The role of crack branching in stress corrosion cracking of aluminium alloys

1 Introduction

Stress corrosion cracking (SCC) is an insidious failure mechanism that is still the subject of considerable debate (Gangloff, 2008; Holroyd, 1990; Lynch, 2012; Newman & Procter, 1990; Parkins, 1979; Sadananda & Vasudevan, 2010; Sieradski & Newman, 1987; Turnbull, 1993). An applied stress, a corrosive environment, and a susceptible microstructure are the three prerequisites for failure by SCC. Accordingly, the approaches that focus predominantly on fracture mechanics, electrochemistry, or microstructural aspects to understand the underlying mechanisms often have limited applicability, such as the inextricable relationship between the stress, environment, and microstructure. Similarly, studies often focus either on macroscale observations (e.g. average crack growth rates or the time to failure of test specimens; Holroyd & Scamans, 2011) or at the grain to nanometre scale [e.g. transmission electron microscopy (TEM) studies and SEM fractography; King, Johnson, Engelberg, Ludwig, & Marrow, 2008; Lozano-Perez, Rodrigo, & Gontard, 2011]. As we will go on to describe in this paper, it is the mesoscale that links the macroscale of the test component to the nanoscale of the crack tip that is often overlooked. Consequently, there is a need for a better understanding of how the results of test components connect to the “real” conditions experienced at the crack tip if we are to effectively explore the proposed SCC mechanisms.

The propensity to SCC of high-strength aluminium alloys, as a function of alloy temper, is often characterised by v-K curves, where the crack propagation rate, v, promoted in macroscale precracked fracture mechanic-type specimens, is plotted as a function of the applied stress intensity factor, K^applied, calculated by assuming a single crack propagating with a relatively uniform through-thickness crack length. A typical v-K curve is provided in Figure 1b. It shows a highly K-dependent region of crack growth (region I) at low K’s, a relatively K-insensitive region for intermediate K’s (region II, often referred to as the “plateau” region), and a second K-sensitive region (region III) at high K’s, typically only generated under rising load conditions. An alloy or temper of a given alloy is deemed more SCC resistant, as the minimum applied stress intensity factor, K_1SCC (the start of region I), needed to generate a significant SCC growth rate (e.g. >10^-11 m/s) increases and the region II “plateau” velocity decreases. Typical v-K data for AA7032 (one of the materials used in this study) as a function of alloy temper when totally immersed in 2% NaCl solution with a sodium dichromate inhibitor addition are shown in Figure 1a, where the SCC propensities of the -TSCC, -T6, -T76, and -T73 tempers would be deemed to be highly susceptible, susceptible, relatively resistant, and highly resistant, respectively.

Figure 1:

(a) Average stress corrosion crack rate (or velocity), v, as a function of the applied stress intensity factor, K, often referred to as a v-K curve, for AA7032 in various alloy tempers totally immersed in an acidified inhibited saline solution and (b) schematic v-K curve showing regions I–III cracking along with suggested minimum K conditions for the types of microcracking proposed after Speidel (1971).

K _M, minimum stress intensity required for microbranching; K_P, stress intensity at the start of the region II plateau; K_B, minimum stress intensity required for macrobranching.

The geometry and environment at the crack tip has also been the focus of investigations at the nanoscale (Lozano-Perez et al., 2011; Staehle, 2010). At the grain scale, the local chemistry and crystallography can have a strong influence on the path that the stress corrosion cracks follows. Consequently, the susceptibility of specific grain boundaries has received quite a lot of attention, holding the promise of grain boundary-engineered materials that are resistant to SCC (Babout, Marrow, Engelberg, & Withers, 2006). The use of multiple different techniques applied across a range of scales is one of the themes of this paper and is receiving increasing attention as a method to investigate materials science engineering challenges (Burnett et al. 2014).

Crack branching is a mesoscale phenomenon taking place at a scale between that of the crack tip and the test component. When considering the SCC of aluminium and steel, it has remained largely at the periphery of our consideration when proposing mechanisms for the advancement of the cracks at the nanoscale or the features of v-K curves obtained from macroscale tests. Speidel (1971) and Carter (1971) described the presence and the circumstances under which crack branching under SCC conditions could occur. In this respect, the region II crack velocity plateau was cited as being of critical importance. The proposition was that crack branching only became possible after critical stress intensity was achieved (approximately double, later revised to ~1.4, the value at which the SCC began to measurably propagate, K_1SCC) and that the crack velocity was then constant across a range of stress intensities. The reason that branching could only occur once the SCC had reached a plateau in crack velocity (region II) was because otherwise crack branching would be stifled because the formation of any branches would have no chance to propagate as the original propagation direction would quickly run ahead and leave undeveloped crack branches behind.

Two types of crack branching were defined by Speidel (1971), namely, microbranching and macrobranching. Microbranching is where the crack front splits into several local cracks with separation distances of the order of a grain diameter, and macrobranching is where the crack separates into two or more macroscopic components that tend to diverge. Figure 1b illustrates a schematic v-K curve showing regions I to III cracking along with suggested minimum K conditions for the types of microcracking proposed by Speidel (1971). K_M is minimum stress intensity required to initiate microbranching, K_p is the stress intensity at the start of the region II plateau, and K_B is the stress intensity required to initiate macrobranching.

The phenomenon of crack branching is a fundamental characteristic of SCC and will have a significant effect on the stress intensity at the stress corrosion crack tip. Hence, we have examined the morphology of stress corrosion cracks in high-strength aluminium alloys using X-ray computed tomography (CT) to examine the importance of crack branching in more detail in three dimensions (3D). Until now, the direct evidence for crack branching has been limited, often relying on nonspatially resolved measurements (e.g. potential drop) or observations of cracks where they intersect the surface. Neither of these approaches offers the opportunity to accurately describe the morphology of the crack and the extent of crack branching. Here, we have used X-ray CT imaging to analyse stress corrosion cracks in two alloys showing quite different macroscopic SCC characteristics: the first a commercial AA7032 alloy that was heat treated to enhance susceptibility (-TSCC) before testing under controlled conditions and the second a commercial AA5083 alloy that was sensitised and failed in service.

2 Materials and methods

2.1 AA7032

Extruded AA7032 alloy (Al-6.0Zn-2.0Mg-2.0Cu-0.2Cr, wt%) was solution heat treated for 1 h at 475°C, cold water quenched, and given an underaged temper -TSCC (24 h at 100°C) to maximise SCC susceptibility. The typical grain structure is shown in Figure 2, with the grain size being in the 20–50 μm range. Although there is some variation in the orientation of the microstructure, it is also evident from Figure 2b that the sample orientation is such that the grain structure is not oriented exactly at 90° to the applied stress (i.e. Z-direction).

Figure 2:

(a) Photograph of the AA7032-TSCC microcompact tension sample with the X-, Y-, and Z-directions representing the transverse, forming, and normal direction, respectively.

Polarised light micrographs showing the typical grain structures in the (b) YZ plane, (c) XZ plane, (d) YZ plane showing a large area overview, and (e) XZ plane. Red arrows indicate the crack propagation direction. For (c) and (e), the crack growth direction is perpendicular to the page.

A mini-compact tension sample, 25×25×10 mm (see Figure 2a), was prepared and loaded to an initial K of 19.9 MNm^-3/2. When no crack initiation was visible after 1533 h in dry air, the notch was sharpened by electrodischarge machining (EDM) and the K increased to 24 MNm^-3/2. After a further 400 h exposure to dry air (~11% relative humidity; Johnson, 2002), no crack initiation was visible. At this time, the loaded test sample was removed from the dry air (with the load maintained) and exposed to an ambient environment for 11 years during which extensive SCC developed.

Subsequently, X-ray CT scanning was carried out on a Zeiss Xradia Versa 520 system (Carl Zeiss X-ray Microscopy, Pleasanton, CA, USA) with a tungsten target at 140 kV using a 4× optical magnification with an exposure time of 15 s per projection and a total of 2001 projections. The effective voxel size was 18 μm³. Data were reconstructed using a Feldkamp-Davis-Kress (FDK) reconstruction (Maire & Withers, 2014).

An FEI Helios plasma focussed ion beam (PFIB) (FEI Company, 5350 NE Dawson Creek Dr, Hillsboro, OR, USA) was used to create the cross-section of the fracture surface. Much higher milling rates are possible with the PFIB compared with the more traditional Ga ion FIB, allowing for the creation of the 700 μm wide and 250 μm deep cross-section shown with comparison to the X-ray CT data.

An FEI Magellan (FEI Company, 5350 NE Dawson Creek Dr, Hillsboro, OR, USA) scanning electron microscope (SEM) with MAPS™ software (FEI Company, 5350 NE Dawson Creek Dr, Hillsboro, OR, USA) was used to capture the fracture surface. Secondary electrons were collected using the ETD detector to create the image as an array of small tiles collected and automatically stitched together using the MAPS software. An array of 736 (23×32) tiles was collected with a pixel size of 244 nm. The total stitched image size was 62,352×38,701 pixels (15.2×9.4 mm). This provided an overview of the entire fracture surface and the ability to investigate fine features across the entire surface.

The crack opening displacement (COD) map was created using a MATLAB code developed in-house. The X-ray CT data of the crack was segmented to identify the crack and was used to calculate the total width of the crack at every point along the Z-direction. The regions of overlapping cracks were summed together to give one value of the COD.

2.2 AA5083

The AA5083 alloy (Al-4.76Mg-1.58Mn-0.59Si, wt%) was supplied as a 25-mm-thick rolled plate and designed for high strength and toughness. The plate was loaded in service and became sensitised during service over time and ultimately was removed from use after 40 years in the atmosphere at ambient temperature. The component was loaded in tension in the Z-direction (Figure 3a).

Figure 3:

(a) Photograph of the AA5083 sample showing the crack and the whole sample.

Electron backscattered diffraction (EBSD) maps showing grain orientation (in Euler colours) in the (b) YZ plane and (c) XZ plane. The crack propagation direction is shown with a red arrow.

X-ray CT scanning was carried out on a Nikon XTH 225/320 system (Nikon Metrology Europe, Derby, UK) with a tungsten target at 200 kV using an exposure time of 1 s per projection and a total of 3201 projections. The effective voxel size was 32 μm³. Data were reconstructed using an FDK reconstruction.

3 Results

The YZ view in Figure 4a shows a virtual cross-section taken from the centre of the X-ray tomograph of the AA7032 specimen displaying the crack “sideways on”. The stress corrosion crack can be seen propagating from the EDM notch. The XZ virtual cross-section (Figure 4b) appears to show multiple separated cracks inclined with respect to the conventional crack plane (the XY plane). The XY virtual cross-section (Figure 3c) again shows what appear to be distinct inclined cracks.

Figure 4:

Virtual (a) YZ, (b) XZ, and (C) XY cross-sections from the reconstructed X-ray CT data showing three orthogonal views of the AA7032-TSCC sample.

In the section in (a), the steel bolt and the machined notch are evident. The coloured lines indicate the trace of the YZ (red), XZ (blue), and XY (green) slices in the other orthogonal slices.

The crack morphology is more clearly seen in Figure 5, where the crack has been segmented and the alloy rendered transparent digitally (see Maire & Withers, 2014). This visualisation allows the exploration of the 3D shape of the crack from which it is clear that, contrary to the impressions gained from the YZ virtual cross-section shown in Figure 4a (or similarly from the visual inspection of the specimen surface), there is not a single crack lying in the XY plane. Instead, the 3D analysis in Figure 8 reveals that three major macroscopic cracks have propagated from the notch. A closer inspection (not shown) at the notch reveals that at least 10 separate cracks initiated from the notch 4 of which arrested, whereas the others coalesced to form the three major macroscopic cracks seen in Figure 5c. It is clear in Figure 6 that, after initiation, the SCC cracks have reoriented to follow the microstructure instead of following the orientation of the original notch (i.e. 90° to the applied stress) as shown in Figure 6c.

Figure 5:

(a) Virtual cross-section showing how regions of crack can be delineated (i.e. segmented) and then (b) visualised in 3D by rendering the solid transparent and (c) three orthogonal views where the notch and the three primary cracks have been artificially coloured differently.

The magnified inset shows the finger-like nature of the crack front.

Figure 6:

Virtual cross-sections of the stress corrosion crack perpendicular to the crack propagation direction (i.e. the Y-direction) for the AA7032 sample (a) just below the notch (b) and 10 mm away from the notch (half-way down the sample).

Virtual cross-sections of the AA5083 sample (c) near the crack origin and (d) close to the crack front. (e) Polarised light micrograph showing the representative grain structure of the AA7032 sample. (f) EBSD orientation map in Euler colours showing the grain structure for the AA5083 sample. The crack propagation direction is indicated by the red arrow.

For the AA5083 sample, there is a single primary crack (Figure 6c and d) that is oriented at 90° to the crack opening stress. This is in good alignment with the grain structure (Figure 6f).

The CT images enable a detailed examination of the morphology of the two different types of crack (Figure 6). A closer inspection reveals that, in both cases, the cracks propagate at the crack front via multiple branched microcracks. In the case of AA7032, the microcracks are inclined to the XY plane and those separated by significant ligament cracks persist (Figure 6b) to form large individual cracks. In contrast to AA5083, the multiple cracks found at the crack front are separated by very fine ligaments (Figure 6d), such that almost all the separate protruding fingers of the crack front coalesce when the bridging ligaments fail plastically as the crack opening increases, ultimately forming a single crack. This is due to the aligned nature of the microstructure with respect to the applied stress (Figure 6f) in the AA5083 sample, which means that the branched cracks are nearly coplanar, whereas they are parallel but not coplanar for the AA7032 (Figure 6c). In each case, the result of crack branching is the creation of metal bridging ligaments. In the former case, as the crack front moves forward and the COD increases, these metal bridges between the coplanar cracks are eventually ruptured, because the bridges are narrow and relatively easily ruptured but also because any lateral growth of the cracks leads to merging with another crack. In the case of the AA7032 sample, because the cracks are parallel but not coplanar, it was possible for three major cracks to propagate on parallel planes without having to coalesce. The metal ligament separating these cracks is substantial, making their rupture more unlikely. However, within these major cracks, once the orientation of the microstructure has been adopted, the branching and coalescence progresses in a similar way to the AA5083, as any branched cracks are now close to coplanar.

Multiple crack branching events are evident across the entire crack in the case of the AA7032 sample as shown in Figure 7. Such secondary cracks lie essentially parallel to the main crack plane and are left behind either because they became unfavourable for growth or other nearby cracks have superseded them by propagating these cracks and leaving these behind. Using correlative tomography techniques (Maire & Withers, 2014) to correlate the CT X-ray image with large sections milled destructively using a plasma dual-beam electron microscope in Figure 8, it is both clear that the cracks lie along grain boundaries and that there are many more such secondary cracks that lie below the resolution limit of the current CT images.

Figure 7:

(a) Virtual cross-section of the YZ plane showing the SCC crack propagating from the machined notch in the AA7032 sample.

A secondary crack is highlighted by the black arrow, and the crack propagation direction is shown by the red arrow. (b) 3D rendering of the same SCC crack (green) in plane view showing the secondary cracks in blue. The secondary crack indicated in (a) is also indicated by a black arrow. The red arrow shows the crack propagation direction. The right-hand edge of the crack shown here is in the bulk of the overlapping with the central (red) crack in Figure 5.

Figure 8:

(a) Virtual X-ray CT YZ cross-section of AA7032 showing a secondary crack and (b) the same cross-section recovered by large area destructive dual-beam plasma microscopy (FEI Helios PFIB) showing many more smaller microcracks and grain boundary contrast.

The small arrows in (a) and (b) denote the same locations; the bold red arrows denote the crack growth direction.

In the majority of cases, however, the microcracks coalesce with the main crack, such that only the failed metal ligaments and the stepped topography of the fracture surface remain when investigating the cracks post-mortem (see Figure 9).

Figure 9:

Features common to SCC fracture surfaces are shown in these two different examples.

(a) 3D rendering from the X-ray CT data showing the fracture surface of the AA5083 specimen with discrete steps over the fracture surface, a result of the closely spaced coplanar cracks. The broken metal bridges are not visible at this scale. (b) Large stitched array of SEM images of the entire fracture surface of the AA7032-TSCC sample. The small red arrows highlight thin layers of metal, which were metal bridges and have been pulled out of plane as the sample was pulled apart. The large red arrows indicate the crack propagation directions.

4 Discussion

We have investigated the 3D SCC crack morphology of two quite different samples. These suggest that crack branching, rather than being a phenomena that occurs under specific conditions, is an inherent feature of the growth of such cracks. In the case of the AA7032 sample, crack branching is macroscopically visible due to the misorientation of the grain structure close to the notch and the applied stress. This has resulted in some of the cracks that initiate from the notch (Figures 4 and 5a) being preserved as separate cracks rather than coalescing with neighbouring cracks. It is also noteworthy from Figure 6a and b that the cracks while being inclined tend to bend towards the conventional crack plane orientation as they grow laterally.

In the case of the AA5083 sample, where the microstructure was aligned with the loading orientation, we have to look very carefully for the evidence of crack branching. The highly aligned texture of the microstructure (Figure 6f) meant that the crack branching led to a number of very closely spaced parallel cracks at the crack front each no more than a grain or two from the XY crack plane. As the crack continued to propagate and the COD increased, the thin ligaments of metal separating these cracks ruptured leading to the coalescence of the multiple cracks to ultimately form a single crack albeit with multiple crack fronts. Thus, the evidence of crack branching is more difficult to find. It can only be seen in using nondestructive 3D CT imaging just behind the crack front (Figure 6d) and via the remnants of broken metal bridges left on the fracture surface (Figures 6c and 9a).

We argue as Kitagawa, Yuuki, and Ohira (1975) and Nakasa, Takei, and Yoshida (1979) have previously reported that the region II plateau should be considered as the symptom of crack branching and not its prerequisite. Further, our observations of the 3D morphology of stress corrosion cracks suggest that assuming a single crack to estimate the stress intensity at the crack tip is wholly inappropriate. The misorientation and the propagation of three primary cracks instead of a single crack in the AA7032 sample make the estimation of stress intensity (using geometric arguments from the C-T sample) at the crack front completely unreliable. In addition, the bridging ligaments will help to hold the crack closed reducing the COD (see Figure 10). Hence, the effective stress intensity, K^effective, will be a fraction of the K^applied. Consequently, plotting crack velocity against stress intensity in a v-K curve immediately encounters uncertainty both with the likely crack velocity, which relies on a measurement of the crack length, and the stress intensity at the crack front. It should be clarified at this point that we are only considering the morphology of the crack at the mesoscale (i.e. from the grain scale to macroscale). We are not considering the behaviour and character of individual grain boundaries. These considerations fit into the hierarchy of crack propagation behaviour at a scale below (i.e. the grain to nanoscale, which has received significant attention elsewhere; Engelberg, Marrow, Newman, & Babout, 2008; King et al., 2008). However, further investigations are certainly required to see how nanoscale and grain-scale phenomena link up to the mesoscale.

Figure 10:

Crack opening displacement (COD) (in microns as shown in the legend) mapped as a cumulative total over the three primary AA7032 cracks from the CT image.

Although the COD profile is surprisingly even across the primary cracks, branch-style variations can are evident.

Nonspatially resolved methods for measuring crack length (e.g. potential-drop or back-face strain-gauge monitoring) are open to misinterpretation when it comes to assessing the role of crack branching in SCC. Similarly, measurements of the crack length from the outer surface of the specimen offer limited insight. The presence (or not) of branching as measured at the outer surface is no guarantee of the behaviour through the thickness of the specimen and should be treated more as an indication. Fracture surface analysis is limited to postmortem investigations but does offer some insight to events that occurred across the entire fracture plane. However, interpretation is not always straightforward because, when the sample is broken apart for analysis, parts of the surface may be deformed, remaining bridging ligaments broken, and failed ligaments lost from one fracture surface or both. However, fracture face matching can add some certainty to this analysis.

There are severe shortcomings in most methods of assessment, as they do not allow a full appreciation of the real 3D morphology of the crack. Abramson, Evans, and Parkins (1985) showed some early indications of the true 3D morphology of the stress corrosion cracks by sectioning the cracks in different directions. Other more recent studies (Singh et al., 2014; Zhu et al., 2014) have begun to discover the true morphology of stress corrosion cracks using X-ray CT and, in particular, proving what appear as crack jumps as observed from the outer surface of the samples are shown to be continuous cracks internally. Thus, it is shown that crack jumps also depend on crack branching. We have shown using 3D analysis, as suggested by Blain, Masounave, and Dickson (1984), that the stress corrosion crack is composed of a “multitude of closely spaced parallel intergranular cracks”. The presence of secondary cracks is the clearest example of this, but also, when we look at the crack front, we can observe that it is split into multiple, closely spaced parallel fingers. The crack morphology revealed by 3D analysis shows features that can be easily overlooked when only analysing the fracture surface [i.e. (1) the presence of metal bridges and (2) the stepped topography of the fracture surface]. The presence of metal bridges is a direct result of crack branching and the emergence of the crack onto different, closely spaced grain boundaries. Close to the crack front, the metal bridges actually separate distinct crack fronts, but as the crack progresses these metal bridges are ruptured and the cracks coalesce. Remnants of ruptured metal bridges can be found all across the fracture surface back to the crack origin (Figures 6 and 9). On a fracture surface, it is the more substantial metal bridges that can most easily be observed, as they are deformed out of plane as the stress corrosion test specimen is broken apart manually by overloading. In fact, these regions can often be peeled back to reveal what would have been a secondary crack path. The metal bridges that were ruptured during SCC are more difficult to observe, as they are often smaller and remain relatively flat to the fracture surface. Figure 9 shows broken and unbroken metal bridges. Lynch (2007) previously suggested that such metal bridges and their rupture could play a role in controlling the crack velocity and the discontinuous nature of SCC. The stepped topography of the fracture surface is more subtle. In a sample of any reasonable size, the stress corrosion crack must traverse many grains, and these grain boundaries do not directly follow on one from another in a continuous plane and local deviations will occur. This was a view also previously highlighted by Rhodes and Radon (1978) but one that we can now directly observe. The crack front splits into closely spaced parallel planes with a separation of only a few grains between the parallel planes. Figure 10 shows that the crack front is not linear, adding uncertainty to non-3D assessments of crack length, and also that there are irregularities in the crack front showing regions of arrest. Close to the front yet still behind the crack front, several uncracked regions can be seen as islands of zero COD. Interestingly, the total cumulative COD of the entire crack gives the appearance of a single crack with a relatively smooth COD gradient from origin to front. This morphology is preserved even once the metal bridges are ruptured and is a characteristic feature that is often difficult to observe with fractographic analysis.

Considerable variability is encountered in v-K measurements (Blain et al., 1984; Holroyd, Scamans, & Hermann, 1984) and the measurement of the crack length and morphology from the outer surface of the specimen can often be unrepresentative of the average properties of the crack. Blain et al. (1984) used fractography to characterise the extent of crack branching, and whilst recognising that such analysis is limited to postmortem evaluation, they were able to measure the amount of branching across the whole through thickness of the crack.

Examining the topography of the entire fracture surface shows this stepped morphology from crack origin to crack tip means that this is a consistent behaviour for the propagation of the crack. Thus, the stress corrosion crack is really a colony of cracks that are nearly coplanar and constantly branching and coalescing. Le Poulain, Touzet, Puiggali, and Aubert (2005) have shown that colonies of cracks interact with each other in a way that either amplifies or shields the stress intensity at the crack tips. As they explained, this is a complex 3D problem, but two points that emerge, which have direct implications on the crack front, are (1) “stacked” cracks (i.e. cracks of close proximity but on parallel planes result in a shielding effect and a reduction of the stress intensity at the crack tip). This effect is described as crack blunting effect by Rhodes and Radon (1978), although this description should be only used to describe a single crack with a larger radius of curvature at the crack tip. (2) When coplanar cracks meet each other, they move around each other with the result that the cracks overlap. Thus, the overlapping cracks thus become a finite region of stacked cracks and therefore result in crack shielding. This we see clearly in Figure 9b.

We have characterised the 3D morphology of several stress corrosion cracks of different high-strength aluminium alloys and seen that crack branching is an essential part of their constitution. This should not be overlooked, as it is clear from the many papers that have explored this phenomena that the result of crack branching is that the effective stress intensity acting at the crack tip is significantly less than the applied stress intensity as calculated from global measurements when assuming a single crack (e.g. Austen, Brook, & West, 1976; Blain et al., 1984; Noronha & Packman, 1978). The reduction of K^applied to K^effective, due to the branches and metal bridges, can explain instances where the apparent stress intensity can be measured at values above the fracture toughness.

The current descriptions of crack branching, which describe microbranching and macrobranching, are too limited and we have attempted to create more rigorous descriptions that include nanobranching and acknowledge crack branching during fast fracture as a distinct further type of crack branching. The new descriptions are as follows.

Nanobranching: This new definition is introduced to describe minute forays of crack branches that typically only propagate away from the main propagation direction and do not reach a parallel grain boundary to continue propagating forward. The lengths of the branches are typically one grain diameter or less and can be in any direction but will be along a grain boundary, twin boundary, or slip band. They are most often observed going into the fracture surface. They are generally too small to observe microscale X-ray CT but can be seen using SEM or TEM (e.g. Lozano-Perez et al., 2009).

Microbranching: This is geometrically constrained crack branching where the branches propagate close to parallel with each other at a separation of typically no more than tens of grains. The constraint is supplied to a greater or lesser extent by the stress localisation and/or the microstructure. The emphasis in the description of microbranching should not be on the scale but on the degree of divergence. As dictated by a strong texture in the microstructure, microbranching leads to independent cracks propagating in parallel but with limited separation. We have found that this may be tens of grain diameters, but certainly the separation is limited.

Macrobranching: This is geometrically unconstrained crack branching that often leads to a divergence of the multiple crack fronts to a degree that appears only limited by the number of grain boundaries that the crack has propagated past i.e. opportunities for divergence, when considering intergranular cracking. Again, the emphasis in the description is really on the divergence of the macrobranching, which is not restricted by the microstructure in the same way as microbranching. This can be realised either through an equiaxed microstructure of limited texture or due to the SCC mechanism being fully or partially transgranular in nature instead of intergranular. It is likely to be enhanced by residual stress.

The region II plateau in the v-K curve results from a balance of the driving forces for the propagation of the SCC crack and the drag associated with the crack front multiplicity and the metal bridges. We believe that the result is that K is effectively self-regulating in region II. As described by Kitagawa et al. (1975) and Nakasa et al. (1979), the stress intensity drops when the crack branches but, as one of the branches proceeds and leaves the other crack behind, the stress intensity begins to rise again. This process is constantly occurring and is averaged out across the entire plane of the crack. There are no direct measurements of this to date, but the fact that the spacing of crack arrest markings appear to be relatively insensitive of K (Holroyd & Scamans, 2011; Scamans, 1982) directly supports this proposal.

5 Conclusions

Our work demonstrates that it is essential to use 3D imaging to understand the morphology of stress corrosion cracks. It is only when the crack morphology is understood in 3D that we can effectively link the properties recorded on macroscale test specimens to the real conditions experienced at the crack tip. It is these real conditions that we need to understand before we can confidently predict or describe the mechanisms taking place at the crack tip.

Crack branching is complicated, and because it can have significant effects on crack bridging and the COD, quantifying K^effective is a major challenge.

The region II plateau is the result of the crack branching phenomena. In region II, the K^applied is reduced by the presence of what is a colony of cracks that are close to coplanar and the presence of the resultant metal bridges. Indeed, region II may only exist due to a self-regulating K that balances the driving forces of crack propagation against the retarding forces of crack multiplicity and the presence of metal bridges. It is possible that, in the future, time-lapse CT imaging of the branching phenomena while collecting v-K data could help to better explain SCC behaviour and provide a means of relating K^applied to a more fundamental understanding through a measure of K^effective or some other fracture mechanics description of the crack driving force experienced at the crack front.