Vous êtes sur la page 1sur 10

27 2003 June JOM

Weld Analysis Using an


Inverse-Problem Approach Based on
Quasi-One-Dimensional Interpolation
S.G. Lambrakos, J.O. Milewski, and P.G. Moore
Research Summary Welding
In this article, interpolation properties
are examined that are inherent in the
physical character of heat-deposition
processes such as welding and that are
relevant to using an inverse-problem
approach for their analysis. This study
considers interpolation properties of
both parametric and non-parametric
representations of the temperature eld
for heat-deposition processes. In the
case of parametric representations of the
temperature eld, linear combinations
of solutions to the heat-conduction
equation are adopted as interpolation
functions rather than strictly explicit
physical-model representations of heat-
deposition processes. In the case of
non-parametric representations of
the temperature field, interpolation
procedures that are applicable to discrete
representations of the temperature eld
are adopted. Issues related to parameter
INVERSE-PROBLEM APPROACH
In the inverse-problem approach to the analysis of welding processes, the temperature
eld across a weld is predicted using a model representation whose form is relatively
convenient for the adjustment of parameters.
15,14,15
The parameters are adjusted
according to experimental data concerning the actual temperature eld at various
locations that are sufciently distributed spatially throughout the workpiece. The model
representations adopted for an inverse-problem approach can be based on either
parametric or non-parametric mathematical formulations. Parametric formulations
can range from those that include detailed descriptions of the underlying physical
processes to those characterized by three-dimensional (3-D) interpolation functions whose
mathematical-function forms are relatively simple. Non-parametric formulations consider
a discrete 3-D representation of the temperature eld. In this formulation, the discrete
temperature eld quantities themselves represent the parameters to be adjusted. Reference
14 provides a relatively extensive overview of the inverse method based on parametric
representations as well as a comprehensive review of studies concerning the mathematical
foundation of inverse methods applied to heat conduction.
It is noted, therefore, that an inverse-problem approach to the analysis of heat-deposition
processes can be applied via any conveniently adjustable parametric or non-parametric
representation of the temperature eld. This follows since any well-behaved mapping
between spatially distributed temperature values, having been obtained according to either
experimental measurements or quantitative characterizations of the process, will represent
an accurate interpolation function. In this article, the terms inverse method and inverse
model denote methods of analysis based on the inverse-problem approach.
Before proceeding, one should note that there are many aspects of the inverse-problem
(Continued on page 34.)
optimization and the practical applica-
tion of methods based on the inverse-
problem approach related to interpola-
tion properties are also discussed.
INTRODUCTION
An inverse-problem approach to the
analysis of heat deposition or joining
processes allows the calculation of
temperature histories at locations within
a material during the heat-deposition
process.
15
These temperature histories
are relevant in that the associated mate-
rial response determines the resultant
properties and performance of the
material. The problem of calculating
the temperature history (i.e., the time-
dependent temperature eld associated
with a dynamic process involving heat
deposition) is becoming progressively
better posed for a variety of different
processes in that a large fraction of the
temperature eld can be experimentally
determined. It follows, therefore, that
methods based on the inverse-problem
approach, which are formulated such
that a large fraction of experimental
information concerning the temperature
eld can be mapped conveniently into
the predicted temperature history, are
well posed for this type of problem.
Throughout the discussion that follows,
reference is made interchangeably to
welding, heat deposition, and joining
processes. The concepts presented
are equally applicable to all these
processes.
The direct-problem approach to the
analysis of welding processes can be
dened as an approach in which the
temperature eld across the weld is
predicted via either an explicit numerical
solution of the coupled equations of
energy, momentum, and mass transport
a 2 mm
b 2 mm
Figure 1. (a) Longitudinal and (b) trans-
verse cross sections showing solidication
boundary of bead-on-plate weld.
a
b
b
28 JOM June 2003
Figure 4. A comparison of calculated
temperature history and thermocouple
measurement.
Figure 2. Longitudinal slices of a calculated temperature eld of bead-on-plate weld along (a)
the symmetry plane and (b) the line of thermocouple placement.
Figure 3. Locations of thermocouples
relative to friction stir weld on the bottom
surface of a workpiece.
250
200
150
100
50
0
T
e
m
p
e
r
a
t
u
r
e

(

C
)
120 130 140 150 160 170 180 190
Time (s)
or an explicit physical model based
on analytical solutions to the heat-
conduction equation for a specific
set of boundary conditions.
69
The
direct-problem approach requires an
a priori knowledge of the physical
characteristics of the energy source
and of the nature of its coupling to
the workpiece. Further, this approach
requires knowledge of the thermal and
uid-ow properties, as a function of
temperature, of the material making up
the workpiece. Reference 9 provides
a comprehensive review of the direct-
problem approach applied to a wide
range of welding processes. This
approach has yielded signicant contri-
butions to the analysis of friction stir
welding processes.
1013
Reference 13 and
its collected references provide a good
review of the direct-problem approach
applied to friction stir welding.
A spatial feature that is characteristic
of heat-deposition processes, joining
processes involving relative translation
of a workpiece, is that their dynamic
steady state may be partitioned into two
regions with respect to the motion of the
energy source relative to the workpiece.
One region contains the energy source
(an upstream region) and the other
contains the heat-affect zone (HAZ),
a downstream region. The ability
to observe an inherent partitioning
between upstream and downstream
regions implies a quasi-one-dimensional
interpolation property of temperature
elds associated with heat-deposition
processes. In addition, a well-dened
characterization of both upstream and
downstream conditions that bound
a given spatial region implies that
an interpolation function, having a
relatively simple functional form and
dened in terms of a convenient set
of parameters, can be adopted for
representation of the temperature eld.
It follows that inverse methods (see
the sidebar) that are based to some
reasonable degree on interpolation
should be well posed for calculation
of temperature elds associated with
dynamic heat deposition (or joining
processes) because of the inherent
quasi-one-dimensional trend features
that are characteristic of these processes.
Further, the ability to observe an inherent
upstream-to-downstream trend of the
dynamic temperature eld relative to a
Figure 5. (a) A longitudinal
slice at y = 0 and (b) a trans-
verse slice at x = 75, respec-
tively, of a calculated tem-
perature eld constrained
according to thermocouple
measurement #3 (1 unit =
0.127 mm).
a
b
29 2003 June JOM
b
a
Figure 6. (a) A longitudinal slice
at y = 0 and (b) a transverse
slice at x = 110, respectively, of
a calculated temperature eld
where values in HAZ are scaled
accordi ng to thermocoupl e
measurement #4 and the tem-
perature eld in TMAZ is calcu-
lated numerically by application
of elliptic solver (1 unit = 0.127
mm).
single dimension whose orientation is
along the direction of relative motion
between energy source and workpiece
(i.e., a quasi-one-dimensional trend)
imposes a restraint on the range of
values of the temperature eld as well as
on the character of its overall trend (i.e.,
the trend must be monotonic).
In this article, a series of case-study
analyses are presented that demonstrate
the interpolation properties inherent in
the physical character of heat-deposition
processes such as welding or joining
processes, and that are relevant to
inverse methods of analysis. This
study considers interpolation proper-
ties concerning both parametric and
non-parametric representations of the
temperature eld for heat deposition
or joining processes. In the case
of parametric representations of the
temperature eld, linear combinations
of solutions to the heat-conduction
equation are adopted as interpolation
functions rather than strictly explicit
physical model representations of heat-
deposition processes.
23,24
In the case
of non-parametric representations,
this study demonstrates the use of
an interpolation procedure that is
applicable to discrete representations
of the temperature eld. The theoreti-
cal foundation of this interpolation
procedure is that of discrete numerical
methods for solving second-order
elliptic boundary value problems.
68

BEAD-ON-PLATE WELDING
This case-study, which demonstrates
the use of an effective heat source
and thermal diffusivity as adjustable
parameters, considers an analysis of the
gas tungsten arc weld whose longitudinal
and transverse cross sections are shown
in Figure 1 and whose model parameters
are given in Table I. This analysis adopts
a downstream constraint condition
on the temperature eld based on a
thermocouple measurement whose
point of attachment is on the bottom
surface of the workpiece. Although this
weld represents a very simple case, it
is signicant in that it demonstrates the
inherent interpolation property of heat-
deposition processes such as welding
and the use of an effective heat source
and thermal diffusivity as adjustable
parameters. The transverse position of
the thermocouple attachment point is
30 JOM June 2003
a
b
Figure 7. (a) A longi-
tudinal cross section
showing the solidi-
cation boundary and
(b) calculated tem-
perature eld at the
midplane of deep-
penetration weld.
given in Table I.
The rst stage of this analysis entails
the construction of a three-dimensional
(3-D) distribution of elemental heat
sources according to Equations AD.
This distribution is constructed accord-
ing to the observed shape of the weld
cross sections but without a priori
knowledge of the material ow proper-
ties of the melt pool. For this case
study, a single point source is sufcient
for calculating a temperature field
according to Equations AD. For these
calculations, non-conducting boundary
conditions are assumed for both the top
and bottom surfaces of the workpiece.
The values of the effective heat source
and diffusivity are adjusted so that there
is good correspondence between the
calculated and experimental transverse
cross sections (Figure 1). From the
perspective of quasi-one-dimensional
interpolation, this stage represents the
assignment of temperature values at
centralized positions near the solidica-
tion boundary that are within the spatial
range of the temperature eld.
The second stage of this analysis
entails a ne adjustment of the scale
of the temperature eld so that there
is good correlation between calculated
values of the temperature eld on the
bottom surface of the work piece
and a thermocouple measurement.
From the perspective of quasi-one-
dimensional interpolation, this stage can
be interpreted as assigning temperature
values at the downstream region of the
spatial range over which the temperature
eld is dened. For this calculation, a
scale factor was determined according
to the thermocouple measurement given
in Table I such that the strength of
the source was assigned a value q
k
=
q
est
where q
est
is the source strength
calculated according to the transverse
cross section. This procedure represents
a ne adjustment of the strength of
the heat source.
The third stage of this analysis entails
assigning a maximum-temperature
isothermal surface whose location is
correlated with that of the energy source
at the top surface of the work piece. As
in the second stage, this stage represents
an assigning of temperature values
at upstream regions of the spatial range
over which the temperature eld is
dened. The numerical operation applied
at this stage involves the application of
a discrete elliptic solver (i.e., Equation
E) for generation of the temperature
eld between the isothermal surface
of maximum temperature and the
solidication boundary. This numerical
operation is described in Reference
5 and is essentially that of a quasi-
one-dimensional interpolation between
upstream and downstream values of
the temperature eld. Shown in Figure
2 are two-dimensional (2-D) slices of
the 3-D temperature eld calculated
following the procedure described in this
section. The total energy per distance is
523 kW/m for this calculation, a value
assumed as reasonable for the present
study. In principle, however, if more
detailed information were available
concerning the energy per distance
coupled into the workpiece (e.g.,
efciency), the values of q and could
be readjusted to satisfy a specified
target value of the total energy per
distance.
FRICTION STIR WELDING
(HAZ)
As in the previous analysis, this case-
study analysis demonstrates the use of
an effective heat source and thermal
diffusivity as adjustable parameters of
an interpolation function. A signicant
aspect of this analysis, however, is
that, in principle, the relative computa-
tional cost is not small for calculating
temperatures at downstream locations
of the workpiece. This is due to the
fact that, in this case, the energy source
spans a relatively large volume of the
workpiece. This case-study analysis
considers a prototype friction stir weld
whose process and model parameters
are given in Table I. This prototype
friction stir weld has been constructed
using experimental data presented
in Reference 17. In this analysis,
downstream and midstream constraint
conditions of the temperature eld are
adopted that are based on thermocouple
measurements whose attachments are
distributed along the bottom surface
of the workpiece. Shown in Figure 3
are the relative locations and numeric
labeling of thermocouples on the bottom
surface of the workpiece. This analysis
considers calculation of the temperature
eld in regions of the weld dened by the
31 2003 June JOM
Table I. Model Parameters
k
c
C
p
C
1
V
k
D l T
M
T
G
T
MAX
Case W/(mK) (Mg/m
3
) J/(KgK) (m
2
/s) (Eq. 5) (mm/s) (mm) (mm) (C) (C) (K) Notes
Bead-on-Plate Welding
304 stainless steel 35 7.2 805.56 9.8 10
6
3.76 1.27 12.8 0.256 a
Friction Stir Welding
6061-T6 Al 167 2.7 896 1.88 10
5
1.09 10
4
2.17 6.35 0.127 582 b, c
Deep-Penetration Welding
304 stainless steel 5 10
6
0.159 1,426.85 2,741.0 d
Consumable-Electrode Welding
304 stainless steel 35 7.2 805.56 9.8 10
6
3.76 4.37 12.8 0.256
Calculation of Top Surface Boundary Condition
304 stainless steel 2.276 10
2
1,700 2,700 e, f
+ 3.285 10
5

T2.762 10
9
T
2
NOTES:
a) Thermocouple measurementmaximum temperature: 550C; location on workpiece: bottom and 8.0 mm from center of weld.
b) Thermocouple #3 (see Figure 3)maximum temperature: 244C; location on workpiece: 3.8 mm from top and 14.3 mm from center of weld.
c) Thermocouple #4 (see Figure 3)maximum temperature: 353C; location on workpiece: 3.8 mm from top and 4.7 mm from center of weld.
d) Type: electron-beam weld; bead depth below surface: 8.0 mm; bead width at surface: 1.66 mm; plate thickness: 25.4 mm; weld speed: 3.0 cm/s.
e) is effective diffusivity.
f) T is in units of K.
g) Distance between heat source at T
MAX
and solidication boundary at T
M
along midplane and at top surface of workpiece: 10 mm.
HAZ, which are external to the stirred
region dening the thermomechanically-
affected zone (TMAZ).
This analysis entails the construction
of a 3-D distribution of elemental
heat sources according to Equations
AD. This distribution is constructed
according to the shape of the stirring
tool, whose geometric parameters are
given in Table I, but without a priori
knowledge of the material properties or
ow characteristic of the stirred region
dening the TMAZ or of any details
concerning the specic geometry of the
stirring tool. All elemental heat sources
making up this distribution are given
the same value. The strength of the
source distribution and the value of the
diffusivity are adjusted so that there
is good correspondence between the
calculated thermal history and that
of a thermocouple measurement at a
specied downstream location of the
workpiece (#3 in Figure 3). From the
perspective of quasi-one-dimensional
interpolation, this stage can be inter-
preted as assigning temperature values
at the downstream region of the spatial
range over which the temperature eld
is dened. The calculated temperature
eld is constrained according to the
thermocouple measurement #3 shown
in Figure 4. This constraint is according
to both the measured peak temperature
and temperature history. Shown in
Figure 5 are 2-D slices of the 3-D
temperature eld constrained according
to the thermocouple measurement #3.
Referring to Figure 4, it is seen that
the effective heat-source distribution
and diffusivity have been adjusted
such that the calculated temperature
eld and thermocouple measurement
temperature assume not only the same
peak temperature but also very similar
trends as a function of time. Referring
to Figure 5, it is signicant to note that
temperatures close to the stirring tool
are greater than the solidus temperature
of 6061-T6 aluminum (582C). This
result is inconsistent with the nature of
friction stir welding processes, which
are characterized by solid-state mass
transfer, and for temperature values
that cannot exceed that of melting.
This result is consistent, however, with
the assumption of unrealistically long
temperature histories at downstream
locations. This implies the presence
of a delay in response time, which is
a consistent experimental artifact of
thermocouple measurements.
FRICTION STIR WELDING
(TMAZ)
This case study analysis demonstrates
the use of thermocouple measurements
to calculate the temperature eld at
locations within the TMAZ of a friction
stir weld. This prototype analysis
entailed an adjustment of the scale of
the temperature eld shown in Figure 5
so that there was an agreement between
calculated values of temperature eld at
the edge of the observed TMAZ on
the bottom surface of the workpiece
and a thermocouple measurement.
For this calculation, a scale factor
was determined according to the peak
temperature of TC #4 (353C) such that
the strength of the elemental sources
were each assigned a value q
k
= q
est

where q
est
was the source strength
calculated according to thermocouple
measurement #3 shown in Figure 4.
The next stage of this analysis entailed
assigning a maximum-temperature
isothermal surface whose shape was
correlated with that of the stirring tool.
The temperature values assigned to this
surface were each that of the solidus
temperature of 6061-T6 aluminum
(582C). As discussed in the previous
analysis, this temperature represents a
reasonable estimate of an upper bound
on temperatures in the region of the
stirring tool. This stage represents
an assigning of temperature values at
upstream regions of the spatial range
over which the temperature eld is
dened. Finally, the temperature eld
was calculated numerically between
the isothermal surface corresponding
to the surface of the stirring tool and
the isothermal surface at the measured
edge of the TMAZ, having the peak
temperature of TC #4. This numerical
operation was effected via application of
b
c
a
32 JOM June 2003
Figure 8. Slices of a calcu-
lated temperature eld for a
consumable-electrode weld
constrained according to
the shape of reinforcement
structure. Slices are for (a)
a longitudinal cross section
along midplane, (b) a trans-
verse cross section of melt
pool, and (c) a longitudinal
cross section of melt pool
along midplane.
a discrete elliptic solver (see Reference
5) over the region bounded by these
isothermal surfaces. Shown in Figure
6 are 2-D slices of the 3-D tempera-
ture field generated following this
procedure.
DEEP-PENETRATION
WELDING (HAZ AND MELT
POOL)
This case study analysis demonstrates
that an inverse-problem approach
provides a means for compensating
for incomplete explicit information
concerning the temperature dependence
of material properties, uid properties of
materials, and for incomplete informa-
tion concerning the spatial characteristics
of the energy source. In addition, this
study demonstrates that the inverse
method is well posed for analysis
of processes characterized by large
temperature gradients, or in general,
sharp transitions in scale. The prototype
analysis considered is that of the
deep-penetration electron-beam weld
whose longitudinal cross section of
the solidication boundary is shown in
Figure 7, and whose dimensions and
process parameters are given in Table
I. For this analysis, three types of cross
sections of the solidication boundary
have been adopted for the specication
of constraints: the transverse cross
section, the extent of the melt pool at
the top surface of the workpiece, and
the solidication boundary along the
midplane of the weld, parallel to the
direction of motion of beam.
The rst stage of this analysis entails
generating a 3-D isothermal surface at
the melting temperature T
M
such that the
different projections of this surface map
onto the experimentally measured cross
sections of the solidication boundary.
This procedure is undertaken by a line
distribution of elemental heat sources,
Equations AD, extending into the
region of the workpiece. The second
stage of this analysis consists of partition-
ing that region into two subdomains. One
subdomain is dened by an isothermal
surface at the vaporization temperature
T
G
(the upstream surface), a section of
the midplane boundary, a section of the
top surface boundary, and the isothermal
surface at T
M
(the downstream surface).
The other subdomain is dened by the
region of the workpiece at temperatures
33 2003 June JOM
Figure 9. The temperature
eld between the energy
source and solidification
boundary at the intersec-
ti on of the top surface
and the symmetry plane of
workpiece. The temperature
eld is specied by one-
dimensional boundary value
problem.
10 20 30 40
1,750
2,000
2,250
2,500
X (1 unit = 0.256 mm)
T
e
m
p
e
r
a
t
u
r
e

(
K
)
less than T
M
.
Proceeding with the analysis, it is
to be noted that with respect to the
HAZ, the distribution of elemental
heat sources used for generating the
isothermal surface at T
M
represents
a physically consistent apparent heat
source. This follows since the linear
combination of solutions dened by
Equations AD are consistent with
boundary conditions for a workpiece of
nite thickness (see Table I). The third
stage of this analysis entails application
of Equation E as a discrete elliptic
solver over the region of the workpiece
bounded upstream and downstream
by isothermal surfaces at T
G
and T
M
,
respectively. The shape of the isothermal
surface at T
G
is a cylinder extending into
the workpiece whose radius is consistent
with the experimentally observed size
of the keyhole structure associated
with the vapor-liquid boundary (see
Reference 1). A zero temperature-
gradient boundary condition is applied
at the surface of the workpiece.
Shown in Figure 7 is the longitudinal
slice at the midplane of the temperature
eld calculated by the inverse method. It
is signicant to note that the temperature
eld within the melt pool is consistent
with both the material and uid-ow
properties of the liquid metal. This
follows because the boundary values,
which extend over an essentially closed
region, provide an implicit representa-
tion of the temperature dependence of
the material properties as well as of
the coupling of heat transfer to uid
convection. This property of closed
regions, which is based on Greens
theorem, is discussed in Reference 5.
An important aspect of this calculation
is that the combined weighting of both
upstream and downstream boundaries
(i.e., quasi-one-dimensional interpola-
tion) is seen to homogenize sharp
transitions in scale. As a result, the
temperature eld was calculated by
applying the discrete elliptic solver,
Equation E, over uniform grid spacing 1
(see Table I). Typically, the temperature
gradients close to the energy source are
very large (see Figure 9, for example).
As a result, any numerical procedure
used for calculating the temperature
eld based only on upstream boundary
conditions (i.e., procedures based on
the direct problem approach) must
consider the singular character of
the temperature eld near the energy
source. The temperature eld shown in
Figure 7 was calculated with the grid
resolution 1 given in Table I.
CONSUMABLE ELECTRODE
WELDING (HAZ AND MELT
POOL)
This study demonstrates the inclusion
of shape information associated with
heat-deposition processes characterized
by reinforcement structures at the
top surface of the workpiece. These
processes include welding processes
based on consumable electrodes and
rapid prototyping involving the melting
of powder or wire. This case study
adopts the observed steady-state shape
of the melt pool and reinforcement
structure as input information. The
shape features of the melt pool and
reinforcement structure that are adopted
as constraints are shown in Figure 8. The
procedure followed for this analysis is
similar to that for the analysis of deep-
penetration welding, and the model
parameters are given in Table I. For this
analysis, a zero temperature-gradient
boundary condition is applied both at
the curved surface of the reinforcement
structure and at the surface of the
workpiece. The generating function
for constructing the shape of boundary
surface of the reinforcement structure
is a linear combination of Gaussian
functions. In contrast to the previous,
deep-penetration welding analysis,
however, there is an additional constraint
condition on the temperature eld at
the intersection of the top surface
and symmetry plane. This constraint
condition, which represents a means
of providing the inverse model with
information concerning the thermal
diffusivity , entails an explicit assign-
ment of the temperature eld along
the line dened by the intersection of
the top surface and symmetry plane of
the weld (consisting of reinforcement
structure and workpiece).
CALCULATION OF TOP
SURFACE BOUNDARY
CONDITION
This case study analysis demonstrates
a procedure for calculating the tem-
perature eld within the melt pool at
the intersection of the top surface and
symmetry plane of the workpiece via the
numerical solution of a one-dimensional
(1-D) second-order boundary value
problem. Again, this method utilizes the
quasi-one-dimensional trend inherent in
heat-deposition processes. Referring to
the longitudinal slice in Figure 7, it is to
be noted that the calculated temperature
eld T has been generated using only
boundary-value information such that
essentially
2
T = 0 is solved over the
region of the melt pool. This calcula-
tion does not adopt any information
concerning material properties or the
inuence of convection other than that
which is represented implicitly by the
upstream and downstream boundary
conditions. Referring to Equation E, it
follows that the quasi-one-dimensional
interpolation procedure used for calculat-
ing the temperature eld can be extended
to include information concerning the
thermal diffusivity as a function of
temperature (i.e., the discrete thermal
diffusivity eld ) can be specied.
Accordingly, one can construct a 1-D
boundary value problem using two
experimentally observable features of
a weld spanning a 3-D workpiece and
34 JOM June 2003
the thermal diffusivity as a function of
temperature. These features are, simply,
the distance between the heat source
and solidication boundary along the
midplane of the weld at the top surface
of the workpiece and the maximum
average temperature at the location of
the heat source. Shown in Figure 9 is the
calculated temperature eld obtained
by numerical solution of a second-order
boundary value problem that assumes
the weld features and thermal diffusivity
function
27
given in Table I. Referring
to Figure 9, it is signicant to note
that although a temperature eld has
been calculated along a single spatial
dimension (i.e., x = Vt), this temperature
eld is to be interpreted as part of
a 3-D eld in that it was calculated
using information concerning 3-D
characteristics of the weld. This point
is made clearer by noting that at the
intersection of the top surface and
symmetry plane of the workpiece the
temperature field must satisfy the
equations

V
T
x x
uT
x
T
x

+ ( )

_
,

(1)

T
y
0
,

T
z
0
, T(x
s
) = T
MAX
, and
T(x
M
) = T
M
(2)
where x
s
and x
M
are the experimentally
observed locations of the energy source
and solidication boundary, respectively,
V is the welding speed and u is
the component of the uid velocity
along the x-coordinate. The calculated
temperature eld shown in Figure 9
does not include contributions from
the terms containing V and u in Equa-
tion 1. Consistent with the inverse-
problem approach, it is assumed that this
information is represented implicitly
by the boundary conditions at x
s
and
x
M
. The inverse-problem procedure
based on Equations 1 and 2, although
computationally simple, can provide
additional boundary-value information
for analyses such as in the deep-
penetration and consumable-electrode
cases. Referring to the calculated
temperature eld in Figure 7 correspond-
ing to a deep-penetration weld, one
notes that the inuence of quasi-one-
dimensional interpolation is relatively
strong and therefore the inclusion of
approach in general
15
and in the analysis of heat-deposition processes that are relevant to
its methodology. Among these are the inclusion of physical information by adaptation of
direct-problem models to inverse methods;
14
the inherent non-uniqueness of parameters
associated with inverse methods; the signicance of the concept of apparent or effective
heat-source distributions; and computational or algorithmic issues related to parameter
adjustment and optimization. Although these aspects are considered, the emphasis here is
primarily on interpolation properties of heat-deposition processes relevant to using inverse
methods. Further, it should be noted that the inverse method examined here has been applied
to a range of processes that include deep-penetration welding,
5
friction stir welding,
1618

and multipass welding.
19,20
Again, although specic aspects of the application of the inverse
method to different types of welding or joining processes are considered, the emphasis is
on the utilization of characteristically general interpolation properties of heat deposition
and joining processes for analysis.
An important aspect of the inverse-problem approach is the use of thermocouple
measurements for obtaining spatially distributed information concerning the temperature
eld.
21
This is especially the case for inverse methods based on interpolation properties
of heat-deposition processes. The issue to be considered when adopting thermocouple
measurements for the purpose of inverse modeling is that their response is characterized
by an inherent time delay. This time delay is a function of the nature of the attachment
of the thermocouple to a given location on the workpiece.
22
It follows then that inverse
methods that adopt thermocouple information must consider variability resulting from a
set of thermocouples being attached differently to the same workpiece, the nature of the
dependence of the characteristic time delay on the type of attachment, and the relationship
or level of correlation between thermocouple measurements and actual temperature histories
at positions of the workpiece.
TEMPERATURE FIELD REPRESENTATIONS
Two representations, parametric and non-parametric, of the temperature eld can be used
for application of inverse methods of analysis. A parametric representation of the temperature
eld T(x ) at positions x = (x, y, z) within the workpiece is given by
T x T T x-x V
A k k k
k=1
N
, ( ) + ( )

and T(x C
m m
) (A)
where T
A
is the ambient temperature of the workpiece and the locations x
m
and temperature
values C
m
specify constraint conditions on the temperature eld. The contribution due to each
individual element of the distribution of sources is given by
T (x-x ,V )=
q
4 k
exp
V (x-x
2
k k k
k
C
k k
)

_
,


1 1
R
exp
V R
2 R
exp
V R
2
i
k i
j
k j
j

_
,

_
,

_
,

_
,

1
]
1
1 i
(B)
where
R
i
[(x x
k
)
2
+ (y y
k
)
2
+ (z 2iD z
k
)
2
]
1/2
(C)
and
R
j
=[(x x
k
)
2
+ (y y
k
)
2
+ (z 2jD + z
k
)
2
]
1/2
(D)
The quantity D is the thickness of the workpiece and x
k
= (x
k
, y
k
, z
k
), k = 1, . . ., N, are
the locations of the elemental heat sources of strength q
k
. The sum dened by Equation A
species a numerical integration over the discrete elements of a distribution of sources that
can be characterized by individual points, line segments, or surfaces of various shapes. For
the analyses presented here, various distributions of heat sources are constructed using sets
of line elements, an example of which is shown in Figure A. Formally, Equation A is a linear
Figure A. A schematic repre-
sentation of the distribution
of elemental heat sources
consisting of contiguous line
segments.
(Continued from page 27.)
q1
q2
q3
qn
qi- strength of line
source element i
line source elements
35 2003 June JOM
additional boundary-value information
(e.g., temperature eld at the top surface)
should not be signicant. This is not
the case for the calculated temperature
eld in Figure 8 corresponding to a
consumable-electrode weld. For this
prototype analysis, the size of the
upstream boundary surface, representing
the source, is relatively small compared
to that of the downstream boundary
surface dened by the solidication
boundary. As a result, calculation of
the temperature eld by application of
an elliptic solver over the region of
the melt pool will tend not to include
appropriate weighting of the inuence
of upstream conditions. Referring to
Figure 8c, it is seen that the inclusion
of top surface boundary information
results in a boundary surface that is
comparable to that of the solidication
boundary. As a result, calculation of the
temperature eld by application of an
elliptic solver will include comparable
weighting of the inuences of both
upstream and downstream conditions.
ACKNOWLEDGEMENTS
The authors would like to thank
the U.S. Department of Energy for
sponsoring most of this research. A Naval
Research Laboratory core program on
friction stir welding sponsored some
of this research. This work was done
during an active collaboration with
the Truchas Team, the Computer and
Computational Science Division, and
the Materials Science and Technology
Division of Los Alamos National
Laboratory. The authors would like
to thank Deniece Korzekwa, David
Korzekwa, Marius Stan, Douglas Kothe,
Kin Lam, and Jim Sicilian for their sup-
port and many discussions concerning
direct and inverse modeling.
References
1. S.G. Lambrakos et al., J. Mater. Eng. Perform.,
3 (5) (1994), p. 639.
2. S.G. Lambrakos and J.O. Milewski, J. Mater. Eng.
Perform., 4 (6) (1995), p. 717.
3. S.G. Lambrakos and J.O. Milewski, J. Mater. Eng.
Perform., 7 (6) (1998), p. 1.
4. S.G. Lambrakos and D.W. Moon, Computer-Aided
Design, Engineering and Manufacturing, Systems
Techniques and Applications, Vol. VI (New York: CRC
Press, 2001), pp. 6-1 to 6-38.
5. S.G. Lambrakos and J.O. Milewski, Science and
Technology of Welding and Joining, 7 (3) (2002),
p. 137.
6. S.V. Patankar, Numerical Heat Transfer and Fluid
Flow (New York: Hemisphere Publishing, 1989).
combination of solutions to the heat-conduction equation. The quantities V
k
are the effective
relative speeds of the elemental heat sources comprising the heat-source distribution. A
consistent assumption is that V
k
(for all k) is equal to the translational speed of the heat
source, or in the case of friction stir welding, the stirring tool. The diffusivity is dened as
= k
c
/C
p
, where the values of the thermal conductivity k
c
, density , and specic heat C
p
are
provided in the table for the case studies presented.
The parametric representation dened by Equations AD may be adopted as a physically
consistent model of the underlying process (i.e., pure thermal diffusion) for positions within
the heat-affect zone (HAZ). This representation may also be adopted, however, for mapping
out the temperature eld at positions within the melt pool, given that one has information
concerning temperature values at positions that are close to or within the region of the energy
source. In this case, the parametric representation dened by Equations AD can be adopted
as a 3-D interpolation function for calculating the temperature eld between upstream (the
region of coupling of the workpiece to the energy source) and downstream (the solidication
boundary) temperature values. Equations AD may also be adopted as a generating function
whose purpose is to generate only downstream values of the temperature eld as well as a
downstream boundary surface. Upstream values of the temperature eld are then calculated
numerically by application of a discrete elliptic solver that establishes a quasi-one-dimensional
interpolation between upstream and downstream boundary values.
A non-parametric representation of the temperature eld T
p
at discrete grid location
x
p
within the workpiece is given by
T
w
T
p
p
i
i=1
6
i


1
and T
m
= C
m
, where w
p i
i=1
6

(E)
and the indices i = 1, . . ., 6 are for discrete nearest-neighbor grid locations along the x, y, and
z coordinate axes relative to the grid location whose index is p. Given a bounded region of
the workpiece, which is essentially closed owing to a spatially distributed set (of sufcient
size) of constraint conditions on the temperature eld, Equation E provides a non-parametric
representation of the temperature eld for purposes of optimization. The direct-problem
formulation upon which this interpolation procedure is based is that of the nite-volume method
of solving the coupled equations of energy, momentum, and mass transport.
68
The formal procedure underlying the inverse method considered here entails the adjustment
of a temperature eld T
p
or T(x
p
) dened over the entire spatial region of the workpiece.
This approach denes an optimization procedure where a temperature eld spanning the
spatial region of the workpiece is adopted as the quantity to be optimized. Depending on
the characteristics of the spatial distribution of constraint information, the temperature
eld is represented either parametrically (Equations AD) or non-parametrically (Equation
E). Accordingly, an optimization criterion may be dened by minimizing the value of
an objective function of the form

Z q q T(x T
max T(x,y z T
ob 1 N n
c
n
c
n=1
N1
m
c
m
c
m
c
( ,..., ) )
, )
( )
+ { }
(

2
))
+

m=1
N2
E
2
Z
(F)
For purposes of convenience related to parameter optimization, the source strengths dened
in Equation B may be interrelated through functions of the form
q
k
= C
k
q
k+1
(G)
for k=1, . . ., N-1, where the coefcients C
k
are adopted as adjustable parameters. The
contribution to the objective function dened by Z
E
is for constraints associated with initial
estimates of the temperature eld which would in principle represent reasonably good
rst approximations and therefore require only relatively small adjustments for further
optimization [see accompanying prototype analyses for bead-on-plate welding and friction
stir welding-thermomechanically affected zone (TMAZ)]. During the course of the prototype
analyses presented here no discussion is given concerning specic procedures for parameter
adjustment relative to minimization of an objective function. Although this represents a
signicant aspect of the practical application of the inverse-problem approach (see Reference
14), it is relatively independent of the interpolation properties examined here.
The input of information into the inverse model dened by Equations AG is effected
by the assignment of individual constraint values to the quantities C
m
(see Equation A and
Equation E); the form of the function adopted for parametric representation (Equation B) for
the analyses presented here; specifying the shapes of the upstream and downstream boundary
surfaces which bound the temperature eld within different regions of the workpiece; and
specifying the shape of and temperature-eld values at the top boundary surface. Specifying
the shapes and temperature eld values of boundary surfaces is equivalent to assigning a set
(Continued on page 36.)
36 JOM June 2003
7. D.A. Knoll et al., SIAM J. Sci. Comput., 23 (2)
(2001), p. 381.
8. M. Pernice and M.D. Tocci, SIAM J. Sci. Comput.,
23 (2) (2001), p. 398.
9. H. Cerjak, editor, Mathematical Modelling of Weld
Phenomena, H.K.D.H. Bhadeshia, series editor,
volumes 1 to 6 (1993, 1995, 1997, 1998, 2001, and
2002, respectively).
10. . Frigaard, . Grong, and O.T. Midling, A Process
Model for Friction Stir Welding of Age Hardening
Aluminum Alloys, Metall. Mater. Trans. A, 32A (2001),
p. 1189.
11. G.J. Bendzsak, T.H. Northand, and Z Li, Acta
Metall. Mater., 45 (4) (1997), p. 1735.
12. K.V. Jata, M.W. Mahoney, R.S. Mishra, S.L.
Semiatin, and D.P. Field, editors, Friction Stir Welding
and Processing (Warrendale, PA: TMS, 2001), pp.
343.
13. H. Shercliff and P.A. Colegrove, Modelling of
Friction Stir Welding, Mathematical Modelling of
Weld Phenomena 6, ed. H. Cerjak (London: Maney
Publishing, 2002), pp. 927974.
14. V.A. Karkhin et al., Inverse Modelling of Fusion
Welding Processes, Mathematical Modelling of
Weld Phenomena 6, ed. H. Cerjak (London: Maney
Publishing, 2002), pp. 10171042.
15. K.A. Woodbury, editor, Inverse Engineering
Handbook (New York: CRC Press, 2003).
16. R.W. Fonda and S.G. Lambrakos, Science and
Technology of Welding and Joining, 7 (3) (2002),
p. 177.
17. S.G. Lambrakos et al., Analysis of Friction
Stir Welds Using the Inverse-Problem Approach,
Proceedings of Sixth International Conference on
Trends in Welding Research (Materials Park, OH: ASM
International, 2002), pp. 153157.
18. S.G. Lambrakos et al., to appear in Science and
Technology of Welding and Joining (2003).
19. E.A. Metzbower, D.W. Moon, and S.G. Lambrakos,
Hardness and Microstructures in an HSLA100 Weld,
Sixth International Conference on Trends in Welding
Research (Materials Park, OH: ASM International,
2002), 690694.
20. D.W. Moon et al., Microstructures, Hardness
and Temperature in an HSLA100 Welds, to appear
in Science and Technology of Welding and Joining
(2003).
21. V.R. Dave et al., Welding Journal Research
Supplement, 80 (10) (2001), p. 246.
22. T. Sawada and N. Nishiwaki, Int. J. Mech. Sci.,
33 (7) (1991), p. 55.
23. . Grong, Metallurgical Modeling of Welding
(London: The Institute of Materials, 1994).
24. H.S. Carslaw and J.C. Jaeger, Conduction of
Heat in Solids, 2
nd
edition (Oxford, U.K.: Clarendon
Press, 1959).
25. S.G. Lambrakos et al., J. Mater. Eng. Perform,
2 (6) (1993), p. 819.
26. Thermal Diffusivity as a Function of Temperature
for 304 Stainless Steel (private communication with
K. Mahin, Center for Materials and Applied Mechanics,
Sandia National Laboratory, 1992).
S.G. Lambrakos is with the Materials Science
and Technology Division of the Naval Research
Laboratory in Washington, D.C. J.O. Milewski
is welding and joining team leader with the
Metallurgy Group of the Materials Science and
Technology Division at Los Alamos National
Laboratory. P.G. Moore is with the Department
of Computer Science at the University of
Maryland.
For more information, contact S.G. Lambrakos,
Naval Research Laboratory, Materials Science and
Technology Division, Code 6324, Washington, D.C.
20375-5000; (202) 767-2601; fax (202) 767-2623;
e-mail lambrakos@anvil.nrl.navy.mil.
of constraint values C
m
(see accompanying analyses of friction stir welding-TMAZ and
deep-penetration welding prototypes).
Before proceeding, it is signicant to note that constrained optimization of the calculated
temperature eld, via the minimization of an objective function such as Equation F,
requires, in principle, a reasonable number of trial iterations in order to nd the appropriate
neighborhood in parameter space containing an optimal set of parameters. Further
iterations are then required to subsequently adjust these parameters (or a set of discrete
temperature values) relative to some designated tolerance on the calculated minimum of
the objective function. For cases where the number of elemental heat sources is large,
(see accompanying analyses of friction stir welding-heat affect zone and friction stir
welding-TMAZ) even a reasonable number of trial iterations can result in a substantial
computational cost due to the large number of calculations at each iteration. Minimization
of computational cost can be effected, however, by sampling temperature eld values
at only a limited number of locations within the workpiece. These locations would
be expected to be either near or at locations for which constraint conditions on the
temperature eld are specied.
Following the inverse-problem approach, the thermal diffusivity can be interpreted as
either a phenomenological quantity or an effective thermal diffusivity whose value
is related to the average material properties of the workpiece. The inverse-problem
approach presented here adopts the thermal diffusivity dened in Equations AD as
an adjustable parameter as well as the distribution of heat sources. This exibility is
supported by the following observations.
The linear combination of solutions to the heat conduction equation dened
by Equations AD is applicable for three different types of representations: an
approximate physical model representation, a 3-D interpolation function for
assigning temperature values over the full range of spatial locations of interest,
and a generating function for assigning temperature values over either upstream
or downstream boundary surfaces.
The linear combination of solutions given by Equations AD is dened in terms of
two parameters C
1
and C
2
given by
C
q
4 k
1
k
C

and C
V
2
k

(H)
An inverse-problem approach that adopts Equations AD as an approximate physical
model representation denes the method of distributed heat sources described in
Reference 23. Typically, the only set of parameters optimized in application of this method
are the relative locations and magnitudes of the heat sources while the diffusivity and
speed V are assigned xed values corresponding to the average diffusivity of the material
and speed of the heat source relative to the workpiece, respectively. This approach is well
founded for cases where there is no quantitative downstream information concerning the
temperature eld. For these cases, the downstream character of the temperature eld is
completely determined by the quantities and V. For cases where quantitative downstream
information is available, however, assigning xed values to and V while adjusting only
values of the elemental heat sources q represents signicant restrictions with respect to the
exibility of using Equations AD to represent the temperature eld over a sufciently wide
spatial range. In addition, one should note that assigning values of the average diffusivity
and relative speed of the heat source to the parameters and V, respectively, is in actuality
only an approximation (see Reference 23 for discussion).
An inverse-problem approach which adopts Equations AD as either a 3-D interpolation
function or generating function where both C
1
and C
2
dened in Equation H are adjustable
parameters provides a relatively high level of exibility with respect to parameter
optimization. The calculated temperature eld is physically consistent if optimization is
constrained with respect to experimental information that is well distributed spatially,
both upstream and downstream, including solidication cross sections, thermocouple
measurements, and positioning of the energy source relative to the solidication boundary.
This denes the inverse-problem approach followed for the prototype analyses in this study.
In addition to the assignment of temperature-eld constraints based on direct measurements
(e.g., cross section and thermocouple measurements), constraint conditions can be assigned
based on correlation with microstructure (see Reference 16, for example).
An example of an inverse model representation based completely on the numerical
solution of the coupled equations of energy, momentum, and mass transport is given in
Reference 1. This inverse model, which uses the direct-problem model of Reference 25,
adopts the thermal coefcient of surface tension as an adjustable parameter. A comparison
of the analyses presented in References 1 and 5 indicates that they are similar, but based
on different inverse model representations of the temperature eld. The inverse model
applied here is that used in Reference 5.
(Continued from page 35.)

Vous aimerez peut-être aussi