Evolution of Residual Stresses with Fatigue Crack Growth

in a Variable Polarity Plasma Arc–Welded Aluminum Alloy

Compact Tension Specimen
C.D.M. LILJEDAHL, O. ZANELLATO, L. EDWARDS, and M.E. FITZPATRICK

The evolution of the residual stresses during fatigue crack growth in a welded compact tension
C(T) specimen was measured using neutron diffraction. The measurements were performed by
growing a fatigue crack in a sample in situ on a neutron diffractometer. The stresses were found
to be unaffected by crack growth through the compressive part of the initial residual stress field.
The residual stresses at the crack tip increased when the crack entered the tensile residual stress
field to maintain residual stress equilibrium. Finite element (FE) modeling of the evolution of
the residual stresses showed good correlation with the experimental results. The residual stress
evolution was found to be governed by redistribution of the initial stress field and only slightly
affected by fatigue-induced effects at the measured spatial resolution (2 mm 9 2 mm 9 7 mm).

DOI: 10.1007/s11661-008-9602-6
Ó The Minerals, Metals & Materials Society and ASM International 2008

I. INTRODUCTION asymmetric compared with the stresses in the original

welded plate. Hence, extreme care has to be taken when
THE use of welding for joining structures is increas- using results from laboratory-scale specimens to assess
ingly being applied in the aerospace industry. This larger structures.
technique permits for larger integral structures to be The evolution of residual stresses around the crack tip
manufactured, and hence the cost and the weight of the with fatigue loading can be estimated experimentally
final product can be reduced. The inherent inconve- with crack closure methods. Brouard et al.[7] measured
nience with welds is the residual stresses caused by the the evolution of the crack closure with crack growth
intense local heating, coupled with the fact that there are using an eddy current transducer in a variable polarity
fewer ‘‘natural’’ crack stoppers in welded structures. plasma arc–welded specimen of 2024-T351 aluminum
Residual stresses can significantly influence the fatigue alloy. The closure was first seen to increase and then
life of engineering components.[1] As a result, accurate decrease as the crack extended. Farrahi et al.[8] mea-
knowledge of the residual stresses is required in order to sured the crack closure with fatigue crack growth using
make effective, sensible calculations to input into back face strain and crack mouth opening displacement
damage tolerance safety assessments; and furthermore, gages for untreated and shot-peened specimens of
the evolution of the residual stresses with fatigue crack 20NiCrMo2. The shot peening increased the crack
growth has to be determined. closure and retarded the fatigue crack growth. However,
Neutron diffraction has been used for measurement of these methods do not give a detailed evolution of the
residual stresses in various types of materials and residual stresses within the structure.
industrial components.[2] This method has been exten- Previously, the evolution of the residual stresses in a
sively applied to measure residual stresses in welds.[3–5] middle tension M(T) specimen was measured.[9] In that
Neutron diffraction is an ideal tool for measurement in specimen, the magnitude of the peak tensile residual
welds as the strains tend to be high, the gage volume stresses was seen to increase with crack growth. In the
typically used is small compared with the weld, and the present work, the evolution of the residual stresses with
penetration depth is large. Pratihar et al.[6] measured the fatigue crack growth in a welded aluminum alloy C(T)
residual stresses in a compact tension C(T) specimen specimen was measured with neutron diffraction and
sectioned from a welded plate. The authors found predicted with an FE simulation.
that the residual stresses were significantly lower and

II. EXPERIMENTAL PROCEDURES

C.D.M. LILJEDAHL, Research Fellow, O. ZANELLATO, Research
Student, and M.E. FITZPATRICK, Professor, are with Materials A. Material and Specimen Geometry
Engineering, The Open University, Milton Keynes, MK7 6AA, United
Kingdom. Contact e-mail: [email protected] L. EDWARDS, Single pass autogenous variable polarity plasma arc
formerly Professor, with Materials Engineering, The Open University, welding was used to manufacture 2024 aluminum plates
is Head of Institute of Materials and Engineering Science with the 500 mm 9 500 mm. The plates were welded with the
Australian Nuclear Science and Technology Organisation, PMB1,
Menai, NSW 2234, Australia.
weld direction parallel to the rolling direction of the
Manuscript submitted December 13, 2007. plate. Further details of the welding process can be
Article published online July 30, 2008 found elsewhere.[10]

2370—VOLUME 39A, OCTOBER 2008 METALLURGICAL AND MATERIALS TRANSACTIONS A

 The plate was skimmed down to a thickness of 7 mm A stress-free reference sample was therefore machined
and an M(T) specimen was machined with the dimen- using EDM to prepare a comb specimen from the
sions and orientations, as shown in Figure 1 (ASTM E original welded plate. The dimensions (3 mm 9 9 mm 9
647-00). The evolution of the residual stresses in the 3 mm) of the comb teeth were small enough to assure
M(T) sample is described elsewhere.[9] A C(T) sample that they were virtually stress free.[15] The interplanar
(Figure 1) was then sectioned from the M(T) with distance could hence be determined as a function of the
electrodischarge machining (EDM). No further skim- distance from the weld center. The measurement gage
ming of the C(T) sample was carried out. volume used for the stress-free sample was 2 mm 9
8 mm 9 2 mm, with the longer dimension along the
‘‘tooth’’ of the comb, corresponding to the through-
B. In-Situ Fatigue Loading and Crack Growth
thickness direction of the original plate.
The fatigue loading, in mode I, was undertaken by use In this research, the specimen was subjected to fatigue
of a 100-kN Instron servohydraulic (Instron, Bucks, loading and the crack was allowed to grow a certain
UK) stress rig (Figure 2). The crackpgrowth ffiffiffiffi was distance. Measurements were then undertaken at loca-
undertaken at a constant DK of 17 MPa m and an R tions along the crack growth direction from the notch
ratio of 0.1. Crack growth was monitored with a tip (Figure 1). This was repeated with increments in the
fractomat gage (FAC-20, Tokyo Sokki Kenkyojo Co., crack length of about 2 to 3 mm. Plane stress conditions
Ltd., Tokyo, Japan). were assumed, and thus only the strain components

C. Neutron Diffraction
Neutron diffraction is an established nondestructive
technique to determine strains within metallic struc-
tures.[11] The interplanar lattice spacing for various
crystal reflections can be determined from the position
of the diffraction peaks, as realized by Bragg.[12] The
interplanar lattice spacing can be used, in conjunction
with a ‘‘stress-free’’ value of the lattice parameter, to
determine the internal strains in the material, which can
then be used to calculate the internal stresses.
The measurements were carried out on the ENGIN-X
diffractometer,[13] which is based at the ISIS pulsed
neutron source, of the Rutherford Appleton Laboratory
in the United Kingdom. There are two detector banks at
±90 deg to the incident beam, which allows for mea-
surements in two directions simultaneously (Figure 2).

1. Stress-free reference specimen

The stress-free lattice parameter can vary significantly Fig. 2—Experimental setup of the servohydraulic stress rig on the
in the heat-affected zone and in the weld itself.[14] ENGIN-X diffractometer at the ISIS neutron source.

Fig. 1—Specimen geometry (thickness 7 mm).

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 39A, OCTOBER 2008—2371

in the longitudinal and transverse directions were The finite element (FE) method, using the commercial
measured. The gage volume used was 2 mm 9 8 mm 9 FE code ABAQUS/Standard Version 6.5[22] was used
2 mm. The longer dimension was oriented in the for all the modeling. The same elastic material proper-
thickness direction of the plate to give through-thickness ties as given in Section II were used throughout. Plane
averaging of the measured strains (as a two-dimensional stress four-noded full integration elements were used.
plane stress assumption was used for the computation of Only half of the real model needed to be meshed due to
the stress). symmetry. An analytical surface and contact elements
were assigned along the symmetry line to avoid over-
2. Strains and stresses closure (penetration of the crack surfaces over the
Rather than fitting each diffraction peak separately, symmetry line). The load was applied to a rigid pin fitted
Pawley[16] refinement was used to determine the lattice into the hole of the specimen. The pin and the hole were
parameters, because this has been shown to yield a good treated as contact surfaces (Figure 3).
approximation of the engineering strain in the compo- In order to be able to compare the experimental
nent.[17] EX-SBA, the ENGIN-X script-based analysis results with the FE model, the FE results were averaged
software, was used for the data analysis.[18] over the equivalent neutron diffraction gage volume
The direct elastic strain (e) in the material in the (Figure 4). In this case, this gave about 225 nodes in the
measured direction can be calculated from the following gage volume for a mesh of 0.1 mm 9 0.1 mm. This was
expression: implemented by writing a postprocessing FORTRAN
routine.[23]
aðx; cÞ
a0 ðxÞ The initial unfatigued stress distribution first had to
eðx; cÞ ¼
a0 ðxÞ be modeled. This is presented in Section A. In Sections B
and C, the approach used to model the effect of crack
where a and a0 are the lattice parameter at the point of
interest and in the stress-free reference specimen, respec-
tively; x is the distance from the weld center; and c is the
crack length.
Assuming an isotropic material and plane stress, the
direct stresses in the longitudinal (rlong) and the trans-
verse (rtrans) directions can be calculated using Hooke’s
law:
E
rlong ðx; cÞ ¼ ðelong ðx; cÞ þ metrans ðx; cÞÞ
1
m2

E
rtrans ðx; cÞ ¼ ðetrans ðx; cÞ þ melong ðx; cÞÞ
1
m2
where E and m are the Young’s modulus and Poisson’s
ratio, respectively. ENGIN-X is a time-of-flight diffrac- Fig. 3—FE mesh for the C(T) specimen.
tometer, where multiple lattice reflections are obtained.
Therefore, many crystal orientations were involved in
the determination of the lattice parameters, and macro-
scopic values of the material properties could be used,
i.e., E = 72 GPa and m = 0.345.
Most of the measurements were carried out at zero
load, i.e., only residual strains were measured, without
the superposition of applied strains. The strains were
also measured at the maximum load in the fatigue cycle
(Pmax) with the crack tip at 19.1 mm from the weld
center.

III. FINITE ELEMENT ANALYSIS

PROCEDURES
The most common approach for numerical modeling
of welding and the resulting residual stress is through
two analysis steps.[19–21] In the first analysis, the tran-
sient thermal history is modeled. In the second analysis,
the thermal history is used in a mechanical analysis to
generate the resulting residual stresses. In this work, the
redistribution of residual stresses, rather than their Fig. 4—Schematic figure of the averaging over the equivalent neu-
generation by the welding process, was considered. tron gage volume for the results of the FE analysis.

2372—VOLUME 39A, OCTOBER 2008 METALLURGICAL AND MATERIALS TRANSACTIONS A

growth on the residual stress redistribution along the C
crack path is presented assuming elastic (Section B) and da ¼ ðr
aÞdepl
cadepl
r0
elastic-plastic (Section C) material behavior.
where C and c are material parameters. The hardening
parameters were automatically determined from the
A. Initial Stress Distribution in the Unfatigued Specimen stress-strain data by a least-squares fitting routine when
Due to Welding running the ABAQUS code. The element size was
0.1 mm 9 0.1 mm. This resulted in a plastic zone
The objective of this research was not to model the
encompassing about seven elements at the final crack
welding process but rather to investigate the effect of
tip.
crack growth on the evolution of the residual stresses in
In the models of the fatigue crack growth, the applied
the welded specimen.
load was cycled and the nodes were released progres-
The stress distribution away from the notch in the
sively. This was modeled with an analysis consisting of
M(T) sample has been measured, and the details are
five steps. In the first step, the initial residual stress field
presented elsewhere.[10] This stress distribution was
was modeled as in the elastic model. In a second step,
assumed to represent the distribution in the welded
the pin was moved with a small displacement to assure
M(T) sample after welding (i.e., assuming that the
contact. In the third step, the load at the initial crack
welded plate was continuously processed) and was read
length was applied. In the fourth step, the fatigue
into the model. This was implemented by writing a
crack growth was simulated. Both the load and the
SIGINI FORTRAN subroutine.[22] In the following
crack length were set as a function of time (Figure 6).
step, the stresses were allowed to equilibrate, simulating
Three cycles were applied between each node release,
the cutting of the M(T) sample to the specified C(T)
with nodes being released at Pmin (Figure 6). Hence, the
geometry (Figure 1).

B. Elastic Redistribution of the Residual Stress Due

to Crack Extension
The evolution of the residual stresses with crack
growth, calculated assuming elastic material properties
only, was predicted by removing the symmetry bound-
ary condition along the crack front. The stresses were
then allowed to re-equilibrate. This was subsequently
repeated until the experimental final crack length was
reached.
The stresses were averaged over the gage volume at
the measured crack lengths (Figure 4). A mesh conver-
gence study was undertaken to determine the necessary
mesh density. The stresses averaged over the gage
volume converged readily at a mesh size of 0.1 mm.
This mesh size was used thereafter.
Fig. 5—Stress-strain curve for the 2024-T3 alloy.[23]
C. Simulation of Fatigue Crack Growth in the CT
Specimen
Nonlinear material properties were included in the
model to investigate the effect of the induced plastic
strain during the fatigue loading on evolution of the
residual stresses. The stress-strain data for the alloy was
taken from Boller and Seeger,[23] as shown in Figure 5.
Nonlinear kinematic hardening was used, because it is
more suited for cyclic loading than isotropic hardening,[22]
predicting both the Bauschinger effect and plastic
shakedown. The constitutive model used in this work is
implemented in ABAQUS, as explained subsequently.
The yield surface is defined by
fðr
aÞ
r0 ¼ 0
where f(r – a) is the equivalent von Mises stress with
respect to the backstress (a) and r0 is the yield stress at
zero plastic strain. The hardening is introduced through
the evolution of the backstress, which is defined as
pffiffiffiffi
follows: Fig. 6—External load vs crack length (DK = 17 MPa m, R = 0.1).

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 39A, OCTOBER 2008—2373

crack was set to grow progressively, releasing a node at a A. Initial Residual Stress Distribution in the C(T)
time. In the fifth step, the specimen was unloaded. Specimen
The redistribution of the stresses that occur when the
IV. RESULTS AND DISCUSSION C(T) sample is sectioned from the M(T) sample is
shown in Figure 7(a). The residual stresses redistribute
The initial weld residual stresses arise due to nonuni- asymmetrically when the C(T) is removed from the
form expansion and contraction during welding caused larger specimen, owing to the presence of the notch in
by the local heating and associated plasticity. The hot the specimen. In this less-constrained specimen, the
weld material yields easily, and on cooling misfit strains residual stresses are significantly lower then in the
are induced between the yielded and the unyielded M(T) specimen and stresses are compressive at the weld
material. This results in large stresses parallel to the weld line in contrast to the M(T). A map of the predicted
line in the longitudinal direction, as seen in the M(T) stresses in the longitudinal direction is shown in
specimen (Figure 7(a)). Figure 7(b).

Fig. 7—(a) Comparison between the initial uncracked longitudinal stress distributions in the M(T) sample and resulting stress redistribution
when the C(T) sample is sectioned from it (longitudinal direction). (b) Map of the predicted stress distribution in the unfatigued CTS (longitudi-
nal direction). (c) Comparison between the measured and predicted stresses in the longitudinal and the transverse direction in the C(T) sample.

2374—VOLUME 39A, OCTOBER 2008 METALLURGICAL AND MATERIALS TRANSACTIONS A

 The predicted stress distribution from the uncracked
distribution in the C(T) sample was in excellent agree-
ment with the experimental results (Figure 7(c)), both in
the longitudinal and the transverse directions. The
transverse stresses are much smaller than the longitudi-
nal stresses. In Section IV–B, only the longitudinal
stresses are discussed, because these are larger and are
primarily responsible for any influence on the mode I
fatigue crack growth.

B. Evolution of the Residual Stresses with Fatigue Crack

Growth
The measured residual stresses are shown in Figure 8.
The residual stresses showed little effect of crack
extension until the crack tip had reached the position
of the first tensile residual strain peak at about 6 mm
from the weld center. This is because as the crack grows
through a compressive portion of a residual stress field,
the compressive tractions can be transferred across the
crack faces.
When the crack enters the tensile portion of the
residual stress field, there is a significant change in the
residual stress profile. The position of the peak tensile
stress ahead of the crack begins to move toward the
back face of the sample. The overall magnitude and
profile of the distribution also changes significantly.
The predictions were compared with the experimental
results, as shown in Figure 9. The predicted trend of the
evolution of the residual stresses was the same as was
found experimentally. The peak tensile stresses were
possibly slightly overestimated. One possible explana-
tion for the apparent overestimation might be the
diversion of the neutron beam, as a result of which the
gage volume will be larger than defined by the slits and
the averaging of the numerical results would have been
carried out over too small an area.

1. The effect of nonlinear material properties

on the simulated residual stress evolution
Residual stresses are also induced when fatiguing an
initially stress-free material. Compressive residual stres-
ses are induced behind the crack tip as a result of the
induced plastic strains, which create misfit strains upon
unloading. Residual stress can also be found in the wake
of the crack in the case where the crack faces come into
contact upon unloading as a consequence of the
deformation left in the wake of the advancing crack
(plasticity-induced crack closure). The effect of these
residual stresses in superposition with the redistributed
weld-induced residual stresses was hence investigated by
simulating the fatigue crack growth and including
nonlinear material properties.
This was carried out for a final crack length of
41.6 mm (19.1 mm from the weld center). The crack Fig. 8—Comparison between the measured and modeled evolution
length of 41.6 mm was chosen because the largest peak of the longitudinal stresses with fatigue crack growth with the crack
stresses were found at this position. The results can be tip at the following positions from the weld center: (a) -9.9, -7.9,
seen in Figure 9. Plastic deformation was seen to affect -5.9, -3.9, and 0.1; and (b) 3.1, 6.1, 8.1, 10.1, and 13.1; and
the residual stresses only slightly. (c) 14.1, 16.1, 19.1, and 22.1.

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 39A, OCTOBER 2008—2375

 residual stress state in depth in the sample being
studied.
2. FE predictions of the residual stresses and their evo-
lution during crack growth were seen to be in good
correlation with the experimental results. The resid-
ual stress evolution during fatigue crack growth was
governed by redistribution of the initial stress field
due to welding. The local plastic deformation associ-
ated with the fatigue crack was seen to have a very
small effect on the redistribution of the stress field.
3. The evolution of the residual stresses in a middle
tension specimen has been reported previously.[9]
The resulting residual stresses when a welded plate
is sectioned to an M(T) or C(T) geometry are seen
to be very different, and the interaction of the weld
stresses with a fatigue crack are therefore markedly
different. The crack-tip stresses in the M(T) geome-
try are predominantly tensile and detrimental to the
Fig. 9—Model and experimental results of the effect of plasticity on structural integrity, whereas the compressive resid-
the residual stresses in the longitudinal direction, for a crack-tip ual stress in the C(T) geometry would slow the fati-
position of 19.1 mm from the weld center.
gue crack growth. Determining the effect of weld
stresses using laboratory samples is therefore highly
dependent on the relaxation of the stresses when
the sample is machined, and care therefore should
be taken in extrapolating results from laboratory
samples to in-service behavior.

ACKNOWLEDGMENTS
The authors gratefully acknowledge the instrument sci-
entist, Dr. Ed Olivier, at the ENGIN-X diffractometer
at ISIS, and Mr. Peter Ledgard for manufacturing of
the fatigue grips.

REFERENCES
1. J.F. Throop and H.S. Reemsnyder: Residual Stress Effects in
Fatigue, ASTM STP 776, ASTM, Philadelphia, PA, 1981, pp. 1–2.
Fig. 10—Comparison between the longitudinal stresses when the 2. M.E. Fitzpatrick and A. Lodini: Analysis of Residual Stress Using
specimen is unloaded and at Pmax (with the crack tip 19.1 mm from Neutron and Synchrotron Radiation, Taylor & Francis, London,
the weld center). 2003, pp. 296–318.
3. L. Edwards, P.J. Bouchard, M. Dutta, D.Q. Wang, J.R.
Santisteban, S. Hiller, and M.E. Fitspatrick: Int. J. Pres. Ves. Pip.,
2005, vol. 83, pp. 288–98.
2. Stresses at Pmax with the crack tip at 19.1 mm from 4. R.A. Oven, R.V. Preston, P.J. Withers, H.R. Shercliff, and P.J.
the weld center Webster: Mater. Sci. Eng. A, 2003, vol. 346, pp. 159–67.
In Figure 10, the predicted stresses, from a combina- 5. Y. Zhang, S. Ganguly, L. Edwards, and M.E. Fitzpatrick: Acta
tion of the residual stresses and the applied Pmax at a Mater., 2004, vol. 52, pp. 5225–32.
crack length 19.1 mm, are compared with the experi- 6. S. Prathar, V. Stelmukh, M.T. Hutchings, M.E. Fitzpatrick
U. Stuhr, and L. Edwards: Mater. Sci. Eng. A, 2006, vol. 437,
mental results. There was a reasonable correlation pp. 46–53.
between the experimental results and the FE analysis 7. J. Brouard, J. Lin, and P.E. Irving: Proc. Fatigue 2006, Atlanta,
prediction. The applied stresses were found to be smaller GA, June 2006.
than the tensile residual stresses at this crack length. 8. G.H. Farrahi, G.H. Majzoobi, F. Hosseinzadeh, and S.M. Harati:
Eng. Fract. Mech., 2006, vol. 73, pp. 1772–82.
9. C.D.M. Liljedahl, M.L Tan, O. Zanellato, S. Ganguly, M.E.
Fitzpatrick, and L. Edwards: Eng. Fract. Mech., 2008, vol. 75,
V. CONCLUSIONS pp. 3881–94.
10. S. Ganguly, M.E. Fitzpatrick, and L. Edwards: Metall. Mater.
1. The evolution of the residual stresses in a welded Trans. A, 2006, vol. 37A, pp. 411–20.
11. A.J. Allen, M.T. Hutchings, C.G. Windsor, and C. Andreani: Adv.
C(T) specimen was measured successfully with neu- Phys., 1985, vol. 34 (4), pp. 445–73.
tron diffraction for what we believe is the first time. 12. W.L. Bragg: Proc. Cambridge Philosophical Society, Royal Soci-
Neutron diffraction allows for determination of the ety, London, 1914, vol. 17, pp. 43–57.

2376—VOLUME 39A, OCTOBER 2008 METALLURGICAL AND MATERIALS TRANSACTIONS A

13. J.A. Dann, M.R. Daymond, L. Edwards, J.A. James, and J.R. 19. R.V. Preston, H.R. Shercliff, P.J. Withers, and S.D. Smith: Sci.
Santisteban: Physica B, 2004, vol. 350, pp. 511–14. Technol. Weld. Join., 2003, vol. 8, pp. 10–18.
14. T.M. Holden, H. Suzuki, D.G. Carr, M.I. Ripley, and B. Clousen: 20. R.V. Preston, H.R. Shercliff, P.J. Withers, and S.D. Smith: Acta
Mater. Sci. Eng. A, 2006, vol. 437, pp. 33–37. Mater., 2004, vol. 52, pp. 4973–83.
15. V. Stelmukh, L. Edwards, and S. Ganguly: Text. Microstruct., 21. L.E. Lindgren: Comput. Meth. Appl. Mech. Eng., 2006, vol. 195,
2003, vol. 35, pp. 175–83. pp. 6710–35.
16. G.S. Pawley: J. Appl. Crystallogr., 1981, vol. 14, pp. 357–61. 22. ABAQUS Version 6.5, Hibbitt, Karlsson & Sorensen, Inc., Rising
17. M.R. Daymond, M.A. Bourke, R.B. Von Dreele, B. Clausen, and Sun Mills, RI, 2005.
T. Lorentzen: J. Appl. Phys., 1997, vol. 82, pp. 1554–62. 23. C.H.R. Boller and T. Seeger: Materials Data for Cyclic Loading,
18. E. Oliver, J. Santisteban, J. James, M. Daymond, and J. Dann: Part D: Aluminium and Titanium Alloys, Materials Science
ENGIN-X User Manual, ISIS, Rutherford Laboratory, Didcot, Monographs, 42D, Elsevier Science Publishers B.V, Amsterdam,
UK, 2004. Netherlands, 1987, pp. 108–24.

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 39A, OCTOBER 2008—2377