Recovery by triple junction motion in aluminium deformed to ultrahigh strains

Tianbo Yu, Niels Hansen, Xiaoxu Huang


Commercial purity aluminium at true strains ε=2∼5.5 was annealed in a wide temperature range (from room temperature to 220°C), and the evolution of microstructure was characterized using transmission electron microscopy (TEM) and electron backscattered diffraction (EBSD) techniques. Triple junctions in an ultrafine lamellar structure are classified into three categories based on the structural morphology, and a relationship is formulated between the density (length per unit volume) of triple junctions and the boundary spacing. The triple junction density increases with increasing strain during plastic deformation and decreases during isochronal and isothermal annealing. Based on TEM and EBSD observations, thermally activated triple junction motion is identified as the key process during the recovery of highly strained aluminium, leading to the removal of thin lamellae with small dihedral angles at the ends and structural coarsening. A mechanism for recovery by triple junction motion is proposed, which can underpin the general observation that a lamellar structure formed by plastic deformation during annealing can evolve into an equiaxed structure, preceding further structural coarsening and recrystallization. Within this framework, the grain boundary surface tension on triple junctions is discussed based on the structural parameters characterizing the deformed and annealed microstructure.

1. Introduction

In the present global strive for the development of strong and light metals, a route has been to take advantage of the well-known phenomenon—strain hardening. An increase of strain refines the microstructure by introducing dislocation boundaries, high angle boundaries and loose dislocations between boundaries, leading to an increase in the flow stress. This structural refinement increases the stored energy (Jm−3) in the metal, i.e. the driving pressure (Nm−2) for structural coarsening and recrystallization, which may counterbalance the gain in strength. Previous kinetic studies have shown that the rate of recovery, which precedes recrystallization, increases with increasing strain in rolled samples of iron and aluminium (Michalak & Paxton 1961; Furu et al. 1995; Vandermeer & Hansen 2008). An important problem, both scientific and technological, is therefore the recovery behaviour of metals deformed to very high strain (Driver 2004; Humphreys & Hatherly 2004), which is investigated in the present phenomenological and kinetic study, where aluminium is deformed to ultrahigh strains and annealed at low temperatures.

The structural morphology at a large strain is represented by a rolled structure, composed of lamellae parallel to the rolling plane (figure 1a,c). The structure is very fine with a width of a few hundred nanometres for the individual lamellae. The lamellae have a finite length and they are subdivided by interconnecting boundaries (see sketch in figure 1d). The lamellar structure illustrated in figure 1 is a typical high strain structure in face centred cubic (fcc) and body centred cubic (bcc) metals deformed by a variety of modes such as rolling (Liu et al. 2002), high pressure torsion (Zhang et al. 2008), plane strain compression (Zahid et al. 2009) and equal channel angular extrusion (Valiev & Langdon 2006). The lamellar boundaries are typically medium and high angle boundaries and the interconnecting boundaries are predominantly low angle dislocation boundaries. Such a structure may also be found in transition or deformation bands, which can be observed both after medium- and high-strain deformation (Dillamore et al. 1972). This structure contains a high stored energy, primarily in different types of boundaries but also in the form of loose dislocations. Furthermore, triple junctions may possess extra energy beyond the energy of adjoining boundaries (Shekhar & King 2008; Godfrey & Liu 2009), and at a given strain the dislocation density at or near triple junctions can also be significantly higher than at grain boundaries (Randle et al. 1996). The annealing behaviour of such a structure has been analysed previously both theoretically and experimentally based on a general assumption that an initial step of recovery is the development of an equiaxed subgrain structure as surface tension forces approach equilibrium (Dillamore et al. 1972; Jazaeri & Humphreys 2004). These analyses have been based on ideal structures where the lamellar boundaries are parallel and have infinite length and where the interconnecting boundaries are perpendicular to the lamellar boundaries. In such structures, triple junctions (lines) connect lamellar boundaries and interconnecting boundaries with a dihedral angle, which is 90°, between the interfaces. However, real deformation structures have a different morphology, consisting of triple junctions with very different characteristics and therefore an expected different behaviour during annealing. This assumption is the basis for the present study encompassing a characterization of triple junctions in a deformed structure and their behaviour during annealing. The experiments consist of ex situ microstructural observations (observation in the same area between annealing steps), and triple junction motion has been identified as an important recovery process during low-temperature annealing, leading to a transition from a lamellar to a near equiaxed morphology. The observations suggest a general mechanism, which is discussed based on present and previous observations.

Figure 1.

(a) An illustration of the rolling geometry, where RD is the rolling direction, ND is the normal direction and TD is the transverse direction. (b) Subgrain boundary structure in the rolling plane (RD–TD plane) of aluminium AA1050 (99.5% purity with an initial grain size approx. 100 μm) cold rolled to a true strain 5.5. (c) Lamellar boundary structure in the longitudinal plane (ND–RD plane) of the same sample. (d) A sketch of lamellar boundaries (bold black lines) and interconnecting boundaries (thin grey lines) in (c).

2. Triple junctions in a deformed lamellar microstructure

(a) Classification of triple junctions

This experimental study concentrates on deformation by rolling, which introduces a lamellar microstructure. Figure 1a shows a sketch of the rolling geometry, where RD is the rolling direction, ND is the normal direction and TD is the transverse direction. Figure 1b shows a typical equiaxed deformation microstructure in the rolling plane (RD–TD plane) of aluminium cold rolled to a thickness reduction of 99.6 per cent or a true strain of 5.5. For the same sample, a very different morphology can be observed in the longitudinal section (ND–RD plane), where a lamellar structure has evolved (figure 1c). The deformation-induced boundaries can be classified into two categories. One is the lamellar boundary, which is approximately perpendicular to ND; the other is the interconnecting boundary, which is approximately parallel to ND (figure 1d). This classification is based not only on the morphology but also on the crystallography—lamellar boundaries can be either high angle boundaries (greater than 15°, original grain boundaries and deformation-induced high angle boundaries) or medium-to-low angle dislocation boundaries (less than 15°), whereas interconnecting boundaries typically are low angle dislocation boundaries. In this material, approximately 60 per cent of the lamellar boundaries are high angle boundaries, the average lamellar boundary spacing is approximately 0.2 μm, and the average interconnecting boundary spacing is approximately 0.8 μm. A similar lamellar boundary structure can be also found in the ND–TD plane. Thus in three dimensions, the lamellar boundaries are lying approximately parallel to the rolling plane, and the interconnecting boundaries are standing approximately perpendicular to the rolling plane. In an idealized lamellar structure, the lamellar boundaries are parallel. However, in real structures, the structural observations in the present study show that at least three types of junctions must be considered (figure 2):

  • — Y-junctions, each formed by three lamellar boundaries;

  • — H-junction pairs, each formed by two lamellar boundaries and an interconnecting boundary between them; and

  • — r-junctions, each formed by three interconnecting boundaries.

Figure 2.

Illustrations and examples of three types of triple junctions in lamellar structures. (a) A Y-junction formed by three lamellar boundaries; (b) two H-junctions (an H-junction pair) formed by two lamellar boundaries and an interconnecting boundary between them; and (c) a random r-junction formed by three interconnecting boundaries. Triple junctions are highlighted in bold lines with dihedral angles 2θ indicated. Both Y-junctions and H-junctions are lying close to the rolling plane, whereas r-junctions are oriented almost parallel to ND.

Regarding the morphology of H-junctions, two types are found (figure 3). For the first type, the lamellar boundary is not deflected by the interconnecting boundary (positions A), whereas for the second type (positions B), the lamellar boundary is deflected.

Figure 3.

(a) A selected area in the longitudinal section of aluminium deformed to a strain of 5.5. Markers A and B indicate two types of H-junctions. (b) Sketch of lamellar boundaries (bold black lines) and interconnecting boundaries (thin grey lines) of the same area. Boundary misorientation angles are indicated above or to the right of each boundary segment.

(b) Triple junction density

A sketch of a lamellar structure is shown in figure 4, and it illustrates that the lamellar boundaries are lying in the rolling plane except where they meet at Y-junctions, forming round, flat volumes (pancake shapes); all the interconnecting boundaries are standing perpendicular to the rolling plane, forming pancake-shaped cells. The density of Y-junctions in a two-dimensional section (number per unit area) can be determined by direct counting and dividing by the area. If the Y-junction density in the longitudinal section is ρND−RDY, the Y-junction density expressed as length per unit volume is then Embedded Image 2.1 since Y-junctions are confined in the rolling plane but inclined randomly with respect to RD, e.g. Embedded Image, where φ is the inclination angle. This ratio, 2/π, is lower than 1 but higher than 1/2, where 1 corresponds to the case that all junctions are perpendicular to the observation plane and 1/2 corresponds to the case that all junctions orient completely randomly in three dimensions (Baddeley & Jensen 2005). Similarly, the H-junction density (length per unit volume) is Embedded Image 2.2 where Embedded Image is the number of H-junctions per unit area counted in the longitudinal plane. In the case of r-junctions, which are parallel to ND, the r-junction density (length per unit volume) is Embedded Image 2.3 where Embedded Image is the number of r-junctions per unit area in the rolling plane.

Figure 4.

(a) An illustration of a lamellar boundary structure in the longitudinal section, where the lamellar boundary (red) spacing is D, the interconnecting boundary (cyan) spacing is l (the lamella ends are also included for this measurement), and the lamella length is L. (b) An illustration of the three-dimensional boundary structure of the marked region in (a). Lamellar boundaries are shown in different shades of red, with the top layer transparent and interconnecting boundaries are shown in cyan. One Y-junction is shown in orange; all H-junctions are shown in blue; and all r-junctions are shown in green. Bold red and cyan lines represent boundaries intersecting the ND–RD plane.

Direct counting is a straightforward way to determine the triple junction density. However, the triple junction density can also be derived from the boundary spacings in a two-dimensional section. In the case of Y-junctions, since each lamella contributes two Y-junctions, it follows that ρND−RDY=2/(DL), where D is the lamella spacing and L is the lamella length measured in the longitudinal plane. Therefore by incorporating equation (2.1), the Y-junction density can be expressed as Embedded Image 2.4 Similarly, if the cell length l (the interconnecting boundary spacing) in the longitudinal plane is measured, the density of H-junctions can be calculated as Embedded Image 2.5 In the case of r-junctions, the density will depend on the detailed morphology of the interconnecting boundaries. If we assume in the rolling plane that all r-junctions are within the central region of each lamella with a diameter 2(Rr), where 2R is the diameter of the lamella and 2r is the diameter of the cell, and R=2L/π and r=2l/π owing to their pancake shape, then the r-junction density is Embedded Image 2.6 It follows that the densities of triple junctions can be either determined by direct counting or estimated from equations (2.4) to (2.6) by measuring the lamella spacing D, the lamella length L and the interconnecting boundary spacing l in the longitudinal section (figure 4a). These equations also predict that the triple junction density increases with decreasing boundary spacing. However, for very short lamellae containing no interconnecting boundaries (l=L), there will be no H-junctions and r-junctions within the lamella. This method can easily be adopted and an example is chosen where the boundary spacings of pure nickel at different rolling reductions have been measured by Hughes & Hansen (2000). The corresponding triple junction density is calculated according to equations (2.42.6), and it is found that the density of all three types of triple junctions increases with increasing strain with the H-junction density being the highest (figure 5).

Figure 5.

Triple junction density in pure nickel deformed by cold rolling to various strains. The calculation is based on boundary spacings reported by Hughes & Hansen (2000). Squares with solid line, Y-junction; circles with solid line, H-junction; triangles with solid line, r-junction. (Online version in colour.)

(c) Structural parameters

Triple junctions are generally assumed not to drag the boundary motion during grain growth, and their role is reduced to preserve the thermodynamically described equilibrium angles at the lines where boundaries meet (von Neumann 1952; Mullins 1956). The underlying assumption is that the triple junction has a practically infinite mobility. However, triple junction drag was first considered by Galina et al. (1987) and then observed by Czubayko et al. (1998) in zinc and by Protasova et al. (2001) in aluminium during boundary migration in tricrystals above the recrystallization temperature. More recently, Gottstein et al. (2010) have derived a kinetic equation of grain growth taking both triple junction drag (finite mobility) and triple junction energy into consideration. An important parameter here is the dihedral angle, which together with the boundary energies, determines the grain boundary surface tension (Nm−1) on triple junctions. In these earlier studies, ideal samples have been considered. However, deformed structures are complex as a high density of dislocations and a high stored energy are present. As to the choice of key parameters, the following are considered: the spacing between and the misorientation angle across different types of boundaries, and the dihedral angle between the boundaries forming the triple junction.

3. Experimental method

The samples used are commercial purity aluminium AA1050 (99.5% purity with an initial grain size approx. 100 μm) cold rolled to true strains (logarithmic strains) 2, 4 and 5.5, corresponding to 86.5, 98.2 and 99.6 per cent thickness reductions, respectively. Owing to the high stored energy in the heavily deformed state, recovery may take place at ambient temperatures. Therefore immediately after cold rolling, one piece of the strain 5.5 aluminium has been stored in a freezer (about −20°C), whereas all other samples have been kept at room temperature. As previous kinetic studies have shown that the recovery rate increases with increasing strain in cold rolled aluminium (Vandermeer & Hansen 2008; Yu et al. 2009), focus has been on aluminium deformed to the highest strain with the highest stored energy. In order to separate recovery from recrystallization and also to study the progression of recovery, low isothermal annealing temperatures (from room temperature to 220°C) have been chosen.

Transmission electron microscopy (TEM) allows for detailed characterization of the morphology of boundary structures and dislocation networks in the highly strained samples. Moreover, boundary misorientation angles can also be determined with high accuracy (approx. 0.1°) using Kikuchi diffraction (Liu 1995). Besides the use of TEM, supplementary electron backscattered diffraction (EBSD) has been applied. The main advantage of the EBSD technique is that bulk samples (not restricted to thin foils as in the case of TEM) can be examined and orientations can be automatically obtained, whereas the main disadvantage is its relatively poor spatial resolution (approx. 20nm with a field emission gun) and poor angular accuracy (approx. 1°; Humphreys 2004). Since both TEM and EBSD methods are two-dimensional characterization techniques, an appropriate observation section has to be selected. In the present study, all the characterizations have been carried out in the longitudinal section (ND–RD plane). Therefore, all lamellar boundaries are viewed approximately edge-on, facilitating the characterization of Y-junctions and H-junctions, and the choice is also consistent with previous microstructural characterization of cold rolled metals (e.g. Hughes & Hansen 2000; Liu et al. 2002).

In order to follow the triple junction motion during recovery annealing, ex situ TEM observations were carried out. TEM thin foils of aluminium in the deformed state were prepared using a modified window technique (Christiansen et al. 2002), and the experiment was carried out in a JEM 2000FX transmission electron microscope, which was operated at a relatively low accelerating voltage of 120 kV in order to minimize beam damage. For selected areas in the thin foils, the deformation structures were imaged, and the boundary misorientation angles were determined. The TEM foils were then removed from the TEM holder and annealed in an air furnace, or simply exposed at room temperature for different time intervals. After annealing, the samples were mounted back in the TEM holder according to predefined macroscopic markers, allowing the same area to be examined repeatedly after annealing, where the accumulated annealing time has been 62 days, 10 h and 1 h for annealing at room temperature, 120°C and 220°C, respectively (table 1).

View this table:
Table 1.

Choice of sample and annealing temperature for ex situ TEM and EBSD observations.

In order to follow the triple junction motion during annealing in bulk specimens, ex situ EBSD observations were carried out for all three strains (table 1). An EBSD bulk specimen in the deformed state was prepared by electropolishing, followed by the characterization of one selected area in a Zeiss Supra-35 field emission gun–scanning electron microscope equipped with an HKL EBSD system. As in the case of ex situ TEM measurements, the specimen was then removed from the microscope sample stage and annealed (table 1). After annealing, the specimen was mounted back in the stage the same way according to macroscopic markers, and an EBSD orientation map was obtained of the original selected area according to the carbon contamination left after the first scan. In principle, the EBSD step size should be selected to be as small as possible for the present highly strained material. However, fine step sampling resulted in a strong electron beam-induced carbon contamination of the scanned region. Therefore in most cases, a step size of 30 nm was chosen. Besides, a partial tilt correction around ND (46.8° tilt correction in software instead of the physical tilt angle 70°) was applied in the scanning geometry, resulting in a real step size of 60 nm in RD. The maps were then expanded in RD to create an artificial step size of 30 nm to be consistent with the step size in ND.

The microstructural evolution characterized by ex situ observations may be affected by the presence of the sample surface (one surface in EBSD samples and two surfaces in TEM foils). Although thermal grooving is not pronounced at these low temperatures (Ivanov et al. 2004), the presence of a surface may still relax the structure near the surface, reducing the driving pressure for annealing processes. Conventional post-mortem EBSD examinations in the sample interior were therefore carried out for the strain 5.5 aluminium sample annealed at 220°C in order to compare the bulk and the surface annealing behaviour.

In the present study, a direct method based on counting was adopted to determine the triple junction density. Most measurements of boundary spacing and triple junction density were carried out manually from TEM images (measurements for strains 2 and 4 were based on micrographs of AA1200 aluminium, see §4) or EBSD orientation maps. However, the lamellar boundary spacings in EBSD datasets were all determined automatically by the line interception method, and the intercept length was determined along ND by defining a threshold misorientation angle at 1.5°. All EBSD maps used in determining the lamella boundary spacing were scanned along ND to minimize errors from image drift. Owing to limited angular resolution of the EBSD technique, the interconnecting boundary spacing and the H-junction density were determined from TEM images.

4. Results

(a) Microstructural parameters in the deformed state

Table 2 shows that the average boundary spacing both between lamellar boundaries and between interconnecting boundaries decreased with increasing strain, but the aspect ratio (l/D) was only slightly changed. The TEM data for strains 2 and 4 were measured from aluminium AA1200 (99.1% purity), i.e. with a slightly lower purity than AA1050 in the present study, but for both materials, the parameters were comparable to previous studies of rolled aluminium (e.g. Liu et al. 2002; Mishin et al. 2010). By direct counting and applying equations (2.1) and (2.2), the Y- and H-junction densities were found to increase with decreasing boundary spacing. The H-junction density was much higher than the Y-junction density, which is consistent with equations (2.4) and (2.5), since the lamella length L is much larger than the cell length l (the interconnecting boundary spacing). EBSD measurements generally agreed with TEM measurements. The density of r-junctions was not measured directly in the present study. However, by combining equations (2.4) and (2.6), the r-junction density ρr can be calculated (table 2) as D, l and ρY were known. The calculated r-junction density also increased with increasing strain as with ρY and ρH. The evolution of triple junction densities with strain is further illustrated in figure 6, showing similarities between aluminium and nickel (figure 5) when cold rolled. However, a much coarser structure is present in aluminium owing to more intrinsic dynamic recovery during rolling.

View this table:
Table 2.

The average values (with standard errors) of the lamellar boundary spacing (D), interconnecting boundary spacing (l) and triple junction density for commercial purity aluminium at different strains.

Figure 6.

TEM measurements of triple junction density in AA1200 (strains 2 and 4) and AA1050 (strain 5.5) aluminium. Error bars for Y-junction densities are smaller than the symbols used. Squares with solid line, Y-junction; circles with solid line, H-junction; triangles with solid line, r-junction. (Online version in colour.)

(b) Transmission electron microscopy observations of triple junction motion

Ex situ TEM observations were carried out for samples deformed to a strain of 5.5 and annealed at 120°C. One selected area in the deformed state is shown in figure 7a. After 1 h at 120°C, most of the boundaries have not changed their positions, since the TEM micrographs taken before and after annealing match very well. However, local changes near Y-junctions were found as illustrated in figure 7b. This structural evolution can be described phenomenologically as a migration of Y-junctions. Following this Y-junction migration, the lamellar boundaries involved consequently migrated laterally. In this area, 14 per cent of Y-junctions migrated different distances causing either shortening or removal of lamellae and local widening of neighbouring lamellae. This Y-junction migration caused the average boundary spacing to increase about 1.5 per cent after annealing at 120°C for 1 h.

Figure 7.

(a) A TEM micrograph montage of a selected area in the longitudinal section of aluminium deformed to a strain of 5.5. (b) A sketch of the same area combining the deformed state and the state after annealing at 120°C for 1 h. Solid thin lines represent lamellar boundaries that are unchanged; dashed lines represent boundaries that have migrated with new positions represented by solid bold lines. The areas swept by migrating boundaries are shown in grey, and the directions of the Y-junction motion are indicated by arrows. An area is marked showing pronounced Y-junction motion.

With longer annealing times, more Y-junctions migrated and they migrated over longer distances, indicating a time effect. Figure 8a shows an example of a migrating Y-junction. It migrated rapidly in the beginning (approx. 0.72 μm in 1 h) before it became pinned by an H-junction (Y–H pinning, figure 8b). With further annealing for up to 5 h, little change occurred in the position of the triple junction. After annealing for 10 h, the junction escaped from the pinning and arrived at the position shown in figure 8c. Another example is shown in figure 9a, where the shortening lamella has a bright contrast. In this case, the migrating Y-junction was initially pinned by an H-junction (figure 9b), but it escaped after annealing for 5 h. However, in the final position shown in figure 9c, the Y-junction was pinned by a neighbouring Y-junction (Y–Y pinning, figure 9d).

Figure 8.

Ex situ TEM observations of Y-junction motion in the longitudinal section of aluminium deformed to a strain of 5.5 and annealed at 120°C. (a) Positions of the Y-junction before and after annealing for 1, 2, 5 and 10 h. The marked region shows pinning by an H-junction and it is sketched in (b) with boundary misorientation angles indicated. (c) The final position of the Y-junction (marked by a circle) after 10 h annealing, where the rectangular marked region corresponds to the region in (a).

Figure 9.

Ex situ TEM observations of Y-junction motion in the longitudinal section of aluminium deformed to a strain of 5.5 and annealed at 120°C. (a) Positions of the Y-junction before and after annealing for 1, 2, 5, and 10 h. The marked region shows pinning by an H-junction which is sketched in (b) with boundary misorientation angles indicated. (c) The final position of the Y-junction (marked by a circle) after 10 h annealing. The rectangular marked region corresponds to the region in (a), and the circular marked region shows pinning by a neighbouring Y-junction, which is sketched in (d).

As a previous analysis by hardness measurements has shown that recovery may take place already at room temperature (Yu et al. 2009), Y-junction migration was also studied by ex situ TEM observations of samples which were originally stored in the freezer and then kept at room temperature for different time intervals. An example is shown in figure 10. The Y-junction migrated downwards until it was pinned by a particle. However, as expected, Y-junction motion was much less frequently observed at room temperature than at 120°C.

Figure 10.

Ex situ TEM observations of Y-junction motion in the longitudinal section of aluminium deformed to a strain of 5.5 and kept at room temperature for 6, 16 and 62 days. (a) Positions of the Y-junction. The marked region shows pinning by a particle, which it is sketched in (b) with boundary misorientation angles indicated.

Y-junction motion at 220°C was found to be much more frequent than at 120°C, illustrating a strong temperature effect. An example is shown in figure 11, where significant Y-junction motion took place in an initially finely spaced region, leading to doubling of the lamella spacing locally after annealing for 1 h. The microstructure during annealing evolved into being more equiaxed, but it kept a clear lamellar morphology.

Figure 11.

Ex situ TEM observations of Y-junction motion in the longitudinal section of aluminium deformed to a strain of 5.5 and annealed at 220°C. (a) Microstructure of the same area before and after annealing for 1, 8 and 60 min. (b) Sketches of the same area at different states corresponding to (a). For the states after annealing, each sketch combines the current state with the previous state to show the structural evolution, with the same markings as in figure 7. The arrows indicate the directions of Y-junction motion and are further illustrated in the marked areas in (a,b).

Y–H pinning was found to be very common (figures 8 and 9), whereas Y–Y pinning was less frequent. Misorientation angle measurements were used in some cases in order to distinguish between the two types. Figure 12a shows an area containing both types of pinning. Y–H pinning can be recognized when neighbouring H-junctions and Y-junctions were identified. Y–Y pinning at site ‘A’ can be easily determined from its morphology; however, at sites ‘B’ it may be difficult to determine whether they are H-junction pairs or Y-junction pairs. However, boundary misorientation measurements (figure 12b) show that sites ‘B’ are also examples of Y–Y pinning. In addition, particles of different sizes may also pin the Y-junction motion (figure 10), although observations of such an effect were infrequent compared with pinning by neighbouring boundaries.

Figure 12.

(a) A selected area in the longitudinal section of aluminium deformed to a strain of 5.5 and annealed at 120°C for 10 h. (b) Sketch of lamellar boundaries (bold black lines) and interconnecting boundaries (thin grey lines) of the same area. Boundary misorientations are indicated above or to the right side of each boundary segment. The markers ‘A’ and ‘B’ indicate sites where two Y-junctions pin each other.

The Y-junction motion caused shortening and finally removal of a specific lamella, followed by widening of the neighbouring lamellae. This Y-junction motion reduced the area of lamellar boundaries per unit volume and increased the average lamellar boundary spacing. Furthermore, this Y-junction motion also removed interconnecting boundaries within the disappearing lamellae and extended the interconnecting boundaries within the neighbouring lamellae. An example is shown in figure 8a, where the interconnecting boundary on the left side was extended after removal of the dark lamella in the middle. It follows that H-junctions and r-junctions within disappearing lamellae also disappear. However, H-junctions within neighbouring lamellae may change characteristics and an example is shown in figure 8, where the pinning H-junction initially connected to a lamellar boundary with a misorientation angle of 53.5° was replaced by a lamellar boundary of misorientation angle 30.4° after disappearance of the middle dark lamella.

(c) Microstructural parameters in the recovered state

Most of the migrating Y-junctions were composed of three high angle boundaries (figure 8b) or a combination of high angle boundaries and medium-to-low angle boundaries (figure 9b), but cases of migrating Y-junctions formed by three medium-to-low angle boundaries (figure 10b) were also found although much less frequently. A similar trend in frequency was also found for the not-migrating Y-junctions; however, it is still not clear whether an effect of boundary misorientation angle on the Y-junction motion is significant.

In addition to structural parameters such as spacing between and misorientation angles across lamellar and interconnecting boundaries, the triple junctions were characterized by their dihedral angles. The dihedral angles 2θ, as shown in figure 2a, of Y-junctions were measured using TEM by aligning the Y-junction line parallel to the electron beam. Measurements showed that these angles were generally between approximately 40° and 120° for aluminium deformed to a strain of 5.5. As an approximation, the dihedral angle of Y-junctions may also be estimated from TEM micrographs (figure 7a), and the two methods give almost similar results. Figure 13 plots dihedral angles versus the lamellar boundary spacing showing that: (i) the estimated dihedral angle appeared to increase with increasing lamella spacing although an exact value must depend on the local configuration (e.g. affected by neighbouring H-junctions); (ii) a migrating Y-junction generally had a small dihedral angle and the corresponding shortening lamella generally had a small spacing; and (iii) neither a lamella spacing below approximately 50 nm nor a dihedral angle below approximately 30° was observed. However, the large dispersion of data shown in figure 13 indicates that other factors may also play a role in determining the stability of a Y-junction.

Figure 13.

Dependence of the dihedral angle on the lamella spacing of Y-junctions in aluminium deformed to a strain of 5.5. The dihedral angles were estimated from TEM micrographs taken from the longitudinal section and the lamella spacing was measured near the same Y-junction. Open circles represent random measurements, whereas solid points represent measurements of Y-junctions observed to have migrated after annealing at 120°C for 5 h. (Online version in colour.)

(d) Electron backscattered diffraction observations of triple junction motion

Y-junction motion observed during ex situ EBSD experiments at the three strain levels and at an annealing temperature of 120°C is shown in figure 14. The motion of the Y-junctions gave rise to either shortening or the removal of lamellae, resulting in widening of neighbouring lamellae. Ex situ EBSD maps showed that the Y-junction motion was always associated with finely spaced lamellae and that the rate of Y-junction motion increased with increasing strain (the example shown in figure 14a is rare whereas that in figure 14c is frequent), indicating an effect of strain. The present deformation structures were subdivided into typical rolling texture components, and Y-junction motion was examined for each rolling texture component but no significant orientation preference was found.

Figure 14.

Examples of Y-junction motion in the longitudinal section (ND–RD plane) of aluminium deformed to strains (a) 2, (b) 4 and (c) 5.5 and annealed at 120°C for 5 h. The maps, both for the deformed state and annealed state, are coloured according to the orientation of TD for each pixel (colour coding is shown by the triangle), and the white pixels correspond to not-indexed points in the raw EBSD data and are mainly distributed along boundaries. The sketches follow the same markings as in figure 7 and marked areas are magnified in these sketches.

Y-junction motion at room temperature was also observed by ex situ EBSD. However, even after about 1 year's exposure, only limited Y-junction motion was found, in agreement with the TEM observations. Also consistent with the TEM observations, significant Y-junction motion was observed by ex situ EBSD in samples annealed at 220°C.

Figure 15a shows the evolution of the lamellar boundary spacing (noted as ‘spacing in interior’) and the Y-junction density after isothermal annealing at 220°C for the strain 5.5 aluminium. The lamellar boundary spacing increased with increasing annealing time, whereas the Y-junction density decreased during annealing. The same trend for the lamellar boundary spacing evolution was also found in the ex situ EBSD observations, i.e. changes of the spacing on a polished sample surface in the longitudinal plane (noted as ‘spacing at surface’ in figure 15a). The coarsening at the surface was only slightly slower compared with that in the sample interior, and it is concluded that the ex situ EBSD observations appear to be representative of the bulk behaviour. In figure 15b, the lamella spacing is plotted versus the Y-junction density. It can be seen that as recovery proceeds, the Y-junction density decreases, leading to increases in the lamella spacing.

Figure 15.

(a) Evolution of the lamellar boundary spacing (both in the sample interior and at the surface) and the Y-junction density during isothermal annealing at 220°C for AA1050 aluminium deformed to a strain of 5.5. (b) Evolution of the lamellar boundary spacing (in the sample interior) versus the Y-junction density. Error bars show the standard errors. Squares with solid line, spacing in interior; triangles with solid line, spacing at surface; circles with solid line, junction density. (Online version in colour.)

5. Discussion

The recovery behaviour in aluminium deformed to large strains has been examined thoroughly by Furu et al. (1995) and Nes (1995), who analysed mechanisms such as climb, glide and annihilation of dislocations. Kinetic studies have also been carried out by Vandermeer & Hansen (2008) and Yu et al. (2009) with the findings that (for AA1050 aluminium):

  • — A lamellar structure coarsens during recovery and evolves towards a more equiaxed morphology.

  • — The recovery rate is high and increases with increasing strain.

  • — The apparent activation energy increases as recovery proceeds.

  • — Recovery starts at room temperature and the recovery rate increases with temperature towards recrystallization.

  • — A recovery window can be defined by a maximum annealing time (approx. 100 h) at 220°C as longer times and higher temperatures lead to the initiation of recrystallization.

Within this frame, recovery by triple junction motion is discussed based on the present findings, which show that: (i) the rate of triple junction motion increases with the strain; (ii) this mechanism operates in the temperature interval from room temperature to 220°C; and (iii) its dependency on annealing time and temperature indicates that it is a thermally activated process.

(a) Triple junction motion

In well-annealed polycrystals, local equilibrium of grain boundaries at a triple junction is generally assumed and in the case of high angle boundaries, a dihedral angle of 120° is expected if the energy of the connecting boundaries is independent of the boundary misorientation angle. However, dihedral angles of 120° are not representative of the deformed lamellar structure as illustrated above. Considering, for example, Y-junctions, most of the boundaries are medium-to-high angle boundaries and have boundary energies of the same order (Read & Shockley 1950). As equilibrium (neglecting torque terms) can be expressed as follows (Herring 1951; King 2010): Embedded Image 5.1 where 2θi(i=1,2,3) is the dihedral angle between two boundaries and γi(i=1,2,3) is the energy (Jm−2) of the third boundary in the junction, it is apparent that the junctions in the deformed structure are not in equilibrium. It follows that there is a grain boundary surface tension (Nm−1) exerted on the triple junction. For the simplest case of a symmetric Y-junction (figure 2a), assuming a constant boundary energy (γ), the grain boundary surface tension (σ) for the motion of the Y-junction can be expressed as Embedded Image 5.2 With dihedral angles varying in the range approximately 40°–120°, the grain boundary surface tension will decrease with increasing dihedral angle as shown in figure 16.

Figure 16.

Normalized grain boundary surface tension as a function of the dihedral angle at a Y-junction. The shaded region corresponds to the experimentally observed range of dihedral angles.

Figure 13 shows that the dihedral angle generally decreases with decreasing lamella spacing in the deformed state, and it follows from figure 16 that the grain boundary surface tension for Y-junction motion increases with decreasing lamella spacing. With increasing strain during plastic deformation, the lamellar boundary spacing decreases (Hughes & Hansen 2000; Liu et al. 2002), therefore increasing the grain boundary surface tension for Y-junction motion and decreasing the stability of the deformation microstructure. The present observations and analyses generally agree with the model suggested by Galina et al. (1987), where the triple junction velocity is assumed to be proportional to the grain boundary surface tension at a given temperature and the grain boundary surface tension increases as a function of dihedral angle and spacing. However, some observations here are not in direct support of the model. For example, the model cannot explain the absence of dihedral angles below 30° or lamellae thinner than 50 nm, although the grain boundary surface tension is still the same order of magnitude (figure 16). This leads to the tentative suggestions that not only the grain boundary surface tension but also the triple junction mobility may depend on its dihedral angle, being very large for a sharp triple junction, or an internal stress may build up at sharp triple junctions during deformation, leading to stress-induced triple junction motion and the removal of very thin lamellae.

Besides a reduction of the boundary energy, there may be other causes for the Y-junction motion. Strain energy from dislocations may contribute to the driving pressure since there may be a difference in dislocation density within different lamellae, although this contribution may not be so significant since the stored energy from these loose dislocations is estimated to be much lower than the contribution from the boundary energy (Godfrey & Liu 2009). Moreover, in the bulk interior, Y-junction lines form loops (connected by quadruple points). Since Y-junctions are one-dimensional defects similar to dislocations, they have a line energy (J m−1) (Zhao et al. 2010). In order to minimize this energy, a Y-junction loop is inclined to shrink, consequently removing the confined lamella. However, if the triple junction line energy in a deformed sample equals that in a fully annealed sample, then its contribution to the total stored energy may be not significant (Gottstein et al. 2010) in the present structure.

For the H-junctions between lamellar and interconnecting boundaries, they combine boundaries with misorientation angles (and energies), which are significantly different. Owing to the low energy and low mobility of interconnecting boundaries, H-junctions should be more stable than Y-junctions, which is observed. The stability of r-junctions has not been examined but their behaviour may be close to H-junctions as r-junctions connect only low angle boundaries. Therefore in the following, only the motion of Y-junctions will be considered.

(b) A recovery mechanism based on Y-junction motion

An important structural change during recovery is that the deformation structure coarsens and evolves from a lamellar structure to a more equiaxed configuration (Prangnell et al. 2004; Xing et al. 2006). It follows that the removal of lamellar and interconnecting boundaries is a key issue when modelling the microstructural evolution. Rotation and coalescence (Li 1962; Jones et al. 1979) may be a possibility for the removal of interconnecting boundaries but may not be operative for lamellar boundaries which have medium-to-high angles. For these boundaries, a possibility is coarsening by strain-induced grain boundary migration (SIBM), which is a common process when recrystallizing samples deformed to low and medium strain (Beck & Sperry 1950). However, this process is not likely to operate easily at low temperatures in highly strained samples where the structure is extremely fine, and it has not been observed in the present study. A third mechanism is based on a lamellar structure with exactly parallel lamellar boundaries and interconnecting boundaries along a direction perpendicular to the lamellar boundaries, i.e. a structure without Y-junctions. This idealized structure has been taken to represent deformation bands observed at a medium strain (Dillamore et al. 1972) and rolling structures observed at a large strain (Jazaeri & Humphreys 2004). During annealing of such a structure, it is suggested that a reduction in boundary energy can be established by a shortening of the interconnecting boundaries and thereby locally pulling the lamellar boundaries together. A further step is a collapse of lamellar boundaries (Jazaeri & Humphreys 2004), annihilating two H-junctions and forming two nodes which are Y-junctions. The resulting structure is a diamond-shaped structure and it is suggested that further spheroidization and growth will occur homogeneously owing to boundary tensions, leading to a more equiaxed grain structure.

The present observations suggest a new mechanism for structural coarsening during the initial stage of recovery in a lamellar structure, namely motion of Y-junctions. In TEM foils, migration of a Y-junction shortens confined lamellae and finally removes them by annihilation of Y-junctions. In the bulk, migration of a Y-junction shortens the Y-junction line and finally removes the confined lamella together with H- and r-junctions within the lamella. As a result, the motion of the Y-junction can lead to removal of both lamellar and interconnecting boundaries and cause a structural coarsening as observed (figure 11). Analysis of the details of structural coarsening shows a decrease in the triple junction density, an increase in the average lamellae spacing and decreases in the aspect ratios (L/D and l/D), towards a more equiaxed morphology. Besides recovery by triple junction motion, other processes may take place simultaneously, for example annihilation of mobile and redundant dislocations, leading to a decrease in the dislocation density in the lamellae and a sharpening of the dislocation boundaries (Humphreys & Hatherly 2004). However, these may not have a significant effect on the structural coarsening since these processes do not lead to the removal of lamellar boundaries. The observation and analysis suggest that recovery of a lamellar structure by motion of Y-junctions is a critical process, and in figure 17 a sketch illustrating how this mechanism can lead to structural coarsening and a transition towards an equiaxed morphology is given.

Figure 17.

A sketch illustrating structural coarsening via Y-junction motion. (a) The original deformation structure with typical lamellar boundaries (bold black lines) and interconnecting boundaries (thin grey lines). This lamellar structure coarsens via Y-junction motion where arrows show the direction of the motions. (b,c,d) Intermediate coarsening steps. (e) A more equiaxed structure has evolved after coarsening via Y-junction motion. (f) A modified equiaxed structure considering also the drag of interconnecting boundaries at H-junctions (i.e. interconnecting boundaries resist widening). (g,h) Examples of such coarsening structures in AA1050 aluminium deformed to a strain of 5.5 and annealed at 220°C for 1 h and 48 h, respectively.

(c) Recovery mechanism and structural observations

The recovery mechanism illustrated in figure 17 leads to a significant increase in the average lamellar boundary spacing, whereas the average lamella length and the average interconnecting boundary spacing may only change marginally. Such a microstructural change leads to a more equiaxed morphology, which is consistent with experimental observations since equiaxed structures are typical observations after recovery annealing of deformed metals (e.g. Prangnell et al. 2004; Xing et al. 2006). However, it must be noted that this equiaxed structure with l/D∼1 after coarsening still retains lamellar features since L/D>1, and a more perfect equiaxed structure requires further Y-junction motion and subgrain growth, also leading to a larger subgrain or grain size.

The mechanism suggests that interconnecting boundaries are stable and resist widening, leading to deflection of lamellar boundaries during coarsening (also seen in figure 8). Also, interconnecting boundaries appear to be stable and are generally not observed to pull lamellar boundaries together as suggested by Prangnell et al. (2004) and Jazaeri & Humphreys (2004). Also, extensive breakdown of lamellae after they are pulled together will lead to a rapid decrease of the lamella length. However, according to equation (2.4), the average lamellar length L=π/(ρYD), and with the data in figure 15 it follows that the average lamella length increases slowly during annealing, in favour of the present mechanism. Moreover, extensive breakdown of lamellae will lead to a rapid evolution of an equiaxed structure with a minor increase in boundary spacing, in contradiction to experimental observations that the lamellar morphology is retained after doubling the lamellar boundary spacing (figure 11).

According to the proposed mechanism, thick lamellae are more stable than thin lamellae, which are terminated by small dihedral angles at Y-junctions and provide large grain boundary surface tensions for Y-junction motion, and coarsening takes place uniformly by continuous removal of the thin lamellae. Such a uniform coarsening may result in a rough scaling behaviour of the distribution of the lamellar boundary spacings as in the case of a structural refinement by rolling (Godfrey & Hughes 2000). However, slight changes in the shape of the scaled distribution may also be expected during uniform coarsening owing to a particular initial distribution of spacings in the deformed state (i.e. the shape of the scaled distribution may approach a stable one, which may be different from that characterizing the deformed state). Since observations (Godfrey & Hughes 2000; Hughes & Hansen 2000) show that the initial distribution is always positively skewed (the tail on the right side is longer than the left side), the skewness of the distribution will decrease slightly owing to uniform coarsening by the removal of thin lamellae and the distribution will become more symmetric. This is confirmed by the present EBSD data on the strain 5.5 aluminium as shown in figure 18, where the skewness decreases at the beginning of recovery. This figure also shows that the skewness increases at longer annealing times, indicating the initiation of non-uniform coarsening, which is not governed by triple junction motion. The initiation of a non-uniform coarsening at the annealing time of 96 h is consistent with a previous determination of the initiation of recrystallization for the same material based on a microstructural characterization (Yu et al. 2009).

Figure 18.

Evolution of the skewness of the lamellar boundary spacing distribution in the strain 5.5 aluminium during isothermal annealing at 220°C. In the definition of skewness, n is the size of the distribution, Di is the ith element in the distribution, Graphic is the mean and s is the standard deviation.

An analysis of the kinetics of recovery based on motion of Y-junctions is not part of the present analysis. However, the observation that Y-junction motion is thermally activated agrees with previous models for recovery kinetics (Humphreys & Hatherly 2004). The finding in one new model (Vandermeer & Hansen 2008) that the apparent activation energy increases from the beginning of recovery to the beginning of recrystallization may relate to the observation that thin lamellae appear to be more unstable than thick ones, which have a lower grain boundary surface tension. As a result, it is to be expected that it will be increasingly difficult to remove lamellae in the course of recovery, leading to an increase in the apparent activation energy as observed.

Y-junctions are stabilized to some extent by H-junction pinning (figure 8). This Y–H pinning, as well as Y–Y pinning and particle pinning, is related to an increase in the dihedral angle of Y-junctions and thus reduces the grain boundary surface tension on Y-junctions. The fact that enhanced Y-junction motion takes place at a high strain indicates that an increase in Y–H pinning has a limited effect when the strain is increased. Stabilization by particle pinning is a possibility and a reduction in the density of Y-junctions is another, for example, by changing the deformation process in such a way that the morphology of the deformed microstructure can be changed from lamellar to equiaxed.

6. Conclusions

  • — Based on boundary morphology, triple junctions in a lamellar structure have been classified into three categories (Y-junctions, H-junctions and r-junctions). The relationship between the density (length per unit volume) of each type of triple junction and boundary spacings has been formulated. The triple junction density increases with increasing degree of strain, and decreases during annealing.

  • — Thermally activated Y-junction motion is identified as a key process during coarsening of highly strained aluminium (ε=2∼5.5) over a wide temperature range (from room temperature to 220°C) both at the sample surface and in the sample interior, with slightly faster kinetics in the sample interior. The motion of Y-junctions, accompanied by lateral movement of lamellar boundaries, causes shortening and removal of lamellae and local widening of neighbouring lamellae.

  • — The stability of a Y-junction is found to be more dependent on its morphology than the misorientation angles of adjoining lamellar boundaries. The dihedral angle of Y-junctions roughly decreases with decreasing lamellar boundary spacing, and the decrease in the dihedral angle increases the grain boundary surface tension for Y-junction motion. The stability of lamellae consequently decreases as the lamella spacing decreases although it is also affected by neighbouring Y-junctions, H-junctions and pinning particles.

  • — A recovery mechanism based on triple junction motion is proposed, and it explains how a lamellar structure evolves into a more equiaxed structure by continuously removing the thin lamellae, which have small dihedral angles at the ends and provide large grain boundary surface tensions for Y-junction motion. This mechanism underpins the observation that the apparent activation energy for recovery increases during recovery.


The authors gratefully acknowledge the Danish National Research Foundation for supporting the Centre for Fundamental Research: Metal Structures in Four Dimensions, and the Danish National Research Foundation and the National Natural Science Foundation of China (grant no. 5091130230) for supporting the Danish-Chinese Centre for Nanometals, within which the present work was performed. Dr O. V. Mishin was thanked for supplying the material; Prof. A. Godfrey and Dr W. Pantleon are thanked for helpful discussions; and Prof. B. Ralph was thanked for critical language corrections.

  • Received February 7, 2011.
  • Accepted June 8, 2011.


View Abstract