Engineering Structures 45 (2012) 307–313

Contents lists available at SciVerse ScienceDirect

Engineering Structures

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

Eurocode 9 to estimate the fatigue life of friction stir welded aluminium panels

Meysam Mahdavi Shahri a,⇑, Torsten Höglund b, Rolf Sandström a

a Department of Materials Science and Engineering, KTH, Brinellvägen 23, S-100 44 Stockholm, Swedenb Division of Structural Design and Bridges, School of Architecture and the Built Environment, Royal Institute of Technology, KTH, Brinellvägen 23, S-100 44 Stockholm, Sweden

a r t i c l e i n f o

Article history:Received 21 November 2011Revised 1 June 2012Accepted 27 June 2012Available online 3 August 2012

Keywords:Friction stir weldingFatigue assessmentEurocode 9Nominal stress

0141-0296/$ - see front matter � 2012 Elsevier Ltd. Ahttp://dx.doi.org/10.1016/j.engstruct.2012.06.039

⇑ Corresponding author. Tel.: +46 87906544; fax: +E-mail address: meysamms@kth.se (M. Mahdavi S

a b s t r a c t

Eurocode 9 is a standard that covers the design of building and engineering structures made fromwrought and cast aluminium alloys. A part of the Eurocode 9 handles the design of aluminium structuressusceptible to fatigue. Eurocode 9 has data for aluminium alloys and welded structures for conventionalwelding methods (fusion welding) except for friction stir welding processes. The present study comparesfatigue test results from friction stir welded joints with fatigue curves of traditional fusion welded jointswhich are presented in Eurocode 9. The results are in reasonable agreement with experimental data andFEM predictions. This suggests that Eurocode 9 can be used for estimating the fatigue strength of frictionstir welded joints.

� 2012 Elsevier Ltd. All rights reserved.

1. Introduction

Friction stir welding (FSW) is a low heat input solid state weld-ing technology especially suitable to low melting point metals,such as Al and Cu [1]. The application of the FSW technology inaerospace, automotive and shipbuilding industry is seen to providesuperior joint integrity. As the temperature remains below themelting point, residual stresses are low. In addition, the absenceof filler material in the weld gives limited risk of porosity forma-tion and a smooth surface of the weld resulting in good fatigueproperties [2,3]. Conventional fusion welding of aluminium alloysresults in solidification cracking and higher residual stresses com-pared to FSW. Using filler material and shielding gas results inrather different fatigue properties of fusion welding componentscompared to FSW ones.

Several studies have been conducted on friction stir welded buttjoints, demonstrating that they have fatigue strength close to thatof the base material, and generally higher than the strength of thejoints obtained with traditional welding techniques [4–9].

Nowadays FSW is widely used. Often, aluminium profiles arewelded together to form large panels used in engineering struc-tures where fatigue is an important design criteria. Eurocode 9[10] is a standard for fatigue assessment of aluminium structuresand weldment based on nominal stresses. Eurocode 9 includes datafor aluminium alloys and welded structures for conventional weld-ing methods but not for the FSW procedure. In the present paperthe capability of Eurocode 9 for estimating the fatigue life time

ll rights reserved.

46 8 207681.hahri).

of FS welded extruded aluminium profiles is investigated. Thepresent study compares experimental fatigue data of friction stirwelded joints with fatigue curves of fusion welded joints in Al6005.

In previously published papers by the present authors [11,12]fatigue assessment of FS welded hollow panels has been performedusing finite element method (FEM) stress analysis and the theoryof critical distance method (TCD). The critical distance methodhas been proposed for fatigue assessment of a notch componentor a body containing a crack. The method is associated with aver-aging stress around a notch or taking stress at a distance from thenotch root and the average stress is used for fatigue life prediction.The procedure of using TCD method is explained in aforemen-tioned references by author.

In the present study Eurocode 9 is used for prediction of fatiguelife at the failure locations (both for base metal and weld material).These predictions are compared with FEM computation in this study.

2. Experimental

2.1. Materials and welding

The profiles used in fatigue tests were made of extruded alu-minium 6005A in the T6 (artificially aged) condition. Chemicalcomposition and mechanical properties of the alloy are shown inTables 1 and 2. The tensile properties were measured on samplesfrom the studied profile according to the standard EN10002-1:2001. These profiles are used as floors in trains and deckpanels in shipbuilding, as well as some application as militarybridges.

Table 1Chemical composition (in wt.%).

Si Fe Cu Mn Mg Cr Zn Ti Al

6005A actual 0.61 0.21 0.07 0.15 0.54 <0.01 <0.01 0.02 Bal6005A nominal 0.5–0.9 <0.35 <0.30 <0.50 0.4–0.7 <0.30 <0.20 0.10 Bal

Table 2Mechanical properties of 6005A.

Yield strength (MPa) Tensile strength (MPa) Elongation (%) Young’s modulus (GPa)

6005A T6 actual 253 280 10 69.56005A T6 typical (AluSelect) 260 285 13 69.56005A T6 nominal min 225 min 270 8

308 M. Mahdavi Shahri et al. / Engineering Structures 45 (2012) 307–313

Friction stir welding technique was used to join these profiles atthe facilities of Sapa Technology in Sweden. The studied profilesare shown in Fig. 1 before and after friction stir welding. Two mainprofiles with different weld geometries (geometry of extruded pro-files around the weld location) are investigated. The two weldgeometries are:

1. hourglass shape,2. half overlap (truss, diagonal ribs).

Two series of hourglass shape were produced (we call these ser-ies A and B). Two series of half over lap was also produced (series Cand D).

Profiles were welded with two welding heads operating fromtop and bottom sides of the panels. Therefore each panel containsweld nuggets both on the top and the bottom side at the centre

Fig. 1. Extruded aluminium profiles. Profiles are welded from top and bottom side; (a) b(truss, diagonal ribs).

part of the panel. Two welding procedures were used. In the firstone, simultaneous welding of the top and the bottom side (seriesB, C, D were welded in this way) was performed. The second pro-cedure involved welding first on the bottom and then on the topside (series A). A Summary of the welding procedure for each seriesis shown in Table 3. Welding parameters such as rotation speed,traverse speed, tilt angle were similar for all series. In a previouslypublished paper [11], welding procedure and clamping conditionsare explained.

2.2. Fatigue testing

Sections of 100 mm width were prepared for fatigue testing.3-Point bending was used in the fatigue tests. This method is em-ployed since it produces more stress at the centre part of the pan-els where the weldment is located. Constant amplitude fatigue

efore welding; (b) after welding; (1) is hourglass shape profile and (2) is halfoverlap

Table 3Summary of fatigue testing conditions and welding procedure for each test series.

Series Type of profile Welding procedure Applied load range (kN) Loading setup Failure location

A Hourglass NS 14–23 Fig. 2a WeldB Hourglass S 21–23 Fig. 2a Base metalC Halfoverlap (truss) S 8–12 Fig. 4a Base metalD Halfoverlap (truss) S 4–9 Fig. 5a (NC) Weld

NS: Not simultaneously welded on the top and bottom side; S: simultaneously welded on the top and bottom side; NC: non-centrally loaded.

M. Mahdavi Shahri et al. / Engineering Structures 45 (2012) 307–313 309

tests were performed in a servo-hydraulic testing machineequipped with an actuator of 250 kN load capacity. A sinusoidalload–time function was used with ratio R = Fmin/Fmax set to 0.1where Fmin and Fmax are the applied minimum and maximumforces on the profiles. Series A and B were loaded in a similarway, this is shown in Fig. 2a. Fig 3a shows the weld location forhourglass profile. Series C was loaded with loading support onthe weld centreline but series D was loaded eccentrically; this isshown in Figs. 4a and 5a. Eccentric loading produces a shear forceat the weld, resulting in a higher stress concentration at the weldinterface notch. The load ranges used for the different series isshown in Table 3.

2.3. Fatigue testing results and failure locations

Two series of hour glass shape profiles were fatigue tested.Although the same loading conditions were used for series A andB, fatigue failure occurred at different positions (see Figs. 2a and3a) resulting in different fatigue strengths. All samples of series A

(a)

(b)

Fig. 2. (a) Series B: Hourglass shape panel and loading system, failure occurred in angassessment.

failed in the weld. The fatigue crack initiated from the sharp inter-face notch and propagated to the weld area. Thus complete frac-ture occurred in the weld. No samples of series B failed in theweld. Initiation site was at the vicinity of a blunt notch in the basemetal and all samples of series B failed at the same position. Twoseries of half overlap (truss, diagonal ribs) were also tested. For acentrally loaded profile (series C) failure occurred in the base metaland for non-centrally loaded profiles failure occurred in the weld(series D). Failure locations are shown in Figs. 4a and 5a. Experi-mental fatigue limit (fatigue strength at 2 million cycles) for eachseries is presented in Table 4. Failure locations, corresponding localstresses and testing setup (including size of the span) are fully ex-plained in [11].

3. Eurocode 9

The European Standard Eurocode 9 for fatigue design of alumin-ium structures estimates the fatigue life on the basis of the nomi-nal stress applied to the joint. The nominal stress is defined based

ular rib by tension force DN. (b) Detail type 1.6 (Eurocode 9) was used for fatigue

(a)

(b)

Fig. 3. (a) Series A: Same profile and loading system as series B but failure occurred in the interface notch at weld location. (b) Detail type 1.6 (Eurocode 9) was used forfatigue assessment.

310 M. Mahdavi Shahri et al. / Engineering Structures 45 (2012) 307–313

on beam theory. Nominal stresses normally include two compo-nents, bending stresses and axial stresses in the section.

After determination of the nominal stress, a detail categoryneeds to be chosen from Eurocode 9 in order to estimate the fati-gue life. Eurocode 9 presents different geometries of welded andnon-welded structures. When doing fatigue assessment the cate-gory which is most similar to actual structure should be chosen.The geometrical similarity, stress direction, and failure locationare the parameters that need to be considered to choose the rightcategory.

Each detail category introduces a reference stress Drc which isthe value of fatigue strength at 2 � 106 cycles. Fatigue life can thenbe predicted using the following equation:

Ni ¼ 2� 106 Drc

Dri

1ab

� �m1

ð1Þ

where Ni is the predicted number of cycles to failure at stress rangeDri; Drc, the reference value of fatigue strength at 2 � 106 cycles,depending on detail category; Dri, the range of calculated nominalstress from beam theory; m1, the inverse slope of the Dr � N curve,depending on detail category; a, the partial factor allowing foruncertainties in the loading spectrum and analysis and b is the par-tial factor for uncertainties in materials and execution.

4. Results

Nominal stress was calculated in the section where failure oc-curred for series A, B, C, D. This is shown in Figs. 2b–5b.

4.1. Nominal stress for series A and B

For series B failure occurred in the ribs therefore the tensileload, DN, in the ribs must be calculated, see Fig. 2a. DN can be eas-ily calculated by resolving the applied force DF in the h direction.Consequently the nominal stress can be obtained when the thick-ness t and width b are known. The same procedure was applied forseries A, but this time failure occurred in the weld. So the tensileload DN applied on the weld was calculated, when the weld thick-ness w and width b are known nominal stress can be calculated, seeFig. 3b.

4.2. Nominal stress for series C and D

For series C and D the tensile load DN was obtained. But a bend-ing moment (DM) also exists in the failure location. The bendingmoment calculated for series C and D is shown in Figs. 4b and5b. The stress produced by the bending moment was added tothe axial stresses in order to get the total applied nominal stress:

DNb � t þ

DM � 6b � t2

In order to verify the calculated nominal stresses they were com-pared with FEM stress computation in [11]. Maximum stressaround the stress concentration location in FEM is divided by stressconcentration factor Kgt. This is indeed the theoretical value fornominal stress. For series A and B the difference between the FEM

(a)

(b)

Fig. 4. (a) Series C: Half overlap shape panel and loading system, failure occurred in a blunt notch but not weld interface. (b) Detail type 1.6 (Eurocode 9) was used for fatigueassessment.

M. Mahdavi Shahri et al. / Engineering Structures 45 (2012) 307–313 311

value and the calculated one is less than 5% but for series C and Dthis reaches up to 15% and 20%.

The next step is choosing the proper detail category from Euro-code 9 in order to estimate fatigue life time. Table 5 shows the de-tail category chosen for the different series. The reason forchoosing the detail categories shown in Table 5 is explained inthe following discussion section. Detail categories are illustratedin Figs. 2b–5b for each series.

Inserting values from Table 5 into Eq. (1) gives an estimated fa-tigue life (Ni) for the corresponding calculated nominal stress range(Dri). a was chosen equal to one since loading was performed withconstant amplitude. Also no partial factor for uncertainties wasconsidered (b = 1).

Fig. 6 shows the predicted fatigue life using Eurocode 9 for allseries. These results are compared with experimental fatigue lifein term of nominal stress. As can be seen in Fig. 6 there is goodagreement between the experimental and predicted fatigue lifefor series B, C and D. The predicted life is slightly conservativei.e. they are not greater than experimental data.

For series A the deviation of the experimental data and predic-tion was rather large. The predicted life is longer than the experi-mental data. This will be analysed in the discussion.

5. Discussion

In the nominal stress approach a nominal section is used. Sincenotch effects are not taken into account they have to be includedon the strength side. This approach of course requires that a suit-able cross section can be identified in the structure that is to beinvestigated. The definition of nominal stress is simple: it is theaverage stress, calculated by beam theory but at a closer look,problems emerge. The first problem is determining the nominalstress in a cross section. For complex loading and structural com-plexity this is not always easy. Secondly a detail category similarto the loaded structure should be identified for life estimation.The problem arises when the loaded structure is not exactly thesame as the detail category within the Eurocode 9. Also variations

(b)

(a)

Fig. 5. (a) Series D: Half overlap shape panel and loading system, failure occurred in weld interface. (b) Detail type 1.6 (Eurocode 9) was used for fatigue assessment.

Table 4Experimental and predicted fatigue strength at 2 million cycles.

Series Experimental fatigue limit (kN) TCD prediction (kN) Eurocode 9 prediction (kN) Deviation from experiment for TCD/Eurocode 9 (%)

A 12.43 127.7 377 +90/+96B 21.74 21.1 18.43 �3/�15C 7.71 7.78 7.09 +1/�8D 3.62 4.8 3.28 +24/�9

Table 5Chosen detail categories for series A, B, C, D. (EN-1999-1-3:2007(E)).

Series Detail type Detail category

Drc m1

A 7.4.1 45 4.3B 1.6 100 7C 1.6 100 7D 13.3 32 4.3

312 M. Mahdavi Shahri et al. / Engineering Structures 45 (2012) 307–313

within the detail in dimensions, welding procedures, etc. may notbe covered in the detail category, resulting in less precision in theprediction.

However, the nominal stress strategy can be used simply withno need of FEM computation. Therefore it is extremely cost effec-tive and time saving. Predicted fatigue strength at two million cy-cle by Eurocode 9 for series A, B, C, D in this paper have beencompared with predicted fatigue limit using FEM and theory ofcritical distance [12]. Prediction for both methods gives a reason-able result within the range of experimental data, Table 4. Predic-tions with Eurocode 9 are conservative in all case (negative errorsign) while TCD is non-conservative for all series (positive error

sign). Predicted data for series A give a big error for both Eurocode9 and TCD. Eurocode 9 and TCD both predict the fatigue limit in

Halfoverlap (angular truss)

10.0

100.0

1000.0

10000 1000000 100000000Number of cycle to failure

Stre

ss R

ange

, MPa

.

Experiment-Series C

Prediction-Euro Code 9-Series C

Experiment-Series D

Prediction-Euro Code 9-Series D

Hourglass

10.0

100.0

1000.0

10000 1000000 100000000Number of cycle to failure

Stre

ss R

ange

, MPa

.

Experiment-Series B

Prediction-Euro Code 9-Series B

Experiment-Series A

Prediction-Euro Code 9-Series A

Prediction-Euro Code 9 (partial penetration)-Series A

(a)

(b)

Fig. 6. Experimental and predicted fatigue life for series A, B, C, D.

M. Mahdavi Shahri et al. / Engineering Structures 45 (2012) 307–313 313

hourglass shape profile for welded region to be much higher thanradii location in base metal. It has been shown elsewhere [12] thatthe reason for the weld failure in series A is excessive residualstresses around the crack tip due to improper clamping and weld-ing procedure. These residual stresses were formed due to localplastic deformation around the crack tip by improper clampingload during the production. This should be noted that these resid-ual stresses are different from those formed due to welding heatand therefore they are not accounted in S–N curves of Eurocode9 and must be taken additionally into account. On the other handin Eurocode 9 only enhancement of fatigue strength can be doneand when the tensile residual stresses are not negligible (like thecase here) the use of enhancement factor is not allowed.

As mentioned above the detail category needs to be chosenproperly in order to get a reasonable estimation of fatigue life.The best category to choose is the one which is similar to theloaded structure in geometry, loading, stress direction, initiationsite, etc.

For series B and C failure occurred in the base metal at the stress con-centration and the detail category can be chosen in a straightforwardway. Detail category 1.6 (from EN 1999-1-3) is the most suitable in thiscase, because it presents the failure in the base metal and in the vicinityof a blunt corner which is the case in series B and C.

Series D resembles a fillet weld (or T joint, with two perpendic-ular welded plates) which can be compared with detail category

13.3 or 13.4. For both categories 13.3 and 13.4 the failure locationis in the weld, 13.3 corresponding to full penetration and 13.4 pre-senting the case with a root crack. They roughly give a good esti-mation and the predicted results are conservative when detailcategory 13.3 and 13.4 are used.

For series A the most similar geometry joint is detail category7.5 which is a butt weld with initiation site at the weld root (partialpenetration) i.e. the weld root defect will propagate as under ten-sion, although the case is different for the hourglass shape in load-ing. For the loaded hourglass shape profile there is no tension atthe root crack (this can be seen through FEM computation). Thisis due to sufficient thickness of the weld and a compression mo-ment applied on the crack face at the weld root which induces al-most no tension to the weld root crack. So detail category 7.4.1 (fullpenetration butt weld without backing) was used for prediction ofseries A. However using this category overestimates the fatiguestrengths (see Fig 6) as residual stresses is not taken into account.Fig. 6 shows the prediction both for detail category 7.5 and 7.4.1.

6. Conclusion

� Eurocode 9 has been used to predict the fatigue life time of FSwelded aluminium profiles.� Prediction results from Eurocode 9 are conservative in all case,

the deviations from the experiment fall between 8% and 15% ofexperimental results.� Both the critical distance method and Eurocode 9 do not pro-

vide accurate prediction when high tensile residual stressesappear in the weld.

Acknowledgments

The authors would like to thank the Sapa Technology Ltd. forfinancial support and providing the test materials and welding.Also the authors would like to thank SIS Förlag AB in Sweden.The figures in the paper are reproduced from SS-EN 1999-1-3:2007 with SIS Förlag AB permission.

References

[1] Mishraa RS, Ma ZY. Frictions stir welding and processing. Mater Sci Eng R2005;R50(1–2):1–78.

[2] Lomolino S, Tovo R, Dos Santos J. On the fatigue behaviour and design curves offriction stir butt-welded Al alloys. Int J Fatigue 2005;27(3):305–16.

[3] Midling OT, Oosterkamp LD, Bersaas J. Friction stir welding aluminium processand applications. In: Seventh international conference INALCO’98, Cambridge,UK; April 1998.

[4] Thomas WM, Johnson KI, Wiesner CS. Friction stir welding – recentdevelopments in tool and process technologies. Adv Eng Mater2003;5(7):485–90.

[5] Matrox SJ. Review of fatigue assessment procedures for welded aluminiumstructures. Int J Fatigue 2003;25:1359–78.

[6] Dickerson TL, Przydatek J. Fatigue of friction stir welds in aluminium alloysthat contain root flaws. Int J Fatigue 2003;25:1399–409.

[7] Bussu G, Irving PE. The role of residual stress and heat affected zone propertieson fatigue crack propagation 2024-T351 aluminium alloys. Int J Fatigue2003;25:77–88.

[8] Lomolino S, Tovo R, Dos Santos J. On the fatigue behaviour and design curves offriction stir butt-welded Al alloys. Int J Fatigue 2005;27:305–16.

[9] Juriçic C, Dalle Donne C, Dressler U. Effect of heat treatment on mechanicalproperties of friction stir welded 6013. In: Third international symposiums onfriction stir welding, September 27–28 2001, Kobe (JP).

[10] Eurocode 9: Design of aluminium structures – Part 1–3: Structures susceptibleto fatigue, EN 1999-1-3.

[11] Mahdavi Shahri M, Sandström R, Osikowicz W. Critical distance method toestimate the fatigue life time of friction stir welded profiles. Int J Fatigue2012;37:60–8.

[12] Mahdavi Shahri M, Sandström R. Influence of fabrication stresses on fatiguelife of friction stir welded aluminium profiles. J Mater Process Technol2012;212:1488–94.