Typically the precision (not accuracy) of EBSD measurements is assumed to be 0.5...2 degrees which is certainly a very rough description since it depends on many factors like pattern and Hough resolution, defects of the microstructure, the number of bands used during indexing, acquisition conditions etc. Recently even precisions of 0.2 degrees and better are discussed but their general reliability is not easy to judge. What is your opinion about the reachable precision of standard EBSD measurements (pattern resolution typically 80x60 pixels)? And what do you think about the necessary accuracy?
I use the EBSD method for a characterization of metals and alloys that are subjected to a large plastic deformation. For such samples an "orientation noise" (or fuzziness of Kikuchi lines) is rather high, thus I normally don't use misorientations below 2 degrees upon an analysis of the grain boundary character distribution. This value is commonly reported in the literature. I think that a very low misorientation (e.g. 0.2 deg) may be applied only for a material that posses a high quality of crystal lattice (namely is not strained or defected). In other cases it may lead to a production of "fake" boundaries.
Kind regards
Very good question!
Down to 0.3 I have seen and can rely on the results of the paper below.
http://www.sciencedirect.com/science/article/pii/S0921509310000195
But how to achieve this needs a bit of experience I believe. Sample type, SEM beam stability, and many other issues as you mentioned can affect this. Depending on what type of analysis you want to obtain, the accuracy can be important, for instance precision is very important when investigating the coherency between two phases in a microstructure...
Any experience you have to share?
Hi Woijciech, for your heavily deformed material the determination of bands and so the precision of local orientation measurements are perhaps really worse than 0.5 degree, what does not mean that the local orientation of the crystalline volume is not better detectable, e.g. using a smaller probe (lower kV) or TKD. The other question of course is, how meaningful a more accurate orientation determination is if the material is highly deformed. Nevertheless, for the evaluation of deformation mechanisms in highly deformed materials it might be a very interesting topic, e.g. for a mapping of the dislocation density.
When I am talking about common conditions I assume not a perfect but also not a very strongly deformed material, i.e. something you get presented if you visit an EBSD company and they want impress you by nice data. This is comparable to many mineralogical applications but also in weldings etc. I must admit that this is not very typical to many EBSD users, but it represents a situation under which everybody can check his system since these materials are everywhere available. In a second part one can certainly discuss, what happens for heavily deformed materials, and which application a more precise crystal orientation would offer in this case. One could reduce this to the question: Would a better precision help you, or not?
Excellent discussion and topic! I'm surprised at the relatively low inter-grain angular resolution (0.3 to 0.5 degrees) possible presently with EBSD. I would have thought a lot higher. For instance, using the real time XRD rocking curve method both in forward and back scattered reflections one can resolve miss-orientations in fractions of arc sec or lower. It is limited by the resolution of the optics used and the 2D XRD detector characteristics.
My fundamental understanding of the EBSD method is weak. So if you gentle folk would post some links, PDF files or other instructive material, I'd be much obliged. Open access links/info would be preferred over those with membership requirements. Thanx!
What is the typical maximum sample size for EBSD work? Just as SEM or TEM?
Can the rocking curve method of oscillating the sample about the "Bragg" condition be used to more precisely resolve small miss-orientations in EBSD?
Dear Ravi, a sample size for the EBSD measurements is similar to that for a normal SEM observations. However, You should noted that a low magnification scans (taken from a relatively large area) are acquired with a higher step scan, so You should adjust it to the size of structural features in Your samples.
Kind regards
Hi, this is a very interesting question that I recently also have been wondering about. However, it is not very clear to me whether you are all talking about precision (reproducibility), or about angular resolution, although both should probably be discussed together?
At the moment I’m working on maps of deformation lamellae in quartz showing point-to-point misorientations often smaller than 1°. Now I’m not sure about the angular resolution that we should take into account when looking at these small misorientations. I think this is a very important question as it can make my results meaningless!
Nice article comparing EBSD with conventional XRD data Wojciech! I'm still trying to absorb all the acronyms:-)
P.S: I'm very much interested in the deconvolution of Nano structural data from super lattice XRD peaks (reciprocal space data) on a VOXEL-by-VOXEL basis for the specimen including poly-crystalline, mono-crystalline and epitaxial materials.
Article EBSD and X-ray diffraction study on the recrystallization of...
Hi Ravi, in fact you are right. The angular resolution can be better, but since the major topic is the orientation determination EBSD mainly benefits from the huge solid angle which enables to cover some very low-indexed zone axis. Usually the vertical size of a screen is so close that it covers vertically 70...90 degrees. Very helpful for a fast indexing of phases with any symmetry. However, the second requirement on EBSD is speed which brought people to the idea to minimize the used patern size. For fast measurement and maximum speed all systems work with a vertical image size of less than 100 pixels, e.g. 60 or 48. This finally means that a pixel size represents between 1 and 2 degrees. The precision of orientation measurements can reach a higher precision only because many bands (5 to 12) are used which automatically means that the result is in the scale of subpixels. Therefore a typical (experimentally proved) estimation is about 0.5 degrees. It is clear that in case of a longer sample-detector distance the captured solid angle is smaller and the visible details becomes clearer so that a higher precision can be reached. The only problems are: you cannot perhaps determine the orientation anymore, and the acquisition takes dramatically longer since for a collection you need now more time. If you know the orientation you can get comparable precisions as typical for channeling patterns. A few SEM manufacturer started to reinvent this beam-movement mode again. :-)
Regarding the scannable size I want to go a bit further in detail. You can combine stage and beam scan to areas which are only limited by the motorized size of the stage and - because of the 70 degree tilt - by the design of the pole piece. Therefore, along the tilt directions only the motorization limits the scanned size. Perpendicular it is often clearly smaller, but still in the scale of 2 or 3 cm so that finally maps of 10x 3 or 15x4cm are possible. Problematic is only the distortions as result of the high tilt angle and a non-proper alignment of the sample which is the classicle error. Therefore many of these stitched maps look strange since the single fields do not match that nicely as they should. A better way would be a scan of large areas, cf. the link. But they are of course limited in size but for many investigations big enough since we are dealing with SEMs which are commonly used for higher resolution applications.
Conference Paper Large area EBSD mapping by single-scan technique
Hi Tine, your comment is correct. It is important to differentiate between precision (the reproducability) and the accuracy of an orientation, which is related e.g. to an ideal mounting of the sample since actually all values are related to the stage and not to the sample, or to the calibration of your system. Here I want focus on the precision, i.e. the orientation "noise" within a perfect grain. Only in this case you can determine the misorientation with a maximum accuracy. The angular resolution is from my point of view an adequate term for precision since it tells you how precise you are measuring the wrong (:-) or hopefully correct) value. In order to increase the resolution (and actually then also the precision) you probably use higher resolving images. Unfortunately the higher resolved images won't help that much since they are binned down internally the software in order to do a Hough transform in a minimal time. To this aim your patterns will be even binarized, i.e. the nice 12bit deep images are lost for any orientation analysis within the systems (and you have no chance to change this). The only option you have is the increase of the Hough resolution but as you perhaps know, this number is also limited (because it makes no sense to increase this number when your original images are anyway of already bad quality). If you like to get a much higher accuracy, from my point of vioew you should apply cross correlation which is mainly linked with the term strain analysis by HiRes EBSD but can be also applied for smaller images. The results are amazing if you compare the results with EBSD maps based on orientation determinations using band detection. I did this also for quartz, cf. the shown link. The comparison with standard EBSD presentation tools like Schmid factor or misorientation maps display that the application of cross correlation even on these low-reduced images will cause clear benefits. My explanation is the use of the original image and not of only some bands which are of less statistical significance and already erroneous. Therefore I don't think itis a good idea to expect too much from orientation data which already containing unpredictable systematic or non-systematic errors in band detection.
Conference Paper EBSD patterns: a mine of information
Ravi, I still forgot to say: the "rocking courve" technique to see even very small misorientations is also reinvented recently. The technique is called ECCI (Electron Channeling Contrast Imaging) and uses a certain Bragg condition for a single grain, e.g. mainly applied for dislocation imaging in monocrystals. There is (as far as I know) no quatification of the misorientation possible. But all these information are part of EBSD patterns and could be extracted if we would take more care in investigation of the original images, cf. the attached link. The so-called forescattered images which can be colourized show amazing information which we are still not able to quantify. Their orientation resolution - see the SEM image of a meteoritic microstructure - is breath-taking and cannot be reached at the moment by EBSD maps. And the funny thing is that this information is somewhere hidden in the EBSD patterns.
Conference Paper EBSD patterns: a mine of information
Hi!
For the question I normally want to answer when I use EBSD, I think the amount of data is better to use to improve the certainty in the measurement, compared to improve accuracy or precision.
Let say for example that you want to see if there is a difference in the misorientation distribution in two microstructures and one of your microstructures has a larger presence of a specific grain boundary, resulting in a larger peak in the misorientation distribution compared to the other microstructure.
Your certainty in the quantification of such a difference would then be how accurate you could quantify the peak area in the distribution. The precision in the EBSD measurement would also here influence the width of the peak, but since it is the peak area that actually are interested in, I think an improvement of the statistics is the best way to go. In this example, the accuracy (peak position) might not be that important if you can be certain that the peak you see is the GB that you are interested in to measure and that you have recorded the two data sets under the same condition. Of coarse, a too poor precision can create overlaps when distinguish between specific GB, but I think the precision possible today is sufficient in most cases.
Also, in my experience, EBSD is more used (or should be used) to measure general aspects of microstructures instead of a specific feature in one specific position in the microstructure. Therefore, larger data sets is generaly good since it increases the certainty that your measurement represents the whole sample.
I would therfore argue that for the general material science community that today uses EBSD, it would be better to improve, if one has to choose :), the technique in making it easier to record larger data sets with high accuracy (stage movement, measureing the actual tilt angle etc.) than to going for improving the precision in the measurements.
Hi Johan, it depends on what you want do with EBSD. For your application of grain or phase boundary characterization and the connected uncertainties a better precision is indeed not very helpful. This is similar to texture measurements where the results in a post-processing step will be convoluted (smeared :-)) using a Gaussian with a FWHM of 10 degrees. Then it is really a waste. However, if your orientation relationship wouldn't change that much and you could correlate properties with even tiny variation in OR or in habit plane you would also call for better precission, especially if you could get them "for free" since they were already visible in your currently collected patterns. Only because of the weak but fast band detection you loose this precision. And you have no chance to change this situation at the moment since there is no option in your software.
That is true. When it comes to the OR, an improved precision would be definetly improve the overall measurement, so yes, I would want that too :).
I just meant on more general scale. I still believe that the development of the technique when it comes to the image recording and indexing are a little bit ahead, compared to stage and automation issues as well as post-processing. But of coarse, all improvements are wanted and I guess there things that can be done with the softwares and to improve the algoritms in the Hough-processing.
Regarding the OR (orientation relationships) it would already help if one would definitely know that the observed "scattering" of orientations are NOT a result of the band detection and derived orientation determination. But in fact this is the same for problems which are related to misorientations within a grain (Tines comments) and not only between adjacent grains.
Yes, that is true.
However, I think that it is also dependent on to what extend one uses EBSD to get a quantified result. My argument that better statistics can compensate for precision is valid only when your final result is a value/distribution. If one use the technique for a qualitative result, e.g. visual comparing two maps with eachother, an improvement of the precision would be much more effective.
A good example maybe is a local distribution map, where an improved precision gives a direct visual improvement in the map and a sharper local misorientation distribution. Both would be good, but it would probably improve any certainty of the result more in the former than the latter.
But as you maybe know, I feel very strongly to get EBSD to be used more quantitatively instead of qualitatively and with better statistics. This might influence my opinion.. :)
There is a relatively simple technique for estimating the precision of an EBSD system: measure the misorientation across a perfectly defined boundary like an annealing twin boundary in fcc crystals (eg Cu). The exact misorientation is 60.0° so a set of EBSD measurements on one side and a second set of measurements on the other gives you 60.0 +/- a spread. A statistical analysis of the spread is then possible to give you the precision. This of course has nothing to do with true accuracy. I think some examples of the method have been published but I cannot recall the exact reference.
Hi Julian, I guess similar experiences have been made by everyone (using Si crystals or sigma 3 boundaries). But as you know, the precision depends on many factors like acquisition time, pattern and Hough resolution, sample prep, acceleration voltage, not talking about the system or detector used. I wonder, is there something like a theoretical limit, and if we need more shouldn't we perhaps think into another direction instead of improving the band detection by longer acquisition. It is also related to the question Tine is worried about (and publications of EDAX-TSL about the influence of superimposing patterns along grain boudaries) where conclusions made from measured misorientations are used to explain geometrically necessary dislocations but in fact it seems so that the band detection only becomes worse the closer you are coming to boundaries. Immediately the question arises: Is this the only source or errors, or is it the only we presently know? It would be nice to get an information about the reliability of each orientation measurement. MAD as single value is not sufficient and only a very rough guidance level. CI (confidence index) only tells us, how reliable the given resolution is in comparison to a second one.
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Materials Science