Dear All,

First of all, I would like to thank Dr. Stuart Wright, Ametek who directed me to this discussion page. It was great going through the answers knowing that this topic still poses a fairly difficult issue to resolve.

We have worked extensively on the segmentation of austenite, martensite, polygonal ferrite and the ferrite in bainite in thermo-mechanically processed TRIP steels using a set of EBSD tools that we have developed over the years.

A paper explaining the method in detail was published recently in Ultramicroscopy. I have attached it to this message for your perusal.

- The paper also contains a detailed review of (almost) all methods of EBSD-based segmentation that have been applied to DP, TRIP and HSLA steel to-date.

- Since we conducted combined EBSD-EDS, we were also able to use the carbon K signal to segment ferrite in granular bainite from the bainitic ferrite laths. The EBSD-EDS data confirmed earlier atom probe work that showed carbon partitioning between the two ferrites in bainite during their growth.

The freeware tools that enabled us to accomplish the segmentation between the bainites were uploaded to: http://www.dmscripting.com and are available to all.

- The methodology is robust in that we are also able to apply it to segment the phases/ferrite morphologies under conditions of external load (for example - during a tensile test). These results are in the process of being written up.

Please do go through the paper whenever you get the time.

I would be very happy to receive any feedback/suggestions and answer comments from you.

Cheers from Australia!

Azdi

Hello Gert,

interesting question indeed... I don't have much experience with steels (too heavy for me !)

quoting from ebsd-image :

http://www.ebsd-image.org/documentation/reference/qualitymetrics.html

"The analysis of the microstructure of commercial steels is a common application for EBSD. Although diffraction patterns of ferrite grains (body centered cubic crystal structure) can easily be differentiated from retained austenite grains (face centered cubic), EBSD is incapable of separating martensite or bainite regions from the ferrite grains. Martensite or bainite are obtained by rapid cooling from the austenite phase leaving the carbon atoms (soluble in the austenite phase) trapped inside a body centered cubic crystal structure. This result is the formation of a metastable body centered tetragonal structure. The c/a ratio of the tetragonal lattice is very close to unity [3] and is therefore very close to a cubic lattice. This slight variation of the c/a ratio is undetectable in normal EBSD operating conditions. The crystal structure of martensite, bainite and ferrite are indistinguishable from one another.

However, the dislocation density of bainite or martensite is higher than that of ferrite. [4] The higher dislocation density deteriorates the quality of the diffraction pattern due to the higher lattice distortion. On average, diffraction patterns of ferrite therefore have a higher quality than those of bainite and martensite. This principle was verified via nano and microhardness measurements by Wu et al. [5]. The separation of bainite from martensite is more complex since the difference in their dislocation densities is smaller. Ryde concludes that the discrimination between these two phases could only be performed by “analyzing the directions of the carbide precipitates with a high resolution microscope”. [3]

Thinking of it, if we had a way to measure absolute elastic distortions on poor quality patterns, that would do the trick (I'm afraid the 3D-Hough transform cannot cope with diffuse Kikuchi bands). It could also be a good idea to see whether the Mean Angular Deviation (Channel's MAD) responds to the difference in c/a in a reliable way, since we are trying to index a tetragonal cell with cubic parameters, we should have a systematic error.

That is not an easy task but possible!

Lattice distortion in martensite induce dislocation density and residual stress in the crystal, results in poor diffraction pattern in EBSD, which can be detected by image quality. Sometimes the band slopes in the Kikuchi patterns can help you distinguishing bainite and martensite.

More details can be found in the paper below:

L. Ryde, Application of EBSD to analysis of microstructures in commercial steels, Mater. Sci. Tech. 22, 2006, 1297-1306

Hi,

For low carbon microstructure, where the carbon content is too low to force a change in crystal symmetry for martensite, it is problematic. There are several definitions of the two based on their appearances and grain/lath morphology, however these are often subjective and open for interpretation. Also, they do not include all morphologies that may be seen.

Regards

Thanks Claire,

I totally agree with the content described. Ferrite is often easy to seperate. Not always since also ferrite can be deformed and could have a lot of dislocations which reduce the pattern quality or band contrast. The tricky thing is the seperation of martensite and bainite... Scratches and boundaries can be filtered out by a procedure called something like "belonging to a grain" (Channel5). I guess also OIM has a similar filter.

The major problem is the lower hit rate of these deformed parts. The phase fractions of bainite and martensite are then obviously smaller than in reality. So it seems that one can perhaps "show" a tendency but the real amount (whatever it finally means for a highly-resolved microstructure) seems to be more an estimation than a usable number. Especially for high resolution measurements required for martensite. And the carbides ususlly cannot be seen in the EBSD maps...

Gert, Martensit als Phase ist selbst in unserer älteren software von Oxford Instruments vorhanden, und zwar mit a=b=2,8590 und c=3,0160. Noch besser, wenn die Tetragonalität des M. in der konkreten Legierung bekannt ist, dann kann man die a,b,c direkt eintragen. Ich hoffe, moderne EBSDs können Martensit von anderen (nichttetragonalen) Phasen auf diesem Wege unterscheiden, unser System wäre dafür zu ungenau.

in iron meteorites also cubic martensite appears since there is no carbon at all.

I also won't claim that it works sometimes under very specific conditions like for "academic steel". The problem I see is that there is no martensite and bainite as phase a user can select from a database, so that a post-processing is required which can be very specific from case to case (as Johan and Ehsan mentioned), from steel to steel, from treatment to treatment etc... The Lena Rydes paper is only one of several published over the years and they are used as leverage and proof that a seperation by EBSD must be possible. But although it has been published already 2006 there is still no general procedure which is in general accepted. Or which procedure you are applying if you get this question and your customer want to see the quantitative fractions or a nearly 100% coincidence with etched images coming from ligh microscopy?

Hi Alexander, (sorry for answering in English :-) )

then you have a very high carbon-containing martensite in your database. Of course you can insert in your database any phase. e.g. also martensite, but martensite is a phase with solid solution of C, i.e. there is not only one martensite. Moreover, martensite cannot be easily discriminated even with XRD which has a clearly higher angular resolution. The peak broadening is that big that the tetragonality is more an estimation than a real detectable effect. Even synchrotron investigations don't help since the blurring effect is intrinsic and not the result of experimental broadening caused by apertures or the wave length distribution. In your case the tetragonality is 5% and theoretically this can be discovered with low speed. However, the bands are so blurred that the EBSD system cannot determine the band positions anymore with the required accuracy (cf. Claires remark above) so that you cannot trust the automatic phase assignment. Even systematic misinterpretations caused by excess deficiency effects for some specifically oriented grains can be automatically interpreted as martensite although you only have ferrite in your sample. My believe is that with a definition of a martensite or bainite phase you cannot fix this problem.

Hello Gert, as already mentioned several times it is a intersting and sometimes challenging task. We are investigating the deformation-induced martesnite formation in a metastable austenite during different mechanical loading conditions. In some cases our initial materials contains among the austenite also some delta-ferrite. In our case it is quite easy to distinguish delta-ferrite and deformation induced martensite via the image quality and as well as via the kernal average misorientation. But in this context I found quite recently some paper where they also separate bainite from ferrite and even lower and upper bainite. As fas as I understood well the paper (I looked only shortly through) they use also the band contrast, the misorientation angle distribution as well as the kernal average misorientation. Please, find attached this paper. Looking forward to meet you again somewhere during a conference. Your Anja

Here is the mentioned attached publication concerning the separation of ferrite and bainite. maybe it helps you. Anja

Besides of confidence index of image quality, you may consider to distinguish bainite with martensite through the morphology difference. EBSD is usually employed with SEM microstructure, so you may use the EBSD grain orientation mapping and the same image-field SEM microstructure after light etching. Bainitic ferrite laths are usually different with the martensite lath structure, such as aspect ratio, misorientation angles, ...... Also, you may consider to the heat treatment history. Preferrably martensite transformation may result into coarse and long martensite laths, while the bainite transformation may result into large misorientation between laths.

Thanks for the paper Anja. I did not have it. However, I hvae another paper which is perhaps of general interest as well:

Characterisation and Quantification of Complex Bainitic Microstructures in High and Ultra-High Strength Linepipe Steels.

S. Zajac; V. Schwinn; K.-H. Tacke

Materials Science Forum Vols. 500-501 (2005) pp 387-394

Nevertheless, the important question is: Since these papers are already quite "old" and describe very well the opportunity of bainite-martensite-ferrite seperation, why there is no general rule for a seperation? Are there too many "may" and "perhaps"? Which meaning extracted numbers have if they cannot be compared between labs or even users applying the same software in the same lab?

We should start a round robin for such investigations. What do you think?

I agree with you. In my opinion there are realy too many "may" and "perhaps"! It would be fine if we could establish/find a general rule on this task. In this sense, a round robin seems to be a good idea.

Very interesting discussion, indeed. We've gone it a million times, yet, it's still unsolved. I have been wondering whether anyone has tried any of these approaches, an if yes, have you got any papers on that:

a) A pattern cross correlation approach, where you should be able to see the pattern rotation caused by the tetragonal distortion in martensite and this way be able to distinguish the phases. Well, don't know in how far bainite qualifies for that.

b) Any approach based on the reconstruction of grain boundaries and grain orientations. If you can identify bainite, and different types of bainite via optical microscopy, would it be possible to work out some microstructure identification algorithm based on the shape of the grain boundaries and grain orientations?

The correlation with optical microscope (LM) images, or before and after SEM images, is always risky because of drift during the measurement (charging), or image distortions caused by the differences between horizontal LM images and 70° tilted SEM images. One could try to use the EBSD patterns as sorce of information to reconstruct BSE images so that the image details fit exactly with the diffraction signal but this is currently still in development (PRIAS, TSL).

Well, I didn't quite mean correlation with any sort of image, but solely the information from the EBSD mapping, assuming the patterns are strong enough to be solved. In other words, a post processing routine looking for grain boundary curvature, grain orientation etc. It's definitely sophisticated, but I would think that it is possible.

About pattern cross-correlation methods, the problem is that you need a reference pattern not too far off the current pattern in term of orientation, that is it will only work within a "grain" (a domain of similar orientation) and I doubt that you'll be able to cross-correlate between a ferrite domain and martensite (or bainite) domain (except maybe in some very particular cases). You could think of using a simulated pattern as the reference, but this needs a geometrical calibration of your EBSD setup to an accuracy level yet to be achieved.

About using averaged backscattered intensity to reconstruct BSE (or other types of images), I don't consider PRIAS having the monopole on this : anyone can calculate localized averaged intensities from EBSPs, Gert you've done it, I've done it, etc.. and yes you can retrieve a wealth of information from that, the problem being the deconvolution of the different contrasts.

In a perfect world, the general definitions of the two constituents would involve quantities that is possible to measure with EBSD, based on the grain boundary network and missorientations. Since the appearance of both types can vary substantially, depending on alloy, it would be the only way to make it possible for objective measurements. However if such definitions would be made, the gain in using EBSD technique in metallography would increase substantially. If such general definitions (that includes all possible appearances) can´t be made, one can ask what the actual benefit of using that categorisation is. Why not only use the quantities that we can measure and define the microstructures from that?

I am sorry if I went to philosophical and radical....:)

The deconvolution of the BSE information superimposed in hardware BSE or 2D EBSD patterns is definitely not simple, but what in science is easy? Background correction in EDS or XRD is also not obvious and still nowadays controversely discussed. I guess we should see the benefits and critically discuss the limitations. At least in grains where no topography exists the signal can be used as better discriminator that EDS which has a worse local resolution and does not see light elements like H or Li. And even in combination with EDS it helps to seperate phases with similar chemistry.

And you are right: PRIAS does not claim a monopole. I only mentioned it as the only comertially available, integrated solution (which is not yet available but obviously soon comming).

Regarding the cross correlation I am also not very optimistic although it is more a feeling than based on a critical analysis of data. There are too many factors which affecting the pattern formation, detection, and interpretation, e.g. from my point of view pattern distortions caused by local sample charging is totally ignored. But also a systematic but wrong (better: incorrect) band detection causes a preferred selection of pseudosymmetric orientations so that it seems the system is able to differentiate between pseudosymmetric solutions but in fact it is an artifact of the applied band detection algorithm which is optimised on speed and low-resolution patterns (even if you are collecting high resolution patterns they are internally binned down (resampled) in the software in order to fit the Hough resolution).

@ Johan: If I correctly understand what you want express then it is exactly the reason why I asked this question. We can certainly develop for specific materials or conditions analytical tools. However, for a general application they are perhaps simply not satisfying. Somtimes also the questions we are asking are not expedient. Regarding the differentiation of martensite, bainite, and ferrite, EBSD as technique is probably really not the best choice, Worth to try but as long as there is no well-defined (standardized) correlation between the observed material properties, visible microstructure, and extracted EBSD data, an application held out in prospective is only dubious. I would appreciate sometimes a paper which would demonstrate when a technique fails than to claim by only one selected example that a developed algorithm delivers any kind of result.

Such a tool exists. Distributions of inter- variant misorientations in martensite and bainite of low carbon steel differ clearly (N. Takayama et al. Acta Materialia 60 (2012) 2387–2396). We tested this method of separation for otherf low carbon steels for conditions of continuous cooling (e.g., Metal Science and Heat Treatment, Vol. 55, Nos. 9 – 10, 2014, 550-558): it works.

Dear Nikolay, is there any chance to get the paper from you (e.g. a preprint or galley proof)?

Nikolay: That is a great paper and a good example on how EBSD can be used on these microstructures, since it focus on their grain morphology. However, there are still a lot more to do to get more general definitions.

I am eager to read the recommended papers but I already feared an answer like that. Nevertheless, never lose your hope. Thanks Johan for your short comment.

Gert, sorry for delayed answer. I can easily send you both papers, but I don't know how to do this. Let me know you e-mail or other way of sending.

Hi Nikolay, my e-mail- address is: [email protected]

I interested mainly in your paper. The Takayama paper I already have.

Many thanks.

Dear All,

First of all, I would like to thank Dr. Stuart Wright, Ametek who directed me to this discussion page. It was great going through the answers knowing that this topic still poses a fairly difficult issue to resolve.

We have worked extensively on the segmentation of austenite, martensite, polygonal ferrite and the ferrite in bainite in thermo-mechanically processed TRIP steels using a set of EBSD tools that we have developed over the years.

A paper explaining the method in detail was published recently in Ultramicroscopy. I have attached it to this message for your perusal.

- The paper also contains a detailed review of (almost) all methods of EBSD-based segmentation that have been applied to DP, TRIP and HSLA steel to-date.

- Since we conducted combined EBSD-EDS, we were also able to use the carbon K signal to segment ferrite in granular bainite from the bainitic ferrite laths. The EBSD-EDS data confirmed earlier atom probe work that showed carbon partitioning between the two ferrites in bainite during their growth.

The freeware tools that enabled us to accomplish the segmentation between the bainites were uploaded to: http://www.dmscripting.com and are available to all.

- The methodology is robust in that we are also able to apply it to segment the phases/ferrite morphologies under conditions of external load (for example - during a tensile test). These results are in the process of being written up.

Please do go through the paper whenever you get the time.

I would be very happy to receive any feedback/suggestions and answer comments from you.

Cheers from Australia!

Azdi