Hi betleenkim,
I am not an expert on such types of material, especially not with such a high Al content. But I am not sure whether your conclusion is correct. If I understand correctly, you identify two problems, namely the wrong (too low) phase fraction of ferrite after reheating, and the "wrong" partitioning of Si. My doubts about your explanation are the following:
1.) Si segregation is mainly due to solidification (liquid--> delta-ferrite). During the solid-solid transformation, the substitutional elements like Si cannot move any more, so they are not partitioning (NPLE or paraequilibrium). Thus, Si is not on the right side of the domain because of partitioning but because of segregation during solidification!
2.) The experiment on Si segregation is showing submicron grains and thus not comparable - on such a small lengthscale, diffusion of Si may also occur at lower temperatures.
3.) The correct amount of ferrite after cooling and reheating requires correct transformation kinetics in the simulation. Without using any special models, MICRESS results suffer from the fact that the slow diffusing elements are not sufficiently resolved - therefore, you should use either the nple model or the paraequilibrium model to get quantitative results (can be selected with the "redistribution_control" flag). NPLE typically should lead to much lower transformation rates, as local equilibrium is achieved and the pile-up of the elements reduces the driving force which is available for transformation.
4.) The temperature range of the simulations you show is more than 1000K! Grid resolution is good only for the upper part, but at temperatures well below 1000K, even the diffusivity of the interstitial elements is such low that their diffusion profiles cannot be properly resolved. As a consequence, the transformation rate is probably wrong without calibration.
Most people doing phase-field simulations of transformations between ferrite and austenite, when they try to do quantitative work, typically use experimentally calibrated interface mobilities (including temperature dependency) to adjust for the specific redistribution behaviour and for stress effects, or they try to apply the nple/para-equilibrium condition under the assumption of a diffusion limited transformation (with respect to the fast interstitial elements). Especially in the latter case it is necessary to work with a sufficiently high grid resolution or to calibrate against highly resolved test calculations (see here).
Best wishes
Solidification simulation under casting
-
betleenkim
- Posts: 41
- Joined: Tue May 07, 2013 1:01 pm
- anti_bot: 333
Re: Solidification simulation under casting
Dear Bernd
Thank you for your reply!
Firstly, NPLE or Para equilibrium assumption is not proper for my simulation since i heat up the domain to 1550k which is the temperature any elements can diffuse with high speed. Many researchers including R. Wei et. al. (Metallurgica and materials transactions A, Vol. 42, p.2189) have shown that the PLE/NPLE transition temperature of Si is around 950K. So i did simulation with NPLE assumption to 950K and i turn it to normal behavior above the temperature. However, there is no difference. This is because 1550K is such high temperature so i think that everything(especially phase transformation kinetics) is going fast.
Secondly, i did thermo_calc calculation for Si partitioning with several different alloying system. Up to quaternary system including FeCMnSi shows that Si is partitioned into BCC. However, as i introduce Al into the system, Si goes opposite direction that is it is partitioned into FCC. I am wondering that this is right behavior of Si. Since i do not know how the database is constructed, i am not so sure that this is right phenomenon or not. Do you think i personally contact the Thermo_Calc company?
Finally, as i attached Al is not diffuse in to the liquid phase during solidification. I think that this is right behavior of Al as it is ferrite stabilizer. But as you can see, Si behaves exactly same with Mn in this system. Best regard
Thank you for your reply!
Firstly, NPLE or Para equilibrium assumption is not proper for my simulation since i heat up the domain to 1550k which is the temperature any elements can diffuse with high speed. Many researchers including R. Wei et. al. (Metallurgica and materials transactions A, Vol. 42, p.2189) have shown that the PLE/NPLE transition temperature of Si is around 950K. So i did simulation with NPLE assumption to 950K and i turn it to normal behavior above the temperature. However, there is no difference. This is because 1550K is such high temperature so i think that everything(especially phase transformation kinetics) is going fast.
Secondly, i did thermo_calc calculation for Si partitioning with several different alloying system. Up to quaternary system including FeCMnSi shows that Si is partitioned into BCC. However, as i introduce Al into the system, Si goes opposite direction that is it is partitioned into FCC. I am wondering that this is right behavior of Si. Since i do not know how the database is constructed, i am not so sure that this is right phenomenon or not. Do you think i personally contact the Thermo_Calc company?
Finally, as i attached Al is not diffuse in to the liquid phase during solidification. I think that this is right behavior of Al as it is ferrite stabilizer. But as you can see, Si behaves exactly same with Mn in this system. Best regard
-
betleenkim
- Posts: 41
- Joined: Tue May 07, 2013 1:01 pm
- anti_bot: 333
Re: Solidification simulation under casting
Dear Bernd
I finally finished a simulation with high resolution. I think that it task around a month to get this result
I cooled the domain for 3000s to 746K and reheated at a rate of 20K/s.
If i apply the interface mobility between ferrite and austenite which is literary taken, i found that no ferrite remains after reheating to 1550K. So i applied very low interface mobility then i could achieve ferrite volume fraction that is closed to experimentally observed ferrite fraction. Now i am trying to find this assumption-low interface mobility- is proper or not.
Anyway, i found a sudden increment of interface during heating as i attached here.
I set autenite nucleation site as the interphase boundary between ferrite and austenite. But it seems that observed nucleation site is inside a ferrite grain as i indicated by red rectangle in the figure.
In addition, there is additional interface at the left side of the domain. But corresponding phase output shows that there is no numerical interface there. How is it happen?
Finally, why do i get such a high number of interface although i am doing 2D simulation? Does it because there is high number of nucleation?
Best regard
I finally finished a simulation with high resolution. I think that it task around a month to get this result
I cooled the domain for 3000s to 746K and reheated at a rate of 20K/s.
If i apply the interface mobility between ferrite and austenite which is literary taken, i found that no ferrite remains after reheating to 1550K. So i applied very low interface mobility then i could achieve ferrite volume fraction that is closed to experimentally observed ferrite fraction. Now i am trying to find this assumption-low interface mobility- is proper or not.
Anyway, i found a sudden increment of interface during heating as i attached here.
I set autenite nucleation site as the interphase boundary between ferrite and austenite. But it seems that observed nucleation site is inside a ferrite grain as i indicated by red rectangle in the figure.
In addition, there is additional interface at the left side of the domain. But corresponding phase output shows that there is no numerical interface there. How is it happen?
Finally, why do i get such a high number of interface although i am doing 2D simulation? Does it because there is high number of nucleation?
Best regard
Re: Solidification simulation under casting
Hi betleenkim,
the .phas output is not a good choice for checking the extension of the numerical interface, as your comparison with the .intf reveals! It shows only the part of interface where the phase-field parameter is between 0.3 and 0.7!
But fact is that there is interface, which I believe is due to numerical artifacts which come from an insufficient grid resolution at low temperatures. You should check the .intf output, which shows that the interfaces are already spread when you start reheating. You could also use the .frac2 output to check whether phase 2 was retained inside phase 1.
You can try to avoid spreading of the austenite ferrite interfaces by reducing mobility at lower temperatures, using a temperature dependent mobilty which can be read from a text file. But to get correct transformation rates at low temperature would require much higher spatial resolution which is not compatible with the length scale of your simulation!
I agree that at the high temperatures after reheating, PLE should be the correct mechanism. But when using nple mode in MICRESS, this should also come out correctly.
With respect to the partitioning, you should ask Thermo-Calc people, if you have doubts about the thermodynamic data. Just contact support (support@thermocalc.com).
Bernd
the .phas output is not a good choice for checking the extension of the numerical interface, as your comparison with the .intf reveals! It shows only the part of interface where the phase-field parameter is between 0.3 and 0.7!
But fact is that there is interface, which I believe is due to numerical artifacts which come from an insufficient grid resolution at low temperatures. You should check the .intf output, which shows that the interfaces are already spread when you start reheating. You could also use the .frac2 output to check whether phase 2 was retained inside phase 1.
You can try to avoid spreading of the austenite ferrite interfaces by reducing mobility at lower temperatures, using a temperature dependent mobilty which can be read from a text file. But to get correct transformation rates at low temperature would require much higher spatial resolution which is not compatible with the length scale of your simulation!
I agree that at the high temperatures after reheating, PLE should be the correct mechanism. But when using nple mode in MICRESS, this should also come out correctly.
With respect to the partitioning, you should ask Thermo-Calc people, if you have doubts about the thermodynamic data. Just contact support (support@thermocalc.com).
Bernd