Dear Bernd,
I would like to compare compositions in dendritic and interdendritic regions upon non-equilibrium solidification in different additive processes. Can a function in DP-Micress or any of the output datas provide this information? I saw the segregation analysis in DP-Micress under data analysis section which plots a value versus cumulative distribution plot. Does increase in cumulative distribution means increase in relative distance to the dendrite core? I cant actually do it with virtual edx since morphologies and amount/thickness of dendrite cores and interdendritic regions naturally vary a lot between processes and conditions.
Moreover, is there any automated possiblity intergrated in DP-Micress to extract mophology information e.g. primary dendrite spacings or should we also do it manually or other post-processing techniques?
Regards,
Ahmet
Analyzing average compositions in dendritic and interdendritic regions
Re: Analyzing average compositions in dendritic and interdendritic regions
Dear Ahmet,
Yes, I think segregation analysis can indeed help you comparing interdendritic segregation pattern for different process conditions. It is important to understand that by sorting the concentration values with respect to a reference component, any direct neighborhood relations are erased in a first instance. However, as a consequence of fast diffusion during solidification, the position in the distribution can be related to a certain solidification time and thus identifies regions in the domain which lie inside or outside a certain segregated interdendritic area.
Apart from segregation analysis, DP_MICRESS can provide properties (e.g. averages) on the domain or the region of the view, and measure interconnected areas characterized by a certain range of field values (Morphology analysis). This can be combined with "Operations" which allow for introducing relations between the same points in different fields, but also between points and their geometrical neighbors. Ralph is the person who knows best about that.
There is still no possibility of automatically determining the primary or secondary dendrite arm spacing. We once tried to implement some pattern recognition algorithms with that aim, but the results were really satisfying...
Bernd
Yes, I think segregation analysis can indeed help you comparing interdendritic segregation pattern for different process conditions. It is important to understand that by sorting the concentration values with respect to a reference component, any direct neighborhood relations are erased in a first instance. However, as a consequence of fast diffusion during solidification, the position in the distribution can be related to a certain solidification time and thus identifies regions in the domain which lie inside or outside a certain segregated interdendritic area.
Apart from segregation analysis, DP_MICRESS can provide properties (e.g. averages) on the domain or the region of the view, and measure interconnected areas characterized by a certain range of field values (Morphology analysis). This can be combined with "Operations" which allow for introducing relations between the same points in different fields, but also between points and their geometrical neighbors. Ralph is the person who knows best about that.
There is still no possibility of automatically determining the primary or secondary dendrite arm spacing. We once tried to implement some pattern recognition algorithms with that aim, but the results were really satisfying...
Bernd
Re: Analyzing average compositions in dendritic and interdendritic regions
Dear Bernd,
Thank you so much for your answer. I tried some segregation analysis in different conditions however I did not understand it completely. Therefore I will try to continue over some simulation results as attached.
The images shows the microstructures and Al distribution simulated via LPBF and LMD. The simulations continued until 1000K after full solidification in order to continue the formation of the second phase, which is Al rich. For comparison, I also selected same growth rates adjusted by cooling rate/temperature gradient. What I can see from these graphs is, LPBF has higher amplitude (higher max Al) and suppose to have higher frequency due to smaller dendrite spacing (can be seen from sharper increase). However when I run Scheil simulations in solute trappping model in ThermoCalc (varying solidification velocities) the calculations suggests that I should have lower amplitude of segregation in interdendritic regions when the growth rate is higher (e.g. LPBF). Here in Micress, I see vice-versa even though trend of Al increase in cumulative distribution (lets say frequency) makes sense for different conditions. Do you have any comments on these microsegregation values between given conditions? I dont know how to perform the analysis to extract relevant trends between processes. even though the lower frequency suggests that interdendritic volumes are bigger in LMD, I dont exactly know how to correlate it. It seems like it might also be causing from liquid pockets (spherical features) which appear to be slightly more enriched in LPBF.
Moreover, I also couldnt understand following comments; "It is important to understand that by sorting the concentration values with respect to a reference component, any direct neighborhood relations are erased in a first instance". Is that mean, the concentration distributions in simualations are somehow simplified to only see overall trend but not local ones?
Do you have any example for morphology analysis or in general post-processing? That would be really helpful to see how it is done.
thank you and regards,
Ahmet
Thank you so much for your answer. I tried some segregation analysis in different conditions however I did not understand it completely. Therefore I will try to continue over some simulation results as attached.
The images shows the microstructures and Al distribution simulated via LPBF and LMD. The simulations continued until 1000K after full solidification in order to continue the formation of the second phase, which is Al rich. For comparison, I also selected same growth rates adjusted by cooling rate/temperature gradient. What I can see from these graphs is, LPBF has higher amplitude (higher max Al) and suppose to have higher frequency due to smaller dendrite spacing (can be seen from sharper increase). However when I run Scheil simulations in solute trappping model in ThermoCalc (varying solidification velocities) the calculations suggests that I should have lower amplitude of segregation in interdendritic regions when the growth rate is higher (e.g. LPBF). Here in Micress, I see vice-versa even though trend of Al increase in cumulative distribution (lets say frequency) makes sense for different conditions. Do you have any comments on these microsegregation values between given conditions? I dont know how to perform the analysis to extract relevant trends between processes. even though the lower frequency suggests that interdendritic volumes are bigger in LMD, I dont exactly know how to correlate it. It seems like it might also be causing from liquid pockets (spherical features) which appear to be slightly more enriched in LPBF.
Moreover, I also couldnt understand following comments; "It is important to understand that by sorting the concentration values with respect to a reference component, any direct neighborhood relations are erased in a first instance". Is that mean, the concentration distributions in simualations are somehow simplified to only see overall trend but not local ones?
Do you have any example for morphology analysis or in general post-processing? That would be really helpful to see how it is done.
thank you and regards,
Ahmet
- Attachments
-
- LPBF vs LMD.jpg (796.11 KiB) Viewed 5090 times
-
- Cumulative distribution of Al (1).PNG (79.79 KiB) Viewed 5090 times
Re: Analyzing average compositions in dendritic and interdendritic regions
Dear Atur,
The type of analysis you do is not trivial, especially with respect to interpretation. I personally did not do anything similar yet, so I can only give you my opinion (and of course my knowledge about how MICRESS and DP_MICRESS work).
I think there are several factors which have an influence on the concentration distribution how it is analyzed by "segregation analysis" in DP_MICRESS. I wanted to make clear with my explanations in my last post that by sorting the pixel values, any neighborhood relations are lost, and you thus cannot (directly) see from the resulting distribution, whether the corresponding microstructure was fine or coarse - you just can see how big the relative extend of each segregation level is (i.e. which extent the segregated area has relative to the total area). In this context, I also do not understand your "frequency" argument: I don't see a reason why a finer microstructure automatically should lead to a steeper concentration distribution.
What I expect to be the major two factors here are:
1.) The solidification temperature: Depending on the dendrite tip undercooling, solidification may start at a much lower temperature compared to Scheil, and the partition coefficients typically are strongly temperature dependent. Also at a later stage, solidification temperature is strongly morphology dependent (e.g. much lower in big liquid pockets).
2.) The size of the primary spacing (at given cooling rate) is very important for diffusion processes (back diffusion in the solid, on one side, and incomplete liquid diffusion, on the other side). This is also not reflected in the Scheil model.
Contrarily, I do not believe that solute trapping is a major factor here. It may have some relevance at the very dendrite tip, but not for the rest of solidification. I do not know all details of the Thermo-Calc solute trapping Scheil model, but I don't think it is the right tool for this application. Solute trapping would be more pronounced, if planar interface solidification would be considered.
Thus, for the kind of comparison you do, I think it is important to focus on a comparable (and of course as correct as possible) microstructure. I fear, this is not completely true in your simulations: While in the LPBF case the primary spacing is evolving by planar front instability, it is fully dominated by the initial seed size in case of 0.1m/s LMD (0.01 m/s LMD is in between). Unfortunately, there is no clear rule how to obtain representative microstructures...
Another point which is not clear for me is whether and where precipitation of a secondary phase happens. Could it be e.g. that in the LPBF case the higher Al-concentration peaks are caused by precipitates which are sufficiently resolved in this case (i.e. high phase fraction in individual grid cells)? This would be supported by the slight inflection of the distribution curve close to the maximum at the right side. Generally, you should also check whether grid resolution has an important impact - only if you can increase it without drastically changing results you can be on the safe side!
Furthermore, it would be interesting for further interpretation to compare the Scheil results with the segregation curves obtained from the MICRESS simulations. Direct comparison is possible if the concentration in the solid is plotted against fraction of solid.
Bernd
The type of analysis you do is not trivial, especially with respect to interpretation. I personally did not do anything similar yet, so I can only give you my opinion (and of course my knowledge about how MICRESS and DP_MICRESS work).
I think there are several factors which have an influence on the concentration distribution how it is analyzed by "segregation analysis" in DP_MICRESS. I wanted to make clear with my explanations in my last post that by sorting the pixel values, any neighborhood relations are lost, and you thus cannot (directly) see from the resulting distribution, whether the corresponding microstructure was fine or coarse - you just can see how big the relative extend of each segregation level is (i.e. which extent the segregated area has relative to the total area). In this context, I also do not understand your "frequency" argument: I don't see a reason why a finer microstructure automatically should lead to a steeper concentration distribution.
What I expect to be the major two factors here are:
1.) The solidification temperature: Depending on the dendrite tip undercooling, solidification may start at a much lower temperature compared to Scheil, and the partition coefficients typically are strongly temperature dependent. Also at a later stage, solidification temperature is strongly morphology dependent (e.g. much lower in big liquid pockets).
2.) The size of the primary spacing (at given cooling rate) is very important for diffusion processes (back diffusion in the solid, on one side, and incomplete liquid diffusion, on the other side). This is also not reflected in the Scheil model.
Contrarily, I do not believe that solute trapping is a major factor here. It may have some relevance at the very dendrite tip, but not for the rest of solidification. I do not know all details of the Thermo-Calc solute trapping Scheil model, but I don't think it is the right tool for this application. Solute trapping would be more pronounced, if planar interface solidification would be considered.
Thus, for the kind of comparison you do, I think it is important to focus on a comparable (and of course as correct as possible) microstructure. I fear, this is not completely true in your simulations: While in the LPBF case the primary spacing is evolving by planar front instability, it is fully dominated by the initial seed size in case of 0.1m/s LMD (0.01 m/s LMD is in between). Unfortunately, there is no clear rule how to obtain representative microstructures...
Another point which is not clear for me is whether and where precipitation of a secondary phase happens. Could it be e.g. that in the LPBF case the higher Al-concentration peaks are caused by precipitates which are sufficiently resolved in this case (i.e. high phase fraction in individual grid cells)? This would be supported by the slight inflection of the distribution curve close to the maximum at the right side. Generally, you should also check whether grid resolution has an important impact - only if you can increase it without drastically changing results you can be on the safe side!
Furthermore, it would be interesting for further interpretation to compare the Scheil results with the segregation curves obtained from the MICRESS simulations. Direct comparison is possible if the concentration in the solid is plotted against fraction of solid.
Bernd
Re: Analyzing average compositions in dendritic and interdendritic regions
Dear Bernd,
Thank you for your detailed reply. Now I seem to have more questions, especially about my simulation domain
Hope this entry will still be suitable for this topic 
Otherwise I will try to open the same topic under pre-processing. I will try to continue in bullet points;
1- I actually intended to use frequency term to indicate the distance between interdendritic regions in L-PBF, as you commented, i.e. which extent the segregated area has relative to the total area. More openly describing, the increase in Al content starts in LMD cases earlier (more left) side that in LPBF case. I tried to correlate it to the total area. Therefore it appeared to me that as the microstructure gets finer, the concentration appears to increase and decrease steeper compared to coarser microstructures. But, anyway, now it is more clear.
2- For comparison, I am trying to keep the simulation domains as similar as possible. But I am not sure to which extend I am successful building them up. I was also wondering if comparing my L-PBF simulations with LMD simulations are like comparing apples and bananas. I actually dont have experimental results from LMD, but I should continue with simulations at this point by setting certain expectations. Here, I would appreciate your inputs to modify them. As I wrote earlier, the aim is to obtain different trends under various thermal conditions during LMD and L-PBF in order to extract segregations, phase distributions and microstructures (roughly, e.g. which ones are finer/coarser, ~relation between dendrite spacings). Only fundamental differences between the LMD and LPBF simulations I varied are actually grid spacings and undercooling values. The interface mobilities are also calibrated to avoid numerical undercoolings.
In both cases, I start with two flat initial grains with slightly different orientations (to mimic competition between orientations with respect to T gradient, grain boundary formation). Both initial grains are flat, so the simulation first starts with the growth of planar(?) front and then breaking to dendrites. But in case of LMD, primary dendrite is emerging from connection point of grains, which appears to be main primary features we observe. Do you think starting with single flat grain would be more comparable? You said that the LMD simulations are actually dominated by initial seed size. Can you please elaborate it? Is it related to the nucleation conditions I set or the initial microstructures? Can you also suggest me which approach/set up would worth to try while setting initial conditions which makes LPBF and LMD comparable? will send you the drive and the GES5 files.
3- Yes, precipitation happens in LMD.I also think it should happen to some extend in LPBF (at least on grain boundaries, experimentally visible). I agree that peaks might cause due to highly resolved precipitates. As you pointed out, the grid spacing might be influencing the peak composition values during LPBF. I use 0.0075 µm during L-PBF simulations, however these enrichments are actually combination of some cells, where the precipitate width is around 10-15 times higher than grid spacing. Might it be an artifact? Just for clarification, do you think I should gradually increase the grid spacing to see if they remain or not? Is there any reasonable way to mask these peak values? To be honest, having almost 40% Al enriched locally or in precipitate is quite unrealistic for my alloy.
4-Furthermore, it would be interesting for further interpretation to compare the Scheil results with the segregation curves obtained from the MICRESS simulations. Direct comparison is possible if the concentration in the solid is plotted against fraction of solid.
This is actually in my to do list, once I am sure Micress resutls are somehow in acceptable range..
thank you and regards,
Ahmet
Thank you for your detailed reply. Now I seem to have more questions, especially about my simulation domain
Hope this entry will still be suitable for this topic Otherwise I will try to open the same topic under pre-processing. I will try to continue in bullet points;1- I actually intended to use frequency term to indicate the distance between interdendritic regions in L-PBF, as you commented, i.e. which extent the segregated area has relative to the total area. More openly describing, the increase in Al content starts in LMD cases earlier (more left) side that in LPBF case. I tried to correlate it to the total area. Therefore it appeared to me that as the microstructure gets finer, the concentration appears to increase and decrease steeper compared to coarser microstructures. But, anyway, now it is more clear.
2- For comparison, I am trying to keep the simulation domains as similar as possible. But I am not sure to which extend I am successful building them up. I was also wondering if comparing my L-PBF simulations with LMD simulations are like comparing apples and bananas. I actually dont have experimental results from LMD, but I should continue with simulations at this point by setting certain expectations. Here, I would appreciate your inputs to modify them. As I wrote earlier, the aim is to obtain different trends under various thermal conditions during LMD and L-PBF in order to extract segregations, phase distributions and microstructures (roughly, e.g. which ones are finer/coarser, ~relation between dendrite spacings). Only fundamental differences between the LMD and LPBF simulations I varied are actually grid spacings and undercooling values. The interface mobilities are also calibrated to avoid numerical undercoolings.
In both cases, I start with two flat initial grains with slightly different orientations (to mimic competition between orientations with respect to T gradient, grain boundary formation). Both initial grains are flat, so the simulation first starts with the growth of planar(?) front and then breaking to dendrites. But in case of LMD, primary dendrite is emerging from connection point of grains, which appears to be main primary features we observe. Do you think starting with single flat grain would be more comparable? You said that the LMD simulations are actually dominated by initial seed size. Can you please elaborate it? Is it related to the nucleation conditions I set or the initial microstructures? Can you also suggest me which approach/set up would worth to try while setting initial conditions which makes LPBF and LMD comparable? will send you the drive and the GES5 files.
3- Yes, precipitation happens in LMD.I also think it should happen to some extend in LPBF (at least on grain boundaries, experimentally visible). I agree that peaks might cause due to highly resolved precipitates. As you pointed out, the grid spacing might be influencing the peak composition values during LPBF. I use 0.0075 µm during L-PBF simulations, however these enrichments are actually combination of some cells, where the precipitate width is around 10-15 times higher than grid spacing. Might it be an artifact? Just for clarification, do you think I should gradually increase the grid spacing to see if they remain or not? Is there any reasonable way to mask these peak values? To be honest, having almost 40% Al enriched locally or in precipitate is quite unrealistic for my alloy.
4-Furthermore, it would be interesting for further interpretation to compare the Scheil results with the segregation curves obtained from the MICRESS simulations. Direct comparison is possible if the concentration in the solid is plotted against fraction of solid.
This is actually in my to do list, once I am sure Micress resutls are somehow in acceptable range..
thank you and regards,
Ahmet
Re: Analyzing average compositions in dendritic and interdendritic regions
Dear Ahmet,
Yes, I think that it would probably be more comparable if you start from only one planar grain, so that primary distances can evolve freely and adapt to the cooling conditions. Exactly the dendrites evolving from the grain edges in your LMD case(s) was what I meant with "dominated by the initial grain size". I think you do not need a grain boundary for your analysis (at least in the first step).
It would be very nice if you could plot the Scheil concentration curves into the same diagram together with the segregation curves obtained with MICRESS (over the whole range from fs=0 to 1)...
Bernd
Yes, I think that it would probably be more comparable if you start from only one planar grain, so that primary distances can evolve freely and adapt to the cooling conditions. Exactly the dendrites evolving from the grain edges in your LMD case(s) was what I meant with "dominated by the initial grain size". I think you do not need a grain boundary for your analysis (at least in the first step).
It would be very nice if you could plot the Scheil concentration curves into the same diagram together with the segregation curves obtained with MICRESS (over the whole range from fs=0 to 1)...
Bernd
Re: Analyzing average compositions in dendritic and interdendritic regions
Dear Bernd,
I attached the figure comparing micress and scheil Al concentration curves against relative distance to the dendrite core. Both of the curves are obtained for identical solidification velocities. As I can understand, micress predicts higher concentration in dendrite core followed by steeper increase compared to Scheil calculations. However as we mentioned earlier, this steep increase and higher amplitude might be related to very localized Al peaks during the simulation. I actually find it hard to obtain such values (~35 at.%) for Al segregation.
Also, I tried to start with planar grains however it takes sometime to obtain dendrites, which makes me loose some simulation domain to planar growth. So I extracted an initial flat grain from DP_micress in .txt format and tried to add some noise to solid/liquid interface by removing some of the solid grids in last layer of the grain and replace them with some liquid grids. Now I can save a good amount of domain (almost 5 to 6 %) however adding the noise periodically resulted in unrealistic and artificial breakdown of the planar front (attached). Do you have any suggestions how should I distribute the noise without manipulating it severely? I thought about replacing randomly 5 to 10% of the solid grids with liquid at the interface. But still, I have doubts whether this approach is legit.The breaking of planar front seems more "natural" without the noise. Therefore the question is; does spacing/magnitude/randomness of noise would impact the microstructure in the end (e.g. secondary dendrite spacing)?
Regards,
Ahmet
I attached the figure comparing micress and scheil Al concentration curves against relative distance to the dendrite core. Both of the curves are obtained for identical solidification velocities. As I can understand, micress predicts higher concentration in dendrite core followed by steeper increase compared to Scheil calculations. However as we mentioned earlier, this steep increase and higher amplitude might be related to very localized Al peaks during the simulation. I actually find it hard to obtain such values (~35 at.%) for Al segregation.
Also, I tried to start with planar grains however it takes sometime to obtain dendrites, which makes me loose some simulation domain to planar growth. So I extracted an initial flat grain from DP_micress in .txt format and tried to add some noise to solid/liquid interface by removing some of the solid grids in last layer of the grain and replace them with some liquid grids. Now I can save a good amount of domain (almost 5 to 6 %) however adding the noise periodically resulted in unrealistic and artificial breakdown of the planar front (attached). Do you have any suggestions how should I distribute the noise without manipulating it severely? I thought about replacing randomly 5 to 10% of the solid grids with liquid at the interface. But still, I have doubts whether this approach is legit.The breaking of planar front seems more "natural" without the noise. Therefore the question is; does spacing/magnitude/randomness of noise would impact the microstructure in the end (e.g. secondary dendrite spacing)?
Regards,
Ahmet
- Attachments
-
- front_with_noise.PNG (5.29 KiB) Viewed 5005 times
-
- micress_vs_scheil.jpg (15.74 KiB) Viewed 5005 times
Re: Analyzing average compositions in dendritic and interdendritic regions
Dear Ahmet,
My interpretation of the shape of the MICRESS segregation curve in comparison to Scheil is as follows:
- At the very left side (i.e. at fs=0) both curves begin at about the same concentration. This is because of the small undercooling of the initial grain at the start of the simulation. This area is retained in the simulation domain and must be considered as an artefact. There is a corresponding artefact at the top of the domain where the dendrites hit the top boundary condition. You can remove the two artefact areas simply by zooming with DP_MICRESS before doing the analysis.
- During the short initial transient, the undercooling of the front increases rapidly, leading to a shift of local equilibrium temperature and thus to an increase of the corresponding composition of Al.
- During most part of the solidification process diffusion in the liquid is not complete (as in Scheil) so that a pile-up of Al is pushed ahead. This is a quasi-stationary process, so that the solidified Al composition remains more or less constant.
- Towards the end of solidification the rest liquid gets smaller, and the diffusion pile-ups overlap. This is a more Scheil-like behavior, and the Al-composition of the solid increases (above around 80% of solidification in your case).
- At the very end one would not expect the MICRESS curve to reach the Scheil level of ~23at% Al, because it was already higher during all the initial part, and because back-diffusion should decrease it even more. Thus, I think the very last part (>99.95%) must be caused by formation of Al-rich precipitates. You can easily check the fraction of the precipitates in .TabF and their composition in .TabC. Alternatively, you could make the analysis based on the phase composition of the primary phase (.c3pha1) to be consistent with your Scheil evaluation. Another problem is that Scheil per default stops at 99%...
Your second question was about the problem of required noise when starting from a planar interface. The problem of your method is that any kind of periodic initial deviation interferes with the instability length, and thus creates strange behavior or artefacts. Therefore it is much better to use noise on the driving force to be specified in the "dG options". This noise is random and not periodic.
Unfortunately, random noise on the driving force is not very efficient because it is partly removed by averaging of the driving force, and it also strongly depends on grid resolution and time-stepping. Therefore I lately prefer to use "thermodynamic noise" which comes from slightly different linearisation parameters existing in different regions of the interface. In the latest version 7.1 of MICRESS there is an option to specify a distance over which common linearisation parameters should be used in case of "global" options. This new option is recommendable anyway for additive processes to avoid averaging over too much different temperatures (and "local" linearisation is not affordable). If the specified distance is smaller than half of the length of the planar interface, you will automatically get rapidly changing zones with slightly differing linearisation parameters, which gives you a nice and scalable long-range noise. You can monitor these linearisation scopes with the .refR output. They will change according to your relinearisation interval, or latest after the interface moved about half of its thickness.
Bernd
My interpretation of the shape of the MICRESS segregation curve in comparison to Scheil is as follows:
- At the very left side (i.e. at fs=0) both curves begin at about the same concentration. This is because of the small undercooling of the initial grain at the start of the simulation. This area is retained in the simulation domain and must be considered as an artefact. There is a corresponding artefact at the top of the domain where the dendrites hit the top boundary condition. You can remove the two artefact areas simply by zooming with DP_MICRESS before doing the analysis.
- During the short initial transient, the undercooling of the front increases rapidly, leading to a shift of local equilibrium temperature and thus to an increase of the corresponding composition of Al.
- During most part of the solidification process diffusion in the liquid is not complete (as in Scheil) so that a pile-up of Al is pushed ahead. This is a quasi-stationary process, so that the solidified Al composition remains more or less constant.
- Towards the end of solidification the rest liquid gets smaller, and the diffusion pile-ups overlap. This is a more Scheil-like behavior, and the Al-composition of the solid increases (above around 80% of solidification in your case).
- At the very end one would not expect the MICRESS curve to reach the Scheil level of ~23at% Al, because it was already higher during all the initial part, and because back-diffusion should decrease it even more. Thus, I think the very last part (>99.95%) must be caused by formation of Al-rich precipitates. You can easily check the fraction of the precipitates in .TabF and their composition in .TabC. Alternatively, you could make the analysis based on the phase composition of the primary phase (.c3pha1) to be consistent with your Scheil evaluation. Another problem is that Scheil per default stops at 99%...
Your second question was about the problem of required noise when starting from a planar interface. The problem of your method is that any kind of periodic initial deviation interferes with the instability length, and thus creates strange behavior or artefacts. Therefore it is much better to use noise on the driving force to be specified in the "dG options". This noise is random and not periodic.
Unfortunately, random noise on the driving force is not very efficient because it is partly removed by averaging of the driving force, and it also strongly depends on grid resolution and time-stepping. Therefore I lately prefer to use "thermodynamic noise" which comes from slightly different linearisation parameters existing in different regions of the interface. In the latest version 7.1 of MICRESS there is an option to specify a distance over which common linearisation parameters should be used in case of "global" options. This new option is recommendable anyway for additive processes to avoid averaging over too much different temperatures (and "local" linearisation is not affordable). If the specified distance is smaller than half of the length of the planar interface, you will automatically get rapidly changing zones with slightly differing linearisation parameters, which gives you a nice and scalable long-range noise. You can monitor these linearisation scopes with the .refR output. They will change according to your relinearisation interval, or latest after the interface moved about half of its thickness.
Bernd
Re: Analyzing average compositions in dendritic and interdendritic regions
Dear Bernd,
Thank you for your reply and comments on segregation curves!
I also though to apply noise from dG, but then I am afraid the noise will be added all the time so I aviod it. My aim was to introduce it only to the interface of the grain in the beginning so it will not take so long to form dendrites. I tried again to introduce noise to the interface by randomly adding liquid grids with a Matlab code. It indeed shortens the time interval to observe dendrites.
Coming to your suggestion I assume that the "thermodynamic noise" is added from phase interactions as follows since I couldnt find any other option relevant. Is it correct or is it introduced differently?
# Which phase diagram is to be used?
# Options: database {local|global|interface|fragment}[<maximal distance>]
# | linear | linearTQ
database global 2.5
I selected the factor of 2.5 just to give it a try. However I actually dont know how should we pick this factor. I think you explained it as follows: If the specified distance is smaller than half of the length of the planar interface, you will automatically get rapidly changing zones with slightly differing linearisation parameters, which gives you a nice and scalable long-range noise. But I am not sure what do you mean by the distance. Is it the maximum distance of planar front to the bottom of the domain before it breaks down to dendrites? Also, what is the way to calibrate/understand what is optimum? Are there any other parameters bound to effect this behavior? What does the factor stands for in this case?
I attached two images at same simluation time, considering I selected the correct option, one is obtained with above linearization settings and other is with randomy added interface noise. I also attached how was interface looking like at t=0s. The one with interface noise appears to grow less secondary dendries compared to the other simulation, but as you can see, the breaking starts earlier. I wonder if any of them is physically incorrect? I also checked interface mobilities from .driv output and process appears to be advancing correctly (driving forces close to 0).
Regards,
Ahmet
Thank you for your reply and comments on segregation curves!
I also though to apply noise from dG, but then I am afraid the noise will be added all the time so I aviod it. My aim was to introduce it only to the interface of the grain in the beginning so it will not take so long to form dendrites. I tried again to introduce noise to the interface by randomly adding liquid grids with a Matlab code. It indeed shortens the time interval to observe dendrites.
Coming to your suggestion I assume that the "thermodynamic noise" is added from phase interactions as follows since I couldnt find any other option relevant. Is it correct or is it introduced differently?
# Which phase diagram is to be used?
# Options: database {local|global|interface|fragment}[<maximal distance>]
# | linear | linearTQ
database global 2.5
I selected the factor of 2.5 just to give it a try. However I actually dont know how should we pick this factor. I think you explained it as follows: If the specified distance is smaller than half of the length of the planar interface, you will automatically get rapidly changing zones with slightly differing linearisation parameters, which gives you a nice and scalable long-range noise. But I am not sure what do you mean by the distance. Is it the maximum distance of planar front to the bottom of the domain before it breaks down to dendrites? Also, what is the way to calibrate/understand what is optimum? Are there any other parameters bound to effect this behavior? What does the factor stands for in this case?
I attached two images at same simluation time, considering I selected the correct option, one is obtained with above linearization settings and other is with randomy added interface noise. I also attached how was interface looking like at t=0s. The one with interface noise appears to grow less secondary dendries compared to the other simulation, but as you can see, the breaking starts earlier. I wonder if any of them is physically incorrect? I also checked interface mobilities from .driv output and process appears to be advancing correctly (driving forces close to 0).
Regards,
Ahmet
- Attachments
-
- global_vs_interface.PNG (270.24 KiB) Viewed 4992 times
-
- initial_inteface_noise.PNG (2.35 KiB) Viewed 4992 times
Re: Analyzing average compositions in dendritic and interdendritic regions
Dear Ahmet,
I think the question is not whether you want to use one or the other type of noise, because it is very inaccurate to use "global" as relinearisation scheme without having a distance option selected: In this case, the whole interface regions between two phases (fcc and liquid in this case) will be described using a common set of linearized thermodynamic parameters. In case of having a strong temperature gradient, this means averaging over a large temperature range!
In the older MICRESS versions the only way to prevent this (bad) approximation was either to use "local" updating of thermodynamics (i.e. for each grid cell), which is extremely time expensive and therefore only recommendable for simple binary or perhaps ternary systems, or using "fragmented" or "interface", which only joins interconnected interface parts or those of the same grain and thus solve the problem only marginally. With the new distance option, regions for common relinearisation are defined by overlapping spheres with given radius (2.5 µm in your case), so that you can easily find a good compromise between accuracy and performance. In version 7.1 you can visualize these regions using the .refR output (to be requested with "out_lin_ref") which contains a reference number for which the common linearisation parameters are given in the .TabLin output.
Thus, getting thermodynamic noise is a by-product of not using a global averaging of thermodynamics, which is bad always if there are big differences in temperature and/or composition over the domain. In your example, this averaging does not hurt at the very beginning of an isothermal flat interface, but gets worse and worse the higher the mushy zone grows. It is also possibly the explanation why the dendrites have advanced to a different height at the same time.
So, using the distance option to "global" is practically inevitable in case of large T-differences along the interfaces, and by introducing noise it perhaps solves a second problem at the same time...
Bernd
I think the question is not whether you want to use one or the other type of noise, because it is very inaccurate to use "global" as relinearisation scheme without having a distance option selected: In this case, the whole interface regions between two phases (fcc and liquid in this case) will be described using a common set of linearized thermodynamic parameters. In case of having a strong temperature gradient, this means averaging over a large temperature range!
In the older MICRESS versions the only way to prevent this (bad) approximation was either to use "local" updating of thermodynamics (i.e. for each grid cell), which is extremely time expensive and therefore only recommendable for simple binary or perhaps ternary systems, or using "fragmented" or "interface", which only joins interconnected interface parts or those of the same grain and thus solve the problem only marginally. With the new distance option, regions for common relinearisation are defined by overlapping spheres with given radius (2.5 µm in your case), so that you can easily find a good compromise between accuracy and performance. In version 7.1 you can visualize these regions using the .refR output (to be requested with "out_lin_ref") which contains a reference number for which the common linearisation parameters are given in the .TabLin output.
Thus, getting thermodynamic noise is a by-product of not using a global averaging of thermodynamics, which is bad always if there are big differences in temperature and/or composition over the domain. In your example, this averaging does not hurt at the very beginning of an isothermal flat interface, but gets worse and worse the higher the mushy zone grows. It is also possibly the explanation why the dendrites have advanced to a different height at the same time.
So, using the distance option to "global" is practically inevitable in case of large T-differences along the interfaces, and by introducing noise it perhaps solves a second problem at the same time...
Bernd