International Journal of Solids and Structures 69–70 (2015) 518–530

Contents lists available at ScienceDirect

International Journal of Solids and Structures

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

The influence of constitutive material models on accumulated plastic

strain in finite element weld analyses
O. Muránsky a,⇑, C.J. Hamelin a, V.I. Patel a,b, V. Luzin c, C. Braham d
a
ANSTO, Institute of Materials Engineering, New Illawarra Road, Lucas Heights, NSW, Australia
b
UNSW, School of Civil and Environmental Engineering, Faculty of Engineering, Sydney, NSW, Australia
c
ANSTO, Bragg Institute, New Illawarra Road, Lucas Heights, NSW, Australia
d
Laboratoire de Procédés et Ingénierie en Mécanique et Matériaux (PIMM-UMR CNRS 8006), ENSAM, 151, Boulevard de l’Hôpital, 75013 Paris, France

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

Article history: Recent studies in computational weld mechanics have revealed the importance of the material plasticity
Received 16 December 2014 model when predicting weld residual stresses. The present work seeks to extend this level of understand-
Received in revised form 17 March 2015 ing to include the effects of the assumed material annealing behaviour, particularly when modelling
Available online 8 May 2015
multi-pass welds that comprise several thermo-mechanical loading cycles. A series of numerical analyses
are performed to examine the variability in predicted residual stress profiles for different material mod-
Keywords: els, using a validated finite element model for a three-pass slot weld in AISI 316LN austenitic steel. The
Residual stresses
material models consider both the work hardening and annealing assumptions for the chosen material.
Plastic strain
Plasticity theory
Model sensitivity is established not only from a weld residual stress perspective, but also from an assess-
Annealing ment of the post-weld plastic strain accumulated in the weldment. Predictions are compared with indi-
Finite element analysis rect measurements acquired using cross-weld micro-hardness maps taken from benchmark specimens.
Sensitivity studies reveal that the choice of annealing behaviour will have a significant impact on plastic
flow predictions, which is dependent on the annealing temperature specified. Annealing assumptions will
have a varying impact on the weld residual stress predictions, such that the extent of sensitivity is depen-
dent on the plasticity model chosen. In contrast, the choice of plasticity model will have a significant
effect on the predicted weld residual stresses, but relatively little effect on predictions of equivalent plas-
tic strain.
Crown Copyright Ó 2015 Published by Elsevier Ltd. All rights reserved.

1. Introduction In recent years, considerable effort has been dedicated to the

standardisation of WRS measurement (Withers, 2001a,b;
Weld residual stresses (WRS) arise as a result of a permanent Hutchings et al., 2005) and prediction in austenitic steel welds
shape misfit between plastically deformed material near the weld (Muránsky et al., 2012a,b; Smith et al., 2012a). Residual stresses
region and the rest of the parent structure. In welding, this misfit is in representative benchmark weld specimens have been measured
caused by a severe thermal gradient that, due to localised thermal and predicted using a variety of techniques. A round-robin
expansion of the heat-affected material, generates plastic deforma- investigation is often used to ensure the highest confidence in the
tion in the vicinity of the weld pool (Withers, 2001a). It is of tech- accuracy of each technique; one example of such collaborative
nological importance to know the magnitude and direction of effort is the research presented through Task Group 4 (TG4) of the
these WRS, as they can superimpose on operational stresses and European Network on Neutron Techniques Standardisation for
thus contribute to premature failure of the welded component or Structural Integrity (NeT) (Muránsky et al., 2012a). Through NeT
structure (Withers, 2001a, 2007). It has also been shown that TG4, a set of AISI 316LN three-pass slot weld benchmark specimens
WRS can provide the driving force for crack initiation and growth have been produced for WRS measurement and finite element (FE)
(Muránsky et al., 2014), thereby affecting the service lifetime of a analysis. The TG4 weld specimens are of particular interest to the
welded structure. nuclear industry, where multi-pass welding processes are often
used for austenitic steel weldments and WRS may adversely affect
the structural integrity of these welded components.
⇑ Corresponding author. Tel.: +61 2 9717 3488; fax: +61 2 9543 7179. Both diffraction (neutron and synchrotron X-ray) (Muránsky
E-mail address: [email protected] (O. Muránsky). et al., 2012a; Martins et al., 2010) and stress relaxation

http://dx.doi.org/10.1016/j.ijsolstr.2015.04.032
0020-7683/Crown Copyright Ó 2015 Published by Elsevier Ltd. All rights reserved.
 O. Muránsky et al. / International Journal of Solids and Structures 69–70 (2015) 518–530 519

(deep-hole drilling and contour cutting) techniques have been three-pass ER316L austenitic steel slot weld in solution
employed for WRS measurement under the NeT TG4 program. heat-treated (45 min @ 1050 °C) AISI 316LN austenitic steel plate.
These independent WRS measurements were then used in the val- The nominal dimensions of the plate, shown in Fig. 1, are
idation of FE analyses (Muránsky et al., 2012a,b; Dewees et al., 194(l)  150(w)  18(h) mm with an 80-mm long and 6-mm deep
2014). It has been shown that welding simulations developed by centreline slot. The slot was filled with three superimposed weld
ANSTO and EDF Energy in Muránsky et al. (2012a,b) accurately pre- passes via a mechanised Tungsten Inert Gas (TIG) welding process.
dict WRS in multi-pass 316LN welds, provided the key simulation The specimens were welded free of constraint, allowing the plates
variables (i.e. material properties and welding parameters) are to distort without inhibition. Three of the benchmark specimens
known. Sensitivity analyses have revealed that the choice of consti- were instrumented with an array of thermocouples (Muránsky
tutive material response can significantly influence the predicted et al., 2012a) to assure welding repeatability and provide transient
WRS distribution (Muránsky et al., 2012b). Specifically, it was thermal data for the heat source calibration in numerical analyses.
determined that adopting a mixed (isotropic-kinematic) cyclic Full details of the welding procedure can be found in Muránsky
plasticity model provides the most accurate prediction of WRS, et al. (2012a).
while using a more conventional (and relatively easy-to-calibrate)
isotropic plasticity model provides conservative1 WRS estimates.
Importantly, it was noted in Muránsky et al. (2012b) that different 3. Numerical weld analysis
constitutive plasticity theories intrinsically predict different levels
of accumulated post-weld plastic strain. Improper characterisation A sequentially coupled thermo-mechanical FE analysis was per-
of post-weld plastic strain might have serious consequences if not formed, whereby the numerical solution from a thermal FE analy-
taken into account because this strain may adversely affect the failure sis is used as input for a mechanical FE analysis. The thermal model
characteristics of welded structures. This level of model accuracy is used a representative heat source calibrated using the FEAT-WMT
often overlooked since the validation of numerical weld analyses is modelling package (Smith, 2010), which was supplied to the
traditionally performed via comparison of predicted and measured ABAQUS 6.13 (SIMULIA, 2014) commercial FE package as a series
WRS, with no consideration of accumulated plasticity. of time- and spatially-resolved volumetric power densities.
In addition, recent advances in the measurement of accumu- Taking advantage of specimen symmetry (see Fig. 1), a 3D
lated plastic strain present an opportunity to re-examine the accu- half-model comprising 38,220 hexahedral quadratic elements
racy of welding simulations based on this characteristic. Model was constructed. Although the parent (AISI 316LN) and filler
validation based on an alternate weld feature allows the analyst (ER316L) materials have a slightly different chemical composition
a more comprehensive assessment of the key simulation variables (Muránsky et al., 2012a), the same physical2 and elasto-plastic3
used in computational weld mechanics. In the present work, mechanical properties were used for both materials over the tem-
post-weld plastic strain in the NeT TG4 benchmark weld speci- perature range of interest (25–1500 °C). While a significant volume
mens is measured using a novel technique, whereby equivalent fraction of metastable d-ferrite is likely to form at high temperatures
plastic strain is indirectly characterised using micro-hardness upon welding, most of this material transforms back to austenite at a
maps. Hence, both post-weld plastic strain and associated temperature close to the steel melting point. Since the temperatures
post-weld residual stresses are compared with experimental over which these solid-state phase transformations occur are well
results to establish the accuracy of current weld modelling proce- above the annealing temperature of the material, they will have a
dures, providing best-practise recommendations to FE analysts. negligible effect on accumulated WRS (Muránsky et al., 2012a;
In the present work, the sensitivity of plastic strain and WRS Martins et al., 2010) and plastic strain, and are thus not considered
predictions is examined when key parameters in the constitutive in the present numerical analyses.
material model are varied. First, the sensitivity of the numerical Within the constitutive material model used, two key simula-
solution is examined when isotropic, kinematic and tion variables have been identified that will significantly affect
isotropic-kinematic (mixed) plasticity theories are adopted under numerical predictions: (i) material plasticity theory (Muránsky
identical high-temperature annealing conditions. Then, model sen- et al., 2012b), which describes temperature-dependent material
sitivity to the assumed high-temperature material annealing beha- yield and hardening behaviour; and (ii) material annealing beha-
viour is examined. Model predictions are compared to viour, which describes the loss of accumulated plastic strain and
measurements of both WRS (previously documented in thus high-temperature metal softening.4 The following sections dis-
Muránsky et al. (2012a,b)) and accumulated plastic strain in the cuss each simulation variable in greater detail.
NeT TG4 benchmark specimens. Since post-weld plastic strain
measurements are inferred from calibrated micro-hardness tests, 3.1. Material plasticity theories
the accuracy of using this novel hardness-based approach for
quantifying plastic strain is also assessed in the present study. If The most commonly employed constitutive plasticity theories
successful, such an approach would provide complementary model used in computational weld mechanics are isotropic plasticity the-
validation alongside the more traditional WRS assessments. ory, kinematic plasticity theory and isotropic-kinematic (i.e.
mixed) plasticity theory (Muránsky et al., 2012b). These theories
differ in the way they capture the response of the material to cyclic
2. Weld specimen loading, in both the material yield and work-hardening behaviour.
Variations in how each theory simulates cyclic loading is shown
A series of benchmark weld specimens were manufactured schematically in Fig. 2. The characterisation of material yield and
under the auspices of NeT TG4 (Muránsky et al., 2012a; Martins, work hardening behaviour under reverse loading has a profound
2009). Multiple specimens were produced using an identical weld effect on the prediction of WRS and plastic strain. This effect is par-
procedure to allow simultaneous WRS measurement as part of an ticularly true for multi-pass welds, since the material in the
international round-robin examination. Weld design comprises a
2
Density (q), thermal conductivity (k), and specific heat (cp).
1 3
When peak tensile WRS are over-predicted using an isotropic hardening model, Young’s modulus (E), Poisson’s ratio (m), and work-hardening behaviour.
4
these stresses may subsequently create large compressive equilibrating stresses, Softening refers to a reduction of the work hardening rate and/or the yield
which are typically non-conservative from a structural integrity standpoint. strength of the material.
520 O. Muránsky et al. / International Journal of Solids and Structures 69–70 (2015) 518–530

It is important to highlight the functionality of Eqs. (1) and (2)

in this work, since the distinction must be made between (i) mate-
rial annealing behaviour controlled by ABAQUS using epl (i.e.
ABAQUS variable PEEQ), and (ii) softening behaviour dictated by
material constants (Q inf and C i ). For example, isotropic softening
is enforced by setting Q inf to 0 at temperatures above 900 °C based
on experimental tests (Smith et al., 2009; Smith, 2006). While this
setting forces the yield surface back to its original size rj0 , the
accumulated plastic flow in the material is maintained; hence,
prior work hardening is kept and the material has not annealed.
A similar effect can be observed for kinematic hardening; C i has
Fig. 1. NeT TG4 weld geometry, showing the symmetry plane for the finite element
been set to 0 at temperatures above 1000 °C (Smith et al., 2009;
simulations, the location of the extracted cross-weld specimen for hardness
measurements, and the line profiles (BD, B2, B16) along which the post-weld Smith, 2006), such that only the nonlinear relaxation term in Eq.
residual stresses and post-weld plastic strain were measured. (2) is active as plastic strain continues to accumulate. As discussed
below, specification of an annealing temperature in ABAQUS will
set epl (and consequently, e_ pl ) to 0 such that all accumulated plastic
vicinity of weld undergoes multiple thermo-mechanical loading
strain is lost once this temperature is exceeded. This annealing will
cycles during the welding process.
cause a similar isotropic softening via Eq. (1), while no kinematic
Isotropic plasticity theory allows the uniform dilation of the
relaxation will occur due to Eq. (2). This distinction is important
yield surface about its centre while maintaining its shape, orienta-
since we are concerned in this work with annealing phenomena,
tion and centroid position in stress space. The equivalent stress
and not material softening due to the lower activation energy
defining the size of the yield surface ðr0 Þ can be defined as follows: required for plastic flow at elevated temperatures (governed by
 pl
 Q inf and C i ).
r0 ¼ rj0 þ Q inf 1  ebe ; ð1Þ

where rj0 is the equivalent yield stress when the equivalent plastic
  3.2. Material annealing behaviour
strain epl is 0; and Q inf and b are temperature-dependent material
parameters. Kinematic hardening allows the yield surface to trans-
late in stress space while maintaining its shape, size and orienta- High-temperature annealing is an important phenomenon that
tion. It is defined by an additive combination of a linear needs to be accounted for in any welding simulation. The anneal-
hardening term and a nonlinear relaxation term, such that ing behaviour of a material is typically defined by setting a temper-
ature above which the material loses all accumulated plastic strain.
X 1

This loss of plasticity naturally leads to the relaxation of the asso-
a_ ¼ Ci ð r  aÞ _ pl  c ae_ pl ;
e ð2Þ
i
i
r
0 ciated stresses in the annealed region when isotropic hardening is
considered (Eq. (1)), as the material is no longer in a
where a_ represents the hardening rate of the backstress tensor a; r work-hardened state. For kinematic models (Eq. (2)), stresses will
is the applied stress tensor; and C i and ci are material constants. remain constant since no additional plastic strain is accumulated
When mixed hardening is considered, the size of the yield surface during annealing (e_ pl ¼ 0). In the present study, two different
due to isotropic hardening (r0 , Eq. (1)) is used instead of rj0 when assumptions on how to capture annealing behaviour during weld-
defining the kinematic hardening rate in Eq. (2). ing are examined:

Fig. 2. The effect of isotropic, kinematic and isotropic-kinematic (mixed) plasticity theory on predicting initial yield and cyclic yield. (i) Isotropic theory allows the expansion
of the initial yield surface, (ii) kinematic theory allows translation of the initial yield surface, and (iii) mixed theory allows both expansion and translation of the initial yield
surface. A = initial yield; B = post-yield work hardening; C, D, E = yield on reverse loading, as predicted by kinematic, mixed and isotropic plasticity theory respectively.
 O. Muránsky et al. / International Journal of Solids and Structures 69–70 (2015) 518–530 521

Fig. 3. The microstructural differences between weld and parent metal, identified using several techniques on a NeT TG4 specimen. (a) Optical macrograph highlighting the
variation in grain dimensions between the weld and parent metal. (b) Neutron diffraction patterns obtained using the ECHIDNA high-resolution powder diffractometer at
ANSTO, showing diffraction peaks corresponding to fcc c-phase (austenite) and bcc d-phase (ferrite) in the weld region. This diffraction data reveals the existence of d-ferrite
in the weld region, while no d-ferrite is present in the parent metal. (c) Bulk texture measurement, comprising {2 0 0} pole figures of weld and parent metal, obtained using the
KOWARI strain diffractometer at ANSTO. The texture shows preferential solidification texture formed in the weld metal, while parent metal texture is fairly random.

1. A single-stage annealing assumption, with a single annealing ECHIDNA high-resolution diffractometer at ANSTO (Liss et al.,
temperature (T0). In this case, any temperature rise above T0 1010), revealed approximately 5% (by volume) d-ferrite exists in
causes the accumulated plastic strain (represented in ABAQUS the weld metal that is not present in the parent metal. This mea-
using the equivalent plastic strain, PEEQ5) at a material integra- surement agrees with phase predictions via the Schaeffler diagram
tion point to be set to zero, thus eliminating all plastic history. (Shchaeffler, 1949), which take into consideration the higher Cr con-
2. A two-stage isotropic annealing approach, whereby two anneal- tent in the weld filler metal. The higher Cr content in the weld metal
ing temperatures (T1, T2) capture the progressive nature of (19 wt%) over the parent metal (17.5 wt%) also suggests an
annealing processes more accurately. At temperatures between increased hardenability of the weld (Muránsky et al., 2012a).
the lower annealing temperature (T1) and the upper annealing Finally, bulk texture measurements presented in Fig. 3(c) show that
temperature (T2), the material is prevented from any further the textures measured for weld and parent metals are markedly dif-
hardening by holding the accumulated plastic strain constant. ferent. Weld metal texture is considerably stronger, with the {1 0 0}
At temperatures above T2, the material loses all accumulated plane-normals oriented towards the sample normal direction
plastic strain (i.e. PEEQ = 0) in a similar manner to (Y(22), Fig. 1) (i.e. the direction of the heat flow). In contrast, the par-
single-stage annealing. Full details of two-stage annealing func- ent metal texture is fairly random. These microstructural variations
tionality within ABAQUS/Standard are given in ABAQUS (2007). are expected to affect the mechanical properties of each material,
thus influencing experimental measurements of both post-weld
plastic strain and WRS. Significant work has been performed in an
4. Experimental characterisation
attempt to properly capture the constitutive weld metal behaviour;
the best results to date have been achieved using solution-annealed
The cross-weld macrograph in Fig. 3(a) shows that the weld
weld metal tensile specimens (Smith et al., 2012b). In the present
metal is comprised of relatively large dendritic grains, which is very
work, cross-weld microstructural variations have not been explic-
different from the smaller equiaxed grains present in the parent
itly considered in experimental measurements, which may lead to
metal. Additional phase analysis, shown in Fig. 3(b) using the
some deviation in measured data that has yet to be quantified. 6
5 p t
PEEQ = e_ pl j0 ½ð 2=3e_ pl : e_ pl Þdt , where e_ pl j0 is initial equivalent plastic strain and
_epl is the plastic strain tensor at time t. In the present work, the initial equivalent 6
Sensitivity analyses would also need to consider the accuracy of the fitting
plastic strain is set to zero because the plate was solution heat-treated prior to technique used on cyclic loading test data, which may considerably influence weld
welding. model accuracy. Such work is outside the scope of this study.
522 O. Muránsky et al. / International Journal of Solids and Structures 69–70 (2015) 518–530

4.1. Characterisation of plastic strain via micro-hardness post-weld plastic strain as presented in Fig. 5(b). This indirect mea-
measurement surement of plastic strain was then used to validate FE predictions.

Plastic deformation in the vicinity of the weld was characterised 4.2. Measurement of post-weld residual stresses
using cross-weld micro-hardness measurements. A Nano Indenter
G300 (Keysight) with a Berkovich tip was used, and measurements A number of residual stress measurement techniques were
were taken under Continuous Stiffness Mode (CSM). The hardness employed in the NeT TG4 round-robin project; neutron diffraction
measurements in Fig. 4(a) converge asymptotically to a true hard- and spiral-slit synchrotron diffraction measurements are pre-
ness value of 200 Hv at imprint depths greater than 2.0 lm; these sented here. Neutron diffraction residual stress measurements
indent depths are proportional to the applied indent load for mea- were carried out using the KOWARI diffractometer at ANSTO
surement. Fig. 4(b) shows typical load–displacement curves for the (Muránsky et al., 2012a; Brule et al., 1040), while synchrotron
316LN parent metal using the 30-gf indent load specified; the diffraction residual stress measurements were carried out using
indent depth (2.35 lm) is sufficient to overcome the size effects ID-15 at ESRF (Martins et al., 2010). It is important to note that
illustrated in Fig. 4(a), thus this load is used for generation of while the neutron diffraction technique can access strain compo-
micro-hardness map data. nents in all three principal strain directions, the synchrotron
The 2(L)  50(T)  18(N)-mm sample used for micro-hardness diffraction technique can access only two principal strain compo-
measurements was extracted via electro-discharge machining nents; the calculation of principal stresses from strain data must
(EDM) from the weld mid-length, as shown schematically in be done assuming plane stress conditions exist through the sample
Fig. 1. The measured hardness map of this sample is presented in thickness. The validity of such an assumption for the current sam-
Fig. 5(a). A motorised X–Y stage mounted on the indenter gave a ple geometry has been previously confirmed in Muránsky et al.
precise spatial resolution of 0.5 mm (i.e. DX, DY = 0.5 mm). To con- (2012a) and Martins et al. (2010).
vert the measured micro-hardness to calculated plastic strain
(PEEQ), a relationship between micro-hardness and plasticity in 5. Results and discussion
AISI 316L austenitic steel has been established using a novel tensile
specimen (shown in Fig. 6) and the method outlined in Ben Moussa Figs. 7 and 8 present contour plots of the predicted longitudinal
(2013). The amount of plastic deformation along the sample tensile (r33 ) post-weld residual stresses (WRS) and post-weld plastic
axis was measured using Digital Image Correlation (DIC). A plot of strain (via PEEQ) respectively, in the vicinity of the weld at the
the variation in plastic strain along the sample tensile axis is pre- weldment mid-length (see Fig. 1) for the following model
sented in Fig. 6(a). The corresponding micro-hardness was charac- assumptions:
terised using the same hardness machine, indenter and load as was
used for weld characterisation in the present work; this profile is (1) Isotropic plasticity theory, single-stage annealing,
shown in Fig. 6(b). The consequent relationship between the hard- T0 = 1050 °C,
ness and plastic deformation along the tensile axis of the specimen (2) Isotropic plasticity theory, single-stage annealing,
is presented in Fig. 6(c). Assuming hardness is directly proportional T0 = 1300 °C,
to yield strength under monotonic tensile loading, the relationship (3) Kinematic plasticity theory, single-stage annealing,
shown in Fig. 6(c) has been fit assuming the formulation of Eq. (1) T0 = 1050 °C,
holds. A good fit can be observed in Fig. 6(c). The relationship was (4) Isotropic-kinematic (mixed) plasticity theory, single-stage
then used to convert the measured post-weld micro-hardness into annealing, T0 = 1050 °C,

Fig. 4. (a) Continuous Stiffness Mode (CSM) hardness measurements showing the effect of depth of indent, which is proportional to the applied load, on the measured micro-
hardness. Size effects are overcome at indent depths greater than 2.0 lm. (b) CSM load–displacement data for a 30-gf indent load, revealing the indent depth (2.35 lm) is
sufficient for accurate hardness measurements.
 O. Muránsky et al. / International Journal of Solids and Structures 69–70 (2015) 518–530 523

Fig. 5. (a) Map of measured Vickers hardness obtained using micro-indentation, with a Berkovich tip and an indent load of 30 gf. The spatial resolution of is map is 0.5 mm.
(b) Post-weld plastic strain calculated using the established hardness-plasticity relationship shown in Fig. 6(c).

(5) Isotropic-kinematic (mixed) plasticity theory, single-stage assumptions cannot be seen using this profile (since
annealing, T0 = 1300 °C, it runs along the symmetry plane), it illustrates the
(6) Isotropic-kinematic (mixed) plasticity theory, two-stage effect of thermo-mechanical cyclic loading on the con-
annealing, T1 = 800 °C, T2 = 1300 °C. stitutive material response.
Line B2: Line B2 runs perpendicular to the weld at the
Only the longitudinal stresses are present here as they are of the mid-length of the plate, 2 mm below the top plate sur-
highest magnitude (Muránsky et al., 2012a,b; Martins et al., face. All of the isotherms represented along line BD are
2010) and are consequently the most sensitive to any model represented here; however, line B2 will only traverse
assumption. Because the model is symmetric along the weld cen- the weld metal deposited during the last weld pass
treline (Fig. 1), half-model results are present in Figs. 7 and 8 under (PASS.3 in Figs. 9 and 10). As such, data collected
the assumption that these results are reflected on the opposite side within the weld bead along this line reflects the mate-
of the specimen. To validate this assumption, all measurements rial response to monotonic loading on cooling
were taken on either side of the weld centreline, along a series of (Muránsky et al., 2012b). The cross-weld measure-
lines as presented in Figs. 9 and 10. Data from three line profiles ments along this profile will be able to confirm sym-
are presented in this work (with locations shown in Figs. 1, 7 metry assumptions.
and 8): Line B16: As with line B2, line B16 runs perpendicular to the
weld at the mid-length of the plate. This line runs
Line BD: Line BD runs through the plate-thickness at the centre 16 mm below the top plate surface; as such, it passes
of the TG4 benchmark specimen. This line crosses a solely through the parent metal. From this position,
number of important isotherms: (i) the 1400 °C iso- it is possible to examine the cross-weld material
therm, defining the fusion boundary; (ii) the 1050°C response when subjected to all three thermo-
and 1300 °C isotherms that define the single-stage mechanical loading cycles, allowing ready comparison
annealing temperatures assumed in this study; and to the monotonic loading behaviour of the weld metal
(iii) the 800 °C and 1300 °C isotherms that define the observed along line B2. Symmetry assumptions within
lower and upper annealing temperatures used under the parent metal may also be validated along this
a two-stage annealing assumption. While symmetry line.
524 O. Muránsky et al. / International Journal of Solids and Structures 69–70 (2015) 518–530

Fig. 6. Method of defining a relationship between plastic strain and micro-hardness in AISI 316L material, using a novel tensile specimen (shown centre of figure). (a) Plastic
strain measurements along the sample loading axis, obtained via Digital Image Correlation (see Ben Moussa, 2013 for details). (b) Hardness measurements along the sample
loading axis obtained using micro-indentation, with a Berkovich tip and an indent load of 30 gf. (c) Relationship between the plastic strain and hardness for AISI 316L.

Fig. 7. 2D maps of predicted longitudinal WRS in the steady-state region in the mid-length of the TG4 specimen. (a) Isotropic plasticity model with a single-stage annealing
assumption: T0 = 1050 °C. (b) Isotropic plasticity model with a single-stage annealing assumption: T0 = 1300 °C. (c) Kinematic plasticity model with a single-stage annealing
assumption: T0 = 1050 °C. (d) Mixed plasticity model with a single-stage annealing assumption: T0 = 1050 °C. (e) Mixed plasticity model with a single-stage annealing
assumption: T0 = 1300 °C. (f) Mixed plasticity model with a two-stage annealing assumption: T1 = 800 °C; T2 = 1300 °C.

Before assessing model sensitivity to key simulation variables, 1. Both the WRS and plastic strain measurements along lines B2
several features of both the measurements and the predictions pre- and B16 are symmetric along the symmetry plane, which vali-
sent in Figs. 9 and 10 must be discussed: dates the use of an FE half-model (Muránsky et al., 2012a).
 O. Muránsky et al. / International Journal of Solids and Structures 69–70 (2015) 518–530 525

Fig. 8. 2D maps of predicted post-weld strain (PEEQ) in the mid-length steady-state region of the TG4 specimen. (a) Isotropic plasticity model with a single-stage annealing
assumption: T0 = 1050 °C. (b) Isotropic plasticity model with a single-stage annealing assumption: T0 = 1300 °C. (c) Kinematic plasticity model with a single-stage annealing
assumption: T0 = 1050 °C. (d) Mixed plasticity model with a single-stage annealing assumption: T0 = 1050 °C. (e) Mixed plasticity model with a single-stage annealing
assumption: T0 = 1300 °C. (f) Mixed plasticity model with a two-stage annealing assumption: T1 = 800 °C; T2 = 1300 °C.

Fig. 9. Comparison of the predicted WRS and post-weld plastic strain along line BD (a,d), line B2 (b,e) and line B16 (c,f) obtained using isotropic, kinematic and mixed
plasticity theory with single-stage annealing assumption (T0 = 1050 °C). WRS predictions are shown alongside synchrotron (open symbol) and neutron (solid symbol)
diffraction measurements; the error associated with the experimental data is approximately 40 MPa. Plastic strain predictions are shown alongside indirect plasticity
calculations using micro-hardness measurements; the error associated with this data has not been confirmed.
526 O. Muránsky et al. / International Journal of Solids and Structures 69–70 (2015) 518–530

Fig. 10. Comparison of the predicted WRS and post-weld plastic strain along line BD (a,d), line B2 (b,e) and line B16 (c,f). Predictions are obtained using mixed plasticity
theory with single-stage annealing assumptions (T0 = 1050 °C, T0 = 1300 °C) and a two-stage annealing assumption (T1 = 800 °C, T2 = 1300 °C). WRS predictions are presented
alongside synchrotron (open symbol) and neutron (solid symbol) diffraction measurements; the error associated with the experimental data is approximately 40 MPa. Plastic
strain predictions are shown alongside indirect plasticity calculations using micro-hardness measurements; the error associated with this data has not been confirmed.

2. The WRS profiles measured using both neutron and syn- Post-weld plastic strain predictions (via PEEQ) are also in general
chrotron diffraction techniques are in good agreement, provid- agreement with measured data (Figs. 9 and 10). As with the WRS
ing a high degree of confidence in the measured residual profiles, there is a discrepancy between measured and predicted
stresses (Muránsky et al., 2012a). This level of confidence can- data that is most notably visible in the PASS.2 and PASS.3 weld
not be claimed for the plasticity measurements however, since metal. However, unlike the WRS measurements, we cannot be con-
a complementary measurement technique has not yet been fident that this difference is due to model inaccuracy; in fact, it is
used on these samples. likely that the indirect strain measurements are inaccurate in this
3. WRS predictions are in general agreement with measurements area. Likely sources of error include the microstructural variations
(Figs. 9 and 10). The only notable deviation from measurement (see Section 4) between weld and parent metals, which may lead to
lies within the weld metal, where all models tend to a different hardness-plasticity relationship than that shown in
over-predict WRS. This deviation increases with each additional Fig. 6. The aforementioned variations in grain size and shape, in
weld pass, and is particularly visible in the weld regions of chemical/phase composition, and in bulk texture will influence
PASS.2 and PASS.3. Since we have a high level of confidence in the hardness-plasticity relationship that was fitted to the parent
the measurements (even though only single phase stress analy- metal.
sis was considered in the diffraction measurement techniques An additional source of error in this comparison is that inferring
(Muránsky et al., 2012a; Martins et al., 2010)), it is most likely a relationship between hardness and plastic strain from a mono-
that the yield strength of the weld metal has not been accu- tonic test will not provide an exact relationship when a complex
rately represented in the model. Since only the parent material (e.g. cyclic) strain path has been applied to the specimen. This rela-
properties are used in the current FE analysis, the tionship is indeed simplified relative to other work (e.g. the
over-prediction of WRS in the weld metal suggests that the cross-weld studies of Smith et al. (2012a)), since the influence of
weld metal has a lower yield strength relative to the parent backstress (a) on the yield strength of material subject to cyclic
metal. This would be expected from the considerably coarser loading is not considered. Here, the accuracy of such a simplifica-
microstructure of the weld metal, as shown in Fig. 3(a). An tion has been assessed. Two important observations from this
increasing deviation in the predicted WRS exists since devia- assessment can be made:
tions in the shape misfit predicted for each weld pass are cumu-
lative. Nevertheless, these inaccuracies do not detract from the 1. PEEQ predictions (under identical annealing behaviour) using
present sensitivity study because the model errors they create isotropic, kinematic, and mixed hardening models are nearly
do not affect model sensitivity results. identical in the weld and HAZ material – see Fig. 9(d). This
 O. Muránsky et al. / International Journal of Solids and Structures 69–70 (2015) 518–530 527

Fig. 11. Comparison of the predicted post-weld plastic strain along line BD, broken into a pass-by-pass analysis. (a) Mixed plasticity theory with a single-stage annealing
assumption: T0 = 1050 °C. (b) Mixed plasticity theory with a single-stage annealing assumption: T0 = 1300 °C. (c) Mixed plasticity theory with a two-stage annealing
assumption: T1 = 800 °C; T2 = 1300 °C. Locations of the relevant isotherms after each pass are shown to identify the onset of material annealing in the analysis.

similarity implies the inclusion of a backstress tensor in the particularly within the parent metal (Y > 8 mm on line BD) where
plasticity model used will not significantly affect the PEEQ pre- the highest amount of plastic strain is expected (Muránsky et al.,
dictions of the FE model. The largest variation in PEEQ predic- 2012b). These bounding results arise since isotropic plasticity the-
tions occurs in the parent metal; however, even in this region ory over-predicts the material yield strength under cyclic loading,
the variation is less than the scatter observed in consequently over-predicting the elastic region (shown schemati-
hardness-inferred measurements. This observation tells us that cally in Fig. 2) and under-predicting the amount of plastic strain
if we can find a way to accurately measure PEEQ, it can be used (PEEQ). In contrast, kinematic plasticity theory under-predicts
reliably to assess the validity of an assumed material annealing the material yield strength under cyclic loading, thus
behaviour for weld simulation. under-predicting the elastic region and over-predicting post-weld
2. The greatest variation between inferred plasticity measurement plastic strain. The permanent shape misfit generated on heating
and calculated PEEQ profiles occurs (i) in the weld metal depos- between the near-weld region and the parent structure upon cool-
ited on the last pass and (ii) in the adjacent high-temperature ing is therefore accommodated to varying degrees within the elas-
HAZ from PASS.2 (Fig. 9(d)). If we focus on the measured data tic regime. Because small variations in elastic strain correspond to
in this region, we see that the inferred plasticity (hence, the large variations in stress (owing to the Young’s modulus of the
hardness) across the weld metal and into the HAZ are relatively metal), numerical solutions appear far more sensitive to the
constant. This trend implies that the potential for significant assumed material plasticity theory from a WRS standpoint than
backstress in the HAZ to increase the measured hardness (thus they do from a plastic strain standpoint. From this analysis it seems
the inferred plastic strain) is not observed; otherwise, we would that PEEQ predictions are largely insensitive to the plasticity model
see a significant increase in the measured hardness of the chosen, provided cyclic hardening behaviour is accurately captured
PASS.2 HAZ metal relative to PASS.3 weld metal. in the plasticity model.

These results indicate it is the microstructural effects and not the

monotonic simplification that produce the greatest source of mea- 5.2. Model sensitivity to the assumed material annealing behaviour
surement error in the present study. Neglecting the influence of
backstress seems to be an acceptable source of error, since the Figs. 7(d–f) and 10(a–c) compare longitudinal (r33 ) WRS predic-
authors have consequently reached a similar conclusion reached tions calculated using mixed plasticity theory with different
by Smith et al. (2012a), who consider backstress in their study of high-temperature annealing assumptions. The results show that
the cross-weld variation in yield strength in a three-pass groove changing the annealing assumption has a limited effect on WRS
weld specimen (Turski et al., 2009). While these results indicate predictions when using mixed plasticity theory. On the other hand,
a source of measurement error exists within the FZ and HAZ Figs. 8(d–f) and 10(d–f) show that the assumed material annealing
regions of the weldment, the numerical sensitivity study with behaviour has a significant effect on post-weld plastic strain pre-
respect to annealing temperature can be assumed valid. dictions (PEEQ). Interestingly, the trends observed when varying
the material annealing behaviour are opposite those observed
when varying the metal plasticity theory. To better understand
5.1. Model sensitivity to the assumed material plasticity theory why significant changes in PEEQ do not necessarily affect predicted
WRS, it is useful to examine the evolution of PEEQ after each weld
Contour plots in Fig. 7(a–c) and line profiles in Fig. 9(a–c) pass; these results are presented in Fig. 11 along the BD line.
clearly show that the choice of plasticity theory has a significant Considering the evolution of plastic strain under a 1050 °C
effect on the predicted WRS, as previously identified in Muránsky single-stage annealing assumption as in Fig. 11(a), the strain pro-
et al. (2012b). This effect is less pronounced for the predictions file after the first weld pass (PASS.1) indicates a sharp drop in
of post-weld plastic strain shown in Fig. 8(a–c), but it appears that post-weld plastic strain at temperatures exceeding the annealing
models using isotropic plasticity theory predict a lower-bound temperature. This characteristic indicates the amount of metal
plastic strain profile, while models using kinematic plastic theory plasticity accumulated in the annealed material on cooling (2%)
predict an upper-bound plastic strain profile. This claim is quanti- is significantly lower than the plastic strain accumulated on heat-
tatively supported in the line profiles shown in Fig. 9(d–f), ing (9%). Similar trends are also observed for single-stage
528 O. Muránsky et al. / International Journal of Solids and Structures 69–70 (2015) 518–530

Fig. 12. Schematic highlighting the variation in yield strength between annealed and work-hardened material during welding, using (a) isotropic, (b) kinematic and (c) mixed
plasticity theories. Line OAB represents the annealed material response on heating for all models. Line OABD represents the cyclic-hardened material response on heating and
subsequent cooling for each plasticity theory, while line OABC represents the annealed material response. Note, for sake of simplification it is assumed (unrealistically) that
the unstrained size of the yield surface (rj0 ) is constant over all temperature ranges.

annealing at 1300°C in Fig. 11(b), and two-stage annealing in reconstruction of Fig. 2 is presented in Fig. 12 to illustrate the rel-
Fig. 11(c). Both annealed and work-hardened material will undergo ative behaviour of annealed and work-hardened weld metal under
a similar plastic deformation on heating that leads to a shape mis- different plasticity theories.
fit, and is thus not directly affected by as yet accumulated plastic Consider material in the weld HAZ subjected to a representative
strain. This strain will indirectly affect WRS predictions however, weld thermal cycle. The following simplifying assumptions are
due to the increase in material yield strength caused by cyclic work imposed for illustrative purposes: (i) the unstrained size of the
hardening. The phenomenon can be shown schematically; a yield surface (rj0 ) is constant over all temperature ranges; (ii)
 O. Muránsky et al. / International Journal of Solids and Structures 69–70 (2015) 518–530 529

Fig. 13. Model sensitivity to the assumed material annealing behaviour, using isotropic plasticity theory. Variation in (a) the transverse and (b) the longitudinal WRS
predictions are presented alongside (c) post-weld plastic strain (PEEQ) predictions, for single-stage annealing temperatures of 1050 °C and 1300 °C.

single-stage annealing occurs at the temperature when perfect compares predicted WRS and post-weld plastic strain using two
plasticity is achieved in the material (when Q inf ¼ C i ¼ 0) and different single-stage annealing temperatures (1050 °C and
(iii) thermo-mechanical loading is uniaxial (parallel to r2 in 1300 °C) and isotropic plasticity theory. For both sensitivity analy-
Fig. 12). As the material is heated during welding, significant tran- ses using mixed and isotropic plasticity theories, the maximum
sient compressive stresses cause plastic deformation (line OAB in variation in longitudinal WRS occurs across the 1300 °C isotherm
Fig. 12) that leads to a permanent shape misfit between this region during the last welding pass (Y = 3 mm). This variation is twice
and the surrounding undeformed parent structure. The predicted as large (120 MPa) when assuming isotropic plasticity as it is when
constitutive material response on heating is identical under each assuming mixed hardening theory (60 MPa), despite the fact that
plasticity theory (Section 3.1), even though the evolution of the the predicted post-weld plastic strain profiles, shown in
material yield surface differs. The shape misfit formed on heating Figs. 10(d) and 13(c), are nearly identical. While such variations
leads to the development of tensile residual stresses parallel to in WRS are not as large as those observed when varying the plas-
r2 on cooling. Each plasticity theory will predict a different yield ticity theory (shown in Fig. 9(a) to be 170 MPa at Y = 8.5 mm), it
strength in tension and thus different WRS (Fig. 7(a–c)), based on is confirmed that weld models assuming isotropic plasticity theory
how that theory treats cyclic behaviour. Fig. 12 represents typical will be more sensitive to the material annealing behaviour, when
variations in yield strength as predicted by isotropic, kinematic predicting both post-weld plastic strain and WRS. These results
and mixed plasticity theories. therefore suggest that an analyst is more likely to have conservative
Should the weld HAZ be heated above the prescribed annealing predictions of both WRS and post-weld plastic strain when using iso-
temperature, two phenomena occur. First, the accumulated plastic tropic plasticity theory with a conservative annealing assumption that
strain is eliminated (epl = PEEQ = 0). Note that the removal of this approaches the melting temperature of the material.
plastic flow does not return the material to its original shape; Similar trends in model sensitivity can be observed when com-
instead a new, undeformed shape is assumed for the material as paring single-stage and two-stage annealing assumptions.
it must maintain compatibility with the surrounding weldment. Fig. 10(d) shows that the post-weld plastic strain predicted using
Hence, the strain shown after annealing in Fig. 12 indicates the two-stage annealing behaviour lies somewhere between the two
shape change relative to the original material geometry, and not different single-stage annealing results, while Fig. 10(a) shows
a residual plastic strain. Second, any further accumulation of plas- how this trend corresponds to a predicted WRS profile that also lies
tic strain is prevented until the temperature drops below T0 (i.e. between the single-stage annealing results. The clear advantage of
e_ pl ¼ 0). The influence of annealing on the material response upon using a two-stage annealing assumption is that it gives the analyst
more flexibility when defining the cyclic thermo-mechanical hard-
cooling can be inferred from Eqs. (1) and (2). Removal of epl will
ening behaviour of welding processes. As previously discussed by
remove all previous isotropic hardening, causing a contraction of
Smith et al. (2012a), two-stage annealing will also eliminate sharp
the yield surface back to rj0 and for a fully isotropic plasticity
discontinuities in PEEQ, which are not representative of the actual
model as in Fig. 12(a), the annealed material will yield at point C
material response. While the upper annealing temperature T2 is
upon cooling while the hardened material will yield at point D.
identical to the single-stage T0 of 1300 °C, significantly less PEEQ
For a fully kinematic model as in Fig. 12(b) the annealed material
is accumulated (e.g. at Y = 3 mm) due to the lack of isotropic hard-
yield surface will retain its size and position in stress space, such
ening at temperatures above T1 (800 °C), as shown in Fig. 11(c).
that the material will yield upon cooling at the same stress (point
While such information can significantly affect model accuracy
C) as the work-hardened material (point D). Consequently, anneal-
when predicting post-weld plastic strain, a two-stage annealing
ing does not influence kinematic hardening behaviour. This insen-
assumption requires the necessary empirical data to define T1
sitivity means that for a mixed hardening model as in Fig. 12(c), the
and T2.
influence of annealing on yield strength will be proportional to the
amount of isotropic hardening removed during the anneal. The
stress difference between points C and D will be less than for a fully 6. Conclusions
isotropic model and greater than for a kinematic model.
Because the difference between annealed and work-hardened An assessment of weld model sensitivity to the assumed consti-
material is greater when using isotropic plasticity theory, WRS pre- tutive material behaviour has been conducted using a validated FE
dictions made using this theory will be more sensitive to the model for a three-pass slot weld in austenitic steel. Model sensitiv-
assumed material annealing behaviour. To prove this fact, Fig. 13 ity to two key material assumptions was assessed: (i) the assumed
530 O. Muránsky et al. / International Journal of Solids and Structures 69–70 (2015) 518–530

material plasticity theory and (ii) the assumed material annealing Acknowledgments
behaviour. The sensitivity was quantified using both weld residual
stresses (WRS) and post-weld plastic strain (PEEQ) predictions. The The authors gratefully acknowledge the assistance of Dr. P.
following conclusions were drawn: Dayal (ANSTO) with cross-weld hardness measurements,
ECHIDNA high-resolution powder diffraction measurements by
1. In general, model predictions of WRS and PEEQ are in good Dr. M. Abdeev (ANSTO) and work by Dr. N. Ben Moussa (L’École
agreement with measurement, however variations between Supérieure des Sciences et Techniques de Tunis) on the correlation
measured and predicted data exist in the weld metal. between plastic strain and micro-hardness. All previous residual
Inaccurate WRS predictions arise in the weld metal due to stress measurements and predictions under the auspices of the
improper specification of the weld metal yield strength in the European Network on Neutron Techniques Standardisation for
FE model. Measurement inaccuracies in post-weld plastic strain Structural Integrity (NeT) programme, Task Group 4 are also
are related to the indirect hardness method used to measure acknowledged.
plastic strain. While this measurement technique accurately
captures accumulated plastic strain in the parent metal, the References
technique fails to account for significant microstructural differ-
ences in the weld and HAZ, which render the trends generated ABAQUS, 2007. A Novel Approach to Annealing using ABAQUS – Behaviour with Soft
Annealing. SIMULIA.
for the parent metal inapplicable in these regions. Ben Moussa, N., 2013. Optimisation des performances des surfaces usinées par le
2. An examination of model sensitivity to the assumed material contrôle des paramètres des procédés: Approches expérimentale et numérique.
plasticity theory reveals the choice of theory will have a signif- Ecole Supérieure des Sciences et Techniques de Tunis, vol. Mécanique. Tunis:
Ecole Supérieure des Sciences et Techniques de Tunis.
icant influence on the predicted WRS, but not the post-weld Brule, A., Kirstein, O., 1040. Physica B, 385–386, Part 2: 1040.
plastic strain. This result is believed to occur since each plastic- Dewees, D.J., Prueter, P.E., Kummari, S.R., 2014. American Society of Mechanical
ity model will predict slightly different elastic behaviour to Engineers. Pressure Vessels and Piping Division (Publication), PVP.
Hutchings, M.T., Withers, P.J., Holden, T.M., Lorentzen, T., 2005. Introduction to the
accommodate the shape misfit between the weld region and
Characterization of Residual Stress by Neutron Diffraction. CRC Press.
remainder of the parent metal on cooling. As slight changes to Liss, K.-D., Hunter, B., Hagen, M., Noakes, T., Kennedy, S., 1010. Physica B, 385–386,
the amount of accumulated elastic strain will result in large Part 2: 1010.
changes to the internal stress, this relative sensitivity (between Martins, R.V., 2009. NeT – Task Group 4: Three-Pass Slot Weld Specimen in
Austenitic Stainless Steel. Protocol for the Destructive and Non-Destructive
WRS and PEEQ) is justified. Determination of Residual Stress in Three-Pass Slot Weld Specimens in
3. An examination of model sensitivity to the assumed material Austenitic Stainless Steel.
annealing behaviour reveals this sensitivity is dependent on Martins, R.V., Ohms, C., Decroos, K., 2010. Mater. Sci. Eng. A 527, 4779.
Muránsky, O., Smith, M.C., Bendeich, P.J., Holden, T.M., Luzin, V., Martins, R.V.,
the plasticity theory used. It is observed that most of the Edwards, L., 2012a. Int. J. Solids Struct. 49, 1045.
post-weld plastic strain is accumulated during heating, where Muránsky, O., Hamelin, C.J., Smith, M.C., Bendeich, P.J., Edwards, L., 2012b. Comput.
compressive stresses in the near-weld region generate plastic Mater. Sci. 54, 125.
Muránsky, O., Smith, M.C., Bendeich, P.J., Hosseinzadeh, F., Edwards, L., 2014. Eng.
flow. Post-weld plastic strain predictions are thus highly sensi- Fract. Mech. 126, 40.
tive to annealing assumptions, since annealing will remove this Shchaeffler, A.L., 1949. Met. Prog. 56.
initial plastic strain. Prior strain hardening is shown to have a SIMULIA, 2014. ABAQUS/Standard. Dassault Systems.
Smith, M.C., 2006. Development of mixed isotropic-kinematic material hardening
variable influence on WRS predictions; a weak influence is models for finite element simulation of austenitic steel welds. E/REP/BDBB/
observed when isotropic hardening is offset by Bauschinger 0092/AGR/06, Revision 000: British Energy.
effects (i.e. mixed plasticity theory), while a strong influence Smith, R.M., 2010. FEAT-WMT: Weld-Modelling Tool User Guide. FEAT-WMT:
Weld-Modelling Tool User Guide: FeatPlus Limited.
is observed when Bauschinger effects are not considered (i.e.
Smith, M.C., Smith, A.C., Nadri, B., Bendeich, P.J., Carr, D.G., 2009. Optimisation of
isotropic plasticity theory). The influence is therefore a function mixed hardening material constitutive models for weld residual stress
of the amount of isotropic hardening present in the material; simulation using the NeT Task Group 1 single bead-on-plate benchmark
however, WRS predictions are less sensitive to annealing beha- problem. In: ASME Pressure Vessels and Piping Conference, vol. 6. Prague, Czech
Republic, p. 303.
viour than they are to the plasticity theory used, in all cases. Smith, M.C., Bouchard, P.J., Turski, M., Edwards, L., Dennis, R.J., 2012a. Comput.
4. It is shown that the mixed plasticity theory in combination with Mater. Sci. 54, 312.
a two-stage annealing approximation gives the most realistic Smith, M.C., Muránsky, O., Austin, C., Bendeich, P.J., Edwards, L., 2012b. Optimised
modelling of weld metal constitutive behaviour in the NeT TG4 international
prediction of both WRS and post-weld plastic strain. It is, how- weld simulation and measurement benchmark, In: 2012 ASME Pressure Vessels
ever, acknowledged that obtaining the experimental data for & Piping Conference Toronto, Canada.
the mixed plasticity theory and proper calibration of Turski, M., Smith, M.C., Bouchard, P.J., Edwards, L., Withers, P.J., 2009. J. Press. Vessel
Technol. 131, 1.
two-stage annealing temperatures might be difficult. Withers, P.J., 2001a. Bhadeshia HKDH. Mater. Sci. Technol. 17, 366.
Therefore, it is recommended that the analyst seeking conserva- Withers, P.J., 2001b. Bhadeshia HKDH. Mater. Sci. Technol. 17, 355.
tive predictions of both peak tensile WRS and post-weld plastic Withers, P.J., 2007. Rep. Progr. Phys. 70, 2211.
strain use an isotropic plasticity theory with a conservative
single-stage annealing temperature (i.e. one that approaches
the melting temperature of the parent metal).