Dear Mr. Böttger,
I hope you are doing well.
This is regarding Micress simulations of austenitic stainless steel in the additive manufacturing process, specifically Powder Bed Fusion. The objective of these simulations is to understand the difference in the microstructure evolution of these steels in additive manufacturing while varying the Ni content between 8 and 12% (weight percent) while keeping other elements constant. The two phases that are present are Delta-ferrite and Gamma-Austenite, only these two phases will be simulated.
So, I ran two types of simulations as a beginning for two alloys, one with 8% Ni and the other with 12% Ni:
1. Equiaxed
2. Directional
Per email under the same subject naming, please find the driving files and the GES file used, I have already sent this email regarding this topic 10 days ago (15.10) but it might have gotten lost in your inbox or junk folder somehow.
Just as a small note, the output locations for the equiaxed simulations are mixed up, the 8 Ni results are in the 12 Ni folder and vice versa.
Simulations specifications:
1. For all simulations a cooling rate of 1E6 K/s was used as to fall within the AM range.
2. The temperature gradient for equiaxed simulations has been set to zero K/m, while being set to 20,000 K/cm for the directional solidification.
3. The interfacial energies between the three phases (Liquid, Delta & Gamma) were calculated via Thermo-Calc in both cases.
4. In the equiaxed simulation for 12% Ni I had to increase the min undercooling of seed 1 (Delta-ferrite) up to 7 K for the nucleation to happen.
Doubts and Questions:
1. In all simulations, especially in the 12 Ni equiaxed simulation, it is clearly visible that late into the simulation, Gamma starts to grow inside the Delta ferrite grains, having dotted-like structure. Is this physically meaningful? Or do I have to use the split from grain command? When I used split from grain, this did not happen.
2. The most important problem is that even with a high Ni content as high as 12 weight percent, given that Ni is an austenite stabilizing element, the Gamma -austenite volume fraction simulated was no way near the real case at around 90 – 95% Gamma-austenite and 5-10% Delta-ferrite. The volume fraction was even falling behind the Scheil Gulliver results for the same alloy with 12% Ni looking at the same temperatures along the solidification. I was wondering how to make the Gamma-austenite growth more aggressive. As you can see in the attached picture (Fractions Vs Time 12 Ni), only when liquid approaches zero, the solid-solid transformation from delta to gamma starts to occur, and Gamma starts to take over till the end of solidification. However, the intended behavior should be like in the micress training example (T02_01_FeCMn_D_DeltaGamma_2D_TQ), Where firstly austenite is growing simultaneously with Delta-ferrite and secondly it looked like the Delta-ferrite dendrite was partially transforming into Gamma-austenite while it is growing, while liquid is still present. So far, I have not been able to capture this behavior which could also solve the volume fraction issue, as Gamma is more aggressive in this case like in the training example. Do you have any tips in this regard?
3. For the equiaxed simulations, as mentioned, I used no thermal gradient, is that okay in a physical sense? Can I rely on this ? This was the only way I could have a fully equiaxed growth.
4. For Directional Solidification, the simulated volume fraction of Gamma is 3 times lower that in the case of equiaxed growth. But here it might be more matching with the real case in terms of the implementation of a thermal gradient. However, the resulting volume fractions are much further from reality.
5. Any other notes from your side so as to improve the simulations would be great.
Thanks in advance for your help. I really appreciate your support.
Delta/Gamma Micress Simulations in Additive Manufacturing
Re: Delta/Gamma Micress Simulations in Additive Manufacturing
Hi Taha,
Somehow I missed your email, sorry for that!
I think, the main point is that you assume diffusion-limited growth for the solid-state transformation which is not correct. I would expect nple-mode for all substitutional elements (like in T02_01_FeCMn_D_DeltaGamma_2D_TQ), while carbon is diffusion-limited (mob_corr). Then, gamma growth is definitively faster. You also should then increase the mobility value of the 1/2-interface (e.g. to 10.) in order not to limit kinetics arteficially.
Another point is that the assumed interface energies are quite low. Thermo-Calc predicts only the chemical part, I would interpret it as a lower limit for coherent boundaries. If you stick to the low values, you need to increase grid resolution. You can easily see the effects of low resolution when the dendrites switch from growing in correct orientation to running along the grid direction.
One important question I think you need to address is whether the primary dendrites grow as ferrite or austenite. Perhaps it is possible to see that from experimental data. I recently assisted in a paper which addressed a similar topic:
https://www.sciencedirect.com/science/a ... via%3Dihub
With respect to your equiaxed simulations, it is certainly wrong to assume no thermal gradient. Even if there would be equiaxed growth during the LPBF-process, the strong temperature gradient is still there. Do you have any experimental evidence for equiaxed growth? I could imagine that only in case of the presence of extremely large seed particles, or at the very bottom of the first layer on a homogenized substrate, leading to fragmentation.
I don't understand what you wanted to achieve with "split_from_grain". An interesting question, however, would be whether you expect nucleation only at the solid-liquid interfaces, or if there also could happen nucleation inside solid (grain boundaries or bulk).
Best wishes
Bernd
Somehow I missed your email, sorry for that!
I think, the main point is that you assume diffusion-limited growth for the solid-state transformation which is not correct. I would expect nple-mode for all substitutional elements (like in T02_01_FeCMn_D_DeltaGamma_2D_TQ), while carbon is diffusion-limited (mob_corr). Then, gamma growth is definitively faster. You also should then increase the mobility value of the 1/2-interface (e.g. to 10.) in order not to limit kinetics arteficially.
Another point is that the assumed interface energies are quite low. Thermo-Calc predicts only the chemical part, I would interpret it as a lower limit for coherent boundaries. If you stick to the low values, you need to increase grid resolution. You can easily see the effects of low resolution when the dendrites switch from growing in correct orientation to running along the grid direction.
One important question I think you need to address is whether the primary dendrites grow as ferrite or austenite. Perhaps it is possible to see that from experimental data. I recently assisted in a paper which addressed a similar topic:
https://www.sciencedirect.com/science/a ... via%3Dihub
With respect to your equiaxed simulations, it is certainly wrong to assume no thermal gradient. Even if there would be equiaxed growth during the LPBF-process, the strong temperature gradient is still there. Do you have any experimental evidence for equiaxed growth? I could imagine that only in case of the presence of extremely large seed particles, or at the very bottom of the first layer on a homogenized substrate, leading to fragmentation.
I don't understand what you wanted to achieve with "split_from_grain". An interesting question, however, would be whether you expect nucleation only at the solid-liquid interfaces, or if there also could happen nucleation inside solid (grain boundaries or bulk).
Best wishes
Bernd
Re: Delta/Gamma Micress Simulations in Additive Manufacturing
Dear Bernd,
Thank you for your suggestions. The results now are much better and closer to the experimental results in terms of the volume fractions of phases.
I used the nple diffusion model for the substitutional elements which seemed to be of a critical importance as you suggested. Howerver, I could not really increase the interfacial energies as altering them or increasing them even a tiny bit seemed to have a detrimental effect on the dendrite shape and growth which looked strange for the most part.
I executed 3 directional solidification simulations with 12,10 and 8 wt% Ni of the same alloy (sent via E-mail).
12 Ni looked quite good in terms of phase fractions and also grain morphology. However, in the case of 10 and 8 Ni, the morphology of austenite grains (white phase) seemed to be a bit random taking a star like shape. The distribution of the austenite grains did not also look right.
It would be great if you could please guide me through this problem and direct me if you see that I am making another mistake in general to optimize the simulations.
Thanks in advance!
Best regards,
Taha
Thank you for your suggestions. The results now are much better and closer to the experimental results in terms of the volume fractions of phases.
I used the nple diffusion model for the substitutional elements which seemed to be of a critical importance as you suggested. Howerver, I could not really increase the interfacial energies as altering them or increasing them even a tiny bit seemed to have a detrimental effect on the dendrite shape and growth which looked strange for the most part.
I executed 3 directional solidification simulations with 12,10 and 8 wt% Ni of the same alloy (sent via E-mail).
12 Ni looked quite good in terms of phase fractions and also grain morphology. However, in the case of 10 and 8 Ni, the morphology of austenite grains (white phase) seemed to be a bit random taking a star like shape. The distribution of the austenite grains did not also look right.
It would be great if you could please guide me through this problem and direct me if you see that I am making another mistake in general to optimize the simulations.
Thanks in advance!
Best regards,
Taha
Re: Delta/Gamma Micress Simulations in Additive Manufacturing
Dear Taha,
I think there are two things you should improve. The first is that your three simulations start with different undercooling. This is because the liquidus temperature changes with the Ni composition. You would have to adapt the initial temperature of the domain for each case. However, it is more recommendable to set the initial grain by nucleation instead. Just define two seed types for phase 1 with "region" to replace the 2 initial grains, and define the region such that only one grid cell is inside (which corresponds to the correct position of each initial grain). Then you should start the simulation at a higher temperature (e.g. 1750K), and the grains will always automatically appear when a specified undercooling is reached (e.g. 10K which you define as critical undercooling of nucleation . With this change, the simulation for the different compositions can really be compared!
The second point is what I already mentioned: The interface energy is too small, especially for the 1/2-interface. This leads to artifacts. For the 0/1-interface, the too low interface energy makes the dendrites grow along the grid direction, while for the 1/2-interaction the interfaces tend to get unstable. If you use my first suggestion, then the grains of phase 1 will grow even if you increase the interface energies (solving your problems with increasing interface energy which you mentioned).
If the austenite grains still look different as compared to the experimental results, you can try to adjust the nucleation conditions of phase 2. The less nuclei you set (e.g. by increasing the critical undercooling or the shield time/radius), the bigger the grains should get. Please note that 2D-simulations generally lead to smaller precipitates compared to 3D-simulations.
Best wishes
Bernd
I think there are two things you should improve. The first is that your three simulations start with different undercooling. This is because the liquidus temperature changes with the Ni composition. You would have to adapt the initial temperature of the domain for each case. However, it is more recommendable to set the initial grain by nucleation instead. Just define two seed types for phase 1 with "region" to replace the 2 initial grains, and define the region such that only one grid cell is inside (which corresponds to the correct position of each initial grain). Then you should start the simulation at a higher temperature (e.g. 1750K), and the grains will always automatically appear when a specified undercooling is reached (e.g. 10K which you define as critical undercooling of nucleation . With this change, the simulation for the different compositions can really be compared!
The second point is what I already mentioned: The interface energy is too small, especially for the 1/2-interface. This leads to artifacts. For the 0/1-interface, the too low interface energy makes the dendrites grow along the grid direction, while for the 1/2-interaction the interfaces tend to get unstable. If you use my first suggestion, then the grains of phase 1 will grow even if you increase the interface energies (solving your problems with increasing interface energy which you mentioned).
If the austenite grains still look different as compared to the experimental results, you can try to adjust the nucleation conditions of phase 2. The less nuclei you set (e.g. by increasing the critical undercooling or the shield time/radius), the bigger the grains should get. Please note that 2D-simulations generally lead to smaller precipitates compared to 3D-simulations.
Best wishes
Bernd