Grain Orientation

aspects of evaluating simulation results and their graphic presentation using either DisplayMICRESS or other software tools. Features and possibilities of DisplayMICRESS
RAGHAV
Posts: 45
Joined: Thu Jul 10, 2008 6:21 pm

Grain Orientation

Postby RAGHAV » Fri Oct 26, 2012 6:16 pm

Dear Mr. Bernd,

I am simulating AlCu as follows: Initial temp. at bottom = 1309 K, Soidification stopped after 2K for all cases, Cooling rates = starting from -10.75 K/s to -25.25 K/s with a interval of 0.25 K/s. This means there are 58 microstructural images for 58 different cooling rates. Here no Temperature gradient was used. Alpha and Beta were the chosen phases to appear.
15.in
Sample input file -15K/s
(17.39 KiB) Downloaded 31 times


The aim is to observe grain size and grain orientation behaviour over the varying cooling rates and use it for neural network. This cooling rate range was selected because the grain size decreased gradually in the range for the chosen set of seed density and radius.We fit the grains as ellipses and calculate grain size and orientation. Orientation is angle between general X-axis and Major axis of the ellipse and grain size was total number of pixels inside the fitted region. Both were averaged to give avg. grain size and orientation for a particular cooling rate. The nature of the variation of average grain orientation is shown in micress based data in
comp3.png
Average grain orientation
comp3.png (22.86 KiB) Viewed 848 times
.

When we look at
Individual_orien.png
Grain orientation vs number of samples
Individual_orien.png (61.06 KiB) Viewed 849 times
, we can see the orientation of individual grains for each microstructural image generated for the first 29 cooling rates. This shows that the grains are oriented from +90° to -90°.

Now, the cooling takes place homogeneously in the simulation domain as no temp. gradient is set. So shouldn't the average grain orientation be zero in all the cases. That's why i don't understand micress based avg. orientation. Why does the average grain orientation vary the way it does? What could be the reason for this?

Best Regards
raghav

Bernd
Posts: 936
Joined: Mon Jun 23, 2008 9:29 pm

Re: Grain Orientation

Postby Bernd » Sat Oct 27, 2012 10:49 pm

Dear raghav,

although I don't know exactly why you try to make an orientation evaluation of this kind, you are right that, without numerical artifacts, there should be no preferential orientation at all!
Unfortunately, it is very difficult to avoid effects of the numerical grid, especially if resolution is low and the interface thickness is small. The effect of the grid is typically to increase growth velocity in the direction of the grid (0°, 90°) and to reduce the velocity in other directions. I think you can directly see this effect in your second image which you submitted with your post.
It is easy to understand that this numerical effect changes with cooling conditions. What I do not understand yet is why this affects the average orientation - but maybe you can understand that with the exact knowledge of the averaging algorithm.
A possible explanation is the "random" selection of grain orientations: If you do not change the seeding integer in the nucleation type, the orientations of the grains will always be the same and not 0 in average because of the probably too low number of nucleated grains. Then, if the number changes systematically due to the cooling conditions, this could change the average orientation systematically...

My advice:
- increase the resolution of the simulation, although this will increase simulation time.
- also increase the interface thickness to at least 5-6 cells! Using the fd_correction of version 6.0 would further help quite a lot in this case.
- change the integers for randomization of the nucleation seed types with each simulation run

I hope this will decrease the numerical artefacts! But I cannot promise that it will be possible to remove them completely without increasing the computational effort too much...

I am interested to hear about your progress!

Best wishes

Bernd

PS: Janin is presently working on a stabilisation mechanism to decrease grid artefacts - it will probably be available in the next version!

RAGHAV
Posts: 45
Joined: Thu Jul 10, 2008 6:21 pm

Re: Grain Orientation

Postby RAGHAV » Tue Nov 06, 2012 8:40 pm

Dear Mr. Bernd,

I want to predict Grain orientation using the Neural Network for training with the data for varying cooling rates and then later also add other conditions that the grain orientations depend upon.

How could i set the integers of randomization to vary for every simulation run?

Best Regards,
Raghav

Bernd
Posts: 936
Joined: Mon Jun 23, 2008 9:29 pm

Re: Grain Orientation

Postby Bernd » Tue Nov 06, 2012 11:51 pm

Hi Raghav,

You can put whichever integers you want, they just determine some "entry point" into the calculation of random numbers! If you later plan to run MICRESS automatically with a script, you could for example just use consecutive numbers!

Bernd

RAGHAV
Posts: 45
Joined: Thu Jul 10, 2008 6:21 pm

Re: Grain Orientation

Postby RAGHAV » Mon Nov 12, 2012 9:08 pm

Hello Mr. Bernd,

The results are better for orientation but unusual as the average size almost remains constant for the four cooling rates shown below. What could be the reason for it? This was not the case earlier. I also send you the .in file
1875.in
(17.4 KiB) Downloaded 33 times
in which the nucleation set is common for all other input files as well. On purpose, i took big intervals to see the variation for the three properties below.

Cooling rate (K/s) Average size (µm) Aspect ratio Avg. grain orientation (°)
-10.75 10.765 1.664 2.3241
-14.75 11.408 1.6738 0.14621
-18.75 11.438 1.6691 0.019653
-22.75 11.164 1.6714 0.86269


Best Regards
Raghav

Bernd
Posts: 936
Joined: Mon Jun 23, 2008 9:29 pm

Re: Grain Orientation

Postby Bernd » Wed Nov 14, 2012 1:33 pm

Hi Raghav,

It is good to hear that your original problem could be at least reduced with my proposed changes!

Apart from that, the question is how your simulation conditions match to experiments. At the present, you are not taking into account any temperature effects, as you are using a given constant cooling rate. This is realistic only in special cases like strong thermal gradients (Bridgman furnace) which is not compatible with equiaxed structures.
If the samples are very small, the "DTA" approximation would be applicable, which is using a constant heat extraction rate and assuming no thermal gradient. This you could do by using "lat_heat".
In real (bigger) castings, this is also not realistic, and I propose the "Homoenthalpic" approach which I described in:

B. Böttger, J. Eiken, M.Apel, Phase-field simulation of microstructure formation in technical castings – A self-consistent homoenthalpic approach to the micro–macro problem, J. Comput. Phys. 228 (2009), 6784-6795.

This I always do if the samples are not small, i.e they are bigger than the thermal diffusion length. In MICRESS it means using a macroscopic 1d-temperature field ("1d_temp") together with "lat_heat" to get a realistic temperature solution, and to do a few iterations to get the microstructure formation consistent with the temperature field. There is a corresponding example in the MICRESS distribution (AlCu_Temp1d_dri).

Nevertheless, with the conditions you are using, a grain size dependence from cooling rate should be expected. I assume that in your case currently the checking frequency for nucleation is determining grain size! You should reduce the checking interval so far until no influence on the result is seen for the fastest cooling rate! I think it should not be a performance issue to reduce it by 2 or 3 orders of magnitude...

Bernd

RAGHAV
Posts: 45
Joined: Thu Jul 10, 2008 6:21 pm

Re: Grain Orientation

Postby RAGHAV » Fri Mar 01, 2013 9:09 pm

Dear Bernd,

The data that we have is from an experimental cast of a Al 4.5% Cu alloy, the chemical composition is (4.5 Cu, 0.05 Si, 0.07 Fe, 0.24 Mn, 0.30 Mg, 0.017 Ti, balance Al, contents in wt. %).

Although, for the simulation i only considered 4.5 Cu and neglected the traces of other elements. I hope it would make sense for the simulation by making it simpler.

The data we have with respect to the experimental work is limited to 5 points:
Cooling rate Grain size (µm) Distance (mm)
1.26 91.29 40
0.49 123 100
0.23 115.57 150
0.17 166.6 200
0.13 166.6 250

The cooling rate is refered to the average one between the interval of 400 to 600C (I have to check this information, as I do not have all the information), the distance is from the chilled end of the ingot.

The sample size is 400 * 400 micrometers^2.


Now, the questions i, keeping in view what you wrote earlier

'' In real (bigger) castings, this is also not realistic, and I propose the "Homoenthalpic" approach which I described in:
This I always do if the samples are not small, i.e they are bigger than the thermal diffusion length. In MICRESS it means using a macroscopic 1d-temperature field ("1d_temp") together with "lat_heat" to get a realistic temperature solution, and to do a few iterations to get the microstructure formation consistent with the temperature field ''

1) I have cooling rates and not heat extraction rate from the experiment, so how do then i take care of the latent heat since latent heat can only be described in combination with heat extraction rate?

2) Microstructure formation consistent with the temperature field should be seen to test if sample size is bigger than thermal diffusion length. But, how can i practically see this?

Best Regards,
Raghav

Bernd
Posts: 936
Joined: Mon Jun 23, 2008 9:29 pm

Re: Grain Orientation

Postby Bernd » Mon Mar 04, 2013 11:10 am

Hi Raghav,

If I understand correctly, the 5 samples are from one casting experiment, where temperature has been measured at the 5 points using thermocouples, right?
The information about "cooling rate" is not really a cooling rate, but rather the time after which a temperature of 400°C is reached - if you could see the temperature time curves you would notice that they are not linear.
So, it does not make sense to use this information as a "cooling rate". You should rather use a more realistic scenario by specifying the heat boundary condition at the surface of the casting!
Unfortunately, you (probably) do not have much information about the cooling conditions. So, what you have to do is to assume a reasonable boundary condition and calibrate it such that it matches your experimental data. You could decide e.g. to use a constant heat flux, a surface temperature-time curve or a constant heat transfer coefficient (see example AlCu_Temp1d_dri).

For iteration of the latent heat data, you need starting data. The most simple way to obtain them is to perform a small simulation without latent heat (e.g. constant cooling rate) and use the option "no_latheat_dsc" in order to write a .dTLat file. This file can be used for the first iteration (see AlCu_Temp1d_dri) and must be replaced iteratively by the new .dTLat files. Consistency between microstructure formation and the macroscopic temperature field is reached once the H(T) curve does not change anymore. You should not change numerical parameters (like nucleation undercooling etc.) after this calibration, or recalibrate.

If you see that the 1d-temperature field which you obtain by this method is practically flat over the whole casting (what I do not believe in your case), then the sample is not big enough compared to the thermal diffusion length. Convergence of the Iterative Homoenthalpic Approach would be difficult, and it would be better to use the DTA approach ("lat_heat" without 1d-temperature field).

Please tell me if you have more doubts or need assistance with the Homoenthalpic Approach!

Bernd

RAGHAV
Posts: 45
Joined: Thu Jul 10, 2008 6:21 pm

Re: Grain Orientation

Postby RAGHAV » Tue Mar 05, 2013 6:35 pm

Dear Bernd,

Bernd wrote:
If I understand correctly, the 5 samples are from one casting experiment, where temperature has been measured at the 5 points using thermocouples, right?


That's correct, there are thermocouples placed at the different distances of the chilled end of the ingot.

Thanks for the very well explained 1D temp. field issue wrt. thermal diffusion length.

best regards

RAGHAV
Posts: 45
Joined: Thu Jul 10, 2008 6:21 pm

Re: Grain Orientation

Postby RAGHAV » Thu Mar 07, 2013 5:11 pm

Dear Bernd,

In what case do i require the connecting points and the temperature gradient values in Z direction
if i already run my simulation with regards to the above mentioned self - consistent homoenthalpic approach.

I mean i am not understanding the difference between the homoenthalpic approach which takes care of the temp. gradients
and the temperature gradient values in Z direction.

Best Regards,
Raghav


Return to “post-processing”

Who is online

Users browsing this forum: No registered users and 1 guest