Vous êtes sur la page 1sur 17

Computational Materials Science 54 (2012) 312–328

Contents lists available at SciVerse ScienceDirect

Computational Materials Science


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

Accurate prediction of residual stress in stainless steel welds


M.C. Smith a,⇑, P.J. Bouchard b, M. Turski b, L. Edwards c, R.J. Dennis d
a
EDF Energy Nuclear Generation, Gloucester GL4 3RS, UK
b
The Open University, Milton Keynes MK7 6AA, UK
c
Australian Nuclear Science and Technology Organisation (ANSTO), PMB 1, Menai 2234 NSW, Australia
d
Fraser Nash Consultancy Ltd., 1 Trinity Street, College Green, Bristol BS1 5TE, UK

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

Article history: Measurements of stress–strain properties and residual stress in purpose-designed one and three-pass
Received 27 June 2011 groove-weld specimens are used to optimise a mixed hardening constitutive model for simulating resid-
Received in revised form 17 August 2011 ual stresses in austenitic Type 316 stainless steel weldments. It is demonstrated that isotropic hardening
Accepted 20 October 2011
over-predicts the tensile magnitude of welding residual stress in the benchmark specimens, while pure
Available online 13 December 2011
kinematic hardening gives an under-prediction of longitudinal stresses in parent material close to the
weld. The most accurate predictions are those based on optimised mixed isotropic–kinematic formula-
Keywords:
tions combined with a multi-pass moving heat source welding analysis.
Weld residual stress
Mixed hardening
Ó 2011 Elsevier B.V. All rights reserved.
Annealing
Finite element method
Measurement
Validation

1. Introduction engineering steels to cyclic thermo-mechanical loading. Mixed


isotropic–kinematic hardening is more realistic, and such models
Finite element (FE) computational methods are being applied are available in commercial finite element codes such as ABAQUS
with increasing frequency to predict residual stresses in welded [17]. But there are many ways of fitting these models and the re-
engineering structures [1–4]. However, these methods involve quired details are rarely published. Usually, their implementation
complex non-linear analyses where many assumptions and is simplified in order to produce a computationally robust model
approximations have to be made by the analyst [5]. Given the com- that may be fitted using a practicable amount of materials charac-
plexity of the analyses being attempted, it is entirely possible to terisation testing.
produce plausible but incorrect residual stress predictions. For In this paper we set out to identify an optimised approach for
example, the deceptively simple problem of a single weld bead simulating the hardening behaviour of austenitic stainless steel
deposited on a flat plate has proved to be a challenge for both weldment materials (i.e. parent and weld metal) that results in
the measurement and the prediction of weld residual stresses the most accurate prediction of residual stresses in benchmark
[6–10]. multi-pass weld specimens.
The material hardening constitutive model used in weld resid-
ual stress calculations can have a crucial influence on the accuracy 2. Approach
of predicted results [11–15]. The plastic response of an austenitic
steel to a welding thermo-mechanical cycle is potentially complex, An optimised strategy for predicting accurate distributions of
both because of the wide range of imposed temperature, strain rate residual stress in stainless steel weldments is developed using
and strain range, and because of the potential for recovery, recrys- multiple measurements from purpose-designed welded bench-
tallisation, creep and visco-plastic effects at high temperatures mark specimens. A set of benchmark specimens are made from
[16]. Pure isotropic and pure kinematic hardening models are often AISI Type 316L austenitic steel plate and contain one, two or three
used because they are easy to apply using limited materials data, superimposed manual metal arc (MMA) weld beads laid in a
but they do not accurately describe the response of austenitic groove. The groove-weld test specimens are manufactured under
closely controlled conditions. Transient temperatures during weld-
ing are recorded using thermocouple arrays and measurements of
⇑ Corresponding author. Tel.: +44 (0)1452 653132; fax: +44 (0)1452 653025. distortion and residual stress made on the completed specimens.
E-mail addresses: mike.smith@edf-energy.com, mike.ann.smith1@me.com Both diffraction and strain relief techniques are used to ensure
(M.C. Smith). greater confidence in the residual stress measurements.

0927-0256/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.commatsci.2011.10.024
M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328 313

First, the measured temperature histories during welding, and Four specimen base-plates were machined from 50 mm thick
cross-sectional macrographs revealing the extent of the weld fu- AISI Type 316L austenitic stainless steel plate remaining from the
sion boundary, are used to calibrate the size, shape and intensity NeT Task Group 1 international round robin [6]. This material
of the heat source used in computational simulations of the had already been extensively characterised for that project. The fi-
groove-weld specimens. Second, the final deformed shape of the nal size of each specimen was (200  180  25) mm with a 10 mm
three-pass benchmark specimen is used to evaluate the merit of deep groove machined into the top surface at the centre parallel to
performing a moving heat source simulation instead of a simpified the long edge (see Fig. 1).
‘‘block-dumped’’ analysis (i.e. where the entire length of the weld After machining, three of the specimens were instrumented
bead is simultaneously introduced into the model). with thermocouples and strain gauges. The thermocouple posi-
Several different strategies are used to derive isotropic–kine- tions were carefully chosen so that transient measured tempera-
matic hardening model parameters suitable for simulating the tures could be used to calibrate the welding heat input
thermo-mechanical behaviour of welding. The required tensile (effectively the weld efficiency, g) during subsequent FE analysis.
data are derived from isothermal strain-controlled cyclic tests on They were thus located sufficiently far from the welding torch to
small-scale parent material and partially annealed weld metal ensure that their response was insensitive to details of the welding
samples representative of the groove-weld test specimen materi- heat source type, size, and shape, but close enough to allow accu-
als. The test parameters (temperature, strain rate and strain ampli- rate determination of the welding efficiency [7,9].
tude) are selected based upon information from previous FE A MMA process was employed to deposit the weld beads with a
simulations of welding. target electrical heat input of 1.8 kJ/mm. The welder was in-
The residual stress measurements from the one and three-pass structed to produce as uniform a weld as possible. All relevant
groove-weld specimens are then compared with predictions from welding parameters were recorded for use in subsequent FE anal-
several hardening model variants to evaluate the most accurate yses, in addition to the transient thermocouple and strain gauge re-
modelling approach. sponses. The plates were left unrestrained during welding, because
Finally, the performance of the preferred material hardening this is the only boundary condition that can be unambiguously
models is checked by comparing the predicted yield strength vari- reproduced in subsequent FE simulations. One plate was welded
ation across the three-pass groove-weld with the measured stress– with a single stringer bead, one with two superimposed stringer
strain response within the weld, heat affected zone (HAZ) and un- beads, and two plates with three superimposed beads. All beads
strained parent material on either side of the weld. The spatially re- were laid in the same direction. A completed, instrumented
solved stress–strain measurements are obtained from cross-weld three-pass specimen is shown in Fig. 2.
tensile specimens extracted from the three-pass groove-weld spec-
imen. This comparison is used to confirm that sufficient isotropic 3.2. Cross-weld samples
hardening is included in the mixed hardening models and to assess
the annealing strategy adopted in the simulations. A 25 mm long slice of material was removed from the weld stop
ends of the single-pass, two pass, and one of the three-pass groove-
3. Test components weld specimens. The transverse cut faces were polished and etched
to reveal the cross-sectional weld bead fusion boundary profiles,
3.1. Benchmark groove-weld specimens see Fig. 3. The fusion boundaries at this cut face, 25 mm from the
stop end, were judged to be representative of mid-length ‘‘steady
A set of groove-weld test specimens, having the pre-weld state’’ welding conditions.
geometry illustrated in Fig. 1, was designed for the simulation opti- A further 3 mm thick slice was then removed from the weld
misation studies. The groove is partially or fully filled with one, stop end of each of the three specimens for production of ‘stress
two or three superimposed stringer beads, producing welds with free’ combs, for the measurement of the unstressed lattice param-
minimum variation in thermal history, dimensions, and residual eter, a0, using neutron diffraction.
stress over much of the length of the specimen. The variation in A 50 mm long slice was then removed from the stop end of the
number of passes produces weld and adjacent HAZ material that remaining three-pass groove-weld specimen. Two flat-sided cross-
has undergone either one, two or three clearly defined thermo- weld tensile specimens having the design shown in Fig. 4 were
mechanical cycles. electro-discharge machined from the inside edge of the block (i.e.

Fig. 1. Layout of groove weld benchmark specimen before welding.


314 M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328

Fig. 2. Completed 3-pass groove weld specimen, showing top surface


thermocouples.

from weld material that would have been laid down under near
steady state conditions). The tensile specimens were extracted in
such a way to ensure that the gauge length passed through the first
and final weld beads.

3.3. Stress–strain test specimens

Isothermal monotonic and uniaxial cyclic (low cycle fatigue)


stress–strain test data are required to fit the parameters for mate-
rial hardening models. No new tests on the groove-weld base
material were performed as sufficient data were available from
other research studies [18,19]. Instead, testing concentrated on
representing the complex behaviour of weld metal.
Single weld pass tensile and low cycle fatigue specimens were
fabricated by extracting cylinders of test material from high heat
input (typically 2.5–3.0 kJ/mm) single-pass groove-welds. Exten-
sion pieces were then electron-beam welded to either end of each
cylinder before machining to the final specimen geometry. This
procedure gave 4 mm gauge diameter weld metal specimens com-
pliant with the relevant standards [20–22]. The test specimens
were then given a ‘‘spike anneal’’ heat treatment in order to simu-
late the effect of adjacent bead deposition in a multi-pass weld.
This was achieved by rapid induction heating to a peak tempera-
ture of 850 °C followed by unassisted air cooling to ambient tem-
perature. This heat treatment reduced the 1% proof stress of the
single-pass material by about 35 MPa to 364 MPa.
Multi-pass weld tensile and low cycle fatigue specimens of the
same design were also made by extracting samples from a large mul-
ti-pass weld pad. This pad was fabricated using MMA electrodes of
the same specification used to manufacture the welded benchmark
Fig. 3. Transverse macrographs taken from 25 mm sections cut from the weld stop
specimens. Tensile tests at room temperature revealed that the mul- end of (from top to bottom) single-pass, two pass, and three-pass groove weld
ti-pass pad material had a 1% proof stress of 424 MPa. A full solution specimens.
heat treatment at 1050 °C for 45 minutes applied to the multi-pass
pad specimens reduced the 1% proof stress of the pad material by for both monotonic and cyclic testing, and strain range for cyclic
about MPa to 325 MPa. While solution heat treatment probably re- testing. Strain rates were chosen to be representative of conditions
sults in mechanical properties most representative of virgin weld during the later part of cooling, when the temperature is below
metal after deposition, it overstates the softening effect of adjacent 600 °C, the material has significant strength, and the final tensile
bead deposition, a process experienced by every weld bead in a mul- residual stress field is developing. Strain ranges for cyclic testing
ti-pass weld except the final capping bead. were chosen to be representative of the average strain range expe-
rienced by weld metal, and by parent material in the HAZ adjacent
4. Measured data to the weld.
Table 1 provides details of the stress–strain data used to derive
4.1. Stress–strain properties hardening parameters for the groove-weld parent and HAZ materi-
als. Monotonic tensile test data for the AISI Type 316L plate mate-
Results from previous FE simulations of a number of weldment rial were obtained from the European network project, NeT [6].
configurations were used to inform the choice of strain rate used Extensive isothermal monotonic and cyclic stress–strain data for
M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328 315

Fig. 4. 50 mm section cut from the weld stop end of a three-pass groove weld specimen, showing locations of cross-weld tensile specimens.

Table 1
Monotonic and cyclic uniaxial test data for Type 316 parent materials used to derive Lemaitre–Chaboche model parameters.

Source Material Available data


NeT task Group 1 AISI type 316L plate Monotonic stress–strain: room temperature, 275 °C, 550 °C, 750 °C; 3 tests at each temperature,
[6] 2.8  104s1
VORSAC project AISI type 316H forging (ex-service steam Monotonic testing at two strain rates and in two orientations from room temperature to 1200 °C, 62
[18,19] header) tests
Circumferential direction, 4.2  105 s1, 100 °C intervals from RT to 1200 °C, plus 275 °C, 525 °C
and 850 °C
Circumferential direction, 4.2  104 s1, 100 °C intervals from RT to 400 °C, then 900 °C, 1000 °C,
1100 °C
Axial direction, 4.2  105 s1, RT and 525 °C
Isothermal cyclic testing at a total strain range of 3%, at 4.2  105 s1 from room temperature up to
900 °C, 19 tests

RT = room temperature.

AISI Type 316H stainless steel were obtained from the VORSAC to produce 3D models of the final welded configuration. These
Euratom Vth Framework project [18,19]. The uniaxial cyclic tests were required primarily to allow design and control of the neutron
employed a high total strain range of 3%, which is a reasonable diffraction residual stress measurements but also provided an
bound to the average cyclic strain ranges expected in HAZ material accurate record of the distortion introduced by welding.
close to a weld.
Details of the weld stress–strain properties obtained at several 4.4. Groove-weld residual stresses by neutron diffraction
temperatures up to 950 °C are listed in Table 2. A strain rate of
4  104 s1 was employed for most of the tests. The cyclic tests Residual stress measurements were made on transverse cross-
were performed at a total strain range of 1.5%, which bounds the sections at mid-length of both the single-pass and the three-pass
average strain range expected for weld metal deposited in a mul- groove-weld specimens, using neutron diffraction and the contour
ti-pass weld. method. Details of measurements on the single-pass weldment can
be found in [24,25].
4.2. Cross-weld stress–strain mapping by ESPI Neutron diffraction residual strain measurements were carried
out on the Engin-X instrument at the ISIS facility of the Rutherford
Tensile tests on cross-weld specimens (Fig. 4) were designed to Appleton Laboratory, UK [26]. The experiments were planned and
obtain a continuous record of the tensile property variation across controlled using the SScanSS software package [27]. By using the
the three-pass groove-weld. During tensile testing surface strain 3D laser scan of the test specimen deformed shape in the software,
was measured using electron speckle pattern interferometry (ESPI) each measurement was positioned with high precision. Measure-
to provide a continuous record of the full field surface strain, there- ments were made in three orthogonal directions at a number of
by allowing the monotonic stress–strain curve to be extracted at locations in a transverse plane at bead mid-length. All diffraction
any point along the specimen. Full details of the technique and measurements were made using a (4  4  4) mm gauge volume.
measurements are given in [23]. Measurements of the stress-free lattice parameter, a0, in both par-
ent and weld metal were made using a comb specimen extracted
4.3. Groove weld deformation by laser scanning from the three-pass specimen, and small cuboids extracted from
a similar specimen manufactured by the same welder on the same
The shortened 170  180  25 mm specimens were scanned day using the same welding parameters and consumables, and the
using a coordinate measuring machine fitted with a laser probe, same 316L parent material [28].
316 M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328

Table 2
Monotonic and cyclic uniaxial test data for Type 316L weld metal used to derive Lemaitre–Chaboche model parameters.

Material Available data


Multi-pass pipe girth weld (VORSAC Monotonic testing at two strain rates and in two orientations from room temperature to 1000 °C, 33 tests
project, [18]) Longitudinal direction, 4.2  105 s1, 20 °C, 275 °C, 525 °C, 700, 850 °C, 1000 °C
Longitudinal direction, 4.2  104 s1, 20 °C, 275 °C, 525 °C, 700, 850 °C, 1000 °C
Transverse direction, 4.2  105 s1, RT and 525 °C
Single-pass weld laid in a groove Monotonic testing at 4  104 s1 in longitudinal direction from room temperature to 950 °C, 15 tests
5 specimens tested at RT (2 as-received, 4 spike-annealed at various temperatures)
9 specimens tested at 200 °C, 400 °C, 600 °C, 700 °C, 900 °C, 2 at each temperature except 900 °C, all spike-annealed at 850 °C
Isothermal cyclic testing at a total strain range of 1.5% and same strain rate, two specimens each at 20 °C, 200 °C, 400 °C, 600 °C,
and 700 °C, all spike-annealed at 850 °C
Multi-pass weld pad Monotonic testing at 4  104 s1 from room temperature to 950 °C, 16 tests
Six specimens tested at RT (1 spike-annealed at 850 °C, 3 solution-treated at 1050 °C)
10 as-received specimens tested at 200 °C, 400 °C, 600 °C, 700 °C, 900 °C, 2 at each temperature
Isothermal cyclic tensile testing at a total strain range of 1.5% and same strain rate, two as-received specimens each at 20 °C,
200 °C, 400 °C, 600 °C, and 700 °C

RT = room temperature.

4.5. Groove-weld residual stresses by the contour method Kinematic hardening is defined as an additive combination of a
purely kinematic term (the linear Ziegler hardening law) and a
The contour method is a destructive strain relaxation technique relaxation term (the recall term), which introduces nonlinearity.
that can map the direct component of residual stress normal to a When temperature and field variable dependencies are omitted,
cut surface in a relatively cheap and time-efficient manner [29]. the kinematic hardening law is:
The technique is based on Bueckner’s superposition principle X 1 
[30], where a specimen containing residual stresses is cut into a_ ¼ C i 0 ðr  aÞe_ pl  ci ae_ pl ð3Þ
i
r
two parts and the resulting relaxed surface displaced shape (con-
tour) is measured. The measured surface contour is then restored where C defines the initial kinematic hardening modulus, and c
to its original flat profile in an elastic FE model, replicating the ori- determines the rate at which the kinematic hardening modulus de-
ginal residual stress field. This procedure allows the full field resid- creases with increasing plastic deformation. r is the stress tensor,
ual stress normal to the cut surface to be measured on the cut r0 is the equivalent stress defining the size of the yield surface,
plane. The contour method was applied to the one and three-pass and e_ pl is the equivalent plastic strain rate.
groove-weld specimens after neutron diffraction measurements The isotropic hardening component of the model defines the
were complete in order to map the distribution of longitudinal evolution of the yield surface size, r0, as a function of the equiva-
residual stress on a transverse plane (i.e. perpendicular to the lent plastic strain, or plastic path length, epl .
pl
weld) located at specimen mid-length [24,25]. r0 ¼ rj0 þ Q inf ð1  ebe Þ ð4Þ
where r|0 is the yield stress at zero plastic strain, and Qinf and b are
5. Lemaitre–Chaboche mixed hardening model material parameters. Qinf is the maximum change in the size of the
yield surface, and b defines the rate at which the size of the yield
The Lemaitre–Chaboche hardening model [31] is designed pri- surface changes as plastic straining develops. The amount of isotro-
marily to represent cyclic inelastic loading of metals. The mixed pic hardening thus depends on the total plastic path length in the
isotropic–kinematic formulation allows the model to describe both structure: a larger number of low amplitude strain cycles are re-
the Bauschinger effect, where the yield is reduced upon load rever- quired to achieve the same amount of hardening as a few large
sal after plastic deformation during initial loading, and cyclic hard- amplitude strain cycles.
ening with plastic shakedown, where soft or annealed metals tend
to harden towards a stable limit during cyclic loading. Both these 5.1. Fitting mixed hardening model parameters
phenomena are important during welding.
The Lemaitre–Chaboche model is implemented in the ABAQUS The five parameters for the mixed-hardening model are nor-
FE code [17] where the non-linear kinematic and isotropic parts mally fitted to the results of isothermal monotonic tensile tests
of the model are defined separately as follows. and cyclic uniaxial strain-controlled tests.
The kinematic component describes translation of the yield sur- The three kinematic parameters C, c, and r|0 are fitted first. The
face in stress space via the backstress tensor a. The pressure-inde- most accurate idealisation of the cyclic response is obtained by fit-
pendent yield surface, f, is defined by the function: ting C and c to the shape of the saturated cyclic stress–strain loop
from a symmetric strain-controlled test. However, this does not
f ðr  aÞ ¼ r0 ð1Þ predict the monotonic response of austenitic stainless steel well.
The model is too ‘‘soft’’, with the proportional limit stress too
where r0 is the radius of the yield surface and f(r  a) is the equiv- low, and insufficient work hardening beyond the peak strain of
alent von Mises stress, defined by: the cyclic test used for fitting. This limitation can have a significant
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi impact on weld residual stress computations. An alternative strat-
3 egy is to fit the parameters C, c, and r|0 to the monotonic tensile
f ðr  aÞ ¼ ðS  adev Þ : ðS  adev Þ ð2Þ
2 response of the material up to a suitable maximum plastic strain.
Integration of the backstress evolution law of Eq. (3) over a half cy-
here adev is the deviatoric part of the backstress tensor, and S is the cle yields the relationship:
deviatoric stress tensor, defined as S = r + pI, where S is the stress X C i  

tensor, p is the equivalent pressure stress, and I is the identity r ¼ rj0 þ 1  expðepl ci Þ ð5Þ
tensor. i
ci
M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328 317

Eq. (5) may be fitted to the monotonic true stress vs. plastic strain adjusted by altering r|0 to achieve the correct 1% proof stress
response of the material up to a chosen peak plastic strain, the mag- using a procedure developed and validated in [32]. This fit is
nitude of which depends on both the expected peak monotonic identified as ‘‘mixed-n’’.
strain excursion expected in the structure, and the need to produce (b) The second fitting approach took advantage of the improved
acceptable cyclic stress–strain loop shapes. functionality available in ABAQUS version 6.8 and higher,
Once the kinematic parameters have been chosen, the isotropic and fitted up to four C, c pairs at each temperature, still
parameter Qinf is obtained from cyclic test data with an appropriate within the constraint of temperature-independent c. This
strain range. It is defined as the difference between the peak stress fit used plate-specific tensile data [6] obtained at a more rel-
at the peak strain of the first loading quarter cycle and one half the evant strain rate of 4  104 s1. The fit was limited to 2%
total stress range, Dr/2, either at saturation or at a cycle number plastic strain. This model fit is identified as ‘‘mixed-rr’’.
deemed appropriate. The hardening rate parameter b is then fitted (c) The third approach fitted the kinematic parameters to the
to achieve an appropriate hardening rate. saturated cyclic response, rather than the monotonic
For welding simulations the above procedure can be used to fit response. These parameters had been obtained during the
the model to measured isothermal stress–strain data at various VORSAC project, and fortuitously predicted a 1% proof stress
temperatures. However, the transient thermo-mechanical history very close to that of the 316L plate used in the current study,
experienced during welding is complex, with potentially a wide so could be used with no further modification. This fit is
variation in strain range, strain rate and temperature. The rela- identified as ‘‘mixed-mm’’.
tively simple mechanical constitutive model cannot explicitly han-
dle all of these variables, and this can present problems with its A paucity of cyclic test data (because many specimens buckled)
implementation, for example: led to an alternative fitting approach for the isotropic parameters.
First, it was observed that the available cyclic tests all cyclically
 The isotropic hardening parameters Qinf and b can only be fitted hardened to close to the 0.2% proof stress of the available multi-
over a single strain range, whereas for stainless steel the pass weldment data. Secondly, it was noted that hardness maps
amount of cyclic isotropic hardening tends to increase with for Type 316 stainless steel weldments show a smooth increase
increasing strain range. through the HAZ with no significant discontinuity at the fusion
 The kinematic parameters Ci and ci cannot vary with strain boundary [23,32]. Qinf was thus derived from the difference be-
range and are used to describe both the monotonic and cyclic tween the 0.2% proof stress of multi-pass weld data [18], taken
response of the material. This can lead to difficulties in match- to represent the cyclically saturated response, and the monotonic
ing both the monotonic and cyclic response. stress at 0.75% total strain. The hardening rate parameter, b, was
 The model is calibrated using the results of isothermal uniaxial selected to achieve saturation in 20 cycles at 1.5% total strain
tests, where the loading is proportional (the principal stress range, with a single value used at all temperatures.
axes do not rotate). The actual hardening behaviour of a mate-
rial during a thermo-mechanical fatigue cycle (TMF loading) or
5.4. Characteristics of the fitted models
under non-proportional loading may differ from that observed
in an isothermal uniaxial test.
Fig. 5 compares the predictions of the fitted weld metal model
with the measured monotonic and cyclic response at room temper-
5.2. Weld metal fitted parameters
ature. The monotonic response is represented well, although the
restriction to a single C–c pair has led to a slight over-prediction
The kinematic parameters C, c and r|0 were fitted to the mono-
of the monotonic response close to the proportional limit. As ex-
tonic true stress vs. plastic strain response up to 2% plastic strain of
pected, the cyclic response is ‘‘over-predicted’’, in the sense that
spike annealed single-pass weld metal under the constraints of
the model predicts a less pronounced Bauschinger effect than ob-
both temperature-independent c, and of only a single C, c pair at
served in the tests, with the predicted cyclic response outside
each temperature.
and more rhomboidal than the observed response. It is also evident
The isotropic parameters Qinf and b were fitted over the first ten
that weld metal cyclically saturates very rapidly, with little change
cycles of loading, because the weld metal was observed to first
in the stress–strain loops after the first full cycle.
cyclically harden, taking 10–20 cycles to achieve saturation, and
The three parent mixed hardening material models predict
then later to soften. It was judged appropriate to fit only to the ini-
rather different cyclic responses. Fig. 6 plots the predicted response
tial hardening part of the response since the total plastic path
for the first half cycle of a test with 2% total strain range. The peak
length and the number of cycles were close to the values expected
stress achieved on initial loading is approximately the same for all
in a typical multi-pass weld.
three models, because they agree at around 1% total strain. This
would not be the case if load reversal occurred after a different ini-
5.3. Parent material fitted parameters
tial loading strain. On unloading, the stress achieved depends
strongly on both the mixed model and the unloading strain incre-
Monotonic and cyclic stress–strain data for AISI Type 316 par-
ment. Thus the onset of reversed yield is predicted over a large range
ent material from several sources, see Table 1, were used to devel-
of stress between +40 MPa and 200 MPa, while at zero strain the
op three model fits to the kinematic parameters:
three models agree closely again. Although material in a multi-pass
weld undergoes several complete loading cycles, the differences be-
(a) The first used similar assumptions to the weld metal fit,
tween the models tend to persist after repeated load reversals.
namely temperature-independent c, and only a single C, c
pair at each temperature. This fit used the largest available
single-cast database of monotonic data; that is VORSAC tests 6. Prediction of weld residual stresses
[18] performed at the slow strain rate of 4.2  105 s1, fit-
ted out to 5% plastic strain to reflect the higher strain excur- A programme of FE modelling was undertaken in order to opti-
sions expected in parent material. The VORSAC cast of 316H mise and validate the mixed hardening material models. A 3D half-
material has a higher yield strength than the 316L plate used model of the groove-weld specimen was constructed for weld
in the current study, so the fitted kinematic parameters were residual stress simulation, as shown in Fig. 7a, using the ABAQUS
318 M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328

500 dimensions as the groove-weld model, but with the mesh designed
Cyclic test t3_217
Weld mixed model
to allow removal and subsequent loading of the cross-weld tensile
400
Monotonic test ZS15 specimens.
Monotonic test ZS13
300
6.1. Thermal analysis of welding
200
Stress (MPa)

An accurate thermal simulation of the welding process is a pre-


100
requisite for obtaining an accurate prediction of residual stresses,
0 since the applied temperature field creates the thermal strain his-
tory that loads the structure. The thermal simulation requires de-
-100 tailed knowledge of the welding conditions.
-200 A specialised welding heat source fitting tool [33], was used to
optimise a simple ellipsoidal volumetric heat source. This enabled
-300 the welding efficiency, g, and heat source geometry to be opti-
mised to match both the observed fusion boundaries and the mea-
-400
sured transient temperatures in the specimen during welding. First
-500 the global heat input was calibrated by matching the measured re-
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
sponse of far-field thermocouples [7,10] that were insensitive to
strain (%) the heat source shape. With g known, the heat source geometry
Fig. 5. Comparison of the weld metal mixed hardening model with the measured
was adjusted to predict the observed fusion boundary, giving an
monotonic and cyclic response at room temperature. optimised near-field solution.
Both block-dumped and MHS thermal simulations of welding
were then performed using ABAQUS. In both cases the analysis
400 commenced with removal of the elements representing beads 2
mixed n and 3. Bead 1 was then heated in situ by the chosen heat source.
In the block-dumped simulation the entire bead was heated simul-
300 mixed mm
taneously, while in the MHS simulation the source moved progres-
mixed rr
sively from one end of the bead to the other. The portion of the
200 bead not already laid down was thermally isolated from the
remainder of the mesh by using ABAQUS field-variable dependent
100
material properties. When the field variable is set to 2, the integra-
Stress (MPa)

tion point concerned has its conductivity reduced by two orders of


magnitude, while when it is set to 1 the conductivity is at its nor-
0 mal level. As the torch approaches, the field variable is changed
smoothly from 2 to 1 over a distance equal to the effective torch
-100 radius, thereby smoothly interpolating the thermal conductivity
[34]. After welding of pass 1 was complete, the specimen was al-
lowed to cool to the inter-pass temperature. The procedure was re-
-200 peated for beads 2 and 3, starting with re-activation of the relevant
elements, and then the entire specimen was allowed to cool to
-300 ambient conditions.
The thermal properties assumed for Type 316L stainless steel
material are given in Table 3. The thermal analyses assumed that
-400 heat was lost from the plate due to free convection and radiation.
-1 -0.5 0 0.5 1 1.5
Radiative heat transfer was based upon heat exchange with a large
Strain (%)
enclosure at 20 °C, with an emissivity of 0.35. Temperature depen-
Fig. 6. Comparing the predicted responses after first load reversal of mixed- dent convective heat transfer coefficients were applied to the top
hardening material models with different parameter fitting strategies. surface of the plate, varying smoothly from 4.2 W/m2 K at 20 °C
to 13.5 W/m2 K at 1400 °C. The sides and bottom of the specimen
reach much lower temperatures than the top surface, so fixed con-
FE code [17]. The model made provision for the introduction of one, vective heat transfer coefficients of 3.0 W/m2 K and 7.0 W/m2 K
two or three superimposed weld beads (not shown in Fig. 7a), and were applied to the plate bottom and sides respectively.
had a mesh design optimised for a full moving heat source (MHS)
weld analysis with simulation of progressive addition of weld filler 6.2. Mechanical analysis of welding
material. Although the groove-weld specimen appears to be almost
two dimensional, and therefore amenable to simplified 3D ‘‘block- The groove-weld plate was assumed to be unrestrained during
dumped’’ analysis techniques in which the entire bead length is welding, with a symmetry condition on the plate longitudinal
deposited simultaneously, it was judged important to make the mid-plane and sufficient additional restraints merely to prevent ri-
most accurate simulations possible. The model was constructed gid body motion. Several small displacement mechanical analyses
using 20-node reduced integration solid elements, with elements were performed, examining the effects of material hardening mod-
close to the weld bead employing a hybrid formulation (see el, high temperature annealing behaviour, and thermal modelling
Fig. 7b). It contained 102,294 nodes and 23,700 elements. Heat strategy on the predicted stress and displacement fields. All fol-
transfer elements substituted for the mechanical elements for lowed the same basic analysis route.
the initial thermal analysis. The analyses commenced with removal of beads 2 and 3, leav-
A second FE model with a finer mesh was constructed to simu- ing bead 1 in situ at ambient temperature. The transient tempera-
late the cross-weld stress–strain tests. This had the same external ture history for pass 1 was then applied. Block-dumped analyses
M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328 319

Fig. 7. 3D half-model of groove weld. (a) Complete mesh with weld beads removed. (b) Transverse cross section at mid-length showing hybrid elements in weld bead and
HAZ.

Table 3
Thermal and physical properties for AISI 316L material.

Temp °C Weld Parent Parent and weld Young’s modulus GPa


Specific heat kJ/Kg/°C Specific heat kJ/Kg/°C Conductivity W/m/°C Density Kg/m3 Thermal Exp  106 mm/mm/°C Parent Weld
20 0.488 0.492 14.12 7966 14.56 195.6 168.2
100 0.502 0.502 15.26 15.39 191.2 164.1
200 0.520 0.514 16.69 16.21 185.7 158.3
300 0.537 0.526 18.11 16.86 179.6 152
400 0.555 0.538 19.54 17.37 172.6 144.9
500 0.572 0.550 20.96 17.78 164.5 136.9
600 0.589 0.562 22.38 18.12 155.0 127.7
700 0.589 0.575 23.81 18.43 144.1 117.4
800 0.589 0.587 25.23 18.72 131.4 105.8
900 0.589 0.599 26.66 18.99 116.8 92.8
1000 0.589 0.611 28.08 19.27 100.0 78.3
1100 0.589 0.623 29.50 19.53 80.0 62.4
1200 0.589 0.635 30.93 19.79 57.0 44.5
1300 0.589 0.647 32.35 20.02 30.0 23.4
1400 0.589 0.659 33.78 20.21 2.0 1.6

Poisson’s ratio was set to 0.29 for parent material and 0.294 for weld metal.

heated the entire bead simultaneously, while the bead was heated cooled to interpass temperature conditions. Beads 2 and 3 were
progressively from one end to the other in MHS analyses. The part then deposited using the same approach. The final residual stress
of the bead not yet deposited was mechanically isolated from the state was obtained by simulating removal of a 25 mm slice of
remainder of the mesh using the same field variable approach material from the stop-end of the specimen.
adopted for the thermal MHS simulations. In this case, the field
variable was initially set to 2, when the integration points con- 6.3. Material annealing behaviour
cerned had their mechanical properties set equal to those normally
assumed at 1400 °C. Here both the yield strength and elastic mod- ABAQUS has a built-in annealing facility that can eliminate the
ulus are very low, so the bead makes no effective contribution to isotropic hardening history at a single temperature. However,
the overall structural response. When the field variable was set applying this facility in weld simulations tends to introduce dis-
to 1 normal temperature dependent mechanical properties were continuities in equivalent plastic strain and yield strength and pro-
assumed. After completion of welding for pass 1, the plate was duce perturbations in predicted longitudinal stress in parent
320 M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328

Table 4
Room temperature mechanical properties of Type 316 material used in welding simulations.

Identifier Weld model Parent model kinematic parameter fit Isotropic annealing Kinematic annealing
behaviour temperature
Isotropic Perfectly plastic, ro = 446 MPa Piecewise linear isotropic, ro = 210 MPa Conventional, 1400 °C n/a
Mixed-n Mixed hardening model for MMA Single C, c pair fitted to monotonic VORSAC tests Conventional, 1050 °C 850 °C
single-pass weld metal ro = 232.9, C = 4997, c = 34 at 20 °C parent, 850 °C weld
ro = 280, C = 18,800, c = 150 at 20 °C Qinf = 191.6, b = 6.9, at 20 °C
Qinf = 69.1, b = 53.2, at 20 °C
Progressive Two-stage, 800 °C and
annealing 1300 °C
Mixed mm Single C, c pair fitted to saturated cyclic response of Conventional, 1050 °C 1000 °C parent,
VORSAC isothermal cyclic tests parent, 850 °C weld 850 °C weld
ro = 100, C = 52,506, c = 295.3 at 20 °C Conventional, 1050 °C
Qinf = 177.4, b = 5.3, at 20 °C parent, 850 °C weld
Mixed-rr Multiple C, c pairs fitted to monotonic tensile response
of NeT plate
ro = 125.6, C1 = 156,435, c1 = 1411, C2 = 6134,
c2 = 47.19, at 20 °C
Qinf = 153.4, b = 6.9, at 20 °C
Non-linear ro = 280, C = 18,800, c = 150 at 20 °C ro = 232.9, C = 4997, c = 34 at 20 °C n/a 850 °C
kinematic Qinf = 0 at 20 °C Qinf = 0 at 20 °C

Notes: Although space permits the presentation only of room temperature data, temperature dependent data were both derived and used for all analyses.
The units of ro, C, and Qinf are MPa; c and b are dimensionless.
The kinematic annealing temperature is the temperature at which C falls to zero.

material close to the weld fusion boundary. It was judged that high heat source analyses were performed for this material
temperature softening behaviour of Type 316 material could be behaviour combination, and the block-dumped solution
much better modelled by defining a two-stage annealing process. was mapped onto the cross-weld simulation mesh.
Above a lower annealing temperature, T1, it is assumed that the (c) The non-linear kinematic hardening analyses used the
material ceases to exhibit any further isotropic hardening, but does mixed-n parent, and mixed hardening weld models, but with
not lose any already accumulated at lower temperatures, while Qinf set to zero to suppress isotropic hardening. Both block-
above an upper annealing temperature, T2, the equivalent plastic dumped and MHS analyses were performed for this material
strain is set to zero, eliminating any prior isotropic hardening. A behaviour combination. The ABAQUS annealing temperature
user subroutine using the UHARD facility within ABAQUS was does not affect the kinematic response.
developed to implement the two-stage annealing scheme with (d) The mixed-rr analysis used a multiple Ci–ci fit to the parent
the Lemaitre–Chaboche mixed hardening model (full details are gi- monotonic response, and the mixed hardening weld model,
ven in [35]). This advanced functionality was implemented for with conventional ABAQUS annealing at 1050 °C for parent
both parent and weld material in one of the studies reported here. and 850 °C for weld metal. Only a MHS analysis was per-
Note that the kinematic hardening history is eliminated (or formed for this material behaviour combination.
‘‘annealed’’) at the temperature when the material becomes (e) The mixed-mm analysis used the single C–c fit to the parent
elastic – perfectly plastic: that is, the back-stress is eliminated, saturated cyclic response, and the mixed hardening weld
and the centre of the yield surface returns to the origin. This oc- model, with conventional ABAQUS annealing at 1050 °C for
curred between 850 °C and 1000 °C for the models used in this parent and 850 °C for weld metal. Only a MHS analysis
study, see Table 4. was performed for this material behaviour combination.
(f) Finally, a single block-dumped analysis was performed using
6.4. Material hardening behaviour the same parent and weld models as the mixed-n analysis,
but using two-stage annealing (see Section 6.3), with lower
A number of different material hardening models were used in and upper annealing temperatures set to 800 °C and
the mechanical simulations. These included pure isotropic harden- 1300 °C respectively. This solution was also mapped onto
ing, pure kinematic hardening, and the mixed isotropic–kinematic the cross-weld simulation mesh
formulations described earlier:
The physical properties used are given in Table 3, while the
(a) The isotropic hardening analyses used a simple piece-wise room temperature hardening parameters are given in Table 4.
linear fit to the monotonic tensile response of parent mate-
rial, with a fully plastic cut-off at 10% plastic strain, coupled 7. Results
with a perfectly plastic weld model, with its yield strength
set to the mean 1% proof stress of multi-pass weld metal. 7.1. 3-pass groove-weld deformation
A conventional ABAQUS annealing temperature of 1400 °C
was assumed for both constituents (see Section 6.3). Both Fig. 8 compares the measured bottom surface deformation
block-dumped and MHS analyses were performed for this profile of the three-pass groove-weld specimen at the mid-length
material behaviour combination, and the block-dumped measurement plane with three FE predictions. Two predictions
solution was mapped onto the cross-weld simulation mesh. used a block-dumped simulation methodology, while the third
(b) The mixed-n analyses used a single C–c fit to the parent used a moving heat source. The groove-weld specimen showed
monotonic response and the mixed hardening weld model, substantial deformation during welding, with the two halves of
with conventional ABAQUS annealing at 1050 °C for parent the plate rotating about a plastic hinge beneath the weld. The
and 850 °C for weld metal. Both block-dumped and moving two block dumped predictions grossly under-predict the mea-
M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328 321

3.5

Measured profile
3 Isotropic hardening, block dumped

mixed hardening, block dumped

2.5 mixed hardening, moving heat source

2
y (mm)
1.5

0.5

0
-100 -80 -60 -40 -20 0 20 40 60 80 100
x (mm)

Fig. 8. Comparison of mid-length bottom surface profile of three-pass groove weld specimen with predicted profiles from block-dumped and moving heat source simulations.

Fig. 9. Predicted stresses on the measurement plane of a groove weld specimen after pass 3 for ‘‘mixed-rr’’ analysis. (a) Transverse stress (S11). (b) Longitudinal stress (S33).

sured deformation, with relatively little sensitivity to material 7.2. 3-pass groove-weld residual stresses
hardening model, while the MHS simulation predicts approxi-
mately the correct deformed shape (the average vertical displace- There is much less difference between residual stresses pre-
ment at the edge of the plate is over-predicted by about 8% dicted by the two bead deposition methodologies. MHS predictions
compared with under-prediction by a factor of nearly 5 for the are subtly better than block-dumped predictions using the same
block-dumped analyses). At first sight this is a surprising discrep- material model, so only these are compared with residual stress
ancy, given that a specimen of this design might be thought ame- measurements. Nevertheless, the block-dumped predictions would
nable to block-dumped analysis. It appears that progressive bead normally be judged perfectly acceptable.
deposition, with elements ahead of the torch isolated from the Fig. 9 plots predicted transverse and longitudinal stresses on the
remainder of the mesh, allows considerably more plastic rotation measurement (mid-length) plane of the three-pass specimen for
than any quasi-2D approach. Note that an additional large dis- the mixed-rr analysis. High residual stresses are predicted near
placement sensitivity analysis produced little difference in the the middle of the specimen, as would be expected, with both trans-
displacement results. verse and longitudinal stresses peaking in weld beads 1 and 2, and
322 M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328

350
ND measured stress
300
Mixed-rr
250 Isotropic hardening

200 Non-linear kinematic hardening

Stress (MPa) 150

100

50

-50

-100

-150
-15 -10 -5 0 5 10
Distance above weld root (mm)
(a)
700
ND measured stress
600 Contour method measured stress
Mixed-rr
Isotropic hardening
500
Non-linear kinematic hardening
Stress (MPa)

400

300

200

100

0
-15 -10 -5 0 5 10
Distance above weld root (mm)
(b)
Fig. 10. Comparison of measured and predicted through-wall stress distribution at mid-length of the three-pass groove weld specimen, showing the effect of moving from
isotropic through kinematic to mixed hardening material models. (a) Transverse stress (S11). (b) Longitudinal stress (S33).

a further peak in transverse stress on the bottom face of the measurements except in weld metal (ordinate values, x > 0), where
specimen. it over-predicts the stress by 100 MPa in the longitudinal direc-
Because both plastic deformation during welding and residual tion. The increase in predicted stress from this model at the fusion
stresses are concentrated in the centre of the specimen, the most boundary may be owing to the use of ‘‘spike-annealed’’ rather than
appropriate residual stress measurement profile to compare with solution heat-treated stress–strain properties in the weld mixed
predictions is a through-thickness ‘‘drill-down’’ line on the weld hardening model.
centre-line at specimen mid-length. A similar through-wall line Fig. 11 compares measured transverse and longitudinal stresses
was found to give most insight into simulation performance for a in the three-pass specimen with predictions made using the three
single weld bead-on-plate benchmark [8,10,11,36]. mixed hardening fits. Predicted stresses in weld metal differ little,
Fig. 10 compares measured transverse and longitudinal stresses because all three simulations use the same weld model. However,
in the three-pass specimen with predictions made using isotropic, there are subtle differences between the predicted stresses in par-
kinematic and a mixed hardening model (mixed-rr with a multiple ent material, particularly in the longitudinal direction. The mixed-n
Ci–ci fit to the parent monotonic response). The results show a very single C–c fit to the monotonic parent response over-predicts lon-
clear ranking in predictive performance of the models. The isotro- gitudinal stresses in parent material more than 5 mm beneath the
pic model over predicts the amplitude of both components of weld root. The mixed-mm fit to the saturated cyclic response
stress by a considerable margin. The kinematic model performs under-predicts the longitudinal stress, but gives the best fit to
well except for a notable under-prediction of the longitudinal com- transverse stress measurements. Conversely the mixed-rr model
ponent of stress in parent material just below the weld root (ordi- gives closest agreement with measurements in the longitudinal
nate values, x < 0). The mixed-rr solution agrees best with the direction.
M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328 323

300
ND measured stress
250
Mixed-n
200
Mixed-rr

150 Mixed-mm

Stress (MPa) 100

50

-50

-100

-150
-15 -10 -5 0 5 10
Distance above weld root (mm)
(a)
500

450

400

350
Stress (MPa)

300

250

200
ND measured stress
150
Contour method measurement
100 Mixed-n
Mixed-rr
50
Mixed-mm
0
-15 -10 -5 0 5 10
Distance above weld root (mm)
(b)
Fig. 11. Comparison of measured and predicted through-wall stress distribution mid-length of the three-pass groove weld specimen, showing the effect of different mixed
hardening material model parameter fitting strategies. (a) Transverse stress (S11). (b) Longitudinal stress (S33).

7.3. 1-pass groove-weld residual stresses 7.4. High temperature annealing

Fig. 12 compares measured transverse and longitudinal stres- Finally, Fig. 14a illustrates how the high temperature annealing
ses in the single-pass specimen with predictions made using model affects mixed hardening stress results in weld metal and a
isotropic, kinematic, and the mixed-rr model. The relative perfor- small annulus of HAZ material close to the weld. The stress varia-
mance of the models is the same as observed in the three-pass tion in the HAZ is marginally smoother in the two-stage annealing
specimen, despite the shorter plastic path and single load cycle, simulation, while the stresses in weld are somewhat higher for
with the mixed-rr model providing the most accurate prediction two-stage annealing. The reasons for this behaviour can be seen
in parent material. However, this model over-predicts the longi- in Fig. 14b, which plots the plastic path length along the same
tudinal component of stress in the weld metal by 200 MPa as through-wall line. In parent material the move from full annealing
opposed to 100 MPa for the three-pass specimen. Also contrast at 1050 °C to two-stage annealing at 800 °C and 1300 °C has the
the transverse stresses which are over-predicted by 75 MPa as beneficial effect of removing the sharp and unrealistic peak in
opposed to the close agreement found with the three-pass the plastic path length. Whereas in weld metal, the change from
specimen. full annealing at 850 °C to two-stage annealing has increased the
Fig. 13 compares measured transverse and longitudinal stresses plastic path in beads 1 and 2, since they are no longer fully an-
in the single-pass specimen with predictions made using the three nealed by deposition of an adjacent weld bead unless they melt.
mixed hardening fits. Here the mixed-mm model gives the most Thus two-stage annealing offers a means of reducing unrealistic
accurate stress predictions for parent material. perturbations in predicted stress in HAZ material adjacent to the
324 M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328

300
ND measured stress
250
Mixed-rr
200
Isotropic hardening
150
Non-linear kinematic hardening
Stress (MPa)

100

50

-50

-100

-150
-15 -10 -5 0 5
Distance above weld root (mm)
(a)
600

500

400

300
Stress (MPa)

200

100 ND measured stress


Contour method measurement
0
Mixed-rr
-100 Isotropic hardening

-200 Non-linear kinematic hardening


-15 -10 -5 0 5
Distance above weld root (mm)
(b)
Fig. 12. Comparison of measured and predicted through-wall stress distribution at mid-length of the single-pass groove weld specimen, showing the effect of moving from
isotropic through kinematic to mixed hardening material models. (a) Transverse stress (S11). (b) Longitudinal stress (S33).

fusion boundary, which is often a region of concern in structural imen, but only for the top surface of the bead 1 cross-weld tensile
integrity assessments. However, two-stage annealing in weld me- specimen. Because the latter test specimen was cut from near the
tal would lead to an unrealistic increase in weld metal residual bottom of the weld groove, the lateral extent of weld metal on the
stress unless combined with other changes to the material model, lower surface was small, and therefore the upper surface was
such as a lower unhardened yield strength. judged most appropriate for comparison. The isotropic predictions
are derived from the full cross-weld test FE simulation described
7.5. Groove weld stress–strain properties earlier. The mixed hardening predictions are made by extracting
the plastic path length and backstress tensor components from
The residual stress comparisons imply that the mixed harden- the relevant line in a weld simulation model and substituting these
ing models examined produce a reasonably accurate prediction into Eqs. (4) and (5) to calculate the stress after a further 1% plastic
of parent material hardening but a less satisfactory over-prediction strain. This simplified method does not require a full cross-weld
of weld metal properties. The performance of the hardening mod- test simulation, and has been shown to give equivalent results [32].
els can be assessed more directly through comparisons with mea- The measured 1% proof stress for weld bead 3 rises smoothly as
sured stress–strain properties from the cross-welds [23], see the fusion boundary is approached, and then remains relatively
Section 4.2. constant across the weld bead. The isotropic model clearly over-
Predicted variations in 1% proof stress across bead 1 and bead 3 predicts the yield strength in both weld and parent material. In
of the three-pass groove-weld specimen are compared with the contrast, the mixed hardening model closely predicts the proof
measurements in Fig. 15. The predictions are shown for both the stress variation at distances >7 mm from the weld centre-line. In
top and the bottom surfaces of the bead 3 cross-weld tensile spec- the HAZ close to the fusion boundary, the two-stage annealing
M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328 325

300
ND measured stress
250
Mixed-rr
200 Mixed-n

150 Mixed-mm

Stress (MPa)
100

50

-50

-100

-150
-15 -10 -5 0 5
Distance above weld root (mm)
(a)
600

500

400
Stress (MPa)

300

200

100 ND measured stress


Contour method measurement
0
Mixed-rr
-100 Mixed-n
Mixed-mm
-200
-15 -10 -5 0 5
Distance above weld root (mm)
(b)
Fig. 13. Comparison of measured and predicted through-wall stress distribution at mid-length of single-pass groove weld specimen, showing the effect of different mixed
hardening material model parameter fitting strategies. (a) Transverse stress (S11). (b) Longitudinal stress (S33).

model correctly predicts a smooth variation in 1% proof stress, al- model using both annealing regimes. But the FE model for this case
beit rising to a higher level than measured, while conventional does not represent the full width of weld metal present nor does it
annealing at 1050 °C results in a sharp peak followed by an unre- represent the change in weld metal width through the thickness of
alistic trough. The 1% proof stress of the final weld bead is over- the test specimen. Thus the discontinuity and apparent under-pre-
predicted by 50 MPa. diction of yield strength near the idealised weld fusion boundary
The measured and predicted 1% yield strengths in the bead 1 may not be real.
specimen are compared in Fig. 15b. The measured 1% proof stress
rises steadily through parent and HAZ material to a peak of approx- 8. Discussion
imately 450 MPa in the weld bead. It is seen that the isotropic
model over-predicts the yield stress in the parent and HAZ materi- If advanced FE based predictions of weld residual stress are to
als, but marginally under-predicts yield of the weld metal (note be used in structural integrity assessments of safety critical com-
that the weld was treated as perfectly plastic in the analysis with ponents, they must be underwritten by reliable measurements
a yield strength of 446 MPa). In contrast the mixed hardening par- on representative welded components (see Section III.15 of [5]).
ent model closely predicts the 1% proof stress at distances >7 mm In this study, different types of high quality measurement data
from the weld bead centre-line. The mixed hardening weld model from advanced experimental techniques have been collected from
predicts a slightly higher 1% proof stress than the isotropic model purpose-designed benchmark specimens and used to develop an
which is very close to the measured 1% proof stress. In the HAZ optimised strategy for predicting accurate distributions of residual
close to the weld, that is in material which has experienced tran- stress in austenitic stainless steel weldments.
sient welding temperatures exceeding 800–850 °C, the yield The comparisons of results clearly show that isotropic hardening
strength appears to be underpredicted by the mixed hardening models over-predict the magnitude of weld residual stresses. In
326 M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328

500
conventional annealing
450
two-stage annealing

400

Stress (MPa)
350

300

250

200

150

100
-15 -10 -5 0 5 10

Distance above weld root (mm)


(a)
0.10
conventional annealing
0.09
Isotropic hardening equivalent strain

two-stage annealing
0.08

0.07

0.06

0.05

0.04

0.03

0.02

0.01

0.00
-15 -10 -5 0 5 10
Distance above weld root (mm)
(b)
Fig. 14. Showing the impact of material annealing assumptions on predicted through-wall stress and plastic path distributions at mid-length of three-pass groove weld
specimen, for mixed hardening material behaviour. (a) Predicted longitudinal stress. (b) Isotropic hardening plastic path length.

contrast, pure kinematic hardening tends to under-predict longitu- pure kinematic hardening. If the constitutive model were capable
dinal stresses in parent material close to the weld. The mixed hard- of correctly modelling both the monotonic and cyclic responses,
ening models perform better in terms of predicting more accurate then this would clearly be the best outcome.
residual stresses. Arguably the best residual stress predictions were The mixed hardening parent models also give a more accurate
achieved for the groove-welds using the fit to the saturated cyclic prediction of the yield strength variation spanning cross-weld
response implemented in the mixed-mm analysis. But this produced specimens for material that has not experienced welding transient
an under-prediction of longitudinal stresses, caused by its poor rep- temperatures exceeding 800–850 °C. Above these temperatures,
resentation of the parent material monotonic stress–strain re- the method for handling high temperature inelastic strains impacts
sponse. Similar under-prediction has been observed by the the accuracy of predicted yield strength and local residual stresses.
authors in other multi-pass weld validation studies (not yet pub- Both single and two-stage ABAQUS annealing regimes were found
lished) and in the VORSAC 65 mm thick multi-pass weld [13] which to give perturbations in predicted longitudinal stresses confined to
used a similar model for the parent material. The multiple Ci–ci fit a small annulus of material adjacent to the fusion boundary,
to the monotonic tensile stress–strain response implemented in although the latter helped to smooth the variation in material
the mixed-rr analyses also gave accurate stress predictions but re- hardening approaching the fusion boundary in the final weld pass
sulted in some under-prediction of transverse stress in parent cross-weld specimen. Both the cross-weld studies and the pre-
material in the three-pass specimen. The simpler single C–c fit dicted stress profiles show that new constitutive models are re-
implemented in the mixed-n analyses performed less well, as would quired to represent hardening/softening behaviour of weld metal
be expected from its representation of the isothermal uniaxial re- and the adjacent heat affected zone under high temperature cyclic
sponse, but is still a marked improvement over pure isotropic or loading conditions associated with welding.
M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328 327

500

450

400

Stress (MPa)
350

300
ESPI -1.0%
mixed hardening, two-stage annealing, bottom
250
mixed hardening, two-stage annealing, top
isotropic hardening
mixed hardening, conventional annealing, top
200
mixed hardening, conventional annealing, bottom

150
-20 -15 -10 -5 0 5 10 15 20
Distance from gauge centreline (mm)
(a)
500

450

400
Proof stress (MPa)

350

300

250 ESPI 1%
isotropic hardening
estimated weld yield strength
200 mixed hardening, two-stage annealing
mixed hardening, conventional annealing

150
-20 -15 -10 -5 0 5 10 15 20
Distance from gauge centreline (mm)
(b)
Fig. 15. Comparison of measured and predicted cross-weld variation of 1% proof stress in a three-pass groove weld specimen, showing the effect of material hardening model
and high temperature annealing assumptions. (a) Weld bead 3 (last pass), cross-weld specimen. (b) Weld bead 1 (first pass), cross weld specimen.

The mixed hardening weld models over-predict the longitudi- phase of the evolutionary cyclic response all appear to be appro-
nal component of residual stress in the single-pass weld by a large priate. However, it must be remembered that the resulting hard-
margin and in the three-pass weld by 50–100 MPa. The cross- ening models are optimised for predicting weld residual stresses,
weld specimens show that the yield strength of single-pass weld and may not be suitable for predicting the response during
metal is over-predicted by 50 MPa, while the almost fully hard- subsequent service loading if the characteristics of that loading
ened bead 1 weld metal is accurately predicted. Together these differ from welding. A possible example is low cycle fatigue,
data suggest that the initial (as-deposited) weld material proper- where it might be appropriate to model cyclic softening from
ties, based on spike-annealed single-pass weld metal, have too the cyclically saturated condition, rather than initial evolutionary
high a yield strength, while properties for the fully hardened state, hardening.
based on the 0.2% proof stress of multi-pass weld metal, are about
right. As the yield strength of fully solution-treated weld metal is 9. Conclusions
about 40 MPa lower than the spike annealed single-pass weld me-
tal used, it is probably more appropriate to base the initial weld 1. An optimised strategy for predicting accurate distributions of
properties on solution treated weld material, provided that it still residual stress in stainless steel weldments has been developed
cyclically hardens to the yield strength of multi-pass weld metal. using multiple measurements from purpose-designed welded
Notwithstanding the caveats above, the approach adopted for benchmark specimens.
materials characterisation testing and model fitting appears to 2. The scope of experimental work performed is an exemplar for
be vindicated by the measured stresses and hardening in the future optimisation studies of this kind. In particular, measure-
benchmark specimens. Thus the choices of strain rate and cyclic ments of spatially varying stress–strain properties across the
strain range, and the decision to fit only the initial hardening weldment have been used to assess the adequacy of material
328 M.C. Smith et al. / Computational Materials Science 54 (2012) 312–328

hardening-softening models used in the FE analysis, and mea- [4] J. Zhou, H.L. Tsai, in: Z. Feng (Ed.), Welding heat transfer, processes and
mechanisms of welding residual stress and distortion, Woodhead Publishing in
surements of residual stress from diverse techniques have been
Materials, 2005, pp. 32–98.
used to evaluate the overall accuracy of the computational pre- [5] British Energy Generation Ltd., Gloucester, UK, 2010.
dictive procedures. [6] C.E. Truman, M.C. Smith, International Journal of Pressure Vessels and Piping
3. It has been demonstrated that isotropic hardening models over- 86 (2009) 1–2.
[7] M.C. Smith, A.C. Smith, International Journal of Pressure Vessels and Piping 86
predict the tensile magnitude of welding residual stress in this (2009) 96–109.
type of structure. In contrast, pure kinematic hardening results [8] M.C. Smith, A.C. Smith, International Journal of Pressure Vessels and Piping 86
in under-prediction of longitudinal stresses in parent material (2009) 79–95.
[9] P.J. Bouchard, International Journal of Pressure Vessels and Piping 86 (2009)
close to the weld. 31–42.
4. Mixed isotropic–kinematic formulations combined with a [10] M.C. Smith, A.C. Smith, R.C. Wimpory, C. Ohms, B. Nadri, P.J. Bouchard, in:
multi-pass moving heat source produce the most accurate pre- ASME PVP2009-77157, American Sociery of Mechanical Engineers, Prague,
2009.
dictions of residual stress. The model of choice is one where the [11] M.C. Smith, B. Nadri, A.C. Smith, D.G. Carr, P.J. Bendeich, L. Edwards, in:
kinematic parameters (which define the shape of the cyclic ASME PVP 2009-77158, American Society of Mechanical Engineers, Prague,
hardening curve) are fitted to the monotonic uniaxial stress– 2009.
[12] L.-E. Lindgren, Journal of Thermal Stresses 24 (2001) 195–231.
strain response as a function of temperature. [13] S.K. Bate, P.J. Bouchard, in: ICRS6, Oxford, UK, 2000, pp. 1511–1518.
5. The initial stress–strain properties of deposited weld metal can [14] K. Ogawa, L.O. Chidwick, E.J. Kingston, R.J. Dennis, D. Bray, N. Yanagida, in:
be based on solution heat treated single pass weld material, ASME PVP2008-61452, American Society of Mechanical Engineers, Chicago,
2008.
provided that it cyclically hardens to the yield strength of
[15] J. Mullins, J. Gunnars, Materials Science Forum 681 (2011) 61–66.
multi-pass weld metal. [16] L.-E. Lindgren, K. Domkin, S. Hansson, Mechanics of Materials 40 (2008) 907–
6. A new two-stage annealing procedure for dealing with plastic 919.
strains accumulating at high temperature in the weld simula- [17] ABAQUS/Standard User’s Manual, Version 6.8, Simulia, Providence, RI, 2008.
[18] Lehrstuhl und Institut für Werkstoffkunde, Universität Karlsruhe, 1999.
tion has been shown to reduce discontinuities in results adja- [19] H. Keinänen, in: VTT Manufacturing Technology, 1999.
cent to the fusion boundary compared with single-stage [20] British Standards Institute, 1990.
annealing, but neither approach is very satisfactory. [21] British Standards Institute, 1992.
[22] British Standards Institute, 2001.
7. Improved constitutive models for stainless steel parent and [23] M. Turski, M.C. Smith, P.J. Bouchard, L. Edwards, P.J. Withers, Journal of
weld metal are needed to represent the monotonic and cyclic Pressure Vessel Technology 131 (2009).
stress–strain response more accurately, and to represent the [24] M. Kartal, M. Turski, G. Johnson, M.E. Fitzpatrick, S. Gungor, P.J. Withers, L.E.
Edwards, Materials Science Forum 524–525 (2006) 6716.
softening behaviour of weld metal and the adjacent heat [25] M. Turski, L. Edwards, J. James, P.J. Bouchard, M.C. Smith, P.J. Withers, in: ASME
affected zone under high temperature cyclic loading conditions. PVP2006-ICPVT-11-93991, American Society of Mechanical Engineers, New
York, Vancouver, Canada, 2006.
[26] J.A. Dann, M.R. Daymond, L.E. Edwards, J.A. James, J.R. Santisteban, Physica B:
Acknowledgements Condensed Matter 350 (2004) E511–E514.
[27] J.A. James, J.R. Santisteban, L. Edwards, M.R. Daymond, Physica B 350 (1–3)
This paper is published with permission from EDF Energy Nu- (2004) 743–746.
[28] P.J. Bouchard, M. Turski, M.C. Smith, in: ASME PVP2009-77234, American
clear Generation Limited. Professor Bouchard gratefully acknowl-
Society of Mechanical Engineers, New York, Prague, Czech Republic, 2009.
edges Royal Society Industry Fellowship support and Dr Mike [29] M.B. Prime, Journal of Engineering Materials and Technology 123 (2001) 162–
Smith acknowledges a Royal Academy of Engineering Global Re- 168.
search Award. [30] H.F. Bueckner, Transactions of the American Society of Mechanical Engineers
80 (1958) 1225–1230.
[31] J. Lemaitre, J.L. Chaboche, Mechanics of Solid Materials, Cambridge University
References Press, 1990.
[32] M.C. Smith, in: British Energy, 2006.
[1] J. Mackerle, Modelling and Simulation in Materials Science and Engineering 4 [33] R.M. Smith, in: FEATPLUS Limited, Bristol, 2009.
(5) (1996) 501–533. [34] R.J. Dennis, N.A. Leggatt, A. Gregg, in: ASME PVP2006-ICPVT11-93907,
[2] J. Mackerle, Modelling and Simulation in Materials Science and Engineering 10 American Society of Mechanical Engineers, Vancouver, BC, Canada, 2006.
(3) (2002) 295–318. [35] ABAQUS, 2007.
[3] L.-E. Lindgren, Computational welding mechanics. Thermomechanical and [36] P.J. Bendeich, M.C. Smith, D.G. Carr, L.E. Edwards, in: ASME PVP2009-77584,
microstructural simulations, Woodhead Publishing Ltd., Cambridge, 2007. American Society of Mechanical Engineers, Prague, 2009.

Vous aimerez peut-être aussi