Escolar Documentos
Profissional Documentos
Cultura Documentos
X
M
i=1
(f
i)
(C))
2
, (2)
where M is the total number of time steps. f
i)
(C) is the error
between thermomechanical FE analysis and the hot com-
pression test at time step i) or height reduction R
i)
.
In order to determine C, in the ow curve, the error
estimator (C) should be minimized. The vector C, which
minimizes the error estimator (C), can be solved iteratively
by the Lavenberg-Marquardt method,
30)
which is expressed
by
[J
T
(C
(k)
)J(C
(k)
) '
(k)
I]C
(k)
= J
T
(C
(k)
)f
i)
(C
(k)
), (3)
where (k) denotes the k-th iteration step in the minimization
procedure. Components of the Jacobi matrix J in eq. (3) can
be calculated by
J
io
=
of
i)
(C)
oC
o
. (4)
From eq. (4), it can be seen that the size of the Jacobi matrix J
is M N, where N is the total number of independent
parameters of the ow curve. I in eq. (3) is the unit matrix
with the size of N N. The components of the Jacobi matrix
are calculated by the numerical dierentiation of the error
function f
i)
(C) P
+i)
P
i)
(C) in thermomechanical FE
analysis.
We need to introduce constraint conditions in the
minimization procedure of the error estimator.
" oo = " oo(C, " cc,
_
" cc" cc, T) > 0 is a representative constraint condition
for example. For the case with a constraint condition
expressed by eq. (5), the modied error estimator can be
expressed by
k
[
= k
[
(C) > 0, (5)
+
(C) = (C)
X
B
[=1
oW
[
ok
[
(C)
, (6)
where k
[
is the [-th constraint function, and B is the number
of constraint functions. W
[
is the penalty function weight.
oW
[
,ok
[
(C) in eq. (6) is the inverse barrier function proposed
by Carroll.
31)
During the iterative procedure, the penalty
function weight W
[
is gradually reduced to zero, in order to
eliminate the eect of the constraint condition on the nal
solution of the ow curve. In this case, the modied
Lavenberg-Marquardt method
32)
can be applied to minimize
the error estimator with the constraint condition.
[J
T
(C
(k)
)J(C
(k)
) '
(k)
I H
(k)
]C
(k)
= J
T
(C
(k)
)f
i)
(C
(k)
) B
(k)
(7)
B
o
=
o
oC
o
X
B
[=1
W
[
k
[
(C)
!
(8)
H
o,
=
o
2
oC
o
oC
,
X
B
[=1
W
[
k
[
(C)
!
(9)
Fig. 1 General construction of proposed method.
2304 A. Yanagida, J. Liu and J. Yanagimoto
Here, 1 _ o _ N and 1 _ , _ N.
2.2 Thermomechanical FE analysis for the hot com-
pression test
A mesh system of thermomechanical FE analysis used in
inverse analysis is shown in Fig. 2. Thermal and deformation
analyses are repeated alternatively until reaching the nal
stroke in the hot compression test. The Lagrange multiplier
rigid-plastic nite element method is used in the deformation
analysis of the workpiece. Its variational principle is ex-
pressed by
Z
V
o
/
ij
o_ cc
ij
dV
Z
S
F
"
TT
i
o _ uu
i
dS
Z
V
'o_ cc
kk
dV
Z
V
o' _ cc
kk
dV = 0. (10)
Thermal analysis of the workpiece as well as the tool used for
upsetting is conducted, as shown in Fig. 2. The Arbitrary
Lagrangean Eularian (ALE) scheme is used in the thermal
analysis of the tool, because the radial coordinate of the
workpiece node at the interface moves due to the plastic ow
of the workpiece. Then, the following equation is applied in
thermal analysis.
,c
oT
ot
v
m
v
g
oT
or
=
K
r
r
o
2
T
or
2
r
o
2
T
oz
_
QQ (11)
Here, ,, c and K are the density, specic heat and thermal
conductivity, respectively. v
m
is the mass velocity of the tool
in the radial direction, and v
g
is the grid velocity of the tool in
the radial direction. r and z are radial and vertical
coordinates.
_
QQ is the heat generation due to plastic deforma-
tion and friction. Here in the thermal analysis of the tool, the
mass velocity v
m
is zero, and the grid velocity v
g
coincides
with the radial velocity of the workpiece v
r
at the tool-
workpiece interface. Then, eq. (11) yields following eq. (11
/
).
,c
oT
ot
v
r
oT
or
=
K
r
r
o
2
T
or
2
r
o
2
T
oz
_
QQ (11
/
)
It is assumed that all of energy dissipations are transferred to
heat generation, such as
_
QQ = o
_
cc. The Galerkin method and
the Crank-Nicholson dierential form are used to discretize
eq. (11
/
). Radiation and heat transfer to the atmosphere are
used as boundary conditions. They are expressed as eqs. (12)
and (13), respectively.
_ qq = H(T T
a
) (12)
_ qq = Es(T
4
T
a
4
) (13)
Here, H is the heat transfer coecient, E is the radiation
coecient, s is the Stephan-Boltzmann constant and is the
shape factor. The temperature of the surrounding atmosphere
T
a
is assumed to be 293 K. Parameters in thermal analysis
33)
are listed in Table 1. An example of thermal analysis for
obtaining initial temperature before upsetting is shown in
Fig. 3. The temperature of surface point A, which locates the
height center of workpiece z = 0.0, is the controlling
Fig. 2 Mesh system of tool and workpiece implemented in inverse analysis
of ow curve.
Table 1 Parameters used in thermal analysis of tool and workpiece.
Layer Workpiece Die Mica Scale
Material S20C Si
3
N
4
Mica FeO
Density
7800 3200 2100 4300
,/kg m
3
Specic heat capacity,
996 710 880 790
c/J kg
1
K
1
0.05
Thermal Conductivity,
23.6 29.3 1.22
(before hot compression test)
k/Wm
1
K
1
2.5
(during hot compression test)
Heat transfer coecient,
11.6 11.6 11.6 11.6
H/Wm
2
K
1
Thickness,
- - 200 30
/mm
Radiation coecient,
0.78
E
Shape facter,
1
Flow Curve Determination for Metal under Dynamic Recrystallization Using Inverse Analysis 2305
temperature in the hot compression test. An example of a heat
treatment curve in the hot compression test is shown in Fig. 4.
The normalizing time is 30[s] and the initial austenite grain
size after normalization is about 50 mm. In Fig. 3, the axial
temperature distribution at time t
I
, which is the end of
temperature holding, is shown. Induction heating is used to
elevate the temperature of the workpiece. Heat generation of
the workpiece by induction heating
_
QQ
I
is approximated by
_
QQ
I
=
_
QQ
I
(z) =
_
QQ
I
0
1
2z
h
10
!
. (14)
Here, h is the initial height of the workpiece, and z is an axial
position. The parameter
_
QQ
I
0
is controlled during the thermal
analysis of heating to monitor the measured temperature
change of point A before the hot compression test. From
Fig. 3, it is clear that temperature analysis is successfully
conducted with sucient accuracy, when we use the
parameters listed in Table 1. Needless to say, temperature
analysis is conducted continuously from heating until the end
of hot compression.
Comparison of the bulge proles between the experiment
and thermomechanical FE analysis revealed that they are in
good agreement throughout the analysis in this investigation,
when the friction coecient j = 0.2 is used. Thus, in
thermomechanical FE analysis in the inverse analysis of the
ow curve, j = 0.2 is used along with the thermal
parameters listed in Table 1.
2.3 Flow curve formulation of metal under dynamic
recrystallization
Various descriptions of ow curves are proposed and
implemented in inverse analysis,
2729)
as is described before.
Unfortunately, these formulations do not reect the eect of
dynamic recrystallization in a correct manner. For example,
the description of the ow curve using an internal state
variable
29)
does not express steady-state ow stress. Figure 5
shows the ow curve of metal under work hardening,
dynamic recovery and dynamic recrystallization. From
metallurgical consideration, the ow curve formulation
should be capable of describing its acute drop after the onset
of dynamic recrystallization. Also, it should be capable of
describing its steady-state value after a certain amount of
plastic strain, when deforming hot metal is under perfect
dynamic recrystallization. As a ow curve for the deforma-
tion analysis of metal forming, a Swift-type formulation of
the ow curve is desirable, because the ow stress is
Fig. 4 Heat treatment condition in hot compression test.
Fig. 5 Flow stress of metal under dynamic recrystallization.
Fig. 3 Example of thermal analysis and measured temperature before hot compression test.
2306 A. Yanagida, J. Liu and J. Yanagimoto
calculated explicitly as a function of strain, strain rate and
temperature. It is better to describe the ow curve by a simple
form, in which parameters C of the ow curve have physical
meanings. Here, the following ow curve formulation is
used.
" oo = F
1
" cc
n
( " cc _ c
c
)
" oo = F
2
exp[a( " cc c
max
)
2
] F
3
( " cc _ c
c
)
(15)
The ow curve expressed by eq. (15) shows its standard form
under isothermal condition and strain rate
_
" cc" cc = 1.0. c
c
in eq.
(15) denotes the critical strain for the onset of dynamic
recrystallization. When the equivalent strain is smaller than
c
c
, work hardening and dynamic recovery take place. A
simple power law is used to describe the ow curve of
smaller strain. c
max
is the strain when the ow stress reaches
its maximum value, and F
3
corresponds to the steady-state
stress o
s
when the deforming metal is under perfect dynamic
recrystallization. a, F
2
and c
max
are dependent parameters,
and they are calculated by continuity of the ow curve in
zero-order, rst-order and second-order derivatives at " cc = c
c
as
a =
nF
1
c
c
n1
2(c
c
c
max
)(F
1
c
c
n
F
3
)
, (16)
F
2
=
F
1
c
c
n
F
3
exp[a(c
c
c
max
)
2
]
, (17)
c
max
= c
c
F
1
c
c
n
F
3
nF
1
c
n1
c
(n 1)c
1
(F
1
c
n
c
F
3
)
. (18)
Here, c
c
, F
1
, n and F
3
are four independent parameters for the
standard description of the ow curve proposed in this
investigation. They are the components of parameter C.
The inverse analysis of the ow curve implemented in the
proposed method allows inhomogeneous distribution of
plastic deformation and temperature. It necessitates ow
stress which can describe its dependency on temperature and
strain rate in thermomechanical FE analysis, around their
standard values. Then, the modied form of the ow curve,
" oo
+
= " oo
_
" cc" cc
m
exp(A,T)
exp(A,T
0
)
, (19)
is used in the inverse analysis of the ow curve, after
substituting eqs. (15) to (19). T
0
is the reference temperature,
and the initial upsetting temperature is used as the reference
temperature. m is the rate sensitivity, and A is the temperature
sensitivity. It is assumed that m = 0.13 and
A = 3414[K
1
].
34)
The above mentioned m and A are
regarded as hypothetical values, and they could be modied
after ow curves are determined for dierent initial upsetting
temperatures T
0
and ram speeds.
2.4 Flow chart of proposed method
Figure 6 shows the ow chart of inverse analysis. The
convergence conditions are summarized in Table 2. The
proposed inverse analysis method fort the ow curve is
composed of a typical optimization problem of nonlinear
phenomena, thus the analysis with dierent initial guesses of
parameter C is conducted, and the converged solution of C
and its convergent status are carefully examined to ensure
that C is stationary. It is worth noting that one additional
external loop in the ow chart, which is drawn by a broken
line, is evident in Fig. 6. This external loop is for calculating
the rate sensitivity m and the temperature sensitivity A, by
using the ow curves obtained by the proposed method for
dierent initial temperatures and strain rates.
3. Results and Discussions
3.1 Experimental conditions
Table 3 shows the conditions of the hot compression test
and the chemical composition of the steel used for the
experiment. A THERMECMASTER Z machine is used in
the hot compression test. Standard initial specimen dimen-
sions are 8 mm diameter and 12 mm height, but two other
tests with dierent initial heights are added to check the
eect of geometry on the ow curve obtained by the
proposed method. Forming temperatures are 1373 K and
1173 K. Three strain rates, 1, 10 and 50 [s
1
], are tested. The
heat treatment conditions are shown in Fig. 4, for the forming
Fig. 6 Flow chart of proposed method.
Table 2 Convergence conditions.
C f
i)
(C)
Convergence 1
N
X
N
o=1
[C
o
(k)
[
C
o
(k)
_ 0.005
1
M
X
M
i=1
[f
i)
(C
(k)
)[
P
+i)
- 0.03
Conditions
Flow Curve Determination for Metal under Dynamic Recrystallization Using Inverse Analysis 2307
temperature of 1373 K. Thermomechanical FE analysis is
conducted immeduately after heating has started to estimate
the initial temperature eld of the tool and the workpiece
precisely until the end of the hot compression test.
3.2 Flow curve and its rate sensitivity
A typical curve for upsetting force versus height reduction
is shown in Fig. 7. Here, points i), which are used to estimate
the error between thermomechanical FE analysis and the hot
compression test, are shown by symbols (please see eq. (2)).
The upsetting force is evaluated in every 1% height
reduction, and its starting point is 2%. The upsetting forces
in FE analysis and the experiment are in good agreement,
although a rapid increase in upsetting force is observed at
several points due to folding of a free surface.
The ow curves determined by the proposed method are
shown in Fig. 8 and Fig. 9, for dierent forming temper-
atures. Flow curves shown in these gures represent those
under isothermal condition, which are obtained by inverse
analysis coupled with thermomechanical FE analysis pro-
posed in this paper. Flow curves obtained by Kadas
approach
14)
are also shown in these gures as references. It
is clearly shown that typical ow curves of metal under
dynamic recrystallization are obtained by the proposed
method, which cannot be obtained only by compensation
for an equivalent strain by FE analysis,
14)
where uctuation in
ow stress can be seen even at a larger strain when the metal
is under perfect dynamic recrystallization. At a glance, the
dierence between the ow stress values obtained by Kadas
approach and our proposed method may not be signicantly
large, but this dierence is crucial from the metallurgical
viewpoint. The ow stress should keep a steady-state value
when the deforming metal is under perfect dynamic
recrystallization, as shown in Fig. 5, but this steady-state
value could not be obtained by the conventional approach.
14)
Flow curves for dierent specimen sizes are shown in
Fig. 10. The ow curves are almost identical, thus we can
conclude that the proposed method does not include the size
eect, when the aspect ratio of the specimen h,D is larger
than 1.5. The steady-state stress for the aspect ratio of 1.0 is
slightly smaller than those for other aspect ratios, but its
relative dierence is less than 7%.
The parameters of the ow curves are summarized in
Table 4. Here, coecients of the ow curves for dierent
Table 3 Conditions of hot compression test and chemical composition of
workpiece.
Fig. 7 Transient change in upsetting force in experiment and thermo-
mechanical FE analysis.
Fig. 8 Flow curve of plain carbon steel at 1373 K.
Fig. 9 Flow curve of plain carbon steel at 1173 K.
Fig. 10 Flow curve obtained by dierent size of workpiece.
2308 A. Yanagida, J. Liu and J. Yanagimoto
strain rates and temperatures are listed. The critical strain for
the onset of dynamic recrystallization, c
c
, increases accord-
ing to the increase in strain rate.
26,35)
Senuma et al.
36)
reported
that the critical strain c
c
is only a function of the forming
temperature. From the results listed in Table 4, it can be
concluded that the critical strain c
c
is also dependent on strain
rate.
As indicated by eq. (19), the dependency of the ow curve
on strain rate is not included in the standard description of the
ow curve shown in eq. (15). If the rate sensitivity of m =
0.13 is correct, the independent parameters F
1
and F
3
should
be identical for the same initial temperature T
0
and dierent
strain rates. In other words, the rate sensitivity m could be
modied using discrepancies in F
1
and F
3
for dierent strain
rates, as shown by the external loop in Fig. 6. It is clear that
the rate sensitivity m = 0.13 is approximately true for higher
temperature T
0
= 1373 K, if we consider F
1
for the strain
rates
_
" cc" cc = 1, 10 and 50 [s
1
]. Also, F
3
for the strain rates
_
" cc" cc = 1, 10 and 50 is almost the same. Therefore, the accuracy
of the rate sensitivity m = 0.13 is conrmed.
3.3 Sensitivity of ow stress to temperature
The temperature sensitivity A could be estimated from
reliable F
1
and F
3
values at dierent forming temperatures
T
0
. If there are two values F
1
A
, F
1
B
and F
3
A
, F
3
B
for dierent
temperatures T
A
, T
B
and the same strain rate, their ratios can
be related by the following equations.
F
1
A
F
1
B
=
exp
A
T
A
exp
A
T
B
= exp A
1
T
A
1
T
B
!
(20)
F
3
A
F
3
B
=
exp
A
T
A
exp
A
T
B
= exp A
1
T
A
1
T
B
!
(21)
Since the parameters of the ow curve at the strain rate
_
" cc" cc = 1.0 are the most reliable, they are used to estimate the
temperature sensitivity A from presented results. Here, from
Table 4, F
1
A
= 115.67, F
3
A
= 66.59 for T
A
= 1373 K, and
F
1
B
= 180.81, F
3
B
= 96.60 for T
B
= 1173 K. Substituting
these parameters into eq. (20) using F
1
values, we obtain
A = 3598 [K
1
]. In the same manner, we obtain A = 2996
[K
1
] from eq. (21) using F
3
values. From both results, we
could see that the temperature sensitivity of plain carbon steel
used in a previous research study, A = 3414[K
1
],
34)
is
approximately correct. More important is the fact that the
temperature sensitivity A and the rate sensitivity m can be
easily estimated by the results obtained by our proposed
method. To evaluate these values more precisely, the external
loop, which is drawn by a broken line in Fig. 6, should be
used. Needless to say, it is not a dicult task to obtain these
values using our proposed method. Consistent investigations
for obtaining ow curves as a function of strain, strain rate
and temperature for metals with dierent alloy compositions
will be continued, on the basis of the analytical and
experimental methods proposed in this paper.
4. Conclusion
A new method for determining ow curves of hot bulk
metal is proposed, and it is applied to hot compression of
plain carbon steel under dynamic recrystallization. The
proposed method is eective even though there exist
inhomogeneous plastic deformation and temperature distri-
butions, which is inevitable in a hot compression test. The
ow curve of metal under dynamic recrystallization can be
easily obtained by the proposed method, as a function of
strain, strain rate and temperature. Also, the obtained
coecients of the ow curve have a physical meanings,
which correspond to the critical strain for the onset of
dynamic recrystallization and the steady-state stress for
example.
The obtained ow curve could be easily implemented in
FE analysis of bulk metal forming processes, as it is
formulated in the Swift-type form. Flow curve determination
for dierent metals under hot bulk forming is really
important, because it aects the accuracy of FE solutions
directly. Accumulated data on ow curves for various metals
will be one of the main components of the material genome
for realization of digital metal forming for the optimization of
the metal forming process and the development of an
innovative metal forming technology.
REFERENCES
1) A. Yoshie, T. Fujita, M. Fujioka, K. Okamoto and H. Morikawa: Testu
to Hagane 80 (1994) 38-43.
2) M. V. Rastegaev: Zavodskaya Laboratoriya 6 (1940) 354.
3) R. Kopp and F. D. Philipp: Steel Res. 63 (1992) 392-398.
4) R. Kopp, J. M. Heussen, F. D. Philipp and K. Karhausen: Steel Res. 64
(1993) 377-384.
5) R. Kopp, R. Luce, B. Leisten, M. Wolske, M. Tschirnich, T. Rehrmann
and R. Volles: Steel Res. 72 (2001) 394-401.
6) A. Laasraoui and J. J. Jonas: Metall. Trans. 22A (1991) 1545-1558.
7) O. Pawelski, U. Rudiger and R. Kaspar: Stahl Eisen 98 (1978) 181-189.
8) J. L. Uvira and J. J. Jonas: Trans. Metall. AIME 242 (1968) 1619-1626.
9) J. R. Douglas and T. Altan: Trans. ASME, J. Eng. Ind., 97 (1975) 67-
76.
10) GLEEBLE series brochure, Dynamic Systems Inc. Poestenkill NY,
12140.
11) S. Torizuka, T. Inoue and K. Nagai: Tetsu-to-Hagane 86 (2000) 801-
806.
12) K. Osakada, T. Kawasaki and K. Mori: Ann. CIRP 30 (1981) 135-138.
13) E. Parteder, R. Bunten: J. Mater. Proc. Tech. 74 (1998) 227-233.
Table 4 Determined parameters in ow curve.
Aspect
Strain Temperature
ratio, F
1
n c
c
c
max
F
3
Error
+
rate,
_
" cc" cc T
0
/K
h,d
1 1373 1.5 115.67 0.229 0.189 (0.271) 66.59 2.70%
10 1373 1.5 104.75 0.184 0.325 (0.491) 70.96 1.97%
10 1373 1.0 106.05 0.192 0.315 (0.495) 66.63 1.50%
10 1373 2.0 106.14 0.189 0.315 (0.472) 70.79 1.77%
50 1373 1.5 103.80 0.197 0.458 (0.700) 72.55 1.81%
1 1173 1.5 180.81 0.185 0.381 (0.666) 96.60 1.55%
10 1173 1.5 180.10 0.201 0.404 (0.705) 93.91 0.98%
50 1173 1.5 158.69 0.190 0.543 (0.949) 88.25 2.22%
+
Average error in upsetting force between analysis and experiment.
Flow Curve Determination for Metal under Dynamic Recrystallization Using Inverse Analysis 2309
14) O. Kada, T. Miki, M. Toda and K. Osakada: Trans. Jpn. Soc. Mech.
Eng. 66C (1998) 4914-4919.
15) P. Ludwik: Julius(Ed.): Element der Tech. Mech, (Springer, Berlin
1909) 32.
16) J. H. Hollmon: Trans. AIME 162 (1945) 268-290.
17) H. W. Swift: J. Mech. Phys. Solids 1 (1952) 1-18.
18) E. Voce: J. Inst. Metals. 74 (1948) 537-562.
19) Y. Bergstrom: Mater. Sci. Eng. 5 (1969-1970) 193.
20) Y. Bergstrom and B. Aronsson: Metall. Trans. 3 (1972) 1951-1957.
21) H. Mecking and U. Kocks: Acta. Metall. 29 (1981) 1865-1875.
22) Y. Estrin, and H. Mecking: Acta. Metall. 32 (1984) 57-70.
23) A. Yoshie, H. Morikawa, and A. Onoe: Trans. ISIJ 27 (1987) 425-431.
24) L. Kong, P. Hodgson and B. Wang: J. Mater. Proc. Technol. 89 (1999)
44-50.
25) J. H. Beynon, X. K. Li and A. R. S. Ponter: Mater. Sci. Forum 113-115
(1993) 293-298.
26) J. M. Cabrera, A. Al Omar, J. J. Jonas and J. M. Prado: Metal. Trans.
28A (1997) 2233-2245.
27) M. Michino, M. Tanaka and T. Kitaoka: J. Jpn. Soc. Technol. Plast. 37
(1996) 219-224.
28) M. Pietrzyk, J. Kusiak and J. G. Lenard: Proc. NUMIFORM95, eds,
S.-F Shen and P. Dawson, Ithaca, (1995) 277-282.
29) M. Pietrzyk, J. Kuisak, R. Kawalla and H. Pricher: J. Mater. Proc.
Technol. 60 (1996) 445-451.
30) D. W. Marquardt: J. Soc. Indust. Appl. Math. 11 (1963) 431-440.
31) C. W. Carroll: Operations Res., 9 (1961) 169-184.
32) V. Ravi and A. A. Jennings: Soil. Sci. Soc. Am. J. 54 (1990) 13-19.
33) Thermophysical properties hand book: (Youken-do, Tokyo, 1990) 187.
34) K. Misaka, T. Yoshimoto: J. Jpn. Soc. Technol. Plast. 8 (1967) 414-
422.
35) C. M. Sellers: Hot Working and Forming Process, C. M. Sellers and G.
J. Davies, eds., (The Metal Society London, 1980) 3-15.
36) T. Senuma, H. Yada, Y. Matsumura, T. Futamura: Tetsu to Hagane 70
(1984) 2112-2119.
2310 A. Yanagida, J. Liu and J. Yanagimoto