Substructure Development and Damage Initiation in a Carbide-Free Bainitic Steel upon Tensile Test

: Carbide-free bainitic (CFB) steels belong to the family of advanced high strength steels (AHSS) that are struggling to become part of the third-generation steels to be marketed for the automotive industry. The combined e ﬀ ects of the bainitic matrix and the retained austenite confers a signiﬁcant strength with a remarkable ductility to these steels. However, CFB steels usually show much more complex microstructures that also contain MA (Martensite–Austenite) phase and auto-tempered martensite (ATM). These phases may compromise the ductility of CFB steels. The present work analyzes the substructure evolution during tensile tests in the necking zone, and deepens into the void and crack formation mechanisms and their relationship with the local microstructure. The combination of FEG-SEM imaging, EBSD, and X-ray di ﬀ raction has been necessary to characterize the substructure development and damage initiation. The bainite matrix has shown great ductility through the generation of high angle grain boundaries and / or large orientation gradients around voids, which are usually found close to the bainite and MA / auto-tempered martensite interfaces or fragmenting the MA phase. Special attention has been paid to the stability of the retained austenite (RA) during the test, which may eventually be transformed into martensite (Transformation Induced Plasticity, or TRIP e ﬀ ect). one the to the rotations phase


Introduction
In the ever-increasing interest of the automotive sector to achieve better mechanical performance of the steel and lightweight new components, there is a clear tendency to develop new steel grades. In this line, the use of multiphase steels tries to take advantage of the optimized combination of the mechanical properties of each of the micro-constituents. The effect of the different phases on the mechanical properties in DP (Dual Phase) and TRIP (Transformation Induced Plasticity) steels has been widely reported [1]. DP steels consist of a mixture of a ferritic matrix with martensitic islands. The ferritic matrix imparts ductility, whereas martensite allows the accomplishment of significant strain hardening rates that simultaneously improve the yield strength and the tensile strength. On the other hand, TRIP steels are more complex from the microstructural point of view. The matrix is mainly ferritic, whereas the hard phase is bainite/martensite. The retained austenite (RA), through its characteristic transformation to martensite during strain, is crucial to understanding the enhanced ductility of these TRIP steels [2].

Experimental Section
The present study has been carried out on samples machined out from cold-rolled lab casts of composition 0.22% C, 2.0% Mn, 1.3% Si, 0.03% Al, and 0.005% N. They have been hot-rolled to a 4 mm thickness and, finally, cold-rolled down to 1 mm thickness before machining. The thermal cycle consisted of an annealing treatment at 920 • C for 2 min, cooling at 50 • C·s −1 down to 420 • C and an overaging time of 120 s in the bainitic region. Then, the material was cooled to room temperature at 2 • C·s −1 . The selection of this condition was carried out after a detailed analysis of the microstructure results derived from the BASEFORM project [7].

Tensile Test
The tensile tests were run in an Instrom tensile machine 5982 (Instron, Grove City, PA, USA) on subsize specimens with a gauge length of 3.0 mm and with 2.5 mm width, which were machined to facilitate the tracking of the deformation and the onset of the necking. An extensometer of 3 mm length and 1.5 mm width were used to control the amount of applied strain during the tensile test. The test was stopped at about every 0.05 of longitudinal strain to perform the microstructural analysis on the RD (rolling direction)-TD (transverse direction) plane.
The plastic deformation at each step was determined according to the following methodology: 1.
The width of the specimen was optically measured after every strain.
• The measurement of the thickness was ruled out since thickness observations are not compatible with, on the one hand, the EBSD preparation and, on the other hand, the tracking of a selected zone (typically by a grid of indentations) during uniform deformation due to limitations inherent to the geometry of the tensile specimen.

•
The longitudinal strain, measured through the lengthening with the extensometer, is not readily measurable once the necking phenomenon takes place. The printing of a grid (by a square net of low weight indentations) on the sample surface from the initial material was the other option, but it has been dismissed. Two reasons are behind this decision. The surface roughening is expected with strain, and extra preparation would be necessary for further EBSD analysis. This may remove the grid. The second reason is that the grid itself could be a set of points on which strain can accumulate, leading to biasing results.

2.
It is possible to correlate the relationship between initial (w 0 = 2.5 mm) and current width (w) with the initial (t 0 = 1 mm) and current thickness (t) after every test interruption. According to [18], if the strain hardening exponent (n) is known, as well as the thickness and width relation, it is feasible to get this relationship; see Figure 1, obtained from data extracted from [18] after some linear extrapolation. In the present work, w 0 /t 0 = 2.5 and the n 0.16.

3.
Thus, with the help of the latter work and using the equation ε t = −ln t t 0 , the thickness strain would be directly determined throughout the test irrespective of the stage: Uniform or necking. 4.
The output strain will thus be the thickness plastic strain.
In the post-mortem specimen, a grid of micro-indentations separated by 200 µm was printed in the necking region to determine unambiguously the width of the specimen where the EBSD (Electron Backscatter Diffraction) scans were performed. set of points on which strain can accumulate, leading to biasing results. 2. It is possible to correlate the relationship between initial (w0 = 2.5 mm) and current width (w) with the initial (t0 = 1 mm) and current thickness (t) after every test interruption. According to [18], if the strain hardening exponent (n) is known, as well as the thickness and width relation, it is feasible to get this relationship; see Figure 1, obtained from data extracted from [18] after some linear extrapolation. In the present work, w0/t0 = 2.5 and the n ≅ 0.16.

Microstructural Characterization
The microstructural characterization was carried out through FEG-SEM (Field Emission Gun SEM) imaging in a JEOL JSM-7000F microscope (JEOL Ltd., Tokyo, Japan), and EBSD scanning with the program Oxford Channel 5 (5.0.9.0, HKL Tech. A/S, Hobro, Denmark). The experimental data have been transformed into a format whereby the OIM 5.0 program can handle them. Over 10 micrographs at ×10000 at the center and at 1 4 of the thickness were taken to measure the number of secondary phases in the initial microstructure after etching with Klemm III reactant through the point counting method. No significant changes were observed in through-thickness observations, nor were any band structures that can be originated in this kind of steel, usually due to Mn segregation.
The EBSD study has been split into two parts:

1.
A detailed analysis of the retained austenite evolution with the strain for the low strain region (below 0.13 longitudinal strain). A fixed region of about 25 × 25 µm 2 was selected and scanned to follow its evolution in the first stages of deformation. Four indentations in the gauge length were arranged in a square to track the region of interest. The step size was set at 100 nm.

2.
The evolution of the microstructure across the necking region. Areas of about 25 × 25 µm 2 with a step size of 100 nm were run every 200 µm in the longitudinal direction.
Two crystalline structures were considered in the EBSD work: BCC (body centered cubic-ferrite) and FCC (face centered cubic-austenite). The bainitic ferrite and the auto-tempered martensite (ATM) are indexed as BCC. The distinction between these two micro-constituents relied on the average EBSD pattern quality. In general, the ATM has a lower pattern quality (PQ)-darker in the PQ maps-than bainite-brighter in the PQ maps. Nevertheless, the amount of ATM is indeed low-below 2%. The sample preparation went through a soft grinding and was finally gently polished down to colloidal silica (0.05 µm) to avoid austenite transformation. Special attention was paid to the clean-up procedure to achieve reliable methods. In general, the scans showed a rate of success above 85-90%. As a result of the clean-up procedure, less than 5% of the data were re-indexed.
Regarding X-ray diffraction, the scans were performed in a Philips PW1825 diffractometer (Philips Analytical, Almelo, The Netherlands) with an angular step size of 0.02 • and with Cu-Kα radiation to determine the amount of volume fraction of retained austenite after every interruption before necking in the angular range 2θ, 40 • -120 • . The calculations of the volume fraction of retained austenite were done according to ASTM E975-13 for 4 diffraction peaks: {200} and {220} for austenite and {200} and {211} for ferrite, and the constants for the Cu-Kα radiation were extracted from Jarvinnen's work [19]. The carbon content of the austenite in the initial state was determined according to equation a γ = 3.572 + 0.033 wt.% C + 0.0012 wt.% Mn-0.00157 wt.% Si [20].
Once the necking occurred, X-ray profiles were also recorded across the necking region at several distances from the center of the necking, using a collimator of 0.3 mm and a slit width of 1 mm.

Results and Discussion
This section has been divided into two parts that correspond to the uniform deformation and the post-necking stages during the tensile test. In both sections, the development of the substructure in the bainitic matrix and the mechanical stability of the RA austenite are tackled and discussed with regard to the applied thickness plastic strain.

Uniform Strain Region
In Figure 2a,b, the microstructure corresponding to the initial state is illustrated. The FEG-SEM micrograph shows a bainitic matrix in which a significant volume fraction of secondary phases is distinguished in light grey color. In this sheet, the banding of secondary phases is not detected at low magnifications. In general, the secondary phases have a blocky type morphology, which is usually related to granular upper bainitic (GB) structures [21]. The amount of lath-like upper bainitic (LLUB) structures is relatively low. The quantitative results achieved from X-ray diffraction as well as the EBSD map, Table 1 and Figure 2c, indicate that a large amount of retained austenite is generated as a result of the thermal cycle applied, leading to almost 14% volume fraction of this phase. The difference in the measurement of the retained austenite between these two techniques can arise from the lack of the EBSD to detect film-like retained austenite and the mis-indexing of the regions close to interfaces between the matrix and the austenitic islands. Other secondary phases are also observed to be associated with the largest islands, specifically auto-tempered martensite (ATM). In general, the size of the retained austenite islands exceeds 0.2 µm. Thus, they have a suitable size to be studied by the EBSD technique. The stress-strain curve is plotted in Figure 3a. The full summary of the initial microstructure and parameters derived from tensile test is gathered in Table 2. In addition to the stress-strain flow curve, the strain hardening rate was also plotted as a function of the longitudinal strain in Figure 3b. The shape is the typical one for steels that undergo a significant TRIP effect in the early stages highlighted by the presence of a local maximum (a maximum around 0.05 longitudinal strain) and a progressive decline of the strain rate afterwards. This is the usual shape of the carbide-free bainitic steels [22], and the appearance of a peak is related to the balance of the decrease in the work hardening of bainite and the increase in the work hardening due to transformation of RA to martensite.
The same region as the one in Figure 2b was analyzed at a higher plastic strain level, Figure 4. EBSD-KAM (Kernel Average Misorientation) and PQ (Pattern Quality) + austenite maps prove that increasing the amount of the plastic strain brings about significant changes in the KAM parameter together with a decline in the amount of RA. In particular, the KAM parameter is widely used to determine the degree of deformation in dislocated microstructures [23]. In this work, a methodology analogous to the one developed by Kamaya in [24] and already applied in [8] has been adopted to measure the local orientation gradients. In the Appendix A section, an explanation of the method to calculate the local orientation gradients is provided. This methodology aims at minimizing the effect of the experimental data acquisition parameters and the EBSD step size on the misorientation measurements with distance [24]. In this work, two maximum values, 5 • and 10 • , have been set to evaluate the evolution of this parameter with the applied strain. These maxima restrict the misorientation between points, which must not exceed the imposed angular threshold. Expectedly, the orientation gradients increase almost linearly with the plastic strain in the uniform region, as seen in Figure 4e. This kind of linear trend has been reported elsewhere [25] for the low plastic strain region.
stress-strain flow curve, the strain hardening rate was also plotted as a function of the longitudinal strain in Figure 3b. The shape is the typical one for steels that undergo a significant TRIP effect in the early stages highlighted by the presence of a local maximum (a maximum around 0.05 longitudinal strain) and a progressive decline of the strain rate afterwards. This is the usual shape of the carbidefree bainitic steels [22], and the appearance of a peak is related to the balance of the decrease in the work hardening of bainite and the increase in the work hardening due to transformation of RA to martensite.     The quantification of the volume fraction of retained austenite exhibits a steep drop for low strains within the uniform deformation region. Some previous austenitic grains, as just explained above, are either not indexed anymore or are partially indexed. This lack of indexing is commonly reported to be a consequence of the presence of high-carbon martensite, which shows a very poor or no pattern in conventional EBSD, or it is observed as an area with low pattern quality (PQ) and a high dislocation density [30]. Thus, blackened regions are related to retained austenite that undergoes   Looking at the film-like retained austenite, conventional EBSD is not a suitable technique for its characterization, mainly when the sizes do not exceed 100 nm. However, its likely transformation is expected to be delayed at larger strains due its higher carbon content [33] and its higher hydrostatic pressures [34] compared to the large islands of retained austenite. The mechanical stability of the retained austenite is a complex function of its morphology, size, surrounding phases, crystal orientation, and composition [26]. Specifically, the stability of retained austenite, when exhibits a block type morphology, has been observed to be low with respect to film-like retained austenite [27]. Here, the blocky type RA is the predominant morphology and the phases that are surrounding the RA grains can be either bainitic ferrite, high-carbon martensite belonging to MA islands or auto-tempered (low-carbon) martensite. The comparison of the maps of same region in Figure 4c,d for two strain levels allows observing that:

•
The retained austenite partially bounded by non-indexed regions, which correspond to martensitic phases, seem to exhibit a poor stability with the strain. This is evident in the regions that are indicated with the red arrows. The distinctly different mechanical behavior of the phases in these microstructures has been put forward through nanohardness measurements [8]. The high-carbon martensite and auto-tempered low-carbon martensite display the highest nanohardness, whereas bainite is the softest phase. In addition, the stability is lower for the retained austenite located between two bainitic packets, indicated by the yellow arrows. Thus, in either case, the stress triaxiality is enhanced due to the mechanical contrast arising from the surroundings. This mechanical contrast is a consequence of either the difference in crystal orientation of adjacent bainitic packets against externally applied stresses [28] or to an even more pronounced variation of the mechanical behavior between bainite and (high-carbon) martensitic-type phases.

•
The austenitic islands that are surrounded only by bainite within the same bainitic packet (see blue arrows in Figure 4c,d) have a greater capability to accumulate deformation without any apparent transformation to martensite. In contrast to the case above, the bainite that embeds the austenite islands now has almost the same orientation. This greater stability against transformation has also been observed in austenite embedded in large ferrite grains in TRIP steels [29].
The quantification of the volume fraction of retained austenite exhibits a steep drop for low strains within the uniform deformation region. Some previous austenitic grains, as just explained above, are either not indexed anymore or are partially indexed. This lack of indexing is commonly reported to be a consequence of the presence of high-carbon martensite, which shows a very poor or no pattern in conventional EBSD, or it is observed as an area with low pattern quality (PQ) and a high dislocation density [30]. Thus, blackened regions are related to retained austenite that undergoes induced martensitic transformation (TRIP effect). In the present study, the transformation of retained austenite mostly occurs before 0.025 thickness strain is achieved. This agrees with the shape of the strain hardening rate, which reaches a maximum close to 0.025 thickness. This strain corresponds to 0.05 longitudinal strain in Figure 3a. Afterwards, the amount of retained austenite remains almost constant at a value of 4-5%, nearly 30% of the initial RA volume fraction, up to 0.08 thickness strain and 0.13 longitudinal strain. Therefore, the enhancement of the ductility based on the TRIP effect is limited to the very first stages of the deformation and, after 0.05 thickness strain, dislocation plasticity and crystal rotations rule the ductility of the austenite grains, also reported in [30]. The analysis of the behavior of the untransformed austenite grains with strain has been performed through the ratio of the GOS (Grain Orientation Spread) parameter and the grain size (D) for grains of austenite with diameters greater than 300 and 500 nm. The mean values of these two grain populations are about 500 and 800 nm, respectively. The GOS parameter measures the average misorientation of any point belonging to a grain with regard to the mean orientation of the grain. Therefore, this parameter looks at the whole grain, in contrast to the previously analyzed local orientation gradients. This kind of angular misorientation to distance-related parameters has also been proposed in [31]. The GOS/D rises monotonically with an amount of thickness strain up to 0.05 and can reach 3 • µm −1 in spite of the small sizes of RA (Figure 5a). Depending on the minimum grain size considered, the mean austenite grain size is close to 500 and 800 nm. In terms of dislocation densities, the density of GNDs amounts to 3.7 × 10 14 m −2 at the end of the uniform strain according to the formulation proposed by Mortensen and Kubin [32]. On the other hand, probably even slightly overestimated, the density is close to 1.4 × 10 14 m −2 at the initial state. The rotation of the crystalline orientation of the austenite grains is readily observed in the point to origin misorientation profiles. In fact, in the RA grains corresponding to the initial state, some misorientation above the noise level is found, typically below 0.5-1 • , reaching a value of 1.5 • from one boundary to the opposite one. This is due to the crystalline rotations induced in RA during γ (austenite)→α B (bainite) phase transformation.

Necking Region
The increase in the strain in the uniform region involves a steady rise of the substructure whose main consequence is the development of local orientation gradients, due to the creation of low angle boundaries or local crystalline curvature with the strain, as seen in Figure 6a-c. At low strain, there are several places close to the high angle boundaries in which the strain accumulates, indicated by the black arrows in Figure 6b, and a relatively coarse subgrain structure is about to form. When the applied thickness strain exceeds 0.2, the bainitic matrix clearly develops a subgrain-type structure and even high angle boundaries are developed due to bainite fragmentation, indicated by the white arrows in Figure 6c. This subgrain structure has also been observed in another CFB steel with filmlike austenite [35]. The fragmentation of the grains is depicted in Figure 6a,b for different strain levels. This is not only reflected by the increase of the density of high angle boundaries (HAB) per unit area (Figure 6d), but also by the shape of the misorientation distribution, in which the fraction of boundaries between 50 and about 62° are lowered by the strain (Figure 6e) as a result of the grain boundaries' rotations. The effect of the transformation induced martensitic formation on the mesotexture should have enhanced this region because of the typical K-S (Kurdjumov-Sachs) and N-W (Nishiyama-Wasserman) orientation relationships between parent austenite and product martensite, which are located at 53 and 60° [36]. Nevertheless, that is insufficient to compensate for the generation of new boundaries that mainly contribute to the angles between 15 and 50°. Of course, the frequency for low misorientation angles is promoted in this strain range.  Looking at the film-like retained austenite, conventional EBSD is not a suitable technique for its characterization, mainly when the sizes do not exceed 100 nm. However, its likely transformation is expected to be delayed at larger strains due its higher carbon content [33] and its higher hydrostatic pressures [34] compared to the large islands of retained austenite.

Necking Region
The increase in the strain in the uniform region involves a steady rise of the substructure whose main consequence is the development of local orientation gradients, due to the creation of low angle boundaries or local crystalline curvature with the strain, as seen in Figure 6a-c. At low strain, there are several places close to the high angle boundaries in which the strain accumulates, indicated by the black arrows in Figure 6b, and a relatively coarse subgrain structure is about to form. When the applied thickness strain exceeds 0.2, the bainitic matrix clearly develops a subgrain-type structure and even high angle boundaries are developed due to bainite fragmentation, indicated by the white arrows in Figure 6c. This subgrain structure has also been observed in another CFB steel with film-like austenite [35]. The fragmentation of the grains is depicted in Figure 6a,b for different strain levels. This is not only reflected by the increase of the density of high angle boundaries (HAB) per unit area (Figure 6d), but also by the shape of the misorientation distribution, in which the fraction of boundaries between 50 and about 62 • are lowered by the strain (Figure 6e) as a result of the grain boundaries' rotations. The effect of the transformation induced martensitic formation on the mesotexture should have enhanced this region because of the typical K-S (Kurdjumov-Sachs) and N-W (Nishiyama-Wasserman) orientation relationships between parent austenite and product martensite, which are located at 53 and 60 • [36]. Nevertheless, that is insufficient to compensate for the generation of new boundaries that mainly contribute to the angles between 15 and 50 • . Of course, the frequency for low misorientation angles is promoted in this strain range. Looking at the in-grain scale, the orientation gradients grow as a result of the applied strain beyond the uniform deformation stage (Figure 6f). Average gradients as high as 12° μm −1 are observed, which corresponds to geometrically dislocation density,ρGND, of about 1.5 × 10 15 m −2 . This high amount of geometrically necessary dislocations is in agreement with the dislocation densities reported in [37]. However, the high amount of statistically stored dislocations inside the lath is also responsible for the isotropic hardening at any stage, and the micro-mechanical behavior simulations in this respect have led to values as high as 10 16 m −2 [7]. Therefore, a large density of dislocations, which has no contribution in the misorientation development, are not considered in the present analysis.
Looking at the retained austenite, its evolution after the whole test is shown in Figure 7a. The onset of the localized necking is related with the decrease in the amount of retained austenite. The change in the local stress conditions triggers the austenite to martensite transformation and, after 0.2, the retained austenite does not exceed 1%. The effect of the stress conditions on the mechanical Looking at the in-grain scale, the orientation gradients grow as a result of the applied strain beyond the uniform deformation stage (Figure 6f). Average gradients as high as 12 • µm −1 are observed, which corresponds to geometrically dislocation density, ρ GND , of about 1.5 × 10 15 m −2 . This high amount of geometrically necessary dislocations is in agreement with the dislocation densities reported in [37]. However, the high amount of statistically stored dislocations inside the lath is also responsible for the isotropic hardening at any stage, and the micro-mechanical behavior simulations in this respect have led to values as high as 10 16 m −2 [7]. Therefore, a large density of dislocations, which has no contribution in the misorientation development, are not considered in the present analysis.
Looking at the retained austenite, its evolution after the whole test is shown in Figure 7a. The onset of the localized necking is related with the decrease in the amount of retained austenite. The change in the local stress conditions triggers the austenite to martensite transformation and, after 0.2, the retained austenite does not exceed 1%. The effect of the stress conditions on the mechanical stability of retained austenite has been explained and discussed in [16], and the higher degree of triaxiality in the necking region is expected to resume the transformation to martensite of the remaining austenite. The X-ray profiles extracted at various positions of the tensile specimen demonstrate that the amount of retained austenite in the necking region is below 1% (Figure 7b) which corresponds to the limit of the resolution of the X-ray technique [38]. By comparison of this result with the volume fraction determined by EBSD in the localized necking region, it can be concluded that the fraction of film-like RA, more resistant to strain than RA with island morphologies, is almost null in this region. Considering the FEG-SEM pictures in Figure 2a,b, the amount of film-like retained austenite does not represent more than 2% of the initial state and, hence, its effect on the ductility in this specimen turns out to be rather limited. stability of retained austenite has been explained and discussed in [16], and the higher degree of triaxiality in the necking region is expected to resume the transformation to martensite of the remaining austenite. The X-ray profiles extracted at various positions of the tensile specimen demonstrate that the amount of retained austenite in the necking region is below 1% (Figure 7b) which corresponds to the limit of the resolution of the X-ray technique [38]. By comparison of this result with the volume fraction determined by EBSD in the localized necking region, it can be concluded that the fraction of film-like RA, more resistant to strain than RA with island morphologies, is almost null in this region. Considering the FEG-SEM pictures in Figure 2a,b, the amount of film-like retained austenite does not represent more than 2% of the initial state and, hence, its effect on the ductility in this specimen turns out to be rather limited.
Regarding the change in GOS/D, a large scatter in the data extracted in the localized necking area is observed (Figure 7b). The minimum number of austenite grains measured in conditions larger than 300 nm was 10, whereas for 500 nm, it was two to five (Figure 7d). These minima are the ones achieved in the most strained conditions (0.24-0.26 thickness strain). The heterogeneity of the local deformation and the orientation of the retained austenite in the analyzed regions are possible causes for such scatter [30]. However, irrespective of the minimum austenite grain size, the average grain curvature parameter (GOS/D) seems to level off at a value of 3.5° μm −1 . Compared to the bainitic matrix, the local orientation gradients, estimated like they were for bainite, show a clear stagnation at the large strain regime (Figure 7c). Instead, the bainite is able to accumulate the strain by formation not only of continuous misorientations that accumulate with distance, but also a subgrain structure, as illustrated in Figure 6c. This kind of subgrain structure is usually reported to increase its related misorientation as strain increases [39]. The stagnation value in the austenite is around 5.4 and 6.1 °μm −1 for the 5 and 10° angular thresholds. These values are lower than the ones for bainite at large strains, which exceed 8 and 11° μm −1 . The stagnation of these two parameters in RA may arise from the combination of several factors. Larger strains may rotate RA grain to positions in which the transformation to martensite is induced. Another reason could be that, due to their small size (typically below 1 μm), the significance of the grain boundaries is relevant in the absorptiongeneration of dislocations and in the inhibition of the generation of substructure [40]. The formation of voids in this kind of bainitic matrix steel is limited to the necking region as opposed to that carried out in DP steels as has been widely observed, where voids are already observed in the uniform strain region [13]. Another interesting observation is the low density of void events for the range of thickness strains (0.1-0.3) that requires a very careful observation of the specimen to find the voids in the microstructure. In the present case, the damage is linked to the void formation in the interfaces between secondary phases and the matrix (Figure 8a), fragmentation of the secondary phases (Figure 8b), and the fragmentation or decohesion of inclusions from the matrix (Figure 8c). The latter is relatively negligible for steels with clean compositions as the one studied in this work. After an extensive observation of the necking region, the very first voids have been associated with a thickness strain of about 0.1-0.12. At this stage of the damage formation, the voids are connected to the bainite-MA island interfaces (Figure 8a). As positions closer to the center of the necking are analyzed, not only matrix-MA related voids are detected. Particularly, in the region subjected to a thickness strain higher than 0.2, the other void formation mechanisms are active, as illustrated in Figure 8b. The fragmentation of secondary phases is closely related to medium-large sized MA islands (typically over 1 μm size) with complex shapes, like the one in Figure 8b. The microcracks/voids are often located between two separated regions of the same MA island in a zone where it gets narrower; there is an acute angle, as illustrated by the red arrows in Figure 8b. From the mechanical point of view, the MA phase and bainite can show a significant contrast, as reported elsewhere [8]. The MA island-which consists of pure austenite/martensite or a mixture of both of them-behaves as a hard phase, whereas the surrounding bainite deforms and builds up stresses around the acute angle. The observations in the uniform region section allow the conclusion that a large amount of retained austenite with complex shapes associated with medium-large MA islands Regarding the change in GOS/D, a large scatter in the data extracted in the localized necking area is observed (Figure 7b). The minimum number of austenite grains measured in conditions larger than 300 nm was 10, whereas for 500 nm, it was two to five (Figure 7d). These minima are the ones achieved in the most strained conditions (0.24-0.26 thickness strain). The heterogeneity of the local deformation and the orientation of the retained austenite in the analyzed regions are possible causes for such scatter [30]. However, irrespective of the minimum austenite grain size, the average grain curvature parameter (GOS/D) seems to level off at a value of 3.5 • µm −1 . Compared to the bainitic matrix, the local orientation gradients, estimated like they were for bainite, show a clear stagnation at the large strain regime (Figure 7c). Instead, the bainite is able to accumulate the strain by formation not only of continuous misorientations that accumulate with distance, but also a subgrain structure, as illustrated in Figure 6c. This kind of subgrain structure is usually reported to increase its related misorientation as strain increases [39]. The stagnation value in the austenite is around 5.4 and 6.1 • µm −1 for the 5 and 10 • angular thresholds. These values are lower than the ones for bainite at large strains, which exceed 8 and 11 • µm −1 . The stagnation of these two parameters in RA may arise from the combination of several factors. Larger strains may rotate RA grain to positions in which the transformation to martensite is induced. Another reason could be that, due to their small size (typically below 1 µm), the significance of the grain boundaries is relevant in the absorption-generation of dislocations and in the inhibition of the generation of substructure [40].
The formation of voids in this kind of bainitic matrix steel is limited to the necking region as opposed to that carried out in DP steels as has been widely observed, where voids are already observed in the uniform strain region [13]. Another interesting observation is the low density of void events for the range of thickness strains (0.1-0.3) that requires a very careful observation of the specimen to find the voids in the microstructure. In the present case, the damage is linked to the void formation in the interfaces between secondary phases and the matrix (Figure 8a), fragmentation of the secondary phases (Figure 8b), and the fragmentation or decohesion of inclusions from the matrix (Figure 8c). The latter is relatively negligible for steels with clean compositions as the one studied in this work. After an extensive observation of the necking region, the very first voids have been associated with a thickness strain of about 0.1-0.12. At this stage of the damage formation, the voids are connected to the bainite-MA island interfaces ( Figure 8a). As positions closer to the center of the necking are analyzed, not only matrix-MA related voids are detected. Particularly, in the region subjected to a thickness strain higher than 0.2, the other void formation mechanisms are active, as illustrated in Figure 8b. The fragmentation of secondary phases is closely related to medium-large sized MA islands (typically over 1 µm size) with complex shapes, like the one in Figure 8b. The microcracks/voids are often located between two separated regions of the same MA island in a zone where it gets narrower; there is an acute angle, as illustrated by the red arrows in Figure 8b. From the mechanical point of view, the MA phase and bainite can show a significant contrast, as reported elsewhere [8]. The MA island-which consists of pure austenite/martensite or a mixture of both of them-behaves as a hard phase, whereas the surrounding bainite deforms and builds up stresses around the acute angle. The observations in the uniform region section allow the conclusion that a large amount of retained austenite with complex shapes associated with medium-large MA islands already transforms before necking, which would impact the formation of cracks/voids. The retained austenite associated with medium-large MA islands with complex shapes are more prone to transform into pure martensite. Hence, when it transforms into martensite, the latter inherits its high carbon content and is expected to exhibit low plasticity due to its high hardness [41]. This, in turn, would affect the strain partitioning between the matrix and the phase during further strain. The formation of the crack brings about the release of energy. This energy release is absorbed by the bainitic matrix. Bainite shows its great ability to deform plastically before fracture takes place through the generation of high angle boundaries. Specifically, the bainitic matrix is also able to develop high angle boundaries around the voids. These high angle boundaries could prevent their growth and final coalescence, as shown in Figure 8c. The most complex morphologies are usually associated with the prior austenite grain boundaries, as shown in Figure 8e, where the retained austenite grains with the same IPF (Inverse Pole Figure) color belong to the same previous austenite grain. In the map at the bottom of Figure 8e, the misorientation with regard to point A (grains colored in greyscale) is depicted. It is clear that the bainitic transformation produces the rotation of the austenite in the same manner as described in Figure 5b. The presence of retained austenite with low stability against strain related to the prior austenite can be regarded as detrimental in terms of ductility and toughness. The early transformation to martensite of these large retained austenite islands makes them hard and brittle at the same time. The appearance of brittle phases at the (prior) grain boundaries, as with the MA phase, has been reported to impair the ductility in TRIP steels [42] and the energy absorption in the Charpy test in low alloyed [43] and high strength steels [44].
All in all, the morphological features of the MA phase play a relevant role as a second stage in the void formation, just after decohesion of the matrix and the MA phase interfaces. Other aspects, such as the local texture of the bainitic matrix, are of importance in the generation of high stresses that produce the void formation [45]. The role in the damage nucleation of the film-like austenite can be considered negligible. The onset of voids around the film-like structures has not been detected in the first stages of necking, nor in the large strain regime.
Another point of interest is the absence of shear bands in the matrix for the applied maximum strain in this specimen. This substructural feature has been reported in several works about DP and TRIP steels, and it appears to be related to a combination of several factors such as the nature/size/morphology of the involved phases and their spatial arrangements [46,47]. For instance, in DP steels, coarse grains of ferrite with small martensite islands are more prone to developing such shear bands, which in turn favors the formation of new voids [46]. In the present case, the bainitic matrix ensures a very fine microstructure (about 1.8 µm average bainitic packet size) and a lower difference in hardness among the various micro-constituents. Thus, these microstructures are more resistant to the onset of a generalized formation of shear bands, which has often been related to the final fracture.

•
In the uniform deformation region, the TRIP effect is active for low strains (below 0.05 thickness strain), giving rise to a decrease in the fraction of retained austenite from about 14% to 4%. The transformation almost halts, leading to a stable region in which retained austenite deforms plastically. Finally, in the necking region, the strain-induced martensitic transformation resumes, and less than 1% of austenite is measured for 0.26 thickness strain with conventional EBSD.

•
From the morphological point of view, a significant number of large blocky austenite grains is stable in the uniform deformation region. These retained austenite grains exhibit a great ability to develop orientation gradients at the local scale, with lengths of less than 200 nm. Apparently, the RA grains fully surrounded by the bainite with similar crystal orientation present a higher stability than when they are located between bainitic packets with different orientations or adjacent to another harder secondary phase. This has been partially attributed to the local stress triaxiality conditions during tensile tests.

•
The onset of the void formation is closely related to the interfaces between bainite and MA phase. The fast drop in RA associated with blocky type morphologies makes the interfaces of transformed RA and the bainitic matrix suitable locations for the onset of the void generation. The fragmentation of blocky MA appears in a later stage of strain as a new mode for void formation. At this point, the ability of the bainitic matrix to create high angle boundaries around voids plays a relevant role in the control of their growth and eventual coalescence.

•
The morphology of the MA phase influences its fracture. In order to improve the cold formability of the TRIP-aided bainitic steels, the number of MA islands associated with previous austenite grain boundaries should be eliminated or at least minimized. Figure A1. Explanation of the method applied to calculate the local orientation gradients.