, for me (after a "quick" optimisation) it took 55000s!
Bernd
-
ilovemicress
- Posts: 26
- Joined: Tue Sep 20, 2011 8:41 am
- anti_bot: 333
Re: temperature dependent mobilities
Dear Bernd,
Thank you for your reply!
>The alloy composition of your Scheil calculation is not exactly the same as
>in the MICRESS simulation, are you considering different alloys?
Sorry, you are right. I sent the results of another one of slightly different
composition. Then, I think it's not serious.
Now, the new simulations are running faster than previous one with new time-step values!
However, another question has arisen. Please see the attached file.
Two simulations, Case A and Case B, are now runnning. I can't understand why
the Case B spends more time than Case A. Does the unaccountable constant
time-step below 1700K work against the simulation?
I'd like your additional advice.
Regards,
Taka
Thank you for your reply!
>The alloy composition of your Scheil calculation is not exactly the same as
>in the MICRESS simulation, are you considering different alloys?
Sorry, you are right. I sent the results of another one of slightly different
composition. Then, I think it's not serious.
Now, the new simulations are running faster than previous one with new time-step values!
However, another question has arisen. Please see the attached file.
Two simulations, Case A and Case B, are now runnning. I can't understand why
the Case B spends more time than Case A. Does the unaccountable constant
time-step below 1700K work against the simulation?
I'd like your additional advice.
Regards,
Taka
- Attachments
-
- TimestepTEST.zip
- (142.56 KiB) Downloaded 229 times
Re: temperature dependent mobilities
Hi Taka,
it is difficult to tell without having more information about what was happening above 1700K, during the short period when Case B was limited by the time-step and Case A wasn't. Thus, the first important step is to check which differences occured, and whether this was caused by the limit (and thus by local reduction of the mobility):
1.) Check differences of microstructure using DP_MICRESS and .conc*, .c*Pha0 and .frac0/.frac outputs.
2.) Check .mueS output using DP_MICRESS to see which grid cells were affected by limiting in case B (cells with reduced mobility value). You can use DP_MICRESS ("Display/Display differences") for systematic checking.
Furthermore, you can also check, which grid point(s) is (are) causing the lower timestep value below 1700K in case B:
3.) Check the .TabT output (ASCII format). Probably, the segregation criterion (column 9) will dominate the time step (i.e. its value is smaller than the Phase-field criterion in column 5). The next 3 columns give you the phase interaction pair and the pointer for the corresponding cell.
4.) With DP_MICRESS and "Data informations/Pinpoint cell pointer" you can find the position of the cell which is responsible for the small time step value in case B. Around there, the problem (if any) should be located.
Following this procedure, with a bit of luck, you will find the reason for the different behaviour. My guess is that in Case B, due to the reduced mobility in one or few cells, there are some traces of liquid left which cause the lower time step.
Bernd
it is difficult to tell without having more information about what was happening above 1700K, during the short period when Case B was limited by the time-step and Case A wasn't. Thus, the first important step is to check which differences occured, and whether this was caused by the limit (and thus by local reduction of the mobility):
1.) Check differences of microstructure using DP_MICRESS and .conc*, .c*Pha0 and .frac0/.frac outputs.
2.) Check .mueS output using DP_MICRESS to see which grid cells were affected by limiting in case B (cells with reduced mobility value). You can use DP_MICRESS ("Display/Display differences") for systematic checking.
Furthermore, you can also check, which grid point(s) is (are) causing the lower timestep value below 1700K in case B:
3.) Check the .TabT output (ASCII format). Probably, the segregation criterion (column 9) will dominate the time step (i.e. its value is smaller than the Phase-field criterion in column 5). The next 3 columns give you the phase interaction pair and the pointer for the corresponding cell.
4.) With DP_MICRESS and "Data informations/Pinpoint cell pointer" you can find the position of the cell which is responsible for the small time step value in case B. Around there, the problem (if any) should be located.
Following this procedure, with a bit of luck, you will find the reason for the different behaviour. My guess is that in Case B, due to the reduced mobility in one or few cells, there are some traces of liquid left which cause the lower time step.
Bernd
-
ilovemicress
- Posts: 26
- Joined: Tue Sep 20, 2011 8:41 am
- anti_bot: 333
Re: temperature dependent mobilities
Dear Bernd,
Thank you for your reply!
Your informative advice helped me be able to guess the reason for different behaviour between the Case A and B! Crystallization of γ-phase from residual liquid phase is thought to relate to the reduction of time-step just above 1700K.
The details of the verification will be reported for you next week.
Have a nice weekend!
Regards,
Taka
Thank you for your reply!
Your informative advice helped me be able to guess the reason for different behaviour between the Case A and B! Crystallization of γ-phase from residual liquid phase is thought to relate to the reduction of time-step just above 1700K.
The details of the verification will be reported for you next week.
Have a nice weekend!
Regards,
Taka
-
ilovemicress
- Posts: 26
- Joined: Tue Sep 20, 2011 8:41 am
- anti_bot: 333
Re: temperature dependent mobilities
Dear Bernd,
First, re-simulation related to the discontinuity of ferrite fraction curve have finished.
The result of "new" first run without line 4 (Case 1) was same with that of the "old" first run (first run). Then, the old first run must have been done without line 4 just like you said. Sorry.
The "new" first run with line 4 and the restart (Case 2) didn't show the discontinuity. The interpolation was performed well.
However, a trivial question has arisen. I don't know why the red curve doesn't have the discontinuity. The time-step setting "max=1.5E-4, min=1.5E-6" was used for the black and the green curve. And the setting "max=1.0, min=1.5E-6" was used for the red and the blue curve. Does this difference relate to the existence or non-existence of the discontinuity?
Attached file, "1.zip", show the details.
Second, I have checked the tabT to find the reason for too small time-step just above 1700K. Then, Pointer column let us clearly know that the grid cells related to the crystallization of γ-phase from residual liquid phase.
However, I intended to simulate the F mode, which must be without the crystallization of γ from liquid.
To avoid this problem, I decreased the max. nucleation temperature of γ from 1710K to 1702K. As a result, the crystallization of γ from liquid and the small time-step is vanished.
Attached file, "2.zip", show the details.
This "1710K" is decided by the result of the Scheil module on Thermo-Calc. The modification of nucleation temp. is valid or invalid?
Regards,
Taka
First, re-simulation related to the discontinuity of ferrite fraction curve have finished.
The result of "new" first run without line 4 (Case 1) was same with that of the "old" first run (first run). Then, the old first run must have been done without line 4 just like you said. Sorry.
The "new" first run with line 4 and the restart (Case 2) didn't show the discontinuity. The interpolation was performed well.
However, a trivial question has arisen. I don't know why the red curve doesn't have the discontinuity. The time-step setting "max=1.5E-4, min=1.5E-6" was used for the black and the green curve. And the setting "max=1.0, min=1.5E-6" was used for the red and the blue curve. Does this difference relate to the existence or non-existence of the discontinuity?
Attached file, "1.zip", show the details.
Second, I have checked the tabT to find the reason for too small time-step just above 1700K. Then, Pointer column let us clearly know that the grid cells related to the crystallization of γ-phase from residual liquid phase.
However, I intended to simulate the F mode, which must be without the crystallization of γ from liquid.
To avoid this problem, I decreased the max. nucleation temperature of γ from 1710K to 1702K. As a result, the crystallization of γ from liquid and the small time-step is vanished.
Attached file, "2.zip", show the details.
This "1710K" is decided by the result of the Scheil module on Thermo-Calc. The modification of nucleation temp. is valid or invalid?
Regards,
Taka
Re: temperature dependent mobilities
Dear Taka,
it is interesting that now, after having understood why you got the discontinuity in the first run, you are not able to reproduce the artefact
- I cannot imagine that it has to do with the maximum time step, there must be something else what you changed in Case 1, compared to the first run...
As you found out, the small time step values and the associated numerical issues are connected to nucleation of austenite from the small amount of rest liquid. But if you decrease the nucleation temperature for austenite in order to avoid these problems, you are "changing physics", because (also from the Scheil results) there is no reason that austenite should not nucleate in the liquid! The only valid argument would be that the results are not strongly changed by this trick (if that is true).
My approach would be rather to remove the numerical issue by proper choice of mobilities: If you use temperature dependent mobility also for the 0/1 and 0/2 interfaces, you should be able to remove the problem and also to further improve performance. The high mobility for these solid-liquid interfaces which you have chosen are only needed for the initial dendritic solidification, i.e. to get realistic truncs and side branches. At the lower temperature of final solidification, you can easily reduce the mobility of these two phase interactions by a factor 10-100 without affecting the solidification kinetics (you can use the same mobility file for both). Thus you would increase the calculated time steps correspondingly and also remove the numerical issues. I use this trick for nearly all solidification simulations!
Bernd
it is interesting that now, after having understood why you got the discontinuity in the first run, you are not able to reproduce the artefact
- I cannot imagine that it has to do with the maximum time step, there must be something else what you changed in Case 1, compared to the first run...As you found out, the small time step values and the associated numerical issues are connected to nucleation of austenite from the small amount of rest liquid. But if you decrease the nucleation temperature for austenite in order to avoid these problems, you are "changing physics", because (also from the Scheil results) there is no reason that austenite should not nucleate in the liquid! The only valid argument would be that the results are not strongly changed by this trick (if that is true).
My approach would be rather to remove the numerical issue by proper choice of mobilities: If you use temperature dependent mobility also for the 0/1 and 0/2 interfaces, you should be able to remove the problem and also to further improve performance. The high mobility for these solid-liquid interfaces which you have chosen are only needed for the initial dendritic solidification, i.e. to get realistic truncs and side branches. At the lower temperature of final solidification, you can easily reduce the mobility of these two phase interactions by a factor 10-100 without affecting the solidification kinetics (you can use the same mobility file for both). Thus you would increase the calculated time steps correspondingly and also remove the numerical issues. I use this trick for nearly all solidification simulations!
Bernd
-
ilovemicress
- Posts: 26
- Joined: Tue Sep 20, 2011 8:41 am
- anti_bot: 333
Re: temperature dependent mobilities
Dear Bernd,
Sorry for my delayed reply.
I found the cause of the discontinuity. I must have simulated the "old" first run without line 4.
Please check the attached file. The behaviour of the Case 3 (i.e. line3→line4) is same that of the "old" first run. Your guess is completely right!
Thank you for your kindly support.
Then, I tried simulating with lower 0/1 interface mobility after temperature range of dendrite growth. However, in the result, figure of the dendrite was unshaped. As I checked the mueS, I found that interpolation was already started at 70 K higher than the temperature we applied new interface mobility as a parameter on the drive file. Therefore, the unshaped dendrite growth is likely to relate to the interpolation. Do you have any idea to resolve this problem?
Regards,
Taka
Sorry for my delayed reply.
I found the cause of the discontinuity. I must have simulated the "old" first run without line 4.
Please check the attached file. The behaviour of the Case 3 (i.e. line3→line4) is same that of the "old" first run. Your guess is completely right! Thank you for your kindly support.
Then, I tried simulating with lower 0/1 interface mobility after temperature range of dendrite growth. However, in the result, figure of the dendrite was unshaped. As I checked the mueS, I found that interpolation was already started at 70 K higher than the temperature we applied new interface mobility as a parameter on the drive file. Therefore, the unshaped dendrite growth is likely to relate to the interpolation. Do you have any idea to resolve this problem?
Regards,
Taka
- Attachments
-
- interpolation_of_interface_mobility.zip
- (486.41 KiB) Downloaded 254 times
Re: temperature dependent mobilities
Dear Taka,
nice to hear that the discontinuity issue is solved!
If you want to reduce the 0/1 mobility at lower temperatures - as I proposed in order to improve numerical stabilitiy and performance - you have to take care that due to interpolation this reduction is not already applied at temperatures where the dendrite tips are still developing and growing rapidly. You can obtain that by building in a "step" at a sufficiently low temperature T2, the mobility input file could look like:
T1 (>Tliq) ........... high_mobility
T2 .................... high_mobility
T2 -dT................ low mobility
T3..................... low mobility
Then you can be sure that the mobility is not decreasing above T2 due to interpolation!
Bernd
nice to hear that the discontinuity issue is solved!
If you want to reduce the 0/1 mobility at lower temperatures - as I proposed in order to improve numerical stabilitiy and performance - you have to take care that due to interpolation this reduction is not already applied at temperatures where the dendrite tips are still developing and growing rapidly. You can obtain that by building in a "step" at a sufficiently low temperature T2, the mobility input file could look like:
T1 (>Tliq) ........... high_mobility
T2 .................... high_mobility
T2 -dT................ low mobility
T3..................... low mobility
Then you can be sure that the mobility is not decreasing above T2 due to interpolation!
Bernd
-
ilovemicress
- Posts: 26
- Joined: Tue Sep 20, 2011 8:41 am
- anti_bot: 333
Re: temperature dependent mobilities
Dear Bernd,
Thanks for your advice.
I will try building in a "step" in the mobility file.
Then, I aspire to further improve performance. According to tabT, the 2/2 mobility determines the Time-step used after the dendrite developing and growing. Does the performance improve if the lower 2/2 mobility would be applied without other serious effects?
Regards,
Taka
Thanks for your advice.
I will try building in a "step" in the mobility file.
Then, I aspire to further improve performance. According to tabT, the 2/2 mobility determines the Time-step used after the dendrite developing and growing. Does the performance improve if the lower 2/2 mobility would be applied without other serious effects?
Regards,
Taka