Transverse fatigue behaviour and residual stress analyses of double sided FSW aluminium alloy joints

3.1 Fatigue tests

The experimental analysis of fatigue testing was conducted using an ElectroPuls E10000 dynamic test instrument provided by Instron. This test rig is able to achieve high loading frequency (up to 100 Hz) at loads as high as 10 kN. Nevertheless, for the purpose of the present study, only the HCF regime was considered; therefore, the frequency was limited to 25 Hz. Such constrain was imposed in order to reduce the effect of thermal heating, which may have affect the outcome of the tests.

A preliminary tensile monotonic test allowed for determining both the ultimate tensile stress (σ_UTS) and yield stress (σ_Y), which were assessed, respectively, to be 180 and 152 MPa. It is worth reporting that the assessment of the yield stress was done by assuming a 0.2% plastic strain offset. Therefore, the fatigue tests were performed at a maximum load being lower than the σ_Y. In the present campaign of fatigue tests, the samples were cyclically loaded at load ratio R = 0.1. This choice was mainly aimed at reducing the effect of plasticity induced crack closure taking place while the crack propagates through the welded region. By decreasing the effect of crack closure, the crack surface shows a less amount of plastic deformation generated by the contact between crack flanks, which in turns allows improved optical inspection of the crack surface morphology. Maximum cyclic stresses σ_max were thereby chosen to range between 80 and 140 MPa; which corresponds to stress ranges ∆σ of 72 to 126 MPa. Seven samples were employed in total for the fatigue tests.

3.2 FIB-DIC ring-core and synchrotron based XRD for the evaluation of residual stress

RS in materials can be evaluated through a number of experimental techniques ranging from destructive to non-destructive. However, only few of them provide absolute estimation of RS values independently from the material microstructural characteristics; these are based on the concept of stress relaxation. Stress relaxation takes place when a cut (removal of material) is performed in the material, which in turn it gives rise to deformation of the surrounding material due to the pre-existence of RS. Based on this concept, several techniques have been developed in the last decades,28-31 and more recently, it also has been implemented at the nano- and micron-scales.32-35 One of the two methods employed in this study is based on such a physical phenomenon, and it is applied at the micron-scale. Such a method is also known as FIB-DIC micro-ring-core technique.26, 27, 36, 37 In this acronym, FIB stands for focused ion beam, which is the technique utilised for the material removal at such a small scale, and DIC is the digital image correlation technique adopted to track and to measure the surface feature displacements occurring on the surface of the “core” while the milling proceeds. Such core is the pillar-like volume of material left after the machining is completed. Latest developments of this method have allowed the evaluation of depth profiling of RS within coatings or treated surface with depth resolution up to 50 nm. However, as concerns the present study, since depth profiling was not required but rather an averaged local value, a more classic strategy was used for the back calculation of RS from the measured strain relief provided by DIC analysis. This calculation framework consists in the fitting of the relief curve with a master curve obtained by FEM simulation, assuming the RS is constant throughout the whole depth reached by the ring-core milling. It is assumed that the RS depth profile is constant at each local measurement. This assumption is realistic seeing as the total ring-core milling depth is only 5 μm. It is important to highlight that this calculation accounts for the presence of several sources of errors, ie, DIC tracking and master curve fitting,38 elastic local anisotropy,39 and FIB-induced amorphisation.40-42 Such a propagation of errors will be illustrated by using error bars for each local measurement. It is worth stressing that in comparison with the SXRD technique, FIB-DIC relies on the assessment of RS within a much smaller gauge volume. It is therefore challenging to estimate the RS variability within the workpiece induced by the manufacturing technique, ie, sum of physical errors.

One of the limitations of the FIB-DIC method is the rather prohibitive possibility of obtaining full-field maps of RS over large areas, given the relevant time required for each measurement point (about 20 minutes each). Since in this study one of the purposes was to find correlations between the presence of the RS and fatigue crack nucleation and propagation, it was considered of great importance to consider a full picture of RS present in the weld. Moreover, SXRD method provided an unparalleled amount of statistical information, which allowed for the assessment of RS variability throughout the sample thickness.

To achieve this, SXRD was used on the B16 beamline at the diamond light source (DLS) facility in Harwell (United Kingdom). The 2-mm-thick cross-section sample was scanned using 18 keV X-ray beam in transmission mode. The region of interest (ie, area including the nugget, HAZ, and TMAZ) was mapped by raster scanning strategy, using a step size of 0.25 mm was chosen. At each scanned point, full Debye-Scherrer patterns were acquired. The quantification of the <311> peak shifting enabled the evaluation of the lattice strain at each point regarding the two perpendicular components, ie, transverse and normal. Upon this assessment, the calculation of the RS was quite straightforward, and it was conducted by invoking the bi-axial form of the Hooke's law. Nevertheless, one of the main issues usually faced when evaluating RS with this class of techniques is the unknown value of the unstrained lattice. Conversely from other authors found in the pertinent literature,43, 44 where a comb-like sample was used for the determination of the unstrained lattice value, in this study, a different approach is employed. This approach relies on the fact that in order to obtain body equilibrium in a beam, the sum of the transverse forces and moments must be zero, along any normal section of the beam. In the present case, for the sake of simplicity only the forces were accounted and therefore the condition assumed was

where σ_xx is the transverse stress according to Figure 1, t is the sample thickness, and y is the normal direction. This method allowed for the independent full-field 2D evaluation of RS. Error estimation was conducted essentially by evaluating the accurateness of the peak fitting, in the first place, and the confidence bounds were assessed by appropriately invoking Bragg's law.

3.3 Optical, SEM, and EBSD for microstructural investigation

The fatigue mechanical testing was followed by detailed inspection of the fracture surface to investigate the cracking mechanisms operative during propagation. Firstly, optical examination was carried out by using an Alicona Infinite Focus optical microscope, which provided 2-mm lateral resolution and accurate focus ensured by the out-of-plane scan reconstruction implemented in its dedicated software. As an outcome, the 3D reconstruction of the crack surface morphology was recreated. For this analysis, two representative fatigued samples were studied.

A representative fractured sample was then examined using SEM microscopy. Detailed fractographic analysis was performed over specific regions of the sample, for instance: initiation point, steady fatigue crack growth region, and monotonic failure.

The microstructural characteristics of the material in the vicinity of the fracture were studied by invoking the EBSD method. One end of a representative fractured sample was used for this task. The analysed surface can be identified as the xy-plane depicted in Figure 1. Such a surface was progressively polished using several grades of sand papers, followed by diamond suspensions up to 1 μm and colloidal silica solution as a last step. Besides the region close to the fracture, thanks to the large size of the map, it was also possible to appreciate the gradient of microstructure size and texture degree across the three dominant regions identified in FSW joints, namely, nugget, HAZ, and TMAZ. The map true pixel size was set to 4 μm.

4.1 Fatigue tests

The results of the fatigue test campaign are summarised in the Wohler diagram in Figure 2. As it is possible to see, the sample fatigued at 72 MPa of stress range experienced fatigue life greater than the cut-off threshold, above which the test is considered as a run-out. On the other hand, other samples tested at different stress ranges showed a general trend of load being inversely proportional to the number of cycles, as expected and well-known from fatigue theory. The experimental data are fitted with the classic empirical Wohler curve. It is worth remembering that this manuscript is not devoted to the full fatigue characterisation of these joints but rather at investigating some failure characteristics. For this reason, the number of fatigue tests was limited to only seven. Needless to say that the actual curve requires a larger number of experimental tests to be statistically representative; however, the present experimental results provide an insight on the trend. In addition, the fatigue behaviour of FWS joints of the same material, extracted from the literature, are reported in the same plot,45 indicated as “Ref.”, along with the fitted power curve.

4.2 Microstructure characterisation and fractography

The investigation of the fracture surface was performed on two samples using optical depth-resolved microscopy. The samples loaded at higher and lower fatigue stress ranges, respectively, at 112.5 and 126 MPa are shown in Figure 3. As it is possible to observe from both the contour plots and the 3D reconstruction, the crack followed a path not strictly perpendicular to the loading direction, as expected for homogeneous materials. Thereby, a first conclusion is that the crack path was strongly influenced by the underlying microstructure and possibly from the RS field for all the samples and fatigue load amplitudes. For all the cases studies, macroscopic features were observed. These features can be seen as marks that are not associated with the direction of crack propagation, as will be shown later, but rather linked to the tool advance, as also observed in the literature46; Figure 3A shows the location of these marks and are indicated as “tool rotation marks.”

A large EBSD map over the regions of interest was acquired and then overlaid onto a secondary electron image obtained from SEM, shown in Figure 4. This maps clearly enables the visualization of the crack path and how this is linked to the microstructure and texture. Two main regions could be identified in the map, the first belonging to the TMAZ on the left-hand side and the second showing a transition to the unaffected microstructure on the right-hand side. A further region can be recognised at the top and bottom sides of the weld, in correspondence of the tool contact region where a prominent grain refinement took place, forming a shallow layer which is named as tool's shoulder layer.

A more detailed inspection of the fractured surface was done by SEM, and the results are shown in Figures 5 and 6. The overview image presented in Figure 5 not only highlights again the tool advance macroscopic marks but also provides enough resolution to identify the typical fatigue striation, which are directly associated to the length of crack propagation at each load cycle. With the purpose of facilitating the comprehension of the image, continuous red lines are manually overlapped to enhance visualisation of the crack front at some stages of propagation. In addition, orange arrows indicate the direction of propagation.

It is important to identify features of this joint when compared with other types of FSW. For instance, despite the nature of the double sided FSW,45 the effect of the tool's shoulder is only predominant on the side where crack nucleation occurred, making the single side type of joint not much different from this point of view. On the contrary, if the monotonic behaviour of the joint was studied, we would experience a difference because of the additional tool's shoulder layer being present at the bottom part (visible in Figure 4). In general, the literature reports an improved performance of double sided FSW, especially in terms of strength.47-49 This effect is thought to be related to the accessibility of the material from both sides, which makes the processed material more homogeneous. Another aspect that may be worth elucidating is the crack path, which differs from what is observed for single sided FSW. In fact, in contrast with the double sided joint seen from the Figure 4, the single side joint presents a V-groove usually extending all the way through the thickness. It can be expected that the crack path would not differ from the traditional single side joint for one third of the crack extent. Instead, as the effect of the double side becomes more and more prominent, the crack path diverts assuming a more characteristic shape resembling the groove (or the TMAZ).

Another visible characteristic is that all the nucleation of cracks took place within the region affected by the tool's shoulder contact. This region can be identified only by observing the fractured surface, as visible at the crack nucleation site in Figure 3, or in more details in the Figure 6A,C. After detailed imaging inspection, it turned out that this can be considered as a semi-detached layer, about 100-um thick. It is indeed at this outer layer where crack first originated, most probably facilitated by the pre-existence of a large defect, as underlined in Figure 6A. It is also important to remark that the interface between the shoulder layer and the core material is rather discontinuous, making the fatigue crack propagating in different manners. In fact, Figure 6C illustrates how at the right-hand side of the specimen the crack propagated in fatigue mode in the core, but the outer layer showed ductile fracture, imputed to monotonic load failure.

4.3 Residual stress evaluation and FEM stress analysis

As explained earlier, two experimental techniques were employed for the study of RS across the weld regions. The analysis of the acquired diffraction patterns, quantification of lattice strain first, and calculation of RS second, allowed for the reconstruction of the RS maps reported in Figure 7A,B. The lack of detailed information at the very top and bottom parts of the maps was attributed to free-edge effects, being this not perfectly even and producing artefacts and therefore unrealistic values of stress that were ultimately discarded.

Although it is generally very challenging to extract quantitative evaluation of the plastic strain present in the weld, the analysis of peak broadening effect may sometimes provide some insights.50 With this end, a peak broadening analysis was carried out, using the same crystal plane (<311>) previously employed for the lattice strain analysis. The resulting contour map is shown in Figure 7C.

To corroborate the results from SXRD, a line of spatially resolved FIB-DIC ring-core measurements was performed across the FSW section. The line is shown by the black arrows in Figure 7A,B, and the x-axis shares the same origin with the SXRD contour plots. For the sake of comparison, RS line profiles were extracted from the SXRD maps at the same locations and plotted in Figure 8. It is worth noting that the error bars concerning the yy component of RS are significantly larger than corresponding xx ones. This is due to the SEM fast scan direction, being in this specific case, corresponding to the x-direction. In any case, the magnitude of error bars is still tolerable.

The sample selected for the study of RS was also imaged by optical microscopy with the purpose of generating a geometrical model. A 2D model of the sample cross section was then imported in an FEM environment and the uniaxial loading condition simulated, neglecting material, and geometrical nonlinearities. After mesh refinement at the high local curvature regions and appropriate implementation of load and boundary conditions, it was possible to estimate the nominal SCF. A contour plot of the SCF is reported in Figure 9.