Construction and Building Materials 54 (2014) 224–235

Contents lists available at ScienceDirect

Construction and Building Materials

journal homepage: www.elsevier .com/locate /conbui ldmat

Residual stress modeling of narrow gap welded joint of aluminum alloyby cold metal transferring procedure

0950-0618/$ - see front matter � 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.conbuildmat.2013.12.056

⇑ Corresponding author. Tel./fax: +86 451 86418746.E-mail address: shufengyuan1987@163.com (P. He).

Fengyuan Shu a,b, Yaohui Lv b, Yuxin Liu b, Fujia Xu a,b, Zhe Sun b, Peng He a,⇑, Binshi Xu a,b

a State Key Laboratory of Advanced Welding and Joining, Harbin Institute of Technology, Harbin 150001, Chinab National Key Laboratory for Remanufacturing, Academy of Armored Forces Engineering, Beijing 100072, China

h i g h l i g h t s

� A novel FEM model for narrow gapCMT welding was established.� Impulse input of weld wire and weld

heat was simplified.� Global stress distribution changed

little after cooling down of the firstweld pass.� Bigger strain resulted in shaper

decreasing of residual stress in re-melted zone.� The vulnerable areas of the joint were

obtained.

g r a p h i c a l a b s t r a c t

Cooperation between wire feeding and heat input was simplified so that the CMT welding process couldbe simulated. The evolution of residual strained was obtained, then the mechanism of residual stress evo-lution was observed with the assistance of mechanical properties.

a r t i c l e i n f o

Article history:Received 9 July 2013Received in revised form 6 December 2013Accepted 17 December 2013Available online 17 January 2014

Keywords:Residual stress evolutionCMT + P MIX welding procedureFE methodMechanism

a b s t r a c t

Research on FE method simulation of CMT + P MIX welding process has been suspended because of theabsence of qualified model for characterizing cooperation between wire feeding and heat input. A novel3D FE thermo-mechanical model was established, in which impulse input of weld wire and weld heatwas simplified by decreasing the impulse frequency. Materials were modeled as elastic-perfectly-plasticwith composite heat sources utilized. Numerically simulated results were validated by thermal cyclecurves and residual stress distribution obtained with infrared photography instrument and X-ray diffrac-tion tester, respectively. The numerical residual stress distribution also got well supported by the testedresults obtained by the blind-hole method. Residual stress evolution was analyzed, whereas specialattention was paid to the influences on residual stress distribution by the subsequent weld passes. Basedupon this was the vulnerable area of the joint identified under the conduction of different strength the-ories. The global stress distribution was found to be dominated by the first weld pass while more residualstress in the re-melted zone than that in the HAZ could be released by the last two welding passes. Themechanism was studied through the yielding process of residual strain which was divided into two steps.

� 2013 Elsevier Ltd. All rights reserved.

Table 1Chemical composition (wt.%) of the base plates AA7A52 aluminum alloy.

Zn Mg Cu Mn Cr Ti Zr Fe Si Al

4.0 2.0 0.05 0.20 0.15 0.05 0.05 60.30 60.25 Bal.�4.8 �2.8 �0.20 �0.50 �0.25 �0.18 �0.15

F. Shu et al. / Construction and Building Materials 54 (2014) 224–235 225

1. Introduction

The benefits including thin heat affected zone, little distortionand improved productivity made the cold metal transferring(CMT) process incomparable when applied in welding thin platesand weld-brazing, as reported by Shang et al. [1] and Cao et al.[2]. However, the application of CMT technology was limited dueto insufficient heat input. Bright foreground turned up with pulsecurrent adopted into the CMT process. The CMT mixed with pulses(CMT + P MIX) process could make up heat input insufficiency yetpreserved all the advantages of the CMT process. Particularity ofthe CMT process was the backward drawing of weld wire and itssynchronous conjunction with the sharply dropped weld heatinput. Pulses were added independent of CMT cycles and the heatinput could be adjusted by setting the parameters including peakcurrent, base current and duty cycle, as reported by Pickin et al. [3].

Low heat input and high welding speed in narrow-gap weldingcontributed to the excellent joint properties. It was found in 1976by Henderson and Steffens [4] that the critical region of the nar-row-gap joint was the weld metal rather than the heat affectedzone (HAZ) or the weld metal-parent metal boundary. Narrow-gap welding was common in welding thick steels by submergedarc welding [5], electron beam welding [6] and gas tungsten arcwelding (GTAW) [7]. The double-V groove was recommended byASME according to the industry standards in the process pipingpart, although, none of the groove types were proved to be withabsolute advantages. Negligible change in magnitude of residualstress was found between U groove and V groove in thick pipewelds according to the work by Sattari-Far and Farahani [8]. TheI type groove and the U type groove were firstly proposed by theBattelle Institute in 1963 [9]. Narrow-gap I type joint could alsobe obtained with the assistance of magnetic field as reported byStarling et al. [10]. With regard to gas tungsten arc welding, thedynamic molten pool behavior of the joint was significantly influ-enced by the angle of the groove. Bigger angle gave birth to betterpenetration, but problems such as overheating and underfillingmight be caused simultaneously as indicated by Cho et al. [11]and Chen et al. [12].

Residual stress was one of the major concerns that might beeither beneficial or detrimental to the performance of the weldedstructures. For instance, compressive residual weld stress woulddecrease crack growth rate while tensile residual stress contrib-uted to the high local stress concentration factor (SCF) that oftenled to inter-granular stress corrosion cracking (IGSCC) during ser-vice life [13]. However, fatigue strength of as-welded specimenswas found higher than that of heat-treated specimens due to com-pressive residual stresses induced at the weld toe areas accordingto the work by Kang et al. [14]. Attentions have been paid to thedistribution of tensile and compressive residual stress in weldedjoint. Compressive residual stresses could be obtained on the sur-face of the joint. Compressive residual stress could also be obtainedin the HAZ with the most tensile stress situated away from theweld center in the fusion zone, as was found in welded aluminumplates with thickness of 12 mm by variable polarity plasma arcwelding in single pass [15]. However, according to the numericalinvestigation upon residual stresses in a multi-pass steel weld joint[16], peak tensile stress was identified in the heat affected zone(HAZ) adjacent to the weld bead, which was validated by experi-mental results with X-ray diffraction method. Kong et al. [17]studied the stress distribution of the welded joint obtained byhybrid heat sources of laser and arc. Numerical simulation indi-cated that higher residual stress is distributed in the weld beadand the adjacent heat affected zone (HAZ). Longitudinal and nor-mal components of residual stress showed a bimodal distributionacross the welded joint with a low trough at the weld centre, as

was indicated by the experimental results by Kumar et al. [18].However the axial stress on the outside surface exhibited a dou-ble-valley distribution and the peak of tensile stress was locatedat the weld center. The width of the central trough was indicatedto be significantly greater for higher heat input weld.

Evolution process of stress field became complex under multi-pass welding situation. Finite element (FE) method has been apractical, effective and non-expensive way of capturing the imme-diate and detailed stress distribution during and after welding pro-cess. With regard to the steel joints, the stress distribution could beexplained by the strains related to the austenite to martensite so-lid-state transformation while cooling down as indicated by Kumaret al. [18]. Deng and Kiyoshima [19] focused their attentions on theweld start/end side as numerically investigating residual stress dis-tributions induced by tungsten inert gas arc welding in a steel pipe.It was found that both the hoop and axial stresses strongly variedwith weld passes and the last two passes seemed to contributedmost to the final stress distribution. Brown [20] found that the ten-sile strength was reduced by 7% after the entire weld passes, inaddition, weld metallurgy was only slightly changed in the HAZdue to the overaging caused by the following passes. The peak lon-gitudinal residual stress was reduced with increasing number ofpasses, which could also be found in the work by Jiang et al.[21]. Mark [22] studied the evolution of residual stress in athree-pass steel weld. It was found that the peak tensile residualstresses in the weld became lower as the subsequent weld passedwas finished. The longitudinal residual stress in the weld beadchanged from compressive in the one-pass specimen to tensile inthe three-pass specimen, which was supposed to result from theincrease in transformation start temperature.

Due to the immaturity of the novel welding method, there islittle information available about immediate and detailed stressdistribution and its evolution in welds by the CMT + P MIX proce-dure. The study adopted the CMT + P MIX procedure into weldingthick aluminum plates with narrow gap multi-pass welding meth-od. A model of cooperation between wire feeding and heat sourcewas established under the principle of equivalent input. Compositeheat source model was proposed for the three weld passes. A sym-metric elastic-perfectly-plastic FE model based upon the CMT + PMIX welded butt joint was established. The stress field waspredicted and the thermal mechanical behavior of the joint wasanalyzed, based on which the vulnerable areas were obtained.

2. Experiments

The AA7A52 plates with a size of 100 mm � 50 mm � 20 mm were welded bythree passes with a CMT + P MIX weld source and automatic weld robot. ER5356with a diameter of 1.6 mm was chosen as filler metal and argon with a purity of99.99% was chosen as shielding gas. The chemical composition of the base materialwas shown in Table 1. The plates were assembled with an initial gap of 2 mm. Asimple type of narrow-gap groove with high practicability was made by a groovemachine. The two plates were horizontally fixed and the dimension of the groovewas shown in Fig. 1.

Prior to welding, the surface of the specimen was degreased by wiping with ace-tone and then cleaned with a brush that had stainless bristles. All the three weldpasses were carried out in the same direction along the positive Z axis. The coolingtime was set as 140 s, 140 s and 150 s in sequence after each weld pass so that theinter-pass temperature was kept below 50 �C, thus the whole welding procedurecould be depicted as Fig. 1. Welding thermal cycle of the welded plates wasrecorded with infrared (IR) camera.

Fig. 1. The assemble criterion and welding procedure in CMT + P MIX welding for beveled aluminum plates of 20 mm thick.

Fig. 2. Schematic plan of coordinates establishment and boundary conditions detailed in Section 3.3.2.

226 F. Shu et al. / Construction and Building Materials 54 (2014) 224–235

After welding, a non-destructive examination method with X-ray diffractiontester was utilized to determine the surface residual stress. The results were em-ployed another support for the accuracy of the FE model. Before the measurements,the upper surface of the joint was firstly prepared by mechanical polishing with theSiC water polishing papers to eliminate the effect of oxidation layer, then electro-chemically polished using 10% NaCl in water at 10 V and 1.0 A, maintained for30 s at room temperature, to remove the mechanically affected surface layer. Trans-verse and longitudinal residual stress was measured at different positions (point A,B, and C) and along eight lines, of which the relative location was shown in Fig. 2.The final residual stress distribution was also obtained using blind-hole technique.The experiment was performed with a YC-IV stress analyzer which could calculateand directly print out the residual stress. A strain gauge rosette of type BE120-2CA-K was attached on the carefully polished surface, then a hole with a diameter of2.0 mm was drilled in the center of the strain gauge rosette.

3. Finite element model

A three-dimensional FE model with refined meshes within andaround the fusion zone was developed in the presented investiga-tion. The element size increased progressively with distance fromthe symmetric face. The model was composed of four parts includ-ing the base plates and three weld passes as shown in Fig. 1, corre-spondingly, four collections of meshes were divided. Schematicplan of the model located in the Cartesian coordinates and bound-ary conditions detailed in Section 3.3.2 were shown in Fig. 2.

The numeric analysis was composed of two relevant stepsincluding the thermal cycle analysis and the thermal–mechanicalanalysis. Both steps of the model were calculated as non-linearproblems. The solution of thermal cycle analysis was employedas the thermal load in the subsequent analysis. The convergenceof the numerical model was directly correlated with the meshingmethod, the boundary conditions and the convergence criterion.Residual checking was adopted in our model which was based onthe magnitude of the maximum residual load compared to themaximum reaction force. The method has the benefit that conver-gence can be satisfied without iteration. There maight be noreaction forces which was a drawback in residual checking. Thiscould occur under the condition of free thermal expansion orunavailable mechanical boundary conditions. Then the resolution

automatically turned to displacement checking in either of thesecases, although neither would appear in our research.

3.1. Cooperation between wire feeding and heat input

The weld fillers in the model were added in sequence withdeath-to-birth method. Death-to-birth method had played animportant role in simulation of adding weld fillers as reported byDai [23] and Bonifaz [24]. Death was attributed to all the elementsof each pass first. The elements were given birth in sequence asgetting covered by the arc.

Synchronous evolution of wire feeding speed and heat inputalong time was shown as Fig. 3. A whole cycle of the CMT + PMIX process was composed of a CMT cycle and a pulse cycle. Upto eleven parameters could be independently set manually underthe non-synergic mode. There were eight parameters of the heatinput curve including two peak currents (ICP and IPP) and two basecurrents (ICB and IPB) with four duration time (tCP, tPP, tCB, and tPB).There were three parameters of the wire feeding speed curveincluding feeding speed within the CMT period (FP), withdrawingspeed (FR) and feeding speed within the pulse period (FPP). Whenthe heat input jumped from the peak value (ICP) to the bottom va-lue (ICB) the weld wire was pulled backward. Heat input could beincreased by increasing the peak current (IPP and ICP) or the corre-sponding duration time (tPP and tCP). The main process parameterswere given in Table 2. The frequency of the CMT + P MIX processcould get up to 70 Hz meaning a overwhelmingly high frequencyof droplet transferring. So, there would be 1680 times of droplettransferring (cycle number N in Fig. 3) for a 100 mm length beadwith weld speed set to be 250 mm/min. Subsequently, it wouldbe impossible to divide the meshes into 1680 parts according tothe transferred droplets. Moreover, the quantity of the parametersin the two evolution curves was eleven times of the cycle number,meaning heavy fore treatment work and huge amount ofcomputation.

In order to decrease the cycle number, several cycles of thewelding process (the cycle number assumed to be N as mentioned

Fig. 3. Schemata of synchronous evolution of wire feeding speed and heat input along time The original curve (a) The simplified curve (b).

Table 2Experimental parameters set-up.

ICP/A ICB/A IPP/A IPB/A tCP/s tPP/s tCB/s tPB/s FP/(m/min) FR/(m/min) FPP/(m/min)

350 150 440 360 0.03 0.005 0.01 0.005 11.5 2.0 5.0

F. Shu et al. / Construction and Building Materials 54 (2014) 224–235 227

above) were recombined into one. The novel or simplified cyclewas composed of a novel CMT cycle and a novel pulse cycle. Thepeak and base duration in the simplified curve was N�tCP, N�tPP

and N�tCB, N�tPB, respectively, as shown in Fig. 3(b). The peak andbase values were kept the same as those in the original weld heatevolution curve. Equivalent amount of added weld filler could beassured by properly setting the peak and base wire feeding speed.The base wire feeding speed was set to be zero and the peak wirefeeding speed was hence obtained with Eq. (1).

FAC ¼FP � tCP � FR � tCB

tCP þ tCBð1Þ

3.2. Heat sources

Double ellipsoid body model proposed by Goladk et al. [25] hasbeen the most advanced form among all the heat source modelsincluding classical Rosenthal source, Gauss source and doubleellipsoid source. In the double ellipsoid method the heat sourcewas illustrated as two uniform parts as reported by Fachinottiet al. [26]. Single heat source is competent under the condition ofstable regular arc length, e.g. in thin plate arc welding. Hybrid heatsources model was often utilized in illustrations of composite heatflux (e.g. laser-TIG heat source [27] and laser-GMA heat source[28]), high energy density beam welding and artificial arc shape(e.g. by the way of external magnetic field as reported in literature[10]). According to the presented work, the arc was ignited, main-tained and squeezed between the plates resulting in an elongatedarc along the thickness direction.

Surface of base plates got melted as covered by the arc and thencooled down as the arc passed by. Weld bead formed as transferreddroplets flowed to the cooled molten pool just opposite to weldingdirection under the control of surface tension. The molten depthwas 10 mm, 6 mm and 4 mm in the three subsequent weld passes,respectively as shown in Fig. 1. It was hereby necessary for the arcto fluctuate periodically along the thickness direction so as to guar-antee available fusion of the groove wall. With regard to the singleheat source the arc was stable without length fluctuation. Lack offusion would arise from the low current density due to the in-creased current flow area covered by the arc. It was therefore veryliable that the single heat source model was unqualified for thepresent investigation. The energy density was balanced by dividing

the single heat source into two throughout each weld pass. Theoriginal points of heat fluxes were situated at the middle and theupper surface of each bead, thus a total of six points were presetcorresponding to the six double ellipsoid heat fluxes.

3.3. Material properties and boundary conditions

3.3.1. Constitutive model of materialsBesides of the welding parameters, residual stress was signifi-

cantly influenced by dissimilar materials as indicated by Josephet al. [29], thus it was essential to establish the model accordingto the base metal and the filler metal. Constitutive equations wereestablished by Liu [30] on basis of rheological behaviors of highstrength aluminum alloys. A visco-elasto-plastic model forAA2024 was developed and the computed transverse strain wasemployed to predict weld centering cracking. According to theinvestigation by Wu et al. [31], the true stress–strain curve at roomtemperature for high strength aluminum alloys was composed ofan elastic stage and a plastic stage. It was characterized by absenceof obvious yielding process between the stages and slight strain-hardening process after yielding. Besides, strain-softening tendedto appear as experimental temperature increased as a result ofenhanced recovery and re-crystallization as reported by Ma andDen Ouden [32].

The constitutive equations usually supply an expression for therate of stress in terms of deformation rate. With regard to thewelding thermal cycle, the temperature ranged from room temper-ature to above the molten point of welded materials. Moreover,plastic strain was expected to occur mainly in high-temperaturerange. So the material during heating process of welding behavedalmost perfectly plastic, while the increase in strength during cool-ing process should recieve enough attentions. Thus, the plasticstage of constitutive curve was then defined as perfectly plasticstyle, that is, no strain hardening. The constitutive curve of weldedmaterials was defined as an elastic-perfectly-plastic one in the pre-sented work. We needed only specify the elastic constants and theyield stress. The model could be simplified as the elastic stage andthe plastic stage. The two stages were divided by the yieldingstrain and the yielding stress and expressed in the sectionedequation:

r ¼ Ee when e 6 ey

r ¼ ry ¼ Eey when e P ey

�ð2Þ

228 F. Shu et al. / Construction and Building Materials 54 (2014) 224–235

in which E, ey and ry were the elastic module, the yielding strainand the yielding strength, respectively. The elastic module wasthe function of temperature (T), which could be expressed by Eq.(3):

EðTÞ ¼ E0 � wðTÞ ð3Þ

in which E0 and w(T) were the elastic module under room tem-perature and the magnitude of the elastic module compared to thatunder room temperature. Then equation (1) could be transformedinto another form and expressed by Eq. (4):

r ¼ E0wTe ¼ r0wT when e 6 ey

r ¼ ry ¼ E0wTey ¼ r0;y;wT when e P ey

�ð4Þ

in which r0 and r0, y were the stress under room temperature andthe yield stress under room temperature, respectively.

With regard to the base and filled material, the correspondingyield strength and Young’s module under different temperatureswere defined according to the work by Zhan et al. [33] and Alankaret al. [34], respectively. The yielding strength of AA7A52 under dif-ferent temperature could be plotted as Fig. 4. Then, the ratio ofyielding strength under different temperatures to that under roomtemperature could be obtained and plotted in Fig. 4. The advantageof the numerical simulation lay in the interpolation method whichdirectly gave the function value rather than by solving the complexfitting equations. Interpolation curve was used to adopt the consti-tutive relation into the FEM model. The final constitutive curves for

280 300 320 340 360 380 400 420 440 46040

50

60

70

80

90

100

110

120 Yield strength Strength ratio at different temperatures

Temperature (°C)

Yie

ld s

tren

gth

(MPa

)

10

12

14

16

18

20

22

Strength ratio at different temperatures (%

)

Fig. 4. The yield strength of AA7A52 under different temperatures and thecorresponding strength ratio to that under room temperature (20 �C).

0.0 0.2 0.4 0.6 0.8 1.00.00E+000

3.00E+007

6.00E+007

9.00E+007

1.20E+008

1.50E+008

1.80E+008

20°C

1350°C

(b(a)

Strain-δ ( )

Stre

ss- σ

(MPa

)

%

Fig. 5. Constitutive of filler metal ER5356 (a) and base

the base metal and the filler metal were obtained as shown inFig. 5(a and b), respectively.

With regard to application of the established model, severalassumptions were proposed. Firstly, the materials were isotropic.Secondly, the yield strain for both the base metal and the filled me-tal was assumed to be 0.2%. Thirdly, residual strain which was as aresult of the yielding stress should be ignored, because the yieldingprocess was ignored in the constitutive model. Fourthly, the dis-count of strength caused by high temperature for both the stageswas in consistence. Above all, the constitutive equations shouldbe used only in specific processing techniques with rapid coolingspeed such as welding, quenching and hot rolling.

Property parameters concerning with the temperature field andstress field were illustrated as functions of temperature as shownin Fig. 6. Moreover, the property data were non-linearly extrapo-lated to high temperature ranges. The latent heat of materialswas considered in the way of equivalent specific heat capacityaccording to the work by Kousksou et al. [35].

3.3.2. Thermal and mechanical boundary conditionsExterior boundary conditions gained importance through the six

exterior surfaces, of which five were assumed to be in contact withambient air and one fixed onto the copper liner plates. Heat lossthrough air included nature convection and heat conduction forthe four vertical side walls and the upper surface [36]. With regardto the lower surface tightly pressed onto the copper liner plate,conduction played a dominant role in heat elimination, thus thelower heat conduction factor among the contact plates was selectedas the heat conduction factor. The welded plates were clamped by aself-locking device in four corners, as shown in Fig. 2. Hence,corresponding nodes were fixed with no displacement to occur.Additionally, the displacement constraint along thickness directionwas applied for nodes in contact with the liner plate.

4. Results and discussions

Four points marked as A, B, C and R were selected and situated asshown in Fig. 7. Point A and C was situated on the top of the first andsecond weld pass, respectively, with point B located 1 mm beneathpoint A. Stress distribution along several specific paths was discussedand utilized. The location and direction of the seven lines was shownin Fig. 7. The other path marked as 8 was shown in Fig. 2. The residualstress was tested with blind-hole method in the upper surface alongline 4 and line 5, because satisfactory measurement results could beachieved by this method providing the residual stress did not exceed50–80% of the material yield stress [37].

0.00E+000

1.00E+008

2.00E+008

3.00E+008

4.00E+008

20°C

1350°C

)

Stre

ss- σ

(MPa

)

0.20.0 0.4 0.6 0.8 1.0

Strain-δ ( )%

material AA7A52 (b) under different temperatures.

0 100 200 300 400 5000.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0(b)(a)

Temperature (°C)

Prop

erty

par

amet

ers

of A

A 7

A52 Specific heat capacity(×1132 J•Kg-1•K-1)

Young's module(×2e11 Pa)

Entropy(×1.697e9 J•k-1)

Thermal conduct(×180 W•m-1•K-1) Thermal expansion (×2.676e-5)

Prop

erty

par

amet

ers

of E

R53

56

Young's module(×6.7e10MPa) Thermal expansion(×2.61e-5)

Thermal conduct(×138W•m-1•K-1)

Specific heat capacity(×1089J•Kg-1•K-1)

0 100 200 300 400 500

Temperature (°C)

Fig. 6. Property parameters changing along temperature for the filled metal ER5356 (a) and the base metal AA7A52 (b).

Fig. 7. Location of the selected points (point A, B, C and R) and the seven research path with the direction defined by the arrow.

0 100 200 300 400 500 600 7000

306090

120150180

Tem

pera

ture

(o C

)

Time (s)

Simulation Experimental

Fig. 8. Temperature evolution along time at point A obtained with IR camera and FEmethod.

F. Shu et al. / Construction and Building Materials 54 (2014) 224–235 229

4.1. Comparison of simulation and experiment results

There was no effective way of directly measuring the principalresidual stresses by experimental test method. What’s more,similar distribution between the max principal stress and thetransverse stress was found with reference to the computationalresolution as discussed in 4.3. Even though the failure judgmentwas dependent upon elemental principal residual stresses ratherthan compositional stresses along any axis, according to the classi-cal strength theories, compositional stresses along the coordinateaxes were firstly researched instead of principal stresses.

Good agreement between numerical and experimental resultswas found in temperature evolution of a specific point (point Rin Fig. 2) as shown in Fig. 8. Distributions of final transverse andlongitudinal residual stresses along line 2, line 3, line 4, line 5, line6 and line 7 were shown in Fig. 9. Agreement with comprehensiblediscrepancy was also found in the distribution of transverse andlongitudinal stresses along line 4, line 5, line 6 and line 7 in theupper surface as shown in Fig. 9(e and f). Deficiencies in the

experiment such as shortage of the tested points and linear fittingof the obtained data together with the assumptions by simulationsuch as the isotropic character for materials could result in discrep-ancy between the simulated and the experimental results.

It was indicated by Fig. 9 that transverse residual stress wasgenerally higher than longitudinal residual stress. Concentrationof transverse residual stress to the weld bead and the adjacent areawas evident as indicated by the distribution along line 3 and line 5.Transverse residual stress along line 6 was mainly tensile exceptfor a short sect of compressive status at the arc starting side. Itwas evident that the value was much bigger in the bottom surfacethan that in the upper surface for both transverse and longitudinalstress. Transverse and longitudinal stress was found in synchro-nous trend between line 3 and line 5, line 4 and line 6. Accordingto the work by Dong [38], stress distribution could be divided intotwo types including the self-equilibrating type and the bendingtype. Distribution of both kinds of stresses in line 4 and line 6was shown to be of self-equilibrating type. The distribution oflongitudinal stress was found in synchronous trend between line5 and line 7.

4.2. Evolution of residual stress

It was told by the global contour bands of compositional stres-ses at different time steps that peak extreme value for all thecompositional stresses was presented in the symmetric face ofthe joint. Highest transverse stress, longitude stress and stressalong thickness were tensile, compressive and compressive,respectively as shown in Fig. 10. Contour band of compositionalstress distribution in the symmetric face was given below the dashline in. It was indicated that tensile transverse stress got trans-ferred from the arc-starting side to the arc-ending side as the first

0.00 0.01 0.02 0.03 0.04 0.05

0.00E+000

5.00E+007

1.00E+008

1.50E+008

2.00E+008

2.50E+008

3.00E+008

3.50E+008

4.00E+008

0.00 0.02 0.04 0.06 0.08 0.10-2.00E+008

-1.50E+008

-1.00E+008

-5.00E+007

0.00E+000

5.00E+007

1.00E+008

1.50E+008

0.00 0.01 0.02 0.03 0.04 0.05

-9.00E+007

-6.00E+007

-3.00E+007

0.00E+000

3.00E+007

6.00E+007

9.00E+007

1.20E+008

1.50E+008

0.00 0.02 0.04 0.06 0.08 0.10

-1.00E+008

-5.00E+007

0.00E+000

5.00E+007

1.00E+008

1.50E+008

2.00E+008

2.50E+008

3.00E+008

0.00 0.01 0.02 0.03 0.04 0.05-1.50E+008

-1.00E+008

-5.00E+007

0.00E+000

5.00E+007

1.00E+008

1.50E+008

2.00E+008

2.50E+008

3.00E+008

3.50E+008

4.00E+008

0.000 0.005 0.010 0.015 0.020-1.50E+008

-1.00E+008

-5.00E+007

0.00E+000

5.00E+007

1.00E+008

1.50E+008

2.00E+008

2.50E+008

3.00E+008

3.50E+008

4.00E+008

Distance (m)

Res

idua

l str

ess

(Pa)

Transverse residual stress Longitudinal residual stress

(c)

Res

idua

l str

ess

(Pa)

Transverse residual stress Longitudinal residual stress

Longitudinal_Blinde hole

Transverse_Blinde hole

Longitudinal_Blinde hole

Transverse_Blinde hole

(d)

Res

idua

l str

ess

(Pa)

Transverse residual stress Longitudinal residual stress

(e)

Res

idua

l str

ess

(Pa)

Transverse_Simulation Transverse_X-ray diffraction

Longitudinal_Simulation Longitudinal_X-ray diffraction

(f)

Res

idua

l str

ess

(Pa)

Transverse_Simulation Transverse_X-ray diffraction

Longitudinal_Simulation Longitudinal_X-ray diffraction

(a)

Distance (m)

(b)

Res

idua

l str

ess

(Pa)

Transverse residual stress Longitudinal residual stress

Distance (m)Distance (m)

Distance (m)Distance (m)

Fig. 9. Simulated final distribution of transverse and longitudinal residual stress along line 2(a), 3(b), 4(c), 5(d), 6(e), 7(f) and experimental data of stress distribution alongline 6(e) and line 7(f).

230 F. Shu et al. / Construction and Building Materials 54 (2014) 224–235

pass was finished, and gradually concentrated as the other twopasses got finished. No evident change was found for the longitu-dinal residual stress besides growing compressive residual stress.The area of tensile stress along thickness got enlarged and concen-trated to the center of the whole joint with a decreasing value,meanwhile, the compressive stress area stayed almost unchangedwith an increasing value. The highest tensile and compressivestress for every time step was the transverse stress and stress alongthickness, respectively.

Contour band for residual stresses was an effective way ofdescribing the residual stress field globally. More informationcould be obtained by stress distribution along different lines at dif-ferent time steps. To better understand the evolution of residualstresses in detail, especially the influences of different passes onthe stress field, distribution of transverse stress along line 1andline 2 at different time steps was shown in Fig. 11.

It was shown by Fig. 11 that distribution of residualtransverse stress along the two lines at different time made a

self-equilibrating type of stress. The tensile and compressivestress reached the peak value as the first pass got finished. Stressevolution at different positions may be different. For instance,tensile stress value in the weld root got firstly decreased afterthe second pass, and kept increasing after the last pass till finallygot a new peak point, while compressive stress at the top surfacehad kept decreasing.

Distributions of transverse residual stress along line 2 immedi-ately before and after each inter-pass cooling were shown inFig. 11(b). It was shown that stress value for the arc-starting sidewas obvious smaller than that of the arc-ending side. The stressturned to be tensile and the value increased after cooling down.Peak point of residual stress was identified at the top of the weldbead immediately after the first inter-pass cooling, which wasdifferent from the position of the peak point after the second orthe third inter-pass cooling. In another word, the point with peaktensile stress value transferred from the prior peak point (pointA) to 1 mm beneath (point B).

Fig. 10. Contour bands of compositional stresses along coordinates axes at different time steps.

F. Shu et al. / Construction and Building Materials 54 (2014) 224–235 231

4.3. Impact of the subsequent weld passes

As was shown in Fig. 11, the changes in value were broughtabout by the last two weld passes, but no essential change evertook place since the first inter-pass cooling got finished, whichwas also found for the longitudinal stress and stress alongthickness. Fig. 12 showed the original transverse residual stressdistribution immediately before and after the second inter-passcooling along line 8 (respective time step of 240 and 390). The ori-ginal residual stress distribution was that of 190 s, when the first

weld pass was just finished. Though stress distribution was chan-ged at the time step of 240 s, the final residual stress distributionat time step of 390 s paralleled well with the original distribution,so did that at final time step. Similarly, Liu et al. [39] found that thefirst passes had a significant impact on the formation of the finalresidual stress distribution. Moreover, the stress value almost keptunchanged after the weld bead was formed to a specific thickness.

Definitely local residual stress was interlocked to two plasticstrain steps as indicated by Paddea et al. [40], during which plasticstrain was opposite in sign. For example, during the first pass the

Fig. 11. Distribution of residual transverse stress along line 1 (a) and line 2 (b) at different time steps.

0.00 0.02 0.04 0.06 0.08 0.10

-9.00E+007

-6.00E+007

-3.00E+007

0.00E+000

3.00E+007

6.00E+007

9.00E+007 440 840

Com

p 11

of

Stre

ss (

Pa)

Distance (m)

190 240 390

Fig. 12. Transverse stress distribution along line 8 at different time steps.

232 F. Shu et al. / Construction and Building Materials 54 (2014) 224–235

molten pool at the arc-starting side was firstly restricted in expan-sion, thus compressive plastic strain was yielded firstly. As coolingdown shrinkage took place in the molten pool and got constrained,while the increased strength added to the constraint. As a result,tensile residual stress was yielded. The compressive and tensileresidual strain in the first and the second step was assumed tobe d1 and d2, respectively. So, the residual stress rr could be simplydescribed by Eq. (5).

rr ¼ �� ðd1 þ d2Þ ð5Þ

From the local point of view, if tensile residual strain was big-ger, residual strain would be tensile. With ignorance of structuraleffect, residual compressive strain of the molten pool from the firststep would be smaller than the tensile residual strain from thesecond step. So, the residual stress at both the arc-starting andarc-ending side should be tensile. However, according to the resid-ual stress along line 8, the residual stress at the arc-starting sidewas compressive after cooling down as shown in Fig. 12. Theresidual stress was essentially caused by plastic strain and thedistribution was dependent upon the structure. Compressive stressyielded interdependent upon the tensile stress and the role was tokeep the structural balance.

Evolution of the transverse residual stress at point A, B and Cwas plotted in Fig. 13. Residual stress for all the points keptincreasing during cooling down. However, for points from the firstpass it was only within the first inter-pass cooling that the residualstrain grew bigger along cooling time, thus the increased residualstress indicated by Eq.(5) was a result of the increased strength.The residual strain for point A sharply dropped when time stepreached 240 s, as shown in Fig. 13(b). This could be explained bythe first step, during which the yielded compressive strain count-erweighed part of the residual tensile strain from the first pass.The phenomenon that attracted us was that the strain for all theremelted point went upward and kept almost horizontal as themarked section in Fig. 13. Arise of tensile residual strain could be

attributed to the shrinkage of the solidified metal. The global struc-tural factor was supposed to help reduce the residual tensile strainthrough counteracting the effect of shrinkage process. The secondstep was absent from the strain evolution curve for both point Aand point B during the third pass, because no phase transformationtook place. During the third pass the residual strain increased firstand then decreased to the original value with little plastic strainyielded for point A and B. Bigger compressive residual strainyielded within the first step for point A during the second pass.As a result, most reduction in residual strain was found for pointA, which could explain the transferring of the peak point from Ato B. Similar impact by the third pass could also be found on thesecond filled pass.

Little change in the residual stress of point B rather than point Awas brought about by the second pass. As conclusion, evidentinfluences by the second pass on the stress distribution could befound only near the top surface of the first weld bead.

4.4. Vulnerable area of the joint

The importance of numerical method lays upon the predictionof test results and assistance to theoretical demonstration in addi-tion to lower cost [41]. In the present study vulnerable areas of thejoint was evaluated under the conduction of strength theories. Fail-ure of the joint was interlocked to stress state of investigated area.The first classical strength theory was therefore adopted for thesafety assessment of material under tri-axial tensile stress condi-tion and the fourth classical strength theory for the other stressstatus.

Contour bands of the principal stresses were shown in Fig. 14.The areas under tri-axial tensile stress condition included the fourcorners under rigid fixation and the molten boundaries betweenthe base plate and the filled metal. Neither crystal crack nor hotcrack was supposed to occur at the four rigidly fixed cornersbecause of the relatively low peak temperature during its thermalcycling as shown in Fig. 8. Tri-axial tensile stress was found alongthe boundary lines resulting in tendency for tensile failure underwork conditions. The other vulnerable area judged by the fourthstrength theory was situated at the symmetric line, where thestress style was of tension-crush style, thus it was the sign of theprincipal stresses that made equivalent Von Mises stress biggerthan that along the molten boundary.

The zones with maximum residual equivalent Von Mises stresstransferred from the arc starting side to the arc ending side as thefirst welding pass was finished, as indicated by its distributionalong line 4. With regard to line 2, the peak value point transferredfrom point A to point B which stayed as the peak point of line 2 tillthe end of the procedure. Stress distribution along line 4 was

Fig. 13. Evolution of residual transverse stress (a) and strain (b) along time at point A, B and C.

Fig. 14. Contour bands of final tri-axial principle stresses.

Fig. 15. Distribution of equivalent Von Mises stress along line 2 (a), 4 (b) and 6 (c) at different time steps.

F. Shu et al. / Construction and Building Materials 54 (2014) 224–235 233

characterized with two peaks at both ends, and the peak value atthe arc ending side was nearly twice of that at the arc starting side.The maximum value along line 6 was found much smaller thanthat of line 1 and line 2, as shown in Figs. 15 and 16, respectively.As conclusion, the top point (A) of the first weld pass at the arcending side was regarded to be most vulnerable just before startingthe second weld pass.

Relationship between equivalent Von Mises stress and trans-verse residual stress could be explained through the maximum

principal stress. It was shown in Fig. 16 that distribution of themax principal stress was mainly dependent upon the first weldpass. Consistency was found between evolution of the maximumprincipal stress and transverse residual stress through the compar-ison of Figs. 16 and 10. The highest tensile stress for every timestep was the transverse stress, as discussed in 4.2, thus the maxi-mum principal stress approximated transverse stress. So it madesense to identify the vulnerable area of the joint directly throughtransverse residual stress.

Fig. 16. Contour bands of max principal stress distribution at the arc-ending side after every inter-pass cooling.

234 F. Shu et al. / Construction and Building Materials 54 (2014) 224–235

5. Conclusions

A required model for simulating the characteristic cooperationbetween wire feeding and heat input was developed. The numeri-cal results for thermal cycling and stress distribution along thesymmetric line on the upper surface and its vertical line were ingood agreement with the experimental results.

(1) Final residual stress distribution was analyzed based uponstress distribution along different lines. Un-uniformity wasfound in the stress field with stress concentrated to the weldbead and the adjacent area.

(2) The evolution of compositional residual stress was firstlyanalyzed by the way of contour bands. It was found thatthe area with maximum extreme value for all the composi-tional stresses was situated in the symmetric face of thejoint; transverse stress at the arc-starting side was muchsmaller than that of the arc-ending side after cooling downto room temperature. The impact of the subsequent weldpasses on the residual stress distribution was studiedthrough evolution of transverse residual stress distributionalong the symmetric lines at both the arc-starting and arc-ending side. Slight change in global stress distribution tookplace since the first weld pass got cooled down in spite ofsharper decreasing of residual stress in the re-melted zonethan that in the HAZ. The mechanism was studied throughthe yielding process of residual strain which was dividedinto two steps. Residual strain yielded was bigger duringthe first step and smaller during the second step for there-melted area than for the heat affected area, leading tomore stress release of residual stress. The area where resid-ual stress could be partially released with the help of the lasttwo weld passes was limited by the moderate heat input ofthe CMT + P process.

(3) The vulnerable area of the joint was obtained under the con-duction of different strength theories. According to the firstclassical strength theory, the molten boundaries betweenthe base plate and the filled metal were vulnerable as aresult of the tri-axial tensile stress. The top of the first filledpass at the arc ending side before starting the second passwas supposed to be most vulnerable with reference to thefourth classical strength theory.

In addition, composition of the molten area and the correspond-ing properties was out of consideration in the present investiga-tion, meaning ignorance of smooth transition between the basemetal and the filled metal. According to the work by Koo et al.

[42] material mismatch could lead to different crack locations bychanging the plastic g factors. Thus it was necessary for thenumerical results to be qualitatively rectified and more workshould be done along this line.

Acknowledgements

This work was supported financially by the major program ofState Key Laboratory of Remanufacturing (Grant No. 9140C850-205120C8501) and the army foundation project of China.

References

[1] Shang J, Wang KH, Zhou Q, Zhang DK, Huang J, Li GL. Microstructurecharacteristics and mechanical properties of cold metal transfer welding Mg/Al dissimilar metals. Mater Des 2012;34:559–65.

[2] Cao R, Yu G, Chen JH, Wang PC. Cold metal transfer joining aluminum alloys-to-galvanized mild steel. J Mater Process Technol 2013;213:1753–63.

[3] Pickin CG, Williams SW, Lunt M. Characterisation of the cold metal transfer(CMT) process and its application for low dilution cladding. J Mater ProcessTechnol 2011;211:496–502.

[4] Henderson ID, Steffens HD. Fracture toughness of narrow-gap welded joints inthe nuclear pressure vessel steel 22NiMoCr37. Nucl Eng Des 1976;36:273–85.

[5] Zhang XP, Dorn L. Investigation on the possibility of using the microshear testas a surveillance method to estimate the mechanical properties and fracturetoughness of nuclear pressure vessel steel, A508CL3, and its joints welded bynarrow-gap submerged-arc welding. Int. J Pres Ves Pip 1999;76:35–41.

[6] Biswas P, Mandal NR, Vasu P, Padasalag SB. A study on port plug distortioncaused by narrow gap combined GTAW & SMAW and electron beamwelding.Fusion Eng Des 2011;86:99–105.

[7] Jang CH, Cho PY, Kima M, Oh SJ, Yang JS. Effects of microstructure and residualstress on fatigue crack growth of stainless steel narrow gap welds. Mater Des2010;31:1862–70.

[8] Sattari-Far I, Farahani MR. Effect of the weld groove shape and pass number onresidual stresses in butt-welded pipes. Int J Pres Ves Pip 2009;86:723–31.

[9] Meister RP, Martin DC. Narrow-gap welding process. Brit Weld J1966;13:252–7.

[10] Starling C, Marques PV, Modenesi PJ. Statistical modelling of narrow-gap GTAwelding with magnetic arc oscillation. J Mater Process Technol1995;51:37–49.

[11] Cho DW, Na SJ, Cho MH, Lee JS. A study on V-groove GMAW for variouswelding positions. J Mater Process Technol 2013;213:1640–52.

[12] Chen J, Schwenk C, Wu CS, Rethmeier M. Predicting the influence of grooveangle on heat transfer and fluid flow for new gas metal arc welding processes.Int J Heat Mass Transfer 2012;55:102–11.

[13] Liu XD, Frankel GS, Zoofan B, Rokhlin SI. In-situ observation of intergranularstress corrosion cracking in AA2024-T3 under constant load conditions. CorrosSci 2007;49:139–48.

[14] Kang HT, Lee YL, Jackson-Sun XC. Effects of residual stress and heat treatmenton fatigue strength of weldments. Mater Sci Eng A 2008;497:37–43.

[15] Zhang Y, Ganguly S, Edwards L, Fitzpatrick ME. Cross-sectional mapping ofresidual stresses in a VPPA weld using the contour method. Acta Mater2004;52:5225–32.

[16] McDonald EJ, Hallam KR, Bell W, Flewitt PEJ. Residual stresses in a multi-passCrMoV low alloy ferritic steel repair weld. Mater Sci Eng A 2002;325:454–64.

[17] Kong FR, Ma JJ, Kovacevic R. Numerical and experimental study of thermallyinduced residual stress in the hybrid laser–GMA welding process. J MaterProcess Technol 2011;211:1102–11.

F. Shu et al. / Construction and Building Materials 54 (2014) 224–235 235

[18] Kumar S, Kundu A, Venkata KA, Evans A, Truman CE, Francis JA, et al. Residualstresses in laser welded ASTM A387 Grade 91 steel plates. Mater Sci Eng A2013;575:160–8.

[19] Deng D, Kiyoshima S. Numerical simulation of welding temperature field,residual stress and deformation induced by electro slag welding. ComposMater Sci 2012;62:23–34.

[20] Brown R, Tang W, Reynolds AP. Multi-pass friction stir welding in alloy 7050–T7451: Effects on weld response variables and on weld properties. Mater SciEng A 2009;513:115–21.

[21] Jiang WC, Wang BY, Gong JM, Tu ST. Finite element analysis of the effect ofwelding heat input and layer number on residual stress in repair welds for astainless steel clad plate. Mater Des 2011;32:2851–7.

[22] Mark AF, Francis JA, Dai H, Turski M, Hurrell PR, Bate SK, et al. On the evolutionof local material properties and residual stress in a three-pass SA508 steelweld. Acta Mater 2012;60:3268–78.

[23] Dai K, Shaw L. Thermal and stress modeling of multi-material laser processing.Acta Meter 2001;49:4171–81.

[24] Bonifaz EA, Richards NL. Modeling cast IN-738 superalloy gas tungsten arcwelds. Acta Mater 2009;57:1785–94.

[25] Goladk J, Chakravariti A, Bibby M. A new finite element model for welding heatsources. Metall Mater Trans B 1984;15:299–305.

[26] Fachinotti VD, Cardona A, Baufeld B, Van der Biest O. Finite-element modellingof heat transfer in shaped metal deposition and experimental validation. ActaMater 2012;60:6621–30.

[27] Zeng Z, Li XB, Miao YG, Wu G, Zhao ZJ. Numerical and experiment analysis ofresidual stress on magnesium alloy and steel butt joint by hybrid laser-TIGwelding. Compos Mater Sci 2011;50:1763–9.

[28] Bang HS, Bang HS, Kim YC, Joo SM. Analysis of residual stress on AH32butt joint by hybrid CO2 laser-GMA welding. Compos Mater Sci 2010;49:217–21.

[29] Joseph A, Rai SK, Jayakumar T, Murugan N. Evaluation of residual stresses indissimilar weld joints. Int J Pres Ves Pip 2005;82:700–5.

[30] Liu WP. Computational analysis and prediction of weld-solidification cracking.Compos Mater Sci 1995;4:211–9.

[31] Wu B, Li MQ, Ma DW. The flow behavior and constitutive equations inisothermal compression of 7050 aluminum alloy. Mater Sci Eng A2012;542:79–87.

[32] Ma T, Den Ouden G. Softening behavior of Al–Zn–Mg alloys due to welding.Mater Sci Eng A 1999;266:198–204.

[33] Zhan LH, Lin JG, Dean AT, Huang MH. Experimental studies and constitutivemodelling of the hardening of aluminium alloy 7055 under creep age formingconditions. Int J Mech Sci 2011;53:595–605.

[34] Alankar A, Wells AM. Constitutive behavior of as-cast aluminum alloysAA3104, AA5182 and AA6111 at below solidus temperatures. Mater Sci EngA 2010;527:7812–20.

[35] Kousksoua T, Jamil A, Zeraouli Y. Enthalpy and apparent specific heat capacityof the binary solution during the melting process: DSC modeling. ThermochimActa 2012;541:31–41.

[36] Holman JP. Heat Transfer. McGraw-Hill Book Company; 1986.[37] ASTM E837-01. Standard Test Method for Determining Residual Stress by the

Hole Drilling Strain Gage Method, American Society for Testing and Materials,Philadelphia, Pennsylvania; 2001.

[38] Dong P. On the mechanics of residual stresses in girth welds. J Press Vess-TASME 2007;129:345–54.

[39] Liu C, Zhang JX, Xue CB. Numerical investigation on residual stress distributionand evolution during multipass narrow gap welding of thick-walled stainlesssteel pipes. Fusion Eng Des 2011;86:288–95.

[40] Paddea S, Francis JA, Paradowska AM, Bouchard PJ, Shibli IA. Residual stressdistributions in a P91 steel-pipe girth weld before and after post weld heattreatment. Mater Sci Eng A 2012;534:663–72.

[41] Cool T, Bhadeshia HKDH, MacKay DJC. The yield and ultimate tensile strengthof steel welds. Mater Sci Eng A 1997;223:186–200.

[42] Koo JM, Huh Y, Seok CS. Plastic g factor considering strength mismatch andcrack location in narrow gap weldments. Nucl Eng Des 2012;247:34–41.