The Effect of Initial Grain Size on the Nanocrystallization of AZ31 Mg Alloy during Rotary Swaging

Nanograins were obtained in the AZ31 Mg alloy bars with different initial grain sizes via cold rotary swaging. Microstructure evolution during deformation was investigated through electron backscatter diffraction analysis and transmission electron microscopy studies. The results indicate that initial grain size had little effect on the mechanism of grain refinement during swaging. The nanocrystallization process of the alloys with different initial grain sizes included extensive twinning followed by the further refinement of the twin lamellae through the formation of massive dislocation arrays. However, as the initial grain size decreased, the formation rate of nanograins increased, resulting in a higher degree of nanocrystallization after the same swaging pass. The mean grain size and yield strength of the sample with the smallest initial grain size were about 91 nm and 489 MPa, respectively. The slower rate and lower degree of nanocrystallization in the alloy with a larger initial grain size were mainly attributed to the less grain boundary areas and higher activity of twinning.


Introduction
Grain refinement is one of the most efficient methods to improve the mechanical properties of metal materials [1]. It is now well-established that significant grain refinement can be obtained through severe plastic deformation techniques, such as high-ratio differential speed rolling [2], equal channel angular pressing [3], and high-pressure torsion [4]. Generally, changing the strain path [5], reducing deformation temperature [6], and increasing strain rate [7] can bring about a better effect of grain refinement in most metal materials. Moreover, the microstructure in the initial billet can also influence the grain refinement process during deformation, through different factors such as the initial grain size [8], artificially formed precipitates [9], shear bands [10], and numerous twins [11] induced prior to the deformation.
Mg alloys, as the lightest metallic structural materials for application in automobiles and aircraft, can achieve a better strengthening effect than steel and Al alloys through grain refinement due to their higher Hall-Petch slope value [12]. In most Mg alloys, the change in the initial grain size can lead to the transformation of slip and twinning behaviors [13][14][15][16][17][18][19][20]. Koike [13] and Agnew [14] indicated small-grained AZ31 Mg alloy could induce the activation of more non-basal slip dislocations by the plastic compatibility stress associated with grain boundaries. Therefore, the non-basal slip dislocation is enhanced by finer initial grain sizes [15] and is even active in the grain interior [16]. Compared with slip dislocation, twinning is more sensitive to the initial grain size [17]. The activity of twinning generally increases with increasing initial grain size, which is attributed to easier interaction with lattice dislocations that "promote" twin growth in larger grains [17,18]. The number of twins per grain is found to increase with increasing grain size [19], while the twin thickness is reported to be independent of the initial grain size [20].
Recently, we successfully prepared bulk nanocrystalline Mg alloys via cold rotary swaging [21]. However, the effect of initial grain sizes on the mechanism of grain refinement during rotary swaging is still unclear. Additionally, the possibility of nanocrystallization by using the initial bar billets with a much larger grain size (≥500 µm) is unknown. In the present work, AZ31 Mg alloy, one of the currently most widely used commercial wrought Mg alloys, was selected as our model material. The microstructure evolution in samples with different initial grain sizes and their effects on the formation of nanograins (NGs) during rotary swaging were studied in detail.

Sample Preparation
The alloy ingots with a diameter of 90 mm were prepared by melting pure Mg, pure Al, pure Zn, and Al-10 wt.% Mn master alloy under a CO 2 and SF 6 atmosphere and casting in an upright semi-continuous casting machine. Homogenization was conducted in an air furnace at 693 K for 12 h, followed by air cooling. Then, a cylindrical-shaped sample with a height of 500 mm and a diameter of 18 mm was cut from the ingot for further rotary swaging and this bar was denoted as "HS". The alloy composition of the homogenized billet was measured by an inductively coupled plasma spectrometer and is listed in Table 1. Two cylindrical-shaped samples with a height of 250 mm and a diameter of 90 mm were cut from the ingot for extrusion. The extrusion was performed using a backward extrusion method under a ram speed of 10 mm/s. The first sample was processed with an extrusion temperature of 623 K and an extrusion ratio of 25. A cylindrical bar with a diameter of 18 mm was obtained and denoted as "E25S". The second sample was processed with an extrusion temperature of 723 K and an extrusion ratio of 9. A cylindrical bar with a diameter of 30 mm was obtained. Then, its diameter was reduced to 18 mm using a lathe machine, and this bar was denoted as "E9S".
The HS, E9S, and E25S samples were further performed via multi-pass rotary swaging under the same process at room temperature. Figure 1 shows the schematic representation of rotary swaging. The accumulated deformation degree after each pass is calculated using Equation (1), where S 0 is the initial and S 1 is the cross-section after swaging, and the values are shown in Table 2. Compared with that for the E9S and E25S samples, the utmost accumulated deformation degree for the HS sample was only 0.14. Further increasing deformation would lead to the failure of the HS sample, probably due to more casting defects existing in the homogenized sample.

Microstructural Characterization and Mechanical Property Evaluation
Vickers microhardness tests were conducted using an HMV-G 21DT (Shimad kyo, Japan) tester with a load of 4.9 N and a dwell time of 15 s. The interval of the a measurement point was 0.5 mm. Dog-bone-shaped tensile samples with a gauge of 15 mm and a diameter of 3 mm were prepared from the center of the alloy ba tensile tests were performed using an Instron 3369 machine at an initial strain rat 10 −3 s −1 at room temperature, with the tensile directions parallel to the feed directio The samples for microstructural observation were all cut from the central re the alloy bars and the observation plane was on the cross-section. Electron bac diffraction (EBSD) measurements were conducted using an Helios Nanolab 60 Hillsboro, OR, USA) scanning electron microscope (SEM) equipped with the HKL nel 5 data acquisition and analysis software. Transmission electron microscopy (TE high-resolution TEM (HRTEM) observations were carried out on a Tecnai G 2 F Brno, Czech) microscope with an accelerating voltage of 200 kV using a double-ti men stage. The cut disks were first mechanically ground to less than 100 μm in th then thinned by twin-jet polishing to an electron-transparent thickness in an ele containing 1 vol.% nitric acid, 2 vol.% perchloric acid, and 97 vol.% ethanol. The po temperature was approximately −40 °C. The average grain size was measured w Nano Measurer software 1.2. At least 500 grains were measured for the statistical a

Microstructural Characterization and Mechanical Property Evaluation
Vickers microhardness tests were conducted using an HMV-G 21DT (Shimadzu, Tokyo, Japan) tester with a load of 4.9 N and a dwell time of 15 s. The interval of the adjacent measurement point was 0.5 mm. Dog-bone-shaped tensile samples with a gauge length of 15 mm and a diameter of 3 mm were prepared from the center of the alloy bars. The tensile tests were performed using an Instron 3369 machine at an initial strain rate of 1 × 10 −3 s −1 at room temperature, with the tensile directions parallel to the feed direction.
The samples for microstructural observation were all cut from the central region of the alloy bars and the observation plane was on the cross-section. Electron backscatter diffraction (EBSD) measurements were conducted using an Helios Nanolab 600i (FEI, Hillsboro, OR, USA) scanning electron microscope (SEM) equipped with the HKL Channel 5 data acquisition and analysis software. Transmission electron microscopy (TEM) and high-resolution TEM (HRTEM) observations were carried out on a Tecnai G 2 F20 (FEI, Brno, Czech) microscope with an accelerating voltage of 200 kV using a double-tilt specimen stage. The cut disks were first mechanically ground to less than 100 µm in thickness, then thinned by twin-jet polishing to an electron-transparent thickness in an electrolyte containing 1 vol.% nitric acid, 2 vol.% perchloric acid, and 97 vol.% ethanol. The polishing temperature was approximately −40 • C. The average grain size was measured with the Nano Measurer software 1.2. At least 500 grains were measured for the statistical analysis of the grain size. Figure 2 shows the microstructure of the as-homogenized and as-extruded samples. The mean grain sizes of the HS-0 sample, the E9S-0 sample, and the E25S-0 sample were about 548 ± 122 µm, 71 ± 27 µm, and 18 ± 5 µm, respectively.  Figure 2 shows the microstructure of the as-homogenized and as-extruded samples. The mean grain sizes of the HS-0 sample, the E9S-0 sample, and the E25S-0 sample were about 548 ± 122 μm, 71 ± 27 μm, and 18 ± 5 μm, respectively.  Figure 3 shows the comparison of microstructure between the HS-1 sample, the E9S-1 sample, and the E25S-1 sample. In the HS-1 sample, different types of twins were observed, including the {101 1}-{101 2} double twins (~1.5%), {101 1} contraction twins (~2.9%), and {101 2} tension twins (~23.7%). These twins refined the initial coarse grains into a thin lamellar structure. In comparison, the double twins and contraction twins were hardly observed in the E9S-1 sample. The main type of twins formed in the E9S-1 sample was the {101 2} tension twins (~24.1%). In the E25S-1 sample, the number of twins observed was much smaller than that found in HS-1 and E9S-1 samples, as shown in Figure 3h.  Figure 3 shows the comparison of microstructure between the HS-1 sample, the E9S-1 sample, and the E25S-1 sample. In the HS-1 sample, different types of twins were observed, including the {1011}-{1012} double twins (~1.5%), {1011} contraction twins (~2.9%), and {1012} tension twins (~23.7%). These twins refined the initial coarse grains into a thin lamellar structure. In comparison, the double twins and contraction twins were hardly observed in the E9S-1 sample. The main type of twins formed in the E9S-1 sample was the {1012} tension twins (~24.1%). In the E25S-1 sample, the number of twins observed was much smaller than that found in HS-1 and E9S-1 samples, as shown in Figure 3h.   Figure 4 shows the typical microstructure in the central region of HS-1, E9S-1, and E25S-1 samples. After one-pass swaging, many lamellar structures with lamellar widths ranging from 100 nm to 1 μm were formed in the samples. After randomly selecting ten lamellae for the selected area electron diffraction (SAED) analysis, most of these lamellae were determined to be {101 2} tension twins (indicated as "T") with misorientations of about 86° at the interface, as shown in the inset SAED patterns in Figure 4a,d,f. In the HS- (d) EBSD IPF image of the E9S-1 sample; (e) corresponding BC map of (d); (f) corresponding misorientation angle distribution map of (e); (g) EBSD IPF image of the E25S-1 sample; (h) corresponding BC map of (g). Figure 4 shows the typical microstructure in the central region of HS-1, E9S-1, and E25S-1 samples. After one-pass swaging, many lamellar structures with lamellar widths ranging from 100 nm to 1 µm were formed in the samples. After randomly selecting ten lamellae for the selected area electron diffraction (SAED) analysis, most of these lamellae were determined to be {1012} tension twins (indicated as "T") with misorientations of about 86 • at the interface, as shown in the inset SAED patterns in Figure 4a,d,f. In the HS-1 sample, two fine twins (T H2 and T H3 ) with a width of about 10 nm could be observed within the twin lamella T H1 , as shown in Figure 4a,b. The misorientation of the boundary between the two fine twins and T H1 was about 88 • , as shown in Figure 4c, which indicates that T H2 and T H3 are {1012} tension twins too. This means that {1012}-{1012} double twinning could occur in the HS-1 sample, which was not observed in E9S-1 or E25S-1 samples. In the E9S-1 sample, many twin-twin intersections were formed, such as T E1 and T E2 shown in Figure 4d. Viewed from a single <1012> axis, both T E1 and T E2 could be well imaged, as shown in Figure 4e, indicating that T E1 and T E2 might correspond to (1012) and (1012) tension twin variants [22]. Compared with HS-1 and E9S-1 samples, most of the twin morphology observed in the E25S-1 sample was nearly parallel, as shown in Figure 4f.

After Three-Pass Swaging
After three-pass swaging, the three groups of samples exhibited different degrees of nanocrystallization. Figure 5a shows that only a small number of nanoscale subgrains (NSGs) were formed in the local region of the HS-3 sample. The proportion of NSGs obviously increased in the E9S-3 sample, as shown in Figure 5b. In the E25S-3 sample, massive NSGs were formed, as shown in Figure 5c. This indicates that a higher degree of refinement could be obtained at the sample with a smaller initial grain size.

After Three-Pass Swaging
After three-pass swaging, the three groups of samples exhibited different degrees of nanocrystallization. Figure 5a shows that only a small number of nanoscale subgrains (NSGs) were formed in the local region of the HS-3 sample. The proportion of NSGs obviously increased in the E9S-3 sample, as shown in Figure 5b. In the E25S-3 sample, massive NSGs were formed, as shown in Figure 5c. This indicates that a higher degree of refinement could be obtained at the sample with a smaller initial grain size.

After Three-Pass Swaging
After three-pass swaging, the three groups of samples exhibited different degrees of nanocrystallization. Figure 5a shows that only a small number of nanoscale subgrains (NSGs) were formed in the local region of the HS-3 sample. The proportion of NSGs obviously increased in the E9S-3 sample, as shown in Figure 5b. In the E25S-3 sample, massive NSGs were formed, as shown in Figure 5c. This indicates that a higher degree of refinement could be obtained at the sample with a smaller initial grain size.     [23,24]. The deviation of the angle from the common 86 • was considered to be a result of the accommodation of local strain after twinning [23,24]. Moreover, NSG1 and NSG2 both had high-angle grain boundaries, namely B 2 (~11 • ) in NSG1 and B 4 (~41 • ), B 5 (~29 • ), and B 6 (~17 • ) in NSG2. This indicates that the NSGs formed in the sample with different initial grain sizes showed similar features. deviation of the angle from the common 86° was considered to be a result of the accommodation of local strain after twinning [23,24]. Moreover, NSG1 and NSG2 both had highangle grain boundaries, namely B2 (~11°) in NSG1 and B4 (~41°), B5 (~29°), and B6 (~17°) in NSG2. This indicates that the NSGs formed in the sample with different initial grain sizes showed similar features.  Figure 7a shows that little difference in contrast was observed between the neighboring grains in the E9S-5 sample. This is probably due to the existence of many low-angle grain boundaries, making it difficult to obtain a convincing statistic of the mean grain size. The unsharp boundaries and relatively discontinuous ring in the inset SAED pattern fur-   Figure 7a shows that little difference in contrast was observed between the neighboring grains in the E9S-5 sample. This is probably due to the existence of many low-angle grain boundaries, making it difficult to obtain a convincing statistic of the mean grain size. The unsharp boundaries and relatively discontinuous ring in the inset SAED pattern further indicate that many NSGs in the E9S-5 sample did not completely transform into NGs. In contrast, the microstructure of the E25S-5 sample showed well-defined and clean boundaries with uniform contrast at the grain interiors. The mean grain size was 91 ± 4 nm. The continuous diffraction ring in the inset SAED pattern indicates that the NGs formed in the E25S-5 sample were randomly oriented. Figure 7 indicates that the sample with a smaller initial grain size showed a higher degree of nanocrystallization after five-pass swaging.  Figure 8 shows the microhardness distributions along the radial direction of the swaged alloy bars after different swaging passes. The three groups of samples with different initial grain sizes all showed significant improvement in microhardness after swaging. As the swaging passes increased, a gradient distribution of microhardness along the radial direction of these samples could be observed, which showed a higher hardness value at the center than at the edge. In comparison, the HS sample only showed a slightly gradient distribution of microhardness after three-pass swaging. The E9S sample exhibited an obvious gradient distribution of microhardness after three-pass swaging, while the E25S sample showed a remarkable gradient distribution of microhardness after twopass swaging. More intensive hardening can be observed at the center of the E25S sample, compared with the E9S and HS samples at the early stage of swaging, as shown in Figure  9.   Figure 8 shows the microhardness distributions along the radial direction of the swaged alloy bars after different swaging passes. The three groups of samples with different initial grain sizes all showed significant improvement in microhardness after swaging. As the swaging passes increased, a gradient distribution of microhardness along the radial direction of these samples could be observed, which showed a higher hardness value at the center than at the edge. In comparison, the HS sample only showed a slightly gradient distribution of microhardness after three-pass swaging. The E9S sample exhibited an obvious gradient distribution of microhardness after three-pass swaging, while the E25S sample showed a remarkable gradient distribution of microhardness after two-pass swaging. More intensive hardening can be observed at the center of the E25S sample, compared with the E9S and HS samples at the early stage of swaging, as shown in Figure 9. gradient distribution of microhardness after three-pass swaging. The E9S sample exh ited an obvious gradient distribution of microhardness after three-pass swaging, wh the E25S sample showed a remarkable gradient distribution of microhardness after tw pass swaging. More intensive hardening can be observed at the center of the E25S samp compared with the E9S and HS samples at the early stage of swaging, as shown in Figu 9.   Figure 10 shows the stress-strain curves of the homogen samples in tension at room temperature. The correspondin listed in Table 3. The yield strength of the HS sample increase than 333 MPa after three-pass swaging. For the extruded sa E9S and E25S samples increased from 186 ± 8 MPa and 202 ± 489 ± 9 MPa, respectively, after five-pass swaging. Three gr significant improvement in strength after rotary swaging. It is E9S-5 samples fractured before reaching the ultimate tensile s sile curves. The former was attributed to the failure to elimi hot extrusion, while the early fracture of the E9S-5 sample wa of incomplete refinement after five-pass swaging.  Figure 10 shows the stress-strain curves of the homogenized, extruded, and swaged samples in tension at room temperature. The corresponding mechanical properties are listed in Table 3. The yield strength of the HS sample increased from 155 ± 21 MPa to more than 333 MPa after three-pass swaging. For the extruded samples, the yield strength of E9S and E25S samples increased from 186 ± 8 MPa and 202 ± 3 MPa to 460 ± 19 MPa and 489 ± 9 MPa, respectively, after five-pass swaging. Three groups of samples all showed significant improvement in strength after rotary swaging. It is worth noting that HS-3 and E9S-5 samples fractured before reaching the ultimate tensile strength according to the tensile curves. The former was attributed to the failure to eliminate casting defects through hot extrusion, while the early fracture of the E9S-5 sample was considered to be the result of incomplete refinement after five-pass swaging. 489 ± 9 MPa, respectively, after five-pass swaging. Three groups of samples all s significant improvement in strength after rotary swaging. It is worth noting that H E9S-5 samples fractured before reaching the ultimate tensile strength according to sile curves. The former was attributed to the failure to eliminate casting defects t hot extrusion, while the early fracture of the E9S-5 sample was considered to be th of incomplete refinement after five-pass swaging.    4 indicate that the refinement of the initial grains of all three groups of samples was mainly due to extensive twinning. Figure 11 shows the further segmentation of the twin lamellae in HS-2 and E25S-2 samples. Massive dislocation arrays were formed within the twin lamellae, as shown in Figure 11a,d, which refined the twin lamellae into many nanoscale subgrains. Boundary 1 (b 1 in Figure 11b) and boundary 3 (b 3 in Figure 11e) are the typical dislocation arrays formed in the Mg alloys [25], which have a small orientation of~5 • . These dislocation arrays would transform into subgrain boundaries with the increase in dislocation pile-ups [25], such as boundary 2 (b 2 in Figure 11c) and boundary 4 (b 4 in Figure 11f) with the misorientation of~10 • and~8 • , respectively. The further formation of high-angle grain boundaries might have resulted from dislocation pile-ups and interactions in the boundaries and subsequent crystal rotation with their neighboring grains, which implies the occurrence of DRX during rotary swaging, as reported in our previous work [21]. Microstructure evolution indicates that the three groups of samples with different initial grain sizes showed the same grain refinement mechanisms during swaging. boundary 4 (b4 in Figure 11f) with the misorientation of ~10° and ~8°, respectively. The further formation of high-angle grain boundaries might have resulted from dislocation pile-ups and interactions in the boundaries and subsequent crystal rotation with their neighboring grains, which implies the occurrence of DRX during rotary swaging, as reported in our previous work [21]. Microstructure evolution indicates that the three groups of samples with different initial grain sizes showed the same grain refinement mechanisms during swaging.  (e,f) magnified HRTEM images of the white dotted boxes "e" and "f" in (d), respectively. Figure 8 shows that all three groups of samples had a gradient distribution of microhardness along the radial direction after swaging. In our previous work [21], it was proven that the higher hardness value at the center of the swaged alloy bars is attributed to the formation of NGs, which could not be formed at the edge even after five-pass swaging. A more obvious gradient distribution indicated more remarkable refinement in the central region of the alloy bars. In addition, Figures 5 and 7 show that a higher degree of grain refinement was obtained in the sample with a smaller initial grain size after the same swaging process. This indicates that the formation rate of NGs decreased with the increase in the initial grain size during swaging. This difference was probably attributed to the different capacities for activating non-basal dislocations. On the one hand, as initial grain sizes increased, the activity of non-basal dislocations decreased due to smaller grain boundary areas [13]. On the other hand, the sample with a larger initial grain size showed higher activity of twinning, as shown in Figures 3 and 4. Higher activity of twinning would inevitably decrease the number of activated movable dislocations due to the release of the local stress concentration through twinning, while dislocation arrays are reported to stem from the piling up of non-basal dislocations in Mg alloy [26,27]. Lower activity of non-basal dislocations in the sample could lead to more difficulty in the formation of dislocation arrays, which further delayed the refinement of twin lamellae [21] after the same swaging pass. As a result, the formation rate of NGs slowed down in the sample with a larger initial grain size.

Conclusions
In summary, three groups of AZ31 Mg alloy bars with remarkably different initial grain sizes were processed via rotary swaging at room temperature. The effect of the initial grain size on the grain refinement process was studied based on the observation of microstructure evolution under the same accumulated deformation degree. The following conclusions were drawn: (1) All three groups of samples showed a significant increase in microhardness and strength after rotary swaging. The improvement in hardness and strength was mainly attributed to the formation of nanoscale grains/subgrains in the central region of the alloy bars; (2) The formation mechanism of nanograins during rotary swaging was independent of the initial grain sizes. The initial coarse grains in the three groups of alloys were all segmented via multiple twinning and followed by the further refinement of twin lamellae through forming massive dislocation arrays; (3) The formation rate of nanograins significantly decreased with the increase in the initial grain size. After three-pass swaging, only local nanocrystallization could be obtained in the sample with a much larger initial grain size (~548 µm). After five-pass swaging, incomplete nanocrystallization was still observed in the sample with a modest initial grain size (~71 µm). In contrast, completely randomly oriented nanograins with well-defined and clean boundaries were obtained in the sample with the finest initial grain size (~18 µm). The lower degree of nanocrystallization in the sample with a larger initial grain size was mainly attributed to the higher activity of twinning at the early stage of swaging.