Engineering Fracture Mechanics 76 (2009) 134148

Contents lists available at ScienceDirect

Engineering Fracture Mechanics

journal homepage: www.elsevier.com/locate/engfracmech

Fatigue life predictions using fracture mechanics methods

T. Ghidini a,*,1, C. Dalle Donne b
a b

ESA, European Space Agency, Product Assurance and Safety Department, Materials and Processes Division, P.O. Box 299, NL-2200AG Noordwijk, The Netherlands EADS Deutschland GmbH, Innovation Works, Munich, Germany

a r t i c l e

i n f o

a b s t r a c t
In the present work, a simple engineering approach which is based on a relatively solid background and which is checked against fatigue test data for various test conditions was developed: it may provide a practical and reliable basis for the analysis of structures under in-service loading conditions, in the presence of previous corrosion attack, or in the presence of a residual stress eld, by using widespread fracture mechanics software. In particular, the approach was checked against an experimental program which consists of the following fatigue tests: base and friction stir welded (FSW) material under constant amplitude loading at different loading ratios (R = 0.1, 0.5, 1); pre-corroded base and FSW material under constant amplitude loading at load ratio R = 0.1; centre hole FSW specimens under the standardised variable amplitude loading spectrum FALSTAFF. Moreover, from the literature fatigue experiments under FALSTAFF of cold expanded as well as not cold expended holes were also used to validate the approach. The predictions were performed with the last version of AFGROW and NASGRO 3.0 software. 2008 Elsevier Ltd. All rights reserved.

Article history: Available online 5 August 2008 This paper is dedicated to the memory of Arij de Koning who passed away the 6th of February 2007. Keywords: Fracture mechanics Fatigue life prediction Pre-corrosion Friction stir welding Variable amplitude loading Cold expansion AFGROW NASGRO

1. Introduction In general, structures contain micro structural defects such as porosity, voids, discontinuities which can lead to the formation of cracks if the service loading exceed a certain level. Once a crack is present, it may grow in a stable manner and after a certain time results in an unstable crack growth and eventually in the ultimate failure of the structure. Fatigue is one of the most difcult, and insidious, design issues to resolve. Experience has shown that the majority of structural failures occur as a result of fatigue: the percentage of such failures in mechanical components is set around the order of 90%. In the aerospace industry, the introduction of damage tolerance requirements has been a milestone in the design of fatigue resistant structures and it was possible only thanks to the level of maturity that linear elastic fracture mechanics has reached. Many crack growth prediction models have been proposed and used by the industrial and scientic community [1,2]. The calculations of this work were performed with AFGROW [3], ESACRACK 4 [4], and NASGRO 3.0 [5]. More recent versions, NASGRO 5.2, exist, NASGRO 3.0 was the version available at the time the work has been performed. AFGROW is a computer program developed by the US Air Force and is available on line. ESACRACK 4 was written by ESA and NASA and NASGRO 3.0 was created by NASA. Since NASGRO 3.0 is also part of the ESACRACK 4 software package, in the following just AFGROW and NASGRO 3.0 will be compared. Both software need the same initial data to perform the crack growth analysis: the crack geometry, the loading conditions, the da/dN DK crack propagation curve of the material as well as static and fracture toughness properties (which can be found in a large database of the software or can be implemented by the user). The programs then use implemented K-factors solutions and crack growth concepts to propagate the crack until failure occurs. During the last three
* Corresponding author. E-mail address: tommaso.ghidini@esa.int (T. Ghidini). 1 The work has been performed during the authors stay at the Institute of Materials Research of the DLR in Cologne, Germany. 0013-7944/$ - see front matter 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.engfracmech.2008.07.008

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148

135

decades considerable work has been devoted to understand the fatigue crack propagation under constant amplitude loading. However, the basic problem is to predict fatigue crack growth in real structures of complex geometry under in-service loading conditions. Fatigue loads in-service generally imply a randomly variable amplitude, rather than constant amplitude loading. Different types of load sequences are known to induce a number of different load interaction effects [6,7], which can result in signicant variation on the crack growth rate. Moreover, it is well known that corrosion unfavourably affects the structural integrity since fatigue cracks nucleate from corrosion pits [8,9], drastically reducing the fatigue life of the component. The computation effort is more complex for structures which are affected from the presence of a residual stress eld. Residual stresses are, in many cases, an undesired consequence of the manufacturing and joining technologies adopted (welding, cold forming) or, in other cases, are intentionally produced by means of proper techniques (e.g. shot peening, cold working) with the aim of improving strength. All these effects should be included in the predictions as they can affect structural health of structures. For practical applications, industrial designers require reliable prediction models, easy to handle, without too many empirical constants (derived from specic tests) and, lastly, capable of running in a reasonably short time on personal computers. In previous works the authors have successfully demonstrated the capability of the previously mentioned software to predict the crack growth life of pristine and welded specimens, both under constant and amplitude loading spectra [1012]. The aim of the present work is to demonstrate the possibility of applying fracture mechanics methods to perform fatigue life predictions. The calculations were compared with experimental results. The following experimental test program was carried out:  Constant amplitude loading fatigue tests of base and FSW material of un-notched specimens at different loading ratios (R = 0.1, 0.5, 1).  Constant amplitude loading tests of pre-corroded base and FSW material at load ratio R = 0.1.  Variable amplitude loading fatigue tests of centre holed FSW specimens under FALSTAFF.  Variable amplitude loading fatigue tests of centre holed cold expanded specimens under FALSTAFF. In predicting the fatigue life of base and friction stir welded structures a basic assumption was done: the crack is starting from constituent particles and particle clusters; moreover, in order to apply to fatigue the fracture mechanics statements, it was also assumed that the cracks formed immediately at the particles and the entire fatigue life was comprised of crack propagation. The pre-corroded base as well as friction stir welded materials were simulated by identifying and quantifying the widespread corrosion damage by many metallographic sections. A model was then developed which creates a single surface crack having the deepest and largest corrosion attack as dimension. In the model no distinction was done between the two observed corrosion types, i. e. pitting and inter-granular corrosion. In predicting the fatigue life of the centre holed FSW specimens under FALSTAFF the same assumptions regarding the starting cracks as the un-notched specimens were adopted, and the load sequence effects were taken into account by using the Willenborg retardation model [13]. Because of the small size of the specimens, the very low transverse residual stresses were not considered. In the calculations for the cold expanded holes, since the residual stresses were too high to be neglected, the Willenborg model was used in combination with a residual stress prole in specimen thickness as well as width direction obtained by tting the X-ray diffraction measurements. 2. Experimental program 2.1. Friction stir welding and specimen geometry Four millimeter thick AA 2024-T3 material was provided from Pechiney, in the form of sheets. Table 1 gives a survey of the chemical and mechanical properties of the alloy investigated in the experimental procedure. The FSW butt joints were produced at the DLR, the German Aerospace Centre following the TWI patent [14]. All the welds were performed parallel to the rolling direction of the sheets. The material was welded in T3 condition and no additional heat treatment was carried out after the welding procedure. The welding speed was 300 mm per minute, the rotating speed of the tool was 850 revolutions per minute and the tilt angle was kept constant at 0. The welding parameters used to produce the joints are summarised in

Table 1 Chemical composition and mechanical properties of the alloys investigated Alloy Si Fe Cu 3.84.9 Mn 0.30.9 Rm (MPa) 476 Mg 1.21.8 Cr 0.1 E 72000 Zn 0.25 Ti 0.15 Zr

Chemical composition of the alloys in wt% AA 2024-T3 0.50 0.50 Material Mechanical properties AA2024-T3 Rp0 (MPa) 329

Elongation (%) 25.7

136

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148

Table 2 Welding parameters and static properties of AA2024-T3 T(L) Alloy AA 2024-T3 Rotational speed (rpm) 850 Weld speed (mm/min) 300 Rp0.2 (MPa) 316 Rm (MPa) 460 RmRm, 96.6
Base

(%)

4 1= DXZ 2= TMAZ 3= HAZ 4= Base Material

160

Base Material
140

HV 1

120

100

AA 2024-T3 FSW
80 -25 -20 -15 -10 -5 0 5 Position [mm] 10 15 20 25

Fig. 1. Hardness prole of the AA 2024-T3 friction stir welded material.

Table 2, where a comparison between the ultimate strength of the FSW and the base material is also given. The hardness prole of the butt joints is presented in Fig. 1: all the measurements were taken in the midsection of the FSW joints. The hardness of AA 2024-T3 base metal is also presented in the diagram of Fig. 1. The diagram shows the same general trend which has been observed in previous works, [15,16]: the hardness distribution results in an M like form with two minima almost 5 mm out of the centre of the weld. The typical different areas of a FSW weld, described in detail in [17], can be identied here. Parent plates, dash bands in the diagram, hardness values were at more than 10 mm from the centre of the weld. The 0 position in the diagram corresponds to the centre of the weld nugget. The hardness distribution in the joints was consistent with the tensile strength. The centre hole friction stir welding fatigue specimens used for the variable amplitude loading testing, Fig. 2, were machined taking the hardness prole into account. The hole was machined in the weakest area of the sample, as shown in [15,16]: The centre of the weld was placed in the middle of the coupon, and the hole was produced 5 mm out of the weld centre in the TMAZ of the retreating side, in correspondence of the hardness minimum. Particular care was devoted to the realisation of the hole, to avoid the presence of residual stress due to the machining process. Un-notched base material and friction stir welding fatigue specimens were machined for the constant amplitude loading part of the testing activity, Fig. 3. All the coupons used for the experimental program, both open-hole and unnotched, were machined in TL orientation. The loading axis was perpendicular to the rolling direction and to the welding direction. The weld nugget was in the middle of the specimen.

5
FSW Weld

12.5 12.5

110

110

Fig. 2. Dimensions of the centre hole FSW fatigue specimens used for testing under FALSTAFF.

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148

137

FSW Weld
12.5

80 R50 220

19

51

Fig. 3. Dimensions of the un-notched FSW fatigue specimens.

2.2. Corrosion treatments Prior to fatigue testing, the 2024-T3 un-notched test samples were corroded with an alternate immersion technique in accordance with ASTM G44-99 standard [18]. All specimens were prepared by manual polishing and by covering the edges with a protective mask (TURCO 5580G), leaving just the centre of the specimen exposed to the corrosion attack, Fig. 4. This was done to avoid any edge effect and to analyse only the propagation of cracks starting from the corroded surfaces. The specimens were cyclic immerse during 10 min in a 3.5% NaCl aqueous solution and they remained 50 min in air (1 h per cycle). 100, 250 and 1000 h were the corrosion times chosen in this research. Air temperature and relative humidity were kept constant (27 C and 45%). After the corrosion procedure the protective TURCO masking was removed. The specimens were fatigue tested without any further treatments. On the other hand, the specimens used for metallographic characterisation of the corrosion attack were cleaned in 50% HNO3 aqueous solution for 15 min, in order to eliminate the residual corrosion products. The characterisation of the corrosion attack on the 2024-T3 base metal and the friction stir welded joints was carried out using light microscopy techniques. In the case of the base metals, depth and width of the corrosion attack of about 200 corroded areas (pitting and intergranular corrosion) were measured for the different exposure times. For the pre-corroded FSW-joints consecutive metallographic preparation was necessary to depict the corrosion attack on the different zones of the weld joint (i.e. weld nugget, HAZ, etc). The FSW specimens were polished and metallographic sections were prepared parallel to weld nugget each 0.5 mm beginning at the centre of the weld nugget and nishing when the unaffected base metal was reached (about 12 mm from the joint centre line). 2.3. Constant amplitude loading The entire SN fatigue testing was conducted using an Amsler resonance machine of 100 kN capability equipped with a TestExpert control software. All the fatigue experiments were performed under constant amplitude loading and 60 Hz at room temperature (21 + 2 C). The 2024-T3 base material specimens were tested in lab environment at load ratio Fmin/Fmax of R = 0.1, R = 0.5 and R = 1 in order to obtain reference curves to check the fracture mechanics software. The diagram reported in Fig. 5 shows the results of the fatigue life tests of 2024-T3 base material and the effect of the load ratio R: decreasing the value of R will result in a shorter fatigue life, especially when the results are plotted in function of the maximum applied stress. The same type of un-notched specimens of 2024-T3 base material and FSW joints were pre-corroded by alternate immersion in 3.5% NaCl solution and they were subsequently tested under constant amplitude loading and lab environment at load ratio Fmin/Fmax of R = 0.1 (loading direction perpendicular to the weld).

Loading Direction

A
Fig. 4. (a) Schematic of a corroded FSW specimen, and (b) appearance of the pits on the exposed surface.

20

138

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148

400 350 Maximum Stress [MPa] 300 250 200 150 100 50 10
4

R=0.1 R=0.5 R=-1

AA 2024-T3 T(L) t=4mm

60 Hz Laboratory air

10

10 Cycles

10

Fig. 5. Comparison between fatigue life of base material 2024-T3 specimens under constant amplitude loading at load ratio Fmin/Fmax of R = 0.1, R = 0.5 and R = 1.

400 350 Maximum Stress [MPa] 300 250 200 150 100 50 10
4

Base Material Not-Corroded FSW Material Not-Corroded Base Material 100h Corroded Base Material 250h Corroded Base Material 1000h Corroded FSW Material 100h Corroded FSW Material 250h Corroded FSW Material 1000h Corroded

AA 2024-T3 t=4mm R=0.1

60 Hz Laboratory air

10

10

10

Cycles
Fig. 6. Comparison between not corroded and 100, 250 and 1000 h corroded polished base and FSW material.

The diagram in Fig. 6 shows the effect of corrosion time on the fatigue life of 2024-T3 friction stir welded joints and base material. Also a comparison between base and FSW not corroded material is given. In general, no signicant difference in fatigue life between the 2024-T3 base metal and FSW-joints can be observed. Most of the un-corroded FSW-specimens failed either in the unaffected base metal (62%) or at the advancing side of the weld nugget (12.5%). Generally, the origin of the failure was found to be a cluster of the second phases present in the microstructure of the 2024-T3 aluminium alloy. The base material has a shorter fatigue life as the FSW material at least on the threshold region: At rst glance, this could be contradictory to the normal fatigue behaviour of FSW structures [1921]: a preliminary explanation of the experimental observation could be that the surfaces of the sheets have been milled before to produce the weld. This could results in a smoother surface, with smaller clusters, and therefore in a better polishing procedure which leads to better fatigue properties. This effect disappears when considering the corroded specimens where practically no difference exists between the pristine and the FSW specimens. In fact the fracture mechanism is now governed by the corrosion pits, whose dimensions are bigger than the clusters and the constituent particles. Moreover, referring to Fig. 6, corrosion has obviously a detrimental effect producing a dramatic drop of the fatigue life of both types of specimens (base metal and friction stir welded joints). The FSW joints show a similar drop in fatigue strength as the base material, which seems to be virtually independent of the corrosion time, also due to the logarithmic scale of the plot. Although the test results of 100, 250 and 1000 h pre-corrosion specimens fall in the same scatter band, the trend is in general consistent (increase exposure leads to reduced fatigue life).

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148

139

450
FALSTAFF

400 Maximum Stress [MPa]

FSW Weld

350 AA 2024-T3 FSW t=4 mm

30 Hz Laboratory air

300

250

200 2 10

10

10 Flights

10

Fig. 7. Experimental results of the centre hole FSW specimen under FALSTAFF spectrum loading.

2.4. Variable amplitude loading The centre hole FSW specimens were tested under a standardised ight simulation load history, FALSTAFF [22]. FALSTAFF is a standard load sequence considered representative for the load time history in the lower wing skin near the wing root of a ghter aircraft. The different specimens were tested at different values of rmax. Fatigue testing under ight simulation loading were carried out on a servo-hydraulic machine of 400 kN capability controlled by Fast Track software. The specimens were enclosed in anti-buckling guides to allow compressive loads during the tests. The tests were carried out in a laboratory air environment at room temperature and at a frequency of 30 Hz. In Fig. 7, the experimental results are plotted in a diagram showing the different levels of rmax plotted against the number of ights at failure. In order to reduce testing time, the tests were carried out at high stress levels. 3. Experimental program from the literature 3.1. Cold expanded holes A very well documented experimental program was taken from the open literature [23] in order to check the developed method under variable amplitude loading and in the presence of a pre-existent residual stress eld. Low-load transfer joint specimens and open-hole specimens of similar geometry were machined of aluminium alloy 2024-T351. The description of the entire test program is given in [23], for brevity purpose only the open-hole testing activity will be presented here. The general dimension of the specimens were 6.35 mm hole diameter, 6.0 mm thickness and 25 mm width. The test program includes non-cold expanded (NCx or plane hole) tests and cold expansion on production (Cx, or 0% Cx). The FTI split sleeve cold expansion method was used. The expansion process consists of pulling an oversized tapered mandrel through the hole, causing extremely high radial pressures on the hole, thus expanding the hole well behind the yield strength of the material. This procedure produces a zone of residual compressive stress that extends approximately one radius from the edge of the hole. The specimens were expanded to a nominal rate of 4.04.4%, resulting in a retained expansion of 2.82.9%. All the specimens, NCx and Cx, were than fatigue cycled under FALSTAFF spectrum. All tests were conducted with a peak net stress of 300 MPa. Crack lengths were measured through the use of acetate replicas, which were examined with an optical microscope. The crack lengths were tracked on both sides of the specimen as well as within the bores of the holes. 4. Fatigue life predictions 4.1. Base and FSW material In modelling the fatigue life of the base material specimens a basic assumption was done: the crack is starting from constituent particles and particle clusters. It was assumed that the cracks formed immediately at the particles and the entire fatigue life was comprised of crack propagation. Previous work has demonstrated that cracks form primarily at single particles or holes, indicating that coalescence is not an issue, at least in the LS direction of propagation [24,25]. In the literature

140

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148

fatigue cracks were observed to start at slip lines around large inclusion clusters containing iron and silicon in 2024-T3 [26] and the cyclic straining was observed to weak and to deteriorate the particle matrix interface, increasing the susceptibility to debonding [27]. In [28,29] it has been pointed out that cracks formed at inclusions and inclusion clusters in thin sheet 2024T3 aluminium and the nucleation site was observed to quickly form a semi-elliptical surface crack approximately the size of the inclusion. The idea of using fracture mechanics for predicting the fatigue life of components by starting from the microstructural features of the materials was rst proposed in [30]. Fig. 8 shows the semi-elliptic crack modelled over a constituent particle observed in a post-mortem failure surface analysis carried out with the scanning electron microscope. The gure is just a schematic, therefore the presented inclusion does not exactly match the crack front since the reported crack front is a mean value measured of the constituent particles. The microstructure of the base metal used in this study consists in attened grains, the so called pancake shape, which mayor axe was parallel to the rolling direction. The constituent particles were aligned at the grain boundaries forming clusters. The maximal sizes of the second phases were to be 3842 lm. SEMEDX analysis has mainly shown the presence of two kinds of constituent particles containing AlCu and AlCuMg. The measured dimension have been used in AFGROW and NASGRO 3.0 as initial crack dimensions for the implemented model semi-elliptical surface crack, which is shown in Fig. 9. The model implemented in both codes utilises a stress intensity factor solution developed by Newman and Raju. The dimensions of the modelled semi-elliptical crack were approximately a = 40 lm and c = 80 lm, referring to the biggest constituent particles also reported in [24,25]: the fatigue strength was assumed to be governed by one critical inclusion and not by the presence of many critical clusters. The assumption was also in agreement with [31]. The dimensions W and t in the model of Fig. 9 are the specimen width and thickness respectively. In modelling the fatigue tests the base material data stored in the NASGRO database was used. The material constants used are those of the AA 2024-T3 orientation T(L). In case the calculated DK was below the propagation threshold, the NASGRO equation (used in AFGROW as well) is taking the small-crack effect into account by allowing to linearly extrapolating the crack growth values below threshold [4]. This approach was adopted in this work. The constants are listed in Table 3. The t obtained by the constants includes small-crack effect [32,33] correction in the near threshold region of the da/dN DK curve. The same dimensions of the inclusion particles were used in predicting the

a 2c

Fig. 8. Schematic of the semi-elliptical surface crack modelled over a constituent particle observed in a post-mortem failure surface analysis carried out with the scanning electron microscope.

2c

Fig. 9. Semi-elliptical surface crack model implemented in AFGROW and in NASGRO 3.0.

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148 Table 3 Material parameters given in the material database of NASGRO for the material AA2024-T3 T(L) UTS (MPa) 448.16 YS (MPa) 330.95 KIC (N/mm3/2) 1007.7

141

DK0 (N/mm3/2)
100.77

C 6.08 1011

n 2.6

p 0.5

q 1.0

a
1.5

400 Maximum Stress [MPa] 350 300 250 200 150 100 50
AA 2024-T3 t=4mm R=0.1
60 Hz Laboratory air

Maximum Stress [MPa]

Experimental Data NASGRO 3.0 AFGROW

400 350 300 250 AA 2024-T3 200 t=4mm 150 100

R=0.5
60 Hz Laboratory air

Experimental Data NASGRO 3.0 AFGROW

10

10

10 Cycles

10

50 4 10

10

10 Cycles

10

400 Maximum Stress [MPa] 350 300 250 200 AA 2024-T3 t=4mm 150 R=-1 100 50
60 Hz Laboratory air

Maximum Stress [MPa]

Experimental Data NASGRO 3.0 AFGROW

400 350 300 250 200 150 100 50

AA 2024-T3 FSW t=4mm R=0.1
60 Hz Laboratory air

Experimental Data NASGRO 3.0 AFGROW

10

10

10 Cycles

10

10

10

10 Cycles

10

Fig. 10. Comparison between base and FSW material experimental results and computer simulations.

life of the FSW joints. Moreover, the observation of the fatigue failed polished FSW specimens demonstrated that all the coupons broke in the base material away from the weld nugget. Neglecting the generally very low transverse residual stresses [34] allows the prediction to use the base material data stored in the NASGRO database also for the friction stir welded material. Therefore, the two predictions referring to base and FSW material are identical. Fig. 10 represents the comparison between base and FSW material experimental data and computer simulations. Both NASGRO 3.0 and AFGROW results are in good agreement with the experimental observation. The small differences between NASGRO 3.0 and AFGROW results are probably due to different integration routines. Both computer programs showed to predict well also the R ratio effect which is implemented in the NASGRO equation even for compressive loads. Concerning the FSW coupons, the difference between the simulated fatigue lives and the experimental ones is higher than in the case of base material ones. Apparently the FSW specimens seem to have a higher fatigue life then the base material (since the simulation was performed with the same basic assumptions, the NASGRO and AFGROW lives are the same in both comparisons). 4.2. Pre-corroded material In all cases (2024-T3 FSW and base metal corroded during 100, 250 and 1000 h), the corrosion type found on the metallographic coupons was a mixture of pitting and inter-granular corrosion (IGC), Fig. 11a. In some cases two or more pits were connected through IGC ramication forming a bigger corroded damaged zone. Seventy-seven percent of the pre-corroded FSW-joint specimens failed at the ne grained region on the advancing side of the weld nugget. At rst glance, this could be contradictory to the corrosion ndings, since it would be then expected that the failure should occur at TMAZ where the greater corrosion damage was found. A detailed study of the fatigued fracture surfaces of different pre-corroded FSWspecimens showed the presence of several pits at the crack initiation site, as it can be seen in Fig. 11b.

142

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148

Fig. 11. Typical corrosion damage zone found in a 250 h pre-corroded 2024-T3 base metal section.

These pits were interconnected by IGC generating a greater corrosion zone, with similar dimensions of the maximal corrosion attack zones found in the TMAZ. Therefore, it is assumed that between defects with similar overall geometry, the crack would start from the material with a greater quantity of corrosion damage areas, in this case, from the ne grained region of the pre-corroded FSW-specimens. In modelling the fatigue tests of the pre-corroded base material and friction stir welded specimens some basic assumptions and simplications were done in order to create a general approach to the corrosion-fatigue of welded structures. On the basis of the results published in [20] the effect of the different weld microstructures on da/ dN DK was neglected. Again residual stresses were not considered. This assumption allowed using the base material data stored in the NASGRO database, as in the previous simulations. The second important simplifying assumption is about the dimensions of the initial defects. The widespread corrosion damage identied and quantied by many metallographic sections such as the one in Fig. 11a, was modelled by a single surface crack having the deepest and largest corrosion attack as dimension. Post-mortem failure surface analysis provided the justication for this assumption. Fig. 11b clearly demonstrates that the fatigue crack started from the semi-elliptical corrosion pit. The SEM pictures were taken perpendicular to the loading directions (section AA in Fig. 4b). As already mentioned not only pitting but also inter-granular corrosion within the pits was found, Fig. 11b. In the model no distinction was made between the two corrosion types, which are therefore referred to as corrosion damage in the following. The schematic view of Fig. 12 claries the fracture mechanical treatment of the corrosion damaged area, which was simulated by a semi-elliptical surface crack. This surface crack comprehended the damage due to pitting and inter-granular corrosion. Since the real surface of the corroded specimen presented several corrosion damaged areas, the dimensions of depth, a, and width, c, of nal single semi-elliptical surface crack of the model were taken separately from the deepest and the widest corrosion damage found in the metallographic sections, Fig. 13. Fig. 14 claries how

Fig. 12. Damage due to pitting and inter granular corrosion is simulated by a surface crack which comprehends the total corrosion damaged area.

2c

2c 2c

a a

Fig. 13. The denitive model surface crack has the dimension of the deepest (a) and widest (2c) corrosion damages.

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148

143

2c 2c

2c 2c

Fig. 14. The denitive model surface crack has the dimension of the deepest (a) and widest (2c) corrosion damages also in the case of coalescence of several pits.

Table 4 Corrosion damage dimensions in AA 2024-T3 specimens Exposure time Maximum depth a (lm) BM 100 250 1000 203.7 170.2 536.6 FSW 104 165 308 Maximum width c (lm) BM 600.3 370.6 762.4 FSW 220 606 864 a/c BM 0.339 0.459 0.356 FSW 0.47 0.18 0.356

the model could also take into account the presence of coalescence of several pits. The effect of exposure time on the corrosion damage dimensions are listed in Table 4. The measurements of the base material were taken from a metallographic section parallel to the rolling direction (T(L)). About 200 randomly selected pits for each exposure time were analysed. Regarding the FSW material, the maximal width and depth of the pits were measured in the TL direction of the specimens (parallel to the weld nugget). The degree of corrosion was evaluated in the different zones of the joint (DXZ, TMAZ, HAZ and unaffected base material). The specimens were iteratively cut, metallographic prepared and measured each 0.5 mm: the measurements began at the centre of the weld nugget and nished until the unaffected metal was reached (about 12 mm from the joint line). Referring to Table 4, the explanation for the un-expected drop of the dimension of the corrosion zone after 250 h alternate immersion of the base material is the following: the corrosion process on the aluminium alloy has a volumetric nature. It means that the pit grows in the three dimensions, not only in depth. At certain period of time the extension of inter-granular corrosion attack led to a coalescence of 2 (or more) contiguous pits. Finally, this coalescence resulted in the separation of a corroded layer (exfoliation corrosion), exposing smaller pits. Moreover, it should be noted once comparing the values of Table 4 to the corresponding fatigue lives of Fig. 6, that the measurements are obtained separately using the model described in Figs. 1216, and not derived from the failed samples. The dimensions reported in Table 4 were used in AFGROW and in NASGRO 3.0 as initial crack dimensions for the implemented model semi-elliptical surface crack, which is shown in Fig. 9. The dimension W and t of the model are the specimen width and the specimen thickness, respectively. No remarkable differences were found between the maximum corrosion damage area (from metallographic studies for the same corrosion time exposition) and the post-mortem ndings, displaying similar fatigue life results. Both, the corrosion damage and post-mortem geometry simulation results, are in good agreement with the experimental results, which validate the assumptions made for the implementation of this model. The simulations results for the polished 100, 250 and 1000 h pre-corroded base and FSW material are presented in Fig. 15. All the simulation results both from NASGRO 3.0 and AFGROW showed a good agreement with the experimental observation. 4.3. Variable amplitude loading In modelling the fatigue life of the open-hole FSW specimens the same basic assumptions as for the base and FSW unnotched simulations were done. Namely, the crack is considered as starting from the same constituent particles and particle clusters, and the entire fatigue life is treated as comprised of crack propagation. Moreover, also here the residual stresses were neglected. The same dimensions of particle inclusions as for the base and FSW un-notched simulations were used. The AA2024-T3 T(L) stored in the NASGRO database, whose constants are listed in Table 3, and the double corner crack at hole model was used for performing the calculations. The model implemented in AFGROW has a stress intensity factor solution developed from Newman and Raju [35]. The sketch of the model is presented in Fig. 16. D is the hole diameter, t is the specimen thickness and W the width. a and c are the crack length in depth and width direction, respectively. The variable amplitude loading spectrum used in the test, introduces the problem of the interaction effects. There are currently six choices of load interaction models in AFGROW: the Closure Model, FASTRAN, the Hsu Model, the Wheeler and the Generalised Willbenborg Model. As also presented in [17], for loading histories with single overloads or with a predominance of overloading, the Generalized Willenborg retardation model showed very good agreement with

144

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148

350 Maximum Stress [MPa] 300 250 200 150 100

AA 2024-T3 t=4mm R=0.1
60 Hz Laboratory air

Maximum Stress [MPa]

100h Corroded Experimental Data NASGRO 3.0 AFGROW

350 300 250 200 150 100

AA 2024-T3 t=4mm R=0.1
60 Hz Laboratory air

250h Corroded Experimental Data NASGRO 3.0 AFGROW

10

10

Cycles

10

10

10

10

Cycles

10

10

350 Maximum Stress [MPa] 300 250 200 150 100

AA 2024-T3 t=4mm R=0.1
60 Hz Laboratory air

Maximum Stress [MPa]

1000h Corroded Experimental Data NASGRO 3.0 AFGROW

350 300 250 200 150 100 50

100h Corroded Experimental Data NASGRO 3.0 AFGROW

AA 2024-T3 FSW t=4mm R=0.1

60 Hz Laboratory air

10

10

10 Cycles

10

10

10

10 Cycles

10

350 Maximum Stress [MPa] 300 250 200 150 100 50

Maximum Stress [MPa]

250h Corroded Experimental Data NASGRO 3.0 AFGROW

350 300 250 200 150 100 50

1000h Corroded Experimental Data NASGRO 3.0 AFGROW

AA 2024-T3 FSW t=4mm R=0.1

60 Hz Laboratory air

AA 2024-T3 FSW t=4mm R=0.1

60 Hz Laboratory air

10

10

10 Cycles

10

10

10 Cycles

10

10

Fig. 15. Comparison between base and FSW pre-corroded material experimental results and computer simulations.

t a

Fig. 16. Semi-elliptical surface crack model implemented in AFGROW.

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148

145

450 Experimental Data AFGROW Willenborg RSO=3

Maximum Stress [MPa]

400

350 AA 2024-T3 FSW t=4 mm FALSTAFF

60 Hz Laboratory air

300

250

200 2 10

10

10 Flights

10

Fig. 17. Comparison between FSW open-hole specimens experimental results under FALSTAFF and computer simulations.

experimental data: this model was used for the present investigation with a shutoff value of the stress ratio, RSO = K OL =K max = 3.0 (which is the suggested value for aluminium). Moreover the option of taking the effect of compressive max stresses into account was switched off. This means, that the original version of the Willenborg Model was used, with no possibility of reducing the retarding effect after an underload reducing the plastic zone ahead of the crack tip. Fig. 17 gives the comparison between the experimental results of the FSW open-hole specimens and the computer simulations. Also here, the calculations are in very good agreement with the fatigue test. 4.4. Cold expanded holes The predictions were performed with the code AFGROW. The base material data stored in the NASGRO database (also present in AFGROW) were used. The material constants are those for the AA 2024-T351 orientation T(L). The material constants are summarised in Table 5. In the predictions of the experimental activity presented in [36] a slightly different approach was adopted: the initial crack size was set at 0.5 mm to correlate with the test data. The double corner crack at hole model [35] was used and corner crack aspect ratio was set at 1. To initiate a 0.5 mm crack, approximately 7370 and 11340 ights were needed for the NCx and Cx holes, respectively. Therefore, these ights were assumed as the lives for initiating a crack of 0.5 mm and added to the AFGROW results. Since the experimental procedure was conducted under FALSTAFF, also here the Generalised Willbenborg Model [13] was employed to account for the overload retardation effect. The shutoff value of the stress ratio, RSO = K OL =K max , was set at 3.0. In Fig. 18 the prediction (smooth line) obtained with max AFGROW is plotted together with the experimental data (scatter symbols) for the AA 2024-T351 NCx specimens in the form of an aN diagram. The calculations show a very good agreement with the measured values. Symbols L and R signify the leftand right-hand sides of the hole when viewed from the front in the test machine, while F and B represent the front and back side of the specimen. The diagram contains four experimental tests. The determination of the residual stress distribution around a cold-worked hole is needed in order to be able to perform a realistic life prediction. The residual tangential stress distribution introduced by cold expansion technique is the main reason for fatigue life extension of the part considered. Various techniques have been developed and used to determine the residual stress distribution; these could be divided into three group: experimental, numerical and analytical approaches. Closed-form solutions are usually obtained by analytical techniques. Compared with the numerical or experimental methods, the closedform theory often provides a simpler and faster method to acquire the residual stress distribution. With the development of cold expansion, a number of such models have been developed, such as those of Hsu and Forman [37], Rich and Impellizzeri [38], Chang [39], Ball [40,41] and many others [42,43]. In [36], the cold work-induced tangential stresses were calculated with closed-form methods as well as measured with X-ray diffraction technique. The residual stress distribution is presented in Fig. 19, [36]. The tangential stress is compressive near the holes edge within the distance of 12 times of the hole radius.

Table 5 Material parameters given in the material database of NASGRO for the material 2024-T351 T(L) UTS (MPa) 468.84 YS (MPa) 358.52 KIC (N/mm3/2) 1424.68

DK0 (N/mm3/2)
90.34

C 1.59 1012

n 3.3

p 0.5

q 1.0

a
1.5

146

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148

4.5 4.0 3.5 Crack Length [mm] 3.0 2.5 2.0 1.5 1.0 0.5 0.0 0 2000 4000 6000
Specimen 2 LF Specimen 2 RF Specimen 2 LB Specimen 2 RB Specimen 3 LF Specimen 3 RF Specimen 3 LB Specimen 3 RB Specimen 4 LF Specimen 4 RF Specimen 4 LB Specimen 4 RB Specimen 5 LF Specimen 5 RF Specimen 5 LB Specimen 5 RB AFGROW Willenborg RSO=3

Not cold Expanded FALSTAFF

8000

10000

Flights
Fig. 18. Comparison between NCx specimens experimental results under FALSTAFF and computer simulations.

200 100 0 -100

res [MPa]

-200 -300 -400 -500 -600 -700 0.0 0.5 1.0 1.5 2.0 2.5

X-Ray Entrance Face X-Ray Exit Face Ball Chang Hsu-Forman Rich-Impellizzeri Linear Fit Entrance Face Linear Fit Exit Face 3.0 3.5 4.0 4.5 5.0

Normalized Distance from Hole Edge r/R

Fig. 19. Residual stress distributions determined by different closed-form models, compared with X-ray diffraction tests.

The diagram summarizes the results obtained with the plane stress closed-form models proposed by Ball, Chang, Hsu-Forman and Rich-Impellizzeri. The scatter symbols refer to the X-ray measurements, both for the entrance and exit face of the cold-worked hole. A linear t of the both experimental results is also present in the diagram of Fig. 19. The diagram shows the stresses versus the normalized distance from the holes edge. Since the experimental results show the most conservative values in the area of interest for the calculations, i. e. between 0.5 mm and 3.6 mm from the edge of the hole, those values were used for performing the predictions. The AFGROW computer code was used for performing the calculations, since it allows accounting for the existence of residual stresses by calculating additive residual stress intensities, Kres, at user dened crack length increments. The linear t of the entrance face (which shows the highest residual stresses and therefore has the highest probability of initiating a crack) was used for the residual stress values in width direction (correcting the c length of the crack in Fig. 16), while a linear interpolation between the entrance and exit face residual stresses was adopted for the residual stresses in depth direction (correcting the a length of the crack in Fig. 16). It should be noted that the linear t is not self-equilibrated. However, the t was able to match the experimental data much better than the other models, therefore it has been considered as a reasonable approximation. AFGROW utilizes than the Gaussian integration method, which uses the point load stress intensity solution from the Tada, Paris and Irwin Stress Intensity Handbook [44]. The stress intensities factors are than merely added to the maximum and minimum stress intensities caused by the applied load: this causes a variation of the true R ratio, Rtr, while DK remains unaffected. The true load ratio of the specimens affected from residual stresses is now dened as

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148

147

6.5 6.0 Cold Expanded 5.5 FALSTAFF Willenborg RSO=3 5.0 No Residual Stresses 4.5 Willenborg RSO=3 4.0 Residual Stresses 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 10000 20000 Flights 30000

Specimen 1 LF Specimen 1 RF Specimen 1 LB Specimen 1 RB Specimen 2 LF Specimen 2 RF Specimen 2 LB Specimen 2 RB Specimen 3 LF Specimen 3 RF Specimen 3 LB Specimen 3 RB AFGROW 1 AFGROW 2

Crack Length [mm]

40000

Fig. 20. Comparison between Cx specimens experimental results under FALSTAFF and computer simulations.

Rtr

K min;app K res K tr;min K tr;max K max;app K res

where Kmax,app and Kmin,app are the maximum and minimum values of the K applied. The presence of the residual stresses is therefore inuencing the da/dN value, by meaning of the Rtr ratio. With this approach and the base material data, the da/dN DK behavior of welded structures can be predicted, as elsewhere published [11,12,17]. Since in [12] and in [11] the authors have successfully shown the potential of the model with constant amplitude and with simpler variable amplitude loading history, the present calculation was considered as a further validation of the model in the presence of an in-service load spectrum. Therefore also here, in the prediction of the Cx specimens the Generalised Willbenborg Model with a shutoff value of 3.0 was used. In Fig. 20 the predictions obtained with AFGROW are plotted together with the experimental data for the AA 2024-T352 Cx specimens in an a-nr. of Flights diagram. The calculations show a very good agreement with the measured values. The experimental data of three specimens (scatter symbols) are compared with two predictions: the calculations obtained taking the residual stresses into account, referred as AFGROW 1 and simulations without the implementation of the rres in the software, AFGROW 2. The predictions obtained without using the residual stress distribution are extremely conservative, while the calculations performed taking into account the presence of a residual stress eld in width and depth direction extremely improve the agreement with experimental results, while remaining conservative.

5. Conclusions In the present investigation it was successfully demonstrated the possibility of using fracture mechanics to predict the fatigue life of structures. In particular the fatigue lives of pristine and pre-corroded base and friction stir welded specimens, even in the presence of residual stresses and under variable amplitude loading conditions, can be predicted with widespread aerospace fracture mechanics based packages. The calculations are in very good agreement with the experimental results once the following basic assumptions have been done:  The weld material is treated as base material and the very low transverse residual stresses are neglected.  The cracks in the base as well as the FSW material are assumed to form at particle inclusions and the entire fatigue life is comprised of crack propagation.  Pitting and inter-granular corrosion are treated as a single corrosion damage source and the model surface crack comprehends this damage.  The several corrosion damaged areas of the specimen surface are simulated with a single semi-elliptical surface crack having the dimensions of the deepest and the widest corrosion damage area.  The Generalised Willenborg retardation model with a shutoff value of the stress ratio, RSO = 3.0 is used for the variable amplitude loading predictions.  A residual stress prole in specimen thickness as well as width direction obtained by tting the X-ray diffraction measurements is used in combination with the Generalised Willenborg retardation model for the predictions of the cold expanded holes specimens under variable amplitude loading.

148

T. Ghidini, C. Dalle Donne / Engineering Fracture Mechanics 76 (2009) 134148

Acknowledgement The Authors wish to acknowledge the work of Mr. Ulises Alfaro related to the corrosion testing and characterisation. References
[1] Lazzeri L, Pieracci A, Salvetti A. An evaluation of fatigue crack growth prediction methods used in aircraft design. In: Proceedings of the 18th symposium of the international committee on aeronautical fatigue; 1995. I: p. 61545. [2] Harter JA. Comparison of contemporary FCG life prediction tools. Int J Fatigue 1999;21:S1815. [3] Harter JA. AFGROW, users guide and technical manual. AFRL-VA-WP-TR-2004-XXXX; September 2004 [Version 4.0011.14]. [4] NASA-ESA, ESACRACK 4 users manual. TOS-MCS/2000/41/In; 2000(1). [5] Forman RG, Shivakumar V, Mettu SR, Newman JC. Fatigue crack growth computer program NASGRO version 3. JSC-22267B; August 2000. [6] Skorupa M. Load interaction effects during fatigue crack growth under variable amplitude loading-a literature review. Part I: empirical trends. Fatigue Fract Engng Mater Struct 1998;21:9871006. [7] Skorupa M. Load interaction effects during fatigue crack growth under variable amplitude loading-a literature review. Part II: qualitative interpretation. Fatigue Fract Engng Mater Struct 1998;22:90526. [8] Wei RP. Corrosion fatigue of aluminium alloys: chemistry, micromechanics and reliability. Final technical report to AFOSR; 1998. [9] Dolley EJ, Lee B, Wei RP. The effect of pitting corrosion on fatigue life. Fatigue Fract Engng Mater Struct 2000;23:55560. [10] Ghidini T, Polese C, Lanciotti A, Dalle Donne C. Simulations of fatigue crack propagation in friction stir welds under ight loading conditions. Materialprfung (Materials Testing) 2006;7/8:3705. [11] Ghidini T, Dalle Donne C. Fatigue crack propagation assessment based on residual stresses obtained through cut-compliance technique. Fatigue Fract Engng Mater Struct 2006;30:21422. [12] Ghidini T, Dalle Donne C. Prediction of fatigue crack propagation in friction stir welds. In: Proceedings of the 4th international conference on friction stir welding, Park City, Utah; 2003. [13] Willenborg J, Engle RM, Wood HA. A crack growth retardation model using an effective stress concept. AFFDL-TM-71-1-FBR, Wright Patterson Air Force Laboratory; 1971. [14] Thomas WM. Friction stir butt welding. International patent application no. 9125978.8; 1991. [15] Dalle Donne C, Biallas G, Ghidini T, Raimbeaux G. Effect of weld imperfections and residual stresses on the fatigue crack propagation in friction stir welded joints. In: Proceedings of the 2nd international conference on friction stir welding, Gothenburg, Sweden; 2000. [16] Biallas G, Braun G, Dalle Donne C, Staniek G, Kaysser W. Mechanical properties and corrosion behavior of friction stir welded 2024-T3. In: Proceedings of the 1st international conference on friction stir welding, Cambridge, UK; 1999. [17] Ghidini T. Fatigue life predictions of friction stir welded joints by using fracture mechanics methods. Fortschritt-Berichte VDI; 2006. p. 155. ISBN 3-18330418-X. [18] ASTM-G44-99, Standard practice for exposure of metals and alloys by alternate immersio in neutral 3.5% sodium chloride solution; 2003. [19] Kristensen JK, Dalle Donne C, Ghidini T, Mononen JT, Norman A, Pietras A, Russell M, Slater M. Properties of friction stir welded joints in the aluminium alloys 2024, 5083, 6082/6060 and 7075. In: Proceedings of the 5th international conference on friction stir welding, Metz, France; 2004. [20] Dalle Donne C, Raimbeaux G, Biallas G, Allehaux D, Palm F, Ghidini T. Fatigue properties of friction stir welded aluminum butt joints. In: Guillaume M. edior. ICAF03, Fatigue of aeronautical structures as an engineering challenge; 2004. ISBN 0954345428. [21] Dalle Donne C, Ghidini T. Fatigue strength of friction stir welded joints. Deutscher Verband fr Materialprfung e. V.-Bericth 2004;236:2818. [22] FALSTAFF, Description of a ghter loading standard for fatigue evaluation. Combined report (LBF, NLR, IABG and F&W); 1976. [23] Zhang X, Wang Z. Fatigue life improvement in fatigue aged fastener holes using the cold expansion technique. Int J Fatigue 2003;911:124957. [24] Laz PJ, Hillberry BM. The role of inclusions in fatigue crack formation in aluminum 2024-T3. In: Proceedings of the fatigue96 international conference; 1996. p. 12938. [25] DeBartolo EA, Hilberry BM. Effects of constituent particle clusters on fatigue behavior of 2024-T3 aluminum alloy. Int J Fatigue 1998;20(10):72735. [26] Grosskreutz JC, Shaw CG. Critical mechanisms in the development of fatigue cracks in aluminum 2024. In: Proceedings of the 2nd international conference on fracture, Brington, UK; 1969. [27] Bowles CQ, Schijve J. The role of inclusions in fatigue crack initiation in an aluminum alloy. Int J Fracture 1973;9:1719. [28] Newman JCJ, Edwards PR. Short crack growth behaviour in an aluminum alloy. AGARD R-732; 1988. [29] Newman JCJ, Philips EP, Swain MH, Everett Jr RA. Fatigue mechanics: an assessment of a unied approach to life prediction. ASTM STP 1992;1122:527. [30] Newman JCJ, Phillips EP, Swain MH. Fatigue-life prediction methodology using small-crack theory. Int J Fatigue 1999;21(2). [31] Murakami Y. Effects of small defects and nonmetallic inclusions on the fatigue strength of metals. JSME Int J, Ser I 1989;32(2):16780. [32] Newman JCJ, Wu XR, Venneri SL, Li CG. Small-crack effects in high-strength aluminum alloys, vol. 1309. Nasa Reference Publication; 1994. [33] Edwards PR, Newman Jr. JC. Short-crack growth behaviour in various aircraft materials. AGARD R-767; 1990. [34] Dalle Donne C, Lima E, Wegener E, Pizalla A, Buslaps T. Investigations on residual stresses in friction stir welds. In: Proceedings of the 3rd international conference on friction stir welding, Kobe, Japan; 2001. [35] Newman JC, Raju IS. Stress intensity factor equations for cracks in three-dimensional bodies subjected to tension and bending loads. Comput Meth Mech Fract. Amsterdam: Elsevier; 1986. [36] Wang Z, Zhang X. Predicting fatigue crack growth life for cold-worked holes based on existing closed-form residual stress models. Int J Fatigue 2003;25(0):28591. [37] Hsu YC, Forman RG. Elasticplastic analysis of an innite sheet having a circular hole under pressure. Trans ASME J Appl Mech 1975;42:34752. [38] Rich DL, Impellizzeri LF. Fatigue analysis of cold-worked and interference t fastener holes. ASTM STP 1977;637:15375. [39] Chang JB. Prediction of fatigue crack growth at cold worked fastener holes. J Aircraft 1977;14:9038. [40] Ball DL. Elasticplastic stress analysis of cold expanded fastener holes. Fatigue Fract Engng Mater Struct 1995;18:4763. [41] Ball DL, Lowry DR. Experimental investigation on the effects of cold expansion of fastener holes. Fatigue Fract Engng Mater Struct 1998;21:17334. [42] Clark G. Modelling residual stresses and fatigue crack growth at cold-expanded fastener holes. Fatigue Fract Engng Mater Struct 1991;14:57989. [43] Link RE, Sanford RJ. Residual strains surrounding split-sleeve cold expanded holes in 7075-T651 aluminium. J Aircraft 1990;27:599604. [44] Tada H, Paris PC, Irwin GR. The stress analysis of crack handbooks. St. Louis, MO: Paris Productions, Inc.; 1985.