I think the following paper will be helpful in understanding the details of surface preparation of light alloys such as Aluminum, brass, zinc, etc.

Good luck

http://www.metallographic.com/Technical/Basics.pdf

Many thanks. I only weakly remember this article since it does not specifically considers the preparation for EBSD. The problem is that the very thin information layer of EBSD (a few to some ten NANO meters!) practically displays the information of the first 30 unit cells (about 60 atomic layers for fcc and bcc).

A 1µm diamond grain is 100 times bigger than the signal-forming layer!

Therefore I am asking: How do you make sure that you measure the information you are interested in (e.g. the deformation structure of your tests) and not the remaining deformation introduced during cutting, grinding and polishing?

Take a soft version (for example fully annealed) of the same chemical composition of the material. In general when it is softer the deformation during preparation is a bigger problem. If you are able to prepare fully annealed sample without deformation artefacts you can apply the same method for deformed sample. What do you think?

Dear Grzegorz, this sounds reasonable (as long as you have this option).

...I guess that once you have seen the great presentation of TSL (Matt Novell did it) where he compared sample preparation (Ni and Ti) , pattern quality and hit rate. Even their best preparation after hours of vibration polishing showed clearly visible scratches (I believe caused by "contamination" during vibration polishing), so that my conclusion is: if the orientation precison of EBSD would be better you would realize even more preparation damages...or as consequence: what do we really see?

Which kind of preparation do you use?

Dear Gert

I think that when you can see scratches then the sample is not well prepared, maybe for grain size determination and less sophisticated analyses it is enough but for misorientation analyses I would not use it.

Usually we are deeply electropolishing samples, it took long time to establish good preparation for all materials we do but it really worth it.

Last week we did a nice experiment with interrupted in-situ tensile test with EBSD scans (about 10 scans within the same area). The sample was soft austenitic stainless steel. In the initial state the misorientations within grains were below the angular resolution of the system, after some deformation we have started to see resolved misorientations.

Dear Grzegorz, how does your misorientation distribution looks like?

Or how did you evaluate the quality of the initial aiustenite. Scratches are certainly visible, but sometimes only the deformation zone around them appears later. For your misorientation measurements: do you use high resolved patterns...did you tried this already?

Best greetings

Gert

You mean misorientation distribution of whole analysed area or misorientation profiles within single grains (I can generate both to show you)?

I think I have seen that before, you mean sometimes you don't see scratches but you see elongated "deformation zones" which can propagate through few grains? It happens when last step of preparation is to short. So you are removing scratches in correct way but you should remove a little bit more material to get rid of the deformation below.

When we are not sure about the quality we look in SE for scratches, then in BSE/FSD for channelling contrast (is it changing within one grain or not). Finally using manual mouse movement we look at the pattern within few grains and we look for little pattern changes but it is only applicable to fully annealed materials.

" high resolved patterns" you mean high resolution patterns?

yes, I mean un- or low-binned patterns.

And regarding the misorientation distribution...KAM? Map? Histogram? Sorry for bothering you...

Our camera is not very fast so most often we use 120x120 px patterns.

In files you can find KAM maps + histograms and orientation deviation - angle maps because it looks nicer for me. The sample is austenite, maps are more less taken in the same area 1 is the initial state 2 and 3 are deformed. KAM is performed for 1st neighbour.

Nice images. I only guess that the small effects I am curious about are simply impossible to see when I consider your misorientation angle distribution. The general tendency is without any doubt very convincing but I assume that you would see much more (less noisy KAM) if your individual orientation determination would be more precise (e.g. 5-10 times better). You did not save the patterns, did you?

...only to show the difference...

This is a P92 after LCF. Whereas the common indexing shows the by M. Nowel in 2005 discussed orientation dependent orientation precision and possibly cause an interesting misinterpretation, everything disappears after a more careful orientation analysis. Since at the scale of 0-2° not that much is visible anymore I added the same data displayed within a scale of 0-1° (white->red->black). Since practically everything is more white or light-pink you can imagine that the misorientation (without any smoothing!) is commonly smaller than 0.1°. Such angular resolution is very beneficial...but finally brought me to this question.... If you don't see these small deviations (which show "everything") what do you really measure? Or turned around: if I can see it, how can I trust a preparation commonly applied? And what happens, if i do not see it that clearly because of a strong deformation overlaid by preparation artifacts (like grinding or cutting)? Such case I recently had, because of the expected deformation and the required high resolution I did not notice it...only the IPF maps showed a touch of unusual stripes much wider than scratches (which could be deformation bands as well). After repreparation it turned out that it was nothing else than artifacts from grinding or cutting.

We have stored the patterns (We always do so). I think we all know that we shouldn't analyse data from border of resolution of the technique and that the better resolution the better sample preparation is required. If the misorientation is smaller than 0.1° what is the resolution of your methodology?

Before I start to explain things which are better visualized, have a look at

Presentation Significant Improvement of the orientation resolution by EBS...

This is very nice technique I've already seen that presentation :). I'm thinking about weaknesses, does it have any? What is the processing speed for example for 120x120 px patterns? I know that for stored patterns speed is not as important as for acquisition in microscope. I wonder if austenite single grains will have much more misorientation than perfect silicon crystal :)

We are using 160x120 pixel images and for a map of 400x300 pixel the reanalysis takes about 5h (single phase). The map above took about 9 h. There is always potential to increase the speed but presently we are satisfied. The processing is always after the measurement, i.e. your patterns could be analyzed as well. Of course, the quality decides, how far you can go. I try to find the optimized settings for the camera. However, in the martensitic P92 I never used longer dwell time than 20ms. o frame averaging etc. This was one of the major requirement: it should work on standard measurements in order to show how much one can extract. High-resolution images are acceptable for special cases but for common investigations they are overkill. (Especially, because they are anyway binned down internally :-) without informing the user. You can use whatever you want... Otherwise, try to Hough transform a high resolution pattern. It takes several seconds!).

Weakness: Only the sample preparation since you see nearly everything and need to judge what is caused by preparation and what is material...however, using the conventional indexing you even do not see it! Therefore my question... Do we really measure what we are looking for?

I forgot your last sentence: The recommended presentation shows the limitation. I assume that the current settings for the refinements simply have their limit as shown for the Si measurement. Indeed, the later shown duplex microstructure does not show a big difference to Si. It is existent (and should exist of course) and is mainly caused by small misorientations, but possibly also be by statistically distributed dislocations which blur the pattern and limit the precision as well. However, this we did not investigate at all.

This looks very reasonably - a piece of magnificent work :). and if it will work for higher resolution patterns it will resolve two for example FCC phases with different lattice parameters?

The magic question is, how different they are. Even more interesting is the question whether the lattice parameter is important at all :-). Look, common EBSD is using the band position, and if this does not work, they try to extract the band width which has certain flaws and limits... However, pattern matching is exclusively working with intensities. It starts with the orientation derived from the EBSD but that's all. In other words: if your phase has the same lattice parameters but a certain difference in the mean periodic number, the band positions and widths are geometrically identical but their intensity distributions are not. You realize this immediately if you compare an Al pattern with a austenite pattern. You can immediately decide which one is Al but you cannot say anything meaningful about the orientation (except of recognizing the symmetry elements). This affects the pattern match in form of the cross-correlation coefficient. You will get for both phases a match, but only one shows a continuously better. I would assume that in case of Ni and austenite the differences are not that big. Might be it works but here another factor raises up: how good is the simulation? For all applications we did until now the match appeared to be perfect but there are always some deviation which are of minor worth compared to all other available simulation models. One example are HOLZ rings which require as many as possible high-indexed hkl what increases the simulation time exponentially. Therefore, the simulation is used is a compromise. Do as much you need, and not as much as possible since then the simulation time is running into the dimension of days or even weeks. If necessary, one can do this of course, but until now it was never required. Then it is better to look for other options like EDS which easily discriminates Fe and Ni :-).