The Effect of Static Recrystallisation of Parent Austenite on the Lath Martensite Intervariant Boundary Network

Abstract

The current study investigated the effect of static recrystallisation (SRX) of parent austenite microstructure on the lath martensite intervariant boundary network. In general, the misorientation angle distribution of lath martensite exhibited a bimodal distribution in the range of 10–22 deg and 47–60 deg for all thermomechanical conditions, closely matching the misorientations expected from the ideal Kurdjumov-Sachs orientation relationship. Nevertheless, the population of 60 deg misorientation angle initially reduced with the extent of softening up to 50 pct beyond which it progressively increased to a maximum in the fully recrystallised condition. This was closely consistent with the trend noticed for the 60 deg/[110] intervariant boundaries having twist character. This was explained through the distinct variant selection mechanism occurring due to the change in the parent austenite grain characteristics (i.e., size and deformed state) upon static recrystallisation. The recrystallisation initially led to the formation of fine dislocation-free parent austenite grains up to 50 pct softening, promoting a 4-variant (V₁V₂V₃V₄) clustering arrangement to decrease the strain related to the martensite transformation. In turn, this promoted the formation of symmetric tilt 60 deg/$\left[ {11\overline{1}} \right]$ intervariant boundaries, which ultimately reduced the 60 deg/$\left[ {110} \right]$ boundaries. Further softening, accompanied with the growth and/or coalescence of parent austenite grains, altered the variant clustering to a 3-variant arrangement (V₁V₃V₅) and the increase in 60 deg/$\left[ {110} \right]$ twist intervariant boundaries. Indeed, the martensite boundary network was largely controlled by the SRX parent austenite characteristics (i.e., size), which can be controlled to enhance the materials performance for a given application.

1 Introduction

The grain boundary is one of the key microstructure constituents, affecting polycrystalline materials processing and performance, namely corrosion behaviour,^[1,2] toughness,^[2,3,4] segregation of alloying elements^[5,6,7] and precipitation/phase transformation.^[8] Different approaches have been developed to architecture the grain boundary network (i.e., population and connectivity) of materials for a given property of interest; known as grain boundary engineering (GBE). Early attempts in the field of GBE have been focused on metals with the face centred cubic (FCC) structure, where high energy grain boundaries are substituted by boundaries with a low energy arrangement, such as Σ3 annealing twin boundaries, using an iterative recrystallisation process.^{[9,10,11,12,13,14]} However, this approach does not lead to a similar result for polycrystalline materials with a distinct crystal structure (e.g., body centred cubic, BCC) and those undergoing phase transformation (e.g., steel). Unlike FCC materials, the recrystallisation alters the grain boundary network of BCC materials (e.g., ferrite) differently, enhancing the population of low angle boundaries at the expense of high angle boundaries due to the promotion of a specific overall crystallographic texture (γ-fibre).^[15] In the latter, the phase transformation replaces the high temperature grain boundary network of the parent austenite by different transformed phases (i.e., ferrite, martensite and bainite), depending on the transformation mechanism (i.e., diffusion and shear).^{[4,16,17,18,19,20,21,22,23]}

In general, the phase transformation in low carbon steels leads to 16 distinct boundaries based on the Kurdjumov-Sachs orientation relationship (KS OR) between the parent austenite and daughter phase/s (ferrite, martensite and bainite).^[18] Among these boundaries, it appears that boundaries with 60 deg/$\left[ {11\overline{1}} \right]$ and 60 deg/$\left[ {110} \right]$ lattice misorientations have the most impact on the material properties in a martensitic microstructure, since the former has the least energy and the latter displays the maximum population with moderately high energy.^[4,5,24] It appears that there is a direct relationship between grain boundary energy and the crack propagation rate, as demonstrated by molecular dynamics simulation.^[24] Therefore, the work to date suggests that an increase in the population of 60 deg/$\left[ {11\overline{1}} \right]$ low energy symmetric boundaries leads to a reduction in the 60 deg$\left[ {110} \right]$ high energy twist boundary, ultimately improving the material toughness.^[4,24]

Recent investigations have revealed that the grain boundary network of low carbon low alloy steels can be manipulated through an alteration in the mechanism of phase transformation (shear vs diffusion)^{[17,18,19,20,21,22]} and/or the parent austenite characteristics prior to the phase transformation, such as composition (i.e., Mn addition),^[23] grain size^[4] and deformation state/retained strain.^[17] Interestingly, these approaches largely alter the martensite start (M_s) and finish (M_f) transformation temperatures (i.e., M_s and M_f, respectively), which has been shown to stimulate the promotion of specific variant cluster/s (variant selection) to accommodate the strain related to the martensitic transformation.^[4,16,23] The dislocation substructure introduced during austenite deformation also enhances the formation of specific martensitic variant/s with their habit planes strongly matched with the primary slip plane and/or the secondary slip plane of the parent austenite.^[25,26] The variant selection introduced through these approaches ultimately promotes the population of low energy 60 deg/$\left[ {11\overline{1}} \right]$ with a symmetric tilt character and, in turn, reduces the high-energy 60 deg/$\left[ {110} \right]$ twist boundary population.^[16]

The static recrystallisation (SRX) of austenite is also one of the common practices occurring during the thermomechanical processing of steel. This involves the nucleation and growth of dislocation-free grains from the deformed parent austenite containing a dislocation substructure, which simultaneously undergoes dislocation annihilation (i.e., recovery) during the post-deformation annealing time. This leads to a change in the characteristics of the parent austenite, depending on the extent of SRX. However, it is not clear how the parent austenite characteristics developed by the SRX phenomenon influence the intervariant boundary network during the martensite transformation. This work, therefore, focused on the deformation of parent austenite and subsequent isothermal holding followed by a martensite transformation upon quenching, to understand the effect of SRX softening on the boundary network characteristics in a lath martensitic microstructure.

2 Experimental Procedure

The as-received material was in the form of hot rolled plate with a thickness of 17 mm, having a composition of 0.043 C, 1.68 Mn, 0.2 Si, 0.017 Ti, 0.021 Nb, 0.002 B (in wt pct). Torsion samples with a 22 mm gauge length and a 6.7 mm diameter were machined, having their longitudinal axis along the rolling direction (RD). The torsion samples were subjected to different thermomechanical processing routes using a hot torsion simulator equipped with an induction furnace along with a computer data acquisition system, as discussed in.^[27] The temperature was measured during the experiment using an N thermocouple inserted in the shoulder of the torsion samples. The temperature difference between both ends of the torsion gauge length was less than 10 °C throughout the test. The torque-twist data were employed to calculate Von Mises equivalent stress-strain values using the Fields and Backofen approach.^[28]

The samples were reheated to 1200 °C at a heating rate of 5 °C/s and subsequently held isothermally for 300 s to achieve a fully austenitic microstructure. Afterwards, they were deformed to a strain of 0.3 at a strain rate of 1 s⁻¹ and subsequently held at 1200 °C for 40 s. This led to full recrystallisation of the austenite microstructure with an average grain size of 92 μm. The samples were then cooled at a rate of 1 °C/s to 1000 °C, which is well above the non-recrystallisation temperature for the experimental steel (i.e., T_nr =  ~ 900 °C,^[16] the temperature below which the recrystallisation is retarded due to the precipitation pinning and/or solute drag effects^[27]). Then, the samples were isothermally held for 60 s to obtain a uniform temperature throughout the gauge length. One specimen was deformed at 1000 °C at a strain of 3 at a strain rate of 1 s⁻¹ to determine the critical strain for the onset of dynamic recrystallisation (ɛ_c) using the Poliak and Jonas approach,^[29] as discussed later (Figure 1). Afterwards, some specimens were subjected to two-pass deformation at 1000 °C to investigate the kinetics of static recrystallisation. Here, the first deformation pass was performed at a strain below the critical strain for the onset of dynamic recrystallisation (i.e., 0.5), and then held for different times followed by the second deformation pass of 0.2 strain (Figure 1). The two flow curves were utilized to determine the softening fraction that resulted from the static recrystallisation at different holding times using Eq. [1].

$$Softening\, Fraction = \frac{{\left( {\sigma_{m} - \sigma_{2} } \right)}}{{\left( {\sigma_{m} - \sigma_{1} } \right)}}$$

(1)

where σ_m is the flow stress at the end of the first deformation, and σ₁ and σ₂ are the 0.2 pct offset yield stress of the first and the second deformations, respectively. Afterwards, some samples were subjected to deformation to a strain of 0.5 followed by water-quenching either immediately after straining or at a specific isothermal holding time to investigate the microstructural changes and texture development in martensite as a result of different post-deformation softening fractions (Figure 1).

Fig. 1

The schematic representation of different thermomechanical routes. Dashed line indicates the non-recrystallisation temperature

To examine the microstructure through the electron back-scatter diffraction technique (EBSD), the water-quenched (martensitic) samples were initially subjected to a tempering treatment at 450 °C for 2 h followed by air-cooling to decrease the dislocation density of martensite through recovery, thereby increasing the confidence index of EBSD mapping. Microstructures were analysed on the tangential section of the gauge length at a depth of ~ 100 μm beneath the surface. The samples for EBSD were prepared through a standard metallography polishing technique followed by a colloidal silica slurry polish. EBSD mapping was performed on a field emission gun Quanta 3D FEI scanning electron microscope, equipped with a fully automated EBSD device, using the working distance, voltage, and current of 10 mm, 20 kV and 4 nA, respectively. The EBSD maps were acquired using a step size of 0.1 μm with a hexagonal grid. Multiple EBSD maps were collected for every thermomechanically processed (TMP) sample, covering an area of ~ 75,000 μm² to obtain reliable measurement for the grain boundary network of martensite at different thermomechanical conditions. Due to the very low carbon content in the steel (~ 0.04 wt pct), it was assumed that the lath martensite has a body-centred cubic (BCC) crystal structure.^[16,18] The average confidence index was altered in the range of 0.60 to 0.70. The parent austenite microstructure was reconstructed from EBSD of phase transformed products (i.e., martensite)^[30] using TSL OIM V8 software, considering the K-S OR. The size of different microstructural features (SRX austenite grain and martensite block and packet sizes) produced for different TMP conditions were measured using the average of the mean linear intercepts measured in the horizontal and vertical directions.

3 Results

The stress-strain curve at 1000 °C displayed a distinct peak stress followed by softening, reaching a nearly steady-state plateau. This is a typical behaviour observed in material undergoing dynamic recrystallisation (Figure 2(a)). To estimate the onset of dynamic recrystallisation according to the method presented by Poliak and Jonas,^[29] the first derivative of stress-strain data (S = ∂σ/∂ɛ) known as the work hardening was plotted as a function of stress (Figure 2(b)). The inflection point in the work hardening curve, which is mathematically equivalent to a minimum value of the second derivative (∂²σ/∂ɛ² = ∂S/∂ɛ = 0, the inset curve in Figure 2(b)) corresponds to the initiation of dynamic recrystallisation. Therefore, the critical strain for the onset of dynamic recrystallisation was estimated to be ~ 0.5. The stress level progressively increased above the critical strain up to peak strain where the work hardening (i.e., dislocation multiplication) was greater than work softening (i.e., dynamic recrystallisation). After exceeding the peak strain, the flow stress continuously decreased, suggesting a progressive increase in the fraction of DRX (Figure 2(a)).

Fig. 2

(a) The flow curve of steel deformed at 1000 °C at a strain of 3 and strain rate of 1/s, and (b) the corresponding work hardening curve as a function of stress. Inset curve in (b) shows the second derivative of stress-strain data

3.1 Kinetics of Static Recrystallization

The first deformation flow stress strain curve at 1000 °C was similar (i.e., ε = 0.5) for all post-deformation holding times, revealing a pronounced power-law work-hardening behaviour (Figure 3(a)). This indicates that the experimental results were well duplicated, and the resultant microstructures and progress of softening were properly reflected for different holding times. At a very short post-deformation holding time of 0.1 s, the second flow curve slightly dropped, estimated at ~ 6 pct softening fraction using Eq. [1]. Therefore, the deformed austenite largely experienced static recovery rather than recrystallisation, as 15–20 pct softening is a minimum fraction for the occurrence of static recrystallisation using the offset approach.^[31] With an increase in the post-deformation holding time, the yield strength of the second deformation pass was progressively reduced, representing a further increase in the softening fraction, reaching 100 pct softening at ~ 20 s (i.e., fully recrystallisation state). The softening fraction data were used to estimate the constants in Johnson-Mehl-Avrami time dependence equation (Eq [2]).

$$x = 1 - \exp \left( { - kt^{n} } \right)$$

(2)

where x is the softening fraction due to the static recrystallisation, t is the post-deformation holding time, n is the Avrami exponent, and k is constant. The fraction of recrystallised grains or softening fraction can be estimated by computing the Avrami coefficients, n, and k, using Eq. [3], which is derived from Eq. [2].

$$\ln \ln \left( {\frac{1}{1 - x}} \right) = \ln K + n\ln t$$

(3)

Fig. 3

(a) The flow stress-strain curves during double hit deformations of steel at 1000 °C and a strain rate of 1 s⁻¹ for different post-deformation annealing times. (b) The softening fraction in the form of lnln(1/1-X) as a function of ln(t) at 1000 °C and a strain rate of 1 s⁻¹ for different post-deformation holding times

The current softening fraction data were used to plot $\ln {\mathbf{ln}} \left( {{1 \mathord{\left/ {\vphantom {1 {1 - x}}} \right. \kern-0pt} {1 - x}}} \right)$ as a function of ln t, where the slope of curve is equivalent to the n value and the corresponding Y-intercept represents the constant value of ln k (Figure 3(b)). Therefore, the softening behaviour of current steel as a function of holding time is described by Eq. [4].

$$x = 1 - \exp \left( { - 0.37 t^{0.72} } \right)$$

(4)

Based on this equation, the post-deformation holding times of 0.7, 2.3, 6.2, and 20 s were chosen to achieve 25, 50, 75, and 100 pct softening fractions, respectively. Therefore, the thermomechanical processing was conducted at 1000 °C to the strain of 0.5 and held isothermally at different times (i.e., 0.7, 2.3, 6.2, and 20 s) followed by water quenching. This led to a fully martensitic microstructure upon quenching for all thermomechanical conditions (Figures 4(a) through (d)).

Fig. 4

(a–d) The band contrast images of martensitic microstructures transformed from the austenite subjected to deformation at 1000 °C at a strain of 0.5 and isothermally held for different times followed by water quenching and (f–i) their corresponding reconstructed parent austenite microstructures using TSL OIM V8 software. The parent austenite grain boundaries delineated by black lines in (f through i). The triangle inset in (f) represents the colour codes referring to normal direction of each IPF map. Arrows in (a) refer to the shear direction and black arrows in (g and h) indicate the SRX grains. (a, f) 0 sec, (b, g) 0.7 sec, (c, h) 2.3 sec and (d, i) 20 sec

The reconstructed high temperature parent austenite microstructure displayed the process of the static recrystallisation, where the nucleation and growth of dislocation-free grains occurs, continuously replacing the deformed microstructure (Figures 4(f) through (i)). It is worth mentioning that the parent austenite reconstruction process used here provides the average grain orientation, disregarding the orientation gradient within the austenite grains, if any. Therefore, the SRX grains were differentiated by their size. At an early stage of the SRX process, the SRX grain size was relatively fine; measured 6 ± 1 μm at 25 pct softening fraction (as shown by white arrows in Figure 4(g)). As the static recrystallisation proceeded, the SRX grains progressively grew, further consuming the deformed microstructure and enhancing the fraction of softening. The average SRX grain size became 18 ± 1.5 μm at 50 pct softening and further coarsened to 33 ± 2 μm and 40 ± 2 μm at 75 and 100 pct softening fractions, respectively (Figures 4(f) through (i)).

The post-deformation annealing times (i.e., extent of static recrystallisation) significantly influenced the characteristics of the martensite texture (Figure 5). The (110) pole figure of martensite transformed from deformed austenite, in the absence of static recrystallisation, manifested a relatively strong overall texture with a strength of ~ 3.5 multiples of the random distribution (MRD, Figure 5(a)). It consisted of multiple peaks corresponding to the simple shear components for body centred cubic metals.^[32] They are mostly located at the positions of D₁, D₂ and E, $\overline{E}$ components along the $\left\{ {hkl} \right\}\left\langle {111} \right\rangle$ fibre, while the F component at $\left\{ {110} \right\}\left\langle {001} \right\rangle$, and J₁ and J₂ components situated on the $\left\{ {110} \right\}\left\langle {uvw} \right\rangle$ fibre had much weaker intensity (Figure 5(a) and (f). The occurrence of static recrystallisation led to a progressive increase in the number of peaks and decrease in the intensity of the overall transformation texture in the martensite microstructure, reaching 2.4 MRD when a fully recrystallised microstructure was achieved (Figure 5).

Fig. 5

(110) pole figures of martensite transformed from high temperature austenite deformed to a strain of 0.3 followed by holding at different times at 1000 °C: (a) 0 s, (b) 0.7 s, (c) 2.3 s, (d) 6.2 s and (e) 20 s. (f) schematic representation of $\left( {110} \right)$ pole figure displaying ideal orientations associated with simple shear deformation of BCC materials and their corresponding $\left\{ {hkl} \right\}\left\langle {uvw} \right\rangle$, reprinted with permission from Ref. [32]

3.2 Effect of SRX on Intervariant Boundary Network Characteristics

The lath martensite misorientation angle distribution exhibited a bimodal distribution revealing a weak peak in the range of 10–22 deg and one strong peak in the 47–60 deg range for all thermomechanical conditions (Figure 6). This is consistent with those expected from the intervariant boundaries formed during martensitic transformation based on the K-S orientation relationship, where theoretically up to 24 variants form from a given parent austenite leading to 16 unique intervariant boundaries because of crystal symmetry (Table I).^[18] However, boundaries in the range of 23–46 deg misorientation angle do not correspond to those expected from the KS OR. Therefore, they can be formed due to the impingement of variants from distinct parent austenite grains at the parent austenite grain boundaries, hereafter called non-KS boundaries (Figure 6).

Fig. 6

The misorientation angle distribution of as-quenched lath martensite obtained after deforming at 1000 °C to a strain of 0.5 and holding at different post-deformation holding times, followed by water-quenching: (a) 0 s (0 pct SRX), (b) 0.7 s (25 pct SRX), (c) 2.3 s (50 pct SRX), (d) 6.2 s (75 pct SRX) and (e) 20 s (100 pct SRX). They largely represent non K-S boundaries that originated from the prior austenite grain boundaries

Table I Possible 24 Variants and Intervariant Boundaries Generated Through Phase Transformation Having the KS Orientation Relationship

The extent of static recrystallisation appeared to influence the lath martensite misorientation angle distribution. The change was largely observed in the range 47–60 deg, where the population gradually reduced with the SRX fraction up to 50 pct softening, beyond which it gradually increased till 100 pct SRX (Figure 6). For instance, the population at 60 deg was ~ 10 pct in the as-deformed condition (Figure 6(a)), reducing to ~ 6 pct at 50 pct SRX (Figure 6(c)) followed by an increase to ~ 12 pct at fully recrystallised specimen (Figure 6(e)). This was inversely reflected in the population of boundaries in the 23–46 deg misorientation angle range (i.e., non-KS boundaries), revealing the highest population at 50 pct SRX (Figure 6).

To evaluate the lath martensite intervariant boundary network through static recrystallisation, the fraction of 16 intervariant boundaries (Table I) was plotted at different SRX conditions in Figure 7(a). For the as-deformed condition (0 pct SRX), the highest intervariant length fraction was related to 60 deg/$\left[ {011} \right]$ with 56 pct, followed by 10.53 deg/$\left[ {0\overline{1}\overline{1}} \right]$, 60 deg/$\left[ {11\overline{1}} \right]$ and 57.2 deg/$\left[ {\overline{6}\overline{2}5} \right]$ having ~ 8, 6, and 6 pct, respectively (Figure 7(a)). Interestingly, the first three intervariant boundaries with the highest population related to the impingement of laths with the same close-packed habit plane (i.e., V1-V3/V5, V1-V2, and V1-V4, Table I). The fraction of intervariant boundaries was influenced by increasing the post-deformation annealing time where the deformed matrix was consumed by the SRX grains. However, there was no specific trend between the population of intervariant boundaries and the extent of static recrystallisation.

Fig. 7

The fraction of total population (a) and length per area (b) of intervariant boundaries based on the KS OR, comparing intervariant interfaces between V₁ and V_i (i = 2–24), for the martensite microstructures transformed from austenite at different static recrystallisation states. The mean values and standard deviations were determined from the values measured from at least five different orientation maps

The 60 deg/$\left[ {011} \right]$ intervariant boundaries displayed the most apparent change in their population compared to other intervariant boundaries. The population of 60 deg/$\left[ {011} \right]$ intervariant boundaries initially decreased from 56 pct for the deformed condition (0 pct SRX) to 40 pct after 50 pct SRX. Then, its population progressively increased to 58 pct for the fully recrystallised condition (Figure 7(a)). This trend was similar for other intervariant boundaries having a twist character (i.e., 10.53 deg/$\left[ {0\overline{1}\overline{1}} \right]$ and 49.47 deg/$\left[ {110} \right]$). On the other hand, some of the intervariant boundaries revealed a slight increase in their population up to 50 pct SRX followed by a decrease with further recrystallisation (e.g., 10.53 deg/$\left[ {11\overline{1}} \right]$, 14.88 deg/$\left[ {13 5 1} \right]$, 57.2 deg/$\left[ {\overline{6}\overline{2}5} \right]$, and 60 deg/$\left[ {11\overline{1}} \right]$, Figure 7(a)). In fact, this trend was dominant for boundaries with a tilt character. A similar observation was made for the length of various intervariant boundary types per unit area (i.e., µm/µm²) at different post-deformation holding times (Figure 7(b)). Overall, the length per unit area of lath martensite intervariant boundaries with the KS OR and the boundaries having twist character continuously decreased up to 50 pct SRX, beyond which further recrystallisation increased their length per unit area. However, the length per unit area of the intervariant boundaries with symmetrical tilt character and non-KS boundaries enhanced with an increase in the recrystallisation fraction up to 50 pct. Further static recrystallisation led to a decrease in their length per unit area (Figure 8).

Fig. 8

Total length per unit area of the martensite interfaces having KS OR, twist character with [110] axis, tilt character with [111] axis, and non-KS OR for martensite transformed from austenite at different static recrystallisation states

It should be emphasized here that the martensitic transformation may locally follow other crystallographic orientation relationship, such as the Nishiyama-Wasserman (N-W) OR.^[18] In the case of the N-W OR, the impingement of different martensitic variants formed within a given parent austenite grain leads to five distinct boundaries, consisting of 60 deg/[110], 53.7 deg/[1 3 3], 50 deg/[3 2 3], 19.5 deg/[100] and 13.8 deg/[1 1 0].^[33] Among these lattice misorientations, 60 deg/[110] is common with those observed in the K-S OR. In addition, the lattice misorientation of 13.8 deg/[110] is within 5 deg deviation angle considered in the current study from 10.5 deg/[110] in the K-S OR (Table I). Therefore, three boundaries of N-W OR (i.e., 53.7 deg/[1 3 3], 50 deg/[3 2 3], 19.5 deg/[100]) were counted as non-KS boundaries in the current study.

As mentioned earlier, each parent austenite grain is divided by distinct intervariant boundaries (see Table I). These boundaries can be employed to identify different microstructure constituents, namely packets and blocks. The block boundaries within a given packet can be defined based on the intervariant boundaries resulting from the impingement of two distinct variants from the same family (i.e., V₁-V₂ to V₁-V₆ in Table I). However, each packet with a given parent austenite grain consists of a set of blocks, which do not share the same habit plane with block/s within neighbouring packet/s. Therefore, the packet boundaries can be differentiated through the intervariant boundaries resulting from the impingement of blocks/variants with distinct habit planes (i.e., from different families, e.g., V₁-V₇ to V₁-V₂₄ in Table I). Therefore, the blocks were delineated considering all boundaries having greater than 5 misorientation angles, i.e., prior austenite grain boundaries, along with all intervariant boundaries presented in Table I. The packets outlined all of the boundary network, excluding the lattice misorientations associated with the block boundaries. Based on this classification, it appears that the post-deformation annealing time significantly influenced the block and packet sizes during the martensite transformation. They both progressively decreased with the extent of static recrystallisation up to 50 pct, beyond which they continuously increased reaching the size of ~ 15.1 ± 0.5 μm and ~ 26.2 ± 0.4 μm, respectively, for the fully SRX state (Figure 9).

Fig. 9

The effect of softening fraction on the size of the packet and block of martensite

4 Discussion

The current thermomechanical processing resulted in a fully martensitic microstructure for different conditions (Figure 4). However, the extent of parent austenite SRX during post-deformation annealing appears to markedly change the characteristics of the intervariant boundaries network in lath martensite transformed upon quenching (Figures 6, 7, and 8). As discussed earlier, the lath martensite transformation reveals a bimodal misorientation angle distribution in the range of 10–22 deg and 47–60 deg misorientation angle (Figure 6), which closely matches the misorientations expected from the ideal KS orientation relationship (Table I). The presence of small fraction of boundaries with 23–46 deg misorientation angles results from the impingement of variants on both sides of deformed and/or dislocation-free SRX austenite parent grains. The parent austenite SRX process significantly alters the population of different intervariant boundaries, more specifically the 60 deg/[110] misorientation and non-KS boundaries, during the post-deformation annealing (Figures 6 and 7).

The grain boundary network is typically influenced through multiple parameters, namely chemical composition,^[23] transformation mechanism,^{[17,18,20,22]} crystallography texture,^[15,34] thermomechanical route (i.e., temperature, strain, strain rate, post-deformation interval time),^{[15,16,20,31]} and grain size.^[4,35] The role of steel composition and phase transformation mechanism is excluded here, as the composition is identical, and all the samples undergo a martensitic transformation.

The influence of crystallographic texture is manifested through the variant selection mechanism upon transformation, resulting in a change in the boundary network. However, the austenite is fully transformed to martensite upon quenching, which makes it difficult to directly evaluate the parent austenite characteristics (texture, dislocation substructure, grain size) and its influence on the variant selection. Here, the EBSD of martensite was used to back-calculate the high temperature parent austenite orientation map,^[30] considering the KS OR for different thermomechanical conditions, as shown in Figure 10. For the (111) pole figure of reconstructed parent austenite at the deformed condition (i.e., 0 s holding time), the overall texture was relatively strong, consisting of the A, $\overline{A}$ texture components as well as the B, $\overline{B}$ and C components, which are typical FCC shear texture^[37] (Figures 10(a) and (i)). In general, the (111) pole figures of reconstructed parent austenite were well correlated with the (110) pole figures observed for martensite at the different thermomechanical processing conditions (comparing Figures 5 and 10(a) through (d)). In other words, the martensite overall texture is inherited from the parent austenite that has been subjected to shear deformation at high temperature. Similarly, the overall texture intensity of parent austenite reduces with an increase in the extent of recrystallisation, from 6.4 MRD at 0 pct softening to 3.2 MRD at 100 pct softening (Figures 10(a) through (d)). The reconstructed parent austenite microstructure was used to estimate the theoretical overall texture of the transformed microstructure in the absence of variant selection. Here, it is assumed that each parent austenite grain transforms into all possible 24 distinct BCC variants in an equal fraction based on the KS OR, as listed in Table I. As expected from the crystallography of phase transformation, the characteristics of (111) overall texture of parent austenite closely matched with the (110) pole figure of transformed martensite at different conditions (Figures 5 and 10(e) through (h)). However, the overall texture intensity for the theoretical transformed condition is significantly lower compared with the ones measured in the transformed martensite for a given TMP condition (Figures 5 and 10(e) through (h)). This discrepancy suggests that the variant selection, to some extent, occurs during the martensitic transformation, but the degree of variant selection depends on the progress of SRX in the deformed parent austenite microstructure. However, it is not easy to define the mechanism of variant selection and how it influences the population of different boundaries during SRX process. This will be discussed later based on the change in the austenite grain characteristics (size and deformed state).

Fig. 10

(a–d) (111) pole figures of high temperature parent austenite, back-calculated from the martensite orientation in Fig. 5 considering the KS OR, at different thermomechanical conditions. (e–h) the corresponding (110) pole figures of transformed BCC structure, forward-calculated using the parent austenite orientation in (a) through (d) using the KS OR. (a, e) 0 s, (b, f) 0.7 s, (c, g) 6.2 s and (d, h) 20 s. The pole figures were plotted using ATEX software.^[36] (i) schematic representation of $\left( {111} \right)$ pole figure displaying ideal orientations associated with simple shear deformation of FCC materials and their corresponding $\left\{ {hkl} \right\}\left\langle {uvw} \right\rangle$, reprinted with permission from Ref. [37]

Based on the thermomechanical route shown in Figure 1, the deformation temperature, strain, and strain rate are similar for all TMP conditions. Therefore, they do not contribute to the changes observed in the martensite intervariant boundary network. However, the post-deformation annealing time varies for different TMP conditions, affecting the extent of softening fraction of austenite prior to the martensitic transformation. During the post-deformation annealing, two restoration processes concurrently occur in the deformed microstructure, namely dislocation annihilation (recovery) and recrystallisation, resulting in the softening of high temperature parent austenite. These microstructural changes may affect the intervariant boundary network differently. The recovery process takes place within the deformed region, progressively disintegrating the dislocation walls and reducing the overall dislocation density.^[38] The presence of a dislocation substructure within the parent austenite is known to enhance the formation of specific variants upon the martensitic transformation, leading to the promotion of 60 deg/$\left[ {11\overline{1}} \right]$, 10.53 deg/$\left[ {0\overline{1}\overline{1}} \right]$, and 49.5 deg/$\left[ {110} \right]$ intervariant boundaries at the expense of 60 deg /$\left[ {011} \right]$.^[16,26] Therefore, the progressive removal of dislocations with annealing time is, to some extent, expected to reveal an opposite trend (e.g., increasing the population of 60 deg/$\left[ {011} \right]$ boundary). However, the current result exhibits a reduction in the population of 60 deg/$\left[ {011} \right]$ boundary with the softening fraction up to 50 pct (Figure 7(a)), where the most recovery is anticipated to take place within the deformed region. This suggests that the dislocation recovery is not the main contributing factor to the intervariant boundary network.

Upon post-deformation annealing, SRX takes place through the nucleation and growth of dislocation-free grains within the deformed microstructure. The nucleation of SRX grains predominantly occurs at regions with high dislocation densities, such as triple junctions and grain boundaries (see white arrows in Figures 4(g) through (h)).^[31] The SRX nuclei grow within the deformed regions/grains through the removal of the dislocation substructure in front of their moving boundaries (Figures 4(f) through (i)).^[31] This ultimately leads to a progressive softening of material as the SRX process continuously replaces the deformed microstructure with dislocation-free grains (Figures 3 and 4).

At an early stage of static recrystallisation, the presence of a high fraction of nucleation sites leads to the formation of small SRX grains (Figure 4(g)). The refinement of the parent austenite grain size decreases the martensite transformation temperature, M_s, as it restricts both the number of variants nucleated within a parent austenite grain and their growth during the martensitic transformation.^[4,39,40,41] This, in turn, leads to the change in the lattice parameters of parent austenite and martensite in which the martensitic transformation takes place for distinct parent austenite grain size.^[4] The phenomenological theory of martensite crystallography demonstrated that the variant selection mechanism alters from the 3-variant clustering (i.e., V₁V₃V₅) in martensite transformed from a coarse parent grain to the 4-variant clustering arrangement (i.e., V₁V₂V₃V₄) in fine austenite parent grain, due to the change in the lattice parameter of the parent and daughter phase in which the martensite transformation occurs, to accommodate the strain related to the martensite transformation. This, in turn, promotes the fraction of 60 deg/$\left[ {11\overline{1}} \right]$, reducing the 60 deg/$\left[ {011} \right]$ intervariant boundaries population.^[4] This is similar to the current result, where the 60 deg/$\left[ {11\overline{1}} \right]$, 10.5 deg/$\left[ {11\overline{1}} \right]$, 14.9 deg/$\left[ {13 5 1} \right]$, 50.5 deg/$\left[ {\overline{7} \overline{5} 5} \right]$ boundaries progressively increase, while the deformed grains are largely replaced by small SRX grains (up to 50 pct softening, where the SRX kinetics is controlled by nucleation) at the expense of 60 deg/$\left[ {011} \right]$ intervariant boundaries (Figure 7(a)). A higher fraction of softening largely occurs because of progressive growth of SRX grains either through consumption of the deformed region and/or their coalescence (i.e., SRX kinetics is controlled by grain growth), leading to the increase in the average of SRX grain size (Figure 4). Therefore, the 3-variant clustering arrangement reveals lower self-strain accommodation predominately formed during the growth of dislocation-free grains promoting the 60 deg/$\left[ {011} \right]$ intervariant boundaries. As a result, the fraction of 60 deg/$\left[ {011} \right]$ intervariant boundaries is expected to progressively increase with the softening fraction beyond 50 pct, as the SRX grain size enhances, which is aligned well with the current observation (Figure 7). The population of non-KS boundaries is also directly related to the SRX grain size. At an early stage of SRX, the average grain size of SRX grains is small, which increases the parent austenite grain boundaries, ultimately enhancing the fraction of non-KS boundaries. The growth/coalescence of SRX grains (i.e., size increment) leads to the reduction of non-KS boundaries population due to a decrease in the parent austenite grain boundary area/fraction (Figures 6, 7, and 8).

5 Conclusions

The influence of static recrystallisation of parent austenite on the grain boundary network of the lath martensite formed upon transformation was examined using interrupted torsion tests. The following conclusions were drawn based on the experimental results:

• The martensite grain boundary network was significantly affected by the extent of SRX fraction. The change in grain boundary network was largely controlled by the size of SRX grains, rather than the extent of dislocation recovery.

• The formation of small dislocation-free grains at an early stage of SRX up to 50 pct softening led to a significant reduction in the population of 60 deg/$\left[ {011} \right]$, due to the promotion of 4-variant (V₁V₂V₃V₄) clustering arrangement to minimize the strain associated with the martensite transformation. This, in turn, promoted the formation of 60 deg/$\left[ {11\overline{1}} \right]$ intervariant boundary.

• Progressive growth of SRX grains either through consumption of the deformed region and/or their coalescence beyond 50 pct softening altered the variant selection mechanism to a 3-variant (V₁V₃V₅) clustering arrangement, leading to the enhancement of 60 deg/$\left[ {011} \right]$ intervariant boundary population.

• The length per unit area (μm/μm²) of overall intervariant boundaries and twist boundaries progressively decreased up to 50 pct SRX, due to a relatively high-volume fraction of small dislocation-free grains. A higher fraction of softening increased the length per unit area (μm/μm²) of the overall intervariant boundaries and twist boundaries due to the growth of SRX grains. However, an opposite trend was found for both symmetrical tilt and non-KS boundaries.\