A correlative approach to segmenting phases and ferrite morphologies in transformation-induced plasticity steel using electron back-scattering diffraction and energy dispersive X-ray spectroscopy 19

Prévia do material em texto

<p>A correlative approach to segmenting phases and ferrite morphologies</p><p>in transformation-induced plasticity steel using electron back-scattering</p><p>diffraction and energy dispersive X-ray spectroscopy</p><p>Azdiar A. Gazder a,n, Fayez Al-Harbi b, Hendrik Th. Spanke b,</p><p>David R.G. Mitchell a, Elena V. Pereloma a,b</p><p>a Electron Microscopy Centre, University of Wollongong, New South Wales 2500, Australia</p><p>b School of Mechanical, Materials and Mechatronic Engineering, University of Wollongong, New South Wales 2522, Australia</p><p>a r t i c l e i n f o</p><p>Article history:</p><p>Received 18 April 2014</p><p>Received in revised form</p><p>18 June 2014</p><p>Accepted 6 July 2014</p><p>Available online 19 July 2014</p><p>Keywords:</p><p>Transformation-induced plasticity (TRIP)</p><p>Steel</p><p>Electron back-scattering diffraction (EBSD)</p><p>Energy dispersive X-ray spectroscopy (EDS)</p><p>Transmission electron microscopy (TEM)</p><p>Carbon partitioning</p><p>Bainite</p><p>a b s t r a c t</p><p>Using a combination of electron back-scattering diffraction and energy dispersive X-ray spectroscopy</p><p>data, a segmentation procedure was developed to comprehensively distinguish austenite, martensite,</p><p>polygonal ferrite, ferrite in granular bainite and bainitic ferrite laths in a thermo-mechanically processed</p><p>low-Si, high-Al transformation-induced plasticity steel. The efficacy of the ferrite morphologies</p><p>segmentation procedure was verified by transmission electron microscopy. The variation in carbon</p><p>content between the ferrite in granular bainite and bainitic ferrite laths was explained on the basis of</p><p>carbon partitioning during their growth.</p><p>& 2014 Elsevier B.V. All rights reserved.</p><p>1. Introduction</p><p>Advanced high strength transformation-induced plasticity</p><p>(TRIP) steels were mainly developed for automotive applications</p><p>as they possess high strength–ductility ratios, formability and</p><p>energy absorption properties [1,2]. They are characterised by a</p><p>complex multiphase microstructure comprising retained auste-</p><p>nite, martensite, polygonal ferrite and carbide-free bainites [3–5].</p><p>Although there are different terminologies in use [6–11], it is</p><p>generally accepted that during the continuous cooling or isother-</p><p>mal holding of low carbon steels, the formation of intermediate</p><p>austenite decomposition products (between diffusional ferrite/</p><p>pearlite and diffusionless martensite) occurs. In TRIP steels, they</p><p>are usually termed as granular bainite and bainitic ferrite. Here</p><p>granular bainite is defined as carbide-free bainite with irregular-</p><p>shaped ferrite or ferrite plates and dispersed blocky martensite/</p><p>retained austenite constituent. On the other hand, bainitic ferrite is</p><p>the arrangement of ferrite laths separated by layers of retained</p><p>austenite and/or martensite [7,10–12]. Both ferrites in these two</p><p>carbide-free bainitic morphologies exhibit a much higher disloca-</p><p>tion density than polygonal ferrite as well as a supersaturation in</p><p>carbon [13–17].</p><p>Under conventional electron back-scattering diffraction (EBSD)</p><p>acquisition conditions, TRIP steel microstructures are typically</p><p>indexed as iron fcc (austenite) and bcc (ferrite). Depending on</p><p>the TRIP steel alloy composition and thermo-mechanical processing</p><p>history, the various bcc phases (martensite and polygonal ferrite)</p><p>and ferrite morphologies (ferrite in granular bainite and bainitic</p><p>ferrite laths) then need to be further segmented during the post-</p><p>processing of the EBSD map. However, to-date the lack of a</p><p>comprehensive method that consistently distinguishes between</p><p>the above phases/ferrite morphologies poses a significant hurdle to</p><p>furthering our understanding of the complex interplay between</p><p>them during loading.</p><p>Over the past 15 years, the methods to segment phases/ferrite</p><p>morphologies have relied on various analytical tools that either:</p><p>(i) quantify the conditions under which the electron back-</p><p>scattering pattern (EBSP) was acquired, or (ii) make use of the</p><p>quality metrics of the acquired EBSP after Hough transformation</p><p>(Table 1). The parameters that describe the conditions under</p><p>which the EBSP was acquired are the least used and include the</p><p>confidence index (CI) and the pattern misfit angle (PM). The CI</p><p>Contents lists available at ScienceDirect</p><p>journal homepage: www.elsevier.com/locate/ultramic</p><p>Ultramicroscopy</p><p>http://dx.doi.org/10.1016/j.ultramic.2014.07.005</p><p>0304-3991/& 2014 Elsevier B.V. All rights reserved.</p><p>n Corresponding author. Tel.: þ61 2 4221 5904; fax: þ61 2 4221 3114.</p><p>E-mail address: azdiar@uow.edu.au (A.A. Gazder).</p><p>Ultramicroscopy 147 (2014) 114–132</p><p>www.sciencedirect.com/science/journal/03043991</p><p>www.elsevier.com/locate/ultramic</p><p>http://dx.doi.org/10.1016/j.ultramic.2014.07.005</p><p>http://dx.doi.org/10.1016/j.ultramic.2014.07.005</p><p>http://dx.doi.org/10.1016/j.ultramic.2014.07.005</p><p>http://crossmark.crossref.org/dialog/?doi=10.1016/j.ultramic.2014.07.005&domain=pdf</p><p>http://crossmark.crossref.org/dialog/?doi=10.1016/j.ultramic.2014.07.005&domain=pdf</p><p>http://crossmark.crossref.org/dialog/?doi=10.1016/j.ultramic.2014.07.005&domain=pdf</p><p>mailto:azdiar@uow.edu.au</p><p>http://dx.doi.org/10.1016/j.ultramic.2014.07.005</p><p>involves a Kikuchi band triplet voting scheme such that within</p><p>a given inter-planar angular tolerance, the ratio between the</p><p>candidate orientation with the highest number of votes and the</p><p>total number of votes is regarded as the most likely solution [18].</p><p>Once a solution is selected, the PM is used to calculate the mean</p><p>angular deviation between the positions of the simulated and</p><p>experimental EBSPs.</p><p>On the other hand, the quality metrics of the acquired EBSP</p><p>that are derived from Hough transformation include the image</p><p>quality (IQ, also known as the pattern quality (PQ) or band contrast</p><p>(BC)) and the band slope (BS). The IQ/PQ/BC defines the average</p><p>intensity of the Hough peaks [19] whereas the BS denotes the</p><p>average slope of the intensity change between the Hough peaks</p><p>and their surrounding background [20]. In practice, the IQ/PQ/BC</p><p>and BS are greyscaled and binned to a byte range between 0</p><p>(black) to 255 (white). Structures with elastically distorted lattices,</p><p>higher density of crystalline defects or residual stresses (causatives</p><p>that can be linked to the transformation of austenite to bainite or</p><p>martensite) present with blurred Kikuchi band edges, diffuse</p><p>Hough peaks and appear darker with lower IQ/PQ/BC and BS</p><p>values [21]. Conversely, polygonal ferrite presents with sharper</p><p>Kikuchi band edges, more intense Hough peaks and has higher IQ/</p><p>PQ/BC and BS values.</p><p>The IQ/PQ/BC are the most commonly used parameters to</p><p>distinguish between features with varying dislocation density by</p><p>thresholding the distribution between areas of low and high</p><p>contrast. In order to accomplish this semi-quantitatively, the</p><p>thresholding procedure relies on the presence of a clear and</p><p>specific inversion point between individual peaks of the IQ/PQ/</p><p>BC distribution. For example, in the case of a bimodal distribution,</p><p>the threshold is conventionally defined as the lowest value</p><p>between the two distinct peaks. Taking advantage of this statistical</p><p>peculiarity, one of the first EBSD studies on Fe–1.57Mn–1.46Si–</p><p>0.91C and Fe–1.57Mn–1.23Al–0.34Si–0.31C (wt%1) TRIP steels by</p><p>De Meyer et al. [22] used the IQ/PQ/BC to distinguish the ferrite in</p><p>bainite from polygonal ferrite. The same technique was used to</p><p>observe/quantify the volume (or area) fractions of: (i) polygonal</p><p>ferrite and martensite in Fe–3.28Ni–0.12C [23,24], Fe–0.09C dual</p><p>phase [25], and Fe–1.8Mn–1.51Si–0.2C quench and partitioned</p><p>TRIP steels [26], (ii) polygonal ferrite and the ferrite in bainite in</p><p>Fe–1.5Mn–1.5Si–0.2C [27], Fe–1.48Mn–1.08Al–0.28Si–0.27C [28]</p><p>and Fe–1.6Mn–1.28Si–0.12C [29] TRIP steels, and (iii) the ferrite</p><p>in bainite and martensite in Fe–1.5Mn–1.5Si–0.6C [30] and Fe–</p><p>1.43Si–0.58Mn–0.56C–0.47Cr SAE 9254 steels [31].</p><p>The BS parameter has been applied less often. Kwon et al. [32]</p><p>used the BS to distinguish the ferrite in bainite from polygonal</p><p>ferrite in austempered Fe–1.5Mn–1.5Si–0.2C TRIP steel. The low BS</p><p>of the ferrite in bainite was ascribed to its formation during the</p><p>Table 1</p><p>The types of EBSD-based segmentation procedures</p><p>regard, carbon exhibits a binding</p><p>energy of 0.75 eV to the core of edge dislocations [69] such that</p><p>the other sinks include dislocations and their arrays forming</p><p>boundaries within bainite. Atom probe studies of Nanobain steel</p><p>[17] after low temperature transformation clearly showed the</p><p>segregation of carbon with concentrations reaching �14 at% at</p><p>dislocation networks in ferrite laths adjacent to retained austenite.</p><p>It follows that the formation of Cottrell atmospheres will restrict</p><p>the re-arrangement of dislocation structures into lower energy</p><p>configurations; as seen in the case of the ferrite in granular bainite</p><p>which comprises dislocation tangles and has the highest defect</p><p>density (Fig. 9b). On the other hand, the lower carbon content in</p><p>bainitic ferrite laths (due to its easier escape to the nearby retained</p><p>austenite) allows for the process of dislocation re-arrangement to</p><p>proceed such that a more ordered substructure consisting of</p><p>dislocation cells and a lower defect density is returned (Fig. 9c).</p><p>The above can also result in lower BC and BS values (Fig. 7b–e) and</p><p>much higher internal misorientation (Fig. 7f) in the ferrite in</p><p>granular bainite compared to the bainitic ferrite laths.</p><p>As seen in Figs. 6 and 7, we have exploited the above variation</p><p>in the relative carbon counts to effectively segment the ferrite in</p><p>granular bainite and the bainitic ferrite laths. However, it is re-</p><p>emphasised that the present understanding of carbon partitioning</p><p>behaviour is only based on two-dimensional EBSD maps. Addi-</p><p>tional three-dimensional EBSD-EDS studies are required to verify</p><p>the differences in relative carbon counts and bainitic morphologies</p><p>and the exact role the location of the martensite/retained auste-</p><p>nite constituent plays in facilitating carbon partitioning.</p><p>6. Conclusions</p><p>A multi-condition segmentation methodology that distin-</p><p>guishes austenite, martensite, polygonal ferrite and bainite in a</p><p>reproducible manner was developed for a thermo-mechanically</p><p>processed low-Si, high-Al TRIP steel. Accurate discrimination</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132130</p><p>between martensite, the ferrite in bainite and polygonal ferrite is</p><p>possible from EBSD-only data when the grain orientation spread</p><p>criterion is applied. However, the use of the relative variation in</p><p>carbon counts from EDS mapping remains indispensable to seg-</p><p>menting the ferrite in granular bainite and bainitic ferrite laths.</p><p>Based on the differences in morphological and internal structure,</p><p>this study suggests that the location of the martensite/retained</p><p>austenite constituent may influence the behaviour of carbon during</p><p>the growth of the two bainites.</p><p>Acknowledgements</p><p>The authors are grateful to Prof. Bruno C. De Cooman (GIFT-</p><p>POSTECH) for providing the source material and Dr. Andrii</p><p>Kostryzhev (UOW) for preparing the FIB section for the TEM study.</p><p>This work was funded in part by the Engineering Materials</p><p>Institute at UOW. The JEOL JSM-7001F FEG-SEM and the JEOL</p><p>JEM-200F FEG-STEM were funded by the Australian Research</p><p>Council (ARC) – Linkage, Infrastructure, Equipment and Facilities</p><p>(LIEF) Grants LE0882613 and LE120100104, respectively. F. Al-</p><p>Harbi and H. Spanke acknowledge the Saudi Basic Industries</p><p>Corporation (SABIC) and BlueScope Steel Ltd. for sponsoring their</p><p>Ph.D. and undergraduate thesis studies, respectively.</p><p>References</p><p>[1] Y. Sakuma, O. Matsumura, H. Takechi, Mechanical properties and retained</p><p>austenite in intercritically heat-treated bainite-transformed steel and their</p><p>variation with Si and Mn additions, Metall. Trans. A 22 (1991) 489–498.</p><p>[2] K.-I. Sugimoto, M. Kobayashi, S.-I. Hashimoto, Ductility and strain-induced</p><p>transformation in a high-strength transformation-induced plasticity-aided</p><p>dual-phase steel, Metall. Trans. A 23 (1992) 3085–3091.</p><p>[3] O. Matsumura, Y. Sakuma, H. Takechi, Enhancement of elongation by retained</p><p>austenite in intercritical annealed 0.4C–1.5Si–0.8Mn steel, Trans. ISIJ 27 (1987)</p><p>570–579.</p><p>[4] B.C. De Cooman, Structure–properties relationship in TRIP steels containing</p><p>carbide-free bainite, Curr. Opin. Solid State Mater. Sci. 8 (2004) 285–303.</p><p>[5] I.B. Timokhina, P.D. Hodgson, E.V. Pereloma, Effect of microstructure on the</p><p>stability of retained austenite in transformation-induced-plasticity steels,</p><p>Metall. Mater. Trans. A 35 (2004) 2331–2341.</p><p>[6] Y. Ohmori, H. Ohtani, T. Kunitake, The bainite in low carbon low alloy high</p><p>strength steels, Trans. ISIJ 11 (1971) 250–259.</p><p>[7] S.W. Thompson, D.J. Colvin, G. Krauss, Continuous cooling transformations and</p><p>microstructures in a low-carbon, high-strength low-alloy plate steel, Metall.</p><p>Trans. A 21 (1990) 1493–1507.</p><p>[8] B.L. Bramfitt, J.G. Speer, A perspective on the morphology of bainite, Metall.</p><p>Trans. A 21 (1990) 817–829.</p><p>[9] G. Krauss, S.W. Thompson, Ferritic microstructures in continuously cooled</p><p>low- and ultralow-carbon steels, ISIJ Int. 35 (1995) 937–945.</p><p>[10] S. Zajac, J. Komenda, P. Morris, P. Dierickx, S. Matera, F. Penalba-Diaz,</p><p>Quantitative structure–property relationships for complex bainitic steels,</p><p>Commission of European Communities Technical Steel Research, ECSC Report</p><p>EUR 21245 EN Contract No-7210-PR/345, Directorate-General for Research and</p><p>Innovation, European Commission, Luxembourg (2005) 157.</p><p>[11] S. Zajac, V. Schwinn, K.H. Tacke, Characterisation and quantification of</p><p>complex bainitic microstructures in high and ultra-high strength linepipe</p><p>steels, Mater. Sci. Forum 500–501 (2005) 387–394.</p><p>[12] M. Takahashi, H.K.D.H. Bhadeshia, A model for the microstructure of some</p><p>advanced bainitic steels mater, Trans. Jpn. Inst. Met. 32 (1991) 689–696.</p><p>[13] H.K.D.H. Bhadeshia, A.R. Waugh, Bainite: an atom-probe study of the incom-</p><p>plete reaction phenomenon, Acta Metall. 30 (1982) 775–784.</p><p>[14] E.V. Pereloma, I.B. Timokhina, M.K. Miller, P.D. Hodgson, Three-dimensional</p><p>atom probe analysis of solute distribution in thermomechanically processed</p><p>TRIP steels, Acta Mater. 55 (2007) 2587–2598.</p><p>[15] F.G. Caballero, M.K. Miller, C. Garcia-Mateo, Carbon supersaturation of ferrite</p><p>in a nanocrystalline bainitic steel, Acta Mater. 58 (2010) 2338–2343.</p><p>[16] I.B. Timokhina, E.V. Pereloma, S.P. Ringer, R.K. Zheng, P.D. Hodgson, Character-</p><p>ization of the bake-hardening behavior of transformation induced plasticity</p><p>and dual-phase steels using advanced analytical techniques, ISIJ Int. 50 (2010)</p><p>574–582.</p><p>[17] F.G. Caballero, M.K. Miller, C. Garcia-Mateo, Atom probe tomography analysis</p><p>of precipitation during tempering of a nanostructured bainitic steel, Metall.</p><p>Mater. Trans. A 42 (2011) 3660–3668.</p><p>[18] TexSEM Laboratories Inc., TSL OIM analysis user manual, Version 5.2 (2007)</p><p>443.</p><p>[19] S.I. Wright, M.M. Nowell, EBSD image quality mapping, Microsc. Microanal. 12</p><p>(2006) 72–84.</p><p>[20] L. Ryde, Application of EBSD to analysis of microstructures in commercial</p><p>steels, Mater. Sci. Technol. 22 (2006) 1297–1306.</p><p>[21] T. Maitland, S. Sitzman, Electron backscatter diffraction (EBSD) technique and</p><p>materials characterization examples, in: W. Zhou, Z.L. Wang (Eds.), Scanning</p><p>Microscopy for Nanotechnology: Techniques and Applications, 1st ed.,</p><p>Springer, London, UK, 2007, pp. 41–75.</p><p>[22] M. De Meyer, L. Kestens, B.C. De Cooman, Texture development in cold rolled</p><p>and annealed C–Mn–Si and C–Mn–Al–Si TRIP steels, Mater. Sci. Technol. 17</p><p>(2001) 1353–1359.</p><p>[23] A.W. Wilson, G. Spanos, Application of orientation imaging microscopy to</p><p>study phase transformations in steels, Mater. Charact. 46 (2001) 407–418.</p><p>[24] A.W. Wilson, J.D. Madison, G. Spanos, Determining phase volume fraction in</p><p>steels by electron backscattered diffraction, Scr. Mater. 45 (2001) 1335–1340.</p><p>[25] B.-Y. Jeong, R. Gauvin, S. Yue, EBSD study of martensite in a dual phase steel,</p><p>Microsc. Microanal. 8 (2002) 700–701.</p><p>[26] W.-J. Feng, Z.-B. Wu, L. Wang, J.G. Speer, Effect of testing temperature on</p><p>retained austenite stability of cold rolled cmnsi steels treated by quenching</p><p>and partitioning process, Steel Res. Int. 84 (2013) 246–252.</p><p>[27] A. Wasilkowska, R. Petrov, L. Kestens, E.A. Werner,</p><p>C. Krempaszky, S. Traint,</p><p>A. Pichler, Microstructure and texture changes in a low-alloyed TRIP-aided</p><p>steel induced by small plastic deformation, ISIJ Int. 46 (2006) 302–309.</p><p>[28] E. Emadoddin, A. Akbarzadeh, R. Petrov, L. Zhao, Anisotropy of retained</p><p>austenite stability during transformation to martensite in a TRIP-assisted</p><p>steel, Steel Res. Int. 84 (2013) 297–303.</p><p>[29] E. Emadoddin, Effect of cold rolling on annealing behavior and retained</p><p>austenite characteristics of multiphase CMnSi steel, ISIJ Int. 53 (2013)</p><p>330–336.</p><p>[30] C. Cabus, H. Réglé, B. Bacroix, Orientation relationship between austenite and</p><p>bainite in a multiphased steel, Mater. Charact. 58 (2007) 332–338.</p><p>[31] J.A. Cruz, T.F.M. Rodrigues, V.D.C. Viana, H. Abreu, D.B. Santos, Influence of</p><p>temperature and time of austempering treatment on mechanical properties of</p><p>SAE 9254 commercial steel, Steel Res. Int. 83 (2012) 22–31.</p><p>[32] E.P. Kwon, S. Fujieda, K. Shinoda, S. Suzuki, Characterization of transformed</p><p>and deformed microstructures in transformation induced plasticity steels</p><p>using electron backscattering diffraction, Mater. Sci. Eng. A 528 (2011)</p><p>5007–5017.</p><p>[33] T. Waterschoot, L. Kestens, B.C. De Cooman, Hot rolling texture development</p><p>in CMnCrSi dual-phase steels, Metall. Mater. Trans. A 33 (2002) 1091–1102.</p><p>[34] J.-H. Wu, P.J. Wray, C.I. Garcia, M.-J. Hua, A.J. Deardo, Image quality analysis: a</p><p>new method of characterizing microstructures, ISIJ Int. 45 (2005) 254–262.</p><p>[35] J.-H. Wu, Development of a new technique to identify and quantify complex</p><p>austenite decomposition products, PhD thesis, Swanson School of Engineer-</p><p>ing, University of Pittsburgh, Pittsburgh (2005) 161.</p><p>[36] J.-H. Wu, P.J. Wray, C.I. Garcia, M.-J. Hua, A.J. DeArdo, On achieving a better</p><p>understanding of the polygonal ferrite microstructure in IF steel using image</p><p>quality analysis, Mater. Manuf. Process. 22 (2007) 281–285.</p><p>[37] R. Petrov, L. Kestens, A. Wasilkowska, Y. Houbaert, Microstructure and texture</p><p>of a lightly deformed TRIP-assisted steel characterized by means of the EBSD</p><p>technique, Mater. Sci. Eng. A 447 (2007) 285–297.</p><p>[38] L. Mujica, S. Weber, H. Pinto, C. Thomy, F. Vollertsen, Microstructure and</p><p>mechanical properties of laser-welded joints of TWIP and TRIP steels, Mater.</p><p>Sci. Eng. A 527 (2010) 2071–2078.</p><p>[39] S.I. Wright, M.M. Nowell, D.P. Field, A review of strain analysis using electron</p><p>backscatter diffraction, Microsc. Microana.l 17 (2011) 316–329.</p><p>[40] J.-Y. Kang, D.-H. Kim, S.-I. Baik, T.-H. Ahn, Y.-W. Kim, H.N. Han, K.H. Oh, H.</p><p>C. Lee, S.H. Han, Phase analysis of steels by grain-averaged EBSD functions, ISIJ</p><p>Int. 51 (2011) 130–136.</p><p>[41] S. Zaefferer, P. Romano, F. Friedel, EBSD as a tool to identify and quantify</p><p>bainite and ferrite in low-alloyed Al-TRIP steels, J. Microsc. 230 (2008)</p><p>499–508.</p><p>[42] E.M. Lehockey, Y.-P. Lin, O.E. Lepik, Mapping residual plastic strain in materials</p><p>using electron backscatter diffraction, in: A.J. Schwartz, M. Kumar, B.L. Adams</p><p>(Eds.), Electron backscatter diffraction in materials science, 1st ed.,Kluwer</p><p>Academic Press, New York, USA, 2001, pp. 247–264.</p><p>[43] O. Man, L. Pantělejev, Z. Pešina, EBSD analysis of phase compositions of TRIP</p><p>steel on various strain levels, Mater. Eng. 16 (2009) 15–21.</p><p>[44] S. Zaefferer, P. Romano, Attempt to identify and quantify microstructural</p><p>constituents in low-alloyed TRIP steels by simultaneous EBSD and EDS</p><p>measurements, Microsc. Microanal. 13 (2007) 944–945.</p><p>[45] K.-Y. Zhu, D. Barbier, T. Iung, Characterization and quantification methods of</p><p>complex BCC matrix microstructures in advanced high strength steels, J.</p><p>Mater. Sci. 48 (2013) 413–423.</p><p>[46] S.L. Shrestha, A.J. Breen, P. Trimby, G. Proust, S.P. Ringer, J.M. Cairney, An</p><p>automated method of quantifying ferrite microstructures using electron</p><p>backscatter diffraction (EBSD) data, Ultramicroscopy 137 (2014) 40–47.</p><p>[47] F. Alharbi, A.A. Gazder, A. Kostryzhev, B.C. de Cooman, E.V. Pereloma, The</p><p>effect of processing parameters on the microstructure and mechanical proper-</p><p>ties of low-Si transformation-induced plasticity steels, J. Mater. Sci. 49 (2014)</p><p>2960–2974.</p><p>[48] D.R.G. Mitchell, Scripting Resources for DigitalMicrograph, 2013.</p><p>[49] A.A. Gazder, W.Q. Cao, C.H.J. Davies, E.V. Pereloma, An EBSD investigation of</p><p>interstitial-free steel subjected to equal channel angular extrusion, Mater. Sci.</p><p>Eng. A 497 (2008) 341–352.</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132 131</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref1</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref1</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref1</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref2</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref2</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref2</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref3</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref3</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref3</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref4</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref4</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref5</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref5</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref5</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref6</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref6</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref7</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref7</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref7</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref8</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref8</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref9</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref9</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref11</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref11</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref11</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref12</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref12</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref13</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref13</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref14</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref14</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref14</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref15</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref15</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref16</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref16</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref16</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref16</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref17</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref17</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref17</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref19</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref19</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref20</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref20</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref864</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref864</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref864</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref864</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref22</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref22</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref22</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref23</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref23</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref24</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref24</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref25</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref25</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref26</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref26</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref26</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref27</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref27</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref27</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref28</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref28</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref28</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref29</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref29</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref29</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref30</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref30</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref31</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref31</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref31</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref32</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref32</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref32</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref32</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref33</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref33</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref34</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref34</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref36</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref36</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref36</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref37</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref37</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref37</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref38</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref38</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref38</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref39</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref39</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref40</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref40</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref40</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref41</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref41</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref41</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref852</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref852</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref852</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref852</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref43</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref43</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref44</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref44</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref44</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref45</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref45</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref45</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref46</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref46</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref46</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref47</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref47</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref47</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref47</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref48</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref48</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref48</p><p>[50] A.A. Gazder, M. Sánchez-Araiza, J.J. Jonas, E.V. Pereloma, Evolution of recrys-</p><p>tallization texture in a 0.78 wt.% Cr extra-low-carbon steel after warm and</p><p>cold rolling, Acta Mater. 59 (2011) 4847–4865.</p><p>[51] S.S. Hazra, E.V. Pereloma, A.A. Gazder, Microstructure and mechanical proper-</p><p>ties after annealing of equal-channel angular pressed interstitial-free steel,</p><p>Acta Mater. 59 (2011) 4015–4029.</p><p>[52] A.A. Gazder, A.A. Saleh, E.V. Pereloma, Microtexture analysis of cold-rolled and</p><p>annealed twinning-induced plasticity steel, Scr. Mater. 65 (2011) 560–563.</p><p>[53] A.A. Gazder, V.Q. Vu, A.A. Saleh, P.E. Markovsky, O.M. Ivasishin, C.H.J. Davies, E.</p><p>V. Pereloma, Recrystallisation in a cold drawn low cost beta titanium alloy</p><p>during rapid resistance heating, J. Alloys Compd. 585 (2014) 245–259.</p><p>[54] S.I. Wright, A review of automated orientation imaging microscopy, J . Comp.-</p><p>Assist. Microsc. 5 (1993) 207–221.</p><p>[55] S. Zaefferer, J. Ohlert, W. Bleck, A study of microstructure, transformation</p><p>mechanisms and correlation between microstructure and mechanical proper-</p><p>ties of a low alloyed TRIP steel, Acta Mater. 52 (2004) 2765–2778.</p><p>[56] P. Hovington, D. Drouin, R. Gauvin, CASINO: a new monte carlo code in C</p><p>language for electron beam interactions—Part I: description of the program,</p><p>Scanning 19 (1997) 1–14.</p><p>[57] D. Drouin, P. Hovington, R. Gauvin, CASINO: a new monte carlo code in C</p><p>language for electron beam interactions—Part II: tabulated values of the mott</p><p>cross section, Scanning 19 (1997) 20–28.</p><p>[58] P. Hovington, D. Drouin, R. Gauvin, D.C. Joy, N. Evans, CASINO: a new monte</p><p>Carlo code in C language for electron beam interactions—Part III: stopping</p><p>power at low energies, Scanning 19 (1997) 29–35.</p><p>[59] D. Drouin, A.R. Couture, D. Joly, X. Tastet, V. Aimez, R. Gauvin, CASINO V2.42—a</p><p>fast and easy-to-use modeling tool for scanning electron microscopy and</p><p>microanalysis users, Scanning 29 (2007) 92–101.</p><p>[60] H.K.D.H. Bhadeshia, Bainite in steels: transformations, microstructure and</p><p>properties, 2nd ed., IOM Communications Ltd., London, UK (2001) 735.</p><p>[61] I. Stark, G.D.W. Smith, H.K.D.H. Bhadeshia, The distribution of substitutional</p><p>alloying elements during the bainite transformation, Metall. Trans. A 21 (1990)</p><p>837–844.</p><p>[62] E. Pereloma, H. Beladi, L.-C. Zhang, I. Timokhina, Understanding the behavior</p><p>of advanced high-strength steels using atom probe tomography, Metall. Mater.</p><p>Trans. A 43 (2012) 3958–3971.</p><p>[63] F.G. Caballero, M.K. Miller, C. Garcia-Mateo, J. Cornide, M.J. Santofimia,</p><p>Temperature dependence of carbon supersaturation of ferrite in bainitic</p><p>steels, Scr. Mater. 67 (2012) 846–849.</p><p>[64] F.G. Caballero, M.K. Miller, C. Garcia-Mateo, J. Cornide, New experimental</p><p>evidence of the diffusionless transformation nature of bainite, J. Alloys Compd.</p><p>577 (Suppl. 1) (2013) S626–S630.</p><p>[65] H.K.D.H. Bhadeshia, Carbon in cubic and tetragonal ferrite, Philos. Mag. 93</p><p>(2013) 3714–3725.</p><p>[66] N. Takayama, G. Miyamoto, T. Furuhara, Atomistic effects in migrating</p><p>interphase interfaces – recent progress and future study, in: 2012 TMS Annual</p><p>Meeting and Exhibition, Orlando, FL, USA, 2012.</p><p>[67] M. Peet, S.S. Babu, M.K. Miller, H.K.D.H. Bhadeshia, Three-dimensional atom</p><p>probe analysis of carbon distribution in low-temperature bainite, Scr. Mater.</p><p>50 (2004) 1277–1281.</p><p>[68] I.B. Timokhina, H. Beladi, X.Y. Xiong, Y. Adachi, P.D. Hodgson, Nanoscale</p><p>microstructural characterization of a nanobainitic steel, Acta Mater. 59</p><p>(2011) 5511–5522.</p><p>[69] A.W. Cochardt, G. Schoek, H. Wiedersich, Interaction between dislocations and</p><p>interstitial atoms in body-centered cubic metals, Acta Metall. 3 (1955)</p><p>533–537.</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132132</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref49</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref49</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref49</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref49</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref50</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref50</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref50</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref51</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref51</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref52</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref52</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref52</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref987</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref987</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref54</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref54</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref54</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref55</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref55</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref55</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref56</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref56</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref56</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref57</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref57</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref57</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref58</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref58</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref58</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref846</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref846</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref60</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref60</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref60</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref61</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref61</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref61</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref62</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref62</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref62</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref63</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref63</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref63</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref64</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref64</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref65</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref65</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref65</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref66</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref66</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref66</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref67</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref67</p><p>http://refhub.elsevier.com/S0304-3991(14)00128-4/sbref67</p><p>A correlative approach to segmenting phases and ferrite morphologies in transformation-induced plasticity steel using...</p><p>Introduction</p><p>Experimental and analytical procedure</p><p>EBSD and EDS-based phase and ferrite morphology segmentation</p><p>Transmission electron microscopy</p><p>Discussion</p><p>The use of internal misorientation criteria</p><p>The drawbacks and disadvantages of using EBSP quality metrics for segmentation</p><p>The use of elemental carbon in segmenting the ferrite morphologies in the two bainites</p><p>On the formation of bainite</p><p>Conclusions</p><p>Acknowledgements</p><p>References</p><p>undertaken to-date on multi-phase steels.</p><p>Segmentation method</p><p>Steel</p><p>type</p><p>Steel composition (wt%) Phases/constituents Ref.</p><p>Thresholding the distribution of one parameter</p><p>IQ/PQ/BC TRIP Fe–1.57Mn–1.46Si–0.91C and Fe–</p><p>1.57Mn–1.23Al–0.34Si–0.31C</p><p>Polygonal ferrite, ferrite in bainite [22]</p><p>Fe–1.8Mn–1.51Si–0.2C Polygonal ferrite, martensite [26]</p><p>Fe–1.5Mn–1.5Si–0.2C Polygonal ferrite, ferrite in bainite [27]</p><p>Fe–1.48Mn–1.08Al–0.28Si–0.27C Polygonal ferrite, ferrite in bainite [28]</p><p>Fe–1.6Mn–1.28Si–0.12C Polygonal ferrite, ferrite in bainite [29]</p><p>Fe–1.5Mn–1.5Si–0.6C Ferrite in bainite, martensite [30]</p><p>Fe–1.9Si–1.43Mn–0.21C Ferrite in bainite, martensite [43]</p><p>DP Fe–3.28Ni–0.12C Polygonal ferrite, martensite [23,24]</p><p>Fe–0.09C Polygonal ferrite, martensite [25]</p><p>SAE 9254 Fe–1.43Si–0.58Mn–0.56C–0.47Cr Ferrite in bainite, martensite [31]</p><p>BS LC bainite – Polygonal ferrite, ferrite in bainite, martensite [20]</p><p>TRIP Fe–1.5Mn–1.5Si–0.2C Polygonal ferrite, ferrite in bainite [32]</p><p>(Sub)grain ECD DP – Polygonal ferrite, martensite [20]</p><p>3rd near neighbour KAM TRIP Fe–1.5Al–1.5Mn–0.2C Polygonal ferrite, ferrite in bainite [41]</p><p>2nd near neighbour KAM TRIP Fe–1.9Si–1.43Mn–0.21C Ferrite in bainite, martensite [43]</p><p>Normalised EDS carbon counts TRIP Fe–1.5Al–1.5Mn–0.2C Polygonal ferrite, ferrite in bainite [44]</p><p>Multi-peak modelling of the distribution of one parameter</p><p>IQ/PQ/BC IF Fe–0.15Mn–0.002C Polygonal, non-polygonal, acicular and bainitic ferrite,</p><p>martensite, carbon-rich micro-constituents</p><p>[34–36]</p><p>DP Fe–1.55Mn–1.09Al–0.15C</p><p>HSLA Fe–1.3Mn–0.078C</p><p>TRIP Fe–1.5Mn–1.5Si–0.2C–0.2Ni Proeutectoid ferrite, ferrite in bainite [37]</p><p>Fe–23.94Mn–0.86Cr–0.51C–0.28Si–</p><p>0.14Ni</p><p>Polygonal ferrite, ferrite in bainite, martensite [38]</p><p>Thresholding the distributions of two parameters</p><p>CI and IQ/PQ/BC or IQ/PQ/BC and (sub) grain size</p><p>or BS and (sub) grain size</p><p>DP – Polygonal ferrite, martensite [20]</p><p>CI and IQ/PQ/BC – Fe–1.39Mn–0.69Cr–0.1Si–0.08C Polygonal ferrite, ferrite in bainite, martensite [33]</p><p>Average (sub)grain IQ/PQ/BC and BS – Fe–2.2Mn–1.0Si–0.06C Polygonal ferrite, martensite [40]</p><p>Multi-peak modelling of average</p><p>(sub) grain IQ/PQ/BC and GAM</p><p>TRIP Fe–1.8Mn–0.5Si–0.2C Polygonal ferrite, ferrite in bainite, martensite [40]</p><p>Thresholding the distributions of multiple parameters</p><p>Average (sub) grain BS, GOS,</p><p>(sub) grain aspect ratio and area</p><p>– Fe–1.9Mn–0.2Si–0.2Cr–0.15C and</p><p>Fe–0.93Mn–0.7Cr–0.2Si–0.07C</p><p>Polygonal ferrite, ferrite in bainite, martensite [45]</p><p>–</p><p>Boundary misorientation profiles, (sub)grain size,</p><p>aspect ratio and average internal misorientation</p><p>CASTRIP Nb-free, 0.04 Nb, and 0.08 Nb Polygonal, non-polygonal and acicular ferrite, ferrite in</p><p>bainite</p><p>[46]</p><p>1 Throughout the text, chemical compositions are in weight per cent unless</p><p>specified otherwise. With the exception of C, elements o0.1 wt% are not stated.</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132 115</p><p>austempering process. The ferrite in bainite contained a higher</p><p>density of geometrically necessary dislocations as a result of the</p><p>local shear stress concentration induced by the volume expansion</p><p>during transformation. In the same study, the martensite formed</p><p>during tensile loading up to 3% engineering strain was manually</p><p>identified using the BC maps. However, the polygonal ferrite, the</p><p>ferrite in bainite and the martensite fractions were not individu-</p><p>ally segmented.</p><p>It is more often the case that the IQ/PQ/BC or BS return</p><p>asymmetric distributions with a single peak; following which</p><p>the phase/ferrite morphology segmentation becomes significantly</p><p>more difficult. In this situation, Waterschoot et al. [33] used a</p><p>combination of CI and IQ/PQ/BC to qualitatively distinguish</p><p>between polygonal ferrite, the ferrite in bainite and martensite</p><p>in Fe–1.39Mn–0.69Cr–0.1Si–0.08C steel. Here structures with the</p><p>highest CI and IQ/PQ/BC values were considered polygonal ferrite</p><p>whereas structures with intermediate CI and IQ/PQ/BC values</p><p>were ferrite in bainite and the ones with the lowest CI and IQ/</p><p>PQ/BC values were denoted as martensite. The classification was</p><p>based on the polygonal ferrite grains having the lowest dislocation</p><p>density, the ferrite in the bainite containing a higher dislocation</p><p>density as a result of their isothermal transformation from</p><p>austenite and the martensite grains possessing the highest defect</p><p>density due to the strain associated with their transformation. The</p><p>martensite grains also recorded the lowest CI (or largest PM) as the</p><p>nominal bct crystal2 deviates slightly from a perfect bcc crystal</p><p>[33].</p><p>In order to overcome issues related to segmentation in asym-</p><p>metric distributions with a single peak, Wu et al. [34–36] sug-</p><p>gested a mathematical multi-peak model. Here the IQ/PQ/BC</p><p>values were first normalised following which the distribution</p><p>was deconvoluted into multiple, symmetric Gaussian sub-</p><p>distributions. The number of sub-distributions was the same as</p><p>the number of phases/ferrite morphologies present such that the</p><p>sum of their peaks was equal to that of the original single</p><p>asymmetric peak. The model successfully distinguished between</p><p>polygonal, non-polygonal, acicular and bainitic ferrite, martensite</p><p>and carbon-rich micro-constituents in IF, dual phase and high-</p><p>strength-low-alloy steels; with the separate phases/ferrite</p><p>morphologies verified by micro and nano-hardness measurements</p><p>[34,35]. Thereafter, Petrov et al. [37] used the multi-peak model</p><p>to distinguish ferrite in bainite from proeutectoid ferrite in</p><p>Fe–1.5Mn–1.5Si–0.2C–0.2Ni TRIP steel. This study confirmed that</p><p>the volume fractions from multi-peak modelling were similar to</p><p>those obtained from the image processing of the optical micro-</p><p>graphs after tint etching, conventional thresholding of the IQ/PQ/BC</p><p>distribution and from magnetic saturation experiments. Mujica</p><p>et al. [38] made use of the multi-peak model assumption that the</p><p>IQ/PQ/BC distribution was a superposition of sub-distributions</p><p>belonging to the various constituent phases/ferrite morphologies</p><p>to manually threshold and segment the polygonal ferrite, the ferrite</p><p>in bainite and martensite in Fe–23.94Mn–0.86Cr–0.51C–0.28Si–0.14Ni</p><p>TRIP steel. Verification of the segmentation was undertaken semi-</p><p>quantitatively by comparing the boundarymisorientation distributions</p><p>of the individual phases/constituents.</p><p>Ryde [20] reviewed the segmentation procedures used by</p><p>various research laboratories via a blind round-robin test on a</p><p>standard set of low carbon bainitic and dual phase steels. Poly-</p><p>gonal ferrite, the ferrite in bainite and martensite were segmented</p><p>in the low carbon bainitic steel by thresholding their BS values;</p><p>with polygonal ferrite having the largest average values whereas</p><p>the ferrite in bainite and the martensite had intermediate and the</p><p>lowest BS values, respectively. On the other hand, martensite was</p><p>distinguished from polygonal ferrite in a dual phase steel [20] on</p><p>the basis of: (i) its grain size, such that martensitic substructures</p><p>had o2 mm equivalent circle diameter (ECD) and a majority of</p><p>o151 or 4501 boundary misorientation angles due to their fast</p><p>transformation rate, or (ii) a combination of IQ/PQ/BC and (sub)</p><p>grain size thresholding for critical (sub)grain boundary misorien-</p><p>tations of 1.51 and 51, respectively, or (iii) a combination of IQ/PQ/</p><p>BC and CI thresholding as per Ref. [33], or (iv) a combination of BS</p><p>and (sub)grain size. In that study, it was noted that method</p><p>(i) would only work when a clear (sub)grain size difference exists</p><p>between martensite and polygonal ferrite and that method (iv)</p><p>was disadvantaged by the need to have successfully indexed</p><p>pixels; which is not always the case for poorly indexing phases</p><p>like martensite.</p><p>Another issue with the IQ/PQ/BC and BS parameters is that the</p><p>pixels at grain boundary interfaces nominally present with smaller</p><p>values as a result of the combined EBSP from a diffracting volume</p><p>that contains contributions from neighbouring but differently</p><p>oriented substructures [39]. In order to reduce grain boundary</p><p>effects, the IQ/PQ/BC and BS values of (sub)grains can</p><p>be averaged</p><p>such that variations within individual (sub)grains are lost but the</p><p>ability to compare between (sub)grains is enhanced [34,39]. Kang</p><p>et al. [40] successfully used this strategy to segment martensite</p><p>from polygonal ferrite in Fe–2.2Mn–1.0Si–0.06C dual phase steel.</p><p>After imposing a 51 critical boundary misorientation criterion and</p><p>averaging the pattern quality within each (sub)grain, an IQ/PQ/BC</p><p>distribution containing two distinct peaks with minimal overlap</p><p>was generated; following which the two phases were easily</p><p>segmented.</p><p>In order to overcome the above limitations associated with the</p><p>IQ/PQ/BC or BS parameters, Zaefferer et al. [41] suggested using</p><p>the third nearest neighbour kernel average misorientation (KAM3)</p><p>and distinguished ferrite in bainite from polygonal ferrite in an</p><p>intercritically annealed Fe–1.5Al–1.5Mn–0.2C TRIP steel. The KAM</p><p>captures short range, in-grain orientation gradients and is calcu-</p><p>lated as the average of the misorientation between the pixel at the</p><p>centre of the kernel and the individual pixels at the perimeter of</p><p>the kernel; all of which must belong to the same (sub)grain [42]. In</p><p>Ref. [41], the KAM threshold was determined mathematically as</p><p>that value where the boundary interface between the polygonal</p><p>ferrite and the ferrite in bainite was smooth and no scattered</p><p>pixels belonging to either ferrite morphology appeared within the</p><p>individual fractions/subsets. Substructures with KAM values less</p><p>than the threshold were designated as polygonal ferrite whereas</p><p>those with KAM values greater than or equal to the threshold were</p><p>quantified as ferrite in bainite.</p><p>Man et al. [43] tracked the deformation in Fe–1.9Si–1.43Mn–</p><p>0.21C TRIP steel subjected to ex-situ uniaxial tensile loading up to</p><p>�23% engineering strain. In order to segment the martensite</p><p>islands from the “grainy” polygonal ferrite and the ferrite in</p><p>bainite, the efficacies of the IQ/PQ/BC thresholding [34,35,37]</p><p>and the second-nearest neighbour KAM schemes were evaluated.</p><p>With increasing strain, IQ/PQ/BC thresholding resulted in errors</p><p>caused by surface relief effects whereas the initially bimodal-like</p><p>KAM distribution of the bcc phases/ferrite morphologies evolved</p><p>into a broad single peak distribution. As a consequence, the</p><p>authors concluded that the second-nearest neighbour KAM</p><p>scheme was only applicable to undeformed or slightly deformed</p><p>TRIP steel microstructures.</p><p>2 In bcc crystals, the lattice parameters are equal such that a¼b¼c. For bct</p><p>crystals, the lattice parameters are a¼bac. In modern EBSD acquisition systems,</p><p>accurate distinction between bcc ferrite and bct martensite is only possible when</p><p>there is a Z10% difference in the “c” lattice parameter value.</p><p>3 In order to calculate KAM, the misorientation between the centre pixel and</p><p>its surrounding neighbours has to be lower than the user-defined critical (sub)grain</p><p>boundary angle. Note here that the size of a kernel is defined by its nth nearest</p><p>neighbour.</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132116</p><p>Zaefferer et al. [44] also took advantage of the developments in</p><p>detector hardware/software integration to obtain EBSD and energy</p><p>dispersive X-ray spectroscopy (EDS) information simultaneously</p><p>during the mapping of Fe–1.5Al–1.5Mn–0.2C TRIP steel. Since the</p><p>EDS information was obtained concurrently as the electron beam</p><p>rastered over the sample area during EBSD mapping, the variation</p><p>in the relative elemental counts between pixels was attributed to</p><p>differences in chemistry. Consequently, the relative carbon counts</p><p>at every pixel were normalised to the total (nominal) carbon</p><p>content of the TRIP steel in order to enhance the differences in the</p><p>carbon distribution between austenite, polygonal ferrite and the</p><p>ferrite in bainite.</p><p>The experience gained from single or two step segmentation</p><p>methodologies over the past decade has led to the realisation that</p><p>accurate and reliable phase/ferrite morphology segmentation in</p><p>TRIP steels is only possible when multi-condition schemes are</p><p>applied. In this regard, one of the first such studies was by Kang</p><p>et al. [40] on Fe–1.8Mn–0.5Si–0.2C TRIP steel. Since the raw IQ/PQ/</p><p>BC distribution was asymmetric and comprised a single peak,</p><p>(sub)grain pattern quality averaging was unable to segment the</p><p>various phases/ferrite morphologies. Consequently, a combination</p><p>of (sub)grain pattern quality averaging, Multi-peak modelling and</p><p>grain average misorientation (GAM4) was used to distinguish</p><p>martensite and the ferrite in bainite from polygonal ferrite. Here</p><p>the GAM denotes short range in-(sub)grain orientation variations</p><p>and is defined as the mean misorientation between adjacent</p><p>in-grain pixel pairs.</p><p>More recently, Zhu et al. [45] developed a multi-condition</p><p>segmentation procedure for intercritically annealed Fe–1.9Mn–</p><p>0.2Si–0.2Cr–0.15C and thermo-mechanically processed Fe–0.93Mn–</p><p>0.7Cr–0.2Si–0.07C steels. After imposing a 21 critical boundary</p><p>misorientation criterion, a combination of EBSP quality metrics,</p><p>internal misorientation and morphological criteria were used to</p><p>segment the equiaxed polygonal ferrite, the ferrite in bainite and</p><p>martensite in the two steels. First, the (sub)grain band slope average</p><p>was applied to roughly segment the martensite from the ferrite in</p><p>bainite and the polygonal ferrite. Thereafter, the ferrite in bainite</p><p>was segmented from the polygonal ferrite using the grain orienta-</p><p>tion spread (GOS5) parameter by thresholding such that substruc-</p><p>tures with GOS o1.51 were classified as polygonal ferrite whereas</p><p>those with GOS Z1.51 were ferrite in bainite. From the polygonal</p><p>ferrite fraction (i) the residual ferrite in bainite was removed using</p><p>a (sub)grain aspect ratio (42.5) threshold, and (ii) the residual</p><p>tempered martensite was removed using a (sub)grain area</p><p>(o0.5 mm2) threshold.</p><p>The latest development in automated identification used a</p><p>combination of EBSD and Matlab to quantify the area fractions of</p><p>polygonal ferrite, ferrite in bainite and acicular ferrite in CASTRIPs</p><p>steel [46]. The study made use of boundary misorientation profiles</p><p>and (sub)grain size, aspect ratio and average internal misorienta-</p><p>tion to return an area fraction of the various ferrite morphologies.</p><p>While this technique provides distinct advantages over earlier</p><p>manual point counting methods, the results cannot be correlated</p><p>back to substructures in the EBSD map as it does not inherently</p><p>identify a (sub)grain (or pixel grouping) as belonging to any</p><p>particular phase/ferrite morphology.</p><p>With the above outlook in mind, the present study is the first to</p><p>develop a multi-condition segmentation methodology that distin-</p><p>guishes austenite, martensite, polygonal ferrite and the ferrite in</p><p>bainite in a reproducible manner for a thermo-mechanically</p><p>processed low-Si, high-Al TRIP steel. As opposed to short-range</p><p>in-grain orientation variations like KAM and GAM which return</p><p>localised misorientation gradients within single (sub)grains and</p><p>are sensitive to the map step size, this study implements the long-</p><p>range GOS criterion to perform the initial segmentation. It will be</p><p>shown that this criterion provides for more effective segmentation</p><p>between various substructure types as it assigns a single value to</p><p>all pixels within a particular (sub)grain. This is also one of the first</p><p>studies to utilise the concurrently acquired EBSD and EDS infor-</p><p>mation to successfully distinguish the ferrite in granular bainite</p><p>from the bainitic ferrite laths. While the method described in the</p><p>following paragraphs should not be automated (as due care and</p><p>attention to phase/ferrite morphology detail needs to be taken for</p><p>each map), the novel procedures/tools developed here can be</p><p>readily applied to study the microstructural variations in a wide</p><p>variety of engineering alloys.</p><p>2. Experimental and analytical procedure</p><p>An Fe–0.15C–2.00Mn–0.30Si–1.00Al–0.05P wt% TRIP steel was</p><p>received as a 6 mm thick hot rolled plate from GIFT-POSTECH. The</p><p>plate was electro-discharge machined into an 8 (RD)�20 (TD)�6</p><p>(ND) mm3 sample and processed on a Gleeble 3500 thermo-</p><p>mechanical simulator operating in hydra-wedge mode as follows</p><p>[47].</p><p>The sample was heated at 2 K s�1 to 1523 K, held for 120 s</p><p>followed by cooling at 1 K s�1 to 1373 K where a 25% roughing</p><p>reduction was applied. The sample was then held for 120 s in order</p><p>to condition the recrystallised austenite and then cooled down to</p><p>the finish rolling temperature in the non-recrystallised austenite</p><p>region of 1123 K. Following a second 47% finishing reduction, the</p><p>sample was slow cooled at 1 K s�1 to the accelerated cooling start</p><p>temperature of 953 K to form �50% polygonal ferrite. At 953 K,</p><p>the cooling rate was increased to 20 K s�1 to avoid pearlite</p><p>formation. Finally, coiling was simulated by holding the sample</p><p>at 743 K for 1200 s to form bainite and then water quenched.</p><p>The sample was again electro-discharge machined from the</p><p>centre of its width along the normal direction–rolling direction</p><p>(ND–RD) and mechanically fine ground using 15 and 6 mm dia-</p><p>mond stages. Thereafter, a 0.5 cm2 area of the sample was</p><p>electropolished on a Struers Lectropol-5 using an electrolyte of</p><p>330 ml methanolþ330 ml butoxyethanolþ40 ml perchloric acid</p><p>at 50 V, �0.95–1.2 mA, 17 1C for 90 s.</p><p>EBSD and EDS information was obtained simultaneously from a</p><p>96.425�47.5 mm2 area located at the centre of the ND–RD cross-</p><p>section using a JEOL JSM-7001F field emission gun – scanning</p><p>electron microscope operating at 15 kV accelerating voltage and</p><p>�5.1 nA probe current. The microscope was fitted with a Nordlys-</p><p>II EBSD detector and an 80 mm2 X-Max EDS detector which</p><p>interface with the Oxford Instruments AZtec software suite. The</p><p>EBSD mapping conditions were optimised beforehand with 43 and</p><p>32 reflectors employed for the bcc and fcc phases, respectively,</p><p>4�4 binning, 3 background frames, a Hough resolution of 60 and</p><p>concurrently indexing individual Kikuchi patterns up to 8 bands</p><p>with an Advanced Fit Index (AFI) value of 3. The raw EBSD map</p><p>returned an overall indexing rate of 87.16% such that most of the</p><p>zero solutions were concentrated at boundary interfaces.</p><p>The employed map step size of 0.095 μm was equivalent to an</p><p>EDS map resolution of �1024�1024 pixels. Other EDS –based</p><p>settings included a 20 keV energy range, auto-selecting the num-</p><p>ber of channels, a process time of 3 and a detector dead time of</p><p>�45–50%. The carbon K output counts over the full ‘TruMap’ area</p><p>without binning returned a single peak Gaussian distribution</p><p>(relative frequency versus cps) with the highest and maximum</p><p>4 The GAM is calculated for contiguous structures bounded by misorientations</p><p>that are lower than the user-defined critical (sub)grain boundary angle.</p><p>5 The GOS is defined as the mean misorientation deviation between the</p><p>average (sub)grain orientation and each pixel within the (sub)grain. It is calculated</p><p>for contiguous structures bounded by misorientations that are lower than the user-</p><p>defined critical (sub)grain boundary angle.</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132 117</p><p>count rates of �1350 cps and �3400 cps, respectively. When these</p><p>values were multiplied with the mean dwell time of 0.074 s per</p><p>pixel, �100 and �250 counts were returned, respectively.</p><p>While segmentation using EBSD-based parameters was under-</p><p>taken within the Oxford Instruments (OI) Channel-5 software</p><p>suite, the EDS data was exported to Gatan DigitalMicrograph for</p><p>thresholding via its associated freeware scripts for X-ray map</p><p>analysis [48] and then imported back into OI Channel-5 for further</p><p>analysis.</p><p>Since this paper focuses on the development of an EBSD-EDS-</p><p>based segmentation procedure, transmission electron microscopy</p><p>(TEM) was employed in a very limited capacity to confirm the site-</p><p>specific segmentation shown in Section 3. In order to provide</p><p>supporting correlative evidence for the variation in the ferrite</p><p>morphologies, a focused ion beam (FIB) section was cut along the</p><p>red line shown in Fig. 1a and Fig. 1a(inset) using an xT Nova NanoLab</p><p>200 Dualbeam located at the University of New South Wales. The FIB</p><p>section was examined on a JEOL ARM-200F (scanning) transmission</p><p>electron microscope operating at 200 kV. The TEM work was</p><p>restricted to bright-field imaging. Other techniques such as X-ray</p><p>microanalysis (EDS) and electron energy loss (EELS) were unsuccess-</p><p>ful in measuring composition differences between the various ferrite</p><p>morphologies; the former due to its low sensitivity and latter due to</p><p>the excessive thickness (i.e. – greater than 2 mean free paths) of the</p><p>FIB section. It is emphasised that no attempt was made to obtain any</p><p>TEM-based statistical information; as the latter is beyond the scope</p><p>of the present study.</p><p>3. EBSD and EDS-based phase and ferrite morphology</p><p>segmentation</p><p>As shown in the secondary electron image, band contrast,</p><p>band slope and phase maps (Fig. 1a–d), the microstructure of</p><p>thermo-mechanically processed TRIP steel consists of bands of</p><p>large ferrite grains interspersed between layers comprising bainite</p><p>and austenite/martensite constituent. The banded appearance is</p><p>the result of the final deformation in the non-recrystallisation</p><p>region (47% finishing reduction at 1123 K) which lead to the</p><p>pancaking of the austenite grains.</p><p>The EBSD map was initially post-processed as per Refs. [49–53]</p><p>by eliminating any potential wild orientation spikes and filling in</p><p>zero solutions via cyclic extrapolation from 8 to 6 neighbours.</p><p>Throughout the text, low-angle boundaries (LAGBs) are defined as</p><p>misorientations between 21rθo151 whereas high-angle bound-</p><p>aries (HAGBs) extend from θZ151. Subgrain/grain reconstruction</p><p>was undertaken using 21 as the minimum misorientation in order</p><p>to fix the angular resolution limit and retain orientation contrast</p><p>information. A minimum spatial resolution of 3 times the nominal</p><p>step size was also maintained constant.</p><p>The first step consisted of separating out the fcc austenite and</p><p>isolating all pixels belonging to bcc ferrite (blue pixels in Fig. 1d).</p><p>Consequently, the following segmentation steps were undertaken</p><p>only on the subset that was originally discriminated as bcc ferrite.</p><p>It should be noted that irrespective of the type of morpholo-</p><p>gical ((sub)grain size or aspect ratio) or internal misorientation</p><p>criteria used in the following paragraphs, the segmentation</p><p>of the bcc ferrite fraction was accomplished by thresholding the</p><p>normalised cumulative distribution of that criterion via an opti-</p><p>mised cut-off value. Unlike other studies that use a fixed cut-off</p><p>value, we applied a constant rule that computed the optimal cut-</p><p>off value in order to guarantee reproducibility across different</p><p>maps. Consequently, our threshold/cut-off value was defined as</p><p>that number at which the slope (m, cf. Figs. 2a and 3a insets) of the</p><p>normalised cumulative distribution with respect to the origin</p><p>tends to 1 ([50] and the references therein). In doing so, the</p><p>variations in the morphological or internal misorientation criteria</p><p>caused by alloy chemistry and/or thermo-mechanical processing</p><p>history can be inherently accounted for when dealing with multi-</p><p>ple samples.</p><p>Fig. 1. (a) Secondary electron image with a 501 tilted inset demarcating the location of the FIB slice and EBSD maps of the (b) band contrast, (c) band slope, and (d) phase</p><p>distribution (red¼ fcc austenite, blue¼bcc ferrite). In (d) white¼LAGBs and black¼HAGBs. (For interpretation of the references to colour in this figure legend, the reader is</p><p>referred to the web version of this article.)</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132118</p><p>As shown in Fig. 2a, the average GOS criterion was applied to</p><p>the bcc ferrite subset. Thresholding at a critical internal misor-</p><p>ientation angle (θC) of 0.931 resulted in the creation of two</p><p>preliminary subsets containing a combination of (i) martensite</p><p>and polygonal ferrite (Fig. 2b), and (ii) ferrite in bainite and</p><p>polygonal ferrite (Fig. 2c). The polygonal ferrite fractions in</p><p>Fig. 2b and c are hereafter referred to as the first and second</p><p>polygonal ferrite fractions, respectively.</p><p>Martensite and the first polygonal ferrite fraction</p><p>were seg-</p><p>mented from their combined subset (Fig. 2b) by thresholding at a</p><p>critical subgrain size of 1.91 mm (Fig. 3a(inset)); resulting in a</p><p>subset comprising a majority of martensite and some polygonal</p><p>ferrite (sub)grains and another subset comprising only the first</p><p>polygonal ferrite fraction. In order to distinctly segment the</p><p>martensite, thresholding was again undertaken on the former</p><p>subset. Consequently, martensite comprises structures between</p><p>0.285 mm (or 3 times the nominal step size) and 0.495 mm (Fig. 3a).</p><p>(Sub)grains between 0.495 and 1.91 mm were re-assigned to the</p><p>first polygonal ferrite fraction (Fig. 3b).</p><p>Fig. 2c depicts the combined subset comprising the ferrite in</p><p>bainite and the second polygonal ferrite fraction. Here we take</p><p>advantage of the fact that compared to the polygonal ferrite</p><p>Fig. 2. EBSD maps of grain orientation spread (GOS) of the (a) bcc ferrite subset, and the two preliminary subsets obtained after thresholding at a critical internal</p><p>misorientation angle ðθC Þ of 0.931 comprising (b) the preliminary martensite and the first polygonal ferrite fraction, and (c) the preliminary ferrite in bainite and the second</p><p>polygonal ferrite fraction. In (a–c) darker shades of grey are relatively scaled to smaller GOS values. Refer to the inset in (a) for the colour code. (For interpretation of the</p><p>references to colour in this figure legend, the reader is referred to the web version of this article.)</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132 119</p><p>fraction, the ferrite in bainite contains a significant proportion of</p><p>very low-angle subgrain boundaries. Consequently, and only for</p><p>the purposes of segmenting the ferrite in bainite subset, the lower-</p><p>bound angular thresholds for the boundary identification and the</p><p>critical subgrain boundary completion angles were reduced from</p><p>21 to 0.51 and from 21 to 0.11, respectively. The net effect of</p><p>reducing the boundary identification and completion angles to the</p><p>“noise” level of the EBSD map's resolution was that the individual</p><p>subgrains of the ferrite in bainite were uniquely delineated.</p><p>Thereafter, when a subgrain size threshold of 1.05 mm was applied</p><p>to Fig. 2c, it resulted in two subsets comprising most of the ferrite</p><p>in bainite (Fig. 4a) and the second polygonal ferrite fraction</p><p>(Fig. 4b). It should be noted that Fig. 4a still contained pixelations</p><p>that were nominally part of the second polygonal ferrite fraction.</p><p>In order to remove these pixelations, the second polygonal ferrite</p><p>fraction shown in Fig. 4b was dilated by one pixel. These dilated</p><p>pixels were then subtracted from Fig. 4a; resulting in a final</p><p>combined ferrite in bainite subset (Fig. 4c).</p><p>Following this, the boundary identification and the critical</p><p>subgrain boundary completion angles were restored to 21. In order</p><p>to obtain the second polygonal ferrite fraction (Fig. 5a), the final</p><p>ferrite in bainite subset (Fig. 4c) was subtracted from the subset</p><p>shown in Fig. 2c. The first (Fig. 3b) and second (Fig. 5a) polygonal</p><p>ferrite fractions were then combined together to return the final</p><p>polygonal ferrite subset (Fig. 5b).</p><p>It should be noted that the final combined ferrite in bainite</p><p>subset contains two distinct morphologies that can be sub-divided</p><p>into ferrite in granular bainite and bainitic ferrite laths (Fig. 1a).</p><p>However, the application of the conventional metrics of band</p><p>contrast, band slope, subgrain internal misorientation and mor-</p><p>phology (size, aspect ratio, slope etc.) were unable to satisfactorily</p><p>segment them.6 On the other hand, the EDS data of the relative</p><p>carbon counts of the ferrite in bainite subset (Fig. 6a) showed</p><p>that relatively higher carbon counts were returned in areas</p><p>comprising ferrite in granular bainite whereas fewer counts were</p><p>seen in areas corresponding to the bainitic ferrite laths. As a result,</p><p>the following steps segmented the ferrite in granular bainite and</p><p>the bainitic ferrite laths from their combined subset via Digital-</p><p>Micrograph7 [48].</p><p>Grayscaling the image showing the variation in the relative</p><p>carbon counts of the final combined ferrite in bainite subset was</p><p>undertaken by importing a TIFF image into DigitalMicrograph,</p><p>changing the data type by converting the RGB image into a real</p><p>image using the intensity of the colour channel that corresponded</p><p>to the relative carbon counts in the map (Fig. 6a). The “Interactive</p><p>Thresholding” script was then used and an appropriate dynamic</p><p>Fig. 3. EBSD maps of subgrain size of the preliminary martensite and the first polygonal ferrite fraction from Fig. 2b, and the two subsets obtained after (a) double</p><p>thresholding martensite between 0.285 mm and 0.495 mm and (b) the first polygonal ferrite fraction. In (a, b) subgrain size is relatively scaled using the rainbow colour</p><p>scheme from small (blue) to large (red). Refer to the inset in (a) for the colour code. (For interpretation of the references to colour in this figure legend, the reader is referred</p><p>to the web version of this article.)</p><p>6 Also refer to Section 5.2 and Fig. 10 for further details.</p><p>7 Note here that the carbon counts from the austenite, martensite and</p><p>polygonal ferrite fractions were not used for segmentation. Moreover, the absolute</p><p>values of the carbon counts recorded by the ferrite morphologies of granular</p><p>bainite and bainitic ferrite are of relatively minor importance compared to the fact</p><p>a relative variation in carbon counts exists between them. As shown in Fig. 6, it was</p><p>this small variation in carbon counts that was exploited for the sole purpose of</p><p>accurately detecting and demarcating the bounds at which the variation in carbon</p><p>occurred in the ferrite morphologies of the two bainities.</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132120</p><p>threshold cut-off value (in this case 205) selected to binarise the</p><p>image and enhance the contrast by assigning a value of 0 and 1 to</p><p>all the black and white regions, respectively. Following this, the</p><p>“Invert Image Contrast” script was called to invert the above</p><p>binary image (Fig. 6b). Thereafter, the “Local Neighbourhood</p><p>Density Threshold and Dilation” script was used. Here a neigh-</p><p>bourhood around a central pixel was defined as either 8 pixels</p><p>arranged in a 3�3 matrix of first-nearest neighbours or 8þ16¼24</p><p>pixels arranged in a 5�5 matrix of first and second-nearest</p><p>neighbours or 8þ16þ24¼48 pixels arranged in a 7�7 matrix</p><p>of first, second and third-nearest neighbours and so on. In our</p><p>case, the thresholding of the inverted binary image was based on a</p><p>9�9 matrix and involved a neighbourhood “density” search for at</p><p>least 4 white pixels within that matrix.8 If found, the central pixel</p><p>being evaluated was assigned a value of 1 and appeared white. The</p><p>Fig. 4. EBSD maps of subgrain size after reducing the boundary identification and critical subgrain boundary completion angles to 0.51 and 0.11, respectively, of the</p><p>preliminary ferrite in bainite and the second polygonal ferrite fraction from Fig. 2(c), and the subsets obtained after thresholding at a critical subgrain size of 1.05 mm</p><p>comprising (a) ferrite in bainite with some pixels belonging to the second polygonal ferrite fraction and (b) the second polygonal ferrite fraction. In (c) the final combined</p><p>ferrite in bainite subset is obtained after dilating the second polygonal ferrite fraction in (b) by 1 pixel and subtracting the dilated pixels from (a). In (a–c) subgrain size is</p><p>relatively scaled using the rainbow colour scheme from small (blue) to large (red). (For interpretation of the references to colour in this figure legend, the reader is referred to</p><p>the web version of this article.)</p><p>8 The neighbourhood density search involved the interrogation of all 8 first</p><p>neighbour pixels and then only the pixels along the vertical, horizontal and</p><p>diagonals of the 9�9 matrix for the second, third and fourth neighbours.</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132 121</p><p>effect of this approach involving local neighbourhood density</p><p>thresholding was that pixels at the interface between areas of</p><p>high and low carbon counts</p><p>(Fig. 6b) were far less likely to appear</p><p>white. It should be noted that Fig. 6b still comprised a large</p><p>number of black pixels distributed within white areas that</p><p>demarcate high carbon counts. Consequently, each white pixel</p><p>was then dilated into a 3�3 matrix of white pixels (Fig. 6c). This</p><p>had the net effect of filling in the black pixels and retaining the</p><p>interface between areas of high and low carbon counts.</p><p>Following this, a “Median filter” script was called and a filter</p><p>size of 5 was applied to smooth out the edges and fill in any</p><p>remaining black pixels within white areas (Fig. 6d). The image was</p><p>then saved as a text file using the “Export Image as Tabbed Text”</p><p>script. The text file consisted of three data columns comprising</p><p>X and Y coordinates (in increments of 0.095 mm (or the step size))</p><p>and a value of 0 or 1 correlated to each position. The third data</p><p>column was merged with a text file of the original EBSD map and</p><p>imported back into the Channel-5 software. Consequently, the</p><p>pixels with a value of 1 form the preliminary ferrite in granular</p><p>bainite subset. The boundary identification and the critical sub-</p><p>grain boundary completion angles were again set to 21 following</p><p>which only those subgrains that had 50% of their pixels belonging</p><p>to the preliminary ferrite in granular bainite subset were said to be</p><p>members of the final ferrite in granular bainite subset (Fig. 6e). The</p><p>final bainitic ferrite lath subset (Fig. 6f) was obtained after</p><p>subtracting the ferrite in granular bainite subset (Fig. 6e) from</p><p>the final combined ferrite in bainite subset (Fig. 4c). Fig. 6g shows</p><p>the distributions of the relative carbon K counts for ferrite in</p><p>granular bainite and the bainitic ferrite laths obtained after</p><p>segmentation. Here two aspects are emphasised. Firstly, if the</p><p>lateral spread of the interaction volume due the high sample tilt</p><p>and plural scattering events were of any consequence, the relative</p><p>carbon counts from the bainitic ferrite laths would be nominally</p><p>higher than the ferrite in granular bainite as a result of the large</p><p>areas of interleaved austenite in the former ferrite morphology.</p><p>However, the opposite is true. Secondly, when compared with</p><p>polygonal ferrite, it is clear that no enhanced X-ray emissions are</p><p>observed in either of the ferrite morphologies of the two bainites</p><p>(Fig. 6g).</p><p>In Fig. 7a, the final subsets comprising the austenite (7.9%),</p><p>martensite (1.3%), polygonal ferrite (68.9%), ferrite in granular</p><p>bainite (7.9%) and bainitic ferrite lath (14.0%) fractions are shown</p><p>in red, yellow, blue, green and white, respectively. Histograms of</p><p>the band contrast, band slope and kernel average misorientation</p><p>for the various phases and ferrite morphologies are given in</p><p>Fig. 7b–f.</p><p>4. Transmission electron microscopy</p><p>In order to ensure that the segmentation procedure described</p><p>in Section 3 was producing valid data, correlations were sought via</p><p>TEM on a FIB section from a particular region of interest (Fig. 1a</p><p>and Fig. 1a(inset)). Fig. 8 shows a montage TEM bright field image</p><p>along the length of the FIB section. The orientation of the FIB</p><p>section was the same in both images and runs from left to right.</p><p>Fig. 5. EBSD maps of subgrain size after restoring the boundary identification and critical subgrain boundary completion angles to 21. In (a) the second polygonal ferrite</p><p>fraction is obtained after subtracting the final combined ferrite in bainite subset (Fig. 4c) from the subset shown in Fig. 2c. In (b) the final polygonal ferrite subset is obtained</p><p>by combining the first (Fig. 3b) and the second (a) polygonal ferrite fractions together. In (a, b) subgrain size is relatively scaled using the rainbow colour scheme from small</p><p>(blue) to large (red). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132122</p><p>The dark band along the upper surface of the section is the Pt</p><p>capping layer deposited during FIB sectioning to protect the</p><p>surface from ion beam erosion. The colour inset in Fig. 8 shows</p><p>the identification of the microstructural constituents derived from</p><p>Fig. 7a. The section intersects four different regions at the surface</p><p>with austenite, polygonal ferrite, ferrite in granular bainite and</p><p>Fig. 6. Sequential DigitalMicrograph-based imaging processing of the relative distrubtion of carbon counts after (a) grayscaling, (b) dynamically threholding at 205,</p><p>binarising to increase contrast and inverting the binary data, (c) applying a 9�9 matrix for a local neighborhood density threshold of 4 white pixels and then dilating each</p><p>white pixel into a 3�3 matrix, and (d) applying a median filter of size 5. EBSD maps of subgrain size of the final (e) ferrite in granular bainite and (f) bainitic ferrite lath</p><p>subsets after importing back to OI Channel-5. (f) The distributions of the relative carbon K counts for the ferrite in granular bainite and bainitic ferrite laths obtained after</p><p>segmentation. The carbon counts of polygonal ferrite are shown for comparison.</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132 123</p><p>Fig. 7. (a) EBSD map of the final austenite (7.9%), martensite (1.3%), polygonal ferrite (68.9%), ferrite in granular bainite (7.9%) and bainitic ferrite lath (14.0%) fractions shown</p><p>in red, yellow, blue, green and white, respectively. The fractions have been rounded-off to one decimal place. In (a), grey¼LAGBs and black¼HAGBs. Histograms of the (b, c)</p><p>band contrast, (d, e) band slope and (f) kernel average misorientation versus (b, d, f) relative frequency and (c, e) area normalised relative frequency of the various phases and</p><p>ferrite morphologies. In (f) the kernel average misorientation was calculated using a 3�3 matrix of first-nearest neighbours for a square grid. (For interpretation of the</p><p>references to colour in this figure legend, the reader is referred to the web version of this article.)</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132124</p><p>bainitic ferrite laths shown in red, blue, green and white, respec-</p><p>tively. It is also interesting to note the effect of etching on the</p><p>polished surface (or TD-view) when observed in cross-section (or</p><p>ND-view). Although microstructure response to etching depends</p><p>on composition and crystallographic orientation, in this particular</p><p>case the polygonal ferrite and bainitic ferrite laths are slightly</p><p>sunken whereas the austenite and the ferrite in granular bainite</p><p>stand proud of the surface.</p><p>In this type of correlative work, there are several potential</p><p>sources of mismatch between the SEM-derived EBSD data and that</p><p>from TEM. The largest uncertainties include the following: (i) the</p><p>measurement error in determining intersects from small features</p><p>with limited pixel resolution within the microstructural constitu-</p><p>ents maps, (ii) the error with which EBSD positions the edges of</p><p>the phase/constituent fields; here small discrete colonies often</p><p>appear larger in metallographic (or TEM) sections than in EBSD</p><p>maps, and (iii) the actual location of the cut FIB section. Since the</p><p>FIB technique destroys the surrounding microstructure, the exact</p><p>location in terms of start, end and lateral position is uncertain by</p><p>perhaps a micron or so. Given these uncertainties, it can be seen</p><p>that the agreement between the phase/constituent positions from</p><p>EBSD (Figs. 1a and 7a) and the morphologically distinct structures</p><p>in TEM (Fig. 8) is quite good.</p><p>Detailed examination of the various regions (grains A–D)</p><p>identified in Fig. 8 was carried out. Fig. 9a shows the polygonal</p><p>ferrite grain (grain A in Fig. 8) after tilting to a two beam (200)</p><p>condition close to an ½011� zone axis. The diffraction pattern</p><p>(Fig. 9a(inset)) was consistent with bcc ferrite. Compared to the</p><p>other ferrite morphologies, grain A had the lowest dislocation</p><p>density. The adjoining structure located at the top of Fig. 9a was</p><p>identified as ferrite in granular bainite.</p><p>In Fig. 9b, the ferrite in granular bainite (grain B in Fig. 8) was</p><p>found to be very different to</p><p>polygonal ferrite (Fig. 9a) such that an</p><p>�0.2 mm wide and several microns long plate-like structure was</p><p>present. The plates contained the highest density of dislocations</p><p>(after disregarding the sub-surface black regions in Fig. 9b–d</p><p>containing an extremely high dislocation density and deemed to</p><p>be martensite) compared to polygonal ferrite. The inset diffraction</p><p>pattern was consistent with a ½111� oriented bcc phase. Also</p><p>present were some off axis intense reflections which arose from</p><p>adjoining regions due to the use of the smallest selected area</p><p>aperture and the inherent uncertainty in localising the diffraction</p><p>information. Most were randomly oriented but some were aligned</p><p>with the matrix reflections. These may be due to the existence of</p><p>cementite or may be a sputtering artifact and were not investi-</p><p>gated further.</p><p>In Fig. 9c, bainitic ferrite laths (grain C in Fig. 8) returned</p><p>dislocation densities that were relatively higher than polygonal</p><p>ferrite (Fig. 9a) but lower than the ferrite in granular bainite</p><p>(Fig. 9b). Dislocation clustering into dislocation cells was also</p><p>evident within this grain.</p><p>Fig. 9d is representative of an austenite grain (grain D in Fig. 8)</p><p>that was elongated parallel to the FIB section surface. It is clear</p><p>that the dislocation density was much lower than that of the</p><p>ferrite in granular bainite and that no other plate/lath structure or</p><p>secondary phase was present. Another example of austenite</p><p>adjacent to the ferrite in granular bainite is shown in Fig. 9e.</p><p>On the basis of the TEM examination, the microstructure can be</p><p>summarised as follows: (i) polygonal ferrite was relatively</p><p>dislocation-free and contained no subgrain structure, (ii) the</p><p>ferrite in granular bainite had an �0.2 mm wide and several</p><p>microns long plate-like structure and was the most defective</p><p>compared to polygonal ferrite, (iii) bainitic ferrite laths presented</p><p>with dislocation cells and was relatively more defective than</p><p>polygonal ferrite, (iv) in comparison to the ferrite morphologies</p><p>in bainite, austenite was relatively defect-free. It should be noted</p><p>that due to the uniform thickness of the FIB slice, the qualitative</p><p>comparison of dislocation densities is acceptable; whereas the</p><p>quantitative estimation could be misleading due to the changes in</p><p>defect structure during FIB milling. The above TEM results are</p><p>similar to the optical, scanning and transmission microscopy</p><p>analyses of the same steel undertaken in Ref. [47] and the</p><p>characterisation of the same phases/constituents in other TRIP</p><p>steels [5,14,16].</p><p>5. Discussion</p><p>5.1. The use of internal misorientation criteria</p><p>The internal misorientation and IQ/PQ/BC and BS values are</p><p>extremely sensitive to the surface preparation methodology, the</p><p>EBSD map acquisition settings, individual orientations within</p><p>distinct phases, grain boundaries and surface topology. Conse-</p><p>quently, the sample used in this study was prepared by electro-</p><p>polishing in order to obtain the highest quality EBSD map.</p><p>Thereafter, a reproducible phase/ferrite morphology segmentation</p><p>method that inherently accounts for variations caused by proces-</p><p>sing history needs to employ internal misorientation criteria. Here</p><p>it should be noted that short range orientation gradient para-</p><p>meters like the KAM provide orientation variations at a pixel-level</p><p>and not at a substructure level. Consequently, short range</p><p>Fig. 8. TEM montage of the FIB section cut from the region of interest shown in Fig. 1a and Fig. 1a(inset). The colour inset shows the phase identification transposed from the</p><p>EBSD phase map shown in Fig. 7c. The austenite, polygonal ferrite, ferrite in granular bainite and bainitic ferrite lath are shown in red, blue, green and white, respectively.</p><p>Grains A to D are detailed in Fig. 9. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132 125</p><p>Fig. 9. Bright field images of surface (a) polygonal ferrite, (b) ferrite in granular bainite, (c) bainitic ferrite laths and (d) austenite denoted as grains A to D in Fig. 8,</p><p>respectively. (a) Polygonal ferrite has a relatively low dislocation density. The diffraction pattern (inset) was obtained after tilting to the bcc [011] zone axis. (b) Ferrite in</p><p>granular bainite has a plate-like substructure. The inset diffraction pattern in (b) was obtained from the circular region after tilting to the bcc [111] zone axis. (c) Bainitic</p><p>ferrite laths with dislocation cell structure. This region is bounded by austenite grains located at the bottom left and top right. In (d) an elongated surface grain of austenite</p><p>with ferrite in granular bainite located at the top right. (e) A two beam image showing the interface between austenite (left) and bainitic ferrite laths (right). The inset</p><p>diffraction pattern in (e) was obtained after tilting to the bcc [011] zone axis. Rotate the micrographs in Fig. 9 clockwise to correlate with the montage in Fig. 8.</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132126</p><p>orientation gradient parameters are only effective when they are</p><p>used in conjunction with an additional conditionality. For example</p><p>– in Ref. [41] ferrite in bainite was distinguished from polygonal</p><p>ferrite in an intercritically annealed TRIP steel when the output</p><p>from the third nearest neighbour KAM calculation was combined</p><p>with a conditionality criterion that interrogated every (sub)grain</p><p>boundary pixel for its immediate neighbourhood.</p><p>Based on the above results, we can conclude that preliminary</p><p>discrimination between martensite, the ferrite in bainite and</p><p>polygonal ferrite is possible from EBSD-only data when the grain</p><p>orientation spread criterion is applied. As seen in Fig. 2b and c, the</p><p>average grain orientation spread criterion provides for an easy and</p><p>more effective segmentation between the preliminary martensite</p><p>and the first polygonal ferrite fraction (Fig. 2b) and the preliminary</p><p>ferrite in bainite and the second polygonal ferrite fraction (Fig. 2c).</p><p>This is accomplished by assigning a single value to all pixels within</p><p>a particular (sub)grain. It is emphasised here that the GOS</p><p>parameter is a measure of orientation spread and is a function of</p><p>the relative change in orientation with spatial variation. In turn,</p><p>the latter is a function of the substructure size and the employed</p><p>step size. It follows that since the GOS parameter is based on a</p><p>user-defined value, the criterion is better suited to discriminate</p><p>between crystallographically similar but morphologically distinct</p><p>structures over large length scales.</p><p>5.2. The drawbacks and disadvantages of using EBSP quality metrics</p><p>for segmentation</p><p>When analysing a series of EBSD maps, the use of the IQ/PQ/BC</p><p>and BS thresholding schemes presupposes that the microscope</p><p>focusing conditions and the EBSD hardware settings were consis-</p><p>tent and uniform across all samples and that the software correctly</p><p>indexed the EBSP during its acquisition (i.e. – with a high CI free</p><p>from pseudo-symmetric orientation errors and a low PM). Given</p><p>real-world issues such as microscope environment and hardware</p><p>limitations, beam-time restrictions and multi-user sites, this may</p><p>not always be the case. Furthermore, each EBSD map is in some</p><p>ways unique as the calibration and optimisation of the EBSD</p><p>camera hardware is undertaken at the start of every map. During</p><p>mapping, beam instabilities can result in smaller IQ/PQ/BC and BS</p><p>values whereas dynamic gain control and background collection</p><p>algorithms compensate for any losses in IQ/PQ/BC and BS values by</p><p>automatically changing the video conditions [39]. Any combina-</p><p>tion of the above factors can therefore lead to issues in delineating</p><p>unique greyscale contrast ranges for the various phases/ferrite</p><p>morphologies in multiple samples.</p><p>Over the years, the mathematical Multi-peak model has proven</p><p>very useful in providing area fraction estimates by deconvoluting</p><p>the IQ/PQ/BC distribution into multiple, symmetric Gaussian sub-</p><p>distributions. However, the main</p><p>disadvantages with this metho-</p><p>dology are as follows: (i) the model assumes that each ferrite</p><p>morphology will return a symmetric IQ/PQ/BC sub-distribution,</p><p>and (ii) the errors increase with the number of ferrite morphol-</p><p>ogies present such that significant overlaps in greyscale contrast</p><p>occur between sub-distributions. As seen in Fig. 10a–d, austenite,</p><p>polygonal ferrite, martensite, the ferrite in granular bainite and</p><p>bainitic ferrite lath fractions return distinct but overlapping,</p><p>asymmetric and in some cases, slightly bimodal BC and BS</p><p>distributions when they are segmented using the procedure out-</p><p>lined in Section 3. It follows that if the multi-peak model was</p><p>applied to the present TRIP steel (whose substructures contain</p><p>local variations and/or overlapping contrast), the segmentation</p><p>procedure would then largely depend on user judgment and</p><p>experience.</p><p>As suggested in Refs. [34,40,54], the averaging of the IQ/PQ/BC</p><p>and BS values of the (sub)grains offers a practical solution to mitigate</p><p>(sub)grain boundary effects and enable comparisons between (sub)</p><p>grains belonging to different phases/ferrite morphologies. This</p><p>enabled Kang et al. [40] to generate a bimodal IQ/PQ/BC distribution</p><p>with minimal overlap and segment martensite from ferrite in a dual</p><p>phase steel. However, it is pointed out that the as-mapped IQ/PQ/BC</p><p>distribution of that steel already contained a bimodal distribution</p><p>(i.e. – before any (sub)grain pattern quality averaging was under-</p><p>taken). The (sub)grain pattern averaging procedure merely served to</p><p>make the difference between the two sub-distributions even more</p><p>apparent. An example of where such a procedure can be undertaken</p><p>is seen in the BS histogram of the full map in Fig. 7e. But it should be</p><p>noted that this strategy would not be very useful in cases when a</p><p>single, asymmetric IQ/PQ/BC or BS distribution is returned in the as-</p><p>mapped condition. An example of the latter situation is seen in the</p><p>BC histogram of the full map in Fig. 7c.</p><p>Another alternative that mitigates (sub)grain boundary effects</p><p>excludes pixels on either side of the boundary interface after</p><p>imposing a critical grain boundary misorientation angle [38].</p><p>However, in doing so, the number of pixels defining the average</p><p>greyscale contrast within the interior of the smallest contiguous</p><p>structure (i.e. – the grain or kernel) must remain statistically</p><p>significant. While this can be achieved through smaller map step</p><p>sizes, the disadvantage of this approach includes significantly</p><p>increased mapping times and its attendant environment/micro-</p><p>scope stability requirements. Here it should be noted that reduc-</p><p>tions in map step size are also physically limited by the interaction</p><p>volume; which in turn can degrade the spatial and angular</p><p>resolution of the map and/or lead to errors in microstructure</p><p>discrimination. Another concern with this strategy is that the</p><p>removal of pixels could lead to misinterpretations; especially</p><p>when attempting to discern ferrite morphologies at/near the grain</p><p>boundaries. An example of this is Ref. [55] where ferrite in bainite</p><p>was deemed to be identified in highly localised areas at the</p><p>grain boundary interface between ferrite and austenite phases in</p><p>Fe–1.4Mn–0.7Al–0.5Si–0.2C TRIP steel.</p><p>Finally, IQ/PQ/BC and BS values can also vary within individual</p><p>phases/ferrite morphologies due to the preferential orientation</p><p>effect. Some crystallographic planes result in Kikuchi bands of</p><p>higher intensity such that the EBSPs containing these reflections</p><p>return large IQ/PQ/BC and BS values [19,39]. Alternatively, if the</p><p>crystallographic planes are oriented such that their EBSPs contain</p><p>low intensity reflections, smaller IQ/PQ/BC and BS values are</p><p>returned.</p><p>The overall inadequacy of the IQ/PQ/BC and BS values for ferrite</p><p>morphology segmentation is further illustrated in Fig. 10. Using the</p><p>DigitalMicrograph-based procedure outlined in the latter half of</p><p>Section 3 and Fig. 6, an attempt was made to segment the ferrites</p><p>in the two bainites via the inverted BS parameter (Fig. 10a). The</p><p>result of this segmentation procedure is shown in Fig. 10b. Clear</p><p>areas of mismatch between Figs. 7a and 10b are marked in the latter</p><p>figure with the aid of black circles. Thus, even when a relatively</p><p>sophisticated image processing procedure such as that described in</p><p>Section 3 and Fig. 6 is applied, the greyscale values of the inverted BS</p><p>parameter lack the sensitivity to accurately discriminate the ferrites</p><p>in the two bainites. Consequently, the use of the relative variation in</p><p>carbon counts from EDS mapping is indispensable to accurately</p><p>segment the ferrite in granular bainite and the bainitic ferrite laths</p><p>(Fig. 6).</p><p>5.3. The use of elemental carbon in segmenting the ferrite</p><p>morphologies in the two bainites</p><p>It is re-emphasised that conventional EDS mapping involves</p><p>rastering over an area of interest multiple times for a fixed time</p><p>period or a user-defined number of counts in order to obtain</p><p>statistically relevant information on elemental distribution. In the</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132 127</p><p>present case, combined EBSD-EDS mapping involves a single raster</p><p>over an area of interest for a given mean dwell time per pixel.</p><p>Given the conditionalities imposed by the high sample tilt, the</p><p>associated interaction volume effects and the rather limited EDS-</p><p>data acquisition conditions, the absolute elemental counts</p><p>recorded by a particular ferrite morphology could be prone to</p><p>error and are therefore of lesser importance compared to the</p><p>relative variation in elemental counts between the ferrite morphol-</p><p>ogies of the two bainites. Thus, while the detailed analysis of the</p><p>absolute elemental counts per morphology is beyond the scope of</p><p>the present study, the relative variation in the elemental counts</p><p>between the ferrite morphologies of the two bainites has been</p><p>exploited to segment the broad regions occupied by the ferrite in</p><p>granular bainite and the bainitic ferrite laths [48].</p><p>Generally speaking, it may be true that X-rays are generated at</p><p>sample depths greater than those for backscattered electrons. In</p><p>the case of carbon in iron, the carbon X-rays can be conventionally</p><p>generated up to �1 mm below the surface. However, the very low</p><p>energy of the carbon X-rays also results in extremely strong</p><p>absorption of all but the very near-surface X-rays. To understand</p><p>the sample depth from which the back-scattered (or EBSP) and</p><p>elemental carbon information arise, we undertook Monte Carlo</p><p>modelling (Casino v2.48 [56–59]) after imposing the same sample</p><p>composition, sample tilt and beam energy conditions (Fig. 11a).</p><p>The simulations show that 95% of the backscattering and carbon</p><p>X-ray emission signals emanate from similar depths at �173 nm</p><p>and �152 nm, respectively (Fig. 11b). In fact, the carbon X-ray</p><p>emission declines very rapidly with sample depth such that �60%</p><p>of the total emission comes from the top �41 nm of the sample</p><p>while for the backscattered signal the equivalent depth is �45 nm.</p><p>On this basis, the carbon X-ray map is more likely to be derived</p><p>from shallower depths than the EBSP data. Moreover, if the TEM</p><p>montage of the FIB section (Fig. 8) is taken as somewhat repre-</p><p>sentative of the various ferrite morphologies, the substructures are</p><p>found to vary from �170 to 600 nm in depth. Thus, on average it is</p><p>more plausible to expect that the carbon X-ray emission data is</p><p>from localised volumes that are contained within the smallest</p><p>substructures defined by the resolution of the EBSD map.</p><p>Considering the high sample tilt and the resultant lateral</p><p>spread of the interaction volume, it could also be possible that</p><p>the relative carbon counts recorded by either of the ferrite</p><p>morphologies in the two bainites contain elemental data from</p><p>plural scattering events arising from neighbouring austenite/mar-</p><p>tensite constituents. As shown in Fig. 11c, the simulations suggest</p><p>that relative to the beam position, 95% of the backscattering and</p><p>carbon X-ray emission signals emanate from effective radii of</p><p>�230 nm and �188 nm, respectively.</p><p>However, both emission</p><p>profiles are heavily weighted towards much smaller radii such that</p><p>60% of the backscattering and carbon X-ray emission signals</p><p>emanate from effective radii of �27 nm and �14 nm, respectively.</p><p>Since the carbon map shown in Fig. 6a is used to differentiate</p><p>micron-scale regions of ferrite in granular bainite and bainitic</p><p>ferrite laths, the segmentation is unlikely to be affected by the</p><p>spatial resolution of the EDS data. Secondly, readers are pointed to</p><p>Fig. 10. EBSD maps of the (a) inverted band slope of the final combined ferrite in bainite subset shown in Fig. 4c. In (b) the phase and ferrite morphology distribution</p><p>obtained after thresholding the inverted band slope via DigitalMicrograph. The austenite (7.9%), martensite (1.3%), polygonal ferrite (68.9%), ferrite in granular bainite (10.1%)</p><p>and bainitic ferrite lath (11.9%) fractions are in red, yellow, blue, green and white, respectively. The fractions have been rounded-off to one decimal place. In (b), grey¼LAGBs</p><p>and black¼HAGBs. Areas of mismatch with Fig. 7a are shown with the aid of black circles. (For interpretation of the references to colour in this figure legend, the reader is</p><p>referred to the web version of this article.)</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132128</p><p>Fig. 11. (a) Monte Carlo simulation of the electron beam interaction with the sample volume. The normalised intensity and cumulative back-scattering electron (BSE) and carbon K</p><p>signals from (b) the sample depth and (c) the effective radius. (d) IPF map showing the spread of orientations along the ND in the final combined ferrite in bainite subset (cf. Fig. 6e and</p><p>f for reference). In (b, c) the red and green colours denote 95% and 60% of the signal, respectively. The inset in (c) is the normalised intensity (scaled by �10�4) versus the effective</p><p>radius of the BSE signal. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)</p><p>A.A. Gazder et al. / Ultramicroscopy 147 (2014) 114–132 129</p><p>the fact that the austenite/martensite constituent between the</p><p>ferrite in granular bainite are fewer and much smaller than the</p><p>areas and widths occupied by the austenite interleaved between</p><p>bainitic ferrite laths (Fig. 7a). Thus, if the lateral spread of the</p><p>interaction volume and plural scattering events were of any</p><p>consequence, the carbon counts returned by the bainitic ferrite</p><p>laths would be nominally higher than those returned by the ferrite</p><p>in granular bainite. In fact, the opposite is true such that the</p><p>carbon counts of the ferrite in granular bainite were higher than</p><p>those of the bainitic ferrite laths (Fig. 6g).</p><p>Lastly, it could also be possible that a channelling artefact explains</p><p>the relative variation in the elemental counts between the ferrite</p><p>morphologies of the two bainites. Electron channelling (resulting in</p><p>enhanced high X-ray emission) may be observed when there is a</p><p>strong preferred orientation such that all the ferrite grains in bainite</p><p>are close to low index zone axes under the imposed imaging</p><p>conditions. To this end, Fig. 11d is an IPF map showing the spread</p><p>of orientations along the ND in the final combined ferrite in bainite</p><p>subset (cf. Figs. 6e, f and 7a for reference). It is clearly seen that:</p><p>(i) there is a large orientation spread within the ferrite in granular</p><p>bainite and the bainitic ferrite laths when each subset is analysed in</p><p>isolation, and (ii) both ferrite morphologies in the two bainites</p><p>comprise local regions of approximately similar orientations which</p><p>differ significantly in their relative carbon counts. Thus, since</p><p>anomalously high X-ray emissions and strong preferred orientations</p><p>are absent in both the ferrite morphologies of the two bainites</p><p>(Fig. 6g, compare with polygonal ferrite and Fig. 11d), it is highly</p><p>unlikely that a channelling artefact is contained in their carbon X-ray</p><p>emission data.</p><p>5.4. On the formation of bainite</p><p>With respect to the bainite transformation mechanism, it is</p><p>commonly accepted that carbon partitions from supersaturated</p><p>ferrite into the residual austenite during the growth stage [60].</p><p>However, even at the conclusion of formation, the carbon con-</p><p>centration of ferrite in bainite exceeds its para-equilibrium con-</p><p>centration. The first explanation of this supersaturation was linked</p><p>to carbon trapping in high dislocation density substructures</p><p>formed in the ferrite in bainite as a result of the shear mechanism</p><p>[61]. However, recent atom probe studies of volumes of ferrite in</p><p>bainite free of any visible carbon segregation have confirmed the</p><p>supersaturation of carbon in the matrix [62–64]. This was</p><p>explained by Bhadeshia [65] in terms of the tetragonality of the</p><p>lattice during the early stages of transformation; resulting in a</p><p>higher solubility of carbon compared to the bcc lattice.</p><p>In agreement with the above, our EDS map data shows that the</p><p>carbon content in the ferrites of both bainitic morphologies is</p><p>relatively higher than in polygonal ferrite (Fig. 6g). With respect to</p><p>the equilibrium phase diagram, the lower carbon content in</p><p>polygonal ferrite can be linked to its lower solubility in the bcc</p><p>lattice. Furthermore, and more importantly, we have clearly shown</p><p>for the first time that the differences in the carbon content</p><p>between ferrite in granular bainite and bainitic ferrite laths in</p><p>the EDS map can be effectively utilised to segment the two ferrite</p><p>morphologies (Fig. 6a–f).</p><p>In our TRIP steel, the formation of granular bainite is assumed</p><p>to take place upon accelerated cooling at 20 K s�1 from 953 K to</p><p>the isothermal holding temperature of 743 K; from which point</p><p>onwards the formation of bainitic ferrite occurs. On undercooling</p><p>from 953 K, nucleation events take place continuously. This results</p><p>in a microstructure of relatively fine ferrite plates or irregular-</p><p>shaped ferrite within granular bainite; which in a majority of</p><p>instances are in contact with each other. Peculiar to regions of</p><p>granular bainite, extremely small area fractions of the martensite/</p><p>retained austenite constituent tend to be located more often at</p><p>granular bainite/polygonal ferrite interface boundaries and very</p><p>rarely between individual ferrite plates (Fig. 7a). Thus, our two-</p><p>dimensional EBSD mapping infers that since the distance for</p><p>carbon from the ferrite in granular bainite to reach the islands of</p><p>retained austenite is rather large, the 10.5 s timeframe during</p><p>undercooling from 953 K to 743 K may not be long enough to</p><p>allow this diffusion to proceed to completion. In addition, our</p><p>calculations show that during the 1200 s holding at 743 K, carbon</p><p>atoms could diffuse by no more than 0.5 mm. As seen in Fig. 7a, in</p><p>order to reach the closest retained austenite, the carbon in the</p><p>ferrite in granular bainite needs to diffuse distances Z1 mm. As a</p><p>consequence, a significant amount of carbon tends to remain</p><p>trapped within the individual ferrite grains in granular bainite</p><p>such that the latter regions present with the highest relative</p><p>carbon counts (Fig. 6a and g).</p><p>On the other hand, the bainitic ferrite laths are separated by</p><p>relatively thick (�100–300 nm) layers of retained austenite. The</p><p>retained austenite layers serve as sinks for carbon such that it is</p><p>easier for carbon to be partitioned into the retained austenite; thus</p><p>leaving the bainitic ferrite laths with a relatively lower carbon</p><p>content (Fig. 6a). Such partitioning of carbon was previously</p><p>observed in bainite using field emission–electron probe micro-</p><p>analysis [66] and atom probe tomography [15,67,68].</p><p>For the purposes of this discussion, if we assume that the</p><p>arrangement of ferrite plates in granular bainite and the location</p><p>of the islands of martensite/retained austenite constituent is the</p><p>same in all three spatial dimensions, we can make use of the TEM</p><p>evidence to further ascertain the differences in carbon partitioning</p><p>behaviour. Since the diffusivity of carbon between 953 K and 743 K</p><p>is high, carbon atoms will also tend to be attracted to other</p><p>microstructural sinks. In this</p>