International Journal of Solids and Structures 80 (2016) 420429

Contents lists available at ScienceDirect

International Journal of Solids and Structures

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

The inuence of welding procedure and plate geometry on residual

stresses in thick components
M.E. Kartal a,b,, Y-H. Kang c, A.M. Korsunsky a, A.C.F. Cocks a, J.P. Bouchard d
a
Department of Engineering Science, Oxford University, Parks Road, Oxford OX1 3PJ, UK
b
Lloyds Register Foundation Centre for Safety & Reliability Engineering, School of Engineering, University of Aberdeen, Aberdeen AB24 3UE, UK
c
Technical Research Laboratories, POSCO, Pohang 790-300, South Korea
d
Materials Engineering, The Open University, Milton Keynes MK7 6AA, UK

a r t i c l e i n f o a b s t r a c t

Article history: This study compares the residual stress proles generated by two different welding techniques (multi-pass
Received 10 September 2014 and single pass welding) in 80 mm thick ferritic steel welds. As residual stresses in such thick welded
Revised 30 September 2015
steel specimens that are representative of important industrial processes are to be determined, the contour
Available online 22 October 2015
method as a practical choice is employed. The results show that the residual stress distribution is sensitive
Keywords: to the details of the welding process, with the maximum tensile residual stress being located in the weld for
Residual stresses the multi-pass weld and in the heat-affected zone for the single pass weld. However, the results obtained
The contour method from the contour method are only valid at the cut plane. In order to predict residual stress variations along
Eigenstrain the depth of the welded specimens from the free surface, the inverse eigenstrain technique is extended. The
Welding eigenstrain-based approach allows us to evaluate the minimum specimen length required for the accurate
Steel estimate of residual stresses in large-scale engineering structures. This technique also allows predicting the
correct magnitude of residual stresses in the full length specimen even if a shorter specimen is used for
experiments. Finally, multiple cut contour measurements are made on the multi pass specimen in order to
compare the distributions of residual stress in the longitudinal and transverse directions.
2015 Elsevier Ltd. All rights reserved.

1. Introduction toughness of components and assemblies as described by Withers

(2007). Residual stress effects are especially signicant when the
Welding is one of the most common techniques used in indus- component is subjected to cyclic loading. Due to the combination
trial assembly to join together components made from similar or dis- of microstructural modications and tensile residual stresses, weld-
similar materials, as described by Weman (2003). As with other heat ments are generally the critical regions of structural components,
treatment processes, such as quenching, hot rolling or forging, weld- which are particularly susceptible to fatigue failure. Many experi-
ing modies the mechano-chemical state of materials (phase compo- mental studies have found that fatigue cracks initiate in and around
sition, solute distribution, etc.). Residual stresses are also developed weld regions, as was concluded by Gurney (1979). In order to derive
in and around the weld region due to temperature gradients gener- reliable estimates of the fatigue life, a quantitative knowledge of the
ated between the weld, heat affected zones and parent material dur- residual stress distribution is essential. A range of techniques can be
ing welding and cool-down. The elevated temperature experienced employed to obtain predictions of the fatigue performance on the ba-
during welding results in local volume expansion in and around the sis of the constitutive relationships for the weld region, heat-affected
fusion zone that is constrained by the cooler surrounding material, zone (HAZ) and the parent material, together with the residual stress
causing thermal stresses that may result in yielding. As the weld bead information. For instance, the effect of the weld residual stresses on
shrinks during solidication, it is again constrained by the colder sur- the fatigue performance has been studied by Liljedahl et al. (2008)
rounding material. This causes tensile stresses in the weld region that and Liljedahl et al. (2010).
are balanced by compressive residual stresses in the region further Residual stresses are particularly signicant in thick welded
away from the weld (Withers and Bhadeshia, 2001b). plates, such as those used in the manufacture of ship hulls. The con-
It is well-established that tensile residual stresses exert a strong ventional welding method currently employed by shipbuilders in-
inuence on the structural integrity, fatigue strength and fracture volves low heat input (less than 2 kJ/mm) multi-passes, which has
been certied for several decades. However, in order to increase pro-
ductivity and eciency of fabrication, new welding techniques are

Corresponding author. Tel.: +44 1224 272495. being proposed and evaluated with the objective of reducing resid-
E-mail address: [email protected] (M.E. Kartal). ual stresses and improving fatigue performance. These techniques

http://dx.doi.org/10.1016/j.ijsolstr.2015.10.001
0020-7683/ 2015 Elsevier Ltd. All rights reserved.
 M.E. Kartal et al. / International Journal of Solids and Structures 80 (2016) 420429 421

generally involve a single pass with ultra-high heat input (more than Mul-pass Welding Single-pass Welding
60 kJ/mm). Since these single pass, high heat input welding tech-
niques are relatively new, few comparisons can be found in the
(a)
literature against conventional methods. Moreover, very few resid-
ual stress measurement techniques summarised by Withers and
Bhadeshia (2001a) and Withers et al. (2008) are available that are
well-suited for bulk stress evaluation in thick steel plates (up to
80 mm) weighing hundreds of kilograms.
Recently, Woo et al. (2013) presented a comparison of residual (b)
stress distributions in 70 mm thick and 230 mm long ferritic steel
welds created by multi-pass (1.7 kJ/mm) and single pass (56 kJ/mm)
processes by using hole drilling, neutron diffraction and contour
methods. A signicant difference in the residual stress distributions
was found between the samples fabricated using the two different
welding techniques. Unfortunately, the results reported by Woo et al.
(2013) were obtained using relatively short sample coupons. As a
consequence, the calculated magnitudes of the residual stresses ob- Fig. 1. (a) Geometries of the grooved plates before welding. (b) Macrographs showing
the fusion boundaries of the welds.
tained from the analysis were lower than those generated in large-
scale components typical of those employed the shipbuilding in-
dustry. The eigenstrain-based modelling framework presented here The principal aims of this work were to carry out a comparative
allows this important sample length effect to be understood and study of two different welding processes, specically, to apply the in-
quantied. A further detailed overview of this issue is given in verse technique of eigenstrain and compare the results with the con-
Sections 6 and 7. ventional contour method, and to determine the minimum specimen
Although non-destructive physical techniques such as neutron or length that allows correct residual stress evaluation relevant to large-
synchrotron X-ray diffraction are increasingly being used for mea- scale components using the contour method. In addition, in order to
suring complex three-dimensional residual stress distributions, the compare the residual stresses in the longitudinal (welding) direction
application of such techniques is limited by the beam penetration with the transverse direction, the multiple cuts contour method was
depth, and access to the facilities remains regulated, expensive and performed on the multi-pass sample.
relatively time-consuming.
In 2001, Prime (2001) introduced a new approach to residual 2. Materials and specimens
stress evaluation, called the contour method. It has become one of
the most powerful techniques used to map the residual stresses in The parent material used in this study was a high strength low car-
a cross section by careful EDM cutting and relief measurement. Like bon steel (EH40-TMU) having nominal chemical composition (wt.%)
many other destructive methods, the technique is based on the char- Mn(1.2), Si(0.1), C(0.05), P(0.01) and balance Fe. The typical mechani-
acterisation of the deformation after material removal. Experimental cal properties of EH40-TMU are as follows: Youngs modulus 219 GPa,
validation of the contour method has been reported with the help Poissons ratio 0.28 and Yield stress 432 MPa. Two different weld-
of other non-destructive and destructive techniques such as neu- ing techniques were used for joining two plates of parent material,
tron diffraction (Kartal et al., 2006; Prime et al., 2006), synchrotron namely multi-pass (MP) and single pass (SP) methods. It is impor-
diffraction (Zabeen et al., 2013), X-ray diffraction (Toparli et al., 2013) tant to note that the groove geometry was different for the two weld-
and slitting (Traore et al., 2013). Recently, the method has been math- ing cases, particularly in terms of the different slopes, as shown in
ematically validated for two dimensional geometries by Kartal (2013). Fig. 1(a). The groove for the multi-pass welding makes an angle of 30
One limitation of the contour method is that it relies on the as- with 6 mm gap from the bottom between two parent blanks, whereas
sumption that the deplanation (displacement out of the cut plane) is for the single pass welding technique the angle was 20 with a 9 mm
assumed to be caused by and only related to the residual stresses on distance from the bottom. The structure of the fusion boundary for
the cut plane. In addition, the residual stress prole obtained from each welding case is illustrated in the macrograph in Fig. 1(b).
the contour method depends on the specimen dimensions. In other Multi-pass welding was carried out using the manual ux cored
words, if further progress is made by introducing additional cuts, the arc welding (FCAW) process, with a target heat input of 1.7 kJ/mm
stresses obtained from the analysis of the second and subsequent cuts generated using a single electrode normal to the welding direction.
are related to the state that arises after any previous cuts. To over- In order to ll the 80 mm thick sample, 75 passes were used. On
come above mentioned restrictions of the contour method and make the other hand, the electro-gas welding (EGW) was used for the case
measurements regardless of a change in the geometry of a body, the of single pass welding with a heat input of 64 kJ/mm generated by
inverse problem can be formulated in terms of nding the unknown means of the two electrodes travelling parallel to the welding di-
eigenstrain. Here the term eigenstrain refers to the incompatible rection. Two different welding methods are schematically shown in
strain eld that gives rise to the residual stress state in a body. As has Fig. 2. The as received specimens were supplied by POSCO with
been extensively discussed in a range of publications on this topic, 500 mm length, 300 mm width, and 80 mm thickness, as shown in
eigenstrain is an intrinsic parameter that can be thought of as the the schematic drawing presented in Fig. 3.
origin of residual stress, as opposed to these stresses themselves that
are extrinsic, i.e. change with the system conguration. The advan- 3. Residual stress measurement methods
tage of using the eigenstrain formulation is that once the underlying
distribution is found, the residual stress can be calculated for any con- In order to investigate the effect of two different welding tech-
guration sectioned from this body. niques on residual stress distributions, the inverse eigenstrain frame-
In this work, we independently studied residual stress elds aris- work and the contour method were employed. The reason for using
ing from single pass (64 kJ/mm) and multi-pass (1.7 kJ/mm) ferritic two different techniques in this study is that the contour method
steel welds, which are similar to the welding processes employed by provides the correct residual stresses on the cut surface only. In con-
Woo et al. (2013), but with different geometric dimensions (500 mm trast, the inverse eigenstrain method allows the determination of the
long, 80 mm thick and 300 mm wide). underlying sources of residual stress. Once these are found in the
422 M.E. Kartal et al. / International Journal of Solids and Structures 80 (2016) 420429

Mul-pass Welding Single-pass Welding at state, thus replicating the original residual stress eld across the
entire cross-section. The contour method allows the residual stress
Oscillaon
maps to be obtained directly from displacement maps with mini-

Welding direcon
mum time and computational cost. Furthermore, this method is not
sensitive to microstructure and does not require complex equipment.
However, as mentioned earlier, the contour method gives the correct
residual stress component only in the plane of cut. Residual stresses
away from the cut surface are not estimated by using this technique.
In order to compare the longitudinal residual stress prole on the
plane normal to the welding direction with the transverse residual
stress prole on the plane parallel to the welding direction, the mul-
tiple cut contour method on the multi pass welded specimen is used.
Fig. 2. Schematic diagram of parent material and the welding position for single and In this method a second cut is introduced to measure stresses on a
multi-pass welding. new cut plane with the contour method. Such stresses would have
been inuenced by the rst cut, but the contour measurement and
calculation for the rst cut also determines the change in stress on
the location of the second cut (and elsewhere). A more detailed de-
scription can be found in the work of Pagliaro et al. 2010.

y 3.2. The inverse eigenstrain problem

x
z
As stated by Mura (1987), eigenstrain is a generic name used to
describe any mist strain eld in a body arising from any inelastic
process, such as plastic deformation, phase transformations, thermal
EDM
phenomena, welding, etc. Elastic strains are generated in the body
transverse
80 mm

cut plane and, when combined with the incompatible eigenstrains, satisfy over-
all strain compatibility. The resulting residual stress eld, which is re-
EDM lated to the elastic strain eld through Hookes law, must also satisfy
300 mm longitudinal equilibrium. Provided that the eigenstrain distribution is known, then
cut plane calculating the corresponding residual elastic strains and stresses is
Fig. 3. Schematic drawing of the welded samples and the location of the transverse relatively straightforward, and requires the solution of an inhomoge-
and longitudinal EDM cut planes. neous elastic problem. We will refer to this as a classical direct prob-
lem. In practice, however, elastic strains (or displacements) may be
known or measured at a nite number of points. Hence, seeking to
sectioning plane, the continuously processed body assumption may reconstruct an unknown eigenstrain distribution from measured or
be applied, the underlying eigenstrain distribution is assumed to be known data requires the solution of an inverse eigenstrain problem
uniform in one of the principal directions. In the case of groove weld- (Korsunsky, 2006). This is generally a much more demanding task.
ing it may be reasonable to assume that the eigenstrain states are The inverse eigenstrain problem has been successfully employed to
identical in each plane normal to the welding direction, since the measure residual stresses at macro (Korsunsky et al., 2007) and mi-
thermo-mechanical processing history is nearly the same for all such cro (Kartal et al., 2012) scales.
planes. It is worth noting, however, that deviations from this assump- In the conventional contour method, residual stresses are directly
tion are likely to take place close to the beginning and end of the run. calculated from the displacement contour map, while for the inverse
Assuming the continuously processed body assumption holds, eigen- eigenstrain method the eigenstrain distribution must be determined
strains can be introduced everywhere within the body, and hence the rst, followed by the solution of the direct problem to reconstruct
residual stresses can be evaluated at all material points. The level of the residual stresses. A particular advantage of the eigenstrain for-
agreement between the two different analytical approaches to the mulation is that provided the body is carefully sectioned, the inher-
same experimental data can be investigated. In addition, the multi- ent eigenstrain distribution remains unchanged, whilst the residual
ple cut contour method described by Pagliaro et al. (2010) was used stress state is modied in response to the change of geometry. Dewald
to obtain the prole of the residual stress distribution along the weld- and Hill (2006) developed the inverse eigenstrain problem to mea-
ing direction of the multi pass specimen. In the following section, the sure the multi-axial residual stress components by using the contour
underlying principles of the techniques are presented. method data. However, their work involves neither an experimental
validation nor a complete comparison against contour method re-
3.1. The contour method sults. Later on, Kartal et al. (2008b) demonstrated the independent
validation of this technique with the help of the neutron diffraction
The contour method is a destructive method that involves section- measurements.
ing, i.e. destruction of the sample. Nevertheless, the method has many Fig. 3 shows the geometry of the welded samples studied here.
advantages such as simplicity of implementation and interpretation The origin of the coordinate system was placed at the centre of the
studied by Zhang (2004) and Kartal (2008a). Unlike other sectioning transverse EDM cut surface. The surface displacements were mea-
methods, such as hole drilling, it can map the residual stress distri- sured using a co-ordinate measurement machine (CMM). Residual
butions in large cross sectional areas using a single cut. The prin- stresses in the transverse cut plane were reconstructed by averag-
ciple of the method is straightforward: a residually stressed speci- ing the displacements normal to the transverse cut plane. Similar to
men is cut into two halves, and the height proles of the resulting the conventional contour method, the technique for residual stress
surfaces are measured. The pre-existing residual stress can then be evaluation uses the superposition principle, which assumes that the
obtained by carrying out a numerical simulation in which the sur- stresses after cutting are relaxed elastically and that sectioning does
face displacements are reversed, the argument being that the simu- not introduce any additional eigenstrains. Therefore, the measured
lation corresponds to the surface being brought back to its original displacements on the cut surface are only caused by the relaxation of
 M.E. Kartal et al. / International Journal of Solids and Structures 80 (2016) 420429 423

residual stress. These assumptions are also employed in many other Hosseinzadeh et al. (2013) were used. For each contour cut the
destructive residual stress evaluation methods. Note that the use of test specimen was symmetrically clamped to the bed of the EDM
the eigenstrain methodology allows this assumption to be veried. using xtures placed close to the cut line. To prevent additional
An additional assumption is generally made to regularise the inverse thermal stresses during cutting, the specimens were submerged
eigenstrain problem, namely, that the eigenstrain is uniform in one into temperature-controlled deionised water. Each specimen was cut
principal direction. In this case this was taken as the longitudinal along a transverse plane at the mid-length of the welded plate, as
(welding) direction. shown in Fig. 3, with the help of a 0.25 mm diameter brass wire.
The determination of the unknown eigenstrain distribution from Subsequent to EDM cutting, the surface contours of each of the
the measured and averaged displacements can be achieved by means two cut pairs of the specimens were measured. Each cut surface was
of an inverse technique, which has been successfully implemented measured on a 0.25 mm spaced grid using Mitutoyo CMM at the Uni-
for several residual stress measurement methods. Examples involve versity of Oxford. The outlines of the cut surfaces were also measured
the diffraction method (Jun et al., 2011), multi-axial contour method for the purpose of FE modelling setup. For the multi-pass sample, one
(Dewald and Hill, 2006; Kartal et al., 2008b) and incremental hole of the halves was then cut along a longitudinal plane located at the
drilling (Schajer, 2010). The formulation to calculate the unknown weld centre line as shown in Fig. 3. The second cut provided the trans-
eigenstrain distribution using the inverse method is briey discussed verse stresses in the component that remained after the rst cut. The
in the next section and a more detailed description can be found else- uncut transverse stresses in the original test component were then
where (Dewald and Hill, 2006; Kartal et al., 2012). reconstructed by elastic superposition of the stresses from both cuts
In the formulation of the inverse problem, it is rst assumed that (Pagliaro et al., 2010). All surface displacement pairs were then aver-
the two-dimensional spatially varying unknown eigenstrain distribu- aged to remove the effect shear stress relief and cut path undulations.
tion zz
(x, y) can be expressed as a series expansion: For the implementation of the conventional contour method, an
additional essential step is data smoothing of the averaged displace-
(m+1
)(n+1)
ments. This is vital as the averaged displacements consist of defor-
zz (x, y) = Ai GI (x, y), (1)
mations resulting from both residual stress release and noise arising
i=1
from cutting-induced surface roughness and measurement errors. As
where Ai represents unknown coecients of the series terms G which the noise is not the result of residual stress relaxation, it is impor-
can be created by multiplying two polynomial functions, such that: tant to remove it from the data before proceeding, otherwise it pro-
Gi (x, y) = Fk (x)Fl (y), (2) duces an amplied effect during the calculation of residual stresses.
The noise is typically removed from the data by tting the averaged
in which k is in the range 0 to m, l ranges from 0 to n, and i = displacements to a smooth function; a cubic spline t has been used
l + (n + 1)k + l. The problem of determining the eigenstrain distri- for this purpose.
bution reduces to a linear elastic system with (m + 1)(n + 1) unknown A three dimensional nite element model for stress calculation
coecients. Considering linear elastic behaviour and hence superpo- was built for one half of each welded specimen using ABAQUS 6.11.
sition, the measured displacements [um ] can then be represented as: The sample geometry was meshed with linear hexahedral elements
with reduced integration and it was assumed that each specimen was
[um ] = [A][C], (3) isotropic and linear elastic. Displacements obtained from the cubic
spline interpolation of the averaged data were applied to the model as
in which, um is a column vector including all measured and averaged
displacement boundary conditions, and linear elastic stress analysis
displacements on the contour cut surface and the model-derived ma-
for each model was performed to calculate the stress eld normal to
trix C contains the displacements computed by solving the eigen-
the plane of sectioning in the welded specimens.
strain problem for each component Gi in Eq. (1). The matrix has one
For the case of the inverse eigenstrain problem, the averaged dis-
column for each polynomial term, and each row corresponds to the
placements were not smoothed into any analytical functions, as the
model-derived displacement vectors at each individual data point.
basis functions used to calculate the eigenstrain (Legendre polyno-
In order to generate the inuence matrix C, each term in G was in-
mials) are already smooth. The reason for using Legendre polyno-
troduced as a pseudo-thermal eigenstrain prole into the FE model.
mials for the denition of the basis functions in this study is that
After satisfying equilibrium, the deformation due to the eigenstrain
they are widely used for the series expansion approach and their
basis function at the cut surface was recorded. Once the inuence
higher-order terms (i.e. two or more) automatically satisfy equilib-
matrix has been calculated, the unknown vector of coecients A is
rium (Prime 1999). Legendre polynomials are valid in the interval [
computed by inversion using a least squares t that minimises the er-
1:1] and hence the coordinates are normalised with the dimensions
ror between the model-derived and the measured data (Dewald and
of the welded samples before computing the values for each eigen-
Hill, 2006):
strain component. Each term in the bivariate Legendre basis function
 T
1 T series was input into the nite-element model as the known eigen-
[A] = [C] [C] [C] [um ], (4)
strain distribution. Then the resulting predicted displacement com-
the eigenstrain distribution can then be calculated by substituting ponent at each node was measured on the free cut surface in the
these coecients into Eq. (1). The elastic strain and the pre-existing nite-element model. After the inuence matrix was assembled, the
residual stress eld in the material are computed using the solution unknown coecients were found by least-squares solution of Eq. (4).
of the direct problem as described above. Note that all other eigen- Then, the unknown eigenstrain component was computed. Finally,
strain components except for zz
are assumed to be zero in this study.
the eigenstrain component was input into the nite-element model
and the resulting residual stresses were calculated on the cut plane.
4. Experimental procedures
5. The effect of the specimen length on residual stresses
Specimens were cut using electric discharge machining (EDM) at
the Open University. Cutting is the rst, and possibly the most crit- As was mentioned in the rst section, choosing the right size of a
ical step of this residual stress measurement method, since the - test specimen to correctly determine the residual stresses is the rst
nal result depends crucially upon the quality of the cut. In order to and most important step in residual stress measurement methods
produce cut surfaces of high quality that minimise associated errors as the remainder of experimental results depends upon the selected
in stress, the optimal cutting conditions identied in the work of specimen dimensions. More particularly, the length of the specimen
424 M.E. Kartal et al. / International Journal of Solids and Structures 80 (2016) 420429

normal to the plane of interest is a very important issue if the resid- (a)
ual stress component to be determined is normal to that plane as mm
the residual stresses at the mid-plane (and hence deformation) will
be inuenced by the proximity of the free ends. The minimum re-
quired critical specimen length in most cases is however unknown
as it entirely depends on the residual stresses yet to be determined
within the sample. Nevertheless, if the residual stress measurements (b)
are taken from a plane suciently remote from the free ends in a suf-
ciently long specimen, the results will be independent of the prox-
imity of the free ends. It follows that the magnitude of the residual
stresses would be the same for case if the specimen length was in-
Fig. 4. (a) Multi-pass (b) single pass averaged surface contour measurements.
nitely long.
In the literature, diffraction based residual stress measurement
methods were performed to investigate the effect of the specimen
length on residual stresses in thin welded specimens. In the work of
Altenkirch et al. (2009), thin welded plates that were progressively
cut-down to form test-pieces of shorter length or narrower width
were used for investigating the extent of residual stresses. An em-
pirical relaxation curve was found to characterise the relaxation be-
haviour for welds. Law et al. (2010) used a pipe containing a girth
weld whose length was rst reduced followed by cutting along the
longitudinal direction to observe residual stress redistribution. The
application of such techniques however requires successive diffrac-
tion measurements after each cutting process and hence it is time
consuming, expensive and requires an access to the facilities.
In the view of the contour method, whether the actual specimen
length used in experiments is sucient can be examined by impos-
ing the contour method data on the suciently long stress free model Fig. 5. Field plots of the longitudinal residual stress obtained from the conventional
and the agreement level with residual stress distribution in the ac- contour method for (a) multi-pass (b) single pass weld specimens.
tual specimen length can tell the answer. In the case of using an in-
suciently long specimen for the actual test, the above mentioned
approach using the contour method however may not give a correct pass specimen. The maximum to minimum range of the contours is
estimation of the magnitude of the residual stresses in real industrial 0.2 mm to 0.2 mm for each welded plate. It should be noted that
components. very few measured data points characterised as noise were removed
An alternative approach is to use the inverse eigenstrain prob- in the analysis procedure due to the high quality of the cuts, which
lem for determining the original residual stress distribution in large increases condence in these results.
scale components. As described earlier, the eigenstrain distribution The nite element results obtained by employing the conven-
in the specimen remains constant when the specimen geometry al- tional contour method for both the multi-pass and single pass welded
ters. This means that the determined eigenstrain in the test speci- specimens are shown in Fig. 5(a) and (b) respectively. The eld plots
men can be incorporated into any specimen geometry to determine demonstrate a clear difference of residual stress distributions caused
residual stresses in that geometry. If the material is continuously pro- by the two different welding techniques. It is found that the multi-
cessed (i.e. the eigenstrain distribution is uniform in one of the prin- pass specimen possesses a tensile residual stress with a peak value of
cipal directions) then the inverse eigenstrain technique can be ap- around 650 MPa at the centre of the weld region, which is balanced
plied to the data from a single cut to determine the residual stress by compressive residual stresses at both the bottom and top regions
component normal to the cut plane. Ueda and Fukuda (1989) im- remote from the weld zone, with a peak value of 475 MPa. This type
plemented continuously processed assumption in welds, where the of residual stress distribution can be observed in many other multi
eigenstrain along the longitudinal (welding) direction is assumed to pass welded components (e.g. Kartal et al., 2006; Elmesalamy et al.,
be constant. The underlying justication for using this assumption 2014). On the other hand, the maximum tensile residual stress with a
is that the weldment is considered to be long enough and welded value of 410 MPa is obtained in the heat affected zone for the single
continuously, so that each plane normal to the weld direction can be pass weld specimen, with compressive residual stresses remote from
thought of as experiencing the same thermal history. the weld region.
With the current projection of the eigenstrain method used in this The cross-sectional prole of the non-zero eigenstrain component
work, the determined eigenstrain distribution from the measured zz
calculated by means of Eq. (1) for each sample is shown in Fig.6(a)

displacements by means of the inverse technique can be assumed and (b). As a part of the inverse method, the order of the Legendre
for any specimen length with the same cross-sectional geometry. In polynomial series was chosen to be 16th-order for the x direction and
this way a critical specimen length can be estimated by installing the 8th-order for the y direction. As the displacement along the x direc-
determined eigenstrain distribution into a suciently long nite ele- tion varies signicantly, as shown in Fig. 4, the order of the polyno-
ment model. mial series had to be fairly high, whereas a lower order was su-
cient for the y direction. As Legendre polynomials are smooth func-
6. Results tions, the resulting linear combination of the solution of Eq. (4) is
smooth, so there is no need for further smoothing. As can be seen
Fig. 4(a) and (b) denotes the measured and averaged displacement from the gure, the calculated eigenstrain proles obtained from the
of two mating surfaces created by the transverse planar cut across measured displacements using the inverse technique are very similar
the multi-pass and single pass samples respectively. As can be seen to the residual stress proles in Fig. 5, but with opposite signs.
from the gures, the width of a compressive zone on the cut surface Comparisons of the longitudinal residual stresses ( zz ) obtained
of the single pass weld specimen is wider than that for the multi- by applying the conventional contour method and by solving the
 M.E. Kartal et al. / International Journal of Solids and Structures 80 (2016) 420429 425

Fig. 6. Eigenstrain proles obtained from the inverse eigenstrain technique for (a) multi-pass (b) single pass weld specimens.

Fig. 7. Comparison of the cross sectional residual stress distribution obtained by (a) the conventional contour method (b) the inverse eigenstrain problem for the multi-pass
specimen.

Fig. 8. Comparison of the cross sectional residual stress distribution obtained by (a) the conventional contour method (b) the inverse eigenstrain problem for the single pass
specimen.

inverse eigenstrain problem are given for the multi-pass and single ponent ( zz ) can be accurately calculated on the cut plane using the
pass specimens in Figs. 7 and 8 respectively. As mentioned above, two inverse problem of eigenstrain.
main assumptions were used: only a single component of eigenstrain In order to investigate the above mentioned effect on the magni-
(zz
) was assumed to be non-zero; and the non-zero eigenstrain com- tude of residual stresses, three different lengths with the same cross-
ponent was assumed to be uniform in the longitudinal (z) direction. sectional geometry were modelled for the multi-pass and single pass
The level of agreement between the two different techniques is very specimens. The conventional contour method was employed to com-
promising, notwithstanding the two assumptions introduced for the pare residual stresses using the same displacement smoothing for
use of the inverse technique in this study. each model. The half-length size was chosen as (i) suciently long
To provide a more quantitative comparison, the residual stress (500 mm), (ii) the original length of the specimen used in this work
component obtained from each method at the mid-plane along (250 mm) and (iii) the length used in the work of Woo et al. (2013),
y = 37.5 mm on the multi-pass specimen and along y = 0 mm on the (115 mm). Fig. 11 compares the effect of the specimen length on
single pass specimen were extracted and plotted in Figs. 9 and 10 re- the longitudinal residual stress distributions in the multi pass speci-
spectively. Here, the solid lines symbolise the values obtained by the men. As can be seen from the gure, residual stress proles obtained
conventional contour method and the dash lines show the residual from the models 500 mm and 250 mm long are in very good agree-
stresses obtained by the inverse eigenstrain problem. There is near- ment. However, the magnitude of the residual stresses in the model
perfect agreement between the variations of the residual stresses. with a length of 115 mm is signicantly lower than the other two
These results demonstrate that the longitudinal residual stress com- results.
426 M.E. Kartal et al. / International Journal of Solids and Structures 80 (2016) 420429

800 A similar trend is observed when the contour method data mea-
sured in the single pass specimen is imposed as a set of displace-
Contour method
600 ment boundary conditions on the stress free single pass models
Eigenstrain
with those three different lengths. Fig. 12 shows the inuence of the
400 single pass specimen length on the longitudinal residual stress dis-
Residual stress (MPa)

tributions. Once again, while there is a very close agreement found

200 between the 250 mm and 500 mm long models the magnitude of
residual stresses in the 115 mm long model is considerably less.
0 To quantitatively compare the results, the residual stress values
for each model on the cut plane along x = 0 mm were extracted and
-200
plotted for the multi-pass and single pass specimens as shown in
Figs. 13 and 14 respectively. As can be seen from the both gures,
-400
the difference of the stress proles between the two models that
-600 are 500 mm and 250 mm long is very small along the plotted line,
-150 -100 -50 0 50 100 150 whereas the stresses in the 115 mm long model are lower by as much
x posion (mm) as 140 MPa. It is evident from Figs. 13 and 14 that there is a critical
length below which the residual stress magnitude may be underesti-
Fig. 9. Comparison of the residual stress variations obtained by employing the con-
mated.
ventional contour method and the inverse eigenstrain problem on the mid-plane of
the multi-pass welded specimen along the line y = 37.5 mm.
One thousand millimetre long multi-pass and single pass speci-
mens were modelled using the eigenstrain distributions calculated
above. The resulting residual longitudinal stress along a path from
the free surface is plotted in Fig. 15 for both the multi-pass and single
pass specimens. As can be seen from the gure, the residual stress
600 rapidly increases over the rst 200 mm from the free surface, and
changes by a negligible amount after the rst 250 mm.
Contour method
400 Residual stress distributions parallel to the welding direction
Eigenstrain
xx in the multi-pass welded specimen for the second cut are given
in Fig. 16. Fig. 16(a) shows the remaining residual stresses, measured
Residual stress (MPa)

200
by the second cut, after residual stress relaxation of the rst cut.
0
Fig. 16(b) shows the original residual stresses prior to making the
rst cut, which were determined by means of the multiple cuts su-
perposition principle (Pagliaro et al., 2010). The maximum tensile
-200
and compressive stress values in Fig. 16(a) are found to be 470 MPa
and 534 MPa respectively and they are 440 MPa and 600 MPa be-
-400 fore the rst cut in Fig. 16(b). It should be noted that while Fig. 16(a)
shows the residual stress distribution for the full specimen length af-
-600 ter the rst cut when the length of the specimen reduced 250 mm,
-150 -100 -50 0 50 100 150 the residual stress distribution in Fig. 16(b) is mapped before the rst
x posion (mm) cut where the length of the specimen is 500 mm. Hence, Fig. 16(b)
Fig. 10. Comparison of the residual stress variations obtained by employing the con- shows the stress transient at the end of the specimen (z = 250) and
ventional contour method and the inverse eigenstrain problem on the mid-plane of that the stresses are nearly constant in the centre (z = 0) after the
the single pass welded specimen along the line y = 0 mm. correction for the rst cut.

Fig. 11. Comparison of the longitudinal residual stress distributions for three different model lengths for the multi-pass specimen.
 M.E. Kartal et al. / International Journal of Solids and Structures 80 (2016) 420429 427

Fig. 12. Comparison of the longitudinal residual stress distributions for three different model lengths for the single pass specimen.

700 800

115 mm 700
600 250 mm
600
500 mm
Residual stress (MPa)
Residual stress (MPa)

500
500

400 400

300
300 Mul pass
200
Single pass
200 100

100 0
-45 -35 -25 -15 -5 5 15 25 35 45 0 50 100 150 200 250 300 350 400 450 500
y posion (mm) z posion (mm)

Fig. 13. Comparison of the residual stress obtained from three different model lengths Fig. 15. Variations of the longitudinal residual stress component along the depth of
along x = 0 mm for the multi-pass specimen. the multi-pass and single pass specimens from the free surface.

As was shown in Fig. 5(a), it is found that a maximum tensile resid-

ual stress of around 650 MPa is obtained within the weld zone in the
multi-pass welded specimen, which is greater than the yield stress of
350 the welded material (570 MPa, Woo et al., 2013). However, the trans-
300 verse stress component was found to be 335 MPa at the position in
which the longitudinal residual stress component is a maximum. The
250
combination of the longitudinal stress zz with the transverse stress
200 xx is equivalent to a von-Mises stress of 563 MPa, which is of the
Residual stress (MPa)

150 order of the yield stress of the weld material. For materials possess-
ing high residual stress, the position of the peak residual stress may
100
be shifted if signicant plastic deformation occurs during the cutting
50 process. However, this effect was not evident in the present contour
115 mm
0 results.
250 mm
-50 500 mm
-100 7. Discussion

-150
Accurate quantication of residual stresses caused by complicated
-45 -35 -25 -15 -5 5 15 25 35 45
manufacturing processes such as welding is a very important engi-
y posion (mm)
neering task, as residual stresses directly affect fatigue strength and
Fig. 14. Comparison of the residual stress obtained from three different model lengths fracture toughness and may result in premature failure if they are not
along x = 0 mm for the single pass specimen. accounted for.
428 M.E. Kartal et al. / International Journal of Solids and Structures 80 (2016) 420429

Fig. 16. Transverse stress map xx (a) after rst cut and (b) before the rst cut in the multi-pass welded specimen using multiple cuts and the superposition principle.

Fig. 5 shows that the maximum critical residual stress is reached use of Legendre polynomials and it was 1.6 m for the cubic spline
in the weld zone of the multi-pass welded specimen, whereas it lies t.
within the heat affected zone for the single pass specimen. It can be In the work of Woo et al. (2013), original welded components
seen that there is a signicant effect of the amounts of heat inputs with a length of 690 mm were sectioned into three pieces of equal
on the residual stress distribution. The difference is also attributed length in order to use multiple residual stress measurement tech-
to the constraints under cooling. Since the amount of heat input for niques. However, it should be noted that a change in the geometry
the single pass specimen is very high, the heat affected zone is wider can signicantly change the residual stress eld and give a misleading
than that in the multi-pass specimen as shown in Fig. 5. indication of their magnitude in the actual component before section,
One of the main targets of this work was to measure accurately particularly if the length of the specimen is reduced signicantly. De-
and compare residual stresses in welded plates typical of those em- ciding the specimen length to be used in the experiments is a very
ployed in the manufacture of ship hulls using two different welding important task. In the present work, by installing the calculated
processes. For this purpose, the conventional contour method was eigenstrain distributions into suciently long models, the critical
employed. As mentioned earlier, measured residual stresses by this specimen length for both multi-pass and single pass specimens have
method are independent of the assumption of a continuously pro- been investigated. It has been found that a minimum half-length of
cessed body in one of the principal directions. However, since the a specimen with cross-sectional dimensions of 300 mm wide and
contour method does not provide stresses away from the cut plane, 80 mm thick should be at least 250 mm (see Figs. 13 and 14). For
the inverse eigenstrain problem was solved as a supplementary tech- instance, at a distance of 250 mm from the free surface, the residual
nique, which employed two assumptions: that the body was continu- stresses reach 98% and 96% of the spatially constant regions of resid-
ally processed in the welding direction; and only one non-zero eigen- ual stresses in the multi-pass and single pass specimens respectively.
strain component exists. Still, the longitudinal stress maps derived by On the other hand, the stresses are only 84% and 69% of these val-
the eigenstrain approach are in very good agreement with that ob- ues in the multi-pass and single pass specimens 115 mm from the
tained via the contour method on the cut plane. Provided the uniform free surface. This suggests that the magnitude of the measured resid-
eigenstrain assumption in the welding direction persists, then the ual stresses in the work of Woo et al. (2013) is well below the actual
calculated residual stresses are correct at each material point within residual stresses in their original welded components.
the specimen. One drawback with the eigenstrain method, however, Since the distribution of residual stresses is not known within a
is that creating the inuence matrix, the size of which depends on the component prior to experimental measurements, choosing a mini-
displacement variation on the cut surface, requires substantial com- mum specimen length is not an easy task. For the implementation
putational effort compared with the conventional contour method. of the contour method, an important parameter is the decaying rate
This level of agreement between the two techniques provided of residual stresses away from the contour cut plane (Kartal, 2013).
the motivation to explore further important questions related to the If the residual stress value rapidly decays in the direction normal to
eigenstrain distribution. The quantitative comparison between two the cut plane, then use of a nite length of specimen may give sim-
methods (e.g. Figs. 9 and 10) provides very close agreement with a ilar results in comparison to a suciently long specimen. A narrow
negligible difference This difference might be due to the fact that the region of tensile stress corresponds to a rapid decay away from the
variation of the eigenstrain in the longitudinal direction is assumed cut surface and hence more tolerance to sample shortening. In addi-
to be constant whereas the conventional contour method is indepen- tion, symmetric displacement elds in the x and/or y directions result
dent of this assumption. In addition, different smoothing functions in a slower decay than from corresponding anti-symmetric displace-
were used for each method. A cubic or quadratic spline smoothing ment elds (Kartal, 2013). Note that the term a nite length is used
is commonly performed for the contour method by tting the raw for the case when the length/width ratio of the specimen is generally
data with a number of piecewise polynomials, as it gives exibil- less than 2.
ity to choose the distance between two consecutive splines, com-
monly referred to as the knot spacing. On the other hand, Legendre 8. Conclusions
polynomials were used as continuous basis functions for the inverse
eigenstrain problem, which are widely used for series expansion ap- In this work we investigated the effect of two extreme welding
proaches, as the higher order terms (i.e., two and more) satisfy equi- heat inputs on residual stress distributions in thick components. For
librium. However, there may be some limitations in the use of poly- this purpose, techniques based on the conventional contour method
nomial series for least square tting (Kartal et al., 2012) and they are and the inverse eigenstrain problem were employed. Very good
less able to t experimental data sets which contain sharp gradients agreement was found between the two different techniques. With
or discontinuities. The average error between the averaged row data the help of the inverse eigenstrain problem, it has been demonstrated
and tted data on the cut surfaces was found to be 2.6 m for the that one could determine the original stresses in the full-length
 M.E. Kartal et al. / International Journal of Solids and Structures 80 (2016) 420429 429

specimen even if measurements are performed on a shorter speci- Kartal, M.E., 2013. Analytical solutions for determining residual stresses in two-
men. It has been shown that for the components used in this study, dimensional domains using the contour method. Proc. R. Soc. A 469, 20130367.
Korsunsky, A.M., 2006. Variational eigenstrain analysis of synchrotron diffraction mea-
very accurate results can be obtained by assuming that only one non- surements of residual elastic strain in a bent titanium alloy bar. J. Mech. Mater.
zero eigenstrain component exists. In addition, the multiple cut con- Struct. 1 (2), 259277.
tour method was employed to determine the original residual stress Korsunsky, A.M., Regino, G.M., Nowell, D., 2007. Variational eigenstrain analysis of
residual stresses in a welded plate. Int. J. Solids Struct. 44 (13), 45744591.
distribution on the welding plane in the multi pass specimen. It has Law, M., Luzin, V., Kirstein, O., 2010. Effects of cutting and specimen size on neutron
been demonstrated that selecting the length of the test specimen is measurement of residual stresses. J. Phys. Conf. Ser. 251 (1), 012044.
very important to accurately characterise residual stress proles in Liljedahl, C.D.M., Tan, M.L., Zanellato, O., Ganguly, S., Fitzpatrick, M.E., Edwards, L.,
2008. Evolution of residual stresses with fatigue loading and subsequent crack
real industrial components.
growth in welded aluminium alloy middle tension specimen. Eng. Fract. Mech. 75,
38813894.
Acknowledgements Liljedahl, C.D.M, Zanellato, O., Fitzpatrick, M.E., Lin, J., Edwards, L., 2010. The effect of
weld residual stresses and their re-distribution with crack growth during fatigue
under constant amplitude loading. Int. J. Fatigue 32, 735743.
The authors would like to acknowledge the nancial support of Mura, T., 1987. Micromechanics of Defects in Solids. Kluwer, Boston, MA.
the Engineering and Physical Sciences Research Council in the UK (EP- Prime, M.B., 1999. Residual stress measurement by successive extension of a slot: the
SRC) under grant reference EP/G004676/1, Micromechanical Mod- crack compliance method. Appl. Mech. Rev. 52, 7596.
Prime, M.B., 2001. Cross-sectional mapping of residual stresses by measuring the sur-
elling and Experimentation. They would also like to thank POSCO for face contour after a cut. J. Eng. Mater. Technol. 123, 162168.
supporting the experimental programme by supplying specimen ma- Prime, M.B., Gnupel-Herold, T., Baumann, J.A., Lederich, R.J., Bowden, D.M.,
terial. We should also like to acknowledge Mr. P. Ledgard for section- Sebring, R.J., 2006. Residual stress measurements in a thick, dissimilar aluminium
alloy friction stir weld. Acta Mater. 54, 40134021.
ing of the samples and Dr. F. Hosseinzadeh for organising the timeline Pagliaro, P., Prime, M.B., Swenson, H., Zuccarello, B., 2010. Measuring multiple residual-
for the samples. stress components using the contour method and multiple cuts. Exp. Mech. 50 (2),
187194.
References Schajer, G.S., 2010. Advances in hole-drilling residual stress measurements. Exp. Mech.
50, 159168.
Toparli, M.B., Fitzpatrick, M.E., Gungor, S., 2013. Improvement of the contour method
Altenkirch, J., Steuwer, A., Peel, M.J., Withers, P.J., 2009. The extent of relaxation of weld
for measurement of near-surface residual stresses from laser peening. Exp. Mech.
residual stresses on cutting out cross-weld test-pieces. Powder Diffr. 24 S313-SS36.
53 (9), 17051718.
DeWald, A.T, Hill, M.R., 2006. Multi-axial contour method for mapping residual stresses
Traore, Y., Paddea, S., Bouchard, P.J., Gharghouri, M.A., 2013. Measurement of the resid-
in continuously processed bodies. Exp. Mech. 46, 473490.
ual stress tensor in a compact tension weld specimen. Exp. Mech. 53 (4), 605618.
Elmesalamy, A., Francis, J.A., Li, L., 2014. A comparison of residual stresses in multi pass
Ueda, Y.M., Fukuda, K., 1989. New measuring method of three dimensional residual
narrow gap laser welds and gas-tungsten arc welds in AISI 316L stainless steel. Int.
stresses in long welded joints using inherent strains and parameters. J. Eng. Mater.
J. Press. Vessels Pip. 113, 4959.
Technol. 111, 18.
Gurney, T.R., 1979. Fatigue of Welded Structures, second ed Cambridge University Press,
Weman, K., 2003. Welding Processes Handbook. CRC Press, New York.
Cambridge.
Withers, P.J., Bhadeshia, H.K.D.H., 2001a. Residual stress: part I measurement tech-
Hosseinzadeh, F., Ledgard, P., Bouchard, P.J., 2013. Controlling the cut in contour resid-
niques. Mater. Sci. Technol. A 17, 355365.
ual stress measurements of electron beam welded Ti-6Al-4V alloy plates. Exp.
Withers, P.J., Bhadeshia, H.K.D.H., 2001b. Residual stress part II nature and origins.
Mech. 53 (5), 829839.
Mater. Sci. Technol. A 17, 366375.
Jun, T.-S., Venter, A.M., Korsunsky, A.M., 2011. Inverse eigenstrain analysis of the ef-
Withers, P.J., 2007. Residual stress and its role in failure. Rep. Prog. Phys. 70, 22112264.
fect of non-uniform sample shape on the residual stress due to shot peening. Exp.
Withers, P.J., Turski, M., Edwards, L., Bouchard, P.J., Buttle, D.J., 2008. Recent advances
Mech. 51 (2), 165174.
in residual stress measurement. Int. J. Press. Vessels Pip. 85 (3), 118127.
Kartal, M., Turski, M., Johnson, G., Gungor, S., Fitzpatrick, M.E., Withers, P.J., Edwards, L.,
Woo, W., An, G.B., Kingston, E.J., DeWald, A.T., Smith, D.J., Hill., M.R., 2013. Through-
2006. Residual stress measurements in single and multi-pass groove weld speci-
thickness distributions of residual stresses in two extreme heat-input thick welds:
mens using neutron diffraction and the contour method. Mater. Sci. Forum 524
a neutron diffraction, contour method and deep hole drilling study. Acta Mater. 61
525, 671676.
(10), 35643574.
Kartal, M.E., 2008a. Advanced Experimental Methods for the Characterization of
Zabeen, S., Preuss, M., Withers, P.J., 2013. Residual stresses caused by head-on and 45
Welded Structures. The Open University, UK Ph.D. thesis.
foreign object damage for a laser shock peened Ti-6Al-4V alloy aerofoil. Mater. Sci.
Kartal, M.E., Liljedahl, C.D.M., Gungor, S., Edwards, L., Fitzpatrick, M.E., 2008b. Deter-
Eng. A 560, 518527.
mination of the prole of the complete residual stress tensor in a VPPA weld using
Zhang, Y., Ganguly, S., Edwards, L., Fitzpatrick, M.E., 2004. Cross-sectional mapping of
the multi-axial contour method. Acta Mater. 56 (16), 44174428.
residual stresses in a VPPA weld using the contour method. Acta Mater. 52, 5225
Kartal, M.E., Dunne, F.P.E., Wilkinson, A.J., 2012. Determination of the complete mi-
5232.
croscale residual stress tensor at a subsurface carbide particle in a single-crystal
superalloy from free-surface EBSD. Acta Mater. 60 (1314), 53005310.