Engineering Materials
and Technology
Published Quarterly by The American Society of Mechanical Engineers
VOLUME 122 • NUMBER 1 • JANUARY 2000
Technical Papers
1 A Simple Model for Stable Cyclic StressStrain Relationship of Type 304 Stainless Steel
Under Nonproportional Loading
Takamoto Itoh, Xu Chen, Toshimitsu Nakagawa, and Masao Sakane
10 Representing the Effect of Crystallographic Texture on the Anisotropic Performance
Behavior of Rolled Aluminum Plate
M. P. Miller and N. R. Barton
18 Anisotropic Nonlinear Kinematic Hardening Rule Parameters From Reversed
Proportional AxialTorsional Cycling
J. C. Moosbrugger
29 Uniaxial Ratchetting of 316FR Steel at Room Temperature—Part I: Experiments
M. Mizuno, Y. Mima, M. AbdelKarim, and N. Ohno
35 Uniaxial Ratchetting of 316FR Steel at Room Temperature—Part II: Constitutive Modeling
and Simulation
N. Ohno and M. AbdelKarim
42 Cyclic Deformation Behavior and Dislocation Substructures of Hexagonal Zircaloy4
Under OutofPhase Loading
Xiao Lin
49 Models for Cyclic Ratchetting Plasticity—Integration and Calibration
Magnus Ekh, Anders Johansson, Hans Thorberntsson, and B. Lennart Josefson
56 Model of Grain Deformation Method for Evaluation of Creep Life in InService
Components
HyoJin Kim and JaeJin Jung
60 Modeling of Irradiation Embrittlement of Reactor Pressure Vessel Steels
S. Murakami, A. Miyazaki, and M. Mizuno
67 High Strain Extension of OpenCell Foams
N. J. Mills and A. Gilchrist
74 Inﬂuence of Heat Treatment on the Mechanical Properties and Damage Development in a
SiC/Ti153 MMC
David A. Miller and Dimitris C. Lagoudas
80 Failure of a Ductile Adhesive Layer Constrained by Hard Adherends
Toru Ikeda, Akira Yamashita, Deokbo Lee, and Noriyuki Miyazaki
86 Interfacial Stresses and Void Nucleation in Discontinuously Reinforced Composites
T. C. Tszeng
93 An Investigation of Yield Potentials In Superplastic Deformation
Marwan K. Khraisheh
98 Numerical Analysis of Welding Residual Stress and Its Veriﬁcation Using Neutron
Diffraction Measurement
Masahito Mochizuki, Makoto Hayashi, and Toshio Hattori
104 Effect of Laser Shock Processing (LSP) on the Fatigue Resistance of an Aluminum Alloy
Tang Yaxin, Zhang Yongkang, Zhang Hong, and Yu Chengye
108 Residual Stress Reduction and Fatigue Strength Improvement by Controlling Welding
Pass Sequences
Masahito Mochizuki, Toshio Hattori, and Kimiaki Nakakado
113 Consistent and Minimal Springback Using a Stepped Binder Force Trajectory and Neural
Network Control
Jian Cao, Brad Kinsey, and Sara A. Solla
This journal is printed on acidfree paper, which exceeds the ANSI Z39.48
1992 speciﬁcation for permanence of paper and library materials. ࠗؕ
TM
85% recycled content, including 10% postconsumer ﬁbers.
Technical Editor
DAVID L. MCDOWELL
GEORGE J. WENG
(Past Technical Editor)
Materials Division
Associate Technical Editors
R. BATRA (2001)
C. BRINSON (2002)
E. BUSSO (2001)
K. CHAN (1999)
N. CHANDRA (2000)
S. DATTA (2000)
G. JOHNSON (2001)
J. W. JU (2000)
S. MALL (2001)
D. MARQUIS (2001)
S. MEGUID (2001)
A. M. RAJENDRAN (2001)
G. RAVICHANDRAN (1999)
H. SEHITOGLU (2000)
E. WARNER (2002)
H. ZBIB (2000)
BOARD ON COMMUNICATIONS
Chairman and VicePresident
R. K. SHAH
OFFICERS OF THE ASME
President, R. E. NICKELL
Executive Director
D. L. BELDEN
Treasurer
J. A. MASON
PUBLISHING STAFF
Managing Director, Engineering
CHARLES W. BEARDSLEY
Director, Technical Publishing
PHILIP DI VIETRO
Managing Editor, Technical Publishing
CYNTHIA B. CLARK
Managing Editor, Transactions
CORNELIA MONAHAN
Production Assistant
MARISOL ANDINO
Transactions of the ASME, Journal of Engineering
Materials and Technology (ISSN 00944289) is published
quarterly (Jan., Apr., July, Oct.) for $205.00 per year by The
American Society of Mechanical Engineers, Three Park
Avenue, New York, NY 10016. Periodicals postage paid at
New York, NY and additional mailing ofﬁces. POSTMASTER:
Send address changes to Transactions of the ASME, Journal
of Engineering Materials and Technology, c/o THE AMERICAN
SOCIETY OF MECHANICAL ENGINEERS, 22 Law Drive, Box
2300, Fairﬁeld, NJ 070072300.
CHANGES OF ADDRESS must be received at Society
headquarters seven weeks before they are to be effective.
Please send old label and new address.
PRICES: To members, $40.00, annually: to nonmembers,
$205.00. Add $40.00 for postage to countries outside the
United States and Canada.
STATEMENT from ByLaws. The Society shall not be
responsible for statements or opinions advanced in papers or
. . . printed in its publications (B7.1, Par. 3).
COPYRIGHT © 2000 by The American Society of Mechanical
Engineers. For authorization to photocopy material for internal
or personal use under those circumstances not falling within
the fair use provisions of the Copyright Act, contact the
Copyright Clearance Center (CCC), 222 Rosewood Drive,
Danvers, MA 01923, tel: 9787508400, www.copyright.com.
Request for special permission or bulk copying should be
addressed to Reprints/Permission Department.
INDEXED by Applied Mechanics Reviews and Engineering
Information, Inc. Canadian Goods and Services Tax
Registration #126148048
Aluminum Tubes,” ASME JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY,
Vol. 105, pp. 242–249.
Taylor, G. I., 1938, “Plastic Strain in Metals,” J. Inst. Met., Vol. 62, pp. 307–324.
Wright, S. I., and Adams, B. L., 1990, “An Evaluation of the Single Orientation
Method for Texture Determination in Materials of Moderate Texture Strength,”
Textures and Microstructures, Vol. 12, pp. 65–76.
Wright, S. I., and Kocks, U. F., 1996, “A Comparison of Different Texture Analysis
Techniques,” Proceedings of the Eleventh International Conference on Textures of
Materials, Liang, Zuo, and Chu, eds., The Metallurgical Society, pp. 53–62.
Zouhal, N., Molinari, A., and Toth, L. S., 1996, “ElasticPlastic Effects During
Cyclic Loading as Predicted by the TaylorLin Model of Polycrystal Viscoplasticity,”
Int. J. Plas., Vol. 12, No. 3, pp. 343–360.
(Contents continued)
119 Cold Compaction of Composite Powders
K. T. Kim, J. H. Cho, and J. S. Kim
129 On the Limit of Surface Integrity of Alumina by DuctileMode Grinding
I. Zarudi and L. C. Zhang
135 Measurement of the Axial Residual Stresses Using the Initial Strain Approach
Weili Cheng
141 Effect of Plasma Cutting on the Fatigue Resistance of Fe510 D1 Steel
M. Chiarelli, A. Lanciotti, and M. Sacchi
Announcement
146 New Reference Format for Transactions Journals
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 17
Takamoto Itoh
Assistant Professor,
Department of Mechanical Engineering,
Fukui University,
91, Bunkyo 3chome, Fukui, 9108507, Japan
email: itoh@mech.fukuiu.ac.jp
Xu Chen
Professor,
Department of Chemical Engng Machinery,
Tianjin University,
Tianjin, 300072, P. R. China
Toshimitsu Nakagawa
Graduate Student.
Masao Sakane
Professor.
Department of Mechanical Engineering,
Ritsumeikan University,
111, Nojihigashi, Kusatsushi Shiga,
5258577, Japan
A Simple Model for Stable
Cyclic StressStrain
Relationship of Type 304
Stainless Steel Under
Nonproportional Loading
This paper proposes a simple twosurface model for cyclic incremental plasticity based on
combined Mroz and Ziegler kinematic hardening rules under nonproportional loading.
The model has only seven material constants and a nonproportional factor which
describes the degree of additional hardening. Cyclic loading experiments with fourteen
strain paths were conducted using Type 304 stainless steel. The simulation has shown that
the model was precise enough to calculate the stable cyclic stressstrain relationship
under nonproportional loadings.
1 Introduction
Many practical components such as pressure vessels and turbine
blades receive nonproportional damage under the combination of
thermal and mechanical loadings. Nonproportional cyclic loading
causes more damage than proportional loading and occasionally
reduces low cycle fatigue life. Type 304 stainless steel is known as
a typical material which shows signiﬁcant additional cyclic hard
ening under nonproportional loading (McDowell, 1983; Doong et
al., 1990; Itoh et al., 1995). Low cycle fatigue lives under non
proportional loading were drastically reduced by the additional
hardening depending on strain history. The maximum reduction in
life occurred by a factor of 10 compared with the proportional
straining (Socie, 1987; Itoh et al., 1995). On the other hand,
aluminum alloys which showed a small additional hardening under
nonproportional loading (Doong et al., 1990; Krempl and Lu,
1983; Itoh et al., 1992a, 1992b, 1997) gave a small reduction in life
(Doong et al., 1990; Itoh et al., 1995, 1997). The amount of
additional hardening also depends on material. Thus, the reduction
in low cycle fatigue life has a close connection with the additional
hardening depending on both loading history and material. Devel
opment of an accurate and convenient inelastic constitutive equa
tion has been needed for the quantitative estimates of additional
hardening.
Studies of plasticity have made great progress in these two
decades to express the cyclic deformation under complex mul
tiaxial loadings. Many inelastic constitutive equations were pro
posed for complex multiaxial loading (McDowell, 1985a, 1985b;
Krempl and Lu, 1984; Benallal and Marquis, 1987; McDowell,
1987; Chaboche and Nouailhas, 1989a, 1889b; Doong and Socie,
1991; Ohno and Wang, 1991) using a relatively large number of
material constants. The twosurface model has been recently re
ceiving attention since it provides the reasonable prediction for
nonproportional cyclic stressstrain response (Krieg, 1975;
Dafalias and Popov, 1976; Lamba and Sidebottom, 1978; McDow
ell, 1985b; McDowell, 1987; Tseng and Lee, 1987; Ellyin and Xia,
1989). One of the authors (Chen et al., 1996) also proposed the
twosurface model superposing Mroz and Ziegler kinematic hard
ening rules to simulate the mechanical ratchetting and the nonpro
portional cyclic deformation of 2014 aluminum alloy. These many
inelastic constitutive models insist the validity of predicting the
amount of additional hardening by comparing with a limited
number of experimental results. However, there are many factors
which inﬂuence the amount of additional hardening (Itoh et al.,
1995, 1997), like phase shift, rotation and change of principal
strain direction, step length etc., so that the validity of constitutive
model must be demonstrated by comparing the model analysis
with the nonproportional results including these factors. In this
meaning, there is no well demonstrated inelastic constitutive
model for additional hardening prediction.
The objective of this paper is to present an inelastic constitutive
model that can predict the stable cyclic nonproportional stress
response. The inelastic constitutive model proposed by Chen et al.
(1996) was improved by introducing the material parameter which
expresses the material dependence of additional hardening and by
the nonproportional factor which expresses the severity of non
proportional loading. The inelastic constitutive model was applied
to the extensive nonproportional data under 14 types of strain paths
and the applicability of the model was discussed.
2 Experimental Procedure of Nonproportional Load
ing
The material tested was Type 304 stainless steel which received
a solution treatment at 1373 K for one hour. Shape and dimensions
of the specimen were a thinwalled tube with 12 mm O.D., 9 mm
I.D. and 4.6 mm gage length, as shown in Fig. 1. The test apparatus
was a tensiontorsion servocontrolled electrichydraulic machine.
Axial and shear strains were measured on the specimen surface by
a tensiontorsion extensometer.
Multiaxial cyclic loading tests were carried out using 14
types of proportional and nonproportional strain histories at
room temperature. Figure 2 shows the 14 strain histories, where
⑀ and ␥ are the axial and shear strains, respectively, and
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division July
27, 1998; revised manuscript received February 5, 1999. Associate Technical Editor:
H. Sehitoglu.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 1
Copyright © 2000 by ASME
alphabets attached indicate representative points along the
strain path which will be referred in the later discussion. Case
0 is a pushpull test which is the basic test for predicting the
nonproportional cyclic stress and strain response. Strain paths
shown in the ﬁgure were determined so as to make clear the
various effects in nonproportional straining. In strain paths
1–13, the total axial strain range, ⌬⑀, had the same strain
magnitude as the total shear strain range, ⌬␥, on Mises’ equiv
alent basis. Two total strain ranges were employed in all the
tests except Case 0. They were ⌬⑀ ϭ 0.5% and 0.8%. In Case
0, the total strain range was varied from 0.5% to 1.5%. The
strain rate was 0.1%/s on Mises’ equivalent base. A detailed
description of the test procedure was presented in the previous
paper (Itoh et al., 1995) together with the nonproportional low
cycle fatigue lives.
3 Nonproportional Low Cycle Fatigue Strain Parame
ter
Nonproportional loading drastically reduces the low cycle
fatigue life accompanied with additional hardening (McDowell,
1983; Doong et al., 1990; Itoh et al., 1995, 1997). The reduction
in fatigue life is closely related to the additional hardening
under nonproportional loading, depending on the loading his
tory and material (Itoh et al., 1995, 1997). The authors proposed
the nonproportional strain parameter for predicting the low
cycle fatigue lives of Type 304 stainless steel and 6061 alumi
num alloy. The proposed parameter is expressed as (Itoh et al.,
1995, 1997)
⌬⑀
NP
ϭ ͑1 ϩ ␣f
NP
͒⌬⑀
I
(1)
where ⌬⑀
I
is the maximum principle strain range under nonpro
portional straining, ␣ the material constant which discriminates the
material dependency of additional hardening, and f
NP
the nonpro
portional factor which expresses the severity of nonproportional
loading.
The value of ␣ is deﬁned as the ratio of stress amplitude
under 90 degree outofphase loading (circular strain path in
␥/
͌
3 Ϫ ⑀ plot) to that under proportional loading. The 90
degree outofphase loading shows the maximum additional
hardening among all the nonproportional histories (Socie, 1987;
Itoh et al., 1995). For Type 304 stainless steel, the stress
amplitude under 90 degree outofphase loading was increased
up to 90% in comparison with the proportional loading, so the
value of ␣ takes 0.9.
The nonproportional factor which accounts for the severity of
nonproportional strain is calculated from strain history and is
deﬁned as
f
NP
ϭ
2T⑀
Imax
͵
0
T
͉͑sin ͑t͉͒⑀
I
͑t͒͒dt (2)
where ⑀
I
(t) and (t) are the absolute value of maximum principal
strain and the angle of maximum principal strain direction at time
Nomencl at ure
E ϭ Young’s modulus for elasticity
G ϭ shear modulus
c ϭ strain hardening modulus
n ϭ cyclic hardening exponent
K ϭ cyclic hardening coefﬁcient
K
NP
ϭ nonproportional cyclic hardening
coefﬁcient
C
s
ϭ material constant
␣ ϭ material constant for additional
hardening
f
NP
ϭ nonproportional factor
⑀, ␥ ϭ axial and shear strains, respec
tively
F ϭ function of yield surface
F* ϭ function of limit surface
R ϭ radius of yield surface
R* ϭ radius of limit surface
X
max
ϭ maximum back stress
1
,
3
ϭ effective axial and shear
stresses, respectively
⑀
1
, ⑀
3
ϭ effective axial and shear strains,
respectively
n
1
, n
3
ϭ orthogonal vectors in the stress
space
X ϭ vector pointing the center of
yield surface
ϭ stress vector
* ϭ stress vector on the limit surface
⑀ ϭ strain vector
d⑀
P
ϭ vector of plastic strain incre
ment
Fig. 1 Shape and dimensions of the specimen tested (mm)
Fig. 2 Proportional and nonproportional strain paths
2 / Vol. 122, JANUARY 2000 Transactions of the ASME
t, respectively. The former parameter shows the strain amplitude
and the latter the principal strain direction change under nonpro
portional loading. In the equation, T and ⑀
Imax
are the time for a
cycle and the maximum value of ⑀
I
(t) in a cycle. Detailed descrip
tion of the nonproportional low cycle fatigue parameter in Eq. (1)
is omitted here and the reader is referred to the previous paper
(Itoh et al., 1995) for details. The values of f
NP
for each case are
listed in Table 1.
The reason for making f
NP
an integral form is that the exper
imental results indicated that nonproportional low cycle fatigue
lives are signiﬁcantly inﬂuenced by the degree of principal
strain direction change and strain length after the direction
change. The value of f
NP
takes zero under proportional straining
such as Case 0 and 5 in which no change of principal strain
direction is occurred. f
NP
takes the maximum value of 1 under
the 90 degree outofphase straining which has the continuous
change of principal strain direction with the constant equivalent
strain corresponding to ⑀
Imax
, which results in the largest addi
tional hardening and the largest reduction in the low cycle
fatigue life (Socie, 1987; Itoh et al., 1995). However, when a
material has a small value of ␣, a small additional hardening
and a small reduction in fatigue life are occurred even though
the value of f
NP
is large (Itoh et al., 1997). Therefore, reduction
in nonproportional low cycle fatigue is connected with the
degree of additional hardening (Socie, 1987) depending on both
the loading history and materials. The material dependency of
the additional hardening can be explained by the mechanism of
slip systems.
In nonproportional loading, the principal strain direction is
changed with proceeding cycles, so the maximum shear stress
plane is changed continuously in a cycle. This causes an interac
tion between slip systems and which results in the formation of
small cells (Doong et al., 1990; Itoh et al., 1992a, 1992b; Kida et
al., 1997) for Type 304 stainless steel. Large additional hardening
occurred by the interaction of slip systems for that steel because of
low stacking fault energy. 6061 aluminum alloy, on the other hand,
is a material of high stacking fault energy and slips of dislocations
are wavy. No large interaction occurred in 6061 aluminum alloy
since dislocations change their glide planes easily following the
variation of the maximum principal strain direction (Itoh et al.,
1992a, 1992b; Kida et al., 1997).
The term, (1 ϩ ␣f
NP
), in Eq. (1) reﬂects the intensity of the
interaction between slip systems which depends on both the ma
terial and strain history (Kida et al., 1997). The equivalent strain
including (1 ϩ ␣f
NP
) in Eq. (1) gave a satisfactory correlation of
fatigue lives under nonproportional straining, and predicted the
additional hardening due to nonproportional loading (Itoh et al.,
1995, 1997). Therefore, it is considered that the term of (1 ϩ
␣f
NP
) becomes a good parameter to evaluate the effect of loading
history and of material dependency on stressstrain relationship
under complex nonproportional loadings. In the following, the
inelastic constitutive equation which includes (1 ϩ ␣f
NP
) is
developed.
4 Twosurface Plasticity Model Based on Kinematic
Hardening Rule
When a thinwalled tube specimen is subjected to combined
axial and torsional loading, the stress and strain state is expressed
in terms of deviatoric vector planes. The deﬁnition of the axial
torsional subspace follows as an Ilyushin’s ﬁvedimensional de
viatoric vector subspace and a stress vector is deﬁned as
ϭ
1
n
1
ϩ
3
n
3
(3)
In this equation,
1
and
3
are the effective axial and shear
stresses, respectively. Based on the Mises’ equivalent value ,
1
and
3
can be given by and
͌
3, where and are the axial and
torsional shear stresses, respectively. n
1
and n
3
are the orthogonal
base vectors in the stress space.
The strain vector is deﬁned as
⑀ ϭ ⑀
1
n
1
ϩ ⑀
3
n
3
(4)
Similarly, ⑀
1
is equal to the axial strain and ⑀
3
is equal to the shear
strain ␥/
͌
3.
Mises’ equivalent stress and strain are expressed by the follow
ing equations using
1
,
3
, ⑀
1
, and ⑀
3
.
e
ϭ ͱ
1
2
ϩ
3
2
⑀
e
ϭ ͱ⑀
1
2
ϩ ⑀
3
2
(5)
The total strain increment, d⑀, is assumed to be decomposed into
the elastic and plastic strain increments, d⑀
e
and d⑀
p
.
d⑀ ϭ d⑀
e
ϩ d⑀
p
(6)
where the elastic strain components are given by
d⑀
1
e
ϭ
d
1
E
, d⑀
3
e
ϭ
d
3
3G
(7)
E and G are Young’s and shear moduli, respectively.
Classical plasticity theory is based on the concept of the yield
surface. The movement of the yield surface under cyclic loading is
described by the kinematic hardening rule and the dimensional
change of the yield surface is described by the isotropic hardening
rule. Mises’ yield surface is expressed in the stress space as
F ϭ ͑ Ϫ X͒ ⅐ ͑ Ϫ X͒ Ϫ R
2
ϭ 0 (8)
Fig. 3 Twosurface model for superposing the two kinematic hardening
rules
Table 1 Values of the nonproportional factor
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 3
where X is a vector indicating the center of yield surface and R is
a radius of yield surface.
The limit surface, F*, in the stress space is required in the
twosurface model which is given as,
F* ϭ * ⅐ * Ϫ ͑R*͒
2
ϭ 0 (9)
where * is a limit stress vector and R* is the radius of limit
surface equated with the radius of yield surface and the maximum
back stress, X
max
.
R* ϭ ͉X
max
͉ ϩ R (10)
In order to describe the Bauschinger effect, a simple kinematic
hardening rule was proposed by Ziegler (1959) given by the next
equation:
dX ϭ d
Z
͑ Ϫ X͒. (11)
On the other hand, Mroz (1969) proposed a kinematic hardening
rule based on the concept of the nested yield and the loading
surface,
dX ϭ d
m
͑
*
m
Ϫ ͒. (12)
For the twosurface model,
*
m
is the stress vector on the limit
surface which has the same direction as the normal vector, n, at the
stressing point, , on the yield surface, i.e.,
*
m
and are located
at similar points on the yield and limit surfaces, respectively.
*
m
is deﬁned as shown in Fig. 3:
*
m
ϭ
R*
R
͑ Ϫ X͒. (13)
In this study, a new superposition of the kinematic hardening
rules, which is proposed by Chen and Abel (1996), is employed. It
is expressed as
Table 2 Material constants used in the analysis
Fig. 4(a) Case 0
Fig. 4(b) Case 4
Fig. 4(c) Case 7 Fig. 4(d ) Case 10
Fig. 4 Stable cyclic stressstrain relationship for Case 0, 4, 7, and 10
4 / Vol. 122, JANUARY 2000 Transactions of the ASME
dX ϭ d
s
͓͑
*
m
Ϫ ͒ ϩ C
s
͑ Ϫ X͔͒ (14)
This equation shows that the movement of yield surface under
cyclic loading complies with the direction determined by the
combination of Ziegler and Mroz hardening rules. The constant,
C
s
, is a function of plastic strain accumulation for transient
plasticity and ratchetting. Taking C
s
ϭ 0 in Eq. (14), the
proposed hardening rule corresponds with Mroz hardening rule
deﬁned in Eq. (12). By the consistency condition dF ϭ 0, d
s
is given by
d
s
ϭ
͑ Ϫ X͒ ⅐ d Ϫ R ⅐ dR
͑ Ϫ X͒ ⅐ ͑
*
m
Ϫ ͒ ϩ C
s
R
2
(15)
For stable cyclic stressstrain relations, the radius of yield surface
keeps the constant value, i.e., dR ϭ 0. Therefore, the movement
direction of yield surface can be expressed as
n
s
ϭ
␦
m
n
m
ϩ C
s
Rn
z
͉␦
m
n
m
ϩ C
s
Rn
z
͉
n
z
ϭ
Ϫ X
͉ Ϫ X͉
, n
m
ϭ
*
m
Ϫ
͉
*
m
Ϫ ͉
,
␦
m
ϭ ͱ͑
*
m
Ϫ ͒ ⅐ ͑
*
m
Ϫ ͒ (16)
The plastic ﬂow rule is given by Eq. (17) for dF ϭ 0 and ( Ϫ
X) ⅐ d Ͼ 0.
d⑀
p
ϭ
1
cR
2
͓͑ Ϫ X͒ ⅐ d͔͑ Ϫ X͒ (17)
where c denotes the strain hardening modulus. d⑀
p
is collinear
with ( Ϫ X) for Mises’ case. For Massing type materials, the
plastic modulus of the stressplastic strain curve (Ramberg
Osgood equation) can be expressed as (Ellyin and Xia, 1989; Lee
et al., 1995)
c ϭ Kn
ͩ
e
K
ͪ
͑nϪ1͒/n
. (18)
There are many ways to deﬁne the strain hardening modulus,
c, under nonproportional loading (McDowell, 1985b). In the
model of this paper, the strainhardening modulus is assumed to
be a function of the distance, ␦
s
, between the current stress
point on the yield surface, , and the stress point on limit
surface,
*
s
, i.e.,
Fig. 5 Stable cyclic stress response for Case 1–13 at ⌬⑀ ؍ 0.8% Fig. 5 (Continued)
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 5
␦
s
ϭ Ϫ ⅐ n
s
ϩ ͓͑ ⅐ n
s
͒
2
ϩ ͑R*͒
2
Ϫ ͉͉
2
͔
1/2
(19)
The maximum distance is
␦
max
ϭ 2͑R* Ϫ R͒. (20)
Let D be normalized distance between ␦
s
and ␦
max
.
D ϭ
␦
max
Ϫ ␦
s
␦
max
(21)
The normalized D varies between zero and one in value.
n and K which characterize the stable cyclic stressstrain must
be modiﬁed to take account of the additional hardening. This study
employed a simple idea,
K
NP
ϭ ͑1 ϩ ␣f
NP
͒K (22)
Therefore, the cyclic stable plastic modulus for nonproportional
loadings can be expressed by
c ϭ ͑1 ϩ ␣f
NP
͒Kn
ͩ
͉2͑R* Ϫ R͒͑D Ϫ 1͒ ϩ R*͉
͑1 ϩ ␣f
NP
͒K
ͪ
͑nϪ1͒/n
. (23)
5 Analytical Results and Discussion
Material constants used for the analysis are listed in Table 2;
Young’s modulus is 210 GPa, shear modulus 82 GPa, radius of
the yield surface 200 MPa, cyclic hardening coefﬁcient 1850
MPa, cyclic hardening exponent 0.29 and the material constant
for additional hardening 0.9 (Itoh et al., 1995).
The material constant C
s
, which is normally a function of plastic
strain accumulated for describing the transient plasticity and ratch
etting, must be determined. Under cyclic stable state, however, no
or small effect of C
s
was observed on the peak stress in stress
strain relationship. Since the stable stressstrain relationship would
be studied and C
s
is mainly related to the ratchetting rate (Chen et
al., 1996), C
s
can be taken as constant. Thus, this study takes the
value of C
s
as unity.
Figures 4(a)–(d) compare the stressstrain relationship between
analysis and experiment for Case 0 at ⌬⑀ ϭ 1.5% and Case 4, 7
and 10 at ⌬⑀ ϭ 0.8%. In these ﬁgures, alphabets indicate the
representative points along the strain path, Fig. 2, and the stress
strain curves, Figs. 4 and 5. In the uniaxial test of Case 0, Fig. 4(a),
the proposed model predicts the stable cyclic hysteresis loop with
satisfactory accuracy. Overall ﬁtting between the analysis and
experiment is excellent but there is a subtle difference during the
unloading stage. The analytical result has a linear shape but the
experimental result has a curved shape. Since the unloading stress
Fig. 5 (Continued) Fig. 5 (Continued)
6 / Vol. 122, JANUARY 2000 Transactions of the ASME
strain relationship must be a linear shape, the difference is pre
sumably resulted from the inaccuracy in experiment.
In Case 4, Fig. 4(b), the model developed in this study predicts
well the experimental stress response. Especially, the peak stresses
of predicted at AϳD agree well with these in experiment. The
shape of the stressstrain curves in the analysis is somewhat
different from that in experiment along the strain path, DA and
BC in the axial component and that of AB and CD in the shear
component. The model estimates the larger stress amplitude along
these strain paths.
In Case 7, Fig. 4(c), the overall ﬁtting of the stressstrain
relationship between the prediction and experiment is good. The
model has satisfactory accuracy for the axial component but there
exists a relatively large difference in the shear component along
the strain paths DE and CЈBЈ. The peak stresses derived by the
model precisely agree with those in experiment whereas the stress
strain curve shape differs between the model and experiment.
In Case 10, Fig. 4(d), the model estimates very close hysteresis
loop to the experimental result in the axial component but it
predicts slightly larger stress response than that in experiment for
the shear component.
From the comparison of the stress response between the analysis
and experiment shown in the above cases together with the other
strain paths which were not graphically presented, the overall
ﬁtting of the stress response is good between the model and
experiment. The axial component in the analysis quite well agreed
with those in experiment but there was a small difference in the
shear component. The model predicted larger stress amplitude than
the experiment for shear component. One of the possible causes of
the larger prediction is resulted from the stress/strain gradient into
the tube wall of specimen and the short gage length. The gage part
of the specimen has 1.5 mm in thickness and 6.4 mm in length. The
Fig. 5 (Continued)
Fig. 5 (Continued)
Fig. 5 (Continued)
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 7
thickness is slightly thicker and the gage length somewhat smaller
for the plasticity study. However, specimens were prone to buckle
under the severe nonproportional strain paths shown in Fig. 2 so
that this specimen geometry was necessary. The similar trend of
the data ﬁtting in the shear components was also found in the
literature (Socie, 1987).
Figure 5 compares the stress response in axialshear stress
diagram at ⌬⑀ ϭ 0.8%. In the ﬁgures of Case 8–12, simulated
results calculated with ␣ ϭ 0 and f
NP
ϭ 0 are shown by dotted
lines to be examine the effect of ␣ on the stress response. In
cruciform strain paths, Case 1 and 2 in Fig. 5, the model predicts
the very close stressstrain shape and estimates the accurate peak
stresses at the points of AϳD. However, there are some points
where the model gives the different stress response. Cases 3 and 4
have the hysteresis shape rotated about 45 degrees from that in
Cases 1 and 2, respectively.
Cases 5–7 have a similar path with the different number of
principal strain direction change steps. The model gives a good
agreement with the experimental results quantitatively at the re
spective strain path. The model follows the trend of the experi
mental results that the stress amplitude increases as the number of
steps decreases.
For the single step and box loading of Cases 8–13, the model
gives an accurate prediction for a simple boxy loading of Cases
10 and 12. The model also gives a good prediction for a double
box loading in Case 13. However, a small difference between
the prediction and experiment exists in Case 9 for axial com
ponent at C and shear component at A. This trend is also found
in Case 8 which is a similar strain path as Case 9. In the loading
path with phase shift, Case 11, the model predicts the larger
stress amplitude along DA and BC. For the case of ␣ ϭ 0, on
the other hand, there is a large difference in stress between the
analysis and experiment shown by the dotted lines for Cases
8–12. These ﬁgures clearly show that the large contribution of
␣ and f
NP
exists in the prediction of stable cyclic stress response
under nonproportional loading.
In conclusion, the model evaluates a relatively accurate stress
response under nonproportional loading except a few cases. The
stress amplitude calculated by the proposed model agrees well
with the experimental results for a wide variety of strain paths
whereas a small difference partly exists. The overall ﬁtting is
concluded to be satisfactory.
Figure 6 shows the comparison of stress range between analysis
and experiment. The axial stresses calculated by the model are
within a factor of 10% scatter band in comparison with experi
mental data in all the tests, Fig. 6(a). The scatter of data is slightly
larger in shear component and the scatter band is a factor of 1.2,
Fig. 6(b). The larger scatter in the shear component may be
resulted from the geometry of the specimen.
Figure 7(a) is a replot of Fig. 6 and shows the comparison of
the equivalent stress between analysis and experiment. The
ﬁgure shows that the simulation estimates the equivalent peak
stress within a factor of 1.2. In Fig. 7(b), where the calculation
was carried out under ␣ ϭ 0 and f
NP
ϭ 0, some of the data
under nonproportional loading are obviously underestimated by
Fig. 6(a) Axial stress range
Fig. 6(b) Shear stress range
Fig. 6 Comparison of axial and shear stress ranges between the anal
ysis and experiment
Fig. 7(a) ␣ ؍ 0.9
Fig. 7(b) ␣, f
NP
؍ 0
Fig. 7 Comparison of the calculated and experimental effective stress
amplitudes
8 / Vol. 122, JANUARY 2000 Transactions of the ASME
more than a factor of 1.2. The larger scatter of the data can be
seen under the larger f
NP
’s tests, such as Cases 10 and 12. The
maximum scatter was almost a factor of 2 which corresponds
with the value of ␣ for Type 304 stainless steel used in this
study. A contribution of (1 ϩ ␣f
NP
) to the simulation is very
large and the simple model proposed has enough precision to
calculate cyclic stable stressstrain relationship along widely
ranged nonproportional straining.
6 Conclusions
1. This study proposed a simple twosurface plasticity model
which has only seven material constants. This model is
based on a superposed Mroz’s and Ziegler’s kinematic
hardening rules for describing the stress and strain relation
ship under nonproportional strain loadings.
2. The term of (1 ϩ ␣f
NP
) which was proposed for the
nonproportional low cycle fatigue life prediction was suc
cessfully incorporated into the plasticity model to describe
the additional hardening quantitatively. ␣ and f
NP
are the
constants related to the material dependency of additional
hardening and the nonproportional factor.
3. The proposed model simulated the stable cyclic stressstrain
response under 14 types of proportional and nonpropor
tional strain histories with a satisfactory accuracy. The
model also predicts the stress amplitude of all the tests
within a 20% scatter band.
References
Benallal, A., and Marquis, D., 1987, “Constitutive Equations for Nonproportional
Cyclic Elastoiscoplasticity,” ASME JOURNAL OF ENGINEERING MATERIALS AND TECH
NOLOGY, Vol. 109, No. 4, pp. 326–336.
Chaboche, J. L., and Nouailhas, D., 1989a, “Constitutive Modeling of Ratchetting
Effects, Part I: Experimental Facts and Properties of the Classical Models,” ASME
JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 111, pp. 384–392.
Chaboche, J. L., and Nouailhas, D., 1989b, “Constitutive Modeling of Ratchetting
Effects, Part II: Possibilities of Some Additional Kinematic Rules,” ASME JOURNAL
OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 111, pp. 409–416.
Chen, X., and Abel, A., 1996, “A TwoSurface Model Describing Ratchetting
Behaviors and Transient Hardening Under Nonproportional Loading,” ACTA Me
chanics Sinica (English Series), Vol. 12, No. 4, pp. 368–376.
Dafalias, Y. F., and Popov, E. P., 1976, “Plastic Internal Variables Formalism of
Cyclic Plasticity,” ASME Journal of Applied Mechanics, Vol. 43, pp. 645–651.
Doong, S. H., Socie, D. F., and Robertson, I. M., 1990, “Dislocation Substructures
and Nonproportional Hardening,” ASME JOURNAL OF ENGINEERING MATERIALS AND
TECHNOLOGY, Vol. 112, No. 4, pp. 456–465.
Doong, S. H., and Socie, D. F., 1991, “Constitutive Modeling of Metals Under
Nonproportional Loading,” ASME JOURNAL OF ENGINEERING MATERIALS AND TECH
NOLOGY, Vol. 113, No. 1, pp. 23–30.
Ellyin, F., and Xia, Z., 1989, “A RateIndependent Constitutive Model for Tran
sient Nonproportional Loading,” J. Mech. Phys. Solids, Vol. 37, pp. 71–91.
Itoh, T., Sakane, M., Ohnami, M., and Ameyama, K., 1992a, “Effect of Stacking
Fault Energy on Cyclic Constitutive Relation Under Nonproportional Loading,” J.
Soc. Mat. Sci., Japan, Vol. 41, No. 468, pp. 1361–1367.
Itoh, T., Sakane, M., Ohnami, M., and Ameyama, K., 1992b, “Additional Hard
ening due to Nonproportional Cyclic Loading—Contribution of Stacking Fault
Energy—,” Proceedings of MECAMAT’92, International Seminar on Multiaxial
Plasticity, Cachan, France, pp. 43–50.
Itoh, T., Sakane, M., Ohnami, M., and Socie, D. F., 1995, “Nonproportional Low
Cycle Fatigue Criterion for Type 304 Stainless Steel,” ASME JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY, Vol. 117, pp. 285–292.
Itoh, T., Sakane, M., Ohnami, M., and Socie, D. F., 1997, “Nonproportional Low
Cycle Fatigue of 6061 Aluminium Alloy Under 14 Strain Paths,” Proceedings of 5th
International Conference on Biaxial/Multiaxial Fatigue and Fracture, Cracow, Po
land, Vol. I, pp. 173–187.
Kida, S., Itoh, T., Sakane, M., Ohnami, M., and Socie, D. F., 1997, “Dislocation
Structure and Nonproportional Hardening of Type 304 Stainless Steel,” Fatigue
Fract. Engng. Struct., Vol. 20, No. 10, pp. 1375–1386.
Krempl, E., and Lu, H., 1983, “Comparison of the Stress Response of an Aluminum
Alloy Tube to Proportional and Alternate Axial and Shear Strain Paths at Room
Temperature,” Mechanics of Materials, Vol. 2, pp. 183–192.
Krempl, E., and Lu, H., 1984, “The Hardening and RateDependent Behavior of
Fully Annealed AISI Type 304 Stainless steel Under InPhase and OutofPhase
Strain Cycling at Room Temperature,” ASME JOURNAL OF ENGINEERING MATERIALS
AND TECHNOLOGY, Vol. 106, pp. 376–382.
Krieg, R. D., 1975, “A Practical Two Surface Plasticity Theory,” ASME Journal
of Applied Mechanics, Vol. 28, pp. 641–646.
Lamba, H. S., and Sidebottom, O. M., 1978, “Cyclic Plasticity for Nonproportional
Paths: Part II, Comparison with Prediction of Three Incremental Plasticity Models,”
ASME JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 100, pp. 104–112.
Lee, Y. L., Chiang, Y. J., and Wong, H. H., 1995, “A Constitutive Model for
Estimating Multiaxial Notch Strains,” ASME JOURNAL OF ENGINEERING MATERIALS
AND TECHNOLOGY, Vol. 117, pp. 33–40.
McDowell, D. L., 1983, “On the Path Dependence of Transient Hardening and
Softening to Stable States Under Complex Biaxial Cyclic Loading,” Proc. Int. Conf.
on Constitutive Laws for Engng. Mater., Tucson, AZ, Desai and Gallagher, eds., pp.
125–135.
McDowell, D. L., 1985a, “A Two Surface model for Transient Nonproportional
Cyclic Plasticity: Part 1: Development of Appropriate Equations,” ASME Journal of
Applied Mechanics, Vol. 52, pp. 298–302.
McDowell, D. L., 1985b, “A Two Surface Model for Transient Nonproportional
Cyclic Plasticity: Part 2: Comparison of Theory with Experiments,” ASME Journal
of Applied Mechanics, Vol. 52, pp. 303–308.
McDowell, D. L., 1987, “An Evaluation of Recent Developments in Hardening and
Flow Rules for RateIndependent, Nonproportional Cyclic Plasticity,” ASME Journal
of Applied Mechanics, Vol. 54, pp. 323–334.
Mroz, Z., 1969, “An Attempt to Describe the Behavior of Metals Under Cyclic
Loading a More General Workhardening Model,” Acta. Mechanica, Vol. 7, pp.
199–212.
Ohno, N., and Wang, J. D., 1991, “Nonlinear Kinematic Hardening Rule: Propo
sition and Application to Ratchetting Problems,” Trans. SMiRT 11, Vol. L, Tokyo, pp.
481–486.
Socie, D. F., 1987, “Multiaxial Fatigue Damage Models,” ASME JOURNAL OF
ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 109, pp. 293–298.
Tseng, N. T., and Lee, G. C., 1987, “Simple Plasticity Model of TwoSurface
Type,” ASCE Journal of Engineering Mechanics, Vol. 109, pp. 795–810.
Ziegler, H., 1959, “A Modiﬁcation of Prager’s Hardening Rule,” Quarterly of
Applied Mechanics, Vol. 7, pp. 55–56.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 9
M. P. Miller
N. R. Barton
Sibley School of Mechanical and
Aerospace Engineering,
Cornell University,
Ithaca, NY 14853
Representing the Effect of
Crystallographic Texture on the
Anisotropic Performance
Behavior of Rolled Aluminum
Plate
Rolled aluminum alloys are known to be anisotropic due to their processing histories. This
paper focuses on measuring and modeling monotonic and cyclic strength anisotropies as well
as the associated anisotropy of the elastic/elasticplastic transition of a commercially
available rolled plate product. Monotonic tension tests were conducted on specimens in the
rolling plane of 25.4 mm thick AA 7075T6 plate taken at various angles to the rolling
direction (RD). Fullyreversed tension/compression cyclic experiments were also conducted.
As expected, we found signiﬁcant anisotropy in the backextrapolated yield strength. We also
found that the character of the elastic/elasticplastic transition (knee of the curve) to be
dependent on the orientation of the loading axis. The tests performed in RD and TD
(transverse direction) had relatively sharp transitions compared to the test data from other
orientations. We found the cyclic response of the material to reﬂect the monotonic anisotropy.
The material response reached cyclic stability in 10 cycles or less with very little cyclic
hardening or softening observed. For this reason, we focussed our modeling effort on
predicting the monotonic response. Reckoning that the primary source of anisotropy in the
rolled plate is the processinginduced crystallographic texture, we employed the
experimentallymeasured texture of the undeformed plate material in continuum slip poly
crystal plasticity model simulations of the monotonic experiments. Three types of simulations
were conducted, upper and lower bound analyses and a ﬁnite element calculation that
associates an element with each crystal in the aggregate. We found that all three analyses
predicted anisotropy of the backextrapolated yield strength and postyield behavior with
varying degrees of success in correlating the experimental data. In general, the upper and
lower bound models predicted larger and smaller differences in the backextrapolated yield
strength, respectively, than was observed in the data. The ﬁnite element results resembled
those of the upper bound when initially cubic elements were employed. We found that by
employing an element shape that was more consistent with typical rolling microstructure, we
were able to improve the ﬁnite element prediction signiﬁcantly. The anisotropy of the
elastic/elasticplastic transition predicted by each model was also different in character. The
lower bound predicted sharper transitions than the upper bound model, capturing the shape
of the knee for the RD and TD data but failing to capture the other orientations. In contrast,
the upper bound model predicted relatively long transitions for all orientations. As with the
upper bound, the FEM calculation predicted gentle transitions with less transition anisotropy
predicted than that of the upper bound.
1 Introduction
Crystallographic texture (hereafter referred to as texture)
evolves during the processing of metals and alloys. Large inelastic
deformations associated with metal forming tend to produce char
acteristic textures, as do recrystallization processes associated with
hot forming and thermal processing. Characterizing texture and
linking it to processing practices forms a main element of alloy
development. Often, a chief processing goal is to produce uniform,
isotropic alloys. The reality, however, is that rolling produces
alloys with varying degrees of heterogeneity and anisotropy. In
thin plate material primarily designed to be loaded inplane, a
uniaxial yield strength that varies with the angle of the tensile axis
to the rolling direction is a manifestation of the processinginduced
anisotropy. A varying material response through the thickness of
the plate is a result of rollinginduced heterogeneities. The physical
sources of the anisotropy depend on the processing schedule and
alloy. While dislocation structures can contribute to anisotropy in
coldrolled materials (cf. Juul Jensen and Hansen, 1990; Hansen
and Juul Jensen, 1992) as can the distribution of precipitates in
some heattreatable aluminum alloys (cf. Hosford and Zeisloft,
1972; Bate et al., 1981; Barlat et al., 1996), it is well accepted that
texture is the primary source of anisotropy in rolled aluminum
plate. Mechanical design with processed alloys, therefore, requires
one to link the texture to properties and performance. In this
process, one seeks to understand what constitutes a “good” or
“bad” texture for a particular application. This is most often
accomplished by recharacterizing the behavior of the alloy when a
processing change is instituted. Simulations that predict the per
formance of an alloy based on its processinginduced internal
structure, including texture, can save time and resources. While a
signiﬁcant amount of research has focused on simulating
processinginduced texture evolution to aid alloy development,
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
February 22, 1999; revised manuscript received June 15, 1999. Associate Technical
Editor: E. Werner.
10 / Vol. 122, JANUARY 2000 Transactions of the ASME Copyright © 2000 by ASME
simulations of alloy performance under small inelastic strain con
ditions, such as those encountered during cyclic loading, that
explicitly represent the plastic slip anisotropy associated with
texture are conspicuously absent from the literature. Employing an
analytic, anisotropic yield surface may enable one to capture
orientationdependent differences in the backextrapolated yield
strength, but cannot easily capture orientationdependent differ
ences in the elastic/elasticplastic transition.
The focus of this paper is the small strain, monotonic and cyclic
response of a rolled aluminum alloy, AA 7075T6. Monotonic
tension tests were conducted on specimens in the rolling plane
taken at various angles to the rolling direction (RD). Fully
reversed, tension/compression, cyclic experiments were also per
formed using the same specimen type. The experimentally
measured texture of the undeformed material and a continuum slip
polycrystal plasticity model were employed to simulate the mono
tonic experiments. Clearly, 7075T6 is a metallurgically compli
cated alloy and texture is only one of the relevant structural
features affecting its behavior. Crystallographic slip is the promi
nent inelastic deformation mechanism in the material and texture is
the major source of the experimentallyobserved elasticplastic
material behavior anisotropy, however, so our goal here is to
investigate the utility of a model that explicitly represents the
effects of the orientations of the crystals in an aggregate for
predicting the anisotropic response of the alloy.
In the following section we present a brief review of the relevant
literature related to measuring and modeling the anisotropy of
rolled plate. We also explore the use of continuum slip models for
simulating small strain elasticplastic deformations. In Section 3
we present our experimental methods and results. Section 4 con
tains the model development and the simulation results. Finally,
we close with discussion and conclusion sections. In this paper,
bold type indicates a tensor quantity of orders one through four.
Operations are as follows:
͑a b͒
ij
ϭ a
i
b
j
͑A ⅐ B͒
ij
ϭ A
ik
B
kj
A:B ϭ A
ij
B
ij
͑ᏸ:A͒
ij
ϭ ᏸ
ijkl
A
kl
ʈAʈ ϭ ͙A:A
2 Background
2.1 Aluminum Plate: Processing and Texture. Rolled AA
7075 (Al Zn Mg Cu) plate is employed in a broad array of
engineering applications, including critical airframe components
that can experience deleterious repeated loading conditions. It is
wellestablished that material property anisotropies arise in this
material due, primarily, to the processinginduced crystallographic
texture (cf. Hatch, 1984). Yield strength, fracture toughness, duc
tility, and fatigue resistance can vary signiﬁcantly as the specimen
loading axis is varied between the rolling direction (RD), the
direction transverse to the rolling direction in the plane of the plate
(TD) and the direction normal to the plane of the plate (ND). While
alloy development practices such as tailoring the volume fraction
and distribution of strengthening precipitates can somewhat miti
gate the strength anisotropy in some alloys (cf. Hosford and
Zeisloft, 1972; Bate et al., 1981; Barlat et al., 1996), the fact
remains that designing with an alloy such as the plate used in this
study requires one to understand the anisotropy of its properties.
The ﬁnal texture of the plate material is a product of its pro
cessing history. For heattreatable alloys such as AA 7075, the
nominal processing of a cast ingot begins with a sequence of
hotrolling and annealing episodes. Depending on the temper, this
sequence may include some early cross rolling passes where the
axis of the ingot is rotated 90°. Once the ﬁnal thickness is attained
the plate is typically stretched to normalize any residual strains,
then annealed and aged to bring out the precipitate structure from
which this material derives its strength. Certainly, small variations
in the processing schedule of the alloy can make large changes in
performance. We don’t know the history of the material investi
gated in this paper and present the above processing schedule as a
generic one for 7XXX alloys.
2.2 Models. Modeling the elasticplastic anisotropic behav
ior of textured alloys on the macroscale often involves an analyt
ical yield surface or ﬂow potential, which is employed to capture
anisotropy of the initial or backextrapolated yield strength (cf.
Hill, 1948; Stout et al., 1983; Harvey, 1985; Barlat, 1987; Lin et
al., 1993; Karaﬁllis and Boyce, 1993). Continuum slip polycrystal
models are alternatives. As we will show, a polycrystal model can
predict anisotropy of the elastic/elasticplastic transition as well as
initial yield.
The philosophy of a continuum slip model is to formulate the
problem at the slip system level and to use the geometry of the
crystal class to construct a “grain.” Inelastic deformation occurs as
the result of net dislocation motion, though the formulation is still
a continuum one in that individual dislocations are not treated
explicitly. The polycrystal response is attained by assembling a
number of grains using an intergranular constraint or mean ﬁeld
assumption (cf. Sachs, 1928; Taylor, 1938; Iwakuma and Nemat
Nasser, 1984; Asaro and Needleman, 1985; Kocks, 1987; Molinari
et al., 1987; Mathur and Dawson, 1989; Lipinski and Berveiller,
1989; Mathur et al., 1990; Cailletaud, 1992; Lebensohn and Tome,
1994; Zouhal et al., 1996; Feyel et al., 1997; Molinari et al., 1997),
or by assigning a ﬁnite element to each grain (cf. Harren and
Asaro, 1989; McHugh et al., 1989; Havlicek et al., 1992; Dawson
et al., 1994; Beaudoin et al., 1995a) or subgrain (Beaudoin et al.,
1995b; Mika and Dawson, 1998, 1999).
While these models have predominantly been used for simulat
ing texture evolution during large strain deformation episodes,
elasticplastic versions have been employed to simulate material
response during small strain cyclic deformation histories (cf. Cail
letaud, 1992; Zouhal et al., 1996; Feyel et al., 1997; Molinari et al.,
1997). Rather than examining cyclic plasticity in textured alloys,
however, these works have generally focused on developing slip
system hardening formulations to model complicated multiaxial
experiments on nontextured polycrystals. It has also been estab
lished that elasticplastic continuum slip models predict a strain
induced yield strength asymmetry or Bauschinger effect (Czyak et
al., 1961; Hutchinson, 1964a, b; Zouhal et al., 1996; Barton et al.,
1999) and a texturedependent elastic/elasticplastic transition
(Barton et al., 1999). We are deﬁning the elastic/elasticplastic
transition or knee of the curve to be the region of the stressstrain
response where the elastic and plastic strain rates are comparable
in magnitude.
Barton et al. (1999) showed that the contribution to the transi
tion behavior from the continuum slip model arises from the
interaction of the texture, the linking assumption, and the initial
residual stress state. Each crystal in the aggregate contributes to the
macroscopic knee through the activation of its individual slip
systems, which depend on the slip system strength and the orien
tation of the slip systems relative to the loading axes, and the
means by which the macroscopic conditions are related to the
crystal (linking assumption). For example, in the case of the upper
bound linking assumption, each crystal experiences the macro
scopic strain rate, producing a range of crystal stresses whose
breadth depends on the orientation distribution of the constituent
crystals. The duration of the transition or “knee length,” therefore,
is determined by the orientation distribution function (ODF) and
initial stress state and depends on the amount of strain necessary
for each crystal to achieve a stress state that will result in fully
developed plastic ﬂow. Single crystal knee lengths are often sev
eral times longer than the strain for initial yield, as initial yield
occurs when the ﬁrst slip system becomes active and substantially
more strain may be required to activate enough slip systems for
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 11
fully developed plasticity. Conversely, the lower bound linking
assumption assigns the macroscopic stress to each crystal. Thus
there is not a similar successive activation of slip systems in
individual crystals. An abrupt transition results as the slip systems
in the weakest crystals in the aggregate yield. A ﬁnite element
formulation that assigns each element an individual crystal orien
tation, predicts a knee duration that was closer to the upper bound
in appearance, though complicated somewhat by neighborhood
inﬂuences. Finally, Barton et al. (1999) showed that a difference in
knee length on reloading and reverse loading after prestrain arises
naturally from the inﬂuence of the residual stress state. It is also
worth noting (and we will show later) that the contributions to the
knee from the continuum slip model dominate any slip system
hardening effects over a comparable strain range for typical hard
ening models with parameters intended to capture behavior for
strains on the order of 0.05 to unity.
3 Experimental Methods and Results
The aluminum alloy investigated was AA 7075T6 25.4 mm
thick plate. We measured the texture of the plate at its centerplane,
producing pole ﬁgures using Xray diffraction. From the pole
ﬁgure data, we determined the orientation distribution function
(ODF) using the popLA (preferred orientation packageLos
Alamos) software (Kallend et al., 1991). The ODF is shown in Fig.
1 on slices taken normal to the ND direction through the Rodrigues
fundamental region of orientation space. Envisioning the rotation
associated with a lattice orientation when employing an angleaxis
description is extremely straightforward. The rotation axis is
described by a vector, whose magnitude is related to the angle of
rotation. These parameterizations of orientation space do not pro
duce the highlydistorted metric tensor and singularities associated
with Euler angles (Frank, 1988; Becker and Panchanadeeswaran,
1989). A point in the fundamental region (a truncated cube for
crystals with cubic symmetry) is an orientation, which, employing
the Rodrigues parameterization, is speciﬁed relative to the RD,
TD, ND frame by a rotation of angle  about an axis, n, by the
vector
r ϭ n tan
ͩ

2
ͪ
(1)
The texture, depicted in Fig. 1, is somewhat typical for rolled, heat
treated plate of this gauge. From the ODF, we randomly selected
a sufﬁcient number of representative orientations, which we em
ploy in the simulations described later.
We machined cylindrical specimens with a gage length of 19
mm and a diameter of 6.4 mm from the centerplane of the plate at
various angles, , to the rolling direction. All mechanical tests
were performed on a servohydraulic test machine operating in
true strain control employing an axial extensometer with a 12.75
mm gage length. The true strain rate for all experiments was 10
Ϫ3
s
Ϫ1
. We ﬁrst conducted tensile tests. The data from these experi
ments are shown in Fig. 2. From Fig. 2(a), we see that beyond the
elastic/elasticplastic transition, the stress strain curves attain a
roughly constant slope with the nearly identical curves from the
RD and TD tests having the highest stress value at the same strain
and the data from the ϭ65° experiment having the lowest value,
just slightly lower than the ϭ 45° data. Figure 2(b) zooms in on
the knee in the curve of the data. We see that the transition
response also is anisotropic, with RD and TD having abrupt
transitions compared to the data from other directions. The aniso
tropic nature of the transitions is best seen in Fig. 3, which depicts
the slope of the stressstrain curves, ⍜, as a function of accumu
lated strain. Each monotonic experiment was repeated at least
once, with stresses at the same strain differing less than 1% from
those depicted in the ﬁgure, in general.
We also conducted fullyreversed cyclic experiments using RD,
TD, and ϭ 45° specimens at strain amplitudes (⑀
a
) of 0.01,
0.0125, 0.015, 0.0175, and 0.02. Figure 4 depicts the stressstrain
response of the ﬁrst 5 cycles of the RD experiment for ⑀
a
ϭ 0.01.
As can be seen, the material rapidly attained cyclic stability,
exhibiting no signiﬁcant difference between the magnitude of the
Fig. 2 (a) The stress strain data from the tensile tests conducted on the
7075 plate. The angle, , is the angle between the specimen axis and the
rolling direction. (b) A closeup of data focusing on the elastic/elastic
plastic transition.
Fig. 1 The ODF at the
1
2
plane of the 7075 plate depicted on slices
through the fundamental region of orientation space parameterized with
the Rodrigues vector deﬁned in Eq. (1). The slices were taken normal to
the plate normal direction. An ODF value of 2.42 corresponds to a
random texture.
12 / Vol. 122, JANUARY 2000 Transactions of the ASME
peak tensile and compressive stress levels. The rapid attainment of
cyclic stability was characteristic of all experiments. Figure 5
shows a plot of strain amplitude versus stable stress amplitude
(
a
), commonly referred to as the cyclic stressstrain curve
(Mitchell, 1978). Again, the anisotropy of the cyclic response is
evident with the 45° data falling well below the RD and TD data.
In contrast to the uniaxial data, we see a slight difference in the RD
and TD stress amplitudes.
4 Model Development and Simulation Results
The model we employ is an elasticviscoplastic continuum slip
polycrystal model described generically in Section 2. For the
details of the model development, the reader is referred to Marin
and Dawson (1998) and Barton et al. (1999). We brieﬂy discuss
the issues most germane to the present work below.
4.1 Continuum Slip Model. The single crystal behavior is
multiplicatively decomposed into elastic and viscoplastic compo
nents with the constitutive relations formulated in an intermediate
conﬁguration. The shearing rates on all active slip systems com
bine to determine the inelastic rate of deformation and spin for
each crystal. A ratedependent ﬂow rule formulated on the slip
system level is employed, i.e.,
␥˙
͑␣͒
ϭ ␥˙
o
sgn ͑
͑␣͒
͒ͯ
͑␣͒
g
͑␣͒
ͯ
1/mf
(2)
where ␥˙
o
is a reference strain rate, m
f
is the rate sensitivity of the
ﬂow stress and g
(␣)
is the slip system yield strength or hardness.
For this work we employ a single hardness for each crystal (Miller
and Dawson, 1997) with the hardness of the I
th
crystal evolving as
g˙
͑I͒
ϭ h
oͫ
1 Ϫ
ͩ
g
͑I͒
g
s
ͪͬ
␣ϭ1
n
͉␥˙
͑␣͒
͉ (3)
where h
o
and g
s
are material parameters.
The internal variables of this model are the elastic strain tensor,
the slip system hardnesses, and the crystal orientations. The system
of equations is highly coupled and while the hardnesses evolve
according to Eq. (2), the evolution of the elastic strains and the
lattice orientations are embedded within the kinematic decompo
sition and constitutive relations (Barton et al., 1999).
We employ two methods for combining the behavior of many
single crystals into the polycrystalline response we seek. In the
ﬁrst, we use an aggregate of crystals to model material behavior at
a continuum point employing a mean ﬁeld assumption to relate the
crystal quantities to their counterparts on the macroscale. A vol
ume average is used to obtain macroscale quantities. We employ
the notation ͗٪͘ to denote the volume average of ٪ over all
crystals in the aggregate so that ͗͘ and ͗L͘ represent the volume
average of the Cauchy stress and velocity gradient, respectively.
We conducted simulations employing either the upper bound or
extended Taylor assumption, L ϭ͗L͘, or the lower bound assump
tion, ϭ ͗͘. These are material point calculations employing no
spatial discretization. The upper and lower bound simulations are
deformationdriven and are initiated by imposing a velocity gra
dient, ͗L͘, on the aggregate of crystals. We seek to model material
behavior during a tensile test, however, which involves a constant
axial straining rate and a stressfree condition on the lateral bound
ary of the specimen. To address this situation, we use a Newton
method to perturb and update ͗L͘ until the conditions on the
speciﬁed components of ͗͘ are satisﬁed, effectively enabling us
to conduct the material point simulation with mixed “boundary
conditions.” The other calculation employs a ﬁnite element for
mulation with a single crystal per element so that element level
(macroscopic) and crystal stress and deformation quantities are
identical (Marin and Dawson, 1998). Equilibrium is satisﬁed in the
usual ﬁnite element weak sense and compatibility is satisﬁed
identically by the chosen element shape functions. The elements
are initially cubes with continuous linear velocity interpolation and
discontinuous piecewise constant pressure interpolation. While a
cubic shape is clearly not representative of a typically alloy mi
crostructure, the initial focus here is on a realistic and
mechanicallysound representation of the intergrain mechanical
interactions. We will see that these interactions play a big role in
determining the character of the elastic/elasticplastic transition of
the aggregate. We will investigate the role of element shape in the
discussion section of the paper.
Fig. 4 The hysteresis loops for the ﬁrst 5 cycles of the fully reversed
cyclic experiment conducted on the 7075 plate in the rolling direction at
a ⑀
a
؍ 0.01
Fig. 5 The stress amplitude,
a
versus the strain amplitude, ⑀
a
, for the
cyclic experiments conducted on the 7075 plate
Fig. 3 The slopes of the stressstrain data, ⍜, in the transition region
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 13
4.2 Simulation Results. We extracted a representative ag
gregate of crystal orientations from the ODF. For the upper and
lower bound calculations we employed 1000 orientations. For the
ﬁnite element calculations we employed a 12 ϫ 12 ϫ 12 cube of
elements (1728 total), each with its own crystallographic orienta
tion as determined by the ODF. We determined the model param
eters for each linking assumption by correlating the model re
sponse to the TD tensile test data. We then ran simulations to
predict the material response in the other orientations. The param
eter values we determined are given in Table 1. In the present
study we have focussed on the monotonic behavior, since we saw
negligible cyclic hardening or softening in the cyclic response. In
this material, capturing the texturedependent elastic/elasticplastic
transition is of ﬁrst order importance to modeling the cyclic
behavior. We employed isotropic elasticity with E ϭ 69 GPa and
ϭ 0.3.
Figures 6, 7, and 8 depict the results for the upper bound, lower
bound, and FEM simulations, respectively. A simulation of the TD
experiment with slip system hardening neglected is also shown for
each simulation. The simulation results are discussed in the fol
lowing section.
5 Discussion
We divide the discussion of the simulation results into two parts,
the elastic/elasticplastic transition region of the stress strain curve
(⑀ Ͻ 1%) and the posttransition region.
5.1 PostTransition Behavior. Each model predicts anisot
ropy of the backextrapolated yield strength with various degrees
of success in correlating the experimental data. The upper bound
model correctly predicts the general trend but overpredicts the
magnitude of the anisotropy. The RD prediction is lower than TD
and the ϭ 45° stressstrain curve lies well below the data. The
lower bound simulations correctly predict the coincidence of the
RD and TD data, but place the stress levels of all the other tests on
roughly the same level, underpredicting the magnitude of the
anisotropy, in general. The ﬁnite element calculations are similar
in nature to the upper bound results, however, a smaller difference
is predicted between the RD and TD stress levels.
As previously mentioned, the elements employed in the FEM
simulations were initially cubic. Increased accuracy in the FEM
simulations was attained by using elements with an initial aspect
ratio more representative of the asrolled microstructure. As with
most rolled metals, we observed ﬂat grains, elongated in the rolling
direction. We reran the FEM calculations for the TD and RD tests
using elements with an aspect ratio of RD:TD:ND ϭ16:4:1. These
results, along with the previously depicted results using cubic
elements, are shown in Fig. 9. We see that the TD stress level
response decreases and the RD response increases, bringing the
curves closer to the coincidence we see in the data. However, as
discussed below, the sharpness of the knee is virtually unaffected
by the aspect ratio change.
Fig. 6 The simulations of the tensile experiments using the upper
bound linking assumption. Also depicted are the RD, TD, and ؍ 45°
data, along with a simulation of the TD experiment with the slip system
hardening neglected (g˙ ؍ 0 in Eq. (3)).
Table 1 Continuum slip plasticity model parameters
Parameter Upper bound Lower bound FEM
␥˙
o
(s
Ϫ1
) 1.0 1.0 1.0
m
f
0.05 0.05 0.05
h
o
350 1200 400
g
s
(MPa) 480 520 600
Initial g
(␣)
(MPa) 250 345 275
Fig. 7 The simulations of the tensile experiments using the lower bound
linking assumption. Also depicted are the RD, TD, and ؍ 45° data,
along with the simulation of the TD experiment with slip system harden
ing neglected (g˙ ؍ 0 in Eq. (3)).
Fig. 8 The simulations of the tensile experiments using the ﬁnite ele
ment formulation. Also depicted are the RD, TD, and ؍ 45°data, along
with the simulation of the TD experiment with slip system hardening
neglected (g˙ ؍ 0 in Eq. (3)).
Fig. 9 The ﬁnite element simulations of the RD and TD experiments
employing initial element aspect ratios of RD:TD:ND ؍ 1:1:1 (cube) and
RD:TD:ND ؍ 16:4:1
14 / Vol. 122, JANUARY 2000 Transactions of the ASME
5.2 Elastic/ElasticPlastic Transition. Mechanistically, the
elastic/elasticplastic transition represents an extremely compli
cated regime. Understanding the sources of transition anisotropy is
an even greater challenge. For histories involving small inelastic
strains (monotonic or cyclic), however, capturing the stress levels
within the transition is of primary importance. Typically, the
strainhardening model is “tuned” to capture the transition behav
ior. However, for the 7075 plate, the hardening model would need
to be anisotropic. It can also be seen in Figs. 6, 7, and 8 that with
the parameters given in Table 1, the slip system hardening has
virtually no effect on the macroscopic stressstrain curves in the
transition regime. For the continuum slip model, the character of
the elastic/elasticplastic transition, therefore, is governed by the
activation of the slip systems, the texture and the mean ﬁeld
assumption. The transitions predicted by each model are shown as
plots of ⍜ versus ⑀ in Figs. 10, 11, and 12. The FEM predictions
shown employed cubic elements. It can seen that, while slight
differences appear between the responses given by the upper
bound and ﬁnite element models, they both predict less sharp
transitions, similar to that of the ϭ 45° data. The spread in the
ﬁnite element predictions is, perhaps, smaller than that of the upper
bound model, especially near the end of the transition and the
ordering of the curves is more reminiscent of the one we see in Fig.
3. However, the sharpness seen in the RD and TD data is not
predicted. The lower bound model, on the other hand, predicts
sharp transitions similar to the RD and TD data for all orientations.
While the transition anisotropy is slight, the ordering of the curves
in the lower bound simulations is similar to that which we see in
the data.
By examining the transition anisotropy from the perspective of
lower and upper bound deformation behavior, possible physical
sources can be explored. One can hypothesize that the sharp RD
and TD transition behavior is more lower boundlike while the
behavior at intermediate values of is more like that of the upper
bound. As described by Kocks et al. (1998) there are plausible
upper and lower bound explanations, rooted in the intergranular
constraint condition, for yielding of a polycrystal. The lower
bound condition (each grain experiences the same stress state)
results in activation of only those systems that are most favorably
oriented for slip among the grains. Since an FCC crystal will, most
likely, have some slip systems so oriented, yielding occurs more or
less simultaneously within each grain and a sharp transition results
as the total cross section quickly yields. The upper bound condition
requires all grains to experience the macroscopic strain rate. The
orientation of some grains will result in stresses large enough to
cause extensive yielding, while others will continue to strain
elastically. Eventually, all grains will yield, but the resulting tran
sition to fully developed plasticity is a gradual one. In reality, the
deformation state experienced by a grain in an aggregate lies
between the upper and lower bound extremes. As shown in the
ﬁnite element results of Barton et al. (1999), the constraint offered
by the neighbors of an individual crystal strongly dictates the
magnitudes of the resulting crystal stress components. They saw
signiﬁcant spread in crystal stress when the neighborhood was
changed. The relative strengths of surrounding crystals will change
as the loading axis is changed. It is conceivable, therefore, that the
manner in which the crystals are placed in the aggregate would
affect the transition and its anisotropy. This may imply that the
ﬁnite element predictions would be improved by choosing crystal
orientations and locations within the aggregate in such a way that
both the ODF and the misorientation distribution function
(MODF) are matched. This will necessitate measuring lattice ori
entations at individual spatial points employing such techniques
such as the automated electron backscatter pattern (EBSP) method
in the SEM (cf. Wright and Adams, 1990; Baudin and Penelle,
1993; Wright and Kocks, 1996; Miller and Turner, 1999). Since
deformation heterogeneity plays such a key role in transition
behavior, a reﬁned mesh (multiple elements per crystal) may also
be necessary. Such a study is currently being undertaken by the
authors.
Bringing emphasis on the fact that the linking assumption can
have such a profound affect on the transition prediction is, perhaps,
as important as exploring the details of the individual model
predictions. This is especially pertinent since the elastic/elastic
plastic transition is typically modeled as a strainhardening phe
nomena, especially in macroscale models. In saturationrecovery
type hardening models, the sharpness of the knee depends on the
time constant employed in the hardening formulation. One can
increase that constant, or, as is common in macroscale cyclic
plasticity models, employ additional variables in order to improve
the transition prediction. As depicted in Figs. 6, 7, and 8, however,
the transition predicted by the elasticplastic continuum slip model
Fig. 10 The slopes of the stressstrain curve, ⍜, predicted by the upper
bound model. Also shown are the RD, TD and ؍ 45°data.
Fig. 11 The slopes of the stressstrain curve, ⍜, predicted by the lower
bound model. Also shown are the RD, TD and ؍ 45°data.
Fig. 12 The slopes of the stressstrain curve, ⍜, predicted by the ﬁnite
element model employing cubic elements. Also shown are the RD, TD
and ؍ 45°data.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 15
is independent of the slip system hardening formulation. Physi
cally, it seems reasonable that the anisotropy of the transition from
elastic to elasticplastic deformation behavior is more a product of
the initial activation of slip systems and related intergranular
constraint effects as of stress elevation due to anisotropic strain
hardening processes.
6 Conclusions
The AA 7075 plate material investigated in this work exhibited
little cyclic hardening, which implies that an accurate representa
tion of the monotonic behavior, especially in the small strain
regime, is of ﬁrst order importance to cyclic model development.
In our monotonic experiments, we saw anisotropy in both the
backextrapolated yield strength and the elastic/elasticplastic tran
sition. As expected, the upper and lower bound models over and
underpredicted the yield strength anisotropy, respectively. A ﬁnite
element model, which associates a single orientation with each
element, most accurately captured the yield strength anisotropy,
especially when an element aspect ratio that resembled that of the
rolled microstructure was employed. The transition behavior and
associated anisotropy, which dictate the character of the small
strain response, were not adequately captured by any of the mod
els. For all three mean ﬁeld assumptions, prediction of the transi
tion behavior was largely independent of the slip system hardening
formulation. Larger, more detailed ﬁnite element simulations of
the experiments may better represent the heterogeneity of slip
activation, and capture the transition anisotropy. Selecting crystal
orientations for the ﬁnite element calculation that would match the
MODF as well as the ODF may also be necessary. These calcu
lations are the focus of current work.
Acknowledgments
The authors acknowledge the Air Force Ofﬁce of Sponsored
Research (grant #F496209810401) and the National Science
Foundation (CAREER grant CMS #9702017) for supporting
this work. In addition, NRB gratefully acknowledges the sup
port of a National Science Foundation Graduate Research Fel
lowship Award. Edward Harley, Heshan Gunawardane and
Todd Turner conducted the mechanical tests and texture mea
surements and are gratefully acknowledged. Thanks also goes
to Professor Paul Dawson of Cornell University for helpful
advice and for the use of his simulation codes. Professor Ar
mand Beaudoin of the University of Illinois is acknowledged
for many helpful discussions.
References
Asaro, R. J., and Needleman, A., 1985, “Texture Development and Strain Hard
ening in Rate Dependent Polycrystals,” Acta Metall., Vol. 33, pp. 923–953.
Barlat, F., 1987, “Crystallographic Texture, Anisotropic Yield Surfaces and Form
ing Limits of Sheet Metals,” Materials Science and Engineering, Vol. 91, pp. 55–72.
Barlat, F., Liu, J., and Weiland, H., 1996, “On PrecipitateInduced Anisotropy
Modeling in Binary AlCu Alloys,” Mat. Sci. Forum, Vols. 217–222, pp. 635–640.
Barton, N., Dawson, P. R., and Miller, M. P., 1999, “Yield Strength Asymmetry
Predictions from Polycrystal Elastoplasticity,” ASME JOURNAL OF ENGINEERING MA
TERIALS AND TECHNOLOGY, (in press).
Bate, P., Roberts, W. T., and Wilson, D. V., 1981, “The Plastic Anisotropy of
TwoPhase Aluminum AlloysI. Anisotropy in Unidirectional Deformation,” Acta
Met., Vol. 29, pp. 1797–1814.
Baudin, T., and Penelle, 1993, “Determination of the Total Texture Function from
Individual Orientation Measurements by Electron Backscattering Pattern,” Met.
Trans. A, Vol. 24A, pp. 2299–2311.
Beaudoin, A. J., Dawson, P. R., Mathur, K. K., and Kocks, U. F., 1995a, “A Hybrid
Finite Element Formulation for Polycrystal Plasticity with Consideration of Macro
structural and Microstructural Linking,” Int. J. Plasticity, Vol. 11, 501–521.
Beaudoin, A. J., Mecking, H., and Kocks, U. F., 1995b, “Development of local
shear bands and orientation gradients,” Simulation of Materials Processing: Theory,
Methods, and Applications, Shen and Dawson, eds., pp. 225–230, Balkema, Rotter
dam.
Becker, S., and Panchanadeeswaran, S., 1989, “Crystal Rotations Represented as
Rodriquez Vectors,” Textures and Microstructures, Vol. 10, pp. 167–194.
Cailletaud, G., 1992, “a Micromechanical Approach to Inelastic Behavior of
Metals,” Int. J. Plas., Vol. 8, pp. 55–73.
Czyzak, S. J., Bow, N., and Payne, H., 1961, “On the Tensile StressStrain Relation
and the Bauschinger Effect for Polycrystalline Materials From Taylor’s Model,” J.
Mech. Phys. Solids, Vol. 9, pp. 63–66.
Dawson, P. R., Beaudoin, A. J., and Mathur, K. K., 1994, “Finite Element
Modeling of Polycrystalline Solids,” Numerical Predictions of Deformation Pro
cesses and the Behaviour of Real Materials, Anderson et al., eds., Riso National
Laboratory, Roskilde, Denmark, pp. 33–43.
Feyel, F., Calloch, S., Marquis, D., and Cailletaud, G., 1997, “F. E. Computation
of a Triaxial Specimen Using a Polycrystalline Model,” Comp. Mat. Sci., Vol. 9, pp.
141–157.
Frank, F. C., 1988, “Orientation Mapping,” Met. Trans. A., Vol. 19A, pp. 403–408.
Hansen, N., and Juul Jensen, J., 1992, “Flow Stress Anisotropy Caused by Geo
metrically Necessary Boundaries,” Acta Metallurgica et Materiala, Vol. 40, pp.
3265–3275.
Harvey, S. J., 1985, “The Use of Anisotropic Yield Surfaces in Cyclic Plasticity,”
Multiaxial Fatigue, ASTM STP 853, ASTM, Philadelphia, PA, 1985, pp. 49–53.
Harren, S. V., and Asaro, R. J., 1989, “Nonuniform Deformations in Polycrystals
and Aspects of the Validity of the Taylor Theory,” J. Mech. Phys. Solids, Vol. 37, pp.
191–232.
Hatch, J. E. (ed.), 1984, Aluminum: Properties and Physical Metallurgy, American
Society for Metals, Metals Park, Ohio, pp. 376–377.
Havlicˇek, F., Tokuda, M., Hino, S., and Kratochvil, J., 1992, “Finite Element
Method Analysis of MicroMacro Transition in Polycrystalline Plasticity,” Int. J.
Plasticity, Vol. 8, pp. 477–499.
Hill, R., 1948, “A Theory of the Yield and Plastic Flow of Anisotropic Metals,”
Proc. Roy. Soc. (London), Series A193, pp. 281–297.
Hosford, W. F., and Zeisloft, R. H., 1972, “The Anisotropy of AgeHardened Al4
Pct Cu Single Crystals During PlainStrain Compression,” Met. Trans., Vol. 3, pp.
113–147.
Hutchinson, J. W., 1964a, “Plastic StressStrain Relations of FCC Polycrystalline
Metals Hardening According to Taylor’s Rule,” J. Mech. Phys. Solids, Vol. 12, pp.
11–24.
Hutchinson, J. W., 1964b, “Plastic Deformation of BCC Polycrystals,” J. Mech.
Phys. Solids, Vol. 12, pp. 25–33.
Iwakuma, T., and NematNasser, S., 1984, “Finite ElasticPlastic Deformation of
Polycrystalline Metals,” Proc. Royal Soc. (London), Series A394, pp. 87–119.
Juul Jensen, D., and Hansen, N., 1990, “Flow Stress Anisotropy in Aluminum,”
Acta Metallurgica et Materiala, Vol. 38, No. 8, 1369–1380.
Karaﬁllis, A. P., and Boyce, M. C., 1993, “General Anisotropic Yield Criterion
Using Bounds and a Transformation Weighting Tensor,” J. Mech. Phys. Sol., Vol. 41,
No. 12, pp. 1859–1886.
Kallend, J. S., Kocks, U. F., Rollett, A. D., and Wend, H., 1991, “Operational
Texture Analysis,” Mater. Sci. and Eng., Vol. A132, pp. 1–11.
Kocks, U. F., 1987, “Constitutive Behavior Based on Crystal Plasticity,” Uniﬁed
Constitutive Equations for Plastic Deformation and Creep of Engineering Alloys,
A. K. Miller, ed., pp. 1–88. Elsevier Applied Science.
Kocks, U. F., Tome´, C. N., and Wenk, H.R., 1998, Texture and Anisotropy,
Cambridge University Press, Cambridge, p. 365 ff.
Lebensohn, R. A., and Tome, C. N., 1994, “A SelfConsistent Viscoplastic Model:
Prediction of Rolling Textures of Anisotropic Polycrystals,” Mater. Sci. and Eng.,
Vol. A175, pp. 71–82.
Lin, H., and NayebHashemi, H., 1993, “Effects of Material Anisotropy on Cyclic
Deformation and Biaxial Fatigue Behavior of Al6061 T6,” Advances in Multiaxial
Fatigue, ASTM STP 191, McDowell and Ellis, eds., ASTM, Philadelphia, pp.
151–182.
Lipinski, P., and Berveiller, M., 1989, “Elastoplasticity of MicroInhomogeneous
Metals at Large Strain,” Int. J. Plas., Vol. 5, pp. 149–172.
Marin, E. B., and Dawson, P. R., 1998, “On Modeling the ElastoViscoplastic
Response of Metals Using Polycrystal Plasticity,” Comp. Methods Appl. Mech. and
Eng., Vol. 165, pp. 1–21.
Mathur, K. K., and Dawson, P. R., 1989, “On Modeling the Development of
Crystallographic Texture in Bulk Forming Processes,” Int. J. Plas., Vol. 5, pp. 67–94.
Mathur, K. K., Dawson, P. R., and Kocks, U. F., 1990, “On modeling anisotropy
in deformation processes involving polycrystals with distorted grain shapes,” Mech.
Mater., Vol. 10, pp. 183–202.
McHugh, P. E., Varias, A. G., Asaro, R. J., and Shih, C. F., 1989, “Computational
Modeling of Microstructures,” Future Gen. Comp. Sys., Vol. 5, pp. 295–318.
Mika, D. P., and Dawson, P. R., 1998, “Effects of Grain Interaction on Deformation
in Polycrystals,” Mat. Sci. Engr., Vol. A257, pp. 62–76.
Mika, D. P., and Dawson, P. R., 1999, “Polycrystal Plasticity Modeling of In
tracrystalline Boundary Textures,” Acta Mat., in press.
Miller, M. P., and Dawson, P. R., 1997, “Inﬂuence of Slip System Hardening
Assumptions on Modeling Stress Dependence of Work Hardening,” Jour. Mech.
Phys. of Solids, Vol. 45, pp. 1781–1804.
Miller, M. P., and Turner, T. J., 1999, “Quantiﬁcation and Representation of
Crystallographic Texture Fields in Processed Alloys,” accepted for publication,
Int. J. Plas.
Mitchell, M. R., 1978, “Fundamentals of Modern Fatigue Analysis for Design,”
Fatigue and Microstructure, M. Meshi, ed., pp. 385–438.
Molinari, A., Canova, G. R., and Ahzi, S., 1987, “A Self Consistent Approach of
the Large Deformation Polycrystal Viscoplasticity,” Acta Metall., Vol. 35, No. 123,
pp. 2983–2994.
Molinari, A., Ahzi, S., and Kouddane, R., 1997, “On the SelfConsistent Modeling
of ElasticPlastic Behavior of Polycrystals,” Mech. of Mat., Vol. 26, pp. 43–62.
Sachs, G., 1928, “Zur Ableitung Einer Fliessbedingung,” A. Ver. dt. Ing., Vol. 12,
pp. 134–136.
Stout, M. G., Hecker, S. S., and Bourcier, R., 1983, “An Evaluation of
Anisotropic Effective StressStrain Criteria for the Yield and Flow of 2024
16 / Vol. 122, JANUARY 2000 Transactions of the ASME
Aluminum Tubes,” ASME JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY,
Vol. 105, pp. 242–249.
Taylor, G. I., 1938, “Plastic Strain in Metals,” J. Inst. Met., Vol. 62, pp. 307–324.
Wright, S. I., and Adams, B. L., 1990, “An Evaluation of the Single Orientation
Method for Texture Determination in Materials of Moderate Texture Strength,”
Textures and Microstructures, Vol. 12, pp. 65–76.
Wright, S. I., and Kocks, U. F., 1996, “A Comparison of Different Texture Analysis
Techniques,” Proceedings of the Eleventh International Conference on Textures of
Materials, Liang, Zuo, and Chu, eds., The Metallurgical Society, pp. 53–62.
Zouhal, N., Molinari, A., and Toth, L. S., 1996, “ElasticPlastic Effects During
Cyclic Loading as Predicted by the TaylorLin Model of Polycrystal Viscoplasticity,”
Int. J. Plas., Vol. 12, No. 3, pp. 343–360.
(Contents continued)
119 Cold Compaction of Composite Powders
K. T. Kim, J. H. Cho, and J. S. Kim
129 On the Limit of Surface Integrity of Alumina by DuctileMode Grinding
I. Zarudi and L. C. Zhang
135 Measurement of the Axial Residual Stresses Using the Initial Strain Approach
Weili Cheng
141 Effect of Plasma Cutting on the Fatigue Resistance of Fe510 D1 Steel
M. Chiarelli, A. Lanciotti, and M. Sacchi
Announcement
146 New Reference Format for Transactions Journals
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 17
J. C. Moosbrugger
Department of Mechanical and Aeronautical
Engineering,
Clarkson University,
Potsdam, NY, 136995725
Anisotropic Nonlinear Kinematic
Hardening Rule Parameters
From Reversed Proportional
AxialTorsional Cycling
A procedure for determining parameters for anisotropic forms of nonlinear kinematic
hardening rules for cyclic plasticity or viscoplasticity models is described. An earlier
reported methodology for determining parameters for isotropic forms of uncoupled,
superposed ArmstrongFrederick type kinematic hardening rules is extended. For this
exercise, the anisotropy of the kinematic hardening rules is restricted to transverse
isotropy or orthotropy. A limited number of parameters for such kinematic hardening
rules can be determined using reversed proportional tensiontorsion cycling of thin
walled tubular specimens. This is demonstrated using tests on type 304 stainlesssteel
specimens and results are compared to results based on the assumption of isotropic forms
of the kinematic hardening rules.
I Introduction
Superposition of multiple ArmstrongFrederick (AF) type ki
nematic hardening rules (Chaboche, DangVan and Cordier, 1979;
Chaboche, 1993) has become a popular approach for phenomeno
logical modeling of cyclic plasticity and viscoplasticity of metals.
The approach has been successfully incorporated into ﬁrstorder
thermodynamics frameworks (Chaboche, 1993; McDowell, 1992)
and used successfully to model responses to strain controlled
nonproportional loading histories (e.g., McDowell, 1985; Moos
brugger and McDowell, 1989; Moosbrugger, 1991, 1993) as well
as uniaxial cyclic loading histories. Recent modiﬁcations using
thresholding functions in the socalled “dynamic recovery term”
have enabled the modeling of ratchetting responses to both uniax
ial and multiaxial nonproportional loading (Chaboche 1989;
Chaboche and Nouailhas, 1989a, b; Bower, 1989; Chaboche,
1991; Ohno and Wang, 1991a, b, 1993a, b; Jiang and Sehitoglu,
1994a, b, 1996a, b; McDowell, 1995; Chaboche and Jung, 1997).
Introduction of anisotropic forms of nonlinear kinematic hardening
rules has been reported by Sutcu and Krempl (1990), Lee and
Krempl (1991), Nouailhas and Chaboche (1991), Nouailhas and
Culie (1991), and Nouailhas and Freed (1992), among others.
These anisotropic models have been applied primarily for the
description of cyclic viscoplasticity of single crystals and or metal
matrix composites. Recently, Chaboche and Jung (1997) discussed
superposed ArmstrongFrederick type kinematic hardening rules,
with ﬁxed anisotropy and with thresholding to correlate ratchetting
experiments, and showed their generalization in the framework of
quasistandard thermodynamics.
An earlier reported methodology for determining the parameters
for superposed, uncoupled AF type kinematic hardening rules
focused on parameters directly obtainable from hysteresis loops
and hardening modulus plots, obtained from experimental re
sponses to proportional strain controlled cycling (Moosbrugger
and Morrison, 1997). This technique however, was applied only to
isotropic forms of superposed AF rules. Such forms have a limited
capacity to correlate responses, even to proportional cyclic loading
histories, if there is signiﬁcant anisotropy present in the specimens.
Of course, one alternative is to adopt anisotropic yield functions or
ﬂow rules. However, such an approach only addresses initial
yielding behavior and plastic strain rate direction in unloading
loading events.
In this paper, the methodology for obtaining the AF rule pa
rameters is applied to determining a limited number of parameters
from tensiontorsion tests on tubular specimens, when the re
sponses exhibit distinctly anisotropic kinematic hardening behav
ior in the axial and shear components. This is seen to correspond
to either orthotropic or transverse isotropic symmetries using co
ordinate axes corresponding to specimen axial, circumferential,
and radial directions. Examples are given wherein the correlation
of responses is shown to be signiﬁcantly improved when param
eters for the axial and shear components of the AF rules are
determined separately, for tests on type 304 stainless steel speci
mens.
II Isotropic Framework for Determining AF Rule Pa
rameters
The forgoing applies to conventional associated, phenomeno
logical incompressible cyclic plasticity or viscoplasticity models
which adopt the approximations suitable for inﬁnitesimal defor
mations and rotations (c.f. Lubliner, 1990; Khan and Huang,
1995). We concentrate on the nonlinear kinematic hardening
model using superposed ArmstrongFrederick (AF) type kine
matic hardening rules originally proposed by Chaboche et al.
(1979) and employed and discussed in numerous works such as
those mentioned in the introduction. The determination of param
eters for thresholding functions (e.g., Ohno and Wang, 1993a, b;
McDowell, 1995; Jiang and Sehitoglu, 1996a, b; Chaboche and
Jung, 1997) is not discussed. Uppercase boldface symbols denote
second rank tensors (except in Section III where Q
i
and C
i
are
fourth rank). Double bars denote the norm ʈAʈ ϭ [A : A]
1/2
where
A : A ϭ A
ik
A
ki
, A
ik
denoting cartesian tensor components.
Summation on repeated subscripts is implied unless otherwise
noted. Lower case boldface symbols denote axialtorsional sub
space (c.f. McDowell, 1985) vectors and single bars their magni
tude, e.g., ͉a͉ ϭ [a ⅐ a]
1/ 2
ϭ [a
i
a
i
]
1/ 2
, a
i
being vector components
in the axialtorsional subspace. All nonbold symbols are scalars
and ͉x͉ denotes the absolute value of x. The subscript a denotes
amplitude of the subscripted quantity as in e
a
or ͉e
a
͉.
The isotropic form of superposed, uncoupled AF type kine
matic hardening rules can be written as
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
October 5, 1998; revised manuscript received July 14, 1999. Associate Technical
Editor: H. M. Zbib.
18 / Vol. 122, JANUARY 2000 Transactions of the ASME Copyright © 2000 by ASME
A
˙
Ј ϭ
iϭ1
M
A
˙
Ј
i
ϭ
iϭ1
M
C
i
D
i
q˙ (1)
with
D
i
ϭ B
i
N Ϫ AЈ
i
(2)
where q˙ ϭ ʈE
˙
p
ʈ and N ϭ (S Ϫ AЈ)/ʈS Ϫ AЈʈ ϭ E
˙
p
/ʈE
˙
p
ʈ; E
˙
p
is
the plastic or inelastic strain rate with cartesian components E
˙
rs
p
.
The tensor D
i
represents the distance in the deviatoric stress plane,
from the current state, to saturation of the ith backstress or kine
matic hardening subvariable where AЈ
i
ϭ A
i
Ϫ ( A
rr
/3)I is the
deviatoric part of A
i
. The “hardening modulus” (i.e., H ϭ S
˙
:
N/E
˙
p
: N) in the absence of cyclic hardening is then
H ϭ
iϭ1
M
C
i
D
i
: N (3)
For a more convenient presentation, the axialtorsional subspace
version of this framework (e.g., McDowell, 1985; Moosbrugger
and Morrison, 1997) can be written as
a˙ ϭ
iϭ1
M
a˙
i
ϭ
iϭ1
M
c
i
d
i
p˙ (4)
with
d
i
ϭ b
i
n Ϫ a
i
(5)
where p˙ ϭ
͌
2/3q˙ , n ϭ (s Ϫ a)/͉s Ϫ a͉ ϭ e˙
p
/p˙ ϭ e˙
p
/͉e˙
p
͉, e˙
p
ϭ
e˙
1
p
j
1
ϩ e˙
3
j
3
, s ϭ s
1
j
1
ϩ s
3
j
3
, a
i
ϭ a
i1
j
1
ϩ a
i3
j
3
, s
1
ϭ , s
3
ϭ
͌
3
, e˙
1
p
ϭ ⑀˙
p
, e˙
3
p
ϭ ␥˙
p
/͌
3
, and j
1
and j
3
form an orthonormal
basis for the two dimensional vector subspace; and are the
axial and shear stress, respectively, and ⑀˙
p
and ␥˙
p
are the axial and
engineering shear plastic strain rates, respectively. The subspace
hardening modulus (h ϭ s˙ ⅐ n/e˙
p
⅐ n) is then
h ϭ
iϭ1
M
c
i
d
i
⅐ n (6)
and the AF rule parameters for the subspace are c
i
ϭ
͌
3/ 2C
i
,
b
i
ϭ
͌
3/ 2B
i
with h ϭ 3H/ 2.
II Anisotropic Form of Superposed Armstrong
Frederick Type Kinematic Hardening Rules in Axial
Torsional Subspace
In a straightforward way, we can generalize Eqs. (4) and (5),
writing the rate of evolution of the subspace backstress in matrix
form as
ͫ
a˙
1
a˙
3
ͬ ϭ
iϭ1
M
ͫ
a˙
i1
a˙
i3
ͬ ϭ
iϭ1
M
ͫ
c
i1
0
0 c
i3
ͬͫ
d
i1
d
i3
ͬp˙ (7)
where
ͫ
d
i1
d
i3
ͬ ϭ ͫ
b
i1
0
0 b
i3
ͬͫ
n
1
n
3
ͬ Ϫ ͫ
a
i1
a
i3
ͬ (8)
The result is an anisotropic axialtorsional subspace version of the
framework, where the axial and shear components of the ith
backstress rate are uncoupled. However, they are not uncoupled
from the respective deformation rate components since n
1
, n
3
and
p˙ all depend on both ⑀˙
p
and ␥˙
p
. The isotropic form, Eqs. (4) and
(5), is recovered when c
i1
ϭ c
i3
and b
i1
ϭ b
i3
. With n deﬁned as
for the isotropic case, we tacitly assume isotropy of the ﬂow rule,
though more general anisotropic forms could be employed (e.g.,
Chaboche and Jung, 1997).
II.1 Parameter Determination Procedure. To determine
the parameters c
i1
, c
i3
, b
i1
, and b
i3
we follow a procedure analo
gous to that presented in (Moosbrugger and Morrison, 1997) for
the isotropic version of the AF rules. The following deﬁnitions are
made under the assumption of pure kinematic hardening for a
single cycle response to an imposed, completely reversed cyclic
loading:
h
1
ϭ
a˙
1
sgn ͑n
1
͒
p˙
ϭ ͯ
d
dp
ͯ , h
3
ϭ
a˙
3
sgn ͑n
3
͒
p˙
ϭ ͱ3ͯ
d
dp
ͯ (9)
Distances to saturation of axial and normalized shear stress can be
written as
d
1
ϭ
iϭ1
M
d
i1
ϭ s
1a
Ϫ s
1
sgn ͑n
1
͒ ϩ ⌬a
M1
,
d
3
ϭ
iϭ1
M
d
i3
ϭ s
3a
Ϫ s
3
sgn ͑n
3
͒ ϩ ⌬a
M3
(10)
where s
1a
ϭ
a
and s
3a
ϭ
͌
3
a
denote the components (axial and
normalized shear stress amplitudes) of the subspace stress vector
amplitude s
a
. The algebraic sign of the subspace plastic strain rate
vector components are denoted by sgn (n
1
) and sgn (n
3
).
If we order the kinematic hardening parameters such that c
i1
Ͼ c
i2
Ͼ. . . Ͼ c
iM
then, providing a
i
for i Ͻ M saturate prior to a reversal,
Eqs. (7)–(10) dictate linearity of h
1
versus d
1
and h
3
versus d
3
at low
h
1
, h
3
. The range of such linearity, if exhibited in the experimental
results, is determined in much the same way as the range of stress
strain employed to estimate Young’s modulus in a uniaxial test is
determined. Quantities ⌬a
M1
and ⌬a
M3
are the distances to saturation
of the axial and normalized shear stress, respectively, at the tips of the
axial and normalized shear stressstrain hysteresis loops. They can be
determined from plots of h
1
versus s
1a
Ϫ s
1
sgn (n
1
) and h
3
versus
s
3a
Ϫs
3
sgn (n
3
) as illustrated schematically in Fig. 1. Thus, d
1
and d
3
can be determined directly from the responses and derivatives of the
responses to the imposed cyclic loading. If the kinematic hardening
behavior is isotropic, the h
1
versus s
1a
Ϫ s
1
sgn (n
1
) and h
3
versus
Fig. 1 Schematic illustrating expected behavior of the axialtorsional
subspace hardening moduli for conformance to the anisotropic, super
posed AF kinematic hardening rules: h
1
versus (s
1a
؊ s
1
sgn (n
1
)) or d
1
and h
3
versus (s
3a
؊ s
3
sgn (n
1
)) or d
3
. The inset indicates the ideal
proportional loading response. Quantities taken from such plots for the
determination of AF rule parameters are also indicated.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 19
s
3a
Ϫ s
3
sgn (n
3
) plots (or h
1
versus d
1
and h
3
versus d
3
) should
coincide. Assuming that the backstress subvariables for i Ͻ M are
saturated (a˙
i
ϭ0 for i ϽM) at the cycle endpoints, c
M1
and c
M3
are the
slopes of the linear region, at low h
1
and h
3
, of the h
1
versus s
1a
Ϫ s
1
sgn (n
1
) and h
3
versus s
3a
Ϫ s
3
sgn (n
3
) plots. The conditions
⌬a
M1
ϩ a
M1a
ϭ b
M1
(11)
⌬a
M3
ϩ a
M3a
ϭ b
M3
(12)
also follow, where [a
M1a
a
M3a
]
T
is the Mth backstress amplitude.
Assumed symmetry of the hysteresis loop responses, with integra
tion of Eq. (7) for proportional cycling, leads to
2a
M1a
ϭ ͑a
M1a
ϩ b
M1
͒͑1 Ϫ e
ϪcM12pa
͒ (13)
2a
M3a
ϭ ͑a
M3a
ϩ b
M3
͒͑1 Ϫ e
ϪcM32pa
͒ (14)
where p
a
ϭ ͉e
a
p
͉ is the effective plastic strain amplitude. For M Ն
2 other equations necessary to determine parameters using this
procedure are
iϭ1
MϪ1
b
i1
ϭ s
1a
ϩ ⌬a
M1
Ϫ ͉s Ϫ a͉n
1
Ϫ b
M1
(15)
iϭ1
MϪ1
b
i3
ϭ s
3a
ϩ ⌬a
M3
Ϫ ͉s Ϫ a͉n
3
Ϫ b
M3
(16)
h
1max
ϭ
iϭ1
MϪ1
c
i1
2b
i1
͉n
1
͉ ϩ c
M1
d
M1max
(17)
h
3max
ϭ
iϭ1
MϪ1
c
i3
2b
i3
͉n
3
͉ ϩ c
M3
d
M3max
(18)
where
d
M1max
ϭ a
M1a
ϩ b
M1a
͉n
1
͉ (19)
d
M3max
ϭ a
M3a
ϩ b
M3a
͉n
3
͉ (20)
For M ϭ 2, the set of Eqs. (11)–(20) constitutes a linear system
sufﬁcient to determine the parameters c
11
, c
13
, b
i1
and b
i3
as well
as a
M1a
and a
M3a
, once c
21
and c
23
and ⌬a
M1
and ⌬a
M3
have been
determined from experimentally derived plots such as those illus
trated in Fig. 1. The stress amplitudes s
1a
and s
3a
, the plastic strain
rate vector components n
1
and n
3
, and the hardening moduli at the
onset of plastic deformation during a reversal (h
1max
and h
3max
) are
obtained from the experimental response to the proportional cy
cling.
For M Ն 3, additional relationships are required which will
result in a nonlinear algebraic system of equations in the unknown
parameters. An overconstrained approach would minimize an ob
jective function, such as sum of squared error between model and
experimental results. For simplicity, we will adopt here the ap
proach used for the isotropic case by Moosbrugger and Morrison
(1997), and determine parameters by matching the number of
experimental data points necessary to obtain the required addi
tional equations (to match the additional number of parameters for
M Ͼ 2, i.e., 4(M Ϫ 2) additional equations). Here, we match the
stress response [s
1k
s
3k
]
T
for a given reverse plastic strain p
k
, i.e.,
s
1k
ϭ Ϫa
M1a
sgn ͑n
1
͒ ϩ ͑a
M1a
sgn ͑n
1
͒ ϩ b
M1
n
1
͒͑1 Ϫ e
ϪcM1 pk
͒
ϩ
iϭ1
MϪ1
͓Ϫb
i1
n
1
ϩ 2b
i1
n
1
͑1 Ϫ e
Ϫci1 pk
͔͒ ϩ ͉s Ϫ a͉n
1
(21)
s
3k
ϭ Ϫa
M3a
sgn ͑n
3
͒ ϩ ͑a
M3a
sgn ͑n
3
͒ ϩ b
M3
n
3
͒͑1 Ϫ e
ϪcM3 pk
͒
ϩ
iϭ1
MϪ1
͓Ϫb
i3
n
3
ϩ 2b
i3
n
3
͑1 Ϫ e
Ϫci3 pk
͔͒ ϩ ͉s Ϫ a͉n
3
(22)
where k ϭ 1 . . . 2(M Ϫ 2) and the reverse plastic strain p
k
ϭ ͉e
k
p
Ϫ e
o
p
͉, e
o
p
being the subspace plastic strain vector at the reversal
point (e
a
p
or Ϫe
a
p
depending on the sign of n). Other possibilities
can be used such as matching hardening moduli or a combination
of stress and hardening moduli at discrete reverse plastic strain
points. Again, this notation tacitly assumes isotropy of the ﬂow
rule. However, ͉s Ϫ a͉n
1
and ͉s Ϫ a͉n
1
in Eqs. (15), (16), (21) and
(22) could be replaced with friction stresses, yield stresses or
thermally activated contributions to the axial and shear compo
nents of the ﬂow stress, which are determined from the experi
mental axial and shear responses, independently. The resulting
nonlinear system of equations must be solved using a suitable
procedure such as NewtonRaphson iteration.
II.2 Demonstration of the Procedure for AxialTorsional
Cycling of Type 304 StainlessSteel Specimens. Strain con
trolled tests were performed on thinwalled tubular specimens of
initially annealed type 304 stainless steel. Details can be found
elsewhere (Moosbrugger, 1988). Here, we report the application of
the described procedure on data taken from loading blocks wherein
the specimens were cycled between two symmetric strain end
points in axialtorsional subspace (i.e., reversed proportional cy
cling) at a constant total effective strain rate of ͉e˙͉ ϭ 10
Ϫ3
s
Ϫ1
. The
procedure is applied to data from two different effective total strain
amplitudes: ͉e
a
͉ ϭ
͌
⑀
a
2
ϩ ␥
a
2
/3 ϭ 6.0(10)
Ϫ3
with e
1a
ϭ ⑀
a
ϭ
4.5(10)
Ϫ3
and e
3a
ϭ ␥
a
/
͌
3 ϭ 3.9(10)
Ϫ3
and ͉e
a
͉ ϭ 7.9(10)
Ϫ3
with e
1a
ϭ 6.0(10)
Ϫ3
and e
3a
ϭ 5.2(10)
Ϫ3
; ⑀
a
and ␥
a
are the
imposed axial and engineering shear strain amplitudes, respec
tively. Experimental hardening moduli, etc. were determined using
the same procedures reported earlier (Moosbrugger and Morrison,
1997) for the isotropic case, but performed on the axial and shear
components of the experimental hysteresis loops separately when
using the anisotropic model. Isotropy of the ﬂow rule was invoked,
and ͉s Ϫ a͉ ϭ 80 MPa was estimated from previous analyses
involving backextrapolation of plastic strain rate directions before
and after nonproportional strain path direction changes (Moos
brugger, 1991). For the proportional cycling discussed here, n
1
and
n
3
were taken as the average of these quantities computed over the
individual cycles by numerically differentiating the response ver
sus time histories.
The experimental hardening moduli plots for ͉e
a
͉ ϭ 6.0(10)
Ϫ3
are shown in Fig. 2, with the full scale shown in Fig. 2(a) and an
expanded scale at low h
1
, h
3
shown in Fig. 2(b). Clearly, c
M1
c
M3
indicating the anisotropy of the kinematic hardening response
in these specimens. The straight lines shown in Fig. 2(b) are the
best ﬁt straight lines through the data at low h
1
, h
3
used to estimate
c
M1
, c
M3
, ⌬a
M1
and ⌬a
M3
. Using the procedure, the parameters
values determined were c
M1
ϭ 339, c
M3
ϭ 597, b
M1
ϭ 103 MPa
and b
M3
ϭ 104 MPa. For M ϭ 2, c
11
ϭ 7240, c
13
ϭ 15600,
b
11
ϭ 120 MPa and b
13
ϭ 90.7 MPa. For M ϭ 3, c
11
ϭ 13200,
c
13
ϭ 17700, c
21
ϭ 572, c
23
ϭ 407, b
11
ϭ 63.7 MPa, b
13
ϭ
80.1 MPa, b
21
ϭ 56.0 MPa and b
23
ϭ 10.7 MPa.
The correlation of the experimental results for ͉e
a
͉ ϭ 6.0(10
Ϫ3
)
using isotropic and anisotropic models is demonstrated in Figs.
3–6. Figure 3 shows the correlation of the hysteresis loops (s
1
ϭ
versus e
1
ϭ ⑀ and s
3
ϭ
͌
3 versus e
3
ϭ ␥/
͌
3) and the stress
subspace response, s
1
versus s
3
, using the isotropic model with
M ϭ 2. Correlation of the hardening moduli represented by h
1
versus d
1
and h
3
versus d
3
is shown in Fig. 4 for the isotropic
model with M ϭ 2. Parameters could not be determined for the
isotropic model with M ϭ 3 at this effective strain amplitude, as
was reported previously (Moosbrugger and Morrison, 1997). Cor
relations using anisotropic models with both M ϭ 2 and M ϭ 3
are shown in Figs. 5 and 6. It is interesting that parameters could
be determined for M ϭ 3 when the axial and shear components of
20 / Vol. 122, JANUARY 2000 Transactions of the ASME
the response were decoupled in this way (anisotropic case). A
larger “expanded range” is shown in Figs. 4(c) and 6(c) compared
to Fig. 2(b) to show the transition from linear to nonlinear behavior
of h
1
versus d
1
and h
3
versus d
3
at low h
1
, h
3
, and so that the
reader can see why the ordinate and abscissa ranges were chosen
in Fig. 2(b). It is seen in Figs. 4–6 that the best overall correlation
of the experimental response is achieved using the anisotropic
model with M ϭ 3. In comparison to the isotropic model, the
correlation of the normalized shear stressstrain hysteresis loop (s
3
versus e
3
) is signiﬁcantly improved, as is the s
1
versus s
3
response.
This is apparently due to the improved correlation of the hardening
moduli at low h
1
, h
3
, since the correlation at high values of the
moduli does not seem to be signiﬁcantly improved over the iso
tropic model (Fig. 4), at least in terms of h
3
versus d
3
.
Experimental hardening moduli plots for ͉e
a
͉ ϭ 7.9(10)
Ϫ3
are
shown in Fig. 7. Again it is seen that the h
1
versus d
1
and h
3
versus
d
3
plots do not coincide, even at low values of the moduli, leading
to c
M1
c
M3
and ⌬a
M1
⌬a
M3
. The straight lines through the
data in Fig. 7(b) are used to determine c
M1
, c
M3
, ⌬a
M1
and ⌬a
M3
.
With the procedure, the parameter values determined were c
M1
ϭ
253, c
M3
ϭ 413, b
M1
ϭ 73.4 MPa, and b
M3
ϭ 86.2 MPa. For
M ϭ 2, c
11
ϭ 4480, c
13
ϭ 4100, b
11
ϭ 131 MPa and b
13
ϭ
94.4 MPa. For M ϭ 3, c
11
ϭ 10400, c
13
ϭ 7100, c
21
ϭ 482,
c
23
ϭ 577, b
11
ϭ 69.4 MPa, b
13
ϭ 78.9 MPa, b
21
ϭ 102 MPa
and b
23
ϭ 67.2 MPa.
Correlation of the experimental hysteresis loops and the stress
subspace response for ͉e
a
͉ ϭ 7.9(10)
Ϫ3
, using isotropic and
anisotropic models, is summarized in Figs. 8–11. Figure 9 shows
the correlation of the hardening moduli using the isotropic model
with M ϭ 3. Again, a larger “expanded range” is shown in Figs.
9(c) and 11(c) compared to Fig. 4(b) to show the transition from
linear to nonlinear behavior of h
1
versus d
1
and h
3
versus d
3
at low
h
1
, h
3
, and so that the reader can see why the ordinate and abscissa
ranges were chosen in Fig. 4(b). The correlations achieved with the
Fig. 2 Experimental hardening moduli curves for a proportional cycle
obtained under total strain control at e
a
؍ 0.0045j
1
؉ 0.0039j
3
, ͦe
a
ͦ ؍
6.0(10
؊3
): (a) full scale and (b) expanded scale at low values of h
1
, h
3
showing the linear region used to determine c
M1
, c
M3
, ⌬a
M1
and ⌬a
M3
Fig. 3 Comparison of experimental response and correlation using M ؍
2 model with isotropic AF rules (M ؍ 2 ISOTROPIC MODEL) at e
a
؍
0.0045j
1
؉ 0.0039j
3
, ͦe
a
ͦ ؍ 6.0(10
؊3
): (a) axial, s
1
versus e
1
response, (b)
normalized shear, s
3
versus e
3
response and (c) stress subspace, s
1
versus s
3
response
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 21
anisotropic models using M ϭ 2 and M ϭ 3 are shown in Figs.
10 and 11. Again, the best correlation among these is achieved
with the anisotropic model using M ϭ 3. This is true for all of the
responses shown, for this effective strain amplitude and ratio of
axial strain amplitude to shear strain amplitude.
Limited independent investigation of the assumption of ﬂow
rule isotropy was obtained by computing friction stress
(KuhlmannWilsdorf and Laird, 1979) or yield stress components,
f 1
and
f 3
from the experimental axial and normalized shear
stressstrain hysteresis loops. A variation of Cottrell’s method
(Cottrell, 1953; KuhlmannWilsdorf and Laird, 1979) was used
with plastic strain offsets of e
1,offset
p
ϭ e
3,offset
p
ϭ 5(10
Ϫ6
), 1(10
Ϫ5
)
and 5(10
Ϫ5
) and the procedure was performed on both branches
(i.e., sgn (n
1
) ϭ sgn (n
3
) ϭ Ϯ1) of the hysteresis loops. Specif
ically, a least squares straight line was ﬁt to points ranging from
the peak stress prior to a reversal to one half of the peak stress
following a reversal. The slope of this line was taken as the elastic
modulus for a given branch of a given hysteresis loop. The friction
stresses for different plastic strain offsets were determined by
ﬁnding the stress corresponding to a given plastic strain offset
measured from the least squares line, in the direction of plastic
straining of the branch, using a running average of ﬁve points to
determine when an offset equal to or greater than a speciﬁed
plastic strain offset was achieved. One half of the absolute value of
Fig. 4 Comparison of experimental hardening moduli plots and exper
imental correlations using M ؍ 2 model with isotropic AF rules (M ؍ 2
ISOTROPIC MODEL) at e
a
؍ 0.0045j
1
؉ 0.0039j
3
, ͦe
a
ͦ ؍ 6.0(10
؊3
): (a) full
scale and (b) expanded scale at low values of h
1
, h
3
Fig. 5 Comparison of experimental response with correlations using
M ؍ 2 model with anisotropic AF rules (M ؍ 2 ANISOTROPIC MODEL)
and M ؍ 3 model with anisotropic AF rules (M ؍ 3 ANISOTROPIC
MODEL) at e
a
؍ 0.0045j
1
؉ 0.0039j
3
, ͦe
a
ͦ ؍ 6.0(10
؊3
): (a) axial, s
1
versus
e
1
response, (b) normalized shear, s
3
versus e
3
response and (c) stress
subspace, s
1
versus s
3
response
22 / Vol. 122, JANUARY 2000 Transactions of the ASME
the difference between the peak stress and the corresponding
plastic strain offset stress was then taken as the friction stress. The
smoothing effect of the straight line ﬁt, as well as the running ﬁve
point average used for the plastic strain offset, tended to average
out inconsistencies associated with ﬂuctuations about the
unloading/loading line due to limitations of analogtodigital data
conversion and any noise in the data. Based on the full scale range
of the extensometry used, the linearity of the extensometer signal
and the analogtodigital data conversion used, the minimum strain
resolution was estimated to be on the order of 10
Ϫ6
. This procedure
was performed on several hysteresis loops other than those used
for illustration in this paper and consistent results were obtained
for the three offsets, even for the smallest plastic strain offset
which is close to the estimated strain resolution limit. Table 1
compares the friction stresses with ͉s Ϫ a͉n
1
and ͉s Ϫ a͉n
3
and
compares the component ratios and the plastic strain amplitude
component ratios for the two effective strain amplitudes. The total
strain amplitude component ratios are e
1a
/e
3a
ϭ 1.15 in each case.
Except for the strain amplitudes, which are control parameters, the
ratios listed should be approximately equal for ﬂow rule isotropy.
Though the ratios deviate somewhat from isotropy in some cases,
indicating ﬂow rule anisotropy might improve correlation of the
data as well, ﬂow rule isotropy is indicated to be a reasonable
approximation. Note the relative agreement between the computed
offset friction stresses using the lower plastic strain offsets and ͉s
Ϫ a͉n
1
and ͉s Ϫ a͉n
3
, the latter having been estimated from strain
path direction changes (Moosbrugger, 1991).
III Relationship to Full Stress Space
We can generalize Eqs. (1) and (2) to an anisotropic form of the
nonlinear kinematic hardening rules by writing evolution of the ith
backstress as
A
˙
ϭ
iϭ1
M
A
˙
i
ϭ
iϭ1
M
͑Q
i
: E
˙
p
Ϫ C
i
: A
i
q˙ ͒ (23)
where the scalar kinematic hardening rule parameters are now
fourth rank tensors C
i
and Q
i
ϭ C
i
: B
i
where [C
i
: B
i
]
rsjl
ϭ
C
irsvw
B
ivwjl
(no sum on subscript i which pertains to the ith back
Fig. 6 Comparison of experimental hardening moduli plots and exper
imental correlations using M ؍ 2 model with anisotropic AF rules (M ؍
2 ANISOTROPIC MODEL) and M ؍ 3 model with anisotropic AF rules
(M ؍ 3 ANISOTROPIC MODEL) at e
a
؍ 0.0045j
1
؉ 0.0039j
3
, ͦe
a
ͦ ؍
6.0(10
؊3
): (a) full scale and (b) expanded scale at low values of h
1
, h
3
Fig. 7 Experimental hardening moduli curves for a proportional cycle
obtained under total strain control at e
a
؍ 0.0060j
1
؉ 0.0052j
3
, ͦe
a
ͦ ؍
7.9(10
؊3
): (a) full scale and (b) expanded scale at low values of h
1
, h
3
showing the linear region used to determine c
M1
, c
M3
, ⌬a
M1
and ⌬a
M3
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 23
stress). Note that the backstress evolution is now written in terms
of the total backstress rate, (i.e., not deviatoric). This is in principle
the form discussed, for example, by Chaboche and Jung (1997),
Nouailhas and Freed (1992), Nouailhas and Chaboche (1991),
Nouailhas and Culie (1991), among others.
Attaching the components of the various tensors to a particular
cartesian basis (coordinates x
1
, x
2
, x
3
) we can write these equa
tions in matrix form by deﬁning
͓E
˙
p
͔ ϭ
΄
E
˙
11
p
E
˙
22
p
E
˙
33
p
2E
˙
23
p
2E
˙
31
p
2E
˙
12
p
΅
, ͓A
i
͔ ϭ
΄
A
i11
A
i22
A
i33
A
i23
A
i31
A
i12
΅
, ͓A
i
͔ ϭ
΄
A
i11
A
i22
A
i33
2A
i23
2A
i31
2A
i12
΅
(24)
Then
͓A
˙
i
͔ ϭ ͓Q
i
͔͓E
˙
p
͔ Ϫ ͓C
i
͔͓A
i
͔q˙ (25)
If we suppose further that both Q
i
and C
i
exhibit orthotropic
symmetry, and the cartesian basis corresponds to the axes of
orthotropy, then we arrive at the following matrices with Voigt
constantlike entries:
Fig. 9 Comparison of experimental hardening moduli plots and exper
imental correlations using M ؍ 3 model with isotropic AF rules (M ؍ 2
ISOTROPIC MODEL) at e
a
؍ 0.0060j
1
؉ 0.0052j
3
, ͦe
a
ͦ ؍ 7.9(10
؊3
): (a) full
scale and (b) expanded scale at low values of h
1
, h
3
Fig. 8 Comparison of experimental response and correlation using M ؍
3 model with isotropic AF rules (M ؍ 3 ISOTROPIC MODEL) at e
a
؍
0.0060j
1
؉ 0.0052j
3
, ͦe
a
ͦ ؍ 7.9(10
؊3
): (a) axial, s
1
versus e
1
response, (b)
normalized shear, s
3
versus e
3
response and (c) stress subspace, s
1
versus s
3
response
24 / Vol. 122, JANUARY 2000 Transactions of the ASME
͓Q
i
͔ ϭ
΄
Q
i11
Q
i12
Q
i13
0 0 0
Q
i12
Q
i22
Q
i23
0 0 0
Q
i13
Q
i23
Q
i33
0 0 0
0 0 0 Q
i44
0 0
0 0 0 0 Q
i55
0
0 0 0 0 0 Q
i66
΅
(26)
and
͓C
i
͔ ϭ
΄
C
i11
C
i12
C
i13
0 0 0
C
i12
C
i22
C
i23
0 0 0
C
i13
C
i23
C
i33
0 0 0
0 0 0 C
i44
0 0
0 0 0 0 C
i55
0
0 0 0 0 0 C
i66
΅
(27)
We can then obtain the special cases of transverse isotropy and
isotropy; for transverse isotropy, with the coordinate x
1
normal to
the plane of isotropy, Q
i22
ϭ Q
i33
, Q
i13
ϭ Q
i12
, Q
i55
ϭ Q
i66
and
Q
i44
ϭ (Q
i22
Ϫ Q
i23
)/ 2; for isotropy Q
i11
ϭ Q
i22
ϭ Q
i33
ϭ
Qi
ϩ
2
Qi
, Q
i12
ϭ Q
i13
ϭ Q
i23
ϭ
Qi
with Q
i44
ϭ Q
i55
ϭ Q
i66
ϭ
(Q
i11
Ϫ Q
i12
)/ 2 ϭ
Qi
(the specializations for C
iJK
, J, K ϭ 1,
2 . . . 6 are analogous using
Ci
and
Ci
). The latter special
Fig. 10 Comparison of experimental response with correlations using
M ؍ 2 model with anisotropic AF rules (M ؍ 2 ANISOTROPIC MODEL)
and M ؍ 3 model with anisotropic AF rules (M ؍ 3 ANISOTROPIC
MODEL) at e
a
؍ 0.0060j
1
؉ 0.0052j
3
, ͦe
a
ͦ ؍ 7.9(10
؊3
): (a) axial, s
1
versus
e
1
response, (b) normalized shear, s
3
versus e
3
response and (c) stress
subspace, s
1
versus s
3
response.
Fig. 11 Comparison of experimental hardening moduli plots and exper
imental correlations using M ؍ 2 model with anisotropic AF rules (M ؍
2 ANISOTROPIC MODEL) and M ؍ 3 model with anisotropic AF rules
(M ؍ 3 ANISOTROPIC MODEL) at e
a
؍ 0.0060j
1
؉ 0.0052j
3
, ͦe
a
ͦ ؍
7.9(10
؊3
): (a) full scale and (b) expanded scale at low values of h
1
, h
3
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 25
(isotropy) case is equivalent to Eqs. (1) and (2) where 2
Qi
ϭ
C
i
B
i
, 2
Ci
ϭ C
i
and B
i
ϭ
Qi
/
Ci
since tr E
˙
p
ϭ E
˙
rr
p
ϭ 0.
Now consider axialtorsional loading of a thinwalled tubular
specimen. We attach the axes of orthotropy to the tube axial ( x
1
),
circumferential ( x
2
) and radial ( x
3
) directions, respectively. We
further deﬁne the plastic Poisson’s ratios
12
p
ϭ ϪE
˙
22
p
/E
˙
11
p
and
13
p
ϭ ϪE
˙
33
p
/E
˙
11
p
and the ratios
12
ϭ ϪAЈ
22
/AЈ
11
and
13
ϭ ϪAЈ
33
/AЈ
11
.
If the ﬂow rule is isotropic or transverse isotropic, then for incom
pressibility
12
p
ϭ
13
p
ϭ
12
ϭ
13
ϭ
1
2
. For orthotropic symmetry
of the kinematic hardening rules, the axial and shear components
of the backstress evolution equations obtainable from axial
torsional tests as described earlier would then be
A
˙
i11
ϭ Q
i,long
E
˙
11
p
Ϫ C
i,long
A
i11
q˙ (28)
A
˙
i12
ϭ Q
i,sh
E
˙
12
p
Ϫ C
i,sh
A
i12
q˙ (29)
with Q
i,long
ϭ Q
i11
Ϫ (Q
i12
ϩ Q
i13
)/ 2, Q
i,sh
ϭ 2Q
i66
, C
i,long
ϭ C
i11
and C
i,sh
ϭ 2C
i66
. In deviatoric components A
˙
Ј
i11
ϭ 2A
˙
i11
/3,
A
˙
Ј
i12
ϭ A
˙
i12
and, using the axialtorsional subspace vectors, a˙
i1
ϭ
A
˙
i11
ϭ 3A
˙
Ј
i11
/ 2, a˙
i3
ϭ
͌
3A
˙
i12
ϭ
͌
3A
˙
Ј
i12
. With N
11
ϭ
͌
2/3
n
1
ϭ E
˙
11
p
/q˙ and N
12
ϭ n
3
/
͌
2 ϭ E
˙
12
p
/q˙ we ﬁnd that
c
i1
b
i1
ϭ Q
i,long
, c
i3
b
i3
ϭ
3Q
i,sh
2
(30)
and
c
i1
ϭ ͱ
3
2
C
i,long
, c
i3
ϭ ͱ
3
2
C
i,sh
(31)
Thus, we can determine 4M parameters, i.e., Q
i66
, C
i11
, C
i66
, and
Q
i,long
, i ϭ 1 . . . M, using the proportional axialtorsional test.
There would, in the general orthotropic case, be 18M of these
parameters to determine. Though there would be less constants in
total to determine for transverse isotropic symmetry, as compared
to the orthotropic case, the two symmetries cannot be distinguished
using the axialtorsional test alone.
VI Discussion and Conclusions
The result, that the correlation of responses to proportional
axialtorsional straining of the tubular specimens can be improved
by adopting anisotropic forms of kinematic hardening rules, is
understandable if the specimens possessed texture. Though a thor
ough characterization, including crystallographic texture analysis
of the specimens is not available, they were machined from 50.8
mm (2 in.) diameter, extruded round bar stock before annealing. It
is likely that a duplex ͗100͘–͗111͘ ﬁber texture corresponding to
the specimen cylinder axes (extrusion direction) was retained (e.g.,
Barrett and Massalski, 1980), along with grain elongation in this
direction. Thus, a transverse isotropic symmetry about the speci
men axis (e.g., Kocks et al., 1998), in some deformation behaviors,
could be expected. The fact that the yield or friction stresses
corresponding to very small plastic strain offsets were, in contrast,
more or less isotropic is understandable since, in spite of texture,
there will always be a large number of grains oriented such that
many slip systems are stressed at or near the maximum possible
magnitude for a given state of stress. Upon reversal of strain rate
direction, therefore, some plastic strain rate should be detectable at
or near a common uniaxial friction stress, no matter what orien
tation is tested. Kinematic hardening on the other hand is the
result, at least to some extent, of the constraint of elastically
deforming grains on the ﬂow of plastically deforming grains.
Texture should impart an anisotropy to this phenomenon, since a
higher percentage of grains will have “harder” orientations with
respect to certain specimen directions than will have those same
orientations in an untextured specimen. In the case of the present
experiments, an apparent ͗100͘–͗111͘ ﬁber texture resulting from
extrusion led to less “elastic, perfectly plasticlike” hysteresis
loops for the axial response than for the normalized shear response.
Loop shape parameters (c.f. Mughrabi, 1978; Morrison et al.,
1990) were computed from the experimental responses as the ratio
of the area inscribed by the stressplastic strain hysteresis loops to
the area of the circumscribing parallelograms. These were
V
H,axial
ϭ 0.78 and V
H,shear
ϭ 0.92 for the axial and normalized
shear stressstrain hysteresis loops, respectively at ͉e
a
͉ ϭ
6.0(10
Ϫ3
). An elasticperfectlyplastic response would yield V
H
ϭ
1 and a purely elastic cycle would yield V
H
ϭ 0. For ͉e
a
͉ ϭ
7.9(10
Ϫ3
) they were V
H,axial
ϭ 0.79 and V
H,shear
ϭ 0.88 for the
axial and shear responses, respectively. Clearly, the normalized
shear stressstrain response appears to be less constrained than
does the axial component. The c
M3
Ͼ c
M1
result, in particular,
reﬂects this tendency. It should be mentioned, however, that the
experiments examined were at the same axial to shear strain
amplitude ratio though at two different effective total strain am
plitudes. A more complete assessment of the technique would
require cycling companion specimens at different ratios, to address
varying angles ϭ arctan (e
3a
/e
1a
) (including in particular ϭ 0,
Table 1 Experimental quantities indicating appropriateness of isotropic ﬂow rule assumption
Effective strain amplitude,
plastic strain offset, Cycle
Branch
f 1
(MPa)
f 3
(MPa)
͉s Ϫ a͉
n1
(MPa)
͉s Ϫ a͉n
3
(MPa)
f 1
/
f 3
n
1
/n
3
e
1a
p
/e
3a
p
e
a
ϭ 0.0045j
1
ϩ 0.0039j
3
͉e
a
͉ ϭ 6.0(10
Ϫ3
)
e
1,offset
p
ϭ e
3,offset
p
ϭ 5.0(10
Ϫ6
)
sgn (n
1
) ϭ sgn (n
3
) ϭ 1 54.7 58.8 60.6 51.7 0.93 1.11 1.10
sgn (n
1
) ϭ sgn (n
3
) ϭ Ϫ1 54.8 57.7 60.6 51.7 0.95 1.11 1.10
e
1,offset
p
ϭ e
3,offset
p
ϭ 1.0(10
Ϫ5
)
sgn (n
1
) ϭ sgn (n
3
) ϭ 1 61.8 66.1 60.6 51.7 0.93 1.11 1.10
sgn (n
1
) ϭ sgn (n
3
) ϭ Ϫ1 61.9 57.7 60.6 51.7 1.07 1.11 1.10
e
1,offset
p
ϭ e
3,offset
p
ϭ 5.0(10
Ϫ5
)
sgn (n
1
) ϭ sgn (n
3
) ϭ 1 86.0 90.1 60.6 51.7 0.95 1.11 1.10
sgn (n
1
) ϭ sgn (n
3
) ϭ Ϫ1 87.6 90.1 60.6 51.7 0.97 1.11 1.10
e
a
ϭ 0.006j
1
ϩ 0.0052j
3
͉e
a
͉ ϭ 7.9(10
Ϫ3
)
e
1,offset
p
ϭ e
3,offset
p
ϭ 5.0(10
Ϫ6
)
sgn (n
1
) ϭ sgn (n
3
) ϭ 1 57.8 49.9 59.5 53.4 1.16 1.17 1.14
sgn (n
1
) ϭ sgn (n
3
) ϭ Ϫ1 58.7 51.3 59.5 53.4 1.14 1.17 1.14
e
1,offset
p
ϭ e
3,offset
p
ϭ 1.0(10
Ϫ5
)
sgn (n
1
) ϭ sgn (n
3
) ϭ 1 64.1 56.5 59.5 53.4 1.13 1.17 1.14
sgn (n
1
) ϭ sgn (n
3
) ϭ Ϫ1 58.7 57.9 59.5 53.4 1.01 1.17 1.14
e
1,offset
p
ϭ e
3,offset
p
ϭ 5.0(10
Ϫ5
)
sgn (n
1
) ϭ sgn (n
3
) ϭ 1 81.9 80.3 59.5 53.4 1.01 1.17 1.14
sgn (n
1
) ϭ sgn (n
3
) ϭ Ϫ1 83.4 81.7 59.5 53.4 1.02 1.17 1.14
26 / Vol. 122, JANUARY 2000 Transactions of the ASME
/2 and 3/2), at constant effective strain amplitude. If the result
ing parameters were found to be relatively independent of , then
the interpretation of the kinematic hardening rule symmetries
would be more conclusive.
By the same texture argument, cyclic loading and, in particular,
nonproportional cyclic loading should have a tendency to diminish
such anisotropy of kinematic hardening behavior. As dislocation
substructures are developed which lead to cyclic hardening (e.g.,
McDowell et al., 1988; Doong et al., 1990), those more highly
deforming grains will tend to cyclically harden somewhat more
rapidly than those with “harder” orientations. Also, reﬁnement of
structure through subgrain cell formation should reduce the mor
phologic anisotropy associated with grain aspect ratio. Figure 12
shows c
M1
and c
M3
versus cycle number for the proportional test at
͉e
a
͉ ϭ 7.9(10
Ϫ3
). It is indicated that while both parameters
decrease as a function of cycle number, the difference between
them decreases very rapidly, more or less stabilizing after a few
cycles. For the same specimen, this proportional cycling was
followed by 25 cycles of nonproportional cycling with the same
component strain amplitudes but with the axial and shear strains 90
degrees out of phase. This was then followed by another 25 cycles
of proportional cycling with the same component strain ampli
tudes. Figure 13 shows the c
M1
and c
M3
determined from the
proportional cycling subsequent to the 90 degrees out of phase
cycling. The ratio c
M1
/c
M3
increased from 0.49 for the last of the
initial proportional cycles to 0.68 for the ﬁrst of the proportional
cycles subsequent to the 90 degree out of phase cycling. Likewise,
the ratio 2Q
M,long
/(3Q
M,sh
) increased from 0.67 to 1.01, for the
same cycles. Thus, the kinematic hardening behavior was signif
icantly more isotropic following the 90 degrees out of phase
cycling. It appears then that processes which tend to homogenize
the deformation, such as cyclic hardening, will diminish the an
isotropy of kinematic hardening, when textures are present.
In conclusion, a procedure has been presented which can be
used to determine parameters, using axialtorsional tests, for aniso
tropic, superposed ArmstrongFrederick type kinematic hardening
rules. The procedure is not sufﬁcient to determine all of the
required constants for transverse isotropic or for orthotropic kine
matic hardening behavior, but it is a convenient means of assessing
some of the possible parameters. Correlation of responses to pro
portional axialtorsional cycling of tubular type 304 stainless steel
specimens, machined from extruded bar stock, was improved
using the anisotropic kinematic hardening rules compared to those
obtained assuming isotropic kinematic hardening behavior. The
axialtorsional test, it would appear, may thus be one convenient
means of assessing the effect of texture producing processes on
subsequent cyclic plasticity behavior. The procedure should be
suitable for assessing such behaviors in other materials, provided
they exhibit anisotropic kinematic hardening behavior that is rea
sonably modeled by the anisotropic ArmstrongFrederick type
rules.
Acknowledgments
The author gratefully acknowledges the support of the U.S.
National Science Foundation through grant no. CMS9634707.
U.S. NSF grant MSM8601889, for support of the experiments on
304 stainless steel, is also gratefully acknowledged. Helpful dis
cussions with Professor D. J. Morrison of Clarkson University are
also acknowledged.
References
Armstrong, P. J., and Frederick, C. O., 1966, “A Mathematical Representation of
the Multiaxial Bauschinger Effect,” C.E.G.B. RD/B/N 731.
Barrett, C., and Massalski, T. B., 1980, Structure of Metals, 3rd revised edition:
Crystallographic Methods, Principles and Data, International Series on Materials
Science and Technology, Vol. 35, Pergamon, New York, pp. 543–544.
Bower, A. F., 1989, “Cyclic Hardening Properties of HardDrawn Copper and Rail
Steel,” Journal of the Mechanics and Physics of Solids, Vol. 37, pp. 455–470.
Chaboche, J. L., DangVan, K., and Cordier, G., 1979, “Modelization of the Strain
Memory Effect on the Cyclic Hardening of 316 Stainless Steel,” Proceedings of 5th
International Conference on Structural Mechanics in Reactor Technology, Berlin,
L11/3.
Chaboche, J. L., 1989, “A New Constitutive Framework to Describe Limited
Ratchetting Effects,” Proceedings of Plasticity ’89, Pergamon Press, p. 211.
Chaboche, J. L., and Nouailhas, D., 1989a, “Constitutive Modeling of Ratchetting
Effects—Part I: Experimental Facts and Properties of Classical Models,” ASME
JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 111, pp. 384–392.
Chaboche, J. L., and Nouailhas, D., 1989b, “Constitutive Modeling of Ratchetting
Effects—Part II: Possibilities of Some Additional Kinematic Rules,” ASME JOURNAL
OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 111, pp. 409–416.
Chaboche, J. L., 1991, “On Some Modiﬁcations of Kinematic Hardening to
Improve the Description of Ratchetting Effects,” International Journal of Plasticity,
Vol. 7, pp. 661–678.
Chaboche, J. L., 1993, “Cyclic Viscoplastic Constitutive Equations, Part I: A
Thermodynamically Consistent Formulation,” ASME Journal of Applied Mechanics,
Vol. 60, pp. 813–820.
Chaboche, J. L., and Jung, O., 1997, “Application of a Kinematic Hardening
Viscoplasticity Model with Thresholds to the Residual Stress Relaxation,” Interna
tional Journal of Plasticity, Vol. 13, pp. 785–807.
Cottrell, A. H., 1953, Dislocations and Plastic Flow in Crystals, Oxford University
Press, London, pp. 111–116.
Doong, S. H., Socie, D. F., and Robertson, I. M., 1990, “Dislocation Substructures
and Nonproportional Hardening,” ASME JOURNAL OF ENGINEERING MATERIALS AND
TECHNOLOGY, Vol. 112, pp. 456–463.
Jiang, Y., and Sehitoglu, H., 1994a, “Cyclic Ratchetting of 1070 under Multiaxial
Stress State,” International Journal of Plasticity, Vol. 10, pp. 579–608.
Jiang, Y., and Sehitoglu, H., 1994b, “Multiaxial Ratchetting under Multiple Step
Loading,” International Journal of Plasticity, Vol. 10, pp. 849–870.
Jiang, Y., and Sehitoglu, H., 1996a, “Modeling of Cyclic Ratchetting Plasticity,
Part I: Development of Constitutive Relations,” ASME Journal of Applied Mechan
ics, Vol. 63, pp. 720–725.
Jiang, Y., and Sehitoglu, H., 1996b, “Modeling of Cyclic Ratchetting Plasticity,
Part II: Comparison of Model Simulations with Experiments,” ASME Journal of
Applied Mechanics, Vol. 63, pp. 726–733.
Fig. 12 Evolution of c
M1
and c
M3
with cycle number for initial 25 propor
tional cycles at e
a
؍ 0.0060j
1
؉ 0.0052j
3
, ͦe
a
ͦ ؍ 7.9(10
؊3
)
Fig. 13 c
M1
and c
M3
versus cycle number for 25 proportional cycles at e
a
؍ 0.0060j
1
؉ 0.0052j
3
, ͦe
a
ͦ ؍ 7.9(10
؊3
) following 25 cycles with strain
components 90 degrees out of phase
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 27
Khan, A. S., and Huang, S., 1995, Continuum Theory of Plasticity, Wiley, New
York.
Kocks, U. F., Tome´, C. N., and Wenk, H.R., 1998, Texture and Anisotropy:
Preferred Orientations in Polycrystals and their Effect on Materials Properties,
Cambridge University Press, Cambridge, U.K.
KuhlmannWilsdorf, D., and Laird, C., 1979, “Dislocation Behavior in Fatigue II.
Friction Stress and Back Stress as Inferred from an Analysis of Hysteresis Loops,”
Materials Science and Engineering, Vol. 37, pp. 111–120.
Lee, K.D., and Krempl, E., 1991, “An Orthotropic Theory of Viscoplasticity
Based on Overstress for Thermomechanical Deformations,” International Journal of
Solids and Structures, Vol. 27, pp. 1445–1459.
Lubliner, J., 1990, Plasticity Theory, Macmillan, New York.
McDowell, D. L., 1985, “An Experimental Study of the Structure of Constitutive
Equations for Nonproportional Cyclic Plasticity,” ASME JOURNAL OF ENGINEERING
MATERIALS AND TECHNOLOGY, Vol. 107, pp. 307–315.
McDowell, D. L., Stahl, D. R., Stock, S. R., and Antolovich, S. D., 1988, “Biaxial
Path Dependence of Deformation Substructure of Type 304 Stainless Steel,” Metal
lurgical Transactions A, Vol. 19A, pp. 1277–1293.
McDowell, D. L., 1992, “A Nonlinear Kinematic Hardening Theory for Cyclic
Thermoplasticity and Thermoviscoplasticity,” International Journal of Plasticity,
Vol. 8, pp. 695–728.
McDowell, D. L., 1995, “Stress State Dependence of Cyclic Ratchetting Behavior
of Two Rail Steels,” International Journal of Plasticity, Vol. 11, pp. 397–421.
Mughrabi, H., 1978, “The Cyclic Hardening and Saturation Behaviour of Copper
Single Crystals,” Materials Science and Engineering, Vol. 33, pp. 207–223.
Moosbrugger, J. C., 1988, “A RateDependent Bounding Surface Model for Non
proportional Cyclic Viscoplasticity,” Ph.D. thesis, Georgia Institute of Technology,
Atlanta, GA.
Moosbrugger, J. C., and McDowell, D. L., 1989, “On a Class of Kinematic
Hardening Rules for Nonproportional Cyclic Plasticity,” ASME JOURNAL OF ENGI
NEERING MATERIALS AND TECHNOLOGY, Vol. 111, pp. 87–98.
Moosbrugger, J. C., 1991, “Some Developments in the Characterization of Material
Hardening and Rate Sensitivity for Cyclic Viscoplasticity Models,” International
Journal of Plasticity, Vol. 7, pp. 405–431.
Moosbrugger, J. C., 1993, “Experimental Parameter Estimation for Nonpropor
tional Cyclic Viscoplasticity: Nonlinear Kinematic Hardening Rules for Two Wasp
aloy Microstructures at 650°C,” Int. J. Plasticity, Vol. 9, pp. 345–373.
Moosbrugger, J. C., and Morrison, D. J., 1997, “Nonlinear Kinematic Hardening
Rule Parameters—Direct Determination From Completely Reversed Proportional
Cycling,” International Journal of Plasticity, Vol. 13, pp. 633–668.
Morrison, D. J., Jones, J. W., and Was, G. S., 1990, “Cyclic Strain Localization in
Ion Beam Microalloyed Nickel,” Scripta Metallurgica, Vol. 24, pp. 2309–2314.
Nouailhas, D., and Chaboche, J. L., 1991, “Anisotropic Constitutive Modeling for
Single Crystal Superalloys using a Continuum Phenomenological Approach,” TIRE A
PART N° 19916, Ofﬁce Nationale d’Etudes et de Recherches Aerospatiales, France.
Nouailhas, D., and Culie, J.P., 1991, “Development and Application of a Model
for Single Crystal Superalloys,” TIRE A PART N° 1991214, Ofﬁce Nationale
d’Etudes et de Recherches Aerospatiales, France.
Nouailhas, D., and Freed, A. D., 1992, “A Viscoplastic Theory for Anisotropic
Materials,” ASME JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 114,
pp. 97–104.
Ohno, N., and Wang, J.D., 1991a, “Nonlinear Kinematic Hardening Rule: Prop
osition and Application to Ratchetting Problems,” Proceedings of the 11th Interna
tional Conference on Structural Mechanics in Reactor Technology, Tokyo, L22/1.,
pp. 481–486.
Ohno, N., and Wang, J.D., 1991b, “Transformation of a Nonlinear Kinematic
Hardening Rule to a Multisurface Form under Isothermal and Nonisothermal Con
ditions,” International Journal of Plasticity, Vol. 7, pp. 879–891.
Ohno, N., and Wang, J.D., 1993a, “Kinematic Hardening Rules with Critical State
of Dynamic Recovery, Part I: Formulation and Basic Features for Ratchetting Be
havior,” International Journal of Plasticity, Vol. 9, pp. 375–390.
Ohno, N., and Wang, J.D., 1993b, “Kinematic Hardening Rules with Critical State
of Dynamic Recovery, Part II: Applications to Experiments of Ratchetting Behavior,”
International Journal of Plasticity, Vol. 9, pp. 391–403.
Sutco, M., and Krempl, E., 1990, “A Simpliﬁed Orthotropic Viscoplasticity Theory
Based on Overstress,” International Journal of Plasticity, Vol. 6, pp. 247–261.
28 / Vol. 122, JANUARY 2000 Transactions of the ASME
M. Mizuno
Y. Mima
M. AbdelKarim
N. Ohno
email ohno@mech.nagoyau.ac.jp
Department of Mechanical Engineering,
Nagoya University,
Chikusaku, Nagoya 4648603, Japan
Uniaxial Ratchetting of 316FR
Steel at Room Temperature—
Part I: Experiments
Uniaxial ratchetting characteristics of 316FR steel at room temperature are studied
experimentally. Cyclic tension tests, in which maximum strain increases every cycle by
prescribed amounts, are conducted systematically in addition to conventional monotonic,
cyclic, and ratchetting tests. Thus hysteresis loop closure, cyclic hardening and visco
plasticity are discussed in the context of constitutive modeling for ratchetting. The cyclic
tension tests reveal that very slight opening of hysteresis loops occurs, and that neither
accumulated plastic strain nor maximum plastic strain induces signiﬁcant isotropic
hardening if strain range is relatively small. These ﬁndings are used to discuss the
ratchetting tests. It is thus shown that uniaxial ratchetting of the material at room
temperature is brought about by slight opening of hysteresis loops as well as by
viscoplasticity, and that kinematic hardening governs almost all strain hardening in
uniaxial ratchetting if stress range is not large.
1 Introduction
Strain accumulation induced by cyclic loading, i.e., ratchetting,
is important in designing structural components. Ratchetting tak
ing place under uniaxial cyclic loading with nonzero mean stress is
referred to as uniaxial ratchetting, which is most fundamental and
has been studied in many works. For 304 and 316 stainless steels,
uniaxial ratchetting experiments have been reported by Yoshida et
al. (1988), Chaboche and Nouailhas (1989), Ruggles and Krempl
(1989, 1990), Yoshida (1990), Sasaki and Ishikawa (1993), Delo
belle (1993), Delobelle et al. (1995), Haupt and Schinke (1996),
and so on.
Closure of stressstrain hysteresis loops, however, has scarcely
been studied experimentally to date. It is obvious that the closure
is directly related with uniaxial ratchetting, because less ratchetting
takes place under uniaxial cyclic loading if the closure is more
complete. It is therefore important to study the closure experimen
tally. The closure can be investigated by performing cyclic tension
tests, which are strain cycling tests with maximum strain increas
ing with every cycle (see Section 2). Such tests, which were
performed ﬁrst by Cofﬁn (1970) to explore the effects of super
imposed cyclic and monotonic strains on the deformation and
fracture of annealed Nickel A, were done recently in a joint work
on the inelastic analysis of liquidlevel induced thermal ratchetting
of 316FR steel cylinders (Inoue and Igari, 1995; Yoshida et al.,
1997).
Ohno et al. (1998) discussed the cyclic tension tests done in the
joint work mentioned above. They thus found that almost complete
closure of hysteresis loops as well as isotropic hardening depend
ing on maximum plastic strain prevailed in the tests. Kobayashi et
al. (1998) then showed this ﬁnding to be essential for predicting
appropriately thermal ratchetting experiments of 316FR steel cyl
inders subjected to spatial variations in temperature; they obtained
good agreement between the experiments and the ﬁnite element
simulations in which they implemented a constitutive model ca
pable of representing complete closure of hysteresis loops. It is,
however, worth noting that the ﬁnding was deduced from cyclic
tension tests performed at high temperatures as high as 300 and
650°C, where the socalled dynamic strain aging can be signiﬁ
cant. It should also be noted that extensive tests were not con
ducted (i.e., strain range was ﬁxed to be 0.2 percent, and maximum
strain was increased by either 0.1 or 0.2 percent every cycle).
One of the structural materials developed for fast breeder reac
tors is 316FR steel, which is based on 316 stainless steel but has
superior high temperature strength (Nakazawa et al., 1988; Taka
hashi, 1998). It has already been reported that 304 and 316 stain
less steels exhibit noticeable viscoplasticity at room temperature
(e.g., Krempl, 1979; Kujawski et al., 1980; Chaboche and Rous
selier, 1983; Yoshida 1990), so that ratchetting of such austenitic
stainless steels at room temperature is affected much more actively
by viscoplasticity than at high temperatures around 550°C where
dynamic strain aging tends to suppress viscoplasticity (Ruggles
and Krempl, 1989; Sasaki and Ishikawa, 1993; Delobelle, 1993).
This suggests that the aforementioned ﬁnding on ratchetting of
316FR steel at high temperatures, i.e., almost complete closure of
hysteresis loops and isotropic hardening depending on maximum
plastic strain, may not hold true at room temperature.
In the present work, therefore, the uniaxial ratchetting charac
teristics of 316FR steel at room temperature are studied with
emphasis on the effects of hysteresis loop closure and viscoplas
ticity on ratchetting. Strain hardening in the presence of ratchetting
is also a matter in the work. Part I of the paper describes the results
of experiments performed to discuss the above factors. The exper
iments include cyclic tension tests, monotonic tension tests, cyclic
straining tests with zero mean strain, and uniaxial ratchetting tests.
Experimental ﬁndings reported in Part I will be taken into account
in Part II, in which a new kinematic hardening model will be
formulated and combined with a viscoplastic equation to simulate
the uniaxial ratchetting experiments.
2 Material and Experimental Program
The specimens tested had gauge sections 8 mm in diameter and
30 mm in axial length. They were machined from a 316FR steel
rolled sheet 50 mm in thickness, so that they had the axis oriented
in the rolling direction of the sheet. The sheet had the chemical
composition shown in Table 1. As seen from the table, 316FR steel
is a lowcarbon nitrogenadded 316 stainless steel. Each specimen
was tested in the asreceived state by using a 100kN closedloop
servohydraulic tensioncompression testing machine with a digital
controller. A clipon extensometer 25 mm in gauge length was
employed to measure the axial elongation. Either the axial elon
gation or the axial load was controlled.
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
October 1, 1998; revised manuscript received August 6, 1999. Associate Technical
Editor: H. Sehitoglu.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 29
Copyright © 2000 by ASME
In the present work, cyclic tension tests were performed in
addition to conventional tests such as monotonic tension, cyclic
straining between ﬁxed strain limits, and uniaxial ratchetting (Ta
ble 2). Figure 1 illustrates the strain history in the cyclic tension
tests, for which cyclic strain range ⌬⑀ and maximum strain incre
ment ␦⑀
max
were prescribed as testing parameters. Six combina
tions of ⌬⑀ and ␦⑀
max
were chosen in the present work to examine
their effects systematically; i.e., for each of three strain ranges of
⌬⑀ ϭ 0.25, 0.4 and 0.8 percent, two tests of small and large
increments of maximum strain, ␦⑀
max
ϭ 0.01 and 0.1 (or 0.05)
percent, were done at a constant strain rate of ⑀˙ ϭ 5 ϫ 10
Ϫ3
percent/s. The cyclic tension tests were interrupted when maxi
mum strain reached a terminal value of ⑀ ϭ 2.0 percent. Uniaxial
ratchetting tests, i.e., cyclic stressing tests with nonzero mean
stresses, were done for four sets of stress ratio R and stress rate ˙
by prescribing maximum stress
max
to be 280 MPa. Here and from
now on R denotes the stress ratio of minimum
min
to maximum
max
. The ratchetting tests were continued for 100 cycles.
3 Experimental Results
3.1 Monotonic Tension Tests. Signiﬁcant viscoplasticity
was observed in the monotonic tension tests performed at constant
strain rates ranging from 5 ϫ 10
Ϫ6
to 5 ϫ 10
Ϫ1
percent/s (Fig. 2).
As seen from the ﬁgure, plastic ﬂow stress became larger by about
ﬁve percent with the tenfold increase in strain rate. Such a signif
icant effect of strain rate can be a feature common to 304 and 316
stainless steels at room temperature (e.g., Krempl, 1979; Kujawski
et al., 1980; Chaboche and Rousselier, 1983; Yoshida, 1990).
3.2 Cyclic Straining Tests. Three tests were done under
cyclic straining between ﬁxed strain limits at ⑀˙ ϭ 5 ϫ 10
Ϫ3
percent/s. Weak cyclic softening and weak hardening/softening
occurred in the tests of ⌬⑀ ϭ 0.25 and 0.4 percent, respectively,
whereas some cyclic hardening took place in the test of ⌬⑀ ϭ 0.8
percent, as shown in Fig. 3 dealing with the change in stress range
⌬ as a function of accumulated inelastic strain p (ϭ ͐ ͉d⑀
p
͉).
Here ⑀
p
denotes inelastic strain. Thus, for 316FR steel at room
temperature, we are allowed to say that cyclic hardening depends
on ⌬⑀, and that cyclic hardening is negligible if ⌬⑀ is small.
Incidentally, the symbols E, ϩ and { in Fig. 3 stand for the cyclic
tension tests discussed in what follows.
3.3 Cyclic Tension Tests. The cyclic tension tests with
large and small values of maximum strain increment ␦⑀
max
gave the
stress versus strain relations shown in Figs. 4(a) to (c) and Figs.
5(a) to (c), respectively. The dashed lines in the ﬁgures indicate for
reference the monotonic tensile curve at ⑀˙ ϭ 5 ϫ 10
Ϫ3
percent/s at
which the cyclic tension tests were performed.
Let us ﬁrst discuss the cyclic tension tests shown in Figs. 4(a) to
(c), in which ␦⑀
max
was prescribed to be large (i.e., ␦⑀
max
ϭ 0.05 or
0.1 percent). It is seen from the ﬁgures that the tests had the
following features: In the tests of ⌬⑀ ϭ 0.25 and 0.4 percent, the
hysteresis loops closed almost completely, and the tensile peak
points came to lie on the monotonic tensile curve; moreover, stress
range varied little with the increase in the number of cycles, N,
(see also Fig. 3). In the test of ⌬⑀ ϭ0.8 percent shown in Fig. 4(c),
on the other hand, the tensile peak points had larger stresses than
in the monotonic tension test, and stress range became noticeably
larger with the increase in N. Therefore we can say that, except for
the effect of isotropic hardening, the hysteresis loops closed almost
completely in the three tests, and that isotropic hardening was
negligible in the tests of ⌬⑀ ϭ 0.25 and 0.4 percent but noticeable
in the test of ⌬⑀ ϭ 0.8 percent. The same features were observed
in the other cyclic tension tests shown in Figs. 5(a) to (c), in which
the increase in maximum strain was much smaller, i.e., ␦⑀
max
ϭ
0.01 percent.
However, a close look at the ﬁgures discussed above reveals
slightly different features in the two cases: Let us pay attention to
the cyclic tensile peak curves formed by linking the tensile peak
points in the cyclic tension tests (Figs. 6(a) and (b)). We then see
that the cyclic tension tests of ␦⑀
max
ϭ 0.1 or 0.05 percent gave
almost the same curves as the monotonic tension test (Fig. 6(a));
on the other hand, in the cyclic tension tests of ␦⑀
max
ϭ 0.01
percent, tensile peak stress came to show smaller increase than
stress in the monotonic tension test, though tensile peak stress
increased rapidly due to cyclic hardening in about the ﬁrst ten
cycles in the cyclic tension test of ⌬⑀ ϭ 0.8 percent (Fig. 6(b)).
Moreover, this smaller increase in tensile peak stress was found to
occur more noticeably when ⌬⑀ was larger in the tests of ␦⑀
max
ϭ
0.01 percent (Fig. 6(b)). It is then suggested that the hysteresis
Fig. 1 Strain history in cyclic tension tests
Fig. 2 Uniaxial tensile stressstrain curves at constant strain rates; the
solid lines indicate the simulated curves to be discussed in Section 4 of
Part II
Table 1 Chemical composition of 316FR steel (wt %)
C Si Mn P S Ni Cr Mo N
0.008 0.54 0.84 0.027 0.004 11.16 16.83 2.10 0.0754
Table 2 Loading conditions in experiments
Monotonic tension ⑀˙ ϭ 5 ϫ 10
Ϫ1
, 5 ϫ 10
Ϫ2
, 5 ϫ 10
Ϫ3
,
5 ϫ 10
Ϫ4
, 5 ϫ 10
Ϫ5
, 5 ϫ 10
Ϫ6
Cyclic straining ⑀˙ ϭ 5 ϫ 10
Ϫ3
⌬⑀ ϭ 0.25, 0.4, 0.8
Cyclic tension ⑀˙ ϭ 5 ϫ 10
Ϫ3
⌬⑀ ϭ 0.25
⌬⑀ ϭ 0.40
⌬⑀ ϭ 0.80
␦⑀
max
ϭ 0.01, 0.05
␦⑀
max
ϭ 0.01, 0.10
␦⑀
max
ϭ 0.01, 0.10
Ratchetting
max
ϭ 280 ˙ ϭ 10 R ϭ 0.0, Ϫ0.5, Ϫ0.75
˙ ϭ 1 R ϭ Ϫ0.75
stress (MPa), strain (%), time (s)
30 / Vol. 122, JANUARY 2000 Transactions of the ASME
loops in the cyclic tension tests opened slightly after saturation of
cyclic hardening as a consequence of cyclic relaxation of mean
stress, as will be discussed later.
3.4 Ratchetting Tests. The stressstrain curves recorded in
the uniaxial ratchetting tests are shown in Figs. 7(a) to (d). The
tensile peak strain versus N relations are plotted in Fig. 8. As seen
from the ﬁgures, ratchetting proceeded without shaking down
under the four loading conditions examined in the present work.
The effect of ˙ or viscoplasticity was marked in the tests: Larger
ratchetting took place in the test at ˙ ϭ 1 MPa/s than in the other
three tests, in which ˙ ϭ 10 MPa/s. The difference in ratchetting
between the two tests of R ϭϪ0.75 at ˙ ϭ1 and 10 MPa/s rapidly
became larger with N in about the ﬁrst ﬁve cycles but slowly in the
subsequent cycles (Fig. 8); in other words, the effect of viscoplas
ticity was signiﬁcant especially in such early cycles. The effect of
R, on the other hand, was not so signiﬁcant: The three tests of R ϭ
0, Ϫ0.5 and Ϫ0.75 at ˙ ϭ 10 MPa/s had nearly the same amount
of ratchetting, as was observed by Yoshida (1990) on 304 stainless
steel at room temperature.
4 Discussion
In this section, the features observed in the cyclic tension tests
are discussed in detail in the context of simulating the ratchetting
tests. Features found in the cyclic tension tests can be summarized
as follows:
(1) Stress range changed little with the increase in the number
of cycles when strain range was small.
(2) The hysteresis loops of stress and strain closed almost
completely except for the effect of cyclic hardening. However, in
the cyclic tension tests in which the increment of maximum strain
was small, tensile peak stress came to show smaller increase than
stress in the monotonic tension test.
Fig. 3 Variation of stress range ⌬ as a function of accumulated inelas
tic strain p; cyclic straining between ﬁxed limits (—) and cyclic tension
(E, ؉, {)
Fig. 4 Stress versus strain relations in cyclic tension tests with large
increment of maximum strain, ␦⑀
max
؍ 0.05 or 0.10 percent, at ⑀˙ ؍ 5 ؋
10
؊3
percent/s
Fig. 5 Stress versus strain relations in cyclic tension tests with small
increment of maximum strain, ␦⑀
max
؍ 0.01 percent, at ⑀˙ ؍ 5 ؋ 10
؊3
percent/s
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 31
Finding (1) implies that for 316FR steel at room temperature
neither accumulated plastic strain nor maximum plastic strain
noticeably induces isotropic or cyclic hardening if strain range is
small; in other words, only kinematic hardening is likely to govern
strain hardening at any strain if strain range is small. Thus, we can
surmise that uniaxial ratchetting proceeds without the effect of
isotropic hardening if stress range is relatively small. To conﬁrm
this, deﬁning the strain range ⌬⑀ in uniaxial ratchetting as illus
trated in Fig. 9, we plot the changes in ⌬⑀ measured in the present
four ratchetting tests (Fig. 10). We thus ﬁnd that ⌬⑀ was almost
constant and at most 0.45 percent in the four ratchetting tests,
though ⌬⑀ increased very weakly with the increase in N when R ϭ
Ϫ0.75. Here we remember that no noticeable cyclic hardening
took place in the cyclic tension tests of ⌬⑀ ϭ 0.4 percent. Finding
(1) therefore leads to the important conclusion that isotropic hard
ening can be ignored when simulating the ratchetting tests done in
the present work.
Let us take note of the cyclic tension tests of 316FR steel done
previously at 300°C and 650°C with ⌬⑀ ϭ0.2 percent and ␦⑀
max
ϭ
0.1 percent (Inoue and Igari, 1995). Ohno et al. (1998) discussed
these high temperature tests and pointed out that the isotropic
hardening depending on maximum plastic strain was signiﬁcant at
650°C, whereas negligible isotropic hardening took place at
300°C. This implies that ﬁnding (1) in the present work can be
valid at least below 300°C, but must be modiﬁed at temperatures
higher than 300°C.
Now we discuss ﬁnding (2). The smaller increase in tensile peak
stress than monotonic tensile stress in this ﬁnding was marked in
the cyclic tension tests of ⌬⑀ ϭ 0.4 and 0.8 percent with ␦⑀
max
ϭ
0.01 percent (Fig. 6(b)). When ␦⑀
max
ϭ0.01 percent, large numbers
of strain cycling with nonzero mean strain were imposed before
maximum strain became equal to terminal strain, ⑀ ϭ 2.0 percent,
to interrupt the cyclic tension tests. It is cyclic relaxation of mean
stress that in general becomes noticeable with the development of
such strain cycling. Therefore, the smaller increase in tensile peak
stress than in monotonic tensile stress can be ascribed to cyclic
relaxation of mean stress. Then, as a consequence of the cyclic
relaxation, it is concluded that the hysteresis loops in the cyclic
tension tests had slight opening rather than complete closure,
though the opening was too slight to inﬂuence the cyclic tension
tests with large increments of maximum strain shown in Figs. 5(a)
to (c).
Here we note that the smaller increase in tensile peak stress
discussed above is not attributable to cyclic softening since no
decrease in stress range occurred before interrupting the cyclic
tension tests of ⌬⑀ ϭ0.4 and 0.8 percent with ␦⑀
max
ϭ0.01 percent
(see Fig. 3).
Slight opening of hysteresis loops allows ratchetting to proceed
slowly with the increase in N. Viscoplasticity also allows ratchetting
to proceed, as was discussed when comparing the ratchetting tests at
˙ ϭ1 and 10 MPa/s in the preceding section. Nowof interest is which
factor was more dominant in the present ratchetting tests. The effect
of viscoplasticity can be identiﬁed with the increase in strain under
unloading frommaximumstress, as illustrated schematically in Fig. 9.
Fig. 6 Change of tensile peak stress in cyclic tension tests in compar
ison with stress change in monotonic tension test at ⑀˙ ؍ 5 ؋ 10
؊3
percent/s
Fig. 7 Stress versus strain relations in ratchetting tests
32 / Vol. 122, JANUARY 2000 Transactions of the ASME
Such increase of strain under unloading was pronounced in early
cycles but declined with the increase in N (Figs. 7(a) to (d)). This is
in accordance with the aforementioned experimental result that the
difference in ratchetting between the two tests of R ϭ Ϫ0.75 at ˙ ϭ
1 and 10 MPa/s increased quickly with N in early cycles but slowly in
subsequent cycles (Fig. 8). Thus we can say that viscoplasticity was
a dominant factor in ratchetting, especially in early cycles in the
present tests. Slight opening of hysteresis loops, on the other hand, can
occur in any stage of ratchetting if reverse loading takes place under
cyclic stressing. It is therefore appropriate to say that slight opening of
hysteresis loops became more inﬂuential than viscoplasticity as ratch
etting proceeded, except for the test of R ϭ 0, in which no reverse
loading took place and only viscoplasticity was a factor in ratchetting.
5 Conclusions
In Part I of this work, the uniaxial ratchetting characteristics of
316FR steel at room temperature were studied experimentally by
performing cyclic tension tests as well as conventional tests such
as monotonic tension, cyclic straining, and ratchetting tests. The
cyclic tension tests were done systematically by prescribing the
increase in maximum strain to be small and large for three strain
ranges of 0.25, 0.4 and 0.8 percent, respectively. The features
observed in the cyclic tension tests were then discussed in the
context of constitutive modeling for ratchetting. The main results
of Part I are summarized as follows:
(1) Stress range changed little with the increase in the number
of cycles when strain range was small in the cyclic tension tests,
and moreover strain range was almost constant in the ratchetting
tests, in which strain range was at most 0.45 percent.
(2) This suggests that, for 316FR steel at room temperature,
neither accumulated inelastic strain nor maximum inelastic strain
induces isotropic or cyclic hardening markedly if strain range is
small; in other words, only kinematic hardening is likely to govern
strain hardening at any strain if strain range is small.
(3) Stressstrain hysteresis loops closed almost completely
except for the effect of isotropic hardening in the cyclic tension
tests. However, in the cyclic tension tests with small increments of
maximum strain, tensile peak stress came to show smaller increase
than stress in the monotonic tension test.
(4) The above result implies that the hysteresis loops of stress
and strain can have slight opening induced by cyclic relaxation of
mean stress. It is therefore necessary to consider slight opening of
hysteresis loops as well as viscoplasticity when simulating uniaxial
ratchetting of 316FR steel at room temperature.
(5) It was deduced that slight opening of hysteresis loops
became more inﬂuential than viscoplasticity as ratchetting pro
ceeded in the uniaxial ratchetting tests with negative stress ratio.
In Part II of the paper, a new and robust kinematic hardening
model capable of representing slight opening of hysteresis loops
will be developed and combined with a viscoplastic equation, and
thus the effects of slight opening of hysteresis loops and visco
plasticity on the ratchetting behavior reported in Part I will be
discussed quantitatively.
Acknowledgment
The authors are grateful to Mr. Masanaka of Nagoya University
for his technical support in performing the experiments.
References
Chaboche, J. L., and Rousselier, G., 1983, “On the Plastic and Viscoplastic
Constitutive Equations—Part II: Application of Internal Variable Concepts to the 316
Stainless Steel,” ASME JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol.
105, No. 2, pp. 159–164.
Chaboche, J. L., and Nouailhas, D., 1989, “Constitutive Modeling of Ratchetting
Effects—Part I: Experimental Facts and Properties of the Classical Models,” ASME
JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 111, No. 4, pp. 384–392.
Cofﬁn, L. F., Jr., 1970, “The Deformation and Fracture of a Ductile Metal under
Superimposed Cyclic and Monotonic Strain,” ASTM STP 467, American Society for
Testing and Materials, pp. 53–76.
Delobelle, P., 1993, “Synthesis of the Elastoviscoplastic Behavior and Modeliza
tion of an Austenitic Stainless Steel over a Large Temperature Range, under Uniaxial
and Biaxial Loadings, Part I: Behavior,” International Journal of Plasticity, Vol. 9,
No. 1, pp. 65–85.
Delobelle, P., Robinet, P., and Bocher, L., 1995, “Experimental Study and Phe
nomenological Modelization of Ratchet under Uniaxial and Biaxial Loading on an
Austenitic Stainless Steel,” International Journal of Plasticity, Vol. 11, No. 4, pp.
295–330.
Fig. 8 Increase in tensile peak strain in ratchetting tests
Fig. 9 Schematic illustration of a stressstrain hysteresis loop under
stress cycling between
max
and R
max
Fig. 10 Variation of strain range ⌬⑀ in ratchetting tests
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 33
Haupt, A., and Schinke, B., 1996, “Experiments on the Ratchetting Behavior of
AISI 316L(N) Austenitic Steel at Room Temperature,” ASME JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY, Vol. 118, No. 3, pp. 281–284.
Inoue, T., and Igari, T., 1995, “Research on Inelastic Analysis of Thermal Ratch
etting under Moving Temperature Distribution—Interim Report,” Subcommittee on
Inelastic Analysis of High Temperature Materials, the Society of Materials Science,
Japan (in Japanese).
Kobayashi, M., Ohno, N., and Igari, T., 1998, “Ratchetting Characteristics of
316FR Steel at High Temperature: Part II: Analysis of Thermal Ratchetting Induced
by Spatial Variation of Temperature,” International Journal of Plasticity, Vol. 14, No.
4–5, pp. 373–390.
Krempl, E., 1979, “An Experimental Study of RoomTemperature RateSensitivity,
Creep and Relaxation of AISI Type 304 Stainless Steel,” Journal of the Mechanics
and Physics of Solids, Vol. 27, No. 5/6, pp. 363–375.
Kujawski, D., Kallianpur, V., and Krempl, E., 1980, “An Experimental Study of
Uniaxial Creep, Cyclic Creep and Relaxation of AISI Type 304 Stainless Steel at
Room Temperature,” Journal of the Mechanics and Physics of Solids, Vol. 28, No. 2,
pp. 129–148.
Nakazawa, T., Abo, H., Tanino, M., Komatsu, H., Nishida, T., and Tashimo, M.,
1988, “Effects of Carbon, Nitrogen and Molybdenum on Creep Properties of Type
316 Stainless Steel,” Proceedings of the Sixth International Conference on Pressure
Vessel Technology, Vol. 2, pp. 1041–1048.
Ohno, N., AbdelKarim, M., Kobayashi, M., and Igari, T., 1998, “Ratchetting
Characteristics of 316FR Steel at High Temperature: Part I: StrainControlled Ratch
etting Experiments and Simulations,” International Journal of Plasticity, Vol. 14, No.
4–5, pp. 355–372.
Ruggles, M. B., and Krempl, E., 1989, “The Inﬂuence of Test Temperature on the
Ratchetting Behavior of Type 304 Stainless Steel,” ASME JOURNAL OF ENGINEERING
MATERIALS AND TECHNOLOGY, Vol. 111, No. 4, pp. 378–383.
Ruggles, M. B., and Krempl, E., 1990, “The Interaction of Cyclic Hardening
and Ratchetting for AISI Type 304 Stainless Steel at Room Temperature—I:
Experiments,” Journal of the Mechanics and Physics of Solids, Vol. 38, No. 4, pp.
575–585.
Sasaki, K., and Ishikawa, H., 1993, “Experimental Observation on Viscoplastic
Behavior of SUS304 (Creep and Ratchetting Behavior),” Transactions of the Japan
Society of Mechanical Engineers, Series A, Vol. 59, No. 562, pp. 39–45 (in Japa
nese).
Takahashi, Y., 1998, “Evaluation of CreepFatigue Life Prediction Methods for
LowCarbon NitrogenAdded 316 Stainless Steel,” ASME JOURNAL OF ENGINEERING
MATERIALS AND TECHNOLOGY, Vol. 120, No. 2, pp. 119–125.
Yoshida, F., Kondo, J., and Kikuchi, Y., 1988, “ViscoPlastic Behavior of Stainless
Steel SUS 304 under Cyclic Loading at Room Temperature,” Transactions of the
Japan Society of Mechanical Engineers, Series A, Vol. 54, No. 501, pp. 1151–1157
(in Japanese).
Yoshida, F., 1990, “Uniaxial and Biaxial Creepratchetting Behavior of SUS304
Stainless Steel at Room Temperature,” International Journal of Pressure Vessels and
Piping, Vol. 44, pp. 207–223.
Yoshida, F., Kobayashi, M., Tsukimori, K., Uno, T., Fukuda, Y., Igari, T., and
Inoue, T., 1997, “Inelastic Analysis of Material Ratchetting of 316FR at Varying
Temperature—Results of Benchmark Project (B) by JSMS,” Transactions of the 14th
International Conference on Structural Mechanics in Reactor Technology, Vol. 9, pp.
221–228.
34 / Vol. 122, JANUARY 2000 Transactions of the ASME
N. Ohno
email ohno@mech.nagoyau.ac.jp
M. AbdelKarim
Department of Mechanical Engineering,
Nagoya University,
Chikusaku, Nagoya 4648603, Japan
Uniaxial Ratchetting of 316FR
Steel at Room Temperature—
Part II: Constitutive Modeling
and Simulation
Uniaxial ratchetting experiments of 316FR steel at room temperature reported in Part I
are simulated using a new kinematic hardening model which has two kinds of dynamic
recovery terms. The model, which features the capability of simulating slight opening of
stressstrain hysteresis loops robustly, is formulated by furnishing the Armstrong and
Frederick model with the critical state of dynamic recovery introduced by Ohno and Wang
(1993). The model is then combined with a viscoplastic equation, and the resulting
constitutive model is applied successfully to simulating the experiments. It is shown that
for ratchetting under stress cycling with negative stress ratio, viscoplasticity and slight
opening of hysteresis loops are effective mainly in early and subsequent cycles, respec
tively, whereas for ratchetting under zerototension only viscoplasticity is effective.
1 Introduction
Uniaxial ratchetting, i.e., strain accumulation under uniaxial
cyclic loading with nonzero mean stress, is a fundamental phe
nomenon in cyclic plasticity and is important in designing struc
tural components. This ratchetting can occur if materials are vis
coplastic or if stressstrain hysteresis loops have opening under
uniaxial cyclic loading (Chaboche and Nouailhas, 1989). Hence
viscoplasticity and hysteresis loop opening are key factors in
uniaxial ratchetting, especially in the context of constitutive mod
eling.
In Part I of the present work, hysteresis loop opening of 316FR
steel at room temperature was studied experimentally by perform
ing cyclic tension tests besides conventional tests such as mono
tonic tension, cyclic straining, and ratchetting under uniaxial load
ing. It was thus found that, although hysteresis loops have only
slight opening in the saturated state of cyclic hardening, that
opening in cooperation with viscoplasticity can result in noticeable
ratchetting under uniaxial cyclic stressing. This suggests the ne
cessity of a constitutive model which can simulate slight opening
of hysteresis loops and viscoplasticity. It was also found that
isotropic or cyclic hardening hardly develops if strain range is
relatively small under cyclic straining, so that kinematic hardening
governs almost all strain hardening in ratchetting if stress range is
not large.
It is not easy to simulate ratchetting accurately, since ratchetting
is strain accumulation proceeding cycle by cycle. Classical con
stitutive models thus have poor capabilities for predicting ratchet
ting (Ohno, 1990, 1998). Intensive studies in the last decade,
however, have shown that although the kinematic hardening model
of Armstrong and Frederick (1966) usually overpredicts ratchet
ting, modiﬁcations of the dynamic recovery term in the model are
effective for simulating ratchetting properly (Burlet and Caille
taud, 1987; Chaboche and Nouailhas, 1989; Freed and Walker
1990; Chaboche, 1991; Ohno and Wang, 1993, 1994; McDowell,
1994, 1995; Jiang and Sehitoglu, 1994a, 1994b, 1996; Jiang and
Kurath, 1996; Wang and Barkey, 1998). Among them, Ohno and
Wang (1993) showed that complete or almost complete closure of
hysteresis loops is represented by introducing the critical state of
dynamic recovery, which leads to little or no uniaxial ratchetting
except for the effect of viscoplasticity. If the critical state of
dynamic recovery is assumed for each part of back stress, kine
matic hardening develops piecewise linearly or piecewise almost
linearly (Ohno and Wang, 1993), and as a consequence the mate
rial parameters concerned with kinematic hardening can be deter
mined systematically from experiments (Jiang and Sehitoglu,
1996; Jiang and Kurath, 1996).
The constitutive property of complete or almost complete clo
sure of hysteresis loops stated above is, as a ﬁrst approximation,
just an experimental characteristic observed in the cyclic tension
tests of 316FR steel at room temperature, as was discussed in Part
I. This implies that a viscoplastic constitutive model into which the
critical state of dynamic recovery of back stress is incorporated
appropriately can be successful in simulating the ratchetting tests
reported in Part I. Incidentally, a rateindependent constitutive
model expressing complete closure of hysteresis loops based on
the critical state of dynamic recovery was used to successfully
simulate cyclic tension tests and thermal ratchetting tests of 316FR
steel at high temperatures where the effect of creep was supposed
to be insigniﬁcant because of dynamic strain aging (Ohno et al.,
1998; Kobayashi et al., 1998).
Part II is concerned with simulation of the ratchetting experi
ments reported in Part I. A new kinematic hardening model, which
can simulate slight opening of hysteresis loops robustly, is devel
oped by furnishing the Armstrong and Frederick model with the
critical state of dynamic recovery introduced by Ohno and Wang
(1993), so that the evolution equation of kinematic hardening has
two kinds of dynamic recovery terms. The new model is then
combined with a viscoplastic equation, and the resulting constitu
tive model is applied to simulating the experiments. Thus the
effects of viscoplasticity and slight opening of hysteresis loops on
the uniaxial ratchetting of 316FR steel at room temperature are
discussed quantitatively.
2 Kinematic Hardening Model
In this section, a kinematic hardening model is developed to
represent slight opening of hysteresis loops robustly. Here it is
assumed that strain ⑀ is divided additively into elastic part ⑀
e
and
inelastic part ⑀
p
, i.e.,
⑀ ϭ ⑀
e
ϩ ⑀
p
. (1)
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
October 1, 1998. Associate Technical Editor: H. Sehitoglu.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 35
Copyright © 2000 by ASME
It is also assumed that back stress ␣ consists of M parts in order to
accurately express the change of ␣ (Chaboche and Rousselier,
1983):
␣ ϭ
iϭ1
M
␣
i
. (2)
From now on ( ˙) denotes the differentiation with respect to time t,
(:) the inner product between second rank tensors, and ͗ ͘ Ma
cauley’s bracket, i.e., ͗x͘ ϭ ( x ϩ ͉x͉)/ 2.
2.1 Formulation. Now let a
i
be the deviatoric part of ␣
i
. In
order to express the evolution of a
i
, we consider a combination of
the Armstrong and Frederick (1966) model and the ﬁrst version of
the Ohno and Wang (1993) model. The two models have a differ
ence with respect to the dynamic recovery of a
i
. In the Armstrong
and Frederick model, the dynamic recovery of a
i
operates at all
times in proportion to a
i
and the accumulating rate of inelastic
strain,
p˙ ϭ ͑
2
3
⑀˙
p
:⑀˙
p
͒
1/2
. (3)
In the ﬁrst version of the Ohno and Wang model, on the other
hand, the dynamic recovery of a
i
is assumed to take place only on
a critical surface f
i
(a
i
) ϭ 0, on which the dynamic recovery of a
i
is fully activated so that a
i
cannot develop beyond the surface f
i
ϭ
0; in other words, a
i
is allowed to move either inside or on the
surface f
i
ϭ 0. The surface is deﬁned to be a hypersphere of radius
r
i
in the space of a
i
such that
f
i
ϭ
3
2
a
i
:a
i
Ϫ r
i
2
ϭ 0. (4)
Let us assume the two kinds of dynamic recovery terms men
tioned above in order to furnish the Armstrong and Frederick
model with the critical surface of dynamic recovery, so that the
overprediction of ratchetting by the Armstrong and Frederick
model can be overcome. Then, with Heaviside’s step function H
and Macaulay’s bracket ͗ ͘, the evolution equation of a
i
can be
expressed as
a˙
i
ϭ
i
͓
2
3
r
i
⑀˙
p
Ϫ
i
a
i
p˙ Ϫ H͑ f
i
͒͗
˙
i
͘a
i
͔, (5)
where
i
and
i
are material parameters, and
˙
i
is determined to
have the following form using f
˙
i
ϭ 0 and f
i
ϭ 0:
˙
i
ϭ ⑀˙
p
:
a
i
r
i
Ϫ
i
p˙ (6)
The second and third terms in the right hand side in equation (5)
express the dynamic recovery of a
i
based on the Armstrong and
Frederick model and the Ohno and Wang model, respectively,
whereas the ﬁrst term is responsible for strain hardening.
If f
i
Ͻ 0 or
˙
i
Ͻ 0, Eq. (5) is reduced to the Armstrong and
Frederick model
a˙
i
ϭ
i
͑
2
3
r
i
⑀˙
p
Ϫ
i
a
i
p˙ ͒. (7)
On the other hand, if f
i
ϭ 0 and
˙
i
Ն 0, Eq. (5) becomes identical
to the following model, i.e., the ﬁrst version of the kinematic
hardening model formulated by Ohno and Wang (1993):
a˙
i
ϭ
iͫ
2
3
r
i
⑀˙
p
Ϫ H͑ f
i
͒ͳ⑀˙
p
:
a
i
r
i
ʹa
iͬ
. (8)
Therefore, Eq. (5) changes its form from Eq. (7) to (8) as soon as
a
i
becomes located on the surface f
i
ϭ 0. Let us emphasize that if
f
i
ϭ 0 and
˙
i
Ն 0, a
i
moves on the surface f
i
ϭ 0; otherwise
inside the surface f
i
ϭ 0.
It will be shown in Section 3.1 that Eq. (5) represents slight
opening of hysteresis loops if 0 Ͻ
i
Ӷ 1. Incidentally, Eq. (8)
with H( f
i
) replaced by a power function (␣
ieq
/r
i
)
mi
was shown to
represent slight opening of hysteresis loops (Ohno and Wang,
1993). Here ␣
ieq
ϭ [(
3
2
)a
i
:a
i
]
1/ 2
. Moreover, Eq. (8) with H( f
i
)͗⑀˙
p
:
a
i
/r
i
͘ replaced by (␣
ieq
/r
i
)
mi
p˙ was used to simulate ratchetting in
previous works (Ohno and Wang, 1994; Chaboche, 1994; Jiang
and Sehitoglu, 1996; Jiang and Kurath, 1996).
2.2 Generalization. Let us normalize a
i
as
a
i
ϭ
a
i
r
i
. (9)
Then Eq. (5) is cast into an alternative form
a
˙
i
ϭ
i
͓
2
3
⑀˙
p
Ϫ
i
a
i
p˙ Ϫ H͑ f
i
͒͗
˙
i
͘a
i
͔. (10)
where f
i
ϭ (
3
2
)a
i
:a
i
Ϫ 1 and
˙
i
ϭ ⑀˙
p
:a
i
Ϫ
i
p˙. Since equation (10)
satisﬁes the consistency condition
˙
f
i
ϭ 0 irrespective of r
i
, this
equation can be used even when r
i
changes with the development of
inelastic deformation. Equations (9) and (10) thus allow Eq. (5) to be
generalized into
a˙
i
ϭ
iͫ
2
3
r
i
⑀˙
p
Ϫ
i
a
i
p˙ Ϫ H͑ f
i
͒͗
˙
i
͘a
iͬ
ϩ
r˙
i
r
i
a
i
. (11)
Addition of the last term (r˙
i
/r
i
)a
i
to Eq. (8) was proposed by
Takahashi and Tanimoto (1995).
3 Features of Model
Features of Eq. (5) are now explained by showing schematically
the evolution of back stress under uniaxial and multiaxial loading.
An efﬁcient method of numerical integration for Eq. (5) is also
shown as a consequence of a multiaxial feature of the model.
3.1 Uniaxial Loading. Equation (5) has an uniaxial expres
sion
␣˙
i
ϭ
i
͓r
i
⑀˙
p
Ϫ
i
␣
i
͉⑀˙
p
͉ Ϫ H͑ f
i
͒͗
˙
i
͘␣
i
͔, (12)
where f
i
ϭ ␣
i
2
Ϫ r
i
2
and
˙
i
ϭ ⑀˙
p
:
␣
i
r
i
Ϫ
i
͉⑀˙
p
͉. (13)
First let us consider monotonic tensile loading. Equation (12) is
then reduced to
␣˙
i
ϭ ͭ
i
͑r
i
Ϫ
i
␣
i
͒⑀˙
p
, 0 Յ ␣
i
Ͻ r
i
,
0, ␣
i
ϭ r
i
,
(14)
the solution of which can be expressed as
␣
i
ϭ r
iͫ
1 Ϫ ͳ1 Ϫ
1 Ϫ exp͑Ϫ
i
i
⑀
p
͒
i
ʹ
ͬ
. (15)
This tensile change of ␣
i
, which has a corner at ␣
i
ϭ r
i
, is
illustrated in Fig. 1. Two special cases of
i
ϭ 0 and
i
ϭ 1 are
dealt with in the ﬁgure, too. When
i
ϭ 0, Eq. (5) is reduced to
Eq. (8); as a result, ␣
i
changes bilinearly as follows (Ohno and
Wang, 1993):
␣
i
ϭ r
i
͓1 Ϫ ͗1 Ϫ
i
⑀
p
͔͘. (16)
This relation can be derived from Eq. (15) as the limiting case of
i
3 0. On the other hand, when
i
ϭ 1, Eq. (15) becomes
␣
i
ϭ r
i
͓1 Ϫ exp͑Ϫ
i
⑀
p
͔͒. (17)
Hence, if
i
ϭ 1, ␣
i
approaches r
i
only asymptotically; in other
words, ␣
i
does not in effect reach the critical state f
i
ϭ 0. In this
case, therefore, Eq. (5) takes the form of Eq. (7).
Figure 2 illustrates the monotonic tensile evolution of back
stress ␣ obtained by superposing ␣
1
to ␣
M
. As illustrated in the
ﬁgure, if 0 Յ
i
Ͻ 1, the ␣ versus ⑀
p
relation have corners, since
each ␣
i
versus ⑀
p
relation has a corner under tensile loading.
36 / Vol. 122, JANUARY 2000 Transactions of the ASME
Especially if
i
ϭ 0, the ␣ versus ⑀
p
relation is multilinear (Ohno
and Wang, 1993), as shown by the dashed line in Fig. 2. Thus
i
and r
i
in the case of
i
ϭ 0 are related with the coordinates of
corners of the multilinear ␣ versus ⑀
p
relation, ␣
(i)
and ⑀
(i)
p
, as
follows (Jiang and Sehitoglu, 1996; Jiang and Kurath, 1996):
i
ϭ
1
⑀
͑i͒
p
, (18)
r
i
ϭ
ͫ
␣
͑i͒
Ϫ ␣
͑iϪ1͒
⑀
͑i͒
p
Ϫ ⑀
͑iϪ1͒
p
Ϫ
␣
͑iϩ1͒
Ϫ ␣
͑i͒
⑀
͑iϩ1͒
p
Ϫ ⑀
͑i͒
p ͬ
⑀
͑i͒
p
, (19)
where ⑀
(0)
p
ϭ 0 and ␣
(0)
ϭ 0.
Now let us consider such uniaxial cyclic loading that the direc
tion of ⑀˙
p
is changed whenever back stress ␣ reaches either the
maximum or the minimum limit prescribed, ␣
max
or ␣
min
. The
change of ␣ and ⑀
p
under such cyclic loading is illustrated in Figs.
3(a) and (b) for
i
ϭ 0 and 0 Ͻ
i
Ͻ 1, respectively. It is noticed
that Eq. (12) has two kinds of dynamic recovery terms, i.e., the
term operating always and that becoming active in the critical state
f
i
ϭ 0. The former causes the ␣ versus ⑀
p
hysteresis loops to be
opened, as in the Armstrong and Frederick model. The latter, on
the other hand, does not contribute to the opening, since Eq. (12)
with
i
ϭ 0, i.e., the uniaxial form of Eq. (5), describes the ␣
versus ⑀
p
hysteresis loops to be multilinear and closed (Ohno and
Wang, 1993). Thus, the smaller
i
is set, the less opening Eq. (12)
gives to the hysteresis loops, giving rise to the smaller uniaxial
ratchetting; especially if
i
ϭ 0, Eq. (12) expresses no uniaxial
ratchetting except for the effect of viscoplasticity. Therefore the
kinematic hardening model formulated in the preceding section
can represent slight opening of hysteresis loops robustly.
3.2 Multiaxial Loading. Let us remember that Eq. (5) al
lows a
i
to move either inside or on the surface f
i
ϭ 0, and that Eq.
(5) is reduced to Eqs. (7) and (8) when a
i
moves inside and on the
surface f
i
ϭ 0, respectively. This feature of Eq. (5) causes the
direction of a˙
i
to change discontinuously as soon as a
i
reaches the
critical surface f
i
ϭ 0, as shown schematically in Fig. 4.
The feature above implies the following efﬁcient method for
numerical integration of Eq. (5): First, we compute the increment
of a
i
based on Eq. (7), ⌬a
*
i
, using its incremental form
⌬a
*
i
ϭ
i
͕
2
3
r
i
⌬⑀
p
Ϫ
i
͓a
i
͑t͒ ϩ ⌬a
*
i
͔⌬p͖, (20a)
i.e.,
Fig. 1 Evolution of ␣
i
under uniaxial tensile loading
Fig. 2 Evolution of back stress ␣ and its parts under uniaxial tensile
loading (M ؍ 3)
Fig. 3 Hysteresis loops of back stress ␣ and inelastic strain ⑀
p
under
uniaxial cycling between ␣
max
and ␣
min
; (a)
i
؍ 0, (b) 0 <
i
< 1
Fig. 4 Radial return mapping for numerical integration of Eq. (5)
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 37
⌬a
*
i
ϭ
iͫ
2
3
r
i
⌬⑀
p
Ϫ
i
a
i
͑t͒⌬p
ͬ
1 ϩ
i
i
⌬p
, (20b)
where ⌬ indicates increments, and 0 Յ Յ 1. Then, noticing that
Eq. (5) without the critical surface f
i
ϭ 0 gives the above incre
ment, and deﬁning a
*
i
such that
a
*
i
ϭ a
i
͑t͒ ϩ ⌬a
*
i
, (21)
we set a
i
(t ϩ ⌬t) ϭ a
*
i
if a
*
i
satisﬁes f
i
Յ 0; otherwise we use
radial return mapping so that a
i
(t ϩ ⌬t) is located on the surface
f
i
ϭ 0 (Fig. 4), i.e.,
a
i
͑t ϩ ⌬t͒ ϭ r
i
a
*
i
␣
*
ieq
, (22)
where ␣
*
ieq
ϭ [(
3
2
)a
*
i
:a
*
i
]
1/ 2
. The radial return mapping to the
critical surface f
i
ϭ 0 is expected to have almost the same high
accuracy as that to the yield surface of elasticperfectly plastic
materials examined by Krieg and Krieg (1977), because they are
similar to each other. Equation (5) is therefore robust also in the
sense that an efﬁcient method of numerical integration is available
for accurately representing slight opening of hysteresis loops.
4 Viscoplastic Equation and Material Parameters
This section describes the viscoplastic equation and material
parameters used for simulating the ratchetting experiments.
We assume that elastic strain ⑀
e
obeys Hooke’s law
⑀
e
ϭ
1 ϩ
E
Ϫ
E
͑tr ͒I, (23)
where E and indicate elastic constants, and I stress tensor and
the unit tensor of rank two, respectively, and tr the trace. For the
change of inelastic strain ⑀
p
, on the other hand, we employ a
viscoplastic equation with back stress
⑀˙
p
ϭ
3
2
f͑
eff
͒
s Ϫ a
eff
, (24)
where s denotes deviatoric stress, and f(
eff
) represents a material
function of effective stress
eff
deﬁned as
eff
ϭ ͓
3
2
͑s Ϫ a͒:͑s Ϫ a͔͒
1/2
. (25)
The kinematic hardening model formulated in Section 2 is used to
represent the change of back stress in Eq. (24). Isotropic hardening
on the other hand is ignored in the equation, since little isotropic
hardening developed in the ratchetting experiments (Part I).
Then, the uniaxial tensile form of Eq. (24), ⑀˙
p
ϭ f( Ϫ ␣),
allows viscoplastic ﬂow stress in uniaxial tensile deformation to be
expressed approximately as
Ϸ f
Ϫ1
͑⑀˙ ͒ ϩ ␣͑⑀
p
͒, (26)
where f
Ϫ1
indicates the inverse function of f, and ␣(⑀
p
) denotes the
evolution of back stress ␣ in tensile deformation. Now let us
remember that in the ratchetting experiments the hysteresis loops
opened only slightly except for the effect of viscoplasticity (Part I),
and that the smaller
i
is set the less opening of hysteresis loops
Eq. (5) expresses. It is therefore appropriate to assume that
i
Ӷ 1. (27)
Equation (5) then gives almost the same evolution of back stress
under tensile loading as the multilinear one predicted by
i
ϭ 0
(see Fig. 2). As a result, ␣(⑀
p
) in Eq. (5) is related with the material
parameters
i
and r
i
through Eqs. (18) and (19).
We thus determined the material parameters
i
and r
i
as well as
the viscoplastic function f( Ϫ ␣) on the basis of the uniaxial
tensile experiments at constant strain rates shown in Fig. 2 of Part
I. First,
i
and r
i
were determined tentatively by approximating
multilinearly the tensile curve at ⑀˙ ϭ 5 ϫ 10
Ϫ3
percent/s and by
using Eqs. (18) and (19); the functional form of f( Ϫ ␣) was
sought by analyzing the strain rate dependence of 1.0 percent
offset stress. Then, the values of
i
and r
i
as well as the constants
in f( Ϫ ␣) were adjusted to best ﬁt the tensile experiments using
Eqs. (2), (5), (23), and (24). This procedure led to the viscoplastic
function and material parameters given in Table 1 and the simu
lated curves shown by the solid lines in Fig. 2 of Part I.
The material parameter
i
, which was not determined from the
tensile experiments, will be used to discuss the effect of hysteresis
loop opening on ratchetting in the following section. It is worth
while to point out here that the parameter
i
has little inﬂuence on
uniaxial tensile behavior if
i
Ͻ 0.5, as shown in a speciﬁc case
of ⑀˙ ϭ 5 ϫ 10
Ϫ3
percent/s in Fig. 5.
5 Simulation of Experiments
In Part I we discussed four ratchetting experiments, which had
the same maximum stress of
max
ϭ 280 MPa but different sets of
stress ratio R (ϭ
min
/
max
) and stress rate ˙ . Figures 6(a) to (d)
compare the ratchetting experiments and the simulation based on
the present constitutive model, which was computed with respect
to four values of the parameter
i
.
To begin with, we discuss the simulation obtained with
i
ϭ 0.
This choice of
i
renders the ␣ versus ⑀
p
hysteresis loops closed
Table 1 Viscoplastic function and material parameters
Elastic E ϭ 1.96 ϫ 10
5
, ϭ 0.3
Viscoplastic f(
eff
) ϭ 2.48 ϫ 10
Ϫ10
sinh ͗(
eff
Ϫ 75.0)/6.20͘
Kinematic hardening (M ϭ 10)
1
ϭ 1.00 ϫ 10
4
, r
1
ϭ 10.0
3
ϭ 2.50 ϫ 10
3
, r
3
ϭ 17.3
5
ϭ 4.00 ϫ 10
2
, r
5
ϭ 20.6
7
ϭ 1.00 ϫ 10
2
, r
7
ϭ 6.40
9
ϭ 3.33 ϫ 10, r
9
ϭ 9.00
2
ϭ 5.00 ϫ 10
3
, r
2
ϭ 14.0
4
ϭ 1.00 ϫ 10
3
, r
4
ϭ 22.0
6
ϭ 1.84 ϫ 10
2
, r
6
ϭ 14.7
8
ϭ 5.00 ϫ 10, r
8
ϭ 10.0
10
ϭ 2.50 ϫ 10, r
10
ϭ 80.0
stress (MPa), strain (mm/mm), time (s).
Fig. 5 Inﬂuence of parameter
i
on uniaxial tensile curve at ⑀˙ ؍5 ؋10
؊3
percent/s
38 / Vol. 122, JANUARY 2000 Transactions of the ASME
completely (Section 3.1), bringing in only the effect of viscoplas
ticity in simulating uniaxial ratchetting. It is noticed that only
viscoplasticity or creep can induce ratchetting especially if mean
stress is high relatively to stress amplitude (Chaboche and Nouail
has, 1989). As seen from Figs. 6(a) to (d), the present model with
i
ϭ 0 well simulates the experiment of R ϭ 0 but underestimates
the other three experiments of R ϭ Ϫ0.5 and Ϫ0.75. In other
words, viscoplasticity accounts well for the increase in strain under
zerototension cyclic loading, but only viscoplasticity is insufﬁ
cient for predicting the ratchetting experiments of R ϭ Ϫ0.5 and
Ϫ0.75. It is hysteresis loop opening that can be more or less
substantial in uniaxial ratchetting of R Ͻ 0, in which reverse
loading takes place. It is therefore necessary to consider hysteresis
loop opening as well as viscoplasticity for simulating uniaxial
ratchetting of 316FR steel at room temperature.
Here let us point out that if
i
ϭ 0, ratchetting stops eventually
under stress cycling of R ϭ Ϫ0.5 and Ϫ0.75 (Figs. 6(b) to (d)).
This is because Eqs. (2) and (5) with
i
ϭ 0 make mean back
stress grow monotonically as mean inelastic strain increases in
ratchetting. Then, as mean back stress increases, the stress versus
strain hysteresis loops become symmetrized to have the same
shapes in tension and compression sides; when completely sym
metrized, viscoplasticity induces no ratchetting.
We are now in a position to discuss the effect of hysteresis loop
opening, for which the parameter
i
is responsible. As seen from
Figs. 6(b) to (d), a larger value of
i
gives more signiﬁcant
ratchetting under stress cycling of R ϭ Ϫ0.5 and Ϫ0.75, though
i
has little inﬂuence under zerototension cyclic loading. It is
emphasized that all the experiments examined are well simulated
if
i
ϭ 0.02. This value of
i
is so small that the hysteresis loops
of ␣ and ⑀
p
can be opened only slightly to cause very slow
ratchetting except for the effect of viscoplasticity. This is in
accordance with the experimental fact that strain increased only by
about 0.002 to 0.003 percent per cycle in the steady state when
R ϭ Ϫ0.5 and Ϫ0.75 (Figs. 6(b) to (d)). We have thus ascertained
that ratchetting under stress cycling of R Ͻ 0 is driven by both
viscoplasticity and slight opening of hysteresis loops.
Let us compare the simulations of
i
ϭ 0 and
i
ϭ 0.02 shown
in Figs. 6(b) to (d). They start to deviate from each other after
about twenty and ﬁve cycles when R ϭ Ϫ0.5 and Ϫ0.75, respec
tively. Since only the effect of viscoplasticity is incorporated if
i
ϭ 0, we can conclude as follows: In the experiments of R ϭ
Ϫ0.5 and Ϫ0.75, viscoplasticity was effective in ratchetting in
such early cycles, whereas slight opening of hysteresis loops
became inﬂuential as ratchetting proceeded, as was deduced ex
perimentally in Part I.
The stress versus strain relations given by the present constitu
tive model are shown in two typical cases of uniaxial ratchetting.
Figure 7 deals with zerototension cyclic loading at ˙ ϭ 10
MPa/s. It is seen in this case that ratchetting occurs due to visco
plasticity, which is signiﬁcant near tensile peak stress. It is also
seen that the effect of viscoplasticity decreases with the increase in
inelastic strain. This is because the increase in inelastic strain
causes the development of back stress ␣, leading to the decrease in
effective stress Ϫ ␣ near tensile peak. Here we notice again that
opening of ␣ versus ⑀
p
hysteresis loops shows no notable effect in
this case, since no reverse loading takes place, so that the value of
i
has almost no inﬂuence on the ratchetting simulated in Fig. 7,
as was seen in Fig. 6(a). Figures 8(a) and (b), on the other hand,
deal with the case of R ϭ Ϫ0.75 and ˙ ϭ 10 MPa/s, in which the
value of
i
has great inﬂuence. Here, if
i
ϭ 0, ratchetting
develops only in about the ﬁrst ﬁve cycles (Fig. 8(a)), but if
i
ϭ
0.02, ratchetting continues to develop properly in the subsequent
cycles (Fig. 8(b)). In other words, viscoplasticity and slight open
ing of hysteresis loops are, respectively, the main driving forces in
the early and subsequent cycles, as aforementioned. Incidentally,
the dashed lines in Figs. 8(a) and (b) indicate the 100th hysteresis
loop in the experiment. It is seen that the model simulates fairly
Fig. 6 Increase of tensile peak strain under four stress cycling condi
tions (
max
؍ 280 MPa)
Fig. 7 Simulation of uniaxial ratchetting under zerototension cyclic
loading (
max
؍ 280 MPa, ˙ ؍ 10 MPa/s)
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 39
well the shape of hysteresis loops in spite of the complete neglect
of isotropic hardening.
Finally, let us show an example of the simulation of cyclic
tension in Figs. 9(a) to (c). The experiment for this example was
depicted in Fig. 5(b) of Part I. It is seen that the simulation greatly
depends on the value of
i
and is close to the experiment if
i
ϭ
0.02.
6 Concluding Remarks
In Part II, a new kinematic hardening model was ﬁrst developed
by furnishing the Armstrong and Frederick model with the critical
state of dynamic recovery introduced by Ohno and Wang. This led
to a model with two kinds of dynamic recovery terms, i.e., the term
operating at all times and one that becomes active in the critical
state. It was shown that the model is capable of representing slight
opening of hysteresis loops robustly, and that it has an efﬁcient
method of numerical integration based on radial return mapping.
The model was then combined with a viscoplastic equation with
back stress, and the resulting constitutive model was applied
successfully to simulating the ratchetting experiments of 316FR
steel at room temperature discussed in Part I. It was thus shown
that for ratchetting under negative stress ratio, viscoplasticity and
slight opening of hysteresis loops are the main driving forces in
early and subsequent cycles, respectively, whereas for ratchetting
under zerototension cyclic loading only viscoplasticity is effec
tive.
The material parameter
i
was taken to be identical for all the
parts of back stress in the present work. Thus, for stress cycling of
negative stress ratio, the simulation had steady states in ratchetting
soon after saturation of the effect of viscoplasticity, whereas the
experiments exhibited slow decay in ratchetting before reaching
steady states. This slow decay may be expressed if different values
are assigned to
1
,
2
, . . . , and
M
.
Isotropic hardening, which developed slightly in the ratchetting
experiments, was ignored in the simulation accordingly. In gen
eral, however, isotropic hardening takes place and, as a result, the
parameter r
i
may change (McDowell, 1994, 1995; Takahashi and
Tanimoto, 1995; Jiang and Sehitoglu, 1996). Then Eq. (10) with
(9) or Eq. (11) can be used to allow the change of r
i
induced by
isotropic hardening.
Acknowledgment
The authors are grateful to Drs. M. Mizuno and M. Kobayashi of
Nagoya University for their kind discussion of the results in Part II.
References
Armstrong, P. J., and Frederick, C. O., 1966, “A Mathematical Representation of
the Multiaxial Bauschinger Effect,” CEGB Report RD/B/N731, Berkeley Nuclear
Laboratories, Berkeley, UK.
Burlet, H., and Cailletaud, G., 1987, “Modeling of Cyclic Plasticity in Finite
Element Codes,” Proceedings of the 2nd International Conference on Constitutive
Laws for Engineering Materials: Theories and Applications, Vol. II, Elsevier, New
York, pp. 1157–1164.
Chaboche, J. L., and Rousselier, G., 1983, “On the Plastic and Viscoplastic
Constitutive Equations, Part I: Rules Developed With Internal Variable Concept, Part
II: Application of Internal Variable Concepts to the 316 Stainless Steel,” ASME
Journal of Pressure Vessel Technology, Vol. 105, No. 2, pp. 153–164.
Chaboche, J. L., and Nouailhas, D., 1989, “Constitutive Modeling of Ratchetting
Effects, Part I: Experimental Facts and Properties of the Classical Models, Part II:
Possibilities of Some Additional Kinematic Rules,” ASME JOURNAL OF ENGINEERING
MATERIALS AND TECHNOLOGY, Vol. 111, No. 4, pp. 384–392, 409–416.
Chaboche, J. L., 1991, “On Some Modiﬁcations of Kinematic Hardening to
Improve the Description of Ratchetting Effects,” International Journal of Plasticity,
Vol. 7, No. 7, pp. 661–678.
Chaboche, J. L., 1994, “Modelingof Ratchetting: Evaluationof Various Approaches,”
European Journal of Mechanics, A/Solids, Vol. 13, No. 4, pp. 501–518.
Fig. 8 Simulation of uniaxial ratchetting under the condition of
max
؍
280 MPa, R ؍ ؊0.75 and ˙ ؍ 10 MPa/s; (a)
i
؍ 0, (b)
i
؍ 0.02
Fig. 9 Simulation of cyclic tension (⌬⑀ ؍ 0.4 percent, ␦⑀
max
؍ 0.01
percent, ⑀˙ ؍ 5 ؋ 10
؊3
percent/s)
40 / Vol. 122, JANUARY 2000 Transactions of the ASME
Freed, A. D., and Walker, K. P., 1990, “Model Development in Viscoplastic
Ratchetting,” NASA Report TM102509, NASA.
Jiang, Y., and Sehitoglu, H., 1994a, “Cyclic Ratchetting of 1070 Steel under
Multiaxial Stress States,” International Journal of Plasticity, Vol. 10, No. 5, pp.
579–608.
Jiang, Y., and Sehitoglu, H., 1994b, “Multiaxial Cyclic Ratchetting under Multiple
Step Loading,” International Journal of Plasticity, Vol. 10, No. 8, pp. 849–870.
Jiang, Y., and Sehitoglu, H., 1996, “Modeling of Cyclic Ratchetting Plasticity, Part I:
Development of Constitutive Relations, Part II: Comparison of Model Simulations With
Experiments,” ASME Journal of Applied Mechanics, Vol. 63, No. 3, pp. 720–733.
Jiang, Y., and Kurath, P., 1996, “Characteristics of the ArmstrongFrederick Type
Plasticity Models,” International Journal of Plasticity, Vol. 12, No. 3, pp. 387–415.
Kobayashi, M., Ohno, N., and Igari, T., 1998, “Ratchetting Characteristics of
316FR Steel at High Temperature, Part II: Analysis of Thermal Ratchetting Induced
by Spatial Variation of Temperature,” International Journal of Plasticity, Vol. 14, No.
4–5, pp. 373–390.
Krieg, R. D., and Krieg, D. B., 1977, “Accuracies of Numerical Solution Methods
for the ElasticPerfectly Plastic Model,” ASME Journal of Pressure Vessel Technol
ogy, Vol. 99, No. 4, pp. 510–515.
McDowell, D. L., 1994, “Description of Nonproportional Cyclic Ratchetting Be
havior,” European Journal of Mechanics, A/Solids, Vol. 13, No. 5, pp. 593–604.
McDowell, D. L., 1995, “Stress State Dependence of Cyclic Ratchetting Behavior
of Two Rail Steels,” International Journal of Plasticity, Vol. 11, No. 4, pp. 397–421.
Ohno, N., 1990, “Recent Topics in Constitutive Modeling of Cyclic Plasticity and
Viscoplasticity,” Applied Mechanics Reviews, Vol. 43, No. 11, pp. 283–295.
Ohno, N., and Wang, J.D., 1993, “Kinematic Hardening Rules with Critical State
of Dynamic Recovery, Part I—Formulation and Basic Features for Ratchetting
Behavior, Part II—Application to Experiments of Ratchetting Behavior,” Interna
tional Journal of Plasticity, Vol. 9, No. 3, pp. 375–403.
Ohno, N., and Wang, J.D., 1994, “Kinematic Hardening Rules for Simulation of
Ratchetting Behavior,” European Journal of Mechanics, A/Solids, Vol. 13, No. 4, pp.
519–531.
Ohno, N., 1998, “Constitutive Modeling of Cyclic Plasticity with Emphasis on
Ratchetting,” International Journal of Mechanical Sciences, Vol. 40, No. 2–3, pp.
251–261.
Ohno, N., AbdelKarim, M., Kobayashi, M., and Igari, T., 1998, “Ratchetting
Characteristics of 316FR Steel at High Temperature, Part I: StrainControlled Ratch
etting Experiments and Simulations,” International Journal of Plasticity, Vol. 14, No.
4–5, pp. 355–372.
Takahashi, Y., and Tanimoto, K., 1995, “Ratchetting Analysis Pertaining to Fast
Breeder Reactor Design,” Proceedings of Plasticity ’95: The Fifth International
Symposium on Plasticity and its Current Applications, Gordon and Breach, Luxem
bourg, pp. 653–656.
Wang, H., and Barkey, M. E., 1998, “Strain Space Formulation of the Armstrong
Frederick Family of Plasticity Models,” ASME JOURNAL OF ENGINEERING MATERIALS
AND TECHNOLOGY, Vol. 120, No. 3, pp. 230–235.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 41
Xiao Lin
State Key Laboratory for
Mechanical Behavior of Materials,
Xi’an Jiaotong University,
Xi’an 710049,
P.R. China
Cyclic Deformation Behavior
and Dislocation Substructures
of Hexagonal Zircaloy4 Under
OutofPhase Loading
Macroscopic response and microscopic dislocation structures of Zr4 subjected to biaxial
fatigue under different phase angles of 30°, 60°, 90°, and different equivalent strain
ranges of 0.8%, 0.6%, 0.4% were studied. The testing results show that the delay angle
between the stress deviators and strain increment tensors is strongly dependent on phase
angle and the equivalent strain range. When phase angle equals 60°, the delay angle has
the minimum variation range for all specimens. The mean value of the delay angle
decreases with increasing phase angle or the equivalent strain range. The variation range
and average value of the Mises equivalent stress have the maximum in S3 with the phase
angle of 90°. They decrease as the equivalent strain range decreases. Zr4 displays a
pronounced initial hardening followed by a continuous softening for all specimens during
outofphase cycling. The stabilized saturation stresses of Zr4 under outofphase cycling
are much higher than that under uniaxial cycling. It indicates that Zr4 displays an
obvious additional hardening. As the phase angle increases, the typical dislocation
structure changes from dislocation cells to tangles. The dislocationdislocation interac
tions increase resulting in an additional hardening. In essence, the degree of additional
hardening depends, among other factors, on the maximum shear stress ratio of resolved
shear stresses and critical resolved shear stresses (RSS/CRSS).
1 Introduction
It has been observed in a few of experimental studies that
outofphase strain cycling may result in cyclic hardening beyond
the extent obtained in inphase biaxial and uniaxial tests at the
same peak effective strain level (Benallal and Marquis, 1987;
Bettge, 1995; Doong, 1990, 1991; Doquet, 1990, 1993; Feaugas
and Clavel, 1997; Kida and Ohnami, 1997; Krempl, 1984, 1989;
McDowell, 1988; Nishino and Ohnami, 1986). Further, the degree
of addition cyclic hardening is material dependent, and it is a
complex function of the loading history. Up to now, the degree of
outofphase cyclic hardening can be accurately estimated for only
a limited number of loading paths. The inability of many consti
tutive equations to effectively model the outofphase hardening
behavior appears to result from the lack of knowledge on micro
scopic deformation mechanism.
A recent study by Kida and Ohnami (1997) shows that a ladder
or labyrinth structure is a common structure for type 304 stainless
steel in inphase straining and a cell in the outofphase straining.
Doong (1990) showed that the multislip structures (cells and
subgrains) were observed in aluminum under proportional and
nonproportional loading and no additional hardening was dis
played. For copper and stainless steel, singleslip structures (planar
dislocations, veins and ladders) were observed after proportional
loading, whereas multislip structures (cells and labyrinths) after
nonproportional loading were found. The increased cyclic harden
ing of copper and stainless steel under nonproportional loading is
attributed to the change of plastic deformation mode from planar to
wavy slip. It was suggested by Bettge (1995) that the dislocation
walls and cells were the typical substructures for IN738LC after
proportional loading, while, it is tangled owing to the multiple slip
under nonproportional loading. So far, it is widely accepted that
the main parameter governing the degree of nonproportional hard
ening in solid solution steel is the ease of cross slip (Bettge, 1995;
Doong, 1990; Kida, 1997). The materials with lower stacking fault
energies exhibit a signiﬁcant addition cyclic hardening under non
proportional loading. Other materials with higher stacking fault
energies have the same amount of hardening for both proportional
and nonproportional cycling (Itoh, 1992). However, all of works
are limited on cubic metals, for example copper and stainless steel
etc. Especially, little is known regarding the plastic deformation
behavior of hexagonal closepacked metals under biaxial nonpro
portional loading. The slip systems are neither so numerous nor so
dispersal in hexagonal metals as in cubic metals. Some anisotropic
plastic deformation behavior is, therefore, expected. The objective
of this work is to investigate the plastic deformation behavior of
Zr4 resulting from the different outofphase loading paths, and
provides some microscopic structural evidence of additional hard
ening.
2 Experimental Procedure
The material chosen for this study is a nuclear grade Zircaloy4.
The chemical composition of this alloy is (in wt%): Sn1.5, Fe
0.21, Cr0.1, O0.09, Zrbalance. The initial texture analysis shows
a preferential crystallographic orientation of the prism pole (Fig.
1). Thinwalled tubular specimens of outside diameter 19 mm,
inside diameter 16 mm, and gage length 50 mm were machined
from a recrystallized bar of Zr4. The mechanical tests were
carried out on a MTS axialtorsion servohydraulic machine. All
tests were conducted at room temperature in strain control with a
von Mises equivalent strain rate of 3 ϫ 10
Ϫ3
S
Ϫ1
. Strain on the
gage length was measured via a biaxial extensometer. Both axial
and torsional strains were sinusoidal waves. The phase lag, (),
between the two sinusoidal signals was 30°, 60° and 90°, respec
tively. Recording of test results was accomplished with two XY
plotters.
Two groups of loading modes shown in Fig. 2 were employed
as a test program. The ﬁrst test program is the elliptical loading
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
November 3, 1998; revised manuscript received August 2, 1999. Associate Technical
Editor: D. Marquis.
42 / Vol. 122, JANUARY 2000 Transactions of the ASME Copyright © 2000 by ASME
paths with the same von Mises equivalent strain range and the
different phase angles of 30°, 60°, 90°, so as to examine the effect
of the phase angles (Fig. 2(a)). The second program examines how
the equivalent strain range affects cyclic deformation behavior in
circular loading paths (Fig. 2(b)). The test program is summarized
in Table 1, where it can be seen that three phase lags and three
equivalent strains were tested.
After biaxial fatigue failure, thin foils were cut from specimens
perpendicular to the axial direction by a spray machine to observe
dislocation substructures. Subsequently, these foils were mechan
ically and electrolytically thinned. Finally, the dislocation sub
structures were examined using a JEOL200CX transmission elec
tron microscope operated at 200 kV.
3 Experimental Results
3.1 Coaxiality of the Stress Deviators and the Strain In
crement Tensors. Under combined tensiontorsion of the thin
walled tubular specimens, the deviator stress, , and strain tensors,
⑀, can be conveniently described as two dimensional vectors in the
subspaces of the full vector spaces (Xiao, 1996)
ϭ
a
n
1
ϩ ͱ3n
3
(1)
⑀ ϭ ⑀
a
n
1
ϩ ͑␥/ ͱ3͒n
3
(2)
where
a
, are axial and shear stresses, respectively. ⑀
a
, ␥ are
axial and shear strains. The base vectors n
1
and n
3
are the unit
vectors along the axes ⑀
a
and ␥/
͌
3.
A strain increment vector can be deﬁned as
d⑀ ϭ d⑀
a
n
1
ϩ d͑␥/ ͱ3͒n
3
(3)
The accumulated strain is deﬁned as the length of strain path,
⌬l, shown as follows
⌬l ϭ
iϭ1
n
͓͑⑀
a,i
Ϫ ⑀
a,iϪ1
͒
2
ϩ ͑␥
i
Ϫ ␥
iϪ1
͒
2
/3͔
1/2
(4)
Regarding biaxial cyclic deformation, the equivalent stress
range, ⌬
eq
, and the equivalent strain range, ⌬⑀
eq
, are deﬁned as
⌬
eq
ϭ ͑
1
2
ϩ 3
2
͒
1/2
(5)
⌬⑀
eq
ϭ ͑⑀
a
2
ϩ ␥
2
/3͒
1/2
(6)
The coaxiality of the stress deviators and the strain increment
tensors is checked along the straining paths of combined tension
torsion specimens.
The orientation of strain increment vector with respect to the
unit vector n
1
is
d⑀
ϭ tan
Ϫ1
͑d␥/ ͱ3d⑀͒ (7)
The orientation of stress vector with respect to the unit vector n
1
is
ϭ tan
Ϫ1
͑ ͱ3/͒ (8)
The coaxiality of the stress deviators and the strain increment
tensors requires that
d⑀
ϭ
. Whereas, the stress deviator is often
delayed to the strain increment tensor under strain controlled
biaxial test. The delay angle equals
ϭ
d⑀
Ϫ
(9)
can be given by
Fig. 1 {101
0} Prism pole ﬁgure of the initial Zr4. RDrolling direction,
TDtransverse direction.
Fig. 2 Outofphase loading paths with different (a) phase angles, (b)
equivalent strain ranges
Table 1 Loading modes employed in this work
No. ⌬⑀
a
, (%) ⌬␥, (%) ⌬⑀
*
eq
, (%) , (°)
S1 0.581 0.865 0.794 30
S2 0.644 1.036 0.789 60
S3 0.788 1.383 0.798 90
S4 0.597 1.012 0.597 90
S5 0.403 0.642 0.403 90
⌬⑀
*
eq
is the maximum of Mises equivalent stress ranges for Zr4 under
outofphase loading; is the phase lag between axial and torsional strain
waves.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 43
ϭ cos
Ϫ1
ͩ
⅐ d⑀
͉ʈd⑀͉
ͪ
(10)
Figures 3(a)–(c) illustrate the variation of the delay angles with
strain path length for Zr4 under outofphase loading with the
same equivalent strain range of 0.8% and different phase angles. It
is revealed that the delay angles have the maximum variation
ranges in S1 with the minimum phase angle of 30° during cycling
(Fig. 3(a)). It is noteworthy that the variation range of delay angles
has the minimum in S2 with a phase angle of 60° (Fig. 3(b)). The
variation range of delay angle is larger in S3 with a phase angle of
90° than that in S2, whereas, it is smaller in S3 comparing with S1
(Fig. 3(c)). The larger the phase angle, the smaller the average
value of delay angles.
Figure 3(c) and Figs. 4(a), (b) depict the variation of delay
angles of Zr4 during cycling with the same phase angle and
different equivalent strain ranges. The variation range decreases,
as the equivalent strain range decreases, while the average value
increases.
It can be concluded from the above that the variation range and
average value of the delay angles are dependent on phase angle
and equivalent strain range. When phase angle equals 60°, the
delay angle has the minimum variation ranges for all the speci
mens. The mean value of delay angle decreases with increasing
phase angle or the equivalent strain range.
3.2 Macroscopic Stress Response Curves During Nonpro
portional Cycling. The effective stress response, ͉͉, vs. strain
path length curves for Zr4 under the same equivalent strain range
and different phase angles are depicted in Figs. 5((a)–(c)). The
approximate sinusoidal hardening rule is obtained during circular
and elliptical loading. When phase angle equals 60°, the variation
range of effective stress has the minimum value (Fig. 5(b)).
Whereas, the variation range and the average value of effective
stress in S3 with phase angle of 90° have the maximum (Fig. 5(c)).
They decrease as the equivalent strain range decreases at the same
circular loading paths, as shown in Fig. 5(c) and Figs. 6(a), (b).
A plot of maximum Mises effective stress in each cycle as a
function of the logarithm of the number of cycle is shown in Fig.
7. Zr4 exhibits a pronounced initial hardening followed by a
continuous softening for all specimens. As phase angle and equiv
Fig. 3 Delay angle, , vs. path length increment, ⌬l, under the same
equivalent strain range of 0.8% and different phase angles. (a) 30°, (b) 60°
and (c) 90°
Fig. 4 Delay angle, , vs. path length increment, ⌬l, under circular
loading with different equivalent strain ranges. (a) 0.597%, (b) 0.403%.
44 / Vol. 122, JANUARY 2000 Transactions of the ASME
alent strain range increase, the degree of hardening and softening
is increased. It is worthy to note that no saturation stages are
obtained for Zr4 under biaxial outofphase cycling in comparison
with that under uniaxial cycling. In the latter, it is known that the
cyclic deformation behavior of Zr4 is characterized with three
stages of cyclic hardening, saturation and cyclic softening, and the
ration of three stages depends on cyclic strain range and the testing
temperature (Xiao, 1998).
3.3 Cyclic Hardening Under OutofPhase Loading. Typ
ical cyclic stressstrain curve of Zr4 during biaxial tensiontorsion
cycling is shown in Fig. 8. For comparison, the uniaxial monotonic
and cyclic stressstrain curves are also included, which are quoted
from the previous work (Xiao, 1999). They were obtained at a
constant total equivalent strain rate of 3 ϫ 10
Ϫ3
S
Ϫ1
. Each datum
point under uniaxial and biaxial cycling in the ﬁgure represents an
individual specimen. Mises criterion is used in the deﬁnition of the
equivalent stress and strain. There is a phase lag between the axial
and shear stress loads in the case of outofphase straining. The
equivalent stabilized cyclic stress for specimens under nonpropor
Fig. 5 Equivalent stress response, ͦͦ, versus path length increment, ⌬l,
under the same equivalent strain range of 0.8% and different phase
angles. (a) 30°, (b) 60°, and (c) 90°.
Fig. 6 Equivalent stress response, ͦͦ, versus path length increment, ⌬l,
under circular loading with the different equivalent strain ranges. (a)
0.597% and (b) 0.403%.
Fig. 7 Peak stress versus cyclic number curves of Zr4 under different
phase angles and different equivalent strain ranges
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 45
tional loading is, therefore, the maximum value of the von Mises
stresses. The equivalent strain range is the von Mises equivalent
strain at the place of the corresponding maximum Mises stress,
because the maximum Mises stress is not synchronous with the
maximum Mises strain. The testing results show that the cyclic
hardening behavior is displayed for Zr4 by the fact that the cyclic
stressstrain curves lie above the monotonic one at both uniaxial
and biaxial cycling. Further study shows that there is an obvious
difference in the cyclic hardening levels of Zr4 under uniaxial and
biaxial outofphase cycling. An over 50 percent increase in the
stabilized stress has arisen, when Zr4 deformed under outof
phase loading in comparison with uniaxial cycling. These results
imply that hcp Zr4 displays an obvious additional hardening. The
amount of hardening, which is given simply by the difference of
the stress in the stressstrain curves, increases as the equivalent
strain range increases. The biaxial test data under circular paths at
the same phase angle of 90° and different equivalent strain ranges
can be described in a powerlaw curve. However, the degree of
additional hardening is decreased with decreasing the phase angle
at the same equivalent strain ranges. The specimen S3 with the
maximum phase angles of 90° has the highest additional harden
ing. When phase angle equals 30° and 60°, the difference of
hardening degree is small between them.
3.4 Dislocation Substructures. The dislocation substruc
tures were examined for Zr4 after biaxial cyclic deformation.
Typical dislocation structures produced by deforming Zr4 speci
mens under various phase angles and the same equivalent strain
range are shown in Figs. 9((a)–(c)). A few of elongated dislocation
cells, as shown in Fig. 9(a), were observed in S1 with the phase
angle of 30°. As the phase angle increases to 60°, cell walls
become thick and the amount of dislocations within cells increases
in S2 (Fig. 9(b)). When the phase angle equals 90°, no slip bands
or dislocation cells were observed, dislocations are homoge
neously distributed in the volume, and dislocation tangles are
welldeveloped (Fig. 9(c)). As phase angle increases at the same
equivalent strain range the dislocation structures change from
dislocation cells to dislocation tangles and the homogeneity of
dislocation distribution increases.
Figure 9(c) and Figs. 10((a), (b)) show the dislocation structures
formed in Zr4 specimens under circular loading paths with dif
ferent equivalent strain ranges. It is known that the typical dislo
cation conﬁguration is parallel dislocation lines resulting from
prismatic slippage for Zr4 under uniaxial cycling at RT (Xiao,
1997). Comparison of the dislocation substructures of Zr4 formed
in uniaxial cycling with biaxial cycling shows that nonproportional
loading changes slip modes of dislocation from planar to wavy
(Fig. 9(c)). With increasing strain range, multislip structures ap
pear to occur in circular nonproportional cycling of specimen S5
and S6 (Fig. 9(c) and Fig. 10(a)). In the lower equivalent strain
range, the typical dislocation structure is some individual disloca
tion lines. On close examination, some dislocation crossslip can
be determined in Fig. 10(b). However, cells and tangled structures
have been suggested to be impossible to occur in Zr4 under
uniaxial cycling at RT, because their formations need pyramidal
slip to be activated. Pyramidal slippage is much difﬁcult for Zr4
fatigued at RT (Xiao, 1997), and takes place only at the much high
strain range and the elevated testing temperature. The major effect
of the outofphase loading on Zr4 is, therefore, to decrease the
minimum strain range or the testing temperature for operating
pyramidal slip and forming multislip structures such as dislocation
cells and tangles.
Fig. 9 Dislocation structures of Zr4 under the same equivalent strain
range of 0.8% and different phase angles. (a) 30°, g ؍ 101
0, incident
beam ʈ[1
21
9], (b) 60°, g ؍ 101
0, incident beam ʈ[1
21
9] and (c) 90°, g ؍
101
0, incident beam ʈ[1
21
9].
Fig. 8 Inﬂuence of the degree of nonproportionality and equivalent
strain ranges on the additional hardening. Auniaxial, Mmonotonic,
Ccyclic, Bbiaxial
46 / Vol. 122, JANUARY 2000 Transactions of the ASME
4 Discussion
Under outofphase loading, cyclic hardening is stronger than
that in proportional loading. The different cyclic hardening behav
ior can be rationalized by considering that the plane of maximum
shear strain (maximum slip) changes with the direction of strain
ing. This raises a question why some materials display the signif
icant additional hardening such as austenite stainless steel (Doong,
1990; Kida, 1997). Whereas, the others show a slight or no
additional hardening such as aluminum alloy (Doong, 1990;
Doner, 1974) for the same cubic metals under nonproportional
cycling. The additional hardening during nonproportional loading
is, therefore, material dependent. This material dependence cannot
be explained by the increase of dislocationdislocation interactions
due to the rotation of maximum shear planes during nonpropor
tional cycling. Another question is that stainless steel with lower
stackingfault energy displays planar slip. However, the deforma
tion mode changes from planar to wavy slip, when the cyclic strain
range is enough large under uniaxial cycling (Xiao, 1996). In this
case, dislocation cells can be formed under inphase loading,
which is the same as that formed under outofphase loading. Thus,
the additional hardening can not be expected. The slip systems
distribute homogeneously in the threedimension space for cubic
metals. Under outofphase cycling, the amount and the type of the
activated slip systems are the similar as the maximum shear strain
plane rotates in the space for the cubic metals with the higher
stacking fault energies. Calculations have shown that in the fully
outofphase case usually ﬁve of the twelve possible slip systems
of face central cubic metals in each grain will experience maxi
mum Schmid factors at some point in each cycle (McDowell,
1988), i.e. ﬁve independent plastic deformation modes are pro
vided to sufﬁcient to fulﬁll von Mises criterion, and the difference
of the critically resolved shear stress (CRSS) between different slip
systems is small. Thus, the additional hardening cannot occur for
these materials under outofphase loading. Only for cubic metals
with lower stackingfault energies, the slippage is generally re
stricted on the stacking fault plane. The singleslip structures such
as veins are predominant under a smallstrain proportional loading.
The deformation mode shows an obvious anisotropic. It will ex
perience a change of plastic deformation mode from planar slip
under proportional loading to wavy slip under outofphase load
ing. Multislip and crossslip should result in numerous
dislocationdislocation intersections, increasing the dislocation
dislocation interactions between different slip planes and promote
signiﬁcant additional hardening.
The macroscopic deformation behavior of Zr4 under outof
phase loading can be elucidated by considering the microscopic
deformation mechanism. Zr4 is a hexagonal close packed metals.
It has been known that only the prismatic slip is activated for Zr4
under uniaxial cycling at RT (Pochettino, 1992; Tenckhoff, 1978;
Xiao, 1997). The prismatic slip only offers two independent slip
modes resulting in a planar slip character. During outofphase
cycling, glide on the other systems like pyramidal or basal planes
with the ͗c͘ or ͗c ϩ a͘ component dislocations is expected to play
a major role as a result of the maximum shear strain plane rotating.
Otherwise, in Zr4 the operable deformation system changes from
single slip under inphase cycling to multislip and cross slip during
outofphase cycling. The number of active slip systems increases
by reducing the mean free path of the mobile dislocations. As a
result multislip structures such as dislocation tangles become
dominant. The dislocationdislocation interactions between differ
ent slip planes increase. The high additional hardening is antici
pated. On the other hand, the critical resolved shear stresses
(CRSS) are different in hcp metals for various slip systems. The
slip plane listed in the order of CRSS from low to high is prismatic
plane, pyramidal plane and basal plane (Pochettino, 1992; Tenck
hoff, 1978; Xiao, 1997). The CRSS of pyramidal and basal slips is
higher than that of prismatic slip. Therefore, it is believed that the
applied stress to operate the new slip systems other than prismatic
slip is increased. The increasing applied stress contributes to the
additional hardening. In essence, whether additional hardening
depends, among other factors, on ratio of the resolved shear
stresses (RSS) and the CRSS. As the maximum shear stress plane
rotates during outofphase cycling, deformation systems with a
maximum shear stress ratio RSS/CRSS become operative. The
CRSS is different for various slip planes of cubic metals with the
lower stacking fault energy and hcp metals. In this case, an
obvious additional hardening is expected. For cubic metals with
the higher stacking fault energy, the CRSS of different slip systems
is approximately constant. Little additional hardening is antici
pated. The degree of additional hardening is dependent on the
variation of the shear stress ratio RSS/CRSS. When phase angle is
60°, equal resolved shear stresses for different deformation sys
tems are presented. Therefore, the degree of additional hardening
is small for Zr4 in S2. The RSS/CRSS is increased as the phase
angles and the equivalent strain range increase, the degree of
additional hardening is, therefore, increased.
5 Conclusions
1. The stress deviators are retarded with the strain increment
tensors for Zr4 under outofphase straining. The variation range
and average value of the delay angle are dependent on the phase
lag and the equivalent strain range. The delay angles have the
maximum variation ranges with the minimum phase angle of 30°.
When the phase angle equals 60°, the variation of delay angles has
the minimum. The larger the phase angle, the smaller the average
value of delay angles. The variation range of delay angles de
creases as the equivalent strain range decreases, while the average
value increases.
2. Macroscopic stress response curves show that the approx
imate sinusoidal hardening rule is obtained for Zr4 during ellip
tical and circular cycling. The variation ranges and mean values of
effective stresses have the maximum, when phase angle equals
90°. They decrease as the equivalent strain range decreases at the
same circular loading paths.
Fig. 10 Dislocation structures of Zr4 under circular loading with the
different equivalent strain ranges. (a) 0.597%, g ؍ 101
0, incident beam
ʈ[2
42
3] and (b) 0.403%, g ؍ 101
0, incident beam ʈ[1
21
1].
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 47
3. Zr4 exhibits a pronounced initial hardening followed by a
continuous softening for all specimens. As the phase angle and
equivalent strain range increase, the degree of hardening and
softening increases. No saturation stage is obtained for Zr4 under
biaxial outofphase cycling in comparison with uniaxial cycling.
4. An obvious additional hardening is displayed for Zr4 under
outofphase cycling by the fact that the cyclic stressstrain curve
lies above the uniaxial cyclic one. The hardening degree increases
as the equivalent strain range and the phase angle increase.
5. The typical dislocation structures change from dislocation
cells to dislocation tangles and the homogeneity of dislocation
distribution increases, as the phase angle increases. It remains the
dislocation tangles under circular paths at the different equivalent
strain ranges. The smaller the equivalent strain range, the lower the
dislocation density. The major effect of the outofphase loading
on Zr4 is to decrease the minimum strain range or the testing
temperature for operating pyramidal slip and forming multislip
structures such as dislocation cells and tangles.
Acknowledgments
The ﬁnancial support from the National Natural Science Foun
dation and the Nuclear Industrial Science Foundation of China is
gratefully acknowledged. I am also grateful for the ﬁnancial sup
port provided by the State Key Laboratory of Structural Strength
and Vibration, Xi’an Jiaotong University.
References
Benallal, A., and Marquis, D., 1987, “Constitutive Equations for Nonproportional
Cyclic ElastoViscoplasticity,” ASME JOURNAL OF ENGINEERING MATERIALS AND TECH
NOLOGY, Vol. 109, No. 4, pp. 326–336.
Bettge, D., Osterle, W., and Ziebs, J., 1995, “Temperature Dependence of Addi
tional Cyclic Hardening of a NickelBase Superalloy during OutofPhase Multiaxial
Deformation,” Scripta Metallurgica Materialia., Vol. 32, No. 10, pp. 1601–1606.
Doner, M., Chang, H., and Conrad, H., 1974, “Plastic Flow and Dislocation
Structure at Small Strains in OFHC Copper Deformed in Tension, Torsion and
Combined TensionTorsion,” Journal Mechanics Physics Solids, Vol. 22, pp. 555–
573.
Doong, S. H., Socie, D. F., and Robertson, I. M., 1990, “Dislocation Substructures
and Nonproportional Hardening,” ASME JOURNAL OF ENGINEERING MATERIALS AND
TECHNOLOGY, Vol. 112, No. 4, pp. 456–464.
Doong, S. H., and Socie, D. F., 1991, “Constitutive Modeling of Metals under
Nonproportional Cyclic Loading,” ASME JOURNAL OF ENGINEERING MATERIALS AND
TECHNOLOGY, Vol. 113, No. 1, pp. 23–30.
Doquet, V., and Pineau, A., 1990, “Extra Hardening Due to Cyclic Non
proportional Loading of an Austenitic Stainless Steel,” Scripta Metallurgica Mate
rialia, Vol. 24, No. 3, pp. 433–438.
Doquet, V., 1993, “Twinning and Multiaxial Cyclic Plasticity of a Low Stacking
FaultEnergy F.C.C. Alloy,” Acta Metallurgica Materialia, Vol. 41, No. 3, pp.
2451–2459.
Feaugas, X., and Clavel, M., 1997, “Cyclic Deformation Behavior of an ␣/
Titanium Alloy (Parts I and II),” Acta Materialia, Vol. 45, No. 7, pp. 2685–2701 and
pp. 2703–2714.
Itoh, T., Sakane, M., Ohnami, M., and Ameyama, K., 1992, “Additional Hardening
Due to Nonproportional Cyclic Loading—A Contribution of Stacking Fault Energy,”
Mecamat’92, Inter. Seminar on Multiaxial Plasticity, LMTCachan, France, pp.
43–50.
Kida, S., Itoh, T., Sakane, M., Ohnami, M., and Socie, D. F., 1997, “Dislocation
Structure and Nonproportional Hardening of Type 304 Stainless Steel,” Fatigue &
Fracture of Engineering Materials & Structures, Vol. 20, No. 10, pp. 1375–1386.
Krempl, E., and Lu, H., 1984, “The Hardening and RateDependent Behavior of
Fully Annealed AISI Type 304 Stainless Steel Under Biaxial InPhase and Outof
Phase Strain Cycling at Room Temperature,” ASME JOURNAL OF ENGINEERING MA
TERIALS AND TECHNOLOGY, Vol. 106, No. 4, pp. 376–381.
Krempl, E., and Lu, H., 1989, “The Path and Amplitude Dependence of Cyclic
Hardening of Type 304 Stainless Steel at Room Temperature,” Biaxial and Multiaxial
Fatigue, EGF3, M. W. Brown and K. J. Miller, eds., Mechanical Engineering
Publications, London, pp. 89–106.
McDowell, D. L., Stahl, D. R., Stock, S. R., and Antolovich, S. D., 1988, “Biaxial
Path Dependence of Deformation Substructure of Type 304 Stainless Steel,” Metal
lurgical Transactions A, Vol. 19, No. 5, pp. 1277–1292.
Nishino, S., Hamada, N., Sakane, M., Ohnami, M., Matsumura, N., and Tokizane,
M., 1986, “Microstructural Study of Cyclic Strain Hardening Behavior in Biaxial
Stress States at Elevated Temperature,” Fatigue & Fracture of Engineering Materials
& Structures, Vol. 9, No. 1, pp. 65–77.
Pochettino, A. A., Gannio, N., Edwards, C. V., and Penelle, R., 1992, “Texture and
Pyramidal Slip in Ti, Zr and Their Alloys,” Scripta Metallurgica Materialia, Vol. 27,
No. 12, pp. 1859–1863.
Tenckhoff, E., 1978, “The Development of the Deformation Texture in Zirconium
During Rolling in Sequential Passes,” Metallurgical Transactions A, Vol. 9, No. 10,
pp. 1401–1412.
Xiao, L., and Kuang, Z. B., 1996, “Biaxial Path Dependence of Macroscopic
Response and Microscopic Dislocation Substructure in Type 302 Stainless Steel,”
Acta Materialia, Vol. 44, No. 8, pp. 3059–3067.
Xiao, L., and Gu, H. C., 1997, “Dislocation Structures in Zirconium and Zircaloy4
Fatigued at Different Temperatures,” Metallurgical Materials Transactions, Vol.
28A, No. 4, pp. 1021–1033.
Xiao, L., and Gu, H. C., 1998, “Plastic Energy Dissipation Model for Lifetime
Prediction of Zirconium and Zircaloy4 Fatigued at RT and 400°C,” ASME JOURNAL
OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 120, No. 2, pp. 114–118.
Xiao, L., and Gu, H. C., 1999, “Low Cycle Fatigue Properties and Microscopic
Deformation Structures of Zircaloy4 in Recrystallized and Stressrelieved Condi
tions,” Journal Nuclear Materials, Vol. 265, pp. 213–217.
48 / Vol. 122, JANUARY 2000 Transactions of the ASME
Magnus Ekh
Anders Johansson
1
Hans Thorberntsson
2
B. Lennart Josefson
Department of Solid Mechanics,
Chalmers University of Technology,
SE41296 Go¨teborg, Sweden
email: lejo@solid.chalmer.se
Models for Cyclic Ratchetting
Plasticity—Integration and
Calibration
Three wellknown ratchetting models for metals with different hardening rules were
calibrated using uniaxial experimental data from Bower (1989), and implemented in the
FE code ABAQUS (Hibbitt et al., 1997) to predict ratchetting results for a tensiontorsion
specimen. The models were integrated numerically by the implicit Backward Euler rule,
and the material parameters were calibrated via optimization for the uniaxial experimen
tal data. The algorithmic tangent stiffnesses of the models were derived to obtain efﬁcient
FE implementations. The calculated results for an FE model of the tensiontorsion
specimen were compared to experimental results. The model proposed by Jiang and
Sehitoglu (1995) showed the best agreement both for the uniaxial and the structural
component case.
1 Introduction
The rolling contact fatigue between railway wheels and rails
may result in initiation of fatigue cracks on, or close to, the surface
of the rail head. These surface cracks may eventually grow into the
rail head, and also possibly branch, which may ultimately lead to
sudden breakage of the rail. This form of damage is not associated
with material or manufacturing defects or imperfections. It is
rather a consequence of the increased loads (both magnitudes and
intensity) in modern rail trafﬁc. The annual cost for rail removal
and renewal including preventive grinding is very high, hence
there is a strong need for methods that can predict the occurrence
of rolling contact fatigue cracks in rails. This, in turn, calls for
efﬁcient simulation tools that use constitutive relations that can
accurately model the material behavior in the rail head during the
passing of railway wheels.
For a material subjected to cyclic loading with nonzero mean
load, ratchetting and/or shakedown may occur. These phenomena
are described by among others Beynon and Kapoor (1996) and
Chaboche and Nouailhas (1989). Ratchetting in the material occurs
when deformation continues to grow during each cycle. This
enlarges the deformations in the material, and may lead to initia
tion of cracks. For some materials the deformation growth slows
down, stabilizing to a cycle where the material behaves completely
elastic. This is called elastic shakedown. If the deformation growth
stabilizes to a cycle where the material behaves both elastically
and inelastically, then the shakedown is denominated plastic.
Realistic macroscopic material models for the description of the
ratchetting characteristics in metals can be obtained within the
framework of plasticity, using the von Mises yield criterion com
bined with the Armstrong and Frederick (1966) type of nonlinear
hardening rule (AF). However, a well known fact (see McDowell,
1985; Bower, 1989; Chaboche and Nouailhas, 1989) is that the
AF hardening rule, in its pure form, is often insufﬁcient to model
cyclic material behavior. Therefore, Chaboche et al. (1979) in
cluded several kinematic hardening variables that follow the AF
hardening rule (besides one linear kinematic hardening variable) in
the plasticity model. Other modiﬁcations of the kinematic harden
ing rule have been suggested by, for example, Bower (1989),
Chaboche (1991), Jiang and Sehitoglu (1996a, b), Ohno and Wang
(1993) and McDowell (1985), to mimic the experimentally ob
served decaying ratchetting rate in engineering materials.
With increasing complexity of the phenomena modeled, the
number of parameters and the nonlinearity in the mathematical
models used tend to increase. These models consist of nonlinear
differential equations, which in general must be solved numeri
cally with some incremental algorithm. The implicit Backward
Euler algorithm is favored by authors, such as Ortiz and Popov
(1985), Chaboche and Cailletaud (1996), for large increments
because of its stability and accuracy properties. Since the Back
ward Euler method is implicit, a nonlinear system of equations
must be solved in each increment, which is often referred to as the
local problem.
Calibration of the mathematical models, i.e., determination of
the model parameters, can be done by either a mechanistic ap
proach or a simultaneous determination, as described by Mahnken
and Stein (1996). In the mechanistic approach the parameters are
given a physical interpretation, and a sequence of ideal experi
ments are performed (see Jiang and Sehitoglu, 1996b). The exper
iments are ideal in the sense that they are designed to activate only
a few of the parameters in the model. In the simultaneous deter
mination, a sufﬁcient number of independent experiments are
done, and all the parameters in the model are determined from a
leastsquares ﬁt.
In order to simulate the ratchetting behavior in a structural
component, the material model must be implemented into an
FEcode. Since the model is nonlinear, an iteration procedure on
the global (FE) level is necessary. An efﬁcient iteration procedure
is obtained by using the algorithmic tangent stiffness (ATS), which
gives the good second order local convergence properties of a
Newton method (cf. Simo and Taylor, 1985).
Previous works describe the material models for ratchetting.
However, it has not focused on the use in realistic FEanalyses of
structures, hence the numerical integration technique and the cor
responding local and global problems have not been addressed
sufﬁciently.
In this paper we investigate three different ratchetting models.
We integrate these models by using the BackwardEuler technique,
and show how to solve the local system of equations. Then the
material parameters are determined by a leastsquares ﬁt to uni
axial experimental data. Furthermore, we describe a general tech
nique for deriving the ATStensor by using results from the local
problem. Finally, the models are implemented in the commercial
FEcode ABAQUS (Hibbitt et al., 1997), and results from tension
1
Present Address: Frontec Research Technology AB, Gårdava¨gen 1, SE41250
Go¨teborg, Sweden.
2
Present Address: VOLVO Information Technology AB, SE40508 Go¨teborg,
Sweden.
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
November 2, 1998; revised manuscript received September 9, 1999. Associate Tech
nical Editor: H. Sehitoglu.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 49
Copyright © 2000 by ASME
torsion simulations of a specimen are compared to experimental
data. This tensiontorsion test is designed to resemble the loading
in a rail head during the passing of a railway wheel (cf. Bower and
Johnson, 1989, 1991).
2 Description of the Models
2.1 Preliminaries. Small strains are assumed in the formu
lation of the models. Hence, the total strain ⑀ can be additively
decomposed into an elastic ⑀
e
and an inelastic strain ⑀
p
⑀ ϭ ⑀
e
ϩ ⑀
p
(1)
Hookes law deﬁnes the linear relation between the elastic strain
and stress in the material
1
ϭ E
e
: ⑀
e
where E
e
ϭ 2GI
dev
ϩ K
b
␦ ␦ (2)
where G is the shear modulus, and K
b
is the bulk modulus.
The stress , the backstress X and the dragstress K deﬁne the
stress state in the material. The material is elastic when ⌽(, X,
K) Ͻ 0 and plastic when ⌽(, X, K) ϭ 0, where the yield
function ⌽ is deﬁned according to von Mises as
⌽͑, X, K͒ ϭ ͱ
3
2
͉
dev
͉ Ϫ K Ϫ
Y
with
dev
ϭ
dev
Ϫ X,
͉
dev
͉ ϭ ͱ
dev
:
dev
(3)
where
Y
is the initial uniaxial yield stress.
The development of the plastic strain ⑀
p
follows the associative
ﬂow rule
⑀˙
p
ϭ
˙
ͱ
3
2
n
dev
and n
dev
ϭ
def
dev
͉
dev
͉
(4)
where the plastic multiplier
˙
was introduced.
The yield condition can be summarized in KuhnTucker form as
˙
Ն 0, ⌽ Յ 0, ⌽
˙
ϭ 0 (5)
2.2 Kinematic Hardening Laws. The different evolution
laws for the kinematic hardening that we examine in this paper, are
all based on the nonlinear Armstrong and Frederick (1966) law
X
˙
ϭ
˙
͑c ͱ
2
3
n
dev
Ϫ ␥X͒ (6)
where c and ␥ are material parameters. When ␥ ϭ0, the kinematic
hardening is purely linear. On the other hand, if ␥ 0, this law is
nonlinear and gives a constant ratchetting rate for a stress con
trolled cyclic test with constant stress amplitude and nonzero mean
stress (if no other type of hardening is included).
The kinematic hardening rule according to Bower in Table 1
includes an additional state variable Y, which gives the possibility
to model decaying and ﬁnally arresting ratchetting rate.
Finally, the kinematic hardening rule according to Jiang and
Sehitoglu (1995) is summarized in Table 2. This rule includes
several backstresses X
(i)
, and the total backstress X is obtained
simply by adding all X
(i)
. The evolution law for the backstresses
makes it possible to simulate a decreasing ratchetting rate. Note
that this kinematic hardening law has been further developed and
reﬁned, (see Jiang and Sehitoglu, 1996a, b). By introducing a
socalled memory surface, the effect of a sudden change of the
stress magnitude in a cyclic experiment can be accurately mod
elled.
2.3 Isotropic Hardening Law. We express the nonlinear
evolution law for the isotropic hardening as
K
˙
ϭ
˙
b
ͩ
1 Ϫ
K
K
ϱ
ͪ
(7)
where K
ϱ
is the saturated drag stress due to the isotropic harden
ing, and the parameter b governs the initial rate of the isotropic
hardening.
2.4 Ratchetting Models. We investigate three different
ratchetting models below. In the ﬁrst model, which we call A, the
ArmstrongFrederick kinematic hardening law (6) and the isotro
pic hardening law (7) are combined. The isotropic hardening law
is needed in this model to simulate decreasing ratchetting rate. This
model is included in the commercial FE package ABAQUS (Hib
bitt et al., 1997). In both, what we have called, model B and JS
only kinematic hardening is considered. The evolution laws for
these two models are described in Tables 1 and 2.
3 Integration of Models
3.1 General Procedure. In this section we describe a pro
cedure to apply the implicit Backward Euler (BE) integration
technique, in a strain controlled fashion, to model A, B, and JS.
This step is performed in each Gaussian point in the elements of an
FE mesh to determine the stress state for a given strain increment.
In order to evaluate whether the material response is elastic or
plastic the KuhnTucker type of conditions (5) are used. We
assume that the response is elastic when
˙
ϭ 0, and therefore
⌬ ϭ
nϩ1
Ϫ
n
ϭ 0, therefore the hardening stresses do not
change, i.e.
nϩ1
K ϭ
n
K and
nϩ1
X ϭ
n
X. This is true if ⌽ Յ 0, and
hence a new strain step can be taken. However, if ⌽ Ͼ 0 the
assumption does not hold, and a set of nonlinear equations ⌽( x) ϭ
0 must be solved. We refer to this system of equations as the local
problem. Note that the local problem for model JS can be reduced
to 7 ϩ 2M equations, thus dim (⌽) ϭ 7 ϩ 2M. In the case of
constant exponents
(i)
a reduction to 7 ϩ M equations is possible
(see Chaboche and Cailletaud, 1996). The unknowns x are deter
mined by the NewtonRaphson method
x
͑kϩ1͒
ϭ x
͑k͒
Ϫ ͑J͒
Ϫ1 nϩ1
⌽ where J ϭ
Ѩ
nϩ1
⌽
Ѩx
͑k͒
(8)
1
The operators “:” and “R” deﬁnes the contraction a : b ϭ a
ij
b
ij
and the open
product (a R b)
ijkl
ϭ a
ij
b
kl
, respectively.
Table 2 Kinematic hardening rule due to Jiang and
Sehitoglu (1995)
Development of the backstresses X
(i)
:
X
˙
͑i͒
ϭ
ͱ
3
2
˙
c
͑i͒
ͩr
͑i͒
n
dev
Ϫ
ͭ
͉X
͑i͒
͉
r
͑i͒ ͮ
͑i͒
X
͑i͒
ͪ, i ϭ 1, 2, . . . , M
Deﬁnition of the exponents
(i)
:
͑i͒
ϭ
0
͑i͒
ͫ
2 Ϫ n
dev
:
X
͑i͒
͉X
͑i͒
͉
ͬ
, i ϭ 1, 2, . . . , M
The total backstress X:
X ϭ
iϭ1
M
X
͑i͒
Material parameters: r
(i)
, c
0
(i)
,
0
(i)
, i ϭ 1, 2, . . . , M
Table 1 Kinematic hardening rule according to Bower (1989)
Development of the backstress X:
X
˙
ϭ
˙
͑c ͱ
2
3
n
dev
Ϫ ␥
1
͑X Ϫ Y͒͒
Development of the additional variable Y:
Y
˙
ϭ
˙
␥
2
͑X Ϫ Y͒
Material parameters: c,␥
1
,␥
2
50 / Vol. 122, JANUARY 2000 Transactions of the ASME
After the local problem has been solved, the stress and the hard
ening stresses can be updated, and a new strain step can be taken.
The local problems and the stress updating are summarized in
Table 3 and Table 4 for model JS and in Appendix A for model
A and B. Note that the local problem of model A and B can both
be reduced to 1 scalar equation, see Tables A.1 and A.3.
4 Models Calibration
4.1 Calibration Procedure. The ratchetting models are cal
ibrated against a uniaxial stress controlled experiment for a cylin
drical specimen of rail steel (BS11A). The experiment was carried
out by Bower (1989). In the experiment the stress in the steel
varies in a sawtooth fashion (t) with a small offset value
m
,
and the output is given as the strain values ⑀ for m time values t
i
,
i ϭ 1, . . . , m.
The calibration is done by minimizing the objective function f,
which we deﬁne as the square root of the sum of the squared
distances between the experimental strain values ⑀ (t
i
) and the
corresponding calculated strain ⑀(t
i
)
f͑͒ ϭ
ͱ
iϭ1
m
͑⑀ ͑t
i
͒ Ϫ ⑀͑t
i
, ͒͒
2
(9)
where is the vector of material parameters.
We minimize the objective function by applying the Downhill
Simplex optimization algorithm (Nelder and Mead, 1965), which
is available in the program Matlab (The Math Works, 1996).
The method proposed by Jiang and Sehitoglu (1996b) to deter
mine the material parameters for model JS is not applicable here,
since the type of experimental data we use is not the same (in our
case all parameters are determined at the same time from one set
of experiment).
4.2 Calibration Results. To ﬁnd the global minimum of the
objective function, different initial guesses of the material param
eters were tried. This was particularly important for model JS
(M ϭ 3), which we assume is because of the large number of
material parameters used. When calibrating this model we found
that no variation of the yield stress was needed, and therefore we
used
y
ϭ 400.1 MPa which was obtained by Bower (1989).
The results of the calibrations are summarized in Table 5, and
the experimental data (dotted lines) and the results from the
calibrated models (solid lines) are shown in Figs. 1–3. Young’s
modulus E was for all the models chosen as the value from Bower
(1989), i.e., E ϭ 2.0982 ⅐ 10
11
Pa.
Figures 1 and 2 show that neither the A model nor the B model
can simultaneously model the nonlinear behavior of the material
and the correct ratchetting rate. However, if only one cycle is used
in the calibration a better ﬁt is obtained to that particular cycle, but
then the ratchetting rate and the ﬁt to the other cycles are far from
satisfactory, which is illustrated in Fig. 4. The isotropic hardening
Table 4 Updating stresses and backstresses for model JS
Updating of the stresses:
nϩ1
dev
ϭ
nϩ1
dev
e
Ϫ ͱ6
nϩ1
n
dev
G⌬,
nϩ1
m
ϭ
n
m
ϩ 3K
b
⌬⑀
m
nϩ1
X
͑i͒
ϭ a
X
͑i͒
͑
n
X
͑i͒
ϩ ͱ
3
2
c
͑i͒
r
͑i͒
⌬
nϩ1
n
dev
͒,
nϩ1
X ϭ
iϭ1
M
nϩ1
X
͑i͒
Table 3 Local problem for model JS
Nonlinear system of equations
nϩ1
⌽ ϭ 0:
nϩ1
⌽ ϭ
΄
ͱ
3
2
͉
nϩ1
dev
͉ Ϫ
y
nϩ1
dev
Ϫ ͱ
2
3
y
nϩ1
n
dev
0
͑i͒
͓2͉
nϩ1
X
͑i͒
͉ Ϫ a
X
͑i͒
͑n
dev
:
n
X
͑i͒
ϩ ͱ
3
2
c
͑i͒
r
͑i͒
⌬͔͒ Ϫ
nϩ1
͑i͒
͉
nϩ1
X
͑i͒
͉, i ϭ 1 . . . M
a
X
͑i͒
͉
n
X
͑i͒
ϩ ͱ
3
2
c
͑i͒
r
͑i͒
⌬n
dev
͉ Ϫ ͉
nϩ1
X
͑i͒
͉, i ϭ 1 . . . M
΅
where
nϩ1
dev
ϭ
nϩ1
dev
e
Ϫ
iϭ1
M
a
X
͑i͒n
X
͑i͒
Ϫ ⌬ ͱ
3
2
͑2G ϩ
iϭ1
M
a
X
͑i͒
c
͑i͒
r
͑i͒
͒
nϩ1
n
dev
nϩ1
dev
e
ϭ
n
dev
ϩ 2G⌬⑀
dev
, a
X
͑i͒
ϭ ͩ1 ϩ c
͑i͒
⌬
ͱ
3
2
ͭ
͉
nϩ1
X
͑i͒
͉
r
͑i͒ ͮ
nϩ1 ͑i͒
ͪ
Ϫ1
Unknown internal variables x:
x ϭ ͓⌬
nϩ1
n
dev
͉
nϩ1
X
͑1͒
͉
. . .
͉
nϩ1
X
͑M͒
͉
nϩ1
͑1͒
. . .
nϩ1
͑M͒
͔
T
Table 5 Summary of the parameter values from the calibrations and the cycles used for the different material models
Model Parameters Optimized cycles
A c ϭ 6.49 GPa, ␥ ϭ 0.81, b ϭ 10.7 MPa 15, 50, 100, 200, 400 and 600
K
ϱ
ϭ 22.8 MPa,
y
ϭ 543 MPa
B c ϭ 6.61 GPa, ␥
1
ϭ 1.28 15, 50, 100, 200, 400 and 600
␥
2
ϭ 0.56,
y
ϭ 550 MPa
JS c
(1)
ϭ 476, r
(1)
ϭ 126 MPa,
0
(1)
ϭ 0.91 1, 2, 15, 50, 100, 200, 400 and 600
c
(2)
ϭ 75.6, r
(2)
ϭ 63.4 MPa,
0
(2)
ϭ 7.8
c
(3)
ϭ 8.02, r
(3)
ϭ 81.4 MPa,
0
(3)
ϭ 3.1
y
ϭ 400.1 MPa (ﬁxed)
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 51
in the A model causes the desired decreasing ratchetting rate and
decreasing width of the stressstrain cycles. In Fig. 3 model JS
shows the best representation of both the curve shape, the width of
the curves and the ratchetting rate (accumulated strain). We un
derline that the results from this comparison are not surprising,
since the models differ much in complexity and number of model
parameters. Nevertheless this comparison motivates why a more
complex model such as model JS must be used. There are other
competitive models available in literature (see Chaboche, 1991;
Ohno and Wang, 1993).
To illustrate the ability to describe the correct ratchetting rate,
the mean strain ⑀
a
for the calibrated models A and JS are shown
in Fig. 5. Model B is for clarity not included in the ﬁgure, since its
result almost coincides with the result for model A. We deﬁne the
mean strain as the mean value of all the strain values in each cycle.
The agreement between the experimental values (o) and the cal
ibrated models is satisfactory for all the models. Note that this
behavior can not be simulated using Eq. (6) with ␥ ϭ 0, that is
using a linear kinematic hardening rule.
5 Finite Element Implementation
5.1 Preliminaries. Following a standard ﬁnite element pro
cedure, the structure is divided into elements, each with an as
sumed displacement ﬁeld. A linearization of the Principle of Vir
tual Displacements and use of the NewtonRaphson method yields
the global system of FEequations to be solved iteratively for the
displacement increment in one time (load) step (Bathe, 1996).
nϩ1
K
͑iϪ1͒
⌬U
͑i͒
ϭ
nϩ1
R Ϫ
nϩ1
F
͑iϪ1͒
(10)
where
nϩ1
K
(iϪ1)
is the tangent stiffness matrix,
nϩ1
R is the consis
tent nodal force (which is assumed to be known beforehand) and
nϩ1
F
(iϪ1)
is the internal nodal force vector consistent with the
present stress ﬁeld in each element.
Information about the material behavior must be supplied in the
tangent stiffness matrix and the internal nodal force vector. It should
be noted that to obtain the full convergence beneﬁts of the Newton
Raphson method in the solution of the global FEequations, an accu
rate tangent stressstrain matrix Ꮿconsistent with the numerical stress
integration used in deriving F must be used. This is accomplished by
employing the algorithmic tangent stiffness.
5.2 Algorithmic Tangent Stiffness. The ATStensor is de
ﬁned as:
Ꮿ ϭ
d
nϩ1
d
nϩ1
⑀
(11)
To obtain the ATStensor, we ﬁrst summarize the stress update for
the three ratchetting models using the following expression
nϩ1
ϭ
nϩ1
m
␦ ϩ
nϩ1
dev
e
ϩ f͑x͑
nϩ1
⑀͒͒ (12)
Fig. 2 Cyclic stressstrain response for model B after calibration
Fig. 1 Cyclic stressstrain response for model A after calibration
Fig. 3 Cyclic stressstrain response for model JS with M ؍ 3 after
calibration
Fig. 4 Cyclic stressstrain response for model A after calibration using
measure points from cycle 600 only
Fig. 5 Accumulated strain versus number of cycles for model A and JS
52 / Vol. 122, JANUARY 2000 Transactions of the ASME
where the functions f( x(
nϩ1
⑀)) are obtained from Table 4, Table
A.2, and Table A.4. Hence, the ATStensor becomes
Ꮿ ϭ E
e
ϩ
Ѩf
Ѩx
Ѩx
Ѩ
nϩ1
⑀
(13)
where Ѩx/Ѩ
nϩ1
⑀ is derived from the local problem
⌽͑
nϩ1
⑀, x͑
nϩ1
⑀͒͒ ϭ 0, ᭙
nϩ1
⑀ f
d⌽
d
nϩ1
⑀
ϭ
Ѩ⌽
Ѩ
nϩ1
⑀
ϩ J
Ѩx
Ѩ
nϩ1
⑀
ϭ 0 f
Ѩx
Ѩ
nϩ1
⑀
ϭ ϪJ
Ϫ1
Ѩ⌽
Ѩ
nϩ1
⑀
(14)
5.3 FE Simulation of Tension Torsion Test. The condition
in the material of a cylindrical specimen which is subjected to
combined tension (extension) and torsion, can be assumed (Bower
and Johnson, 1989, 1991) to be similar to the condition in the
material of rail heads subjected to combined rolling and sliding
contact by railway wheels. To evaluate the prediction capability of
the ratchetting models in such conditions, an FE model of a
cylindrical specimen was created in IDEAS (SDRC, 1994). The
geometry of the FE model and the applied load are shown in Fig.
6 and Fig. 7, respectively.
The FE model consists of two parts, one which is assumed to
behave elastically through the whole load history and another
where the plastic deformation occurs. The elastic part is built up of
two beam elements and thin shell elements (rigid interface ele
ments), which are connected to 20nodes solid elements
(C3D20R). A reduced integration technique is used to decrease the
CPUtime for the analysis. The iteration technique used for ﬁnding
the global equilibrium is the NewtonRaphson method or the
QuasiNewton method (Hibbitt et al., 1997). Moreover, when
solving the system of equations in (10), a direct wavefront solver
was employed, denoted “wavefront” in Table 6 below. In some
cases a direct multi wavefront sparse solver (Hibbitt et al., 1997)
was used, denoted “sparse” in Table 6. For the convergence
criteria, we have chosen the default values in ABAQUS for the
residuals. However, in Table 6 the tolerance of largest residual to
the corresponding average ﬂux norm is smaller (10
Ϫ8
) than the
default value (10
Ϫ4
) (Hibbitt et al., 1997).
The load is applied to the beam node at the end of the specimen
as prescribed displacement and torque. In total, the FEmodel
contains approximately 1100 elements, 5000 nodes and 15,700
degrees of freedom. Furthermore, there are 50 load cycles and in
each cycle 40 number of timesteps are taken. Also, it should be
noted that geometrically linear conditions are assumed.
Figures 9–11 show the calculated displacement and rotation of
the node where the load was applied. Only model JS gives a
satisfactory agreement between prediction and experimental data
from Bower (1989), see Fig. 8.
Although the trend from the calculations is quite clear, there are
some limitations and uncertainties in the FEcalculations which
could contribute to the poor correspondence between experiments
and predictions by use of model A and B. The approximation with
the elastic beam and shell elements in the FE model was created to
reduce the computation time. Further, the results of the computa
tions were found to be dependent on the chosen timestep, and the
predictions must therefore be seen as approximate. It is also noted
that, for model JS, the timestep had to be decreased during the
analysis to achieve convergence of the global iteration.
Table 6 summarizes the CPU time on an HP 735 unix work
station for the FE calculations of one load cycle. It is seen, as
expected, that use of an elastic stiffness matrix leads to excessively
large computational times for this case of a strongly weakening
Fig. 8 Experiment data from Bower (1989)
Fig. 6 The geometry of the FEmodel of the specimen
Fig. 7 Load cycle of tension and torsion, which approximates the stress
cycle of surface elements in rolling and sliding contact
Table 6 CPUtimes depending on solution method
(ABAQUS5.7). * ؍ convergence problems, calculation
stopped, ** ؍ memory problems, calculation stopped
Model
Iteration
method Stiffness
CPUtime,
tolerance ϭ 1e Ϫ 8
JS Newton ATS 2 h 22 min (sparse)
M ϭ 3 Newton Elastic 30 h 25 min (sparse)
QuasiNewton Elastic 2 h 23 min (wavefront)
QuasiNewton ATS 1 h 17 min (wavefront)
JS
M ϭ 1 Newton ATS 3 h 30 min (wavefront)
B Newton ATS 1 h 26 min (sparse)
Newton Elastic 93 h 11 min (sparse, **)
QuasiNewton Elastic 20 h 59 min (wavefront, *)
QuasiNewton ATS 32 min (wavefront)
A Newton ABAQUS
internal
39 min (sparse)
QuasiNewton ABAQUS
internal
31 min (wavefront)
Fig. 9 Result from FEcalculation with model A
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 53
material in the loading phase. For model JS the results for M ϭ
3 are displayed. Model B and A (ABAQUS internal version) use
lower computational times, primarily since a larger reduction of
the local problem is possible and that fewer state variables need to
be stored in every Gauss point. For a given material model the
QuasiNewton method combined with the use of the ATStensor
seems to be the best choice.
6 Concluding Remarks
From the graphs in Section 4.2, we conclude that the capability
of model A and B to mimic the ratchetting rate and the shape of the
stress strain cycles in a uniaxial experiment is not satisfactory. This
conclusion is in agreement with other references (see McDowell,
1985; Chaboche and Nouailhas, 1989) experience, at least for
model A. However, the more complex JS model, with a larger
number of material parameters, gives satisfactory results. There
are other sophisticated models available in literature (Chaboche,
1991; Ohno and Wang, 1993), which are not evaluated in this
paper.
The material parameters in the models were determined via
calibration using the uniaxial experimental data. To determine the
correct parameters, it is necessary to obtain enough independent
data by using measure points from several cycles, see Fig. 4. Still,
for model JS the optimal set of material parameters may be
difﬁcult to ﬁnd. This indicates that the total number of parameters
should be kept reasonably low, that is using a low value for M in
Table 2. In addition, the reduced local problem arising when
applying the fully implicit Backward Euler rule becomes more
involved, i.e., more unknowns and state variables are needed, for
model JS as compared to model A and B, which increases the total
computational time as seen in Table 6.
Also in the FE prediction of a tensiontorsion experiment on a
cylindrical specimen, with the calibrated material parameters,
model JS showed the best capability to mimic experimental
results. To perform the FE calculations efﬁciently the algorithmic
tangent stiffness was derived for the models. Still there are nu
merical uncertainties in the results due to the limited computer
power, which limits the realistic number of applied timesteps.
Finally one may note two limitations of the present approach
which may be subject to further improvement. The three
dimensional FEsimulation, which aimed at reﬂecting a realistic
loading situation, was based on values for entering material pa
rameters determined from a simpler uniaxial simulation. This was
due to the limited knowledge of the local behavior during the
tensiontorsion test, but mainly to the desire to reduce the com
plexity in the optimization process for the determination of the
material parameters. Moreover, the present paper employs engi
neering stresses and strains only. This seems to be a reasonable
approximation based on the levels of the stresses and strains in the
uniaxial and three dimensional experiments used for comparison.
Formally, large displacement, or large strain approach could have
been employed, but this was avoided to reduce the computer time
needed in the simulations. It is believed that the qualitative results
in the present paper will hold also when, for example, large
displacements are used.
Acknowledgments
We wish to acknowledge ﬁnancial support from the Brite Euram
IIIproject ICON (Integrated study of rolling CONtact fatigue).
Valuable advice and support with the FE calculations from Dr.
Hans Bjarnehed at Frontec Research and Technology (Go¨teborg,
Sweden) are gratefully acknowledged.
References
Armstrong, P. J., and Frederick, C. O., 1966, “A Mathematical Representation of
the Multiaxial Bauschinger Effect,” CEGB Report No. RD/B/N731.
Bathe, K. J., 1996, Finite Element Procedures, PrenticeHall, London.
Beynon, J. H., and Kapoor, A., 1996, “The Interaction of Wear and Rolling Contact
Fatigue. Reliability Assesment of Cyclic Loaded Engineering Structures,” Varna,
Bulgaria, Proc. NATO Advanced Research Workshop.
Bower, A. F., 1989, “Cyclic Hardening Properties of HardDrawn Copper and
Railway Steel,” J. of the Mechanics and Physics of Solids, Vol. 37, pp. 455–470.
Bower, A. F., and Johnson, K. L., 1989, “The Inﬂuence of Strain Hardening on
Cumulative Plastic Deformation in Rolling and Sliding Line Contact,” J. of the
Mechanics and Physics of Solids, Vol. 37, pp. 471–493.
Bower, A. F., and Johnson, K. L., 1991, “Plastic Flow and Shakedown of the Rail
Surface in Repeated WheelRail Contact,” Wear, Vol. 144, pp. 1–18.
Chaboche, J. L., 1991, “On Some Modiﬁcations of Kinematic Hardening to
Improve the Description of Ratchetting Effects,” Int. J. Plasticity, Vol. 7, pp.
661–678.
Chaboche, J. L., and Cailletaud, G., 1996, “Integration Methods for Complex
Plastic Constitutive Equations,” Comp. Meth. Appl. Mech. Engng., Vol. 133, pp.
125–155.
Chaboche, J. L., and Nouailhas, D., 1989, “Constitutive Modeling of Ratchetting
Effects—Part I,” ASME JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol.
111, pp. 384–392.
Chaboche, J. L., Dang Van, K., and Cordier, G., 1979, “Modelization of the Strain
Memory Effect on Cyclic Hardening of 316 Stainless Steel,” Transactions of the Fifth
International Conference on Structural Mechanics in Reactor Technology, Div. L,
Berlin, L 11/3.
Hibbitt, Karlsson, and Sorensen, 1997, ABAQUS Version 5.7.
Jiang, Y., and Sehitoglu, H., 1995, “Rolling Contact Stress Analysis with the
Application of a New Plasticity Model,” Wear, Vol. 191, pp. 35–44.
Jiang, Y., and Sehitoglu, H., 1996a, “Modeling of Cyclic Ratchetting Plasticity,
Part I: Development of Constitutive Relations,” ASME Journal of Applied Mechan
ics, Vol. 63, pp. 726–733.
Jiang, Y., and Sehitoglu, H., 1996b, “Modeling of Cyclic Ratchetting Plasticity,
Part II: Comparison of Model Simulations with Experiments,” ASME Journal of
Applied Mechanics, Vol. 63, pp. 726–733.
Mahnken, R., and Stein, E., 1996, “Parameter Identiﬁcation for Viscoplastic
Models Based on Analytical Derivatives of a LeastSquares Functional and Stability
Investigations,” Int. J. Plasticity, Vol. 12, pp. 451–479.
McDowell, D. L., 1985, “A Two Surface Model for Transient Nonproportional
Cyclic Plasticity, Part I,” ASME Journal of Applied Mechanics, Vol. 52, p. 298.
Nelder, J. A., and Mead, R., 1965, “A Simplex Method for Function Minimiza
tion,” Computer J., Vol. 7, pp. 308–313.
Ohno, N., and Wang, J.D., 1993, “Kinematic Hardening Rules with Critical State
of Dynamic Recovery. Part I: Formulation and Basic Features for Ratchetting Be
havior,” Int. J. Plasticity, Vol. 9, p. 391.
Ortiz, M., and Popov, E. P., 1985, “Accuracy and Stability of Integration Algo
rithms for Elastoplastic Constitutive Relations,” Int. J. Num. Meth. Eng., Vol. 21, pp.
1561–1575.
SDRC, Milford, OH, USA. 1994, IDEAS Master Series Version 2.1.
Simo, J. C., and Taylor, R. L., 1985, “Consistent Tangent Operators for Rate
Independent Elastoplasticity,” Comp. Meth. Appl. Mech. Engng., Vol. 48, pp. 101–
118.
The Math Works Inc, 1996, Matlab 5.0.
Fig. 10 Result from FEcalculation with model B
Fig. 11 Result from FEcalculation with model JS
54 / Vol. 122, JANUARY 2000 Transactions of the ASME
A P P E N D I X A
Local Problem and Stress Update for the Chaboche and the Bower Model
A.1 Local Problem.
Table A.1 Local problem for model A
Nonlinear equation
nϩ1
⌽ ϭ 0:
nϩ1
⌽ ϭ
͌
3
2
͉
nϩ1
dev
͉ Ϫ
y
Ϫ
nϩ1
K
where
͉
nϩ1
dev
͉ ϭ ͉
nϩ1
dev
Ϫ a
X
n
X͉ Ϫ
2
3
(3G ϩ ca
X
)⌬,
n؉1
dev
e
ϭ
n
dev
ϩ 2G⌬⑀
dev
a
X
ϭ ͑1 ϩ ␥⌬͒
Ϫ1
,
nϩ1
K ϭ
1
1 ϩ ⌬b
͑
n
K ϩ ⌬bK
ϱ
͒
Unknown internal variable x:
x ϭ ⌬
Table A.2 Updating stresses and backstresses for model A
Updating of the stresses:
nϩ1
dev
ϭ a
1
nϩ1
dev
e
ϩ ͑1 Ϫ a
1
͒a
X
n
X,
nϩ1
m
ϭ
n
m
ϩ 3K
b
⌬⑀
m
nϩ1
X ϭ a
2
nϩ1
dev
e
ϩ ͑1 Ϫ a
2
͒a
X
n
X,
nϩ1
K ϭ
1
1 ϩ ⌬b
͑
n
K ϩ ⌬bK
ϱ
͒
where
a
1
ϭ 1 Ϫ
ͱ6G⌬
͉
nϩ1
dev
e
Ϫ a
X
n
X͉
, a
2
ϭ
ͱ
2
3
c⌬a
X
͉
nϩ1
dev
e
Ϫ a
X
n
X͉
Table A.3 Local problem for model B
Nonlinear equation
nϩ1
⌽ ϭ 0:
nϩ1
⌽ ϭ ͱ
3
2
͉
nϩ1
dev
͉ Ϫ
y
where
͉
nϩ1
dev
͉ ϭ ͉
nϩ1
dev
e
Ϫ b
X
n
X Ϫ (1 Ϫ b
X
)
n
Y͉ ϩ
2
3
(3G ϩ cb
X
)⌬
nϩ1
dev
e
ϭ
n
dev
ϩ 2G⌬⑀
dev
, b
X
ϭ
1 ϩ ␥
2
⌬
1 ϩ ͑␥
1
ϩ ␥
2
͒⌬
Unknown internal variable x:
x ϭ ⌬
Table A.4 Updating stresses for model B
Updating of the stresses:
nϩ1
dev
ϭ b
1
nϩ1
dev
e
ϩ (1 Ϫ b
1
)b
X
n
X ϩ b
2
n
Y,
nϩ1
m
ϭ
n
m
ϩ 3K
b
⌬⑀
m
nϩ1
X ϭ b
3
nϩ1
dev
e
ϩ (1 Ϫ b
3
)b
X
n
X ϩ b
4
n
Y,
nϩ1
Y ϭ b
Y
(
n
Y ϩ ␥
2
⌬
nϩ1
X)
where
b
1
ϭ 1 Ϫ
ͱ6G⌬
͉
nϩ1
dev
e
Ϫ b
X
n
X Ϫ ͑1 Ϫ b
X
͒
n
Y͉
, b
2
ϭ ͑1 Ϫ b
1
͒͑1 Ϫ b
X
͒
b
3
ϭ
ͱ
2
3
c⌬b
X
͉
nϩ1
dev
e
Ϫ b
X
n
X Ϫ ͑1 Ϫ b
X
͒
n
Y͉
,
b
4
ϭ ͑1 Ϫ b
3
͒͑1 Ϫ b
X
͒, b
Y
ϭ
1
1 ϩ ␥
2
⌬
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 55
HyoJin Kim
JaeJin Jung
R&D Center,
Korea Heavy Industries &
Construction Co., Ltd.,
555 GuygokDong,
Changwon, 641792, Korea
email: hyojin@hanjung.co.kr
Model of Grain Deformation
Method for Evaluation of Creep
Life in InService Components
The creep life can be evaluated by the degree of grain deformation since the grains of
CrMo base material deform in the direction of stress. The grain deformation method
using imageprocessing technique is developed for life assessment of inservice high
temperature components. The new assessment model for the method is presented to apply
to inservice components and is veriﬁed by interrupted creep test for exserviced material
of 1Cr0.5Mo steel. The proposed model, which is irrespective of stress direction, is to
evaluate mean of the absolute deviation for the measured aspect ratios of the grains
sections.
Introduction
Life extension of fossil fuel power plants beyond their classi
cally deﬁned design life is being practiced today by many utilities.
The emergence of this aftermarket has lead to recent advances in
component inspection, monitoring techniques, and remaining life
assessment techniques. Failure of header and steam piping of
boiler is induced by material degradation and/or creep damage.
The main damage mode in HAZ regions is grain boundary cavi
tation. It can cause the component failure if cavities coalesce in
cracks. In order to perform the assessment on these regions,
Aparameter technique for quantifying cavitation mainly is applied
(Roberts et al., 1989). In contrast, the primary damage occurring in
CrMo base material is material softening or headerbody swelling.
In the material, cavities nucleate at the very last stage of creep life.
Therefore, a qualitative assessment (Roberts et al., 1989) from
coarsening of carbides or quantitative grain deformation method
(Soji, 1994) is applied to remaining life assessment on the mate
rial.
Grain deformation method is applied to materials in which the
damage due to intergranular creep is predominant. The method is
nondestructive technique and easy to apply to base material of
inservice high temperature component. But the method by mi
croscopy is time consuming and subjects to errors for measuring
size and deformation of grain. In this paper, imageprocessing
technique is introduced to contribute to accuracy and speediness of
the measurement. The new assessment model of the method is
presented and conﬁrmed. For validation of the model, creep tests
are performed using standard testpieces of recovered and ex
serviced materials. Replicas are taken on the specimen at intervals
of 500–1000 hours, which are repeated by the specimen rupture.
From the prepared replicas, grain deformation is measured, and
creep damage and microstructures are examined.
Experimental Procedure
Specimen. The specimen is 1Cr0.5Mo steel from a boiler
superheater header which have experienced 180,000 hours of
operation at 538°C. To reproduce the original microstructure of the
steel, a portion of specimens is reheattreated according to the
manufacture’s initial heat treatment schedule. To evaluate the
mechanical properties, tensile and creep tests are performed for the
reheattreated and exserviced materials. The results of tensile test
are presented in Table 1 and are compared with the requirement for
strength of 1Cr0.5Mo steel (13CrM044). It is seen that the reheat
treated material satisﬁes the requirement for mechanical proper
ties. The results of creep rupture test are shown in Fig. 1. In the
Larson Miller Parameter of xaxis, T and t
r
are operating temper
ature and rupture time, respectively. It is seen that the creep
strength for exserviced material reduces considerably and that the
creep strength for reheattreated material is identical with that for
new material of NRIM (Tanaka, 1990). Therefore, it is conﬁrmed
that the reheattreated material is substituted for new material. The
reheattreated material is called the recovered material in this
paper. Specimens for replication are mechanically polished.
Creep Test. Creep tests are performed using standard test
pieces of recovered and exserviced materials. Replicas are taken
on the specimen at intervals of 500–1000 hours, which are re
peated by the specimen rupture. The replica (ASTM, 1990) is
produced by coating one side of a sheet of plastic ﬁlm with a
suitable solvent, such as acetone or methyle acetate, and applying
the wetted side of the ﬁlm to the prepared specimen surface. By
replicas taken, grain deformation is measured, and creep damage
and microstructures are examined.
ImageProcessing Technique
Measuring the degree of deformation for a great deal of grains
by microscopy is timeconsuming and subjects to errors. Here
imageprocessing technique (Vrhovac et al., 1996; Louis et al.,
1995) is introduced to contribute to accuracy and speediness for
the measurement of grains and analysis. It is assumed that the
twodimensional orientations represent threedimensional grain
orientations. The assumption is valid for the microstructures of a
cylindrical component that deforms in the direction of hoop or
axial. In this paper, the technique is applied to header and steam
piping of boiler that the deformation in the direction of hoop or
axial is predominant.
Image Input System and Input. Input of metallurgical mi
crostructure images is performed by several methods, which are
from specimen, replica, or scanner. Here images are inputted from
replicas taken on the specimens of interrupted creep tests. The
imageprocessing system consists of input apparatus such as mi
croscopy and CCD camera, main system such as A/D converter
and controller, and output apparatus such as printer. The degree of
grain deformation can differ from positions of specimen in the
interrupted creep tests. To obtain precise data, images are acquired
at constant position of specimen and input of images is performed
using the microscopy with automatic transfer equipment and the
controller that can control imageprocessing program.
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division June
8, 1998; revised manuscript received July 9, 1999. Associate Technical Editor:
S. K. Datta.
56 / Vol. 122, JANUARY 2000 Transactions of the ASME Copyright © 2000 by ASME
Image Early Processing. The image which computer can
recognize is made from digital image obtained from CCD camera
via early processing. To measure grain deformation, grain bound
ary need to be clearly distinguished. But grain boundary is not
clear by several factors such as material, material degradation,
etching and replica conditions. To best image, several image
processing techniques are used. Filtering operations reduce or
increase the rate of change that occurs in the intensity transitions
within an image. Arithmetic and logic operations upon the active
image are performed between two image, or between an image and
a constant value. Background correction is used to better distin
guish image background from image objects, making it easier to
extract the objects during a counting or measurement operation.
Data Measurement. To measure grain deformation, principal
stress axis is determined after early processing and grain angles
about principal stress axis,
m
, are measured for grains, as shown
in Fig. 2(a). The proposed model, which is irrespective of stress
direction, is to measure the ratios that are diametrical maximum to
minimum dimensions for grains, as shown in Fig. 2(b). For a
consistent data measurement, a standard area with constant bound
ary is applied for input images. Also, standard deviation method
and mean method of the absolute deviation are applied for pro
cessing of measured data.
Grain Deformation Method
Creep damage occurring in base material generally is material
softening or deformation before creep cavities. A new assessment
technique, which is not a quantitative assessment method for creep
cavities, need to be developed for the ductile material such as base
material of CrMo steel. In this paper, grain deformation method,
which is a quantitative method for the degree of grain deformation,
is introduced. The microstructure of base material for low alloy
ferrite CrMo steel consists of ferrite plus bainite (or pearlite).
Creep damage is induced during a component continues at the
hightemperature and constant load. Grain does not deform at
lowductile bainite and deforms toward principal stress axis at
highductile ferrite. As the deformation degree of grains indicates
creep life expenditure, a quantitative assessment for the degree is
grain deformation method. The method can evaluate more exact
remaining life of base metal with master curve that gets from creep
tests in the laboratory. The application range of the method extends
through development of assessment model. The model (Soji, 1994)
assesses creep damage by evaluating the standard deviation for
deformation angles about principal stress axis since grains deform
toward the principal stress axis.
The quantiﬁcation of grain deformation is applied using image
processing technique after replicas are taken, as shown in Fig. 3.
Figure 4 shows distribution of grain deformation using image
processing technique for recovered material and material of creep
life fraction, 0.87. It is seen that distribution of deformation angles
for grains is random in the case of recovered material and the
Table 1 Mechanical properties
Property
Material
Yield strength
(MPa)
Tensile strength
(MPa)
Elongation
(%)
Recovered 362.8 541.3 29.3
Exserviced 273.6 492.3 33.0
DIN 17175:13CrMo44 275 min. 440–590 19.0 min
Fig. 1 Creeprupture strength of recovered and exserviced materials
Fig. 2 Measurement of grain deformation
Fig. 3 Measurement procedure of grain deformation using image
processing technique
Fig. 4 Distribution of grain deformation using imageprocessing tech
nique
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 57
exserviced material for the end of life has normal distribution
centering around the principal stress axis. The master curve are
obtained by evaluating the standard deviation for distribution of
grain deformation and is in the temperature range of 500°C to
650°C and low stress, as shown in Fig. 5. The established model
can subject to many errors because the principal stress axis is
determined in a component. In this paper, the new model is
presented to apply to the creep life assessment for the components
irrespective of principal stress axis. The proposed model is to
evaluate mean of the absolute deviation for the measured ratios
which are diametrical maximum to minimum dimensions for
grains. During interrupted creep tests are performed at 550°C and
14 kg/mm
2
for exserviced material, mean of the absolute devia
tion for the measured ratios which are diametrical maximum to
minimum dimensions for grains is shown in Fig. 6. The deforma
tion distribution for an early stage shows a little uniform pattern
and the absolute deviation for the end of life grows large as the
grains deform largely. Therefore, it is conﬁrmed that the proposed
model can be applied to the creep life assessment of base material
for inservice components.
Results and Discussion
For recovered material, Fig. 7 shows the degree of grain defor
mation by the established model and Fig. 8 shows the degree of
grain deformation by the proposed model during interrupted creep
tests are performed at 538°C and 16 kg/mm
2
. It is shown that
standard deviation is 46 before creep test and is 41 after creep
rupture. Also, it is shown that mean of absolute deviation is 0.42
before creep test and is 0.44 after creep rupture. From the results,
it is shown that the deformation distribution for grains makes no
difference between an early stage and the end of creep life. It is
seen that the degree of grain deformation is very small because
material degradation is not induced and intergranular fracture is
induced by high stress and low temperature. To obtain actual
results in the case of recovered material, It need to be lower stress,
high temperature, and longterm creep test.
For exserviced material, Fig. 9 shows the degree of grain
deformation by the established model and Fig. 10 shows the degree
of grain deformation by the proposed model during interrupted
creep tests are performed at several temperatures and stresses. The
standard deviation is 45 before creep test. It means that the
directions of grain boundaries are disordered. Because the thick
ness of the boiler header is thick and the applied stress is very
small, the directions of grain boundaries are disordered as though
the header have experienced 180,000 hours of operation. The
standard deviation for the end of life is 20 to 30. Also, it is shown
that mean of absolute deviation is 0.42 before creep test and is 0.7
to 0.9 for the end of life. The degree of grain deformation for the
end of life of exserviced material is large because grains deform
easily by material degradation. The material degradation is due to
the increase of distance between precipitates according to coales
Fig. 5 Relation between creep life fraction and deformation index
Fig. 6 Distribution of deformation index for creep life fraction of ex
serviced material at 550°C and 4 kg/mm
2
Fig. 7 Distribution of deformation index for creep life fraction of recov
ered material at 538°C and 16 kg/mm
2
Fig. 8 Distribution of deformation index for creep life fraction of recov
ered material at 538°C and 16 kg/mm
2
58 / Vol. 122, JANUARY 2000 Transactions of the ASME
cence and coarseness of precipitate, or decrease of solute inside
matrix by long term operation. From these results, it is seen that
standard deviation and mean of absolute deviation change as the
creep life decreases. Therefore, it is conﬁrmed that the proposed
model of grain deformation method can be applied to the creep life
assessment for inservice components.
The actual conditions of creep behavior for components of
power plants are similar to these of tests at low stress and high
temperature, as shown in Figs. 9 and 10, for exserviced material.
It is appropriate to make master curve under the conditions. On the
basis of the master curve, the introduction of imageprocessing is
signiﬁcantly contributing to accuracy and speediness of the life
assessment for inservice high temperature components. Exact
results can be obtained in the laboratory by the established model
because grain deformation by the model is measured at specimens
of uniaxial creep test which principal stress axis is unidirectional.
But principal stress axis must be determined in the actual compo
nent, which can subjects to errors. In this paper, the new model is
presented to apply to the creep life assessment for the component
irrespective of principal stress axis.
Conclusions
This study is performed to characterize the correlation between
grain deformation and creep life assessment for inservice high
temperature components. The imageprocessing technique is in
troduced to contribute to accuracy and speediness of the assess
ment for the components such as header and steam piping of boiler
that the deformation in the direction of hoop or axial is predomi
nant. From results, the new model is presented to apply to the
creep life assessment for the components irrespective of principal
stress axis and conﬁrmed.
References
Roberts, B. W., Ellis, F. V., and Henry, J. F., 1989, RemainingLife Estimation of
Boiler Pressure Parts, Volume 4: Metallographic Models for WeldHeatAffected
Zone, EPRI CS5588, pp. 37–39.
Roberts, B. W., Avernue, K., and Askins, M. C., 1989, RemainingLife Estimation
of Boiler Pressure Parts, Volume 3: Base Metal Model, EPRI CS5588, pp. 22–23.
Shoji, T., 1994, Material Degradation & Life Prediction, REALIZE INC., pp.
179–186.
Tanaka, C., 1990, NRIM Creep Data Sheet, No. 35A, NRIM, pp. 5–9.
ASTM Standard, 1990, “Standard Practice for Production and Evaluation of Field
Metallographic Replicas,” Annual Book, E135190, pp. 953–958.
Vrhovac, M., Barata, J. C. R., and Rodrigues, R., 1996, “Automation of Micro
structure Analysis by Artiﬁcial Intelligence Image Processing,” Proceedings of Fail
ures ’96, pp. 297–312.
Louis, P., and Gokhale, A. M., 1995, “Application of Image Analysis for Charac
terization of Spatial Arrangements of Features in Microstructure,” Metallurgical and
Materials Transaction, Vol. 16, pp. 1449–1456.
Fig. 9 Distribution of deformation index for creep life fraction of ex
serviced material
Fig. 10 Distribution of deformation index for creep life fraction of ex
serviced material
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 59
S. Murakami
Professor.
A. Miyazaki
Graduate Student.
M. Mizuno
Assistant Professor.
email mizuno@mech.nagoyau.ac.jp
Department of Mechanical Engineering,
Nagoya University,
Furocho, Chikusaku,
Nagoya 4648603, Japan
Modeling of Irradiation
Embrittlement of Reactor
Pressure Vessel Steels
A model to describe the change in the inelastic and fracture properties of reactor
pressure vessel steels due to neutron irradiation in the ductile region (i.e., irradiation
embrittlement) is developed. First, constitutive equations for unirradiated elastic
viscoplasticdamaged materials are developed within the framework of the irrevers
ible thermodynamics theory. To take into account the effect of hydrostatic pressure on
the nucleation and growth of microvoids, properly deﬁned dissipation potential is
used. Then, the effect of irradiation on the material behavior is incorporated into the
proposed model as a function of neutron ﬂuence ⌽ by taking into account the
interaction between irradiationinduced defects and movable dislocations. As regards
the damage strain threshold p
D
, the mechanism of void nucleation due to pileup of
dislocations at the inclusions in the material is proposed ﬁrst under unirradiated
condition, and then the effect of irradiation on the mechanism is formulated. In order
to demonstrate the validity of this model, it is applied to the case of uniaxial tensile
loading of a low alloy steel A533B cl. 1 for the pressure vessel use of lightwater
reactors at 260°C. The resulting model can describe the increase in yield stress and
ultimate tensile strength, the decrease in total elongation and strain hardening, and
the strain rate dependence of yield stress due to neutron irradiation.
1 Introduction
The structural materials used in nuclear reactors are subject to
neutron irradiation during their operations, and the resulting
knocking out of the constituent atoms leads to the change in
internal structure of the material and thus to the change in its
mechanical properties. Embrittlement of pressure vessel steels of
lightwater reactors due to the irradiation is, in particular, a crucial
issue to be overcome from the viewpoint of the reliability and
integrity of the reactors (Little, 1976; Phythian and English, 1993).
So far, a considerable number of papers (Wronski et al., 1965;
Little, 1976; Kussmaul et al., 1990; Mansur and Farrell, 1990;
Erve and Tenckhoff, 1993; Gage and Little, 1994; Little and Gage,
1994; Scott, 1994) have focused on the change of solely a single
material property due to the irradiation, e.g., the rise of ductile
brittle transition temperatures (Phythian and English, 1993), the
increase of yield stress (Makin and Minter, 1960; Hunter and
Williams, 1971; Williams and Hunter, 1973; Steichen and Wil
liams, 1975; Phythian and English, 1993), etc. Similar studies on
materials employed for fast reactors and fusion reactors have been
also reported (Klueh, 1993; Klueh and Alexander, 1995, 1992;
Klueh and Maziasz, 1992; Klueh and Vitek, 1991; Odette and
Lucas, 1992, 1991). Most of these studies deal mainly with the
change of ductilebrittle transition behavior due to the irradiation,
and hence uniﬁed theory to describe the deformation and fracture
of irradiated materials in the ductile region is, so far, not available.
The present paper is concerned with the mechanical modeling of
the process of damage, embrittlement and strain hardening of
pressure vessel steels for lightwater reactors under neutron irra
diation. A model is developed in the framework of continuum
damage mechanics to describe a uniﬁed process of embrittlement,
damage growth and elasticplastic deformation. First, a dissipation
potential, which can represent the effect of hydrostatic pressure on
damage evolution, is adopted on the basis of the irreversible
thermodynamics theory for constitutive equations, and the consti
tutive equations for an unirradiated material are formulated. Then,
in order to extend the constitutive equations to an irradiated
material, effects of irradiationinduced defects on the material
behavior are taken into account from the viewpoint of the micro
scopic mechanisms. Finally, the resulting constitutive equations
are applied to uniaxial tensile loading of a lightwater reactor
pressure vessel steel, ASTM A533 grade B class 1, and the validity
of the model is discussed in comparison with the corresponding
experiments.
2 Mechanisms of Irradiation Embrittlement
When a crystalline material is irradiated by a neutron ﬂux, a
number of atoms are knocked out from the lattice sites to create the
Frenkel defects comprised of equal numbers of interstitial atoms
and vacancies. A part of the interstitial atoms and vacancies thus
created annihilates by their recombination and by the absorption
by the sink. The point defects escaping such annihilation make up
the irradiationinduced defects or the clusters affecting the material
properties. The irradiationinduced defects will act as an obstacle
or a dislocation source for the dislocation motion on its slip planes,
and bring about the irradiation hardening. This irradiation harden
ing will in turn cause the phenomena of irradiation embrittlement
(Little, 1976; Phythian and English, 1993; Klueh and Alexander,
1995, 1992); i.e., the reduction of fracture strain, the rise of
ductilebrittle transition temperature, and the reduction of upper
shelf energy in the ductile region. The rate of nucleation of Frenkel
defects, however, decreases with the increasing dose of irradiation,
and the change of material properties is gradually saturated (Wron
ski et al., 1965; Little, 1976; Klueh, 1993; Oddete and Lucas,
1992, 1991).
Besides the abovementioned damage due to knocking out of
atoms, irradiation damage caused by nuclear conversion has also
been observed. However, regarding ASTM A533 grade B class 1
steels discussed below, the effects of nuclear conversion are not
signiﬁcant, and thus we will disregard the effects of nuclear
conversion in the following discussion.
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division April
13, 1998; revised manuscript received July 9, 1999. Associate Technical Editor:
S. K. Datta.
60 / Vol. 122, JANUARY 2000 Transactions of the ASME Copyright © 2000 by ASME
3 Modeling of Irradiation Embrittlement
Phenomena of irradiation embrittlement can be modeled as the
effects of the irradiation on the constitutive and the damage evo
lution equations of materials. If the neutron irradiation does not
cause signiﬁcant change in the mechanisms of deformation and
fracture, these effects may be formulated by representing the
material constants in these equations as functions of neutron ﬂu
ence. Namely, we will ﬁrst develop the constitutive and damage
evolution equations for unirradiated materials. Then, by taking
account of internal structural changes brought about by the neutron
irradiation discussed above, the material functions to describe the
effects of the irradiation will be formulated.
3.1 Constitutive Equations for Unirradiated Materials.
The deformation of ASTM A533 grade B class 1 steels depends on
the rate of strain. At elevated temperature around 300°C, the A533
steels show ductile fracture as a result of the nucleation and growth
of microvoids within the material (Shockey et al., 1980). The
description of such a process of deformation and fracture can be
facilitated by incorporating the notion of continuum damage me
chanics (Lemaitre, 1985, 1992; Lemaitre and Chaboche, 1990;
Saanouni et al., 1994) into a general thermodynamical framework
of viscoplastic constitutive equations (Lemaitre and Chaboche,
1990; Saanouni et al., 1994).
We will now assume that the material is isotropic linear elastic
viscoplastic at the initial undamaged state. Since the main purpose
of the present paper is to predict a uniaxial tensile behavior, only
an isotropic hardening variable is considered as an internal vari
able for strain hardening caused by viscoplastic deformation. The
isotropic hardening variable can represent also the effects of strain
aging. However, since the experimental results necessary to dis
cuss the effects of strain aging of A533 steel are lacking, the aging
effects are not taken into account in the present paper. By suppos
ing that damage development is isotropic, a free energy ⌿ of
damaged elasticviscoplastic material may be formulated as fol
lows:
⌿͑⑀
e
, r, D͒ ϭ
1
2
͑1 Ϫ D͒͑tr ⑀
e
͒
2
ϩ ͑1 Ϫ D͒ tr ͑⑀
e
⑀
e
͒
ϩ
1
2
qQ͑1 Ϫ D͒r
2
, (1)
where ⑀
e
, r and D represent the elastic strain tensor, isotropic
hardening variable and the damage variable, respectively, while q
and Q are material constants associated with strain hardening.
Furthermore, the symbol represents the density, and and are
Lame’s constants.
In order to derive inelastic constitutive equations and evolution
equations of internal state variables, the relevant dissipation po
tential should be formulated. The dissipation due to the inelastic
deformation of polycrystalline materials is caused mainly by dis
location motion caused by stress, while the dissipation due to
damage evolution is primarily caused by the internal energy re
lease due to the nucleation and growth of microvoids (Lemaitre
and Chaboche, 1990; Lemaitre, 1992). Thus by taking account of
the difference in the mechanisms of these dissipation phenomena,
two dissipation potentials,
*
in
for the inelastic deformation and
*
D
for the damage evolution, will be employed in the present paper.
According to the power function proposed by Saanouni et al.
(1994), we have the dissipation potential
*
in
as follows:
*
in
͑, R; r, D͒ ϭ
K
n ϩ 1
ͳ
1
K
ͫ
f ϩ
1
2
R
2
Q͑1 Ϫ D͒
Ϫ
1
2
q
2
Q͑1 Ϫ D͒r
2
ͬ
ʹ
nϩ1
, (2)
where represents stress tensor, and R is the thermodynamic
conjugate force associated with r. The symbols K and n are
material constants to represent strain rate sensitivity, and f is a
yield function given as follows:
f ϭ
eq
Ϫ R
͑1 Ϫ D͒
1/2
Ϫ k, (3)
Nomencl at ure
B
1
, B
2
ϭ material constants
D ϭ damage variable
D
c
ϭ critical value of damage vari
able
E ϭ Young’s modulus
f ϭ yield function
I ϭ secondrank unit tensor
k ϭ initial yield stress
k
0
ϭ material constant to represent
the value of k in unirradiated
condition
k
1
, k
2
ϭ material constants
k
⌽
ϭ material function correspond
ing to k
K ϭ material constant associated
with strain rate sensitivity
K
0
ϭ material constant to represent
the value of K in unirradiated
condition
K
1
, K
2
ϭ material constants
K
⌽
ϭ material function correspond
ing to K
l ϭ distance between obstacles
l
0
, l
⌽
ϭ distance between obstacles in
unirradiated and irradiated con
ditions
n ϭ material constant associated
with strain rate sensitivity
nЈ ϭ number of pileup disloca
tions
n
0
, n
⌽
ϭ number of pileup disloca
tions in unirradiated and irra
diated conditions
N ϭ number of defects per unit
volume
p ϭ magnitude of inelastic strain
p
D
ϭ threshold of inelastic strain
for damage initiation
p
D0
, p
D⌽
ϭ threshold of inelastic strain
for damage initiation in unir
radiated and irradiated condi
tions
q, Q ϭ material constants associated
with strain hardening
q
0
ϭ material constant to represent
the value of q in unirradiated
condition
q
1
, q
2
ϭ material constants
q
⌽
ϭ material function correspond
ing to q
r ϭ isotropic hardening variable
R ϭ thermodynamic conjugate
force associated with r
s, S ϭ material constants associated
with damage evolution
Y ϭ thermodynamic conjugate
force associated with D
␣ ϭ material constant
 ϭ material constant associated
with damage evolution
␥ ϭ material constant
⌬
y
ϭ increase in yield stress
⑀ ϭ total strain tensor
⑀
e
ϭ elastic strain tensor
⑀
in
ϭ inelastic strain tensor
ϭ material constant
, ϭ Lame’s constants
ϭ Poisson’s ratio
ϭ density
ϭ stress tensor
Ј ϭ deviatoric stress tensor
c
ϭ critical stress of the initiation of
debonding
EQ
ϭ equivalent stress
ϭ shear stress acting on slip plane
0
,
⌽
ϭ plastic ﬂow stress/shear stress
acting on slip plane in unirradi
ated and irradiated conditions
*
D
ϭ dissipation potential for damage
evolution
*
in
ϭ dissipation potential for inelastic
deformation
⌽ ϭ neutron ﬂuence
⌿ ϭ free energy of damaged elastic
viscoplastic material
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 61
where
eq
and k represent the equivalent stress and the initial yield
stress, respectively.
As regards the dissipation potential
*
D
, on the other hand, the
following potential proposed by Lemaitre (1985) will be em
ployed:
*
D
͑Y; p˙ , D͒ ϭ
S
s ϩ 1
ͩ
ϪY
S
ͪ
sϩ1
p˙
͑1 Ϫ D͒

, (4)
where Y is thermodynamic conjugate force associated with D, p˙
represents the magnitude of inelastic strain rate, and S, s and  are
material constants with respect to damage evolution.
According to the usual procedure of irreversible thermodynam
ics (Lemaitre, 1985, 1992; Lemaitre and Chaboche, 1990; Saa
nouni et al., 1994), Eqs. (1)–(4) furnish the constitutive equations
for unirradiated material as follows:
⑀ ϭ ⑀
e
ϩ ⑀
in
, (5a)
⑀
e
ϭ
1
1 Ϫ D
ͩ
1 ϩ
E
Ϫ
E
tr I
ͪ
, (5b)
⑀˙
in
ϭ
3
2
Ј
eq
p˙ , (5c)
R
˙
ϭ ͑1 Ϫ D͒q
ͭ
Q Ϫ
R
͑1 Ϫ D͒
1/2ͮ
p˙ Ϫ
R
1 Ϫ D
D
˙
, (5d)
D
˙
ϭ
ͭ
ͩ
ϪY
S
ͪ
s
p˙
͑1 Ϫ D͒

, p Ն p
D
,
0, p Ͻ p
D
,
(5e)
p˙ ϭ
1
͑1 Ϫ D͒
1/2
ͳ
f
K
ʹ
n
, (5f)
ϪY ϭ
1
2
͑tr ⑀
e
͒
2
ϩ tr ͑⑀
e
⑀
e
͒ ϩ
1
2
qQr
2
, (5g)
where ⑀ and ⑀
in
represent the total strain tensor and inelastic strain
tensor, respectively, while E and are Young’s modulus and
Poisson’s ratio. The symbols I and Ј represent a secondrank unit
tensor and deviatoric stress tensor, respectively, and p
D
is a thresh
old of inelastic strain for the damage evolution (Lemaitre, 1985,
1992). According to the notion of damage mechanics, we will
further assume that material ruptures when the damage variable D
attains to the critical value D
c
(Lemaitre, 1985, 1992).
3.2 Effects of Neutron Irradiation on Material Properties.
By noting that the essential mechanisms of inelastic deformation
and damage do not change due to irradiation, the inelastic consti
tutive equation and the damage evolution equation of irradiated
material may be given only by modifying the material constants in
the corresponding equations of the unirradiated material as mate
rial functions of neutron ﬂuence ⌽. In the following discussion, we
will represent the effects of neutron irradiation on the material
constants in view of the mechanisms of the internal structural
changes due to irradiation.
3.2.1 Young’s Modulus. The elastic deformation is, in gen
eral, a structureinsensitive phenomena, and thus is little sensitive
to microstructural defects such as vacancies and dislocations in
materials (Barrett et al., 1973); we will assume that Young’s
modulus E is not affected by neutron irradiation.
3.2.2 Initial Yield Stress. As already described in Section 2,
the irradiation hardening is caused mainly by the obstruction of
dislocation motion by irradiationinduced defects, and the source
hardening and friction hardening have been proposed as their
major mechanisms (Little, 1976). The former takes place because
irradiationinduced point defects are attracted to the vicinity of
dislocations due to the diffusion, and thus trap the dislocations.
The latter, on the other hand, is a kind of dispersion strengthening
mechanisms by which the point defects or their clusters interact
with movable dislocations as dispersed obstacles. Thus mechanism
of the irradiation hardening will be modeled in the present paper
by noting that irradiation hardening is primarily governed by the
latter mechanism.
The increase in yield stress ⌬
y
brought about by the dispersion
strengthening is inversely proportion to the distance l between
obstacles (Barrett et al., 1973). In view of the saturation of the
defect nucleation with the increase of ﬂuence, the relation between
the number of defects per unit volume N and the ﬂuence ⌽ is
formulated as follows (Little, 1976):
N ϭ ␣͓1 Ϫ exp͑Ϫ␥⌽͔͒, (6)
where ␣ and ␥ are material constants depending on the material
and the irradiation condition. By noting dispersion strengthening
mechanism and Eq. (6), Makin and Minter (1960) proposed the
following relation for the increase of yield stress due to irradiation:
⌬
y
ϭ ͓1 Ϫ exp͑Ϫ␥⌽͔͒
1/2
, (7)
where and ␥ are constants depending on the material and the
irradiation conditions. If we apply Eq. (7) to the initial yield stress
k, the increase of k due to irradiation may be expressed as follows:
k
⌽
ϭ k
0
͕1 ϩ k
1
͓1 Ϫ exp͑Ϫk
2
⌽͔͒
1/2
͖, (8)
where k
0
, k
1
and k
2
are material constants, and k
0
represents the
value of k in the unirradiated condition.
Though the materials for experiments are different from A533
steels, the increase and saturation of yield stress due to neutron
irradiation in CrMo steels and stainless steels have been reported
by Klueh (1993) and Odette and Lucas (1992, 1991).
3.2.3 Strain Hardening. Strain hardening of crystalline ma
terials is attributable to the increased resistances to the dislocation
motion due to the change of the dislocation conﬁgurations and the
increase in dislocation density (Barrett et al., 1973). In addition to
these mechanisms, since irradiationinduced defects act as the
resistance to the dislocation motion, the hardening rate q of Eq.
(5d) in irradiated materials becomes larger than that in unirradiated
material. Then if we assume that the increase of dislocation density
is proportional to the number of irradiation defects on the slip
plane and the effects will be saturated for large dose of irradiation,
the hardening rate q may be expressed as follows:
q
⌽
ϭ q
0
͕1 ϩ q
1
͓1 Ϫ exp͑Ϫq
2
⌽͔͒
1/2
͖, (9)
where q
0
, q
1
and q
2
are material constants, and q
0
represents the
value of q when unirradiated.
Though the effects of neutron irradiation on strain aging can be
incorporated into Eq. (9), the effects and mechanisms of aging in
A533 steels have not been elucidated experimentally. Thus, the
effects of strain aging are not considered here. As regards the
materials for fast reactors and fusion reactors, i.e., mainly CrMo
steels, the effects of aging were reported by Klueh and Maziasz
(1992) and Klueh and Vitek (1991).
As regards the magnitude of saturated hardening Q in Eq. (5d),
on the other hand, the reduction of strain hardening by neutron
irradiation have been observed in the experiments of A533 steel
(Hunter and Williams, 1971; Steichen and Williams, 1975). How
ever, this decrease in hardening may be caused by the reduction of
the threshold for the damage evolution due to irradiation. Thus, we
will disregard the effect of neutron irradiation on Q.
3.2.4 Strain Rate Dependence. The reduction of strain rate
sensitivity by neutron irradiation is observed in A533 steel
(Steichen and Williams, 1975). Since the irradiation defects act as
obstacles to the deformation, they reduce the effect of the thermal
activation on the strain rate. If the cause of the reduction in rate
sensitivity may be found in irradiation defects on the slip plane, K
can be formulated in the same way as in q:
K
⌽
ϭ K
0
͕1 Ϫ K
1
͓1 Ϫ exp͑ϪK
2
⌽͔͒
1/2
͖, (10)
62 / Vol. 122, JANUARY 2000 Transactions of the ASME
where K
0
, K
1
and K
2
are material constants, and K
0
represents the
value of K when unirradiated.
3.2.5 Threshold for Damage Evolution. Now let us discuss
the irradiation effects on the threshold inelastic strain p
D
of Eq.
(5e) for the damage initiation. As shown in Fig. 1, the threshold
strain for damage evolution p
D
is supposed to decrease in irradi
ated materials.
A number of dimples have been observed on the rupture sur
faces of unirradiated A533 steel, and relatively large nonmetallic
inclusions such as Al
2
O
3
and MnS have been detected inside of a
large dimples. Many small dimples supposed to be generated at
submicron particles like Fe
3
C have been also observed between the
large dimples (Shockey et al., 1980). Nonmetallic inclusions
Al
2
O
3
, MnS, etc., in general, have weak bonding with the matrix
because of low wetting, while particles such as carbide and nitride
have strong bonding. Accordingly, debonding of large inclusions
Al
2
O
3
, MnS, etc. from the matrix causes the initiation of damage
in the ductile region. Such debonding is frequently caused by local
stress concentration within the material (Tetelman and McEvily,
1970). In the case of A533 steel, we suppose a mechanism in
which dislocation motion is prevented by the so large inclusions
and the pillup of the dislocations increase the local stress.
In order to evaluate the local stress at the end of the dislocations
piled up, we will employ the result of Koehler’s calculation
(1952). Namely, according to the calculation, when nЈ edge dis
locations pile up on the slip plane subjected to the shear stress ,
the magnitude of the tensile stress which is parallel to the slip
plane at the end of the pileup dislocations is given by:
ϭ nЈ. (11)
In order to discuss the effects of irradiation on the damage
initiation, we take a representative volume element (RVE) of
unirradiated and irradiated materials at the initiation of damage as
shown in Fig. 2. The ﬁgure shows the migration of a dislocation
line on a slip plane. Since the rupture strain of A533 steel is
relatively large even under irradiated condition (Hunter and Wil
liams, 1971), the dislocation line goes through the inclusions
around which dislocation loops remain as shown in the ﬁgure. In
such case, edge dislocations with different signs will pile up on
both sides of inclusion on the same slip plane, and thus the pileup
dislocations causes local tensile stress on the interface. Since the
irradiationinduced defects are distributed randomly within the
material, the effect of irradiation on the interface between the
matrix and the inclusion is not signiﬁcant and can be neglected.
Thus the debonding in the irradiated materials occurs when the
tensile stress on the interface due to the pileup dislocations attains
to the critical stress of the initiation of the debonding in unirradi
ated material
c
. Then, the condition for the initiation of the
debonding may be expressed from Eq. (11) as follows:
c
ϭ n
0
0
ϭ n
⌽
⌽
, (12)
where n
0
,
0
, n
⌽
and
⌽
represent the number of the pileup
dislocations around the inclusion at the initiation of the debonding
and the shear stress acting on the slip plane in unirradiated and
irradiated materials, respectively.
In irradiated materials, the dislocation motion is hindered by
means of the dispersion strengthening mechanism due to the
irradiationinduced defects. Namely, the mean length between
obstacles in an irradiated material l
⌽
is smaller than l
0
in an
unirradiated material because of the existence of irradiation
induced defects as shown in Fig. 2. Accordingly, the application of
the theory developed in 3.2.2 to the plastic ﬂow stress
0
and
⌽
in
Fig. 2 leads to the following relation:
⌽
ϭ
0
͓1 ϩ B
1
͕1 Ϫ exp͑ϪB
2
⌽͖͒
1/2
͔, (13)
where B
1
and B
2
are material constants. By substituting this
equation into Eq. (12), we have the following relation between the
numbers of pileup dislocations at the inclusions before the initi
ation of debonding in unirradiated and irradiated materials:
n
0
ϭ n
⌽
͓1 ϩ B
1
͕1 Ϫ exp͑ϪB
2
⌽͖͒
1/2
͔. (14)
Then Eq. (14) furnishes the relation between the inelastic strain of
RVE of the unirradiated and irradiated materials (Barrett et al.,
1973) before the initiation of debonding:
p
D⌽
ϭ p
D0
͓1 ϩ B
1
͕1 Ϫ exp͑ϪB
2
⌽͖͒
1/2
͔
Ϫ1
. (15)
This equation provides the threshold inelastic strain for the nucle
ation of voids in unirradiated and irradiated materials. The rela
tions among
0
,
⌽
, p
D0
and p
D⌽
in the equation are depicted in
Fig. 1.
In the above discussion, the threshold for the initiation of
damage in an irradiated material was speciﬁed by the magnitude of
the inelastic strain on a single slip plane. However, in reality, the
debonding may be caused by the pileup dislocations on a multi
tude of slip planes around inclusions (Broek, 1973). In these cases,
since the dislocation motion is not affected by the hydrostatic
pressure and is governed mainly by shear stress, the above discus
sion will be developed with respect to the maximum shear stress.
Furthermore, since the effects of the irradiation have been incor
porated qualitatively in the above discussion, the above theory can
be applied only to the slip plane which has the largest effect on the
initiation of debonding. Therefore, the threshold inelastic strain
(15) for the initiation of damage in an irradiated material may be
applied to general three dimensional states of stress.
As regards material constants besides those discussed above, the
experimental results to discuss the effects of irradiation on them
are quite lacking. Thus, these material constants will be assumed to
be unaffected by irradiation.
The effects of temperature also can be taken into account easily
in the framework of the thermodynamic constitutive theory. How
ever, the main purpose of the present paper is to formulate the
Fig. 1 Schematic stressstrain curves of unirradiated and irradiated
materials
Fig. 2 RVE of unirradiated and irradiated materials at damage initiation
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 63
constitutive equation to describe the change in mechanical prop
erties under tensile deformation of A533 steels due to neutron
irradiation by the use of continuum damage mechanics. Thus, we
don’t consider here the temperature effects to avoid complicated
formulation and discussion.
4 Application to Pressure Vessel Steel of LightWater
Reactor
Let us now apply the constitutive equations (5), (8)–(10), (15)
developed in the preceding section to the lightwater reactor pres
sure vessel steel A533, and the tensile behavior of unirradiated and
irradiated materials under uniaxial loading will be simulated. The
validity of the model will be discussed in comparison with the
corresponding experiments. The experimental temperature is
260°C, with the irradiation temperature 266°C. In the following
discussion, true stress and true strain will be employed to eliminate
the effects of the geometrical change of test specimens. However,
since the details of the geometry of the specimens after the onset
of the necking are not available in the literature, we shall use
herewith the average value of strain over the gauge length to
calculate the true stress and true strain.
4.1 Determination of Material Constants. We ﬁrst deter
mine the material constants of Eq. (5) for unirradiated viscoplastic
deformation. These material constants were determined so that the
constitutive equations could describe the experimental results
(Steiche and Williams, 1975; Mahmood et al., 1990) of the tensile
properties at different strain rates:
E ϭ 31.6 GPa, k ϭ k
0
ϭ 293 MPa, n ϭ 30,
K ϭ K
0
ϭ 145 MPa ⅐ s
1/n
, Q ϭ 320 MPa,
q ϭ q
0
ϭ 30, S ϭ 2.0, s ϭ 1.0,  ϭ 0.5,
p
D
ϭ p
D0
ϭ 6.0%, D
c
ϭ 0.45.
Then, we determine the material constants in the material func
tions and Eqs. (8)–(10), (15) for irradiated material so that the
experimental results (Hunter and Williams, 1971; Williams and
Hunter, 1973; Steichen and Williams, 1975) concerning the
change of the tensile properties due to irradiation can be described.
Regarding the material constants in the present model, the
constants characterizing the saturation rate of irradiation effect in
Eqs. (8)–(10) and (13)–(15) were assumed to be equal to one
another; i.e.,
k
2
ϭ K
2
ϭ q
2
ϭ B
2
. (16)
As to the material function K that decreases with neutron irradia
tion, we assume that the strain rate dependence will disappear
eventually as the ﬂuence increases; i.e.,
K
1
ϭ 1.0. (17)
By the use of these assumptions, the material constants associ
ated with the irradiated material were determined as follows:
k
1
ϭ 1.55, K
1
ϭ 1.0, q
1
ϭ 0.5, B
1
ϭ 8.0,
k
2
ϭ K
2
ϭ q
2
ϭ B
2
ϭ 9.0 ϫ 10
Ϫ21
cm
2
/n.
4.2 Results of Calculation and Discussion. Figure 3 shows
the true stresstrue strain curves for unirradiated and irradiated
materials calculated by Eqs. (5), (8)–(10) and (15). The ﬁgure
shows qualitatively the increase in yield stress and tensile strength
together with the decrease in fracture strain as a typical phenom
enon of the irradiated material. The change of these properties will
be discussed quantitatively later in comparison with experimental
results (Hunter and Williams, 1971; Williams and Hunter, 1973;
Steichen and Williams, 1975).
Figure 4 shows the evolution of damage variables corresponding
to the true stresstrue strain curves in Fig. 3. The ﬁgure shows the
trend that increased ﬂuence brings about earlier onset and the
accelerated rate of damage evolution.
Figures 5 to 8, furthermore, show the change in yield stress
(0.2% proof stress), tensile strength, fracture strain and the amount
of strain hardening due to irradiation. Figures 5, 6 and 8 show that
the results of the present prediction are in good agreement with the
corresponding experimental data (Hunter and Williams, 1971;
Williams and Hunter, 1973; Steichen and Williams, 1975); i.e.,
they represent well the signiﬁcant rise in the low ﬂuence region
and the eventual saturation of the irradiation effect with increasing
ﬂuence. Regarding fracture strain in Fig. 7, though signiﬁcant
scatter in experimental data (Hunter and Williams, 1971; Steichen
and Williams, 1975) makes the quantitative comparison difﬁcult,
they can be compared qualitatively with the experimental results.
Though materials are different from A533 steels focused in the
present paper, the qualitatively similar dependence of neutron
irradiation is observed in the experiments on CrMo steels by
Klueh (1993) and stainless steels by Odette and Lucas (1992,
1991). Thus the present model may be applicable to their experi
mental results without any essential modiﬁcation.
Finally, Fig. 9 shows the dependence of yield stress (0.2% proof
stress) on strain rate in unirradiated and irradiated materials.
Throughout the entire range of ﬂuence, the present model can
express the reduction in strainratedependence of yield stress with
Fig. 3 True stresstrue strain curves of unirradiated and irradiated
A533B cl. 1 steel at 260°C
Fig. 4 Damage of unirradiated and irradiated A533B cl. 1 steel at 260°C
64 / Vol. 122, JANUARY 2000 Transactions of the ASME
increasing ﬂuence. The predictions particularly for the conditions
of the ﬂuence of 4.0 ϫ 10
19
n/cm
2
and for the unirradiated
condition can describe well the experimental data (Steichen and
Williams, 1975), although lack of experimental data prevents us
from satisfactory quantitative evaluation. At 1.6 ϫ 10
20
n/cm
2
the
dependence on strain rate is shown to virtually vanish.
5 Concluding Remarks
In order to evaluate the irradiation embrittlement of the pressure
vessel steel of lightwater reactor in the ductile region, uniﬁed
inelastic constitutive equations and evolution equation of damage
variable for an irradiated material were formulated from the view
point of continuum mechanics.
Inelastic constitutive equations and an evolution equation of
damage variable for unirradiated material were ﬁrst formulated by
using the theory proposed by Saanouni and his coworkers (1994)
in the framework of thermodynamics theory for constitutive equa
tions. The resulting equations were extended to the irradiation
condition by expressing the material constants as functions of
neutron ﬂuence. The material functions were expressed by taking
into account the interaction of dislocation motion with the
irradiationinduced defects based on the mechanism of dispersion
strengthening. Finally, the constitutive equations were applied to
simulate uniaxial tensile behavior of a lightwater reactor pressure
vessel steel, ASTM A533B cl. 1.
The results of the present paper are summarized as follows:
Fig. 5 Predicted and experimental yield stress of unirradiated and irra
diated A533B cl. 1 steel at 260°C (Hunter and Williams, 1971; Williams
and Hunter, 1973; Steichen and Williams, 1975)
Fig. 6 Predicted and experimental ultimate tensile strength of unirradi
ated and irradiated A533B cl. 1 steel at 260°C (Hunter and Williams, 1971;
Steichen and Williams, 1975)
Fig. 7 Predicted and experimental total elongation of unirradiated and
irradiated A533B cl. 1 steel at 260°C (Hunter and Williams, 1971; Steichen
and Williams, 1975)
Fig. 8 Predicted and experimental strain hardening of unirradiated and
irradiated A533B cl. 1 steel at 260°C (Hunter and Williams, 1971; Steichen
and Williams, 1975)
Fig. 9 Predicted and experimental strain rate dependence of yield
stress of unirradiated and irradiated A533B cl. 1 steel at 260°C (Steichen
and Williams, 1975)
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 65
1) The present model was able to describe the true tresstrue
strain curves for irradiated materials.
2) In particular, the increase in yield stress and tensile strength
as well as the reduction in fracture strain and the amount of strain
hardening could be presented by the uniﬁed constitutive model.
3) Earlier onset and an accelerated rate of damage evolution in
irradiated materials could be modeled in the framework of contin
uum damage mechanics.
4) Though the signiﬁcant variation in the experimental data of
fracture strain prevented the exact comparison, qualitative agree
ment was observed in the predictions and the experiments.
As mentioned above, by introducing the notion of the continuum
damage mechanics into uniﬁed inelastic constitutive equations, the
present model can describe the mechanical behavior (i.e., stress
strain relation of irradiated materials) up to the fracture under
different conditions of neutron irradiation. However, most of pre
ceding models describe only the change of speciﬁc mechanical
properties due to the irradiation. Therefore, incorporation of the
present constitutive model into a ﬁnite element method facilitates
the crack analysis in nuclear structures. By using this analysis (so
called a local approach), for instance, fracture toughness of the
structural components can be predicted theoretically.
The effects of temperature are not considered in the present
paper. However, it may be straight forward to incorporate the
effects into the present model by formulating material functions as
a function of both neutron ﬂuence and temperature. If the temper
ature effects were incorporated into the present model, DBTT shift
could be easily predicted by analysis.
References
Barrett, C. R., Nix, W. D., and Tetelman, A. S., 1973, The Principles of Engineer
ing Materials, PrenticeHall, Englewood Cliffs, NJ.
Broek, D., 1973, “The Role of Inclusions in Ductile Fracture and Fracture Tough
ness,” Engng. Fracture Mech., Vol. 5, No. 1, pp. 55–66.
Erve, M., and Tenckhoff, E., 1993, “Residual Life Assessment, Technical Improve
ments and Backﬁtting of Components and Systems from the Materials Standpoint,”
Nucl. Engng. Des., Vol. 144, pp. 139–148.
Gage, G., and Little, E. A., 1994, “Materials Characterization for Advanced
Pressurized Water Reactors. Part 1: Mechanical Properties,” Nucl. Energy, Vol. 33,
No. 3, pp. 153–162.
Hunter, C. W., and Williams, J. A., 1971, “Fracture and Tensile Behavior of
NeutronIrradiated A533B Pressure Vessel Steel,” Nucl. Engng. Des., Vol. 17, pp.
131–148.
Klueh, R. L., 1993, “Tensile Behavior of NeutronIrradiated Martensitic Steels: A
Review,” Nucl. Tech., Vol. 102, pp. 376–385.
Klueh, R. L., and Alexander, D. J., 1995, “Embrittlement of CrMo Steels after
Low Fluence Irradiation in HFIR,” J. Nucl. Mater., Vol. 218, pp. 151–160.
Klueh, R. L., and Alexander, D. J., 1992, “Embrittlement of 9Cr1Mo VNb and
12Cr1Mo VW Steels Irradiated in HFIR,” J. Nucl. Mater., Vol. 187, pp. 60–69.
Klueh, R. L., and Maziasz, P. J., 1992, “Effect of Irradiation in HFIR on Tensile
Properties of CrMo Steels,” J. Nucl. Mater., Vol. 187, pp. 43–54.
Klueh, R. L., and Vitek, J. M., 1991, “Tensile Properties of 9Cr1Mo VNb and
12Cr1Mo VW Steels Irradiated to 23 dpa at 390 to 550°C,” J. Nucl. Mater., Vol. 182,
pp. 230–239.
Koehler, J. S., 1952, “The Production of Large Tensile Stress by Dislocations,”
Phys. Rev., Vol. 85, pp. 480–481.
Kussmaul, K., Fohl, J., and Weissenberg, T., 1990, “Investigation of Materials
from a Decommissioned Reactor Pressure Vessel—A Contribution to the Understand
ing of Irradiation Embrittlement,” ASTM STP, Vol. 1046, pp. 80–104.
Lemaitre, J., 1985, “A Continuous Damage Mechanics Model for Ductile Frac
ture,” ASME JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 107, No. 1,
pp. 83–89.
Lemaitre, J., 1992, A Course on Damage Mechanics, SpringerVerlag, Berlin.
Lemaitre, J., and Chaboche, J.L., 1990, Mechanics of Solid Materials, Cambridge
University Press, Cambridge.
Little, E. A., 1976, “NeutronIrradiation Hardening in Irons and Ferritic Steels,”
Int. Metals Rev., Vol. 21, pp. 25–60.
Little, E. A., and Gage, G., 1994, “Materials Characterization for Advanced
Pressurized Water Reactors. Part 2: Microstructure,” Nucl. Energy, Vol. 33, No. 3, pp.
163–172.
Mahmood, S. T., AlOtaibi, K. M., Jung, Y. H., and Murty, K. L., 1990, “Dynamic
StrainAging and Neutron Irradiation Effects on Mechanical and Fracture Properties
of A533B Class 1 PV Steel and 2.25 CrMo Steel,” J. Test. Eval., Vol. 18, pp.
332–337.
Makin, M. J., and Minter, F. J., 1960, “Irradiation Hardening in Copper and
Nickel,” Acta Metall., Vol. 8, pp. 691–699.
Mansur, L. K., and Farrrell, K., 1990, “On Mechanisms by which a Soft Neutron
Spectrum may Induce Accelerated Embrittlement,” J. Nucl. Mater., Vol. 170, pp.
236–245.
Odette, G. R., and Lucas, G. E., 1992, “Deformation and Fracture in Irradiated
Austenitic Stainless Steels,” J. Nucl. Mater., Vol. 191–194, pp. 50–57.
Odette, G. R., and Lucas, G. E., 1991, “The Effects of Intermediate Temperature
Irradiation on the Mechanical Behavior of 300series Austenitic Stainless Steels,”
J. Nucl. Mater., Vol. 179–181, pp. 572–576.
Phythian, W. J., and English, C. A., 1993, “Microstructural Evolution in Reactor
Pressure Vessel Steels,” J. Nucl. Mater., Vol. 205, pp. 162–177.
Saanouni, K., Forster, Ch., and BenHatira, F., 1994, “On the Anelastic Flow with
Damage,” Int. J. Damage Mech., Vol. 3, No. 2, pp. 140–169.
Scott, P., 1994, “A Review of Irradiation Assisted Stress Corrosion Cracking,”
J. Nucl. Mater., Vol. 211, pp. 101–122.
Shockey, D. A., Seaman, L., Dao, K. C., and Curran, D. R., 1980, “Kinetics of Void
Development in Fracturing A533B Tensile Bars,” ASME Journal of Pressure Vessel
Technology, Vol. 120, pp. 14–21.
Steichen, J. M., and Williams, J. A., 1975, “Effect of Strain Rate and Temperature
on the Tensile Properties of Irradiated ASTM A533Steel,” J. Nucl. Mater., Vol. 57,
pp. 303–311.
Tetelman, A. S., and McEvily, A. J., Jr., 1970, Fracture of Structural Materials,
Wiley, New York.
Williams, J. A., and Hunter, C. W., 1973, “Irradiation Strengthening and Fracture
Embrittlement of A533B Pressure Vessel Steel Plate and SubmergedArc Weld,”
ASTM STP, Vol. 529, pp. 5–16.
Wronski, A. S., Sargent, G. A., and Johnson, A. A., 1965, “Irradiation Hardening
and Embrittlement in BodyCentered Cubic Transition Metals,” ASTM STP, Vol. 380,
pp. 69–85.
66 / Vol. 122, JANUARY 2000 Transactions of the ASME
N. J. Mills
A. Gilchrist
School of Metallurgy and Materials,
University of Birmingham,
Edgbaston, Birmingham B15 2IT, U.K.
email: N.J.MILLS@BHAM.AC.U.K.
High Strain Extension of
OpenCell Foams
The high strain tensile deformation of opencell foams is analyzed, using a Kelvin foam
lattice model. The stretching, bending, and twisting of elastic cell edges is analyzed, and
the deformed cell shapes predicted. The stressstrain relation and Poisson’s ratio are
predicted for strains up to 40%, for tension in the [100] and [111] directions of the BCC
lattice. The latter prediction is closest to stressstrain curves for polyurethane foams,
especially when the cell shape anisotropy is taken into account. The change from edge
bending to extension as the main deformation mechanisms, for strains exceeding 20%,
increases the slope of the stressstrain curve. A comparison is made with irregular cell
structure models.
Introduction
Open cell polyurethane (PU) foams are used in seating and
bedding. At ﬁrst sight, these applications appear to involve only
compressive strains. However, near the periphery of an indenter
pressing on a foam sheet, there are high shear strains, so one of the
principal strains can be tensile. The uniaxial tensile foam response
is considered as an initial step in dealing with biaxial tensile plus
compressive deformation.
Although the PU foams have polygonal cells of variable shape
and size, it is easier to analyze a regular structure. Lord Kelvin
(1887) showed that space could be partitioned into identical tetra
kaidecahedral cells of minimal surface area. The vertices are the
interstices of a body centered cubic (BCC) lattice of atoms. The
cells have six planar faces with 4 curved sides, and eight nonplanar
hexagonal surfaces. If the faces are removed, the remaining edges
form an open cell foam. Zhu et al. (1997) analyzed the high strain
compression of a modiﬁed Kelvin structure with initially straight
edges, called a Kelvin foam by Warren and Kraynik (1997). The
structure is almost elastically isotropic for small deformations, and
its response for compression in the [111] direction is a good
approximation to the highstrain response of PU foams.
Gent and Thomas modeled the lowstrain deformation of open
cell foams (1959), emphasizing the stretching of cell edges. They
and Lederman (1971) considered the variation of the foam
Young’s modulus with its density, and not the high strain response.
Their models predict a linear variation of modulus with relative
density R (the foam density divided by the polymer density) for
foams with low R values. Lederman showed that the orientation
distribution of cell edges relative to the tensile stress axis affected
the Young’s modulus. Hence PU foams with anisotropic edge
orientation distributions have anisotropic elastic properties. Ko
(1965) analyzed the moduli of both face centered cubic (FCC) and
close packed hexagonal (CPH) lattices of cells, considering edge
stretching and bending. The CPH model has continuous straight
edges running in some directions, and both models have vertices at
which 8 edges meet. Dement’ev and Tarakanov (1970) analyzed
the Young’s modulus of the BCC Kelvin foam in the [001]
direction, assuming square crosssection edges and only consider
ing edge bending. Warren and Kraynik (1991) analyzed the elastic
nonlinearity due to edge reorientation, considering both edge bend
ing and stretching, for tetrahedral units having four identical half
edges. The elastic properties of a collection of units, with random
orientations, were averaged. The bending moment is assumed to be
zero at edge midpoints, due to pinjoint connection. This is a
reasonable for real foam structures, but randomly oriented tetra
hedra do not ﬁll space, unlike the unit cells of the BCC or FCC
lattice models. They predicted nonlinear compressive behavior at
strains less than 15%.
Finite Element Analysis (FEA) was used to model the response
of foam seating (Pajon et al., 1996). ElRatal and Mallick (1996)
argued that the tensile response of the foams is needed for FEA
modeling, and gave such data for typical PU foams. The uniaxial
tensile response considered here is a ﬁrst step in the modeling the
biaxial strain response.
Modeling
Introduction. The analysis is for foams with relative density
Ͻ0.05, for which the cell edge widths are much smaller than their
lengths. At relative densities Ͼ0.3 the microstructure consists of
connected nearspherical holes (Dawson and Shortall, 1982),
whereas at low densities the polygonal cell structure is more fully
developed. The majority of PU foam for seating and bedding
applications has a relative density less than 0.03.
Elastic anisotropy increases as a Kelvin foam deforms in tension
or compression, since the geometric nonlinearity depends on the
direction of loading. In the low strain analysis (Warren and
Kraynik, 1997, Zhu et al., 1997) the Young’s modulus in the [001]
direction E
001
is the highest, and E
111
is the lowest. The cubic
symmetry means that the Young’s modulus is the same in all
equivalent ͗100͘ or ͗111͘ directions. The loading directions chosen
provide the greatest symmetry of edge deformation, simplifying
the analysis by reducing the number of edges to be analyzed. It is
likely that the [100] and [111] directions will produce the extremes
of high strain behavior.
A Kelvin Foam Loaded in the [001] Direction. For concise
ness, the previous compressive analysis (Zhu et al., 1997) will be
referred to. A uniaxial tensile stress
z
is applied in the [001]
Miller index direction (Fig. 1). The boundary planes of the struc
tural cell (Zhu et al., 1997) in Fig. 2 remain mirror symmetry
planes at high deformations. Symmetry means that the deformation
analysis reduces to that of a single edge BC, which lies in the yz
plane of the BCC lattice. Horizontal edges (those in the xy plane)
lie in mirror planes, and do not bend.
The forces P, acting on the vertices B, R, D, and G, are related
to the applied engineering stress tensile
z
by
P ϭ 2L
2
z
(1)
using the sign convention that P is positive for tension. In the
compressive analysis, edge axial compression was ignored com
pared with edge bending, and the results expressed as elliptic
integrals. However, to consider edge extension in the tensile mod
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
August 20, 1998; revised manuscript received April 22, 1999. Associate Technical
Editor: S. Meguid.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 67
Copyright © 2000 by ASME
eling, numerical methods were used. The shear deformation of the
edges was ignored in comparison with their extension, since the
cell shape change at high foam extensions means that the force
vector P acts nearly parallel with the edge BC.
If the volume of vertices is ignored, the foam relative density R
is related (Zhu et al., 1997) to the edge length L and crosssectional
area A by
R ϭ
3A
2 ͱ2 L
2
(2)
The Plateau border edge section consists of three touching circular
arcs, meeting tangentially. Its crosssectional area A is (͌3 Ϫ
/ 2)b
2
, where b is the edge breadth. Phelan et al. (1996) calcu
lated the volume of a vertex plus 4 half edges, as a function of the
length L between vertexcentres, by minimizing the surface area.
The total volume V (per edge) was
V ϭ ͑ ͱ3 Ϫ /2͒b
2
͑L ϩ 1.50b͒ (3)
if L Ͼ b. They did not give the distance from the vertex to the
parallel sided part of the edge, or the exact vertex shape. A ﬁnite
element calculation of the vertex deformation would show whether
our assumptions, that the edges have uniform crosssections and
the vertices are rigid, produce signiﬁcant errors. Warren et al.
(1997) calculated the second moment of area I and the polar
second moment of area J of the Plateau border section as
I ϭ 0.1338A
2
J ϭ 0.0803A
2
(4)
The shape of the edge BC has twofold rotational symmetry
about its midpoint, at which the moment is zero (Zhu et al., 1997).
A curvilinear coordinate s, with origin O, deﬁnes the position in
the half edge OB (Fig. 2(b)), which lies in the yz plane. is the
angle between the tangent to the edge and the z axis. The moment
M at a general position is P( y
O
Ϫ y) and the differential equation
for the beam curvature is
EI
d
ds
ϭ M (5)
The tensile strain ⑀
s
along the edges is calculated using
⑀
s
ϭ
P cos
EA
(6)
This affects the length ds of the 100 segments into which the half
edge is split. In an iterative solution, the moment M is calculated
from the original edge shape, then the curvature, calculated from
M, is used to estimate a new edge shape. The solution is partially
relaxed towards the new shape and the process repeated until the
edge shape converges. Figure 3 shows the reduced stress (divided
by ER
2
) versus tensile strain for a relative density of 0.025. There
is a steady increase in slope as the slanting edges reorient toward
the tensile axis, and the edge stretching contribution increases
(Table 1). Figure 4 shows how the maximum tensile strain in the
edges, at the edge midpoints, increases with the foam strain and is
about 7% when the foam strain is 40%. The relative proportion of
edge stretching to bending increases for higher relative density
foams.
The maximum bending strain, which occurs at a cusp of the
Plateau border next to a vertex (Table 1), is greater than the foam
macroscopic strain. This indicates that the PU will be in the
Nomencl at ure
A ϭ edge crosssectional area
b ϭ breadth of a Plateau border
C ϭ constant
E ϭ polymer Youngs modulus
E
001
ϭ Youngs modulus of foam loaded
in the [001] direction
L ϭ edge length
I ϭ second moment of area of edge
cross section
J ϭ polar moment of area of edge
cross section
M ϭ bending moment, with compo
nents M
x
, M
y
, M
z
P ϭ force
R ϭ foam relative density (foam
density/polymer density)
s ϭ position coordinate
x, y, z ϭ Cartesian coordinates
␣, , ␥ ϭ angles between the edge CG
and the x, y, z axes, respec
tively
ϭ angle between the edge OD and
the z axis
ϭ twisting angle of edge CG
ϭ rotation angle of vertex C
⑀
z
ϭ foam compressive strain
⑀
x
, ⑀
y
ϭ foam lateral strain
⑀
s
ϭ edge axial strain
ϭ foam Poisson’s ratio
z
ϭ uniaxial tensile stress
Fig. 1 A uniaxial tensile stress
z
applied in the [001] direction of a
Kelvin open cell foam. Chains of edges, like those in bold, take the load.
Fig. 2 (a) The structural cell for [001] direction tension of the Kelvin
foam, (b) the effective load on the halfedge BO. Its shape is shown for
20% foam strain.
68 / Vol. 122, JANUARY 2000 Transactions of the ASME
nonlinear region (Zhu et al., 1997) when the foam strain is greater
than 10%, so the calculations should really be made using a
nonlinear material model. The bending strains are zero at the edge
midlength where the tensile strain is largest.
As the foam strain tends to zero, the slope of the stressstrain
curve becomes the Young’s modulus E
001
. At a strain of 0.1%, for
a relative density R ϭ 0.001, the secant Young’s modulus is
E
100
ϵ
z
⑀
z
ϭ 1.009ER
2
(7)
This is the theoretical low strain value (Zhu et al., 1997). However,
if E
100
is the secant modulus, the constant in Eq. (7) has increased
to 1.046 when the strain is 5% and R ϭ 0.03. The horizontal
projected length y
L
of the half edge (Fig. 2(b)) is used to calculate
the diagonal distance across the top of the deformed structural cell
(Fig. 2(a)), hence the lateral tensile strain (Zhu et al., 1997).
Poisson’s ratio is deﬁned as the ratio Ϫ⑀
y
/⑀
z
of the engineering
strains. Figure 5 shows that this increases slowly with the foam
tensile strain, for strains up to 40%.
A Kelvin Foam Extended in the [111] Direction. The anal
ysis for compression (Zhu et al., 1997) used an approximation for
the edge bending curvature, hence overestimated the low strain
Young’s modulus. For edges with Plateau border crosssections,
the equivalent of Eq. (7) contained the constant 1.10 rather than the
0.96 value from low strain theory. An exact expression for the
edge curvature is now used, and in other parts of the analysis, some
equations are simpliﬁed.
Symmetry Considerations and Boundary Conditions. A new
set of axes are chosen, with the z axis parallel to the lattice [111]
direction, and the x axis parallel to the projection of the unde
formed edge CG onto the plane normal to z. Edges lying initially
in the xy plane are horizontal. Mirror symmetry planes divide the
lattice into triangular prisms, having at their centres a threefold
screw axis parallel to the z axis. The helix of edges in each prism
transmits a force P (Fig. 6), given (Zhu et al., 1997) by
P ϭ
4
ͱ3
L
2
z
(8)
Table 1 [100] Extension of a Kelvin foam of relative density
0.025 with Plateau border edges
Foam
strain
%
Reduced
stress
ER
2
Poisson’s
ratio
Maximum edge
strain %
Reduced
moment at D
M
ER
2
L
3
Tensile Bending
9.9 0.051 0.513 0.4 13 0.074
19.8 0.292 0.570 1.3 29 0.160
30.3 0.628 0.632 3.1 50 0.274
40.0 1.255 0.655 6.5 74 0.411
Fig. 3 The predicted reduced stress–strain relation for the Kelvin foam
loaded in the [001] and [111] directions, using Plateau border edges, plus
Shulmeister’s (1998) random cell prediction – – – – – – with circular sec
tion edges, all for a relative density of 0.025
Fig. 4 The maximum edge tensile strain versus the foam tensile strain,
for the Kelvin foam extended in the [001] and [111] directions, with the
same parameters as Fig. 3
Fig. 5 Predicted variation of Poisson’s ratio with foam tensile strain, for
the Kelvin foam loaded in the [001] and [111] directions as in Fig. 3,
compared with ElRatal and Mallick’s Poisson’s ratio data F (1996) for a
commercial PU foam (increased by 0.15 to make the comparison easier)
Fig. 6 (a) xy projection of a triangular prism showing two sloping edges
and two half horizontal edges, in both the undeformed and deformed
states, for [111] direction extension of the Kelvin foam by 20%. G lies
beneath and R above the plane of the projection (b) a perspective view of
the force and moments applied at G to the edge CG.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 69
The direction of the force P and the sign convention for the stress
are opposite to that used in the compressive case. The symmetry of
the loading direction in the lattice means that
(a) The tensile and shear forces are zero in horizontal edges.
Their midpoints do not rotate about the z axis, and they
deform into circular arcs, lying in a vertical plane.
(b) The helices do not rotate relative to the prisms. The vertex
C and the midpoint of CG do not rotate about the z axis,
and their displacements in the xy plane are directed away
from the screw axis.
(c) The vertex C rotates clockwise about an axis QO by an
angle , which is positive for tension.
(d) The vertex G lies in the xz plane.
The coordinate axis origin is chosen at C (Fig. 6(a)). The angles
between a segment of the edge CG and the x, y, and z axes are ␣,
, and ␥, respectively. A curvilinear coordinate s is used for
position on CG, with s ϭ 0 at C and s ϭ L at G. Zhu et al. (1997)
gave expressions for the orientation of the undeformed edge CG,
and the orientation of the edge at C in the deformed cell, when the
vertex C has rotated by .
The Equilibrium of the Moments Acting at the Vertex C. Zhu
et al. (1997) analyzed the equilibrium of the moments acting at C,
for compression. The vertex rotation angle was negative, al
though this was not stated. In the tensile case is positive. The
constant bending moment M
h
in the horizontal edge BC is
M
h
ϭ
ͱ3 EI
L
(9)
The vector moment transmitted from the vertex G to the edge GC,
has components M
x
G
M
y
G
and M
z
G
. The equilibrium of the moments
acting at C gives the equation
ͱ3 M
x
G
Ϫ M
y
G
ϩ 2P
͵
0
L/2
cos ␣ds ϭ
3EI
L
(10)
Zhu et al. (1997) used an integral for whole edge in the equivalent
equation.
Torsion of the Edge CG. At a general point on the edge, the
total moment M has components
M ϭ
ͩ
M
x
G
Ϫ P
͵
0
s
cos ds,
Ϫ M
y
G
ϩ P
͵
0
s
cos ␣ds, M
z
G
ͪ
(11)
In Zhu et al. (1997) the limits of the second integral were from s
to L, and the sign of the M
y
G
term was positive. The moment of the
component parallel to the edge causes torsion. The increment of
twist d in an element ds of edge length is
d ϭ
M ⅐ S
GJ
ds (12)
where G is the polymer shear modulus, J the edge polar second
moment of area, and S the edge unit vector with Cartesian
components (cos ␣, cos , cos ␥). Equation (12) is integrated
along the half edge, using the boundary condition that (0) ϭ
/2, giving
M
x
G
͵
0
L/2
cos ␣ds Ϫ M
y
G
͵
0
L/2
cos ds ϩ M
z
G
͵
0
L/2
cos ␥ds
Ϫ P
͵
0
L/2
ͩ
͵
0
s
cos ds
ͪ
cos ␣ds
ϩ P
͵
0
L/2
ͩ
͵
0
s
cos ␣ds
ͪ
cos ds ϭ Ϫ
2
GJ (13)
Zhu et al.’s (1997) equivalent equation contains integrals for the
whole edge.
Bending of the Edge CG. The components of M that act
perpendicular to S cause edge bending. The second moment of
area I of the Plateau border edge cross section is independent of
the direction of bending, so the change ⌬S in the unit vector along
an element ds of edge is given by the vector product
⌬S ϭ
M ϫ S
EI
ds (14)
Substituting for M from Eq. (11), the components of ⌬S are
⌬ cos ␣ ϭ
ds
EI
ͩ
ϪM
y
G
cos ␥ Ϫ M
z
G
cos 
ϩ P cos ␥
͵
0
s
cos ␣ds
ͪ
(15)
⌬ cos  ϭ
ds
EI
ͩ
ϪM
x
G
cos ␥ ϩ M
z
G
cos ␣
ϩ P cos ␥
͵
0
s
cos ds
ͪ
(16)
⌬ cos ␥ ϭ
ds
EI
ͩ
M
x
G
cos  ϩ M
y
G
cos ␣
Ϫ P cos ␣
͵
0
s
cos ␣ds Ϫ P cos 
͵
0
s
cos ds
ͪ
(17)
The integral of Eq. (15) along the whole of CG is zero since
␣(0) ϭ ␣(L). Since the integral ͐
0
L
cos ds ϭ 0, the remaining
terms can be analyzed to show
M
y
G
Ϫ P
͵
0
L/2
cos ␣ds ϭ 0 (18)
Equation (16) when integrated along the half edge provides
͵
0
L/2
ͩ
ϪM
x
G
cos ␥ ϩ M
z
G
cos ␣ ϩ P cos ␥
͵
0
s
cos ds
ͪ
ds
ϭ ϪEI cos 
0
(19)
The four equations (10), (13), (18), and (19) determine the un
knowns M
x
G
, M
y
G
, M
z
G
, and P. The tensile axial strain in the edge
is calculated from Eq. (6) with the angle ␥ replacing . The effect
of this strain in reducing the edge moments of area is ignored.
The Stress Strain Relationship. The half edge CG is divided
into 100 segments. For incremental rotation of the vertex C, the
moment M
G
and force P are found, then substituted in Eqs. (15) to
(17) to estimate the new edge shape. The edge is partially relaxed
70 / Vol. 122, JANUARY 2000 Transactions of the ASME
toward the new shape, and the process repeated until the change in
angle is less than 10
Ϫ4
radian.
Figure 3 shows the variation of the reduced stress
z
/ER
2
with
the tensile strain. The initial slope of the graph gives the Young’s
modulus E
111
as 0.96ER
2
for Plateau border edges, conﬁrming the
low strain analysis. The slope of the stressstrain curve rises more
rapidly at high strains than for [001] direction tension, because
sloping edges like CG are initially within 35 deg of the tension
axis, rather than at 45 deg. Figure 7 shows a perspective view of
a single Kelvin foam cell at a tensile strain of 25%, in which closer
edges appear larger. The slanting edges align toward the tensile
stress axis and bow toward the helix axis, and the triangular prism
contracts in diameter. The initially horizontal edges bend into
circular arcs of moderate curvature. Table 2 gives the dimension
less moments applied to G, the largest of which is M
y
. Figure 4
shows that the maximum axial edge extensions, at the midpoints of
the slanting edges, are greater for [111] than for [001] direction
extension, at a particular foam strain. However, the maximum
bending strain in the edge, which occurs near the vertices, is much
larger.
Poisson’s Ratio. The edge length of the triangular prism (Fig.
6) is used to calculate the lateral strain and Poisson’s ratio (Zhu et
al., 1997). Figure 5 shows that the Poisson’s ratio for [111]
direction extension is slightly larger than for the [100] direction,
but it reaches a maximum and decreases for strains exceeding
25%.
ElRatal and Mallick (1996) found for their “commercial” PU
foam of density 17 kg m
Ϫ3
that the contraction in both lateral
directions was the same within experimental error. Their Poisson’s
ratio data in Fig. 5 is about 0.15 smaller than the predictions,
showing the same increasing trend with the foam strain. They
preferred to calculate the apparent Poisson’s ratio from the ratio
of the true strains; these values differ from the Poisson’s ratio if the
strain are high. For high strains, a Poisson’s ratio of 0.5 does not
mean that the foam volume remains constant.
Anisotropic Cell Shapes. The PU upholstery foam tested has
anisotropic cell shapes, resulting from the manufacturing process.
Scanning electron microscopy showed that the cell diameter in the
rise direction was 30% greater than in perpendicular directions.
The Kelvin foam model can be made anisotropic, with cells
elongated or compressed along the [100] or [111] axes, while
maintaining most of the structural symmetry. The angle, between
the sloping edges and the tensile stress axis, changes from the
value for equiaxed cells. However, the transverse isotropy pro
duced is not exactly the same as that of a PU foam extended in the
plane of the sheet.
Comparison With a Random Cell Model. Schulmeister
(1998) made predictions based on regular and randomized Kelvin
foams. His predictions for the regular Kelvin foam in tension along
the [100] direction should be the same as those reported here.
However, the single reduced stress versus true strain graph is
presented on scales that preclude accurate reading of the low strain
data. His predicted stress strain curve for the random model, for a
relative density of 0.025 and circular crosssection edges, is almost
the same as our [111] direction prediction (Fig. 3). However,
changing from Plateau border to circular edge crosssections re
duces the second moment of area by 40%, so the agreement is not
as good as it appears. He noted that the random model stress was
signiﬁcantly higher than the [100] direction Kelvin model stress at
moderate foam strains, attributing this to the chains of edges
aligned close to the stress axis.
Experimental
A commercial upholstery PU foam, of density 26 kg m
Ϫ3
was
obtained in the form of a 25 mm thick sheet. This had been cut
from a continuously foamed block, so the foam rise direction was
perpendicular to the sheet thickness. It was not possible to fully
densify this crosslinked PU, so the bulk density was assumed to be
1230 kg m
Ϫ3
, the same as a partly crosslinked PU, which had a
Young’s modulus of 50 MPa. It was not possible to evaluate the
polymer Young’s modulus. The cell diameter was approximately 1
mm.
Tensile tests were carried out with an Instron, adapted to mea
sure the foam Poisson’s ratio (Mills and Gilchrist, 1997). The ends
of rectangular foam specimens of length 170 mm, width 42 mm
and thickness 25 mm, were bonded to ﬂat steel plates using epoxy
resin, hence the tensile stress acts perpendicular to the foam rise
direction. The sample size is much greater than the cell size, so
edge effects should be small. Lateral displacement transducers are
spring loaded with a force of about 1 N, were ﬁtted with 20 by 40
mm PTFE plates, so that the compressive lateral stress was about
1 kPa. There was low friction as the foam moved past the plates.
The outputs of the load cell and displacement transducers were
taken via an analoguetodigital converter to a computer.
Fig. 7 Predicted Kelvin foam cell shape for 20% tensile strain in the
[111] direction, seen in perspective with the stress axis vertical
Table 2 [111] Deformation of a Kelvin foam of R ؍ 0.025 with Plateau border edges
Foam
strain
%
Reduced
stress
ER
2
Poisson’s
ratio
Maximum edge strain %
Rotation
degree
Reduced moment at G
Tensile Bending
M
x
ER
2
L
3
M
y
ER
2
L
3
M
z
ER
2
L
3
10.0 0.142 0.54 0.7 14 5.8 Ϫ0.022 0.074 0.002
20.5 0.561 0.62 3.3 37 11.6 Ϫ0.066 0.188 Ϫ0.003
30.4 1.50 0.58 9.1 63 14.6 Ϫ0.128 0.314 Ϫ0.008
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 71
The response of the PU foam changes with the deﬂection cycle
(Fig. 8). For the second and later cycles the response is near
constant, with a negative stress at zero strain. The change can be
attributed to viscoelasticity, since the foam has been resting at zero
strain for a long time before the ﬁrst cycle of extension. There is
signiﬁcant hysteresis, contradicting ElRatal and Mallick’s (1996)
observation of no discernible difference between the loading and
unloading response when there was a 1 minute wait at each load.
The response in the absence of viscoelasticity, estimated from the
mean of the loading and unloading stress for a particular strain,
suggests a nonlinear response. Figure 8 shows the predicted re
sponses of the Kelvin model for [111] direction extension, assum
ing a PU Young’s modulus of 100 MPa, and either equiaxed cells,
or cells compressed along [111] by 30% so the slanting edges are
initially at 45 deg to the stress axis. The latter assumption gives the
better ﬁt to the experimental data. The prediction is shifted to the
negative stress start of the cyclic response. The assumed modulus
is realistic for low density PU foams, since the use of high water
contents in the mixture causes an increase in the hard block content
and hence in the Young’s modulus of the PU.
The foam response is anisotropic. When the tensile stress is in
the plane of the sheet, Fig. 9 shows that the contraction measured
in the rise direction gives a higher Poisson’s ratio (0.8) than the
contraction in the other direction (0.5). The foam volume de
creases with the imposition of tensile strain, in agreement with the
observations of ElRatal and Mallick on their commercial foam.
The data in Fig. 9 show slight hysteresis, which may be due to the
pressure and frictional shear stress from the transducers. The near
constant slope of the data implies that Poisson’s ratio is nearly
constant, unlike the data of ElRatal and Mallick shown in Fig. 5.
Discussion
The predicted response of the Kelvin foam for [111] direction
extension is closer to the measured PU foam response than is that
for the [100] direction, the same result as for compression. How
ever, the assumed PU Young’s modulus, although reasonable,
cannot be directly checked experimentally. Shulmeister’s (1998)
model of a randomized Kelvin foam, if it used Plateau border
edges, would predict a 40% higher stress than for a regular Kelvin
foam, of the same density, loaded in the [111] direction to the same
strain. The regular foam model provides the quantitative link
between edge orientation and the stress strain curve missing from
Lederman’s (1971) paper. It is not possible to compare the pre
dictions with Warren and Kraynik (1991) tetrahedral unit model
because they made no tensile predictions. Kraynik et al. (1999)
FEA predictions of the high strain compressive response of the
Kelvin open cell model, for both the [001] and [111] directions, are
identical to the compressive predictions using the method of this
paper, conﬁrming the validity of the analyses.
The predicted Poisson’s ratio remains above 0.5, in agreement
with our experiments, and most of ElRatal and Mallick’s data.
This contrasts with compression where edge buckling is the dom
inating deformation mechanism, and Poisson’s ratio falls to low
values.
Since the edge tensile compliance depends inversely on the ﬁrst
power, and the edge bending compliance on the second power of
the relative density, the low strain tensile response is dominated by
edge bending for low relative density foams. The edge alignment
with the tensile axis is almost independent of the foam density, as
is the relative hardening of the stress strain curve at high strains.
Only when the foam tensile strain is mainly provided by edge
extension does the stress become linearly dependent on the foam
relative density—this region is rarely met in PU foams since
failure intercedes. This contrasts with compression, where the
axial strain in the edges can be neglected, and there is a near
plateau in stress for strains exceeding 10%. The reduced compres
sive collapse stress of about 0.1 is much less than the reduced
tensile stresses shown in Fig. 3.
The response of PU foams is complicated by:
(a) Viscoelasticity. For creep loading, elastic predictions can
be converted into approximate viscoelastic solutions (Zhu
and Mills, 1999), by replacing the Young’s modulus by the
reciprocal of the creep compliance J(t), except in the
vicinity of the glass transition temperature. The majority of
the hysteresis in the tensile loading and unloading response
is due to viscoelasticity.
(b) Polymer nonlinearity. The maximum polymer strain in
creases with the foam relative density, at a particular foam
strain. In commercial PU foams at strains Ͼ10%, the
polymer is in the nonlinear region, so the predicted re
sponses in Fig. 8 are overestimates.
(c) The anisotropy of the cell structure, which makes the foam
response anisotropic.
(d) Temperature, which has a similar effect to strain rate in
changing the response.
Fig. 8 Stress strain curve for a PU upholstery foam of relative density
0.021, loaded in the plane of the sheet, cycled three times to 25% strain
at a nominal strain rate of 0.005 s
؊1
. The Kelvin foam [111] predictions
are ⅐ ⅐ ⅐ ⅐ ⅐ ⅐ for equiaxed cells and – – – – – – for cells with edges initially
at 45 deg to the stress, using E ؍ 100 MPa for the polyurethane.
Fig. 9 Lateral contraction in the rise —— and inplane – – – – – – direc
tions versus tensile strain for the same experiment as Fig. 8, for cycles
2 and 3
72 / Vol. 122, JANUARY 2000 Transactions of the ASME
The simple shear response of opencell foams will be inﬂuenced
by the tensile principal stress, which may be greater than the
compressive principal stress. Micromechanics analysis may be
able to predict whether a large compressive principal strain causes
sufﬁcient edge buckling to affect the tensile response along the
other principal axis. PU foam response under such biaxial strain
states, where one principal strain is tensile, needs to be investi
gated further.
Acknowledgment
The authors acknowledge the support of ESPRC and DERA
under grant number GR/L 81307.
References
Dawson, J. R., and Shortall, J. B., 1982, “The Microstructure of Rigid Polyurethane
Foam,” J. Mater. Sci., Vol. 17, pp. 220–224.
Dement’ev, A. G., and Tarakanov, O. G., 1970, “Model Analysis of Plastics Foams
of the Polyurethane Type,” Mekh. Polim., Vol. 6, pp. 859–865, translated as Polymer
Mechanics, Vol. 6, pp. 744–749.
ElRatal, W. H., and Mallick, P. K., 1996, “Elastic Response of Flexible Polyure
thane Foams in Uniaxial Tension,” ASME JOURNAL OF ENGINEERING MATERIALS AND
TECHNOLOGY, Vol. 118, pp. 157–161.
Gent, A. N., and Thomas, A. G., 1959, “The Deformation of Foamed Elastic
Materials,” J. Appl. Polymer. Sci., Vol. 1, pp. 107–113.
Kelvin, Lord (Thompson, W.), 1887, “On the Division of Space with Minimum
Partitional Area,” Phil. Mag., Vol. 24, pp. 503–514.
Ko, W. L., 1965, “Deformations of Foamed Elastomers,” J. Cellular Plastics, Vol.
1, pp. 45–50.
Kraynik, A. M., Neilsen, M. K., Reinhelt, R. A., and Warren, W. E., “Foam
Micromechanics,” NATO Adv. Sci. Inst. Series E: Appl. Sci., Vol. 354, Foams and
Emulsions, N. Rivier and J. F. Sadok, eds., pp. 187–204, Kluwer.
Lederman, J. M., 1971, “The Prediction of the Tensile Properties of Flexible
Foams,” J. Appl. Polym. Sci., Vol. 45, pp. 693–703.
Mills, N. J., and Gilchrist, A., 1997, “The Effect of Heat Transfer and Poisson’s
Ratio on the Compressive Response of Closed Cell Polymer Foams,” Cell. Polym.,
Vol. 16, pp. 87–119.
Pajon, M., Backacha, M. et al., 1996, “Modelling of PU Foam Behaviour—
Applications in the Field of Automotive Seats,” SAE SP1155, paper 960513.
Phelan, R., Weaire, D., Peters, E. A. J. F., and Verbist, G., 1996, “The Conductivity
of a Foam,” J. Phys. Condens. Matter, Vol. 8, pp. L475–L482.
Schulmeister, V., 1998, “Modelling of the Mechanical Properties of Low Density
Polymer Foams,” Ph.D. thesis, Technical University of Delft.
Warren, W. E., and Kraynik, A. M., 1991, “The NonLinear Elastic Behavior of
OpenCell Foams,” ASME Journal Appl. Mechanics, Vol. 58, pp. 376–381.
Warren, W. E., and Kraynik, A. M., 1997, “Linear Elastic Behavior of a Low
Density Kelvin Foam with Open Cells,” ASME Journal Appl. Mechanics, Vol. 64,
pp. 787–794.
Warren, W. E., Neilsen, M. K., and Kraynik, A. M., 1997, “Torsional Rigidity of
a Plateau Border,” Mechanics Res. Comm., Vol. 24, pp. 667–672.
Zhu, H., Knott, J. F., and Mills, N. J., 1997, “The Elastic Constants of Open Cell
Foams Having Tetrakaidecahedral Cells,” J. Mech. Phys. Solids, Vol. 45, pp. 319–
343.
Zhu, H. X., Mills, N. J., and Knott, J. F., 1997, “Analysis of the High Strain
Compression of OpenCell Foams,” J. Mech. Phys. Solids, Vol. 45, pp. 1875–1904.
Zhu, H. X., and Mills, N. J., 1999, “Analysis of Creep in OpenCell Foams,” J.
Mech. Phys. Solids, Vol. 47, pp. 1437–1457.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 73
David A. Miller
Dimitris C. Lagoudas
Center for Mechanics of Composites,
Department of Aerospace Engineering,
Texas A&M University,
College Station, TX 778433141
Inﬂuence of Heat Treatment on
the Mechanical Properties and
Damage Development in a
SiC/Ti153 MMC
Titanium alloys are commonly heattreated to meet speciﬁc design requirements. In an
effort to possibly create a better composite, the inﬂuence of heat treatments on the damage
evolution and strength of a SiC/Ti153 metal matrix composite (MMC) was studied. Heat
treatments of 450°C and 700°C for 24 hours were performed on axial and transverse
unidirectional specimens. These specimens, in addition to specimens in the asreceived
condition, were tested under nonproportional loading paths and then microstructurally
analyzed to determine the induced damage. The axial composite with the 450°C heat
treatment showed the highest elastic modulus and the lowest stiffness reduction than the
other heat treatment conditions. The transverse composite in the asreceived condition
showed the highest room temperature elastic modulus and the lowest stiffness reduction
compared with other heat treatment conditions. Typical damage modes of Ti MMC’s, such
as ﬁber/matrix debonding and matrix microcracking, were seen in all heat treatments. A
micromechanics model based on the MoriTanaka averaging scheme was implemented to
simulate the effects of microcracking induced damage on composite stiffness reduction.
1 Introduction
Designs for the next generation of supersonic and hypersonic
aircraft as well as engine components will require advanced ma
terials that can withstand elevated temperatures in structural ap
plications. Metal matrix composites (MMC’s) have recently un
dergone serious evaluation to satisfy this need because of their
high strength, low weight, and elevated temperature capabilities.
Titanium has been the primary matrix material considered because
of its formability, high strengthtoweight ratio, and high melting
temperature (1668°C). Titanium also has the advantage that its
microstructure can be altered by alloying with metals which will
stabilize the ␣ or  phase to satisfy many different design require
ments.
Structural titanium MMC’s are often reinforced with continuous
silicon carbide (SiC) ﬁbers fabricated through a chemical vapor
deposition process. To date, the results for the SiC/Ti systems have
shown severe limitations. The composite shows improved perfor
mance in the ﬁber direction compared with monolithic titanium,
but the transverse ﬁber direction properties are considerably less.
Johnson (1992) has shown in an SCS6/Ti153 MMC composite
with off axis plies that ﬁbermatrix separation leads to a knee in the
stress/strain curve well below the matrix yield level. This damage
was conﬁrmed by a reduction in the unloading modulus and edge
replicas. Combined loading paths have been studied by Lissenden
et al. (1995), in SiC/Ti tubes. Results have shown that ﬁber/matrix
interfacial debonding was observed in the stressstrain response
and veriﬁed through microstructural evaluations.
Mujumdar and Newaz (1992a, b, 1993) have performed room
temperature experimental studies of the SCS6/Ti153 composite
identifying the inelastic deformation mechanisms of plasticity and
damage. Lagoudas et al. (1995) studied the effect of surface
damage on oxidized SiC/Ti153 MMC laminates under tensile
loading, and an effective stiffness reduction due to the develop
ment of cracks on the surface of the composite was evaluated.
Newaz et al. (1992) investigated the thermal cycling response of
quasiisotropic SiC/Ti153 MMC’s. Johnson et al. (1990) per
formed room temperature tests to characterize a SiC/Ti153 MMC
in both the asreceived condition and aged at 482°C for 16 hr. This
heat treatment was shown to increase the elastic modulus, the
ultimate strength and the yield stress of the matrix.
Several different temperatures have been utilized for heat treat
ments of MMC’s in previous works, (Johnson, et al., 1990; Lerch
et al., 1990a) although rigorous comparisons were not made
among different heat treatments. In fact, the motivation behind the
heat treatments was not discussed, other than to stabilize the
microstructure. Wolfenden et al. (1996) utilized nondestructive
techniques to investigate the effect of heat treatments on the elastic
modulus, damping and microhardness for SiC/Ti153. Lerch and
Saltsman (1991), advanced Johnson’s room temperature results
with a study of the SCS6/Ti153 MMC at 427°C. Specimens
were tested asreceived and with a heat treatment of 700°C for 24
hrs. However, the effect of heat treatment and resulting micro
structures on the damage development of MMC’s has not been
addressed.
The objective for this work is to study possible improvements
on a SiC/Ti153 MMC through different heat treatments and to
experimentally characterize the deformation and damage mecha
nisms associated with each. The effects of heat treatment on the
thermomechanical response and damage development in unidirec
tional specimens, loaded axially and transverse to the ﬁbers, are
discussed in the second section. The third section utilizes an
averaging micromechanics technique for the correlation of crack
density as a damage parameter with stiffness reduction and com
pares model predictions with experimental results.
2 Experimental Results on Thermomechanical Re
sponse of SiC/Ti153 With Different Heat Treatments
2.1 Material Description and Specimen Preparation. All
tests in this research were performed on a composite material
manufactured by DWA Specialty Metals using the foilﬁberfoil
technique. The composite is a 4ply unidirectional SiC/Ti 153
MMC with a 32% ﬁber volume fraction. The matrix is reinforced
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
August 5, 1997; revised manuscript received May 3, 1999. Associate Technical
Editor: S. A. Meguid.
74 / Vol. 122, JANUARY 2000 Transactions of the ASME Copyright © 2000 by ASME
by SiC ﬁbers manufactured by British Petroleum. The average
ﬁber diameter was 100 m and has a structured carbontitanium
diboride coating (Allen et al., 1994). The alloy utilized in this
composite is a metastable phase alloy commonly referred to as
Ti 153, and has an actual weight percent of 15% V, 3% Al, 3% Cr,
3% Sn and the balance Ti.
The composite was tested in the asreceived condition and with
heat treatments of 450°C for 24 hours and 700°C for 24 hours.
Heat treatments were performed in a vacuum to prevent oxidation
from degrading the material. The asreceived material, see Fig. 1,
contains large ␣phase needles, labeled as A. The ␣particles are
seen primarily along the ﬁber interface creating a depleted zone
around the ﬁber area. Results from an energy dispersive spectros
copy (EDS) line scan show a high carbon concentration in the ␣
rich region surrounding the ﬁber, possibly due to the coating.
Carbon is an ␣stabilizer for titanium, thus explaining the ␣ rich
region surrounding the ﬁber. The ␣ rich region surrounding the
ﬁber seen in this composite has not been observed in other Ti153
systems (e.g., in the work of Lerch and Saltsman, 1991, where the
SCS6/Ti153 MMC has been studied). Since the ﬁber coatings
differ for the Sigma ﬁber and the SCS6 ﬁber, the possibility of
different ﬁber/matrix interfaces exist.
The 450°C heat treatment produces a small ﬁne ␣particle
precipitant with virtually no congregation along the grain bound
ary, as seen in Fig. 2. The distinct ﬁber matrix interface region
remains, but the precipitant accumulation around the ﬁber has
disappeared. The aging has caused the precipitant to distribute
throughout the grain structure. The grains remain the same size,
but are not as apparent as before the redistribution of the ␣phase.
It has been shown (Lerch et al., 1990a) that this heat treatment
produces the highest room temperature modulus for the matrix
material. This is to be expected since the HCP ␣phase has a higher
stiffness than the BCC phase. However, the increase in ␣phase
also decreases the number of active slip planes thus reducing the
ductility of the material.
The 700°C heat treatment, Fig. 3, results in large ␣phase
needles dispersed throughout the phase grains, while the grain
size remains unchanged. The depleted region around the ﬁbers is
still present, as well as the grain boundary phases. It has been
shown that this heat treatment slightly reduces the stiffness of the
titanium from the asreceived condition (Lerch et al., 1990a).
2.2 Thermomechanical Testing Description. All tests
were performed on an MTS servohydraulic load frame under load
control at a rate of 10.4 MPa/sec. Elevated temperature tests were
performed using a threezone clamshell resistance furnace. Strain
measurements for all tests were made with a oneinch gage high
temperature extensometer.
In order to study the evolution of inelastic deformation through
out the entire loading path, cyclic tension tests with an increasing
stress amplitude were performed at room temperature and at 427°C
(800°F). Measurement of the unloading elastic modulus for each
cycle quantiﬁed the damage imparted into the specimen at the
applied stress level. Cyclic loading paths were used to study the
development of inelastic deformation in the composite and mon
itor the degradation of the initial elastic modulus. The loading path
consisted of unloading to 10% of the maximum applied stress in
each cycle with the maximum stress incrementally increasing in
each successive cycle to a value approaching the failure stress of
the composite.
A total of six unidirectional composite specimens loaded in the
ﬁber direction (axial specimens) were investigated in this study
allowing for one cyclic test for each heat treatment and testing
temperature. Also investigated were nine specimens loaded trans
verse to the ﬁber direction (transverse specimens). The transverse
tests consisted of one monotonic test to failure for each heat
treatment at room temperature and one cyclic test for each heat
treatment and testing temperature. Due to the small number of
specimens tested, this study only qualitatively addresses the inﬂu
ence of heat treatment on the material parameters (e.g., elastic
modulus) of the composite. Since limited material did not allow for
multiple tests of the same heat treatment, possible variations of the
material parameters due to material heterogeneity cannot be ad
dressed in this study. However, all specimens were taken from the
same composite plate, thus, the manufacturing process was iden
tical for each specimen. Although the present experiments cannot
identify material parameters without a large margin of error, the
degradation of material properties reported here should be indic
ative of damage induced changes in similarly produced Ti153
systems.
Microstructural evaluations were performed after each test to
determine the state of damage in the composite. The specimens
were carefully sectioned, polished and etched for a detailed mi
crostructural evaluation with light microscopy and scanning elec
tron microscopy.
Fig. 1 Microphotograph of a tested asreceived specimen polished and
etched to reveal the microstructure
Fig. 2 Microphotograph of a tested 450°C/24 hour heat treatment spec
imen polished and etched to reveal the microstructure
Fig. 3 Microphotograph of a tested 700°C/24 hour heat treatment spec
imen polished and etched to reveal the microstructure
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 75
2.3 Axial Tension Tests. Three cyclic tests, one for each
heat treatment, were conducted at room temperature and at 427°C
for axial specimens. For brevity, only the results of the unloading
elastic modulus as a function of the applied stress are shown in Fig.
4. As seen in Fig. 4, the 450°C specimen has the highest elastic
unloading modulus for both room temperature and elevated tem
perature tests. This was expected since the 450°C heat treatment
creates the highest room temperature modulus and yield point in
the matrix due to the high concentration of HCP ␣phase present
throughout the matrix. The 700°C specimen tested at elevated
temperature has the lowest initial elastic modulus of the three
treatments, although the asreceived specimens have similar val
ues. These specimens have similar microstructures with the pri
mary difference being the size of the precipitant as seen in Figs. 1
and 3.
A reduction in elastic modulus is seen throughout the entire test
for each specimen as the applied stress increases, indicating that
damage is present in all specimens. However, the 450°C specimen
tested at room temperature settles to a constant unloading modulus
above 1500 MPa. The inelastic deformation in the axial direction
has been shown to be primarily dominated by matrix plasticity,
rather than damage (Majumdar and Newaz, 1992a), although the
results seen in Fig. 4 show that damage is present. Microstructural
evaluation of the specimens reveal both ﬁber cracks and matrix
cracks in room and elevated temperature tests, however, a uniform
spacing of cracks was not found. Matrix cracks were seen to travel
across ﬁbers, without causing ﬁber failure, although, on failed
specimens the ﬁber/matrix interface was debonded. From the cat
astrophic failure seen and heard when the composite failed, it can
be assumed that the ﬁbers fail almost simultaneously. Regardless
of the heat treatment and testing temperature, all specimens failed
in the range of the ﬁber strain limit of 0.9–1.0%.
2.4 Transverse Tension Tests. Three cyclic tests, one for
each heat treatment condition, were conducted at room tempera
ture and at 427°C. Also, a monotonic tension test to failure was
performed at room temperature for each heat treatment to identify
the regions of inelastic deformation to be studied in the cyclic tests.
For brevity, only the results of the unloading elastic modulus are
presented in Fig. 5. The asreceived specimen tested at room
temperature and the 700°C specimens tested at both room and
elevated temperature showed inelastic deformation with no per
manent strain, thus, the inelastic deformation is attributed only to
damage. The 450°C heat treatment specimen tested at room tem
perature and the asreceived specimen tested at elevated tempera
ture showed inelastic deformation with a permanent strain. How
ever, these tests both contained a knee, or stiffening, in the
unloading curve, most likely due to the closing of the cracks along
the debonded ﬁber/matrix interface. For both of these tests, the
composite would unload to the origin without showing a perma
nent offset if the knee were not present. The 450°C heat treatment
tested at elevated temperature showed inelastic deformation with a
permanent strain, however, the knee is not observed in the unload
ing cycles. This implies that the permanent strain for this specimen
is due to matrix plasticity, in contrast to all other transverse tests
mentioned here.
The asreceived specimen at room temperature, as seen in Fig.
5, showed the highest initial elastic modulus, but damage reduced
the modulus as higher load levels were reached. The 700°C spec
imen had the highest elevated temperature modulus, but also
showed the greatest loss of stiffness, while the asreceived speci
men at elevated temperature and the 700°C specimen at room
temperature showed a slow rate of damage growth with the applied
load. This point could be explained by the accumulation of micro
cracks reaching a saturation point at which the modulus no longer
decreases. The 700°C and 450°C specimens tested at elevated
temperature and the asreceived and 450°C specimen tested at
room temperature show a continuous drop in the unloading mod
ulus, while the remaining tests settle to a nearly constant value.
The initial elastic modulus for the transverse tensile tests closely
match published results for similar systems, however, the inelastic
deformation described above occurs at stress levels well below the
elastic limit seen in similar systems (about 275 MPa for room
temperature SCS6/Ti153, Mujumdar and Newaz, 1992a; Lerch
and Saltsman, 1991). The inelastic deformation is facilitated by the
manufacturing ﬂaws seen in the material, resulting in failure for as
received, 450°C and 700°C specimens at 242 MPa, 274 MPa and
244 MPa respectively at room temperature. Observations of such
a low failure stress similar to those presented here have not been
reported in literature, but microstructural evaluations suggest a
Fig. 6 Example of representative tested unetched transverse speci
mens showing cracks emanating from manufacturing ﬂaw and propagat
ing toward ﬁber
Fig. 4 Unloading axial elastic modulus versus applied stress for axial
specimens
Fig. 5 Unloading transverse elastic modulus versus applied stress for
transverse specimens
76 / Vol. 122, JANUARY 2000 Transactions of the ASME
correlation between the low failure stress and the preexisting
manufacturing defects in the specimens tested in this work.
Microstructural evaluation shows that damage for all of the
specimens primarily occurs in the form of matrix cracking ema
nating from the ﬁber/matrix interface in the direction of the load
ing. Figs. 1, 2, 3, and 6 show examples of crack development in the
transverse specimens. Figure 6 shows crack initiation from man
ufacturing ﬂaws, while Figs. 1, 2, and 3 show the crack develop
ment along grain boundaries revealed through etching. Grain
boundaries are prone to crack development in these specimens due
to the brittle ␣phase accumulation. From the etched microstruc
tures seen in Figs. 1, 2, and 3 the highest level of ␣phase grain
boundary accumulation is seen in the 700°C heat treatment spec
imens and the lowest accumulation is seen in the 450°C heat
treatment specimens. The premature failure of the specimens and
lack of large permanent deformation, leads to the conclusion that
crack development along grain boundaries, initiated by pre
existing manufacturing ﬂaws, led to damage development and
ultimate failure of the specimens at stress levels well below the
composite elastic limit.
3 Thermomechanical Modeling of Damage in SiC/Ti
153 MMC
3.1 Model Development. In this section an attempt will be
made to correlate the experimental observations with theoretical
modeling. The actual deformation mechanisms are complex and
involve damage evolving during the loading. An approximate
averaging micromechanics method will be used to model the
inﬂuence of the loading direction and material properties on the
mechanical response of the composite. The model will be based on
the extension of the MoriTanaka micromechanics averaging
method (Mori and Tanaka, 1973) to include damage effects.
Averaging methods have been developed which model damage
in ﬁbrous composites. In the present study, an attempt will be
made to incorporate evolving damage in an incremental formula
tion. For the present implementation of the incremental formula
tion, the increment of total strain, ⌬⑀
t
, for each increment of
applied overall stress, ⌬ , is expressed as the sum of the elastic
increment, and strain increment due to damage, i.e.,
⌬⑀
t
ϭ ͓M
d

ϩ⌬
Ϫ M
d

͔ ϩ M
d

ϩ⌬
⌬ , (1)
where the ﬁrst term calculates the strain increment introduced by
damage through the change in the elastic damage compliance M
d
,
and the second term is the elastic increment using the current value
of the compliance.
The MoriTanaka method (Weng, 1984; Benveniste, 1987) uti
lizes the Eshelby equivalence tensor, S, to calculate the elastic
stress concentration factor, B
␣
e
, which relates the overall applied
stress, , to the stress in the ␣ phase by
␣
ϭ B
␣
e
, ␣ ϭ f, m (2)
where ␣ indicates the ﬁber or matrix phase. For a twophase
composite, a computationally convenient stress concentration fac
tor based on the MoriTanaka approximation is given by (Gavazzi
and Lagoudas, 1990)
B
f
e
ϭ ͓I ϩ c
m
L
m
e
͑I Ϫ S͒͑M
m
e
Ϫ M
m
e
͔͒
Ϫ1
(3)
where L
␣
e
and M
␣
e
are the elastic stiffness and compliance tensors
respectively, c
␣
is the volume fraction for each phase and I is
identity matrix. The effective elastic compliance for the MMC can
then be found using the following relation (Hill, 1965),
M
e
ϭ M
m
ϩ c
f
͑M
f
Ϫ M
m
͒B
f
e
. (4)
Substituting the elastic modulus for each heat treatment into the
above equations yields an evaluation of effective elastic compli
ance for each heat treatment of the composite.
As the loading is increased, matrix cracks develop in the uni
Table 1 Material properties for each heat treatment and testing temperature utilized in modeling (Lerch et al., 1990a; Johnson
et al., 1990; Wolfenden et al., 1996)
Matrix Properties Fiber Properties
Room temperature Elevated temperature, 427°C Room temperature and elevated temperature
AsRcvd 450°C 700°C AsRcvd 450°C 700°C
E, GPa 92.4 104.8 90.0 78.0 83.0 77.5 395.1
.32 .32 .32 .32 .32 .32 .19
Fig. 7 A comparison between the simulated stress/strain response and
experimental measurements for (a) 700°C specimen tested at elevated
temperature and (b) 450°C specimen tested at room temperature
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 77
directional composite laminate and the above evaluation of the
elastic compliance must be modiﬁed to account for the developing
damage. The effect of cracks on the overall compliance of the
composite is modeled using the methods derived by Laws and
Dvorak (1987), and Dvorak et al. (1985) which relate the com
posite compliance to the density of cracks in the composite.
Motivated by experimental observations, two different types of
cracks are modeled, slitcracks in the transverse specimens and
pennyshaped matrix cracks in axial specimens.
The transverse crack density, 
tr
, a function of the overall
applied stress, is deﬁned as the number of cracks of width, a, in a
square crosssection of material with sides of length a. The dam
aged composite compliance, M
d
, is calculated as a function of the
crack density 
tr
. The MoriTanaka method is used to ﬁrst gen
erate the undamaged effective composite compliance, which is
then revised as a function of the crack density (Dvorak et al.,
1985).
In the axial specimens, cracks are modeled as a collection of
pennyshaped matrix cracks transverse to the ﬁbers. The crack
density, 
a
, is a function of applied load and is deﬁned as the
average number of cracks of speciﬁc diameter, d, in a cube with
sides of length, d. The effective compliance of the composite is
calculated in two steps. First, the matrix is regarded as a volume of
homogeneous material with similar cracks of density 
a
, for which
an effective compliance, M
m
d
, is calculated as a function of the
crack density (Dvorak et al., 1985). The second step utilizes the
MoriTanaka method to generate an effective compliance of the
damaged composite, M
d
, using the properties of the ﬁber and the
damaged matrix, M
m
d
.
3.2 Implementation and Comparison With Experiments.
Implementation of the averaging micromechanics approach to
predict the mechanical response for damage requires knowledge of
the evolution of the crack density, , as a function of applied load.
A damage growth model relating the crack density to the applied
load is not utilized in this paper due to the inherent complexities
those models contain. In the absence of a damage growth criterion,
an experimental measurement will be utilized to determine the
contribution of damage to the total inelastic deformation during a
given loading path. Speciﬁcally, the damage will be determined
from the experimentally measured components of the elastic un
loading stiffness, L
d
(3, 3), shown in Fig. 4 and L
d
(2, 2) shown in
Fig. 5, where L
d
ϭ [M
d
]
Ϫ1
.
Using the experimental data from Figs. 4 and 5, the overall
stress vs. strain response can be simulated accounting for damage
evolution utilizing the method described above. Simulations were
made for all specimens, however, for brevity only two transverse
specimens are presented. A comparison between the simulated
stress/strain response and experimental measurements are shown
in Fig. 7 for the 700°C specimen tested at elevated temperature and
the 450°C specimen tested at room temperature. In the simulation
of the 700°C specimen at elevated temperature, Fig. 7(a), the
inelastic deformation is only due to damage, as evidenced by the
lack of permanent deformation. For the 450°C specimen tested at
room temperature, the simulation shows that the knee in the
unloading curve is the cause of the permanent deformation, sug
gested by the simulation matching the initial unloading modulus
above the knee and fully unloading with no permanent deforma
tion, Fig. 7(b).
A prediction of crack density as a function of applied load can
be determined by solving for the transverse crack density, 
tr
, and
the axial crack density, 
a
, that yields the experimentally mea
sured component of the unloading elastic stiffness. Since uniaxial
stress states are investigated, only the components of compliance
in the loading direction will be calculated. The model prediction of
the crack density as a function of the applied stress is shown in Fig.
8 using the material properties listed in Table 1. To verify the
model predictions for the resulting crack density as a function of
applied load, posttest microstructural evaluations were made. The
average crack densities measured from the specimens are com
pared to the predicted crack densities at the maximum applied
stress shown in Fig. 8. Table 2 contains the measured and pre
dicted values of the crack densities for each heat treatment and test
temperature for the transverse composite.
Table 2 Comparison of experimental and predicted crack densities, 
tr
, after ﬁnal loading for transverse specimens
Room temperature Elevated temperature, 427°C
AsRcvd 450°C 700°C AsRcvd 450°C 700°C
Experiment .33 .35 .22 .23 .39 .28
Model Prediction .50 .69 .38 .35 .76 .92
Fig. 8 Prediction of crack density as a function of applied stress for (a)
axial crack density, 
a
and (b) transverse crack density, 
tr
78 / Vol. 122, JANUARY 2000 Transactions of the ASME
In transverse specimens, the experimental crack density was
measured from a representative microphotograph of each sample,
and calculated as the number of cracks of average measured crack
width, a, divided by the number of squares with sides of length, a,
in a square area encompassing the entire crosssection of the
specimen. The crack width was taken to be the extent of the matrix
crack often developing adjacent to a ﬁber, and perpendicular to the
loading direction, as seen in Fig. 6. Note that the matrix cracks,
which propagate along the grain boundary in the direction of the
load, are not considered in the measured transverse crack density.
The predicted crack densities for the transverse specimens, shown
in Table 2, are higher than the experimentally measured values
especially for the elevated temperature tests. The preexisting
manufacturing defects in the composite may cause a larger exper
imentally measured stiffness reduction, thus leading to an over
prediction of the crack density. In addition, matrix cracks bridging
the ﬁbers in the loading direction, not accounted in the measured
crack densities, may play a role in the higher stiffness reduction
leading to an overprediction of crack density. When the transverse
matrix crack is coupled with the cracks developing along grain
boundaries, the result is a higher stiffness reduction for the ele
vated temperature tests. Caution should be ﬁnally drawn to the fact
that the model assumes nonintersecting matrix slit cracks, while
the observed damage indicates crack coalescence in many of the
examined specimens.
4 Conclusions
A SiC/Ti153 MMC was studied to determine the effects of
different heat treatments on the damage evolution at both room
temperature and at 427°C. The thermomechanical study revealed
that the heat treatments could affect the overall composite com
pliance and damage accumulation. Mechanical tests show that the
450°C heat treatment creates a microstructure with a consistently
high elastic modulus and high damage tolerance for all loading
conditions and temperatures.
Microstructural evaluation identiﬁed the primary damage modes
for both the transverse and axial specimens. The axial specimens
showed evidence of cracks developing perpendicular to the load
ing direction starting from the ﬁber/matrix interface. The trans
verse specimens showed cracks emanating from areas of poor
consolidation resulting in cracks propagating in the loading direc
tion along grain boundaries. A model based on the MoriTanaka
method was developed and implemented which incorporates dam
age in an incremental formulation. The developed method success
fully simulates the stress vs. strain response of the composite
accounting for damage. A prediction of the crack density was
made as a function of overall applied load. It was shown that the
predicted crack density at the ﬁnal load level overpredicts the
crack densities measured from a posttest microstructural analysis
for the transverse specimens, most likely due to the preexisting
damage along the ﬁber/matrix interface.
5 Acknowledgments
The authors acknowledge the support of AFOSR grant No.
F496209410341.
References
Allen, D. H., Eggleston, M. R., and Hurtado, L. D., 1994, “Recent Research on
Damage Development in SiC/Ti Continuous Fiber Metal Matrix Composites,” Frac
ture of Composites, E. A. Armanios, ed., Key Engineering Materials series, Trans
Tech Publications.
Benveniste, Y., 1987, “A New Approach to the Application of the MoriTanaka’s
Theory in Composite Material,” Mech. Materials, Vol. 6, pp. 147–157.
Dvorak, G. J., Laws, N., and Hejazi, M., 1985, “Analysis of Progressive Matrix
Cracking in Composite Laminates I. Thermoelastic Properties of a Ply with Cracks,”
J. Composite Mat., Vol. 19, pp. 216–234.
Gavazzi, A. C., and Lagoudas, D. C., 1990, “Incremental Elastoplastic Behavior of
Metal Matrix Composites Based on Averaging Schemes,” Proceedings IUTAM Sym
posium on Inelastic Deformation of Composite Materials, G. J. Dvorak, ed., Springer
Verlag, pp. 465–485.
Hill, R., 1965, “Theory of Mechanical Properties of FibreStrengthened Materials:
III. SelfConsistent Model,” J. Mechanics and Physics of Solids, Vol. 13, pp. 189–198.
Johnson, W. S., 1992, NASA Technical Memorandum 107597, Lewis Research
Center.
Johnson, W. S., S. J. Lubowinski, and A. L. Highsmith, 1990, “Mechanical
Characterization of Unnotched SCS6/Ti153 Metal Matrix Composites at Room
Temperature,” Thermal and Mechanical Behavior of Metal Matrix and Ceramic
Matrix Composites, ASTM STP 1080, J. M. Kennedy et al., eds., ASTM, Philadel
phia, PA, pp. 193–218.
Lagoudas, D. C., X. Ma, D. A. Miller, and D. H. Allen, 1995, “Modeling of
Oxidation in Metal Matrix Composites,” Int. J. Engng. Sci., Vol. 33, No. 15, pp.
2327–2343.
Laws, N., and Dvorak, G. J., 1987, “The Effect of Fiber Breaks and Aligned
PennyShaped Cracks on the Stiffness and Energy Release Rates in Unidirectional
Composites,” Int. J. Solids and Structures, Vol. 23, No. 9, pp. 1269–1283.
Lerch, B. A., Gabb, T. P., and MacKay, R. A., 1990a, “Heat Treatment Study of
the SiC/Ti153 Composite System,” NASA TP 2970.
Lerch, B. A., and Saltsman, J. F., 1991, “Tensile Deformation Damage in SiC
Reinforced Ti15V3Cr3Al3Sn,” NASA Technical Memorandum 103620, Lewis
Research Center.
Lissenden, Herakovich, C. T., and Pindera, M. J., 1995, “Response of SiC/Ti under
Combined Loading part I: Theory and Experiment for Imperfect Bonding,” Journal
of Composite Materials, Vol. 29, No. 2, pp. 130–155.
Majumdar, B. S., and G. M. Newaz, 1992a, 1992b, 1993, NASA Contractor Report
189095, 189096, 191181, Lewis Research Center.
Mori, T., and K. Tanaka, 1973, “Average Stress in Matrix and Average Elastic
Energy of Materials with Misﬁtting Inclusions,” Acta Metallurgica, Vol. 21, pp.
571–574.
Newaz, G. M., B. S. Majumdar, and F. W. Brust, 1992, “Thermal Cycling
Response of QuasiIsotropic Metal Matrix Composites,” ASME JOURNAL OF ENGI
NEERING MATERIALS AND TECHNOLOGY, Vol. 114, pp. 156–161.
Okada, M., and Banergee, D., 1984, “Tensile Properties of Ti15V3Al3Cr3Sn
Alloy,” Titanium Science and Technology: Proceedings of the Fifth International
Conference on Titanium, Munich, Germany, Luetjering ed., Deutsch Gesellschaft fur
Metallkunde, Pub., pp. 1835–1842.
Sun, C. T., J. L. Chen, G. T. Sha, and W. E. Koop, 1990, “Mechanical Charac
terization of SCS6/Ti64 Metal Matrix Composite,” J. Composite Mat., Vol. 24, pp.
1029–1059.
Weng, G. J., 1984, “Some Elastic Properties of Reinforced Solids, with Special
Reference to Isotropic ones Containing Spherical Inclusions,” Int. J. Ingrg. Sci., Vol.
22, pp. 845–856.
Wolfenden, A., K. D. Hall, and B. A. Lerch, 1996, “The effect of heat treatment on
Young’s modulus, damping, and microhardness of SiC/Ti153,” J. of Materials
Science, Vol. 31, pp. 1489–1493.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 79
Toru Ikeda
Chemical Engineering Group,
Department of Materials Process Engineering,
Graduate School of Engineering,
Kyushu University,
6101, Hakozaki, Higashiku,
Fukuoka, 8128581, Japan
Akira Yamashita
Boiler Plant Division,
Kawasahi Heavy Industry Ltd.,
2111, Minamisuna, Kotoku,
Tokyo 1360076, Japan
Deokbo Lee
Noriyuki Miyazaki
Chemical Engineering Group,
Department of Materials Process Engineering,
Graduate School of Engineering,
Kyushu University,
Fukuoka, 8128581, Japan
Failure of a Ductile Adhesive
Layer Constrained by Hard
Adherends
The evaluation of a fracture from a thin layer constrained by a hard material is important
in relation to the structural integrity of adhesive joints and composite materials. It has
been reported that the fracture toughness of a crack in a ductile adhesive joint depends
on the bond thickness, but the mechanism has not yet been elucidated clearly. In this study,
the Jintegral and the neartip stress of a crack in an adhesive joint are investigated. It
is determined that a decrease of the bond thickness increases the stress ahead of a crack
tip, which results in the decrease of fracture toughness.
1 Introduction
The estimation of fracture behavior of a crack in an adhesive
joint or a thin ductile layer constrained by hard adherends is
important from the viewpoint of the structural integrity of com
posite materials and adhesive structures. Bond thickness is one of
the important design parameters for adhesive structures because of
the dependence of the fracture toughness on the bond thickness.
Several experimental studies of the effect of bond thickness on the
fracture toughness have been performed. For example, Gardon
(1963) determined that the peeling force, which is proportional to
the fracture energy, decreases with the decrease of bond thickness.
Mostovoy and Ripling (1971) reported the same phenomenon in a
tapered double cantilever beam adhesive joint specimen (TDCB)
with an epoxy adhesive. Bascom et al. (1975, 1976) investigated
the TDCB for a rubber toughened epoxy adhesive and found that
the fracture energy is maximized when the bond thickness is equal
to the diameter of the plastic zone size 2r
p
for plane strain. r
p
is the
famous ﬁrst order estimate of the plastic zone size proposed by
Irwin,
Ά
r
p
ϭ
1
2
E
a
G
0
2
Plane Stress
r
p
ϭ
1
6
E
a
G
0
2
͑1 Ϫ
2
͒
Plane Strain
, (1)
where G is the energy release rate of a crack in an adhesive layer,
E
a
Young’s modulus,
a
Poisson’s ratio and
0
the yield stress, all
of them of the adhesive. Kinloch and Shaw (1981) performed the
same experiment as Bascom et al., and veriﬁed the same depen
dence of the fracture energy on the bond thickness, as shown in
Fig. 1. They considered that the fracture energy of the adhesive
joint is attributed to the volume of a plastic zone, as also illustrated
in Fig. 1. Their explanation is as follows. At ﬁrst, the volume of the
plastic zone and the fracture energy G
Ic
are the same as those of a
crack in a bulk adhesive when the bond thickness is much larger
than 2r
p
(t
c
ӷ 2r
p
). Then the plastic zone develops and G
Ic
increases with the bond thickness approaching to 2r
p
due to the
constraint of the adherend (t
b
Ͼ 2r
p
). The plastic zone reaches the
maximum volume and G
Ic
gives a maximum when the bond
thickness is equal to 2r
p
(t
m
ϭ 2r
p
). For t
a
Ͻ 2r
p
, the decrease
of the bond thickness causes restriction in the development of the
plastic zone, and G
Ic
decreases. This qualitative explanation is
plausible. It is, however, questionable that the fracture toughness
directly relates to the volume of the plastic deformation zone
around a crack tip because the plastic deformation zone is not
equal to the process zone. Furthermore, the variation of the plastic
deformation zone with the bond thickness has not been veriﬁed by
any numerical analysis or any experimental observations.
The effect of conﬁnement by adherends on the fracture tough
ness should be argued in relation to a stress ﬁeld around a crack tip.
Varias et al. (1991) investigated stress ﬁelds around a crack tip in
metal foil constrained by ceramic adherends. They determined the
increase of the near tip stress with the decrease of the foil thick
ness. Hsia et al. (1994) studied the effect of conﬁnement by
ceramic adherends on the dislocation around a crack tip in ex
tremely thin metal foil. Tvergaard and Hutchinson (1996) simu
lated the crack extension of a crack in constrained metal foil.
However, the target of these researches is a crack in metal foil
between ceramic adherends. The difference in Young’s moduli
between ceramics and metal is about 5 times, which is much
smaller than that between metal and plastic adhesives. The plastic
strain around a crack tip is dominant around a crack tip within
metal foil; however, the elastic strain still in the vicinity of a crack
in a plastic adhesive can not be ignored. The process zone around
a crack tip in a plastic adhesive is much larger than that in the
metal form. Therefore, it is necessary to investigate the stress ﬁeld
and a plastic zone around a crack tip within an adhesive layer
conﬁned by hard adherends. We investigated the elasticplastic
stress ﬁeld around a crack tip within a constrained adhesive layer,
using a combination of the ﬁnite element method (FEM) and the
boundary element method (BEM) (Ikeda et al., 1995). Daghyani et
al. (1995) conducted fracture tests of compact tension (CT) adhe
sive joints with several bond thicknesses and observed a transition
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
August 5, 1998; revised manuscript received June 22, 1999. Associate Technical
Editor: G. Ravichandran.
80 / Vol. 122, JANUARY 2000 Transactions of the ASME Copyright © 2000 by ASME
from cohesive fracture in thick bonds to interface fracture in thin
bonds. They also performed a large deformation elasticplastic
analysis for investigating stress near the crack tip. In this study, we
examined the bond thickness effect on the Jintegral of the ECP
and the TDCB ﬁrst. Then we performed the detailed large defor
mation elasticplastic analyses of stresses ahead of crack tips in
adhesive joints.
2 Bond Thickness Effect on the JIntegral
Many experiments (Mostovoy and Ripling, 1971; Bascom et al.,
1975; Bascom and Cottington, 1976; Kinloch and Shaw, 1981)
used approximate equations obtained by the beam theory to cal
culate the fracture toughness, in which adhesive layers are ignored.
However, it is possible that the elastic deformation of an adhesive
region affects the estimation of the fracture toughness due to the
small elastic constant. The plastic deformation of the adhesive may
not be ignored. We compared the energy release rate obtained by
the approximate equation, that of the linear elastic analysis and the
Jintegral obtained by the elasticplastic analysis. Both the elastic
and the elasticplastic analyses are assumed to be plane strain. The
energy release rate obtained by the approximate equation and that
by the numerical analysis are denoted by G
app.
and G, respectively,
hereafter distinguished each from the other.
2.1 Method for Jintegral Analysis. The Jintegrals of sev
eral bond thicknesses were obtained for two types of specimens,
the edge cracked plate adhesive joint specimen (ECP) and the
TDCB. The combination method of the FEM and the BEM was
used for the analyses (Ikeda et al., 1998). The adherend and
adhesive regions were assumed to be a linear elastic material and
an elasticplastic material and modeled by the BEM and the FEM,
respectively. Both regions are connected along the interface using
the conjugate gradient method (Mori, 1986). The algorithm of this
combination method is based on the traditional incremental strain
theory. This combination method has the advantage of saving CPU
time and memory storage size over the ﬁnite element method. The
contour integral method was utilized to obtain the Jintegral. The
numerical integration of the contour integral was performed accu
rately by the adaptive automatic integration technique (Miyazaki et
al., 1993). In this method, a Jintegral path can be located both in
the FEM and the BEM regions, and we can obtain the accurate
Jintegrals using integral paths located far from a crack tip. The
elastic energy release rate G was also obtained in the same way by
assuming both adhesive and adherend regions to be linear elastic
materials.
2.2 JIntegral of the ECP. The geometry of the ECP under
uniform tension is illustrated in Fig. 2. The adherend and adhesive
were assumed to be mild steel and a rubber toughened epoxy,
respectively. Combined meshes comprising about 2000 eightnode
isoparametric ﬁnite elements and about 180 quadratic boundary
elements were used to model a symmetrical half of the ECP. The
adherend region was modeled by the BEM as an elastic material,
and the adhesive region was modeled by the FEM as an elastic
plastic material, whose stressstrain relation is given by the
RambergOsgood equation as follows.
⑀
⑀
0
ϭ
0
ϩ ␣
ͩ
0
ͪ
n
, (2)
where ⑀ and are strain and stress, respectively, E Young’s
modulus,
0
the yield stress, ␣ and n the constants. We separated
(2) into two parts, the elastic part and the elasticplastic part, for
the convenience of the numerical analysis.
Ά
⑀
⑀
0
ϭ
0
. . . ͑ Յ
0
͒
⑀
⑀
0
ϭ
0
ϩ ␣
ͩ
0
ͪ
n
. . . ͑ Ͼ
0
͒ ͑⑀
0
ϭ
0
/E͒
. (3)
We can avoid calculating plastic strain at integral points in ﬁnite
elements where the stress is lower than the yield stress at each
incremental step. Both strainstress relationships obtained by (2)
and (3) are almost identical. Although this separation causes a
jump of strain from ⑀
0
to (1 ϩ␣)⑀
0
at being
0
, the effect of this
jump on the stress distribution in the plastic deformation zone is
small. Material constants of each material are shown in Table 1.
These material constants are the same as those of Kinloch’s and
Shaw’s experiment (1981) except for the parameters ␣ and n for
the RambergOsgood equation. The parameters ␣ and n were
assumed to be the same as another rubber toughened epoxy whose
stressstrain curve we have measured using strain gauges and a
testing machine. The Jintegral analyses were performed with
Fig. 1 Concept of the bond thickness effect on the fracture toughness
of a crack in an adhesive layer
Fig. 2 Two types of adhesive specimens
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 81
several adhesive thicknesses. The elastic energy release rate G was
obtained by assuming that both mild steel and epoxy resin are
linear elastic materials. The energy release rate of the ECP without
the adhesive region for plane strain was approximated by (4) and
(5) (Murakami et al., 1987).
ͭ
K
I
ϭ
f
ͱa F͑ ͒, ϭ a/W͑ Յ 0.6, Ϯ0.5%͒
F͑ ͒ ϭ 1.12 Ϫ 0.231 ϩ 10.55
2
Ϫ 21.72
3
ϩ 30.39
4
(4)
G
adherends
Ӏ
K
I
2
͑1 Ϫ
s
2
͒
E
s
ϭ
͕
f
ͱa F͑ ͖͒
2
͑1 Ϫ
s
2
͒
E
s
(5)
To roughly estimate the contribution of an adhesive region to the
energy release rate, the steel adherends are assumed to be a rigid
body in comparison with the adhesive because of the large differ
ence in elastic constant. If we ignore the rotation of the adherends,
the adhesive layer can be modeled into a gripped layer between
rigid adherends under uniform displacement. The energy release
rate of a crack in an adhesive region, subjected to the uniform
displacement in plane strain shown in Fig. 3, is approximated by
the following equation.
G
adhesive
Ӏ
E
a
͑1 Ϫ
a
͒
͑1 ϩ
a
͒͑1 Ϫ 2
a
͒
⅐
0
2
H
(6)
where E
a
and
a
are Young’s modulus and Poisson’s ratio of the
adhesive, respectively. The
0
is the uniform displacement of the
rigid body, the H is half of the bond thickness t. If we assumed the
0
to be equal to the average displacement of the ligament part of
the ECP,
0
can be expected to be
0
Ӏ
HW
f
͑1 Ϫ
a
2
͒
͑W Ϫ a͒E
a
(7)
The total energy release rate of the ECP can be obtained by
combining (5) and (6),
G
app.
Ӏ G
adherends
ϩ G
adhesive
Ӏ
͕
f
ͱa F͑ ͖͒
2
͑1 Ϫ
s
2
͒
E
s
ϩ
͑1 Ϫ
a
2
͒͑1 Ϫ
a
͒
2
W
2
f
2
2͑1 Ϫ 2
a
͒͑W Ϫ a͒
2
E
a
t. (8)
(8) obviously shows a proportional relationship between the en
ergy release rate and the bond thickness.
Figure 4 shows the Jintegral and the G of the ECP with the
bond thickness. The Jintegral obtained by the elasticplastic anal
ysis corresponds well to the G obtained by the liner elastic anal
ysis, but they are not constant and increase with the bond thick
ness. Thus, in this case, the effect of plasticity in the adhesive can
be ignored, but the effect of bond thickness cannot. The reason for
the energy release rate’s dependence on the bond thickness is that
the total compliance is affected by that of the adhesive layer. The
approximated energy release rate G
app.
for the ECP calculated by
(8) is also shown in Fig. 4. (8) approximates the energy release rate
of the ECP relatively well, but it is still smaller than the energy
release rate calculated by numerical analyses. The two interfaces
between an adhesive layer and adherends of an actual ECP are not
parallel because of the rotation of the adherends, which is a
possible reason for the lower evaluation of the energy release rate
obtained by (8).
2.3 JIntegral of the TDCB. The next example is the
TDCB, shown in Fig. 2. The geometry of the TDCB is the same
as that of Kinloch’s and Shaw’s experiment (1981). All of the
material constants used in this analysis are the same as those of
the ECP, as shown in Table 1. The Jintegral and the elastic
energy release rate G were obtained in the same way as the
ECP. Combined meshes comprising about 2500 eightnode
isoparametric ﬁnite elements and about 250 quadratic boundary
elements were used to model a symmetrical half of the TDCB.
The elastic energy release rate of the TDCB for plane strain,
ignoring the adhesive region, can be approximated by the beam
theory as
G
app.
ϭ
4P
2
m͑1 Ϫ
s
͒
E
s
B
2
m ϭ 3a
2
/h
3
ϩ
1
h
͑mm
Ϫ1
͒
(9)
where P is a load, E
s
Young’s modulus of the adherend, B a width
of the specimen, h a height of the beam, a a crack length, and m
a constant (Kinloch and Shaw, 1981). The Jintegral, the elastic
energy release rates G and G
app.
, are shown in Fig. 4 as a function
of the bond thickness t. The Jintegral and the elastic energy
release rate G are almost equivalent. Furthermore, they are inde
pendent of the bond thickness and agree well with G
app.
over the
actual bond thickness. This tendency is very different from that of
the ECP. The effect of the adhesive region is negligible due to the
small amount of strain energy stored in the adhesive region com
pared with that stored in the beams; thus, the Jintegral of TDCB
can be approximated by (9).
Figure 5 shows a comparison of the ratio of stored strain energy
within an adhesive layer in the TDCB and in the ECP. The small
ratio of the energy release rate is stored in the adhesive layer in the
TDCB, whereas the relatively large ratio of it is stored in the
adhesive layer in the ECP, which depends on the bond thickness.
The strain energy released from the adhesive layer at a crack
extension is very limited for the TDCB, but considerable for the
ECP.
Table 1 Material constants of adhesive joint
Fig. 3 A crack in an adhesive region under uniform displacement
Fig. 4 Jintegral, G and G
app.
of the ECP and the TDCB as a function of
bond thickness
82 / Vol. 122, JANUARY 2000 Transactions of the ASME
3 Near Crack Tip Stress Analysis of a Crack in an
Adhesive Joint
3.1 Analysis of a Crack in Homogeneous Material. When
the material property is characterized by the RambergOsgood
equation shown in (1), the stress ﬁeld in the close vicinity of a
crack tip in a plastic zone can be approximated by the HRR ﬁeld
as follows (Rice, 1968; Hutchinson, 1968; Rice and Rosengren,
1968):
1
0
ͩ
rr
r
r
ͪ
ϭ
ͩ
J
␣⑀
0
0
I
n
r
ͪ
1/͑nϩ1͒
ͩ
˜
rr
˜
r
˜
r
˜
ͪ
(10)
where r and are polar coordinates centered at a crack tip, I
n
a
function of n, ˜
rr
, ˜
and ˜
r
functions of n and . The near tip
stress ﬁeld in a plastic zone of a metal can be approximated well
by the HRR ﬁeld. It is, however, not known if the HRR ﬁeld exists
around a crack tip in a plastic material such as epoxy. The stress
around a crack tip in a CT specimen made of a bulk rubber
toughened epoxy, shown in Table 1, was analyzed. The ﬁnite
deformation elasticplastic analysis by the updated Lagrange algo
rithm was performed by the MARC program, which is a commer
cial software of the ﬁnite element method. About 1800 fournode
isoparametric ﬁnite elements were used for modeling a symmet
rical half of the CT specimen. The crack is modeled as a slit,
whose width is 2 m for the ﬁnite deformation analysis. The ﬁnite
element mesh around each crack tip is common to the CT, the
ECP, and the TDCB mentioned in the following sections.
The normalized hoop stress along the xaxis near a crack tip
obtained from the analysis at J ϭ 1500 N/m is compared with the
HRR ﬁeld and the Kﬁeld in Fig. 6. It can be seen that the
analytical results are considerably below the HRR ﬁeld. The HRR
ﬁeld is much higher than the stress ﬁeld of the bulkCT specimen
in the region of 10–100 m distance from a crack tip. The stress
ahead of a crack tip in the bulkCT specimen corresponds with the
Kﬁeld rather than the HRRﬁeld. The HRR ﬁeld in the sufﬁciently
developed elasticplastic region must be lower than the Kﬁeld.
The HRR ﬁeld is derived by assuming that the 1st term of the right
side of (1) can be neglected well within the developed plastic
deformation zone (Hutchinson, 1968; Rice and Rosengren, 1968).
However, the elastic part of the strain cannot be ignored in the
plastic zone of a rubber toughened epoxy because of the high
elastic strain, and the development of the plastic zone is poor in
comparison with that in metal. If we assume the small strain theory
even in the extreme vicinity of a crack tip, the HRR ﬁeld may be
observed within a region less than 5 m distant from the crack tip.
However, this region has already been covered by the large defor
mation zone, and the adaptability of the homogeneous mechanics
is also doubtful because the diameter of the rubber particle is more
than 1 m. We judge the absence of the HRR ﬁeld around a crack
tip in a rubber toughened epoxy from this indirect evidence.
3.2 Analysis of a Crack in Adhesive Joints. The stresses
near a crack tip of the ECP and the TDCB with several bond
thicknesses were also analyzed in the same way as the bulk
adhesive CT specimen. The geometries of specimens and the
material properties are shown in Fig. 2 and Table 1, as well as in
the analyses in the previous chapter. Symmetrical halves of both
specimens were modeled by about 3000 fournode isoparametric
Fig. 7 An example of the ﬁnite element meshes. TDCB specimen with an
adhesive layer whose thickness is 0.2 mm.
Fig. 5 Ratio of the strain energy stored in adhesive (W
adhesive
) to that
stored in whole specimen (W
whole
) with bond thickness
Fig. 6 Comparison among the hoop stress distributions near a crack tip
of a CT specimen, the HRRﬁeld and the Kﬁeld at ؍ 0
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 83
ﬁnite elements. An example of the FEM mesh of the TDCB is
shown in Fig. 7. The distributions of the normalized hoop stresses
yy
/
0
and triaxial stresses
H
/
0
(
H
ϭ (
xx
ϩ
yy
ϩ
zz
)/3)
near a crack tip of both specimens on xaxis at J ϭ 1500 N/m are
shown in Figs. 8 and 9. The
yy
/
0
and
H
/
0
of the bulk CT
specimen are also plotted in these ﬁgures.
The hoop stress and triaxial stress distributions near a crack tip
are considered to be a function only of the Jintegral and the bond
thickness because these stress distributions both for the ECP and
the TDCB with the same bond thickness correspond with each
other. When the bond thickness is thicker than 2.8 mm, the stress
distribution near a crack tip is identical to that of a crack in the
bulk CT specimen. In the case of the bond thickness being thinner
than 0.5 mm, the distributions of both the hoop stress and the
triaxial stress near a crack tip are higher than those of the bulk CT
specimen in the region more than 10 m distant from a crack tip.
The length of a failure region observed around a crack tip in the
rubber modiﬁed epoxy is usually from 100 m to 1 mm (Lee et al.,
1998). The stress around the failure region is expected to have
direct inﬂuence on the fracture toughness. Figure 10 shows the
yy
and
H
normalized by those of the bulk CT specimen at a 50 m
distance from each crack tip as the function of the bond thickness
t. Both the
yy
and
H
increase sharply in the bond thickness less
than 0.5 mm. One of the major mechanisms of the failure of rubber
toughened epoxy is cavitations of rubber particles by the triaxial
stress (Chen and Jan, 1992; Lee et al., 1998). It is expected that the
higher triaxial stress around a crack tip in a thinner bond region
accelerates the failure process, which results in the decrease of the
fracture toughness. A similar mechanism of the decrease of the
fracture toughness was reported by Varias et al. (1991) for metal
foil constrained by hard adherends. They expected cavitation and
void growth in the thin ductile metal layer between hard ceramic
adherends.
The inﬂuence of bond thickness on the crack tip opening dis
placement (CTOD) was investigated for the ECP and the TDCB.
Deformed crack faces around crack tips in several bond thick
nesses of the ECP and the TDCB at J ϭ 1500 N/m are shown in
Fig. 11. As can be seen from this ﬁgure, the CTOD of the ECP
almost corresponds to that of a crack in the homogeneous adhe
sive. The CTOD of the TDCB is not shown in this paper because
it is almost identical with the case of the ECP. The strip yield
model is one of the estimation models of the CTOD for small scale
yielding conditions (Anderson, 1995). This model provides the
CTOD for plane strain,
CTOD ϭ
G
2
0
. (11)
In this case, one half of the CTOD obtained by (11) is about 5.5
m, which corresponds well with one half of the CTOD obtained
by the ﬁnite element analysis, as shown in Fig. 11. The CTOD of
the ﬁnite element analysis was determined by the displacement at
the intersection of a 90 degree vertex with the crack ﬂanks. This
result takes issue with other researchers. The CTOD obtained by
Varias et al. (1991) for a crack in metal foil bonded between two
ceramic substrates depends on metal foil thickness. We obtained a
different result from in our previous paper using coarser mesh than
that of our current paper, in which the CTOD decreases with the
decrease of bond thickness (Ikeda et al., 1995). Daghyani, Lin and
Mai also showed the decrease of the CTOD with the decrease of
the bond thickness (1995). However, the bond thickness depen
dence of the CTOD was almost removed by using ﬁner meshes in
our current study. The numerical analysis of the CTOD of a crack
in an adhesive layer between hard adherends is very sensitive to
the mesh size. We need more detailed research to make a ﬁnal
decision.
The distributions of the plastic deformation zone around crack tips
of several bond thicknesses of the TDCB at J ϭ 1500 N/m are
illustrated in Fig. 12. The areas of plastic zones around the crack tip
Fig. 8 Distributions of hoop stress near a crack tip on the xaxis of the
ECP and the TDCB for several bond thicknesses at J ؍ 1500 N/m
Fig. 9 Distributions of triaxial stress near a crack tip along the xaxis of
the ECP and the TDCB for several bond thicknesses at J ؍ 1500 N/m
Fig. 10 Variation of the hoop stress and the triaxial stress near a crack
tip along the xaxis at r ؍ 50 m as a function of bond thickness
Fig. 11 Variation of the crack tip opening displacement in the several
bond thicknesses of the ECP at J ؍ 1500 N/m
84 / Vol. 122, JANUARY 2000 Transactions of the ASME
in both the ECP and the TDCB are plotted with the bond thickness in
Fig. 13. The area of the plastic zone increases continuously with a
decrease in the bond thickness. There is obviously no direct correla
tion between the volume of the plastic zone and the fracture energy
because the increase of the volume of the plastic zone is not related to
the decrease of the fracture toughness, but to the increase. The fracture
toughness of adhesive joints can not be attributed to the volume of the
plastic zone around a crack tip.
4 Concluding Remarks
First, we investigated the inﬂuence of the bond region on the
Jintegral of two different types of cracked adhesive joints, the
ECP and the TDCB. The Jintegral of the ECP increases propor
tionally with the bond thickness, but that of the TDCB is indepen
dent of the bond thickness and corresponds with the approximate
equation which derives from the beam theory. The reason why the
Jintegral of the TDCB is independent of the bond thickness is that
almost all of the strain energy is not stored in the adhesive region
but in the adherends.
Then, we analyzed precisely the stress ﬁelds near a crack tip of
both the ECP and the TDCB. The following are the results ob
tained from the analyses. The volume of the plastic deformation
zone around a crack tip constrained by hard adherends can not be
related to the fracture toughness directly. The stress distributions
near a crack tip of the ECP and the TDCB agree well with each
other in the case of the same bond thickness. Therefore, the stress
distribution near a crack tip is expected not to depend on the shape
of the specimens, but only on the Jintegral and the bond thick
ness. The CTOD of a crack in an adhesive joint keeps a unique
relationship with the Jintegral.
The distribution of the triaxial stress near a crack tip increases with
the decrease of the bond thickness in the region more than 10 m
distant from a crack tip. It is expected that higher triaxial stress ahead
of a crack tip in a thinner adhesive layer than 0.5 mm accelerates the
failure process and causes smaller fracture toughness. However, we
need other reasons for the drastic decrease of the fracture toughness in
a very thin adhesive layer (Ͻ0.5 mm). For example, the transition
from the cohesive fracture to the interface fracture at a thin adhesive
layer is one of the possible reasons.
References
Anderson, T. L., 1995, Fracture Mechanics, CRC Press, pp. 117–181.
Bascom, W. D., Cottington, R. L., Jones, R. L., and Peyser, P., 1975, “The Fracture
of Epoxy and ElastomerModiﬁed Epoxy Polymers in Bulk and as Adhesives,”
Journal of Applied Polymer Science, Vol. 19, pp. 2545–2562.
Bascom, W. D., and Cottington, R. L., 1976, “Effect of Temperature on the
Adhesive Fracture Behavior of an ElastomerEpoxy Resin,” Journal of Adhesion,
Vol. 7, pp. 333–346.
Chen, T. K., and Jan, Y. H., 1992, “Fracture Mechanism of Toughened Epoxy
Resin with Bimodal Rubberparticle Size Distribution,” Journal of Materials Science,
Vol. 27, pp. 111–121.
Daghyani, H. R., Ye, L., and Mai, Y. W., 1995, “Mode I Fracture Behaviour of
Adhesive Joints. Part I. Relationship between Fracture Energy and Bond Thickness,”
Journal of Adhesion, Vol. 53, pp. 149–162.
Daghyani, H. R., Ye, L., and Mai, Y. W., 1995, “Mode I Fracture Behaviour of
Adhesive Joints. Part II. Stress Analysis and Constraint Parameters,” Journal of
Adhesion, Vol. 53, pp. 163–172.
Gardon, J. L., 1963, “Peel Adhesion. I. Some Phenomenological Aspects of the
Test,” Journal of Applied Polymer Science, Vol. 7, pp. 625–641.
Hsia, K. J., Suo, Z., and Yang, W., 1994, “Cleavage duo to Dislocation Conﬁne
ment in Layered Materials,” Journal of the Mechanics and Physics of Solids, Vol. 42,
No. 6, pp. 877–896.
Hutchinson, J. W., 1968, “Singular Behavior at the End of a Tensile Crack in a
Hardening Material,” Journal of the Mechanics and Physics of Solids, Vol. 16, pp. 13–31.
Ikeda, T., Miyazaki, N., Yamashita, A., and Munakata, T., 1995, “ElasticPlastic
Analysis of Crack in Ductile Adhesive Joint,” Composites for the Pressure Vessel
Industry ASME PVP302, pp. 155–162.
Ikeda, T., Yamashita, A., and Miyazaki, N., 1998, “ElasticPlastic Analysis of
Crack in Adhesive Joint by Combination of Boundary Element and Finite Element
Method,” Computational Mechanics, Vol. 21, No. 6, pp. 533–539.
Kinloch, A. J., and Shaw, S. J., 1981, “The Fracture Resistance of a Toughened
Epoxy Adhesive,” Journal of Adhesion, Vol. 12, pp. 59–77.
Lee, D., Ikeda, T., Todo, M., Miyazaki, N., and Takahashi, K., 1999, “The
Mechanism of Damage around Crack Tip in RubberModiﬁed Epoxy Resin,” Trans
actions of JSME, Series A, Vol. 65, No. 631, pp. 439–446 (in Japanese).
Miyazaki, N., Ikeda, T., Soda, T., and Munakata, T., 1993, “Stress Intensity Factor
Analysis of Interface Crack Using Boundary Element Method; Application of
ContourIntegral Method,” Engineering Fracture Mechanics, Vol. 45, No. 5, pp.
599–610.
Mori, M., 1986, “FORT77; Numerical Programming,” Iwanami, pp. 90–114 (in
Japanese).
Mostovoy, S., and Ripling, E. J., 1971, “Effect of Joint Geometry on the Toughness
of Epoxy Adhesives,” Journal of Applied Polymer Science, Vol. 15, pp. 661–673.
Murakami, Y. (Editor in Chief), 1987, Stress Intensity Factors Handbook 1,
Pergamon Press, p. 9.
Rice, J. R., 1968, “A Path Independent Integral and the Approximate Analysis of
Strain Concentration by Notches and Cracks,” ASME Journal of Applied Mechanics,
Vol. 35, pp. 379–386.
Rice, J. R., and Rosengren, G. F., 1968, “Plane Strain Deformation Near a Crack
Tip in a PowerLaw Hardening Material,” Journal of the Mechanics and Physics of
Solids, Vol. 16, pp. 1–12.
Tvergaard, V., and Hutchinson, J., 1996, “On The Toughness of Ductile Adhesive
Joints,” Journal of the Mechanics and Physics of Solids, Vol. 44, No. 5, pp. 789–800.
Varias, A. G., Suo, Z., and Shih, C. F., 1991, “Ductile Failure of a Constrained
Metal Foil,” Journal of the Mechanics and Physics of Solids, Vol. 39, No. 7, pp.
963–986.
Fig. 12 Distributions of the plastic zone around crack tips in several
bond thickness regions of the TDCB at J ؍ 1500 N/m (
M
: von Mises’s
equivalent stress)
Fig. 13 Variation of the area of plastic deformation zone with the bond
thickness
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 85
T. C. Tszeng
1
Department of Mechanical, Materials and
Aerospace Engineering,
Illinois Institute of Technology,
Chicago, IL 60616
Interfacial Stresses and Void
Nucleation in Discontinuously
Reinforced Composites
This paper presents the theoretical predictions of the stress state at the inclusionmatrix
interface in discontinuous metal matrix composites by the generalized inclusion method.
In the author’s previous works, this method had been extended to the elastoplastic
deformation in the matrix material. The present analysis of the ellipsoidal inclusion
problem indicates that the regions at the pole and the equator of the particle/matrix
interface essentially remain elastic regardless of the level of deformation, although the
size of the elastic region keeps decreasing as deformation becomes larger. It was also
found that, when the composite is undergoing a relatively large plastic deformation
(strain), the maximum interfacial normal stress is approximately linearly dependent upon
the von Mises stress and the hydrostatic stress. Based on the stress criterion for void
nucleation, the author determined the void nucleation loci and nucleation strain for a
composite subjected to an axisymmetric macroscopic stress state. The inﬂuences of
interfacial bonding strength, inclusion shape, and volume fraction on the occurrence of
void nucleation have been determined. The interfacial bonding strength in a SiC
aluminum system was reevaluated by using existing experimental evidence.
Introduction
Stress state at the inclusion/matrix interface relative to the
bonding strength plays a very important role in determining the
mechanical strength of discontinuously (e.g., whisker, short ﬁber,
or particulate) reinforced metal matrix composites. As far as duc
tile fracture is concerned, a large number of studies have been
conducted which focused on the mechanism of void nucleation in
particlestrengthened alloys (Fisher and Gurland, 1981; Brown and
Stobbs 1971; Goods and Brown, 1979; Argon et al., 1975; Hunt et
al., 1989; Embury, 1985; LeRoy et al., 1981; Needleman, 1987;
Tszeng, 1993; Tvergaard, 1990). The previous studies have led to
the conclusion that void nucleation at the interface between the
inclusion and the matrix is one of the controlling phenomena in
ductile fracture in particlestrengthened alloys, especially in cases
where the volume fraction of the second phase particles is high.
There are two prevailing criteria for void nucleation, namely, the
energy criterion and the stress criterion (see, for example, Fisher
and Gurland, 1981). Typical SiC
p
reinforced composites usually
contain SiC particles of a size on the order of 10 m (Weng, et al.,
1992; Lloyd, 1989). According to most of the previous studies (see
Tszeng, 1993 for a brief review), the energy criterion for void
nucleation on a particle of this size is automatically satisﬁed. That
is, the occurrence of void nucleation is solely governed by the
stress criterion. This criterion states that voids will nucleate at the
interface if the tensile normal stress
n
at the inclusionmatrix
interface exceeds the bonding strength
I
of the interface.
While the present paper is mainly focused on interfacial failure,
one has to notice that other forms of failure may actually prevail in
a process, including particle fracture (Barnes, et al., 1995), void
formation in the matrix (Christman and Suresh, 1988), shear
failure in the matrix (Vasudevan et al., 1989). In addition, the
effects of temperature are very often signiﬁcant in determining the
form of failure (Barnes et al., 1995). With this reservation, this
study only intends to provide an investigation on the inﬂuence of
external/macroscopic load on the interfacial stress.
Among the theoretical studies of inclusionmatrix interfacial
stress surrounding the particle, Argon et al. (1975) and Thomson
and Hancock (1984) used a continuum mechanics based ﬁnite
element method (FEM) while Brown and Stobbs (1971) and Le
Roy et al. (1981) employed the theory of dislocations. The ap
proach based on FEM certainly gives a picture of the detailed
stress state in the vicinity of interface, but the choice of unit cell for
computation and the associated boundary conditions becomes
more difﬁcult with increasing volume fraction of second phase
particles. This is the case associated with discontinuously rein
forced metal matrix composites, which is the material considered
in the present study.
Any theoretical prediction of void nucleation from the inclusion/
matrix interface also requires a good estimate of the bonding
strength
I
at the interface. To the author’s knowledge, the only
data of the bonding strength in a SiCAl system was reported by
Flom and Arsenault (1986), who proposed a lower bound of 1690
MPa for the 6061 aluminum alloy reinforced by 1 vol% SiC
p
.
The objective of this paper is to provide a theoretical view of the
stress state at the interface between inclusion and matrix, and to
relate these theoretical predictions to the occurrence of void nu
cleation. The level of bonding strength of a SiCaluminum alloy
system will be reevaluated. The SiC/2124 aluminum alloy com
posite will be used as the model material and the inﬂuences of
hydrostatic pressure will be examined.
Theory
The reinforcements are represented by hard, elastic ellipsoidal
inclusions of an aspect ratio a embedded in a matrix material
which can deform elastoplastically. In a typical short ﬁber
reinforced composite, the orientation and aspect ratio varies from
ﬁber to ﬁber. The effects of variable aspect ratio have been
examined elsewhere (Tszeng, 1994b), and the present study as
sumed that all the ﬁbers have the same aspect ratio. The analysis
method is an extension of the Eshelby inclusion method (Eshelby,
1957) combined with the MoriTanaka mean ﬁeld theory (1973).
The full formulation can be found in Tszeng (1994a, b). In using
this line of approach, one has to know that the legitimacy of the
1
Formerly afﬁliated with Department of Mechanical Engineering, University of
NebraskaLincoln.
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division June
30, 1997; revised manuscript received June 21, 1999. Associate Technical Editor:
J. W. Ju.
86 / Vol. 122, JANUARY 2000 Transactions of the ASME Copyright © 2000 by ASME
method becomes suspect when the material is deformed at a
relatively large plastic strain, or in a system comprising high
volume fraction of reinforcement (Christensen, 1990). In addition,
the use of the secant method in the generalized inclusion method
for ﬁnite plasticity also invites concerns. To work within the
limitations, we will use the model only in a class of proportional,
monotonic loading problems.
In the following, we provide a brief account of the formulation
for calculating interfacial stress by the generalized inclusion
method (Tszeng, 1994a, b). The stress in the considered reinforce
ment,
F
, is calculated by (boldfaced variables denote tensors):
F
ϭ L
F
͑e
A
ϩ e
C
ϩ eЈ͒ Inhomogeneous composite (1)
ϭ L
M
͑e
A
ϩ e
C
ϩ eЈ Ϫ e
T
͒
Reference homogeneous composite (2)
where e
A
is the strain in homogeneous, unreinforced matrix metal,
e
C
is the constrained strain, eЈ is the additional term due to the
interaction between inclusions, e
T
is the incompatible transforma
tion strain, and L
F
and L
M
are the stiffness tensors of the rein
forcement and matrix material, respectively. In fact, L
M
is the
secant stiffness tensor of the matrix material at
M
and e
M
when it
goes into plastic deformation, i.e.,
M
ϭ L
M
e
M
. The secant
Young’s modulus and the corresponding Poisson’s ratio for plastic
incompressibility are, respectively,
1
E
S
ϭ
1
E
M
ϩ
e
p
Y
(3)
S
ϭ
1
2
Ϫ
ͩ
1
2
Ϫ
M
ͪ
E
S
E
M
(4)
where E
M
and
M
are, respectively, the Young’s modulus and
Poisson’s ratio of the matrix material, and e
p
and
Y
are, respec
tively, the effective plastic strain and the ﬂow stress of the matrix
material. In general, the ﬂow stress
Y
is dependent on the effec
tive plastic strain, e
p
. The von Mises yield criterion is used for the
matrix, and the components of the plastic strain in the matrix are
determined by the ﬂow rule
de
P
ϭ
3
2
de
p
Y
s (5)
where s is the deviatoric stress of the average matrix stress
M
.
The reinforcement stress in Eqs. (1) or (2) is the stress in a single
particle of particular orientation in the matrix. For a composite
comprising reinforcement of different orientation, each individual
reinforcement would experience a different stress state depending
on its individual orientation. A volume average procedure is
needed for composites having reinforcements with different ori
entations. The formulation given by Tszeng (1994a) shows the
average transformation strain ͗e
T
͘ is obtained by solving the fol
lowing equation:
͓͑1 Ϫ f ͒L
M
͑͗S͘ Ϫ I͒ Ϫ L
F
͑͑1 Ϫ f ͒͗S͘ ϩ f I͔͒͗e
T
͘
ϭ ͑L
F
Ϫ L
M
͒͑I ϩ L
͒e
A
. (6)
The volume average is deﬁned in Tszeng (1994a). In Eq. (6), f is
the volume fraction of the reinforcement, S is the fourthorder
Eshelby tensor which depends only on the inclusion geometry
(aspect ratio) and the Poisson’s ratio of the matrix, I is the
fourthorder identity tensor and L
ϭ (L
M
)
Ϫ1
(L
R
Ϫ L
M
), L
R
is the
secant stiffness tensor of the matrix material at
A
and e
A
, i.e.,
A
ϭ L
R
e
A
. After the average transformation strain ͗e
T
͘ is ob
tained, the transformation strain, e
T
, associated with the considered
inclusion can be obtained (Tszeng, 1994a).
The stress state at a point immediately outside the inclusion
surface is obtained by modifying the original formulation of
Eshelby (1957). At point P in Fig. 1, the unit normal n is
n ϭ ͑cos cos , sin cos , sin ͒. (7)
The angle is related to by
tan ϭ
1
a
2
tan
(8)
where a is the aspect ratio. The strain at the interface is:
e
͑Int͒
ϭ e
C͑Int͒
ϩ e
M
, (9)
where e
M
ϭ ͗e
T
͘ ϩ eЈ, is the average matrix strain. The hydro
static and deviatoric part of the interfacial constrained strain e
C(Int)
are respectively given by, in indicial form (Eshelby, 1957):
e
o
C͑Int͒
ϭ e
o
C
Ϫ
1
3
1 ϩ
S
1 Ϫ
S
e
o
T
Ϫ
1 Ϫ 2
S
1 Ϫ
S
Јe
ij
T
n
i
n
j
, (10)
and
Јe
il
C͑Int͒
ϭ Јe
il
C
ϩ
1
1 Ϫ
S
Јe
jk
T
n
j
n
k
n
i
n
l
Ϫ Јe
ik
T
n
k
n
l
Ϫ Јe
lk
T
n
k
n
i
ϩ
1 Ϫ 2
S
3͑1 Ϫ
S
͒
Јe
jk
T
n
j
n
k
␦
il
Ϫ
1
3
1 ϩ
S
1 Ϫ
S
e
o
T
ͩ
n
i
n
l
Ϫ
1
3
␦
ilͪ
. (11)
where e
T
is the stress free transformation strain associated with the
considered inclusion. In these expressions, a variable primed at the
left top corner represents the deviatoric part, and a subscript “o”
represents the hydrostatic part. Note that
S
in above equations is
the secant Poisson’s ratio of the matrix material deﬁned in Eq. (4).
The corresponding interfacial stress
(Int)
is obtained through sim
ple calculation, i.e.,
(Int)
ϭ L
(Int)
e
(Int)
, where L
(Int)
is the secant
modulus at e
(Int)
. Since a material point at the “interface” is actually
located immediately outside the inclusion surface, the mechanical
properties of the interface are assumed to be the same as those of
matrix material. For convenience, the interfacial stress is trans
formed to the local coordinate which has the 3axis coincident with
the surface outer normal n (Fig. 1).
The present paper is focused on the occurrence of void nucle
ation based on the stress criterion (Tszeng, 1993), no void growth
or coalescence will be discussed. More precisely, we are investi
gating the limiting conditions for void nucleation, instead of the
postnucleation behavior. The latter subject has been studied by
several investigators in the past (e.g., Needleman, 1987).
Fig. 1 The coordinate system and conventions for ellipsoidal inclusion.
At a surface point P which makes angles (, ) with axes 3 and 1, the
outer normal from the surface has an angle with the horizontal plane.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 87
Results and Discussion
In the following, the mechanical properties are based on SiC
reinforced 2124 aluminum alloys which were recently studied by
several investigators (Christman et al., 1989; Tvergaard, 1990;
Tszeng, 1994a, b). This composite system is chosen simply as a
model material, the general observations to be derived from the
theoretical calculations are rather generic and therefore should be
applicable to other composite systems as well. The Young’s mod
ulus and Poisson’s ratio are E
M
ϭ 60 GPa and
M
ϭ 0.3 for the
matrix and E
F
ϭ 342.6 GPa and
F
ϭ 0.21 for the inclusion. The
ﬂow stress,
Y
, of the matrix material is given by the modiﬁed
Ludwik relation:
Y
ϭ
o
ϩ h(e
p
)
n
, with
o
ϭ 300 MPa, h ϭ
442.1 MPa, and n ϭ 0.478, which is very close to what was used
by Tvergaard (1990). The macroscopic stress is axisymmetric with
11
A
ϭ
22
A
ϭ
33
A
, where is the proportional factor.
Interfacial Stresses on Spherical Particles. To elucidate the
calculations, we ﬁrst consider a simple composite comprising
ellipsoidal reinforcement of an aspect ratio a ϭ 1 (i.e., spherical
particulate), volume fraction f ϭ 0.1, and loaded by a uniaxial
stress
33
A
ϭ 1 MPa in the principal (long) direction (i.e., the
3axis). The distribution of interfacial stress is plotted as a function
of the angle from the pole (see Fig. 1) in Fig. 2. The stresses
shown in Fig. 2 actually represent the speciﬁc stress (stress divided
by the macroscopic stress
33
A
) in the elastic range of the composite
deformation. As shown in Fig. 2, the normal stress (along direction
3Ј in Fig. 1) shows a maximum tension,
nmax
, at the pole ( ϭ 0).
Figure 2 also shows the the von Mises effective stress
(ϭ
͌
1
2
((
11
Ϫ
22
)
2
ϩ (
22
Ϫ
33
)
2
ϩ (
11
Ϫ
33
)
2
)
1/ 2
, where
11
,
22
,
33
are the principal stress in the local coordinate system).
In contrast, the von Mises effective stress (normalized to the
overall uniaxial stress) has a peak value of about 1.5 at ϭ45 deg
from the pole. For a rigid, spherical inclusion in an elastic matrix,
Goodier (1933) found that the maximum tensile stress during
uniaxial loading is
nmax
ϭ 2
33
A
. By increasing the Young’s
modulus of the particle material and decreasing the particle vol
ume fraction, we found essentially the same results as that of
Goodier’s, see Table 1.
In the present analysis, there is no separate yield criterion for the
interface, yielding is allowed to occur only in the matrix which, by
deﬁnition, includes the interface. It is of interest to examine the
possibility of yielding at the interface. We apply the von Mises
yield criterion on the data presented in Fig. 2. According to Fig. 2,
the speciﬁc von Mises effective stress has a peak value of about 1.5
at ϭ 45 deg from the pole. Since the initial yield stress is 300
MPa, yielding at the interface ﬁrst occurs at a macroscopic stress
of about 200 MPa (ϭ300 MPa/1.5) in the region around ϭ 45
deg for a ϭ 1, and spreads along the interface as the macroscopic
stress becomes higher. The macroscopic stress is
33
A
ϭ 324 MPa
when initial yielding occurs globally in the matrix as based on the
average matrix stress
M
. At that moment, the interfacial von
Mises effective stress at ϳ 0 deg actually just exceeds the yield
stress, whereas the interface away from the pole (e.g., Ͼ 70 deg)
remains elastic. As the macroscopic stress continues increasing,
Fig. 3 illustrates the stress distributions at different levels of
deformation. Several interesting phenomena can be seen in Fig. 3.
First, the peak tensile normal stress always appears at the pole, and
increases with the deformation. In this respect, Argon et al. (1975)
determined that the normal stress in the polar region seemed to
have a ﬂat distribution in the case of a cylindrical inclusion,
whereas both Orr and Brown (1974) (on cylindrical inclusion) and
Thomson and Hancock (1984) (on spherical inclusion) indicated
that the maximum normal stress moves away from the pole as the
deformation becomes larger. Although the inclusions studied by
the works of Orr and Brown (1974) and Thomson and Hancock
(1984) had a very low volume fraction (Ͻ1%), we have found
essentially the same pattern of distribution as of Figs. 2 and 3 for
composites with different level of particle volume fraction. Thus,
difference in volume fraction is not the cause of the discrepancy.
Further study is needed to clarify the cause of the discrepancy.
Let the maximum normal stress be expressed in the form after
Thomson and Hancock (1984):
nmax
ϭ C ϩ
m
(12)
where C is the concentration factor, and and
m
are the macro
scopic von Mises effective stress and hydrostatic stress (ϭ(
m1
ϩ
m2
ϩ
m1
)/3), respectively. Note that the original form of Eq.
(12) proposed by Argon et al. (1975) actually assumed C ϭ 1. A
plot of the concentration constant C versus the effective plastic
strain is given in Fig. 4 (the curve with ϭ 0). The concentration
factor is increasing with deformation, as observed by Thomson and
Hancock (1984). However, qualitative differences do exist be
tween our calculation of C and that of Thomson and Hancock.
First, the value of C becomes more level at a larger composite
strain, compared with that of Thomson and Hancock which in
creases almost linearly with the strain. Second, our calculations
Fig. 2 The distribution of interfacial speciﬁc stresses (interfacial stress
divided by
33
A
) as a function of the angle from the pole in a composite
deformed elastically during uniaxial loading. The aspect ratio of the
inclusion a ؍ 1 and 5, volume fraction f ؍ 0.1.
Table 1 Maximum interfacial tensile stress as a function of
volume fraction and ratio of Young’s moduli
Volume
fraction
Ratio of Young’s moduli (inclusion to matrix)
1 10 100
0.1 1.559 1.759 1.782
0.01 1.653 1.906 1.935
0.001 1.663 1.922 1.952
0.0001 1.664 1.923 1.954
Fig. 3 The distribution of normal stress and von Mises stress at the
particlematrix interface in a composite deformed plastically during uni
axial loading at a range of overall uniaxial stress
33
A
. The aspect ratio of
the inclusion a ؍ 1, volume fraction f ؍ 0.1.
88 / Vol. 122, JANUARY 2000 Transactions of the ASME
(not shown) indicate that, in the elastic range, the value of C
remains constant whereas that of Thomson and Hancock shows a
rapid decrease followed by a sudden increase in the transition from
elastic to plastic deformation.
It is desirable to know whether Eq. (12) holds for a more general
state of macroscopic stress. In this respect, we consider an axi
symmetric, proportional macroscopic stress with
11
ϭ
22
ϭ
33
,
where remains constant during the loading history. By changing
the value of , the role of hydrostatic stress can be examined. The
case with ϭ 0 corresponds to the uniaxial loading (
11
ϭ
22
ϭ
0) and a more positive corresponds to a more tensile hydrostatic
stress. For the considered macroscopic stress with Ϫ1 Յ Յ1, the
interfacial normal stress always has a peak at the pole of the
spherical particles. The maximum interfacial normal stress at a
range of effective plastic strain is calculated, and the correspond
ing concentration factor C is plotted in Fig. 4. The concentration
factor, C, increases very rapidly at small plastic strains, and
becomes more level at larger deformation. Note that the curves in
Fig. 4 do not collapse into one single curve, the value of C is
therefore dependent on the type of macroscopic stress state. It is,
however, noticed that the values of C for different ’s seem to be
very close to each other at a larger strain. This means that the
simple form of Eq. (12) is applicable at a relatively large plastic
strain for a more general state of macroscopic stress.
There are two other ways to look at the empirical Eq. (12). First,
at a speciﬁed macroscopic von Mises effective stress, Eq. (12)
implies that the maximum interfacial normal stress is linearly
dependent on the hydrostatic stress with a unity proportionality
constant. The relation according to our calculations is plotted in
Fig. 5 for three levels of von Mises effective stress. Indeed, a linear
relationship does exist between the maximum interfacial normal
stress and the hydrostatic stress. The slopes of these curves are
very close to unity, although it decreases slightly at a higher Mises
stress. Second, at a speciﬁed macroscopic hydrostatic stress, Eq.
(12) implies that the peak stress is linearly dependent on the von
Mises effective stress with a proportionality constant C. The
relation according to our calculations is plotted in Fig. 6 for three
levels of hydrostatic stress. Again, the relationship is approxi
mately linear, although the proportionality constant C becomes
slightly lower at a higher hydrostatic stress. According to the
above elaboration, it seems that the linearity as represented by Eq.
(12) does exist when either the Mises stress or the hydrostatic
stress hold constant.
According to Fig. 3, the von Mises effective stress at the
interface is increasing with the overall stress. However, distinctly
different phenomenon occurs at the pole and the equator. As
discussed earlier, the von Mises stress at the interface between
angles 0 and 70 deg from the pole already exceeds the initial yield
stress of the matrix material (300 MPa) when the composite starts
to yield globally. In the transition from elastic to plastic deforma
tion, the von Mises stress at the pole decreases and becomes lower
than the matrix’s yield stress. To further illustrate the stress state
in the polar region, the von Mises stress at the pole is plotted as a
function of the macroscopic effective stress, shown in Fig. 7. As
indicated, the von Mises stress at the pole does exceed the yielding
stress in the transition from elastic to plastic deformation, followed
by an asymptotic decrease. The interface in the polar region
therefore experiences unloading after the early, very low level of
plastic deformation. To facilitate discussion, let us use two angles
1
and
2
in the way that the plastic zone at the interface is deﬁned
by
1
Յ Յ
2
. That is, the region between 0 Յ Յ
1
is in elastic
deformation. The angles are plotted as functions of the effective
plastic strain in Fig. 8. The angle
1
is equal to 0 deg at the elastic
limit (meaning the pole is in plastic deformation) and increases
very rapidly right after the composite goes into plastic deforma
tion. After reaching a peak, the angle
1
then monotonically
decreases as the deformation becomes even larger. The FEM
calculations of Thomson and Hancock (1984) also indicated the
similar tendency of a very low level of plastic deformation at the
pole (see Figs. 9 and 14 of Thomson and Hancock, 1984). As
pointed out by Thomson and Hancock (1984), the observed yield
ing at the polar element in their study may be a consequence of the
Fig. 4 The concentration factor C as a function of the effective strain for
composites comprising spherical particles of volume fraction 0.1.
11
A
؍
22
A
؍
33
A
Fig. 5 The relationship between the maximum interfacial normal stress
and the hydrostatic stress at several different levels of von Mises stress
(labeled by the corresponding curves). All stresses are in MPa. The
volume fraction of spherical particle is 0.1.
Fig. 6 The relationship between the maximum interfacial normal stress
and the von Mises stress at several different levels of hydrostatic stress
(labeled by the corresponding curves). All stresses are in MPa. The
volume fraction of spherical particle is 0.1.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 89
ﬁnite dimensions of the ﬁnite element mesh, and therefore may not
represent the actual feature.
Similar phenomena also appear at the equator, where the von
Mises stress reaches its peak of 101 MPa (Ͻ
o
ϭ 300 MPa) at the
macroscopic elastic limit, and decreases monotonically after the
composite yields globally (see Fig. 7). The plastic zone at the
interface spreads toward the equator at larger deformation, as
indicated by the increasing
2
in Fig. 8. However, the material in
the region of the equator would not go into plastic deformation,
regardless of the level of macroscopic deformation.
Based on the above observations, we see that the plastic zone
ﬁrst appears at ϳ45 deg and spreads in both directions along the
interface; the pole ( ϭ 0 deg) and the equator ( ϭ 90 deg)
essentially remain elastic throughout the deformation. This con
clusion is generally in accord with the FEM calculations of Thom
son and Hancock (1984) (see Fig. 7 of that paper).
According to Table 1, a lower particle volume fraction leads to
a higher maximum interfacial normal stress,
nmax
at the same
macroscopic stress. However, the same statement does not stand
when the maximum interfacial normal stress is plotted as a func
tion of the macroscopic strain. This point will be reexamined later
in this paper.
Interfacial Bonding Strength. As mentioned earlier, the only
data of the bonding strength in an SiCAl composite system was
reported by Flom and Arsenault (1986), who suggested a lower
bound of 1690 MPa for the 6061 aluminum alloy reinforced by 1
vol% SiC
p
. In determining the actual maximum interfacial normal
stress, Flom and Arsenault used the original form of Eq. (12)
proposed by Argon et al. (1975), i.e., C ϭ 1, for the axisymmetric
tensile specimen with a semicircular, circumferential groove. Due
to the lack of exact material properties of the aluminum alloy
matrix considered by Flom and Arsenault (1986), we are not able
to use the present method for an estimate of the bonding strength.
Further, as shown in the Appendix, the bonding strength of 1690
MPa is not reproducible based on the data provided by Flom and
Arsenault (1986).
A recent work of Weng et al. (1992) on the microfracture of
SiC
p
reinforced 6061T6 aluminum alloy provides other possible
data of the bonding strength of the SiCaluminum system. For the
standard tensile specimen of composite comprising 10 vol% SiC
particles of mean diameter 10 m, matrix debonding starts appear
ing at a uniaxial stress of 253 MPa. By using Eq. (12) with C ϭ
1 and assuming the effective stress in the matrix when the particles
are absent is 253 MPa, one obtained the bonding strength as 337
MPa. Since, in general, the value of C in Eq. (12) is greater than
1, the actual interfacial normal stress, and therefore the bonding
strength, should be higher than 337 MPa. By using the present
method with matrix properties: E
M
ϭ 70 GPa,
M
ϭ 0.33,
o
ϭ
220 MPa, h ϭ 190 MPa, and n ϭ 0.56, the maximum interfacial
normal stress
nmax
at
33
A
ϭ 253 MPa was found to be 416 MPa
and the corresponding macroscopic strain is 0.00619 in the loading
direction. Thus, the bonding strength of the considered composite
system is about 416 MPa.
According to these data for the bonding strength of SiC
aluminum alloy system, the value of 1690 MPa suggested by Flom
and Arsenault (1986) seems to be too high. A plausible bonding
strength should range between 400 MPa and 900 MPa (see the
Appendix). However, it is understood that the bonding strength for
a given material system depends very much upon the processing
history. Further, void nucleation itself is a stochastic process
(Thomson and Hancock, 1984). Consequently, a statistical ap
proach may be needed for a better description of the bonding
strength. Similar studies on the particle cracking have been con
ducted in the past, e.g., Mochida, Taya and Lloyd (1991). It will be
a sizable effort to determine the statistical nature of the interfacial
bonding strength and to include this description in the theoretical
model. That effort is beyond the scope of the present paper.
Instead, two different levels of bonding strength will be used to
explore their inﬂuences on the void nucleation, as will be discussed
in the next section.
Void Nucleation. In the following calculations regarding void
nucleation for the considered SiC2124 aluminum alloy matrix
composite, two levels of interfacial bonding strength
I
are to be
used, namely, 500 and 1000 MPa. A bonding strength of 1000
MPa may not be achievable by the current technology, as will be
seen later. The macroscopic stress is axisymmetric with
11
A
ϭ
22
A
ϭ
33
A
. For a void to nucleate from the particle/matrix interface,
the maximum interfacial normal stress
nmax
which appears at the
pole has to exceed the bonding strength
I
of the interface. The
nucleation loci for both levels of bonding strength are calculated
and shown in Fig. 9 after LeRoy et al. (1981). The same ﬁgure also
shows the loci corresponding to the initial yielding and the hydro
static stress state, respectively. The bonding strength does have an
important inﬂuence on the occurrence of void nucleation. With the
higher bonding strength of 1000 MPa, the composite shows very
high resistance to void nucleation. Note that void nucleation in the
composite with a lower bonding strength occurs approximately at
the elastic limit during uniaxial loading (
11
A
ϭ
22
A
ϭ 0).
The nucleation strains in square root are plotted as a function of
the hydrostatic stress
m
in Fig. 10. A higher hydrostatic (tensile)
stress corresponds to a smaller nucleation strain. This ﬁgure is in
agreement with the widely accepted observation that a compres
sive hydrostatic stress can inhibit or delay the damage formation/
accumulation. Again, the important inﬂuence of bonding strength
is obvious. According to the present calculation, for a composite
comprising spherical particles of volume fraction 10% and a
bonding strength of 1000 MPa, the nucleation strain in uniaxial
loading is as high as 0.169. Obviously, this strain is even larger
than most of the elongation strain at rupture reported in the
literature. That is, a bonding strength of 1000 MPa may be unre
alistically high from the standpoint of current technology.
Fig. 7 The changing of interfacial von Mises at the pole and at the
equator as functions of the effective plastic strain
Fig. 8 The angles (
1
,
2
) deﬁning the plastic zone at the interface are
changing as the deformation proceeds
90 / Vol. 122, JANUARY 2000 Transactions of the ASME
Effects of Aspect Ratio and Volume Fraction. The curves of
speciﬁc interfacial stress for a composite comprising ellipsoidal
inclusions (ﬁbers) of an aspect ratio of 5 are also shown in Fig. 2.
Obviously, at the same macroscopic stress, inclusions of a higher
aspect ratio experience a high peak stress at the inclusionmatrix
interface. Thus, particulatereinforced composites are expected to
possess a better ductility compared with the whiskerreinforced
composites, which is actually what is generally observed (Rack
and Ratnaparkhi, 1988). To further explore this point, the macro
scopic uniaxial stress (i.e., ϭ 0) needed to bring the maximum
interfacial normal stress to the bonding strength is plotted as a
function of the ﬁber aspect ratio in Fig. 11. In these calculations,
the bonding strength is taken to be 1000 MPa. The same ﬁgure also
shows the effects of reinforcement volume fraction. As indicated
in this ﬁgure, the nucleation strain decreases very rapidly with the
aspect ratio. At an aspect ratio of 5, void nucleation actually takes
place in the elastic range. It is also noticed that the nucleation
strain approaches an asymptotic value at a large aspect ratio.
In Fig. 11, the volume fraction has a more important role in the
composite comprising inclusions of a smaller aspect ratio. In this
respect, a larger volume fraction leads to a lower nucleation strain.
This is in agreement with the general observation that the ductility
and fracture toughness of discontinuously reinforced metals is
usually inferior to that of unreinforced matrix metals, and that a
higher volume fraction results in a smaller elongation strain (duc
tility).
According to the curve corresponding to an aspect ratio of 5 in
Fig. 2, initial yielding at the interface may have started at the
location of peak Mises stress at a
33
A
as low as 110 MPa, which is
about onethird of the initial yield stress of the matrix aluminum
alloy. To ensure the composite is free from localized plastic
deformation, the macroscopic stress has to be below the elastic
limit which is about 110 MPa in the considered case. As mentioned
earlier, the corresponding limit stress in composites comprising
spherical particulate is 200 MPa. It is generally accepted that local
concentrations of plastic deformations should be avoided for a
better fatigue resistance (Dieter, 1976). Hence, a higher elastic
limit corresponds to better fatigue resistance, especially in the
cases involving a high level of stress. Since it has been shown that
the elastic limit is lower for composites comprising ﬁbers of a
larger aspect ratio, one may expect composites comprising ﬁbers
of a higher aspect ratio should possess a lower fatigue resistance.
This remark is actually contradictory to what is generally observed
(Rack and Ratnaparkhi, 1988). This is a topic subjected to further
study.
Conclusion
Several concluding remarks can be derived from our calcula
tions and discussion for composites subjected to an axisymmetric
macroscopic stress state.
1. When the composite is undergoing a relatively large plastic
deformation (strain), the maximum interfacial normal stress is
approximately linearly dependent upon the von Mises stress and
the hydrostatic stress (Eq. (12)). That is, the general form of Eq.
(12) is valid only at a relatively large plastic strain. However, the
linearity does exist when either the Mises stress or the hydrostatic
stress remains constant in a loading process.
2. The regions at the pole and the equator of the particle/
matrix interface essentially remain elastic regardless of the level of
deformation, although the size of the elastic region keeps decreas
ing as deformation becomes larger.
3. Interfacial bonding strength is the most inﬂuential single
factor as far as void nucleation is concerned. A plausible value of
bonding strength in SiCaluminum systems is in the range between
400 MPa and 900 MPa.
4. A higher level of reinforcement volume fraction leads to a
lower nucleation strain. Its effects becomes diminishing in com
posites comprising inclusions of a larger aspect ratio.
Acknowledgment
The author would like to thank the reviewers for their elaborated
comments on the original manuscript.
References
Argon, A. S., Im, J., and Safoglu, R., 1975, “Cavity Formation from Inclusions in
Ductile Fracture,” Metallurgical Transactions, Vol. 6A, pp. 825–837.
Fig. 9 A diagram showing the comparison of the calculated nucleation
loci for composites comprising spherical particles of volume fraction f ؍
0.1 subjected to axisymmetric deformation
11
A
؍
22
A
؍
33
A
. Two levels
of interfacial bonding strength are used.
Fig. 10 A plot of the square root of nucleation strain (
͌
⑀
N
) as a function
of macroscopic mean stress for composites comprising spherical parti
cles of volume fraction f ؍0.1 subjected to axisymmetric deformation
11
A
؍
22
A
؍
33
A
. Two levels of interfacial bonding strength are used.
Fig. 11 A plot of the nucleation strain (
͌
⑀
N
) for composites comprising
ellipsoidal inclusions of a range of aspect ratio and volume fraction
during uniaxial loading. The interfacial bonding strength is assumed to
be 1000 MPa.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 91
Barnes, S. J., Prangnell, P. B., Roberts, S. M., and Withers, P. J., 1995, “The
Inﬂuence of Temperature on Microstructural Damage During Uniaxial Compression
of Aluminium Matrix Composites,” Scripta Metallurgica et Materialia, Vol. 33, pp.
323–329.
Brown, L. M., and Stobbs, W. M., 1971, “The Workhardening of CopperSilica,
I. A Model Based on Internal Stresses, with no Plastic Relaxation; II. The Role of
Plastic Relaxation,” Phil. Mag., Vol. 23, pp. 1185–1233.
Christensen, R. M., 1990, “A Critical Evaluation for a Class of Micromechanics
Models,” J. Mech. Phys. Solids, Vol. 38, pp. 379–404.
Christman, T., and Suresh, S., 1988, “Effects of SiC Reinforcement and Aging
Treatment on Fatigue Crack Growth in an AlSiC Composite,” Mater. Sci. Engng.,
Vol. A102, pp. 211–216.
Christman, T., Needleman, A., and Suresh, S., 1989, “An Experimental and
Numerical Study of Deformation in MetalCeramic Composites,” Acta Metallurgica,
Vol. 37, pp. 3029–3050.
Dieter, G. E., 1976, Mechanical Metallurgy, 2nd edition, McGrawHill, New York.
Embury, J. D., 1985, “Plastic Flow in Dispersion Hardened Materials,” Metallur
gical Transactions, Vol. 16A, pp. 2191.
Eshelby, J. D., 1957, “The Determination of the Elastic Field of an Ellipsoidal
Inclusion, and Related Problems,” Proc. Royal Soc., (London), Vol. A241, pp.
376–396.
Fisher, J. R., and Gurland, J., 1981, “Void Nucleation in Spheroidized Carbon
Steels Part 2: Model,” Metal Science, pp. 193–202.
Flom, Y., and Arsenault, R. J., 1986, “Interfacial Bond Strength in an Aluminum
Alloy 6061SiC Composite,” Materials Science and Engineering, Vol. 77, pp. 191–
197.
Goodier, J. N., 1933, ASME Journal of Applied Mechanics, Vol. 55, pp. 39–44
(see Thomson and Hancock, 1984).
Goods, S. H., and Brown, L. M., 1979, “The Nucleation of Cavities by Plastic
Deformation,” Acta Metallurgica, Vol. 27, pp. 1–15.
Hancock, J. W., and Mackenzie, A. C., 1976, “On the Mechanisms of Ductile
Failure in HighStrength Steels Subjected to Multiaxial StressStates,” Journal of
Mechanics and Physics of Solids, Vol. 24, pp. 147–169.
Hunt Jr., W. H., Richmond, O., and Young, R. D., 1989, “Fracture Initiation in
Particle Hardened Materials with High Volume Fraction,” ICCM/VI, Vol. 2, pp.
2.209–223.
LeRoy, G., Embury, J. D., Edward, G., and Ashby, M. F., “A Model of Ductile
Fracture Based on the Nucleation and Growth of Voids,” Acta Metallurgica, Vol. 29,
pp. 1509–1522.
Lloyd, D. J., 1989, “The Solidiﬁcation Microstructure of Particulate Reinforced
Aluminium/SiC Composites,” Composites Science and Technology, Vol. 35, pp.
159–179.
Mochida, T., Taya, M., and Lloyd, D. J., 1991, “Fracture of Particles in a
Particle/Metal Matrix Composite under Plastic Straining and Its Effect on the
Young’s Modulud of the Composite,” Materials Transactions, JIM, Vol. 32, pp.
931–942.
Mori, T., and Tanaka, K., 1973, “Average Stress in Matrix and Average Elastic
Energy of Materials with Misﬁtting Inclusion,” Acta Metallurgica, Vol. 21, pp. 571.
Needleman, A., 1987, “A Continuum Model for Void Nucleation by Inclusion
Debonding,” ASME Journal of Applied Mechanics, Vol. 54, pp. 525–531.
Orr, J., and Brown, D. K., 1974, “ElastoPlastic Solution for a Cylindrical Inclusion
in Plane Strain,” Engineering Fracture Mechanics, Vol. 6, pp. 261–274.
Rack, H. J., and Ratnaparkhi, P., 1988, “Damage Tolerance in Discontinuously
Reinforced MetalMatrix Composites,” Journal of Metals, November, pp. 55–57.
Thomson, R. D., and Hancock, J. W., 1984, “Local Stress and Strain Fields near a
Spherical Elastic Inclusion in a Plastically Deforming Matrix,” International Journal
of Fracture, Vol. 24, pp. 209–228.
Tszeng, T. C., 1993, “A Model of Void Nucleation from Ellipsoidal Inclusions,”
Scripta Metallurgica et Materialia, Vol. 28, pp. 1065–1070.
Tszeng, T. C., 1994a, “Micromechanics of Partially Aligned Short Fiber Compos
ites with Reference to Deformation Processing,” Composites Science and Technology,
Vol. 51, pp. 75–84.
Tszeng, T. C., 1994b, “Micromechanics Characterization of Unidirectional Com
posites during Multiaxial Plastic Deformation,” J. Composite Materials, Vol. 28, pp.
800–820.
Tvergaard, V., 1990, “Analysis of Tensile Properties for a WhiskerReinforced
MetalMatrix Composite,” Acta Metallurgica et Materialia, Vol. 38, pp. 185–194.
Vasudevan, A. K., Richmond, O., Zok, F., and Embury, D., 1989, “The Inﬂuence
of Hydrostatic Pressure on the Ductility of AlSiC Composites,” Mater. Sci. Engng.,
Vol. A107, pp. 63–69.
Weng, B. J., Chang, S. T., and Shiau, J. S., 1992, “Microfracture Mechanisms of
SiC6061 Aluminum Composites after Hipping,” Scripte Metallurgica et Materialia,
Vol. 27, pp. 1127–1132.
A P P E N D I X
Bonding Strength According to the Experiments of Flom
and Arsenault (1986)
According to Flom and Arsenault (1986), the hydrostatic stress
reaches its maximum value at the bottom of the semicircular
groove, which is given by
m
ϭ
c
ͱ1 Ϫ ͑a/a ͒
2
Y
(A1)
where
Y
is the ﬂow stress, c ϭ 0.48 and a ϭ 2.3 mm, and 2a
is the current diameter of the ligament. However, the works of
Hancock and MacKenzie (1976) and Hancock and Brown (1983)
indicated that the hydrostatic stress in the transverse cross section
actually has a minimum value at the bottom of the groove and
reaches maximum at the center. According to Flom and Arsenault
(1986), particle debonding could barely be seen at the bottom of
the circumferential groove at an uniaxial load of 750 kgf on the
specimen. The corresponding ﬂow stress
Y
ϭ 750 ϫ 9.81/a
2
MPa, where a is not available from the paper of Flom and
Arsenault (1986). Since the actual dimensions of the specimen are
not available from the paper, a reverse procedure is to be used to
assess those missing data. In fact, by using the original formula for
a and c (Eqs. (3) and (4) of Flom and Arsenault, 1986) and using
a ranging from 1.85 through 2.25 (initial diameter at the grooved
area), we were not able to obtain the c ϭ 0.48 and a ϭ 2.3 mm
given by Flom and Arsenault (1986). According to our calculation,
(c, a ) ϭ (0.4826, 2.3) at a ϭ 2.2, and (c, a ) ϭ (0.48, 1.97)
at a ϭ 1.97. The corresponding maximum interfacial normal
stress
nmax
ϭ 1285 and 1692 MPa, respectively. Also, the corre
sponding ﬂow stress
Y
ϭ 484 and 670 MPa, respectively. Now,
according to Hancock and MacKenzie (1976), the hydrostatic
stress at the bottom of the groove is
1
3
Y
. Using Eq. (12) with C ϭ
1, we obtained the lower bound of the bonding strength as 645 and
893 MPa, respectively, corresponding to the ﬂow stress of 484 and
670 MPa. Thus, a plausible bonding strength should range between
400 MPa and 900 MPa.
92 / Vol. 122, JANUARY 2000 Transactions of the ASME
Marwan K. Khraisheh
Department of Mechanical Engineering,
King Fahd University of Petroleum and
Minerals (KFUPM),
Dhahran 31261, Saudi Arabia
email mkhraish@kfupm.edu.sa
An Investigation of Yield
Potentials In Superplastic
Deformation
Recent results (Khraisheh et al., 1995 and 1997) have indicated that superplastic mate
rials exhibit a strong degree of anisotropy and that the plastic ﬂow cannot be described
by the isotropic von Mises ﬂow rules. In this study, the yield potential for the model PbSn
superplastic alloy is constructed experimentally for different effective strain rates using
combined tension/torsion tests. A generalized anisotropic “dynamic” yield function is also
proposed to represent the experimentally constructed yield potentials. The anisotropic
function is not only capable of describing the initial anisotropic state of the yield potential,
it can also describe its evolution through the evolution of unit vectors deﬁning the
direction of anisotropy. The anisotropic yield function includes a set of material constants
which determine the degree of deviation of the yield potential from the isotropic von Mises
yield surface. It is shown that the anisotropic yield function successfully represents the
experimental yield potentials, especially in the superplastic region.
I Introduction
There has been a signiﬁcant amount of research done in the area
of superplasticity in recent years due to the potential beneﬁts in the
area of metal forming. Many parts used in the aerospace industry
are currently manufactured from Ti and Al superplastic alloys, and
recently, superplastic forming has been considered for use in the
automotive industry due to the many advantages over conventional
forming techniques (Hamilton, 1985; Ghosh, 1985; and Smith,
1998). Greater design ﬂexibility, lower die cost, and the ability to
form complex shapes from hard materials are some of beneﬁts that
superplastic forming offers. However, for the superplastic forming
process to become more acceptable as an efﬁcient metal forming
technique suitable for high production rate, there is a need for
numerical tools that can predict the deformation behavior of su
perplastic materials under multiaxial loading conditions, as en
countered in actual forming operations. In order to achieve that,
accurate constitutive relations for superplastic deformation must
be developed. One important aspect in constructing the constitu
tive relations is the selection of appropriate yield function, in the
case of rateindependent materials, or equivalently the yield po
tential (dynamic yield function) in the case of ratedependent
materials (e.g., superplastic materials).
Most of the work that has been directed toward studying the
mechanics of superplastic materials is based on uniaxial loading
conditions and assumes isotropic behavior (Holt, 1970; Hamilton
et al., 1991; Dutta and Mukherjee, 1992; and Carrino and Guiliano,
1997). Moreover, the von Mises yield criterion is usually em
ployed to extend the uniaxial models to multiaxial loading condi
tions through deﬁnitions of effective stress and strain. Until re
cently (Khraisheh et al., 1995 and Khraisheh et al., 1997), there
was no or little information reported on anisotropy in superplastic
deformation, which led to the use of the isotropic von Mises ﬂow
rules. The recent results indicate that the PbSn superplastic alloy
exhibits a strong degree of deformationinduced anisotropy. These
results included induced axial stresses in torsion and distortional
effects in tension/torsion tests. The source of the deformation
induced anisotropy is believed to be the severe grain shape distor
tion and grain rotation. For more details on the mechanism(s) of
deformation induced anisotropy, see Khraisheh et al. (1995) and
Mayo and Nix (1989). The results suggest that in order to accu
rately model the deformation behavior of superplastic materials
under general loading conditions, an appropriate anisotropic dy
namic yield function must be used in representing the yield po
tential.
For rateindependent materials subjected to multiaxial loading,
anisotropic yield surfaces are used to describe anisotropic effects
resulting from texture development. There is a considerable
amount of work done in this area, for example, Hill (1950), Barlat
(1987), Barlat and Fricke (1988), and Barlat and Lian (1989). For
ratedependent materials, however, the anisotropic dynamic yield
surface (yield potential) is the equivalent of the anisotropic yield
surface. Very little is known about the yield potentials in super
plastic materials. In this paper, the yield potential for the PbSn
superplastic alloy is constructed experimentally using combined
tension/torsion tests at different effective strain rates. In addition,
a general anisotropic dynamic yield function is used to accurately
represent the experimentally constructed yield potentials.
II Experimental
Combined tension/torsion tests were conducted on a modiﬁed
screwtype Instron machine capable of controlling the axial and
radial motions and especially equipped with data acquisition sys
tem. The tests were conducted on the Sn38.1% Pb superplastic
alloy, for which superplastic forming is possible at room temper
ature. The superplastic alloy was prepared in air by melting com
mercially pure Pb and Sn and casting into a mold. The casting was
then upset in one direction and rolled to a ﬁnal thickness resulting
in a thickness strain of Ϫ2.71. The resulting microstructure was
examined by scanning electron microscope and can be character
ized by equiaxed grains with an average grain size of ϳ5.8 m.
Cylindrical bar test specimens with gage length of 2.54 cm and a
diameter of 0.46 cm were then machined from the rolled sheet. For
more details on the experimental setup, material preparation, and
the mechanical properties of the alloy used here, see Khraisheh et
al. (1995), and Khraisheh et al. (1997).
For ratedependent materials and using the general associated
ﬂow rule, it is shown that for every effective strain rate,
˙
⑀ , there is
one value for the dynamic yield function (Dafalias, 1990 and
Khraisheh et al., 1997). Hence the yield potential in the stress
space can be constructed from experiments by maintaining the
effective strain rate constant for various loading paths. Therefore,
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
November 15, 1998; revised manuscript received May 14, 1999. Associate Technical
Editor: N. Chandra.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 93
Copyright © 2000 by ASME
combined tension/torsion tests at constant effective strain rates
were conducted to construct the yield potential. The effective
strain rate for simple tension, pure torsion, and combined tension/
torsion cases using von Mises deﬁnition are given by:
Simple Tension
⑀
˙
ϭ ⑀˙ (1)
Pure Torsion
⑀
˙
ϭ
␥˙
ͱ3
(2)
Combined TensionTorsion
⑀
˙
ϭ
ͱ
⑀˙
2
ϩ
␥˙
2
3
(3)
where ⑀˙ is the axial strain rate and ␥˙ is the radial strain rate.
Proportional loading tests with ⑀˙ ϭ k␥˙ were conducted, where k is
the proportionality constant. For a speciﬁc effective strain rate,
different combinations of the axial and radial strain rates where
chosen to obtain different points in the  plane. The axial strain
rate can be expressed in terms of the effective strain rate and k as:
⑀˙ ϭ ⑀
˙
ͱ
3k
2
1 ϩ 3k
2
(4)
Five data points were used to construct the yield surfaces. Each
point corresponds to a different value of k. One point corresponds
to a k value of ϱ (simple tension), another point corresponds to a
k value of 0 (pure torsion), and three points corresponds to ﬁnite
values of k (combined tension/torsion).
An important issue in constructing the yield surface is how to
deﬁne the yield point. For most metals, the initial yield point is
clear and the proportional limit or the 0.2% offset methods are
usually used to determine the yield point. For most superplastic
materials, however, there is no well deﬁned yield point and the
deﬁnition of the yield point (or ﬂow stress value) is controversial.
The problem arises because plastic deformation for superplastic
materials at the superplastic temperature takes place almost in
stantly and the value of the ﬂow stress changes signiﬁcantly at the
very early stages of deformation, where the yield point is usually
deﬁned (e.g., 0.2% strain). For the PbSn superplastic alloy used
here, this problem was solved by examining the stress/strain
curves. The ﬂow stress is almost constant and no signiﬁcant
hardening is observed. Therefore, the steady or the “saturated”
value of the ﬂow stress (axial and shear) was used in constructing
the yield surface. Hence the yield surface presented here represents
the collection of points in the stress space, at which superplastic
ﬂow occurs. As an illustration, the axial and shear stress compo
nents for a combined tension/torsion test at an effective strain rate
of 6.5 ϫ 10
Ϫ4
s
Ϫ1
and k ϭ 0.45 are shown in Figs. 1 and 2. It is
clear that the axial and shear components saturate approximately to
values of 16 MPa and 7.75 MPa, respectively. These values are
used to locate one point on the yield potential (Fig. 5).
Fig. 1 The axial stress component in a combined tension/torsion test.
Effective strain rate 6.5 ؋ 10
؊4
s
؊1
, k ؍ 0.45.
Fig. 2 The shear stress component in a combined tension/torsion test.
Effective strain rate 6.5 ؋ 10
؊4
s
؊1
, k ؍ 0.45.
Table 1 Results of experiments performed to construct the yield potential for an effective strain rate of 3 ؋ 10
؊4
s
؊1
Prop. constant (k ϭ ⑀˙ /␥˙ )
Axial strain rate,
⑀˙ (s
Ϫ1
)
Radial strain rate,
␥˙ (s
Ϫ1
)
Saturated axial
stress, MPa
Saturated shear
stress, MPa
0 (pure torsion) 0 5.20 ϫ 10
Ϫ4
2* (induced) 7.65
0.25 1.19 ϫ 10
Ϫ4
4.77 ϫ 10
Ϫ4
8.25 5.70
0.6 2.16 ϫ 10
Ϫ4
3.60 ϫ 10
Ϫ4
12.50 4.50
1.4 2.77 ϫ 10
Ϫ4
1.98 ϫ 10
Ϫ4
15.5 2.0
ϱ (pure tension) 3.00 ϫ 10
Ϫ4
0 20.6 0
* Induced axial stress during ﬁxedend torsion tests, for more details see Khraisheh et al. (1995).
Table 2 Results of experiments performed to construct the yield potential for an effective strain rate of 6.5 ؋ 10
؊4
s
؊1
Prop. constant (k ϭ ⑀˙ /␥˙ )
Axial strain rate,
⑀˙ (s
Ϫ1
)
Radial strain rate,
␥˙ (s
Ϫ1
)
Saturated axial
stress, MPa
Saturated shear
stress, MPa
0 (pure torsion) 0 1.13 ϫ 10
Ϫ3
1.75* (induced) 11.00
0.15 1.60 ϫ 10
Ϫ4
1.09 ϫ 10
Ϫ3
8.75 9.50
0.45 3.9 ϫ 10
Ϫ4
9.0 ϫ 10
Ϫ4
16.00 7.75
1.1 5.7 ϫ 10
Ϫ4
5.4 ϫ 10
Ϫ4
21.25 4.00
ϱ (pure tension) 6.5 ϫ 10
Ϫ4
0 23.50 0
* Induced axial stress during ﬁxedend torsion tests, for more details see Khraisheh et al. (1995).
94 / Vol. 122, JANUARY 2000 Transactions of the ASME
The experimental data used in constructing the yield potentials
for effective strain rates of 3 ϫ 10
Ϫ4
, 6.5 ϫ 10
Ϫ4
, and 1 ϫ 10
Ϫ3
s
Ϫ1
are summarized in Tables 1, 2, and 3, respectively. The corre
sponding yield potentials are plotted in Figs. 3, 4, and 5.
Fore superplastic materials which exhibit signiﬁcant hardening
(e.g., Al7474 superplastic alloys), a consistent meaningful proce
dure need to be employed to determine the yield point. A possible
way is to use the value of the stress at zero plastic strain. However,
this technique involves graphical extrapolation and could present
error and scatter in the results.
III Yield Potential Representation
The selection of the appropriate form for the dynamic yield
function is very important in constitutive modeling. First, the
isotropic von Mises yield function is considered and compared
with the experimental data. It has the form:
f ϭ ͱ3J
2
J
2
ϭ
1
2
S.S (5)
where f is the yield function and J
2
is the second invariant of the
deviatoric stress tensor, S. For combined tension/torsion condition,
the yield function is reduced to:
f ϭ ͱ
2
ϩ 3
2
(6)
is the axial stress and is the shear stress. Equation (6) is
modiﬁed such that the yield function reduces to the tensile yield
stress in simple tension ( f ϭ
y
) leading to:
y
2
ϭ
2
ϩ 3
2
(7)
Equation (7) represents the von Mises yield surface and is plotted
against the experimental data in Figs. 3, 4, and 5. The values for
y
were taken from Tables 1, 2, and 3 for simple tension case (k ϭ ϱ).
It is clear from the ﬁgures that the experimental data cannot be
represented by the isotropic von Mises yield function. This distor
tion of the yield potential can be attributed to the deformation
induced anisotropy that was observed for this alloy (Khraisheh et
al., 1995). Therefore, a general anisotropic dynamic yield function,
Fig. 3 The yield potential for the PbSn superplastic alloy. Effective
strain rate 3 ؋ 10
؊4
s
؊1
. Experimental data versus von Mises and the
anisotropic functions.
Table 3 Results of experiments performed to construct the yield potential for an effective strain rate of 1 ؋ 10
؊3
s
؊1
Prop. constant (k ϭ ⑀˙ /␥˙ )
Axial strain rate,
⑀˙ (s
Ϫ1
)
Radial strain rate,
␥˙ (s
Ϫ1
)
Saturated axial
stress, MPa
Saturated shear
stress, MPa
0 (pure torsion) 0 1.73 ϫ 10
Ϫ3
1.25* (induced) 14.2
0.2 3.25 ϫ 10
Ϫ4
1.625 ϫ 10
Ϫ3
12.5 11.4
0.5 6.25 ϫ 10
Ϫ4
1.3 ϫ 10
Ϫ3
20.0 7.75
1.1 8.9 ϫ 10
Ϫ4
8.0 ϫ 10
Ϫ4
25.0 4.5
ϱ (pure tension) 1.0 ϫ 10
Ϫ3
0 27.8 0
* Induced axial stress during ﬁxedend torsion tests, for more details see Khraisheh et al. (1995).
Fig. 4 The yield potential for the PbSn superplastic alloy. Effective
strain rate 6.5 ؋ 10
؊4
s
؊1
. Experimental data versus von Mises and the
anisotropic functions.
Fig. 5 The yield potential for the PbSn superplastic alloy. Effective
strain rate 1 ؋ 10
؊3
s
؊1
. Experimental data versus von Mises and the
anisotropic functions.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 95
in which anisotropy can evolve is now considered. The yield
function which is deﬁned in reference to the axis x
i
(i ϭ 1, 2, 3)
has the following form (Hill, 1950; Dafalias, 1990; and Khraisheh
and Zbib, 1997):
f
2
ϭ
3
2
S.S ϩ c
1
͑M.S͒
2
ϩ c
2
͑N
1
.S͒
2
ϩ c
3
͑N
2
.S͒
2
N
1
ϭ a
1
ᮏ a
1
N
2
ϭ a
2
ᮏ a
2
M ϭ
1
2
͓a
1
ᮏ a
2
ϩ a
2
ᮏ a
1
͔ (8)
where a
1
, a
2
, and a
3
are orthonormal vectors (a
3
ϭ a
1
ϫ a
2
)
along the axis of anisotropy xЈ
i
(i ϭ 1, 2, 3), c
1
, c
2
, and c
3
are
material parameters, and V is the diadic product. For c
1
ϭ c
2
ϭ
c
3
ϭ 0, the anisotropic yield function reduces to the isotropic von
Mises yield function (equation 5). The directions of a
1
and a
2
are
deﬁned by the angle between x
1
and a
1
measured positive coun
terclockwise (), as shown in Fig. 6. The direction tensors N
1
, N
2
,
and M can then be expressed in terms of as:
N
1
ϭ
ͫ
cos
2
cos sin 0
cos sin sin
2
0
0 0 0
ͬ
N
2
ϭ
ͫ
sin
2
Ϫcos sin 0
Ϫcos sin cos
2
0
0 0 0
ͬ
M ϭ
1
2
ͫ
Ϫ2 cos sin cos
2
Ϫ sin
2
0
cos
2
Ϫ sin
2
2 cos sin 0
0 0 0
ͬ
(9)
For combined tension/torsion case, Eq. (8) reduces to:
f
2
ϭ A
I
2
ϩ A
II
ϩ A
III
2
A
I
ϭ 1 ϩ
c
1
9
cos
2
sin
2
ϩ
c
2
9
͑2 cos
2
Ϫ sin
2
͒
2
ϩ
c
3
9
͑2 sin
2
Ϫ cos
2
͒
2
A
II
ϭ Ϫ
2
3
cos sin ͕c
1
͑cos
2
Ϫ sin
2
͒ Ϫ 2c
2
͑2 cos
2
Ϫ sin
2
͒ ϩ 2c
3
͑2 sin
2
Ϫ cos
2
͖͒
A
III
ϭ 3 ϩ c
1
͑cos
2
Ϫ sin
2
͒
2
ϩ 4c
2
cos
2
sin
2
ϩ 4c
3
cos
2
sin
2
(10)
The values of the material parameters (c
1
, c
2
, and c
3
) and the
angle determine the degree of deviation of the anisotropic yield
surface from the isotropic von Mises yield surface. Equation (10)
can be written in a nondimensional form such as:
1 ϭ A
Iͩ
f
ͪ
2
ϩ A
IIͩ
f
ͪͩ
f
ͪ
ϩ A
IIIͩ
f
ͪ
2
(11)
The nondimensional yield surface is plotted in Figs. 7, 8, 9, and
10 for different values of the angle , and the material parameters
c
1
, c
2
, and c
3
, respectively. The ﬁgures clearly show how the yield
surface is affected by each parameter. In general, the angle
Fig. 6 Schematic illustration of the axis of anisotropy and the angle
Fig. 7 The effect of the angle on the anisotropic yield surface for
tensiontorsion case (equation 11). c
1
؍1, c
2
؍5, and c
3
؍4. The dashed
line represents von Mises surface (c
1
؍ c
2
؍ c
3
؍ 0).
Fig. 8 The effect of c
1
on the anisotropic yield surface for tension
torsion case (Eq. (11)). ؍ 30°, c
2
؍ 5, and c
3
؍ 4. The dashed line
represents von Mises surface (c
1
؍ c
2
؍ c
3
؍ 0).
Fig. 9 The effect of c
2
on the anisotropic yield surface for tension
torsion case (Eq. (11)). ؍ 30°, c
1
؍ 1, and c
3
؍ 4. The dashed line
represents von Mises surface (c
1
؍ c
2
؍ c
3
؍ 0).
96 / Vol. 122, JANUARY 2000 Transactions of the ASME
determines the deviation of orientation of the anisotropic yield
surface from the von Mises isotropic one. The material parameters
c
1
, c
2
, and c
3
determine the magnitude or the intensity of the
deviation.
In order to compare the anisotropic yield surface with the experi
mental data in Figs. 3, 4, and 5, the condition f ϭ
y
is imposed, where
y
is the yield stress in simple tension. This leads to:
f
2
ϭ
1
A
I
͓A
I
2
ϩ A
II
ϩ A
III
2
͔ ϭ
y
2
(12)
Equation (12) is plotted against the experimental data and von
Mises in Figs. 3, 4, and 5.
IV Discussion and Conclusions
It is clear from Figs. 3, 4, and 5 that the yield surface (or ﬂow
curve) of the PbSn superplastic alloy is anisotropic and cannot be
represented by the isotropic Mises surface. This is especially true
for the low strain rate (3 ϫ 10
Ϫ4
s
Ϫ1
) which lies in the superplastic
region (region II of the log Ϫ log ⑀˙ curve). As the strain rate
increases, we move away from the superplastic region, and von
Mises surface presents better representation of the experimental
data. Hence, when modeling the deformation behavior of super
plastic materials in the superplastic region, it is inappropriate to
assume isotropic behavior and use the von Mises stress/strain
equivalents. The departure from isotropy explains why von Mises
based models failed to predict the deformation behavior of the
PbSn superplastic alloy under biaxial loading in bulge forming
(Khraisheh and Zbib, 1999).
The anisotropic yield function proposed in this study is not only
capable of describing the initial anisotropic nature of the yield
surface, it can also describe its evolution through the evolution of
the unit vectors, a
1
and a
2
(i.e., the evolution of the angle ),
which deﬁne the direction of the axis of anisotropy. The material
dependent parameters, c
1
, c
2
, and c
3
have signiﬁcant effects on the
shape of the yield surface. They deﬁne the intensity of the depar
ture from the isotropic yield surface.
After rigorous ﬁtting trials to the experimental data shown in
Figs. 3, 4, and 5, the values (c
1
ϭ 1, c
2
ϭ 5, c
3
ϭ 4, and ϭ
30°) present the best combination of the parameters which resulted
in a very good ﬁt to the experimental data for the three different
strain rates. Of course, a more accurate treatment can result if the
material parameters can be expressed in terms of the strain rate,
however, for simplicity, constant values are used here. The aniso
tropic yield function for these values are compared against the
experimental data and Von Mises in Figs. 3, 4, and 5. It is evident
that this anisotropic yield function represents the experimental data
very well, especially in the superplastic region (⑀˙ ϭ 3 ϫ 10
Ϫ4
s
Ϫ1
).
It is worth mentioning that only experimental data in the ﬁrst
quadrant is available, and these values of the parameters represent
the best ﬁt in the ﬁrst quadrant only. However, if results of other
loading directions are available (i.e., data in other quadrants), the
proposed yield function (with the range of parameters that it
includes) can easily ﬁt almost any experimental data.
In a previous study (Khraisheh et al., 1997), the isotropic von
Mises yield surface was used in constructing a constitutive model
for superplastic materials. Although the model was able to capture
a number of phenomena, including uniaxial behavior and stress
relaxation, however, it failed to predict the actual trend of the
deformationinduced anisotropy. Under preparation is a study in
which the anisotropic yield potential proposed here is employed in
the constitutive model developed earlier to predict the
deformationinduced anisotropy. The angle , which indicates the
deviation of the orientation of the anisotropic function, will play a
signiﬁcant role in predicting the induced anisotropy and its evo
lution.
In summary, the present study clearly showed that superplastic
deformation is anisotropic, as the experimentally constructed yield
potentials deviate considerably from the isotropic von Mises yield
surface. The proposed anisotropic dynamic yield function is capa
ble of representing the experimental results and also in capturing
the evolution of the inducedanisotropy.
Acknowledgment
The author wishes to acknowledge the continuous collaboration
and fruitful discussions with Professor Hussein Zbib of Washing
ton State University which motivated this research. The support of
KFUPM is also acknowledged.
References
Barlat, F., 1987, “Crystallographic Texture, Anisotropic Yield Surfaces and Form
ing Limits of Sheet Metals,” Material Science and Engineering, Vol. 91, pp. 55–72.
Barlat, F., and Fricke, W. G., Jr., 1988, “Prediction of Yield Surfaces, Forming
Limits and Necking Directions for Textured FCC Sheets,” Proceedings of the Eighth
International Conference on Textures of Materials, J. S. Kallend and G. Gottstein,
eds., The Metallurgical Society, pp. 1043–1050.
Barlat, F., and Lian, L., 1989, “Plastic Behavior and Strechability of Sheet Metals.
Part 1: A Yield Function for Orthotropic Sheets under Plane Stress Conditions,”
International Journal of Plasticity, Vol. 5, pp. 51–66.
Carrino, L., and Guiliano, G., 1997, “Modeling of Superplastic Blow Forming,”
International Journal of Mechanical Science, Vol. 39, No. 2, pp. 193–199.
Dafalias, Y. F., 1990, “The Plastic Spin in Viscoplasticity,” International Journal
of Solids Structures, Vol. 26, No. 2, pp. 149–163.
Dutta, A., and Mukherjee, A. K., 1992, “Superplastic Forming: An Analytical
Approach,” Material Science and Engineering, Vol. A157, pp. 9–13.
Ghosh, A. K., 1985, “Superplasticity in Aluminum Alloys,” Superplastic Forming,
S. P. Agrawal, ed., The American Society of Metals, pp. 23–31.
Hamilton, C. H., 1985, “Superplasticity in Titanium Alloys,” Superplastic Form
ing, S. P. Agrawal, ed., The American Society of Metals, pp. 12–22.
Hamilton, C. H., Zbib, H. M., Johnson, C. H., and Richter, S. K., 1991, “Dynamic
Grain Coarsening and its Effects on Flow Localization in Superplastic Deformation,”
Proceedings of the Second International SAMPE Symposium, Chipa, Japan, pp.
272–279.
Hill, R., 1950, The Mathematical Theory of Plasticity, Oxford University Press,
London.
Holt, D. L., 1970, “Analysis of the Bulging of Superplastic Sheet by Lateral
Pressure,” International Journal of Mechanical Science, Vol. 12, pp. 491–497.
Khraisheh, M. K., Bayoumi, A. E., Hamilton, C. H., Zbib, H. M., and Zhang, K.,
1995, “Experimental Observations of Induced Anisotropy During the Torsion of
Superplastic PbSn Eutectic Alloy,” Scripta Metallurgica et Materialia, Vol. 32, No.
7, pp. 955–959.
Khraisheh, M. K., Zbib, H. M., Hamilton, C. H., and Bayoumi, A. E., 1997,
“Constitutive Modeling of Superplastic Deformation. Part I: Theory and Experi
ments,” International Journal of Plasticity, Vol. 13, No. 1–2, pp. 143–164.
Khraisheh, M. K., and Zbib, H. M., 1997, “Constitutive Modeling of Anisotropic
Superplastic Deformation,” Physics and Mechanics of Finite Plasticity and Visco
plastic Deformation, A. S. Khan, ed., NEAT Press, Fulton MD, pp. 45–46.
Khraisheh, M. K., and Zbib, H. M., 1999, “Optimum Forming Loading Paths for
PbSn Superplastic Sheet Materials,” ASME JOURNAL OF ENGINEERING MATERIALS AND
TECHNOLOGY, Vol. 121, No. 3, pp. 341–345.
Mayo, M. J., and Nix, W. D., 1989, “Direct Observation of Superplastic Flow
Mechanisms in Torsion,” Acta Metallurgica, Vol. 37, No. 4, pp. 1121–1143.
Smith, M. T., 1998, “Research Toward the Increased Use of Superplastic Forming
in Lightweight Structures,” Proceedings of the Symposium on Superplasticity and
Superplastic Forming, 35
th
Annual Meeting of the Society of Engineering Science,
M. Khaleel, ed., Paciﬁc Northwest National Laboratory, PNNLSA30406.
Fig. 10 The effect of c
3
on the anisotropic yield surface for tension
torsion case (Eq. (11)). ؍ 30°, c
1
؍ 1, and c
2
؍ 5. The dashed line
represents von Mises surface (c
1
؍ c
2
؍ c
3
؍ 0).
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 97
Masahito Mochizuki
Department of Manufacturing Science,
Osaka University,
Suita, Osaka 5650871,
Japan
(Formerly, Hitachi, Ltd.)
email: mmochi@mapse.eng.osakau.ac.jp
Makoto Hayashi
Toshio Hattori
Mechanical Engineering Research Laboratory,
Hitachi, Ltd.,
Tsuchiura, Ibaraki 3000013,
Japan
Numerical Analysis of Welding
Residual Stress and Its
Veriﬁcation Using Neutron
Diffraction Measurement
Direct measurements and computed distributions of throughthickness residual stress in a
pipe buttwelded joint and a pipe socketwelded joint are compared. The analytical
evaluation methods used were inherent strain analysis and thermal elasticplastic anal
ysis. The experimental methods were neutron diffraction for the internal residual stress,
and Xray diffraction and straingauge measurement for the surface stress. The residual
stress distributions determined using these methods agreed well with each other, both for
internal stress and surface stress. The characteristics of the evaluation methods and the
suitability of these methods for each particular welded object to be evaluated are
discussed.
Introduction
Welding residual stress can eventually result in brittle fracture,
fatigue failure, and stress corrosion cracking in welded structures.
Therefore, when a welded structure needs to be strengthdesigned
or crack propagation needs to be evaluated, it is important to know
the distribution of residual stress in detail including through
thickness distributions (Almen and Black, 1963; Todoroki and
Kobayashi, 1988). A method that can be used to directly measure
residual stress in the welded structure should be established be
cause each structure naturally contains dispersion of welding re
sidual stress. There are many difﬁculties in the stress measurement
because the various conditions of the measurement are limited by
the environment of the welded structure or by the speciﬁcation of
the measurement method. Therefore, we should also establish an
analytical method for welding residual stress at the same time as
that we are developing a measurement method.
We have developed analytical evaluation methods for welding
residual stress by using inherent strain analysis (1995; 1997) and
thermal elasticplastic analysis (1993; 1996). The analytical results
were veriﬁed by comparing them with those obtained from other
evaluation techniques. That is, for the stress on the surface of
welded structures, calculated residual stress were compared with
straingauge measurements and they showed good agreement. We
also compared the throughthickness residual stress from the in
herent strain analysis and the thermal elasticplastic analysis. And
it was shown that both stress distributions agree well. However,
throughthickness residual stress obtained by the analytical method
has hardly been directly compared with measured values for
welded structures.
Neutron diffraction is a common method for measuring internal
stress of a structure (McEwen et al., 1983; Allen et al., 1985). The
principle of neutron diffraction is the same as that of Xray
diffraction. Residual stress at the structure surface can be measured
by using Xray diffraction because the penetration depth of Xrays
is extremely shallow (about 20 m from the surface). On the other
hand, the penetration depth of neutrons is deep, about 50 mm from
the surface even in case of steel, so it can be used to measure the
throughthickness residual stress in steel welded structures.
Throughthickness residual stress in various kinds of welded joints
has been measured using neutron diffraction (Hutchings, 1990;
Wikander et al., 1994; Hayashi and Ishiwata, 1995). However, the
veriﬁcation of measured values by neutron diffraction are seldom
done for welded structures. In particular, the measured values for
throughthickness welding residual stress by neutron diffraction
have not been validated by comparing them with other evaluation
results. Furthermore, it is commonly considered to be difﬁcult for
neutron diffraction to measure residual stress at a stress
concentration point because the minimum volume to be measured
depends on the width of neutron beam.
In this paper, we compare direct measurements and computed
distributions of throughthickness residual stress in welded joints.
Residual stresses are evaluated in a pipe buttwelded joint and a
pipe socketwelded joint. Inherent strain analysis and thermal
elasticplastic analysis are adopted as the analytical evaluation
technique and neutron diffraction is used as the measurement
technique for evaluating throughthickness residual stress. Stress
measurement method using a strain gauge and Xray diffraction
method for measuring residual stress on the structure surface were
simultaneously used for comparing analyzed residual stress distri
butions on the structure surface. It is shown that the most suitable
method for evaluating residual stress can be selected by consider
ing each characteristics of the evaluation technique and the con
dition of the welded object.
Residual Stress In Pipe ButtWelded Joint
The Object for Residual Stress Evaluation. The geometry and
dimensions of a 4in.diameter pipe buttwelded joint, which is made
of carbon steel (JIS STPT410), used to evaluate residual stress are
shown in Fig. 1. The angle of the Vshaped groove is 60 degrees, and
the groove root width is 3 mm. There are four welding layers and four
welding passes. The heat input is 1.2 kJ/mm for each welding pass.
Strainrelief treatment was done before welding in order to remove
residual stress due to machining.
Evaluating Residual Stress
Inherent Strain Analysis. When stress is induced in a body
without the action of external load, the stress is called residual
stress which is selfequilibrating, and the source of residual stress
should exist in the body. The source of residual stress is incom
patible strain which is called inherent strain. In the case of welded
joints, the source of residual stress is plastic strain including
dislocations. The plastic strain is regarded as inherent strain.
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
October 7, 1998; revised manuscript received May 11, 1999. Associate Technical
Editor: N. Chandra.
98 / Vol. 122, JANUARY 2000 Transactions of the ASME Copyright © 2000 by ASME
The following elastic response equations are generally derived
as the relationships between the vectors of inherent strain {⑀*},
elastic strain {⑀
e
}, and residual stress {} which occur at any
location in an elastic body owing to the inherent strain (Ueda and
Fukuda, 1989).
͕͖ ϭ ͓D͔͕⑀
e
͖
ϭ ͓D͔͓H*͔͕⑀*͖
ͮ. (1)
Here the matrix [H*] is the elastic straininherent strain matrix and
the matrix [D] is the elastic stressstrain matrix. The matrix [H*]
links the overall ﬁelds of inherent strain to the overall ﬁelds of
elastic strain, and depends on the domain occupied by the body and
its boundary conditions. And the component H
*
ij
of [H*] can be
derived as the ith component of the elastic response strains subject
to the jth component of the inherent strains being unit.
The distribution zone and the magnitude of inherent strain do
not change as long as no new inherent strains are added by plastic
deformation resulting from mechanical cutting, external loads, and
so on. Therefore, even though the geometry of an object is changed
due to cutting, the inherent strain can be estimated from the
observed value of elastic strain as follows (Mochizuki et al., 1995).
First, to measure the elastic strain induced in a structural mem
ber, as many changes of strain as possible need to be observed by
cutting or the like. Since various kinds of errors in measurement
may be induced in the observed strains, the measurement equation
is as follows.
͕
m
⑀
e
͖ Ϫ ͓H*͔͕⑀ˆ *͖ ϭ ͕r͖ (2)
where {⑀ˆ*} is the most probable value for inherent strain and {r}
is the residual.
The most probable value {⑀ˆ*} for the inherent strain can be
determined from a condition that minimizes the sum of squares of
the residual {r} as in the following equation.
͕⑀ˆ *͖ ϭ ͓͑H*͔
T
͓H*͔͒
Ϫ1
͓H*͔
T
͕
m
⑀
e
͖ (3)
The most probable value {⑀ˆ*} for welding residual stress at any
location can be determined by substituting the most probable value
{⑀ˆ*} obtained for the inherent strain {⑀*} in Eq. (1).
The procedure for evaluating residual stress in the pipe butt
welded joint is shown in Fig. 2. Strain gauges are bonded on the
surface of the welded joint. Cutting the welded joint using an
abrasive cutoff machine and bonding the strain gauges to a section
after cutting are repeated one after the other. Inherent strain is
calculated from measured elastic strain {
m
⑀
e
} by using Eq. (3).
The detailed procedure is as follows.
After the strain gauges are bonded on the surface of the welded
joint, the Tspecimen and the L
i
specimens (i ϭ 1 to n) are cut
from the original pipe buttwelded joint. The inherent strain in the
original welded joint is the same in the Tspecimen and the
L
i
specimens. This is because inherent strain does not change even
if the specimen is cut. Residual stress remaining in each L
i

specimen results only from by circumferential inherent strain ⑀
*
.
The inherent strains in the other directions (radial and axial) do not
contribute to residual stress in the L
i
specimens. The shear com
ponents of inherent strains can be ignored because they are much
smaller. Residual stress remaining in the Tspecimen also occurs
by axial inherent strain ⑀
*
z
and radial inherent strain ⑀
*
r
.
Each released strain is measured after cutting around the strain
gauges on the specimens to form dicelike pieces. By substituting
these released strains into Eq. (3), circumferential inherent strain
⑀
*
can be calculated from the measured data of the L
i
specimens.
And axial inherent strain ⑀
*
z
and radial inherent strain ⑀
*
r
can be
calculated from the measured data of the Tspecimen. The original
inherent strain distribution of the whole welded joint can be
constructed by gathering the calculated inherent strain from each
small specimen. Threedimensional residual stress including
throughthickness stress can be calculated by using the complete
inherent strain in the pipe buttwelded joint. This inherent strain
analysis was done by the elastic ﬁnite element method.
Thermal ElasticPlastic Analysis. Thermal elasticplastic be
havior during each welding step caused by temperature change
was simulated by thermal elasticplastic analysis. The detailed
ﬂow is described as follows (Mochizuki et al., 1993; Mochizuki et
al., 1995).
The material properties at each temperature from room temper
ature to melting point must be known for the materials used in the
analysis. The temperature distribution at each time due to the heat
input of welding was computed by transient heat conduction
analysis during welding procedure. Transient stress during the
welding steps was successively calculated from thermal loads
obtained by heat conduction analysis, which was done by thermal
elasticplastic analysis with the nonlinearity of materials. Residual
stress can be determined as the stress when the material is cooled
down to room temperature after welding. The residual stress in the
pipe buttwelded joint was computed by thermal elasticplastic
analysis using an axisymmetric model.
Neutron Diffraction Measurement. Neutron diffraction is used
for measuring the spacing d between the atomic planes of a crystal
Fig. 2 Sequences of measuring released strains for calculating inherent
strains in a pipe buttwelded joint
Fig. 1 Conﬁguration of a carbonsteel pipe buttwelded joint
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 99
lattice (Allen et al., 1985). A neutron beam of known wavelength
is diffracted by a scattering angle 2 from its incident direction
according to Bragg’s law.
2d ⅐ sin ϭ n (4)
Elastic strain ⑀ is determined by comparing the measured value
of d to the value measured in a suitable stressfree reference d
0
by
using the differentiated equation (4).
⑀ ϭ
⌬d
d
0
ϭ
d Ϫ d
0
d
0
ϭ Ϫcot ⅐ ⌬. (5)
Measurements of the change of a spacing between the atomic
planes of a crystal lattice, ⌬d ϭ d Ϫ d
0
, or a scattering angle ⌬
can be used to calculate elastic strain ⑀. To calculate ⑀, it is
necessary to measure d
0
or
0
under stressfree conditions. Ac
cordingly, the reference scattering angle
0
must be measured after
a part of the specimen of the measured object is cut out and treated
by stressrelief annealing. A characteristic of neutron diffraction is
that the measurement of
0
is essential, whereas it is not necessary
for Xray diffraction.
Strain is measured in a direction that bisects the incident and
diffracted neutron beams. To obtain values for strain in the prin
cipal directions of the object to be measured (that is, axial strain ⑀
z
,
radial strain ⑀
r
, and circumferential strain ⑀
), the object must be
reoriented so that a bisector is located in the required direction, as
Fig. 3 shows. The incident and diffracted beams are shaped by slits
in neutronabsorbing cadmium masks.
Measurements are made within the region of intersection of
these two beams. With three measurements of strain at each
location, the principal residual stresses, axial stress
z
, radial stress
r
, and circumferential stress
, can be calculated through a
generalized Hooke’s law.
z
ϭ
E
N
1 ϩ
ͭ
⑀
z
ϩ
1 Ϫ 2
͑⑀
z
ϩ ⑀
r
ϩ ⑀
͒
ͮ
(6)
r
ϭ
E
N
1 ϩ
ͭ
⑀
r
ϩ
1 Ϫ 2
͑⑀
z
ϩ ⑀
r
ϩ ⑀
͒
ͮ
(7)
ϭ
E
N
1 ϩ
ͭ
⑀
ϩ
1 Ϫ 2
͑⑀
z
ϩ ⑀
r
ϩ ⑀
͒
ͮ
(8)
Here E
N
is Young’s modulus for neutron diffraction and is
Poisson’s ratio.
The diffraction plane dependence due to elastic anisotropy of
the material affect elastic constants for neutron or Xray diffrac
tion. Thus, the Young’s modulus for neutron diffraction E
N
is
different from the mechanical Young’s modulus E which is typi
cally used. The appropriate elastic constraints for neutron diffrac
tion are E
N
ϭ 243 GPa and ϭ 0.28 for elastic strain determined
from shifts in the (211) diffraction peak in a pipe buttwelded joint
made of ferritic steel (Hayashi et al., 1995a), and internal residual
stress was measured. The neutron diffractometer at Chalk River
Nuclear Laboratories of Atomic Energy of Canada Limited was
used in the measurement (Hayashi and Ishiwata, 1995).
XRay Diffraction Measurement. The principle of Xray dif
fraction measurement is the same as that of neutron diffraction.
However, only residual stress near the surface can be measured
because the penetration depth of Xrays is extremely shallow
(about 20 m). The stress component of the normal direction on
the surface can therefore be regarded as 0, and it is not necessary
to measure the scattering angle. Residual stress in the pipe butt
welded joint was measured by a sin
2
Ϫ 6 points method (Society
of Material Science, Japan Ed., 1981). CrK Xrays were used for
the measurement.
StrainGauge Measurement. Measuring stress by strain gauge
can obtain the residual stress on the surface of the structure to be
evaluated (Lu Ed., 1996). Once small pieces are cut off from the
object, residual stress in the small pieces is physically released.
Residual stress is converted into the released elastic strain which is
obtained as the difference between before and after cutting. The
detailed procedure for measuring the residual stress on the surface
of the pipe buttwelded joint is as follows.
First, strain gauges were bonded on the inside and outside
surface of the welded joint. The periphery of the bonded strain
gauges in the welded joint was cut into small cubes with about
10mm sides. Residual stress in the small pieces was released by
cutting the welded joint, and axial released strain ⑀
z
and circum
ferential released strain ⑀
were measured. Axial residual stress
z
and circumferential residual stress
can be obtained from the
following equations using measured strains.
z
ϭ Ϫ
E
1 Ϫ
2
͑⑀
z
ϩ ⑀
͒ (9)
ϭ Ϫ
E
1 Ϫ
2
͑⑀
ϩ ⑀
z
͒ (10)
Here E is Young’s modulus and is Poisson’s ratio for the welded
joint.
Comparison of Residual Stress Distributions. The welding
residual stress in a carbonsteel pipe buttwelded joint was evalu
ated using each method. Throughthickness residual stress was
compared with the results obtained from inherent strain analysis,
Fig. 3 Scattering geometries for measurement of threecomponents of
lattice strain in a pipe buttwelded joint
100 / Vol. 122, JANUARY 2000 Transactions of the ASME
thermal elasticplastic analysis, and neutron diffraction measure
ment. Residual stress on the surface was evaluated by inherent
strain analysis, thermal elasticplastic analysis, Xray diffraction
measurement for the outer surface, and straingauge measurement.
Axial residual stresses along the weldcenter line through the
thickness direction are compared in Fig. 4. The horizontal axis is
the radial distance from the inner surface to the outer surface.
These stress distributions by inherent strain analysis, thermal
elasticplastic analysis, and neutron diffraction measurement agree
well with each other. Axial residual stresses along the heat
affected zone are compared in Fig. 5. Residual stress from all three
methods agree except the measurements near the inner surface by
neutron diffraction. These measurement values are a little lower
than the analytical results.
Circumferential residual stresses along the weldcenter line and
along the heataffected zone through the thickness direction are
shown in Figs. 6 and 7, respectively. Both ﬁgures indicate that the
results for the three evaluation methods completely agree well. It
is thought that these results verify the validity of stress distribution
by comparing each other.
Axial residual stress on the outer surface of the pipe buttwelded
joint is shown in Fig. 8. The stress distributions by inherent strain
analysis, thermal elasticplastic analysis, Xray diffraction mea
surement, and straingauge measurement are compared in the
ﬁgure. Analytical results by thermal elasticplastic analysis are
little higher than those of others, but the other residual stresses
wholly agree well. And thermal elasticplastic analysis gives
slightly different results in this ﬁgure.
Axial residual stresses on the pipe inner surface are compared in
Fig. 9. Residual stresses by using each evaluation method agree
well, except the results of thermal elasticplastic analysis are a
little different to those of others.
Residual Stress in Pipe SocketWelded Joint
Throughthickness residual stress in a pipe socketwelded joint was
also evaluated as another example for the comparison of internal
residual stress. The conﬁguration for the pipe socketwelded joint is
shown in Fig. 10. Two smalldiameter pipes, which are made of
carbon steel (JIS STPT410), are ﬁlletwelded with a socket, which is
made of carbon steel (JIS S25C). The number of welding layers are
two and the number of passes are three. The heat input is 1.4 kJ/mm
for each welding pass. Strainrelief treatment was done before weld
ing in order to remove residual stress by machining.
Throughthickness residual stresses in the welding deposit were
measured by using neutron diffraction. The measuring positions for
Fig. 4 Comparison of axial residual stress across the weld metal center
in a pipe buttwelded joint
Fig. 5 Comparison of axial residual stress across the heataffected
zone in a pipe buttwelded joint
Fig. 6 Comparison of circumferential residual stress across the weld
metal center in a pipe buttwelded joint
Fig. 7 Comparison of circumferential residual stress across the heat
affected zone in a pipe buttwelded joint
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 101
the neutron diffraction are shown in Fig. 11 (Hayashi et al., 1995b).
The sampling volume was about 1 mm ϫ 1 mm ϫ 2 mm with the
long dimension tangent to the curvature of the pipe for the axial and
radial strains, and 1 mm ϫ 1 mm ϫ 4 mm for circumferential strain.
Residual stress in the socketwelded joint was also calculated by using
inherent strain analysis and thermal elasticplastic analysis. The min
imum mesh size for the thermal elasticplastic analysis was nearly the
same as the measured area of the neutron beam.
Axial and circumferential residual stresses in the weldmetal of
the pipe socketwelded joint are compared in Figs. 12 and 13,
respectively. Both distributions are agreed with each other in each
ﬁgure, except near the welding root, in which the stress is con
centrated. It is surely shown to be difﬁcult for neutron diffraction
to measure residual stress at the stressconcentration point, because
the minimum range to be measured depends on the area of the
neutron beam and because a value of the residual stress by neutron
diffraction is obtained as the average in the measuring volume.
Fig. 8 Comparison of axial residual stress on the outer surface in a pipe
buttwelded joint
Fig. 9 Comparison of axial residual stress on the inner surface in a pipe
buttwelded joint
Fig. 10 Conﬁguration of a smalldiameter pipe socketwelded joint
Fig. 11 Measuring points of residual stress in a pipe socketwelded
joint by neutron diffraction
Fig. 12 Comparison of axial residual stress in the deposit in a pipe
socketwelded joint
Fig. 13 Comparison of circumferential residual stress in the deposit in
a pipe socketwelded joint
102 / Vol. 122, JANUARY 2000 Transactions of the ASME
Discussion for Evaluation Methods
We conﬁrmed that residual stress evaluated by using each
method agreed very well for two pipe welded joints. This means
that each method can precisely evaluate welding residual stress for
both internal and surface stresses. Therefore, the characteristics of
each evaluation method are examined as follows by considering
the conditions under which each evaluation method was applied to
an actual welded structure.
Characteristics of Each Evaluation Method
Inherent Strain Analysis. This method can be used to simply
calculate residual stress by elastic analysis. It can easily be applied
to a lot of welded structures with various conﬁgurations, once the
database for inherent strain distribution is constructed (Mochizuki
et al., 1997). On the other hand, the evaluations for structural
shape, welding method, or welding conditions which do not con
form to a database are difﬁcult. Even if these are done, it is
necessary to assume the inherent strain distribution to be used in
the evaluation. Furthermore, transient thermal stress or stress his
tory in the welding process cannot be calculated by using inherent
strain analysis.
Thermal ElasticPlastic Analysis. Welding residual stress can
be accurately calculated by simulating the change of mechanical
behavior caused by thermal elasticplastic phenomenon during the
welding process. In thermal elasticplastic analysis, the time spent
is extremely long. Also, the material properties of the object from
room temperature to melting point should be known. However,
thermal elasticplastic analysis can be used to calculate residual
stress with some degree of precision, even material property can be
deﬁned precisely, and even the welded objects can be treated by
the twodimensional models. In other words, thermal elastic
plastic analysis becomes a powerful technique when the analysis
condition is appropriately known. It is also suitable for a welded
structure with special shape, in which the application of inherent
strain analysis is difﬁcult, and for materials that are difﬁcult to
investigate, such as irradiated material.
Neutron Diffraction Measurement. This technique can be used
directly to obtain internal residual stress by nondestructive mea
surement. However, the objects which can be measured are limited
depending on the condition of the neutron source device. The
spacing between the atomic planes of the crystal lattice or scatter
ing angle under stressfree conditions must be measured as a
reference. There are many difﬁculties in direct application to an
actual structure. But neutron diffraction becomes very effective
when the residual stress in a small test piece is to be measured or
when the veriﬁcation using a mockup specimen is necessary.
XRay Diffraction Measurement. Nondestructive measurement
on the structural surface is possible using Xray diffraction. However,
the measured value depends on the surface conditions of the object to
be measured (Park et al., 1994). Therefore, scrupulous attention to
detail is necessary in measurement. The measurement becomes dif
ﬁcult so that the diffraction proﬁle is hard to obtain at the place where
the crystalline structure of the welded metal is large, as in some
materials. It can be used to measure an actual structure because the
body for the Xray equipment can be miniaturized.
StrainGauge Measurement. Method of stress measurement
using the straingauge can obtain average stress in small pieces
that have been cut into cubes. It is very difﬁcult to apply to actual
structures because straingauge measurement is a destructive
method. However, it is the most standard technique for the residual
stress measurement in a welded specimen. Throughthickness re
sidual stress can be evaluated when the inherent strain method is
used in combination. Measuring a steep grade in a minute domain
such as a thin ﬁlm is difﬁcult with the straingauge method.
Most Suitable Method. In conclusion, the most suitable
method of evaluating residual stress in welded structures should be
selected depending on the purpose of measurement by considering the
evaluating location and the operating conditions of the welded object
to be evaluated. Using several kinds of evaluations together is also
effective to increase the reliability of residual stress evaluation.
Summary
Throughthickness residual stress in a pipe buttwelded joint and
a pipe socketwelded joint was compared. The analytical evalua
tion methods were inherent strain analysis and thermal elastic
plastic analysis and the experimental methods were neutron dif
fraction for the internal stress. And Xray diffraction and strain
gauge measurement for the surface residual stress were also used.
The residual stress distributions determined using these methods
agreed well with each other, both for internal stress and surface
stress. The residual stress distributions obtained by each evaluation
method were shown to agree mutually. The characteristics of the
evaluation methods were summarized and it was found that the
most suitable method under any particular situation can be selected
by considering the evaluated location and the operating conditions
of the welded object to be evaluated.
References
Allen, A. J., Hutchings, M. T., Windsor, C. G., and Andreani, C., 1985, “Neutron
Diffraction Methods for the Study of Residual Stress Fields,” Advances in Physics,
Vol. 34, pp. 445–473.
Almen, J. O., and Black, P. H., 1963, Residual Stresses and Fatigue in Metals,
McGrawHill, New York.
Hayashi, M., Ishiwata, M., Minakawa, N., Funahashi, S., and Root, J. H., 1995a,
“Diffraction Plane Dependence of Elastic Constants in Ferritic Steel in Neutron
Diffraction Stress Measurement,” Journal of the Society of Materials Science, Japan
(in Japanese), Vol. 44, pp. 1115–1120.
Hayashi, M., Ishiwata, M., Minakawa, N., Funahashi, S., and Root, J. H., 1995b,
“Residual Stress Measurement of Socket Welded Joint by Neutron Diffraction,”
Journal of the Society of Materials Science, Japan (in Japanese), Vol. 44, pp.
1464–1469.
Hayashi, M., and Ishiwata, M., 1995, “Residual Stress Measurement of Butt
Welded Carbon Steel Pipe by Neutron Diffraction,” Proceedings of the 13th Inter
national Conference on NDE in the Nuclear and Pressure Vessel Industries, Kyoto,
Japan, pp. 22–25.
Hutchings, M. T., 1990, “Neutron Diffraction Measurement of Residual Stress
Fields, The Answer to the Engineers Prayer,” Journal of Nondestructive Testing and
Evaluation, Vol. 5, pp. 395–403.
Klassen, R. J., Root, J. H., and Holden, T. M., 1991, “Neutron Diffraction
Measurements of Strain in a CANDU3 Rolled Joint Hub,” AECL Report, ANDI50.
Lu, J., Edited, 1996, Handbook of Measurement of Residual Stresses, Society for
Experimental Mechanics, Inc., Fairmont Press, Inc., Lilburrn.
McEwen, S. R., Holden, T. M., Hobson, R. R., and Cracknell, G., 1983, “Residual
Strains in Rolled Joints,” Transactions of the 9th International Conference on
Structural Mechanics in Reactor Technology, Chicago, Div. G, pp. 183–188.
Mochizuki, M., Enomoto, K., Okamoto, N., Saito, H., and Hayashi, E., 1993,
“Welding Residual Stresses at the Intersection of a Small Diameter Pipe Penetrating
a Thick Plate,” Nuclear Engineering and Design, Vol. 144, No. 3, pp. 439–447.
Mochizuki, M., Saito, N., Enomoto, K., Sakata, S., and Saito, H., 1995, “A Study
on Residual Stress of ButtWelded Plate Joint Using Inherent Strain Analysis,”
Transactions of the 13th International Conference on Structural Mechanics in Reac
tor Technology, Porto Alegre, Brazil, Vol. 2, Div. F, pp. 243–248.
Mochizuki, M., Hayashi, M., and Hattori, T., 1996, “Effect of Welding Sequence
on Residual Stress in MultiPass ButtWelded Pipe Joints,” Transactions of the Japan
Society of Mechanical Engineers (in Japanese), Vol. 62A, No. 604, pp. 2719–2725.
Mochizuki, M., Hayashi, M., Nakagawa, M., Tada, N., and Shimizu, S., 1997, “A
Simpliﬁed Analysis of Residual Stress at Welded Joints Between Plate and Penetrat
ing Pipe,” JSME International Journal (Series A), Vol. 40, No. 1, pp. 8–14.
Park, W., Kobayashi, H., Nakamura, H., and Koide, T., 1994, “Measurement and
Evaluation of Welding Residual Stresses for Austenitic Stainless Steel by Xray
Diffraction Method,” Proceedings of the 1994 Annual Meeting of JSME/MMD (in
Japanese), Vol. B, No. 94037, pp. 70–71.
The Society of Materials Science, Japan, Edited, 1981, Standard for Xray Stress
Measurement, Yokendo Ltd. (in Japanese).
Todoroki, A., and Kobayashi, H., 1988, “Prediction of Fatigue Crack Growth Rate
in Residual Stress Field (Application of Superposition Technique),” Transactions of
the Japan Society of Mechanical Engineers (in Japanese), Vol. 54A, No. 497, pp.
30–37.
Ueda, Y., and Fukuda, K., 1989, “New Measuring Method of ThreeDimensional
Residual Stresses in Long Welded Joints Using Inherent Strain as Parameters—Lz
Method,” ASME JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 111, pp.
1–8.
Wikander, L., Karlsson, L., Nasstrom, M., and Webster, P., 1994, “Finite Element
Simulation and Measurement of Welding Residual Stresses,” Modeling and Simula
tion in Materials Science and Engineering, Vol. 2, pp. 845–864.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 103
Tang Yaxin
Zhang Yongkang
Zhang Hong
Yu Chengye
Department of Mechanical Engineering,
Nanjing University of Aeronautics &
Astronautics,
Nanjing, 210016, P.R. China
Effect of Laser Shock
Processing (LSP) on the
Fatigue Resistance of an
Aluminum Alloy
The laser shock induced stress wave is described and measured with a polyvinylidene
ﬂuoride (PVDF) transducer. A principle for selecting laser parameters is proposed. A
small sized laser with a high power is used for Laser Shock Processing (LSP). The fatigue
life of the aluminum alloy 2024T62 is greatly improved after LSP. With 95% conﬁdence,
the mean fatigue life of LSP specimens is 4.5–9.8 times that of unshocked ones. The fatigue
and fracture resistance mechanisms of LSP such as the variation of the surface hardness,
the microstructure and the fracture section of specimens before and after LSP are
analyzed.
1 Introduction/Background
Laser Shock Processing (LSP) uses an intense laser beam with
a high power density (GW/cm
2
) and a short duration (tens of ns)
to irradiate the surface of metals. It forms a stress wave (shock
wave) which is transmitted into the metal. If the peak pressure of
the stress wave is larger than the Hugoniot Elastic Limit (HEL) of
the metal, plastic deformation is produced and a strengthened layer
is formed on the surface of the metal, and the mechanical proper
ties of the metal are improved. Compared with other laser treat
ment methods, the duration of laser shock processing is very short,
so LSP can mitigate harmful heat effects in the metal. LSP has
wide applications in many manufacturing industries, such as en
gines, tankers, ships as well as automobiles [1–4].
LSP can be very useful in fatigue and fracture resistance areas
of aircraft structures, in which the structural material is expected to
be subjected to a large number of fatigue cycles. In some detailed
structures such as small holes, small slots, it is very difﬁcult for
conventional strengthening technology such as the shot peening
and the cold expansion to be used due to limited accessibility. But
LSP can easily reach these areas by using optical methods. LSP
can also provide aircraft structures with highamplitude residual
compressive stresses and an excellent original fatigue qualities of
structural details to lengthen the useful life of airplanes. Therefore,
LSP technology is of great value for increasing fatigue and fracture
resistance of aircraft structures in the development of modern
airplanes [5].
2 Experimental Procedure
2.1 Calculation of the Laser Shock Induced Stress Wave.
A schematic of the LSP process is shown in Fig. 1. When the
surface of a metal is irradiated with a high power laser beam, the
surface temperature of the metal rapidly rises to its vaporization
temperature. For example, if the input power density is 1 GW/cm
2
,
the vaporization temperature of the metal will be reached in less
than 1 ns. The irradiated zone is vaporized violently, which pro
duces an expanding plasma and causes a stress wave in the metal.
According to the theory of explosive gas dynamics [6, 7], the
following can be used for estimating the shock pressure P:
P ϭ
ͩ
k ϩ 1
2k
ͪ
2k/kϪ1
⅐
͓2͑k
2
͒ Ϫ 1͔͒
2/3
k ϩ 1
⅐
1/3
⅐ ͑AI
0
͒
2/3
(1)
where
k — ratio of speciﬁc heat
— metal density
A — ratio of laser absorption
I
0
— laser power density
The ratio of laser absorption A was measured as 0.8. Assuming
the metal steam is an ideal singleatom gas where k Ϸ 1.67,
formula (1) can be reduced as follow:
Ϸ 0.266
1/3
⅐ I
0
2/3
(2)
2.2 Measurement of the LSP Stress Wave Using the PVDF
Transducer. The polyvinylidene ﬂuoride (PVDF) ﬁlm has been
widely used to measure shock loads: plate impacts, explosions, etc.
These ﬁlms have a good reproducibility for measuring a wide
range of pressures up to 20 GPa [8]. In order to measure laser
shock pressures, a special PVDF transducer (shown in Fig. 2) has
been developed. The transducer consists of a 50 m thick PVDF
ﬁlm and two electrodes of 100 m thick aluminum foils. A
resistor, R is connected with the ends of both electrodes. When the
PVDF ﬁlm is shocked by a pulsed laser, it produces an electric
current through the circuit, which is represented by the voltage
across the resistor, R. The output voltage of this resistor is fed to
an oscilloscope (25 MHz, dual channels). When the laser power
density is 5.7 ϫ 10
8
W/cm
2
(wavelength 1.06 m, pulse duration
50 ns), the stress wave induced by laser shock in the PVDF ﬁlm is
shown in Fig. 3, and the measured peak pressure is 0.3 GPa. As the
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division May
21, 1998; revised manuscript received March 10, 1999. Associate Technical Editor:
K. S. Chan.
Fig. 1 Schematic of the LSP process
104 / Vol. 122, JANUARY 2000 Transactions of the ASME Copyright © 2000 by ASME
impedance mismatch between the Al foil (Z
Al
ϭ1.3 ϫ10
7
Kg/m
2
⅐
s) and the PVDF (Z
PVDF
ϭ 0.35 ϫ 10
7
Kg/m
2
⅐ s) is taken into
account, the peak pressure applied on the backside of Al foil can
be determined. In this case, the peak pressure is 0.8 GPa. While
using formula (2), the calculated value of peak pressure under the
same parameters is 1.05 GPa, which is close to the experimental
result. Formula (2) can be accepted to estimate the peak pressure
of LSP.
2.3 Fatigue Specimen. The fatigue specimen is of the dual
dogbone type geometry and made of the aluminum alloy
2024T62. Two fatigue holes with diameter 2 mm are bored with a
precise jig boring machine, which are in both the LSP area and the
unshocked area, respectively, as shown in Fig. 4. It is ensured that
the qualities of the two holes are similar. The roughness is less than
0.1 m on the inside surface of the two holes.
2.4 Laser and Laser Parameters Selection. A Qswitched
Nd: glass laser with wavelength of 1.06 m and pulse duration of
30 ns was used. The laser has three ampliﬁer stages, namely, the
preset ampliﬁer, the triple ampliﬁer and the postset ampliﬁer.
The modulation crystal is a deuterated potassium dihydrogen phos
phate (KD*P) crystal. A storage oscilloscope (7834Tek) with a
resolution of 1 ns is used to monitor the wave shape of the laser at
regular intervals. Eight percent of the laser beam energy was cut by
a plate and transferred directly to an energy meter to achieve
online control.
Eleven specimens were LSP shocked successively on both sides.
The shocked surface was coated with a black paint and covered
with a overlay of K9 optical glass. The selected laser parameters
are shown in Table 1. Using formula (2), the evaluated value of
peak pressure for the stress wave is about 1.2–2.2 GPa, which is
larger than the Hugoniot Elastic Limit of 2024T62 of 0.53 GPa. It
is obvious that the strengthened zone is formed to different extents.
2.5 Fatigue Test. All of the fatigue specimens were tested to
failure at a maximum load of 4.2 kN, and a stress ratio of R ϭ 0.1
under constantamplitude load control with precision Ϯ5%, using
a test frequency 13 Hz, i.e., low frequency fatigue failure.
In order to identify the mechanism of LSP, the surface hardness,
the microstructure, and the residual stress of the LSP specimen
were measured and were compared with those of the unshocked
material. The fatigue fractures of shocked and unshocked speci
mens were also analyzed.
3 Results and Discussion
3.1 Results of Fatigue Test. The fatigue test results are
shown in Table 2. Considering the specimen number 8 seems
unusual, its result is left out. So the fatigue life of 2024T62
increases by 2.2–8.7 times, with an average increase of 5.2 times.
Based on the consideration of 95% conﬁdence, the mean fatigue
life of LSP specimens is 4.5–9.8 times that of the unshocked ones.
3.2 Hardness. The microhardness distribution of the LSP
area is measured with a Model 71 microhardness meter under a
load of 50 g and a pressure period of 15 s. The result is shown in
Fig. 5. It is shown that the hardness of the unshocked area is 72–93
Hv and the hardness of the LSP area is 98–137 Hv. There are large
hardness variations across the LSP area. In general, the hardness of
the center is higher than that of fringe of the LSP area. The
increase of hardness means that it is difﬁcult to produce the grain
boundary slip and form fatigue slipbands, i.e., the fatigue life
increases.
3.3 Microstructure. The microstructure of speciments be
fore and after LSP was examined by using a transmission electron
microscope (TEM). The TEM photos are shown in Fig. 6. It is
shown that the dislocation density of the aluminum alloy increases
distinctly after LSP, which is thought to contribute to the improve
ment in the fatigue life.
3.4 Residual Stress. The residual stress on the surface of the
LSP specimen was measured using Xray stress meter with a spot
of 2.5 ϫ 3.5 mm, and the average residual stress of LSP area is
Ϫ30.1 MPa (acting as a compressive stress), while in the unshock
area the residual stress is approximately zero. The existence of a
Table 1 LSP parameters of 2024T62
Specimen
no. Energy (J)
Pulse
duration
(ns)
Spot
diameter
(mm)
Power
density
(GW/cm
2
)
Peak
pressure
evaluation
(GPa)
1 14.10, 15.10 30 7 1.22, 1.31 1.75, 1.84
2 18.07, 16.72 30 7 1.57, 1.45 2.08, 1.97
3 16.44, 19.03 30 7 1.42, 1.65 1.94, 2.15
4 16.85, 16.04 30 7 1.46, 1.39 1.98, 1.91
5 16.31, 15.63 30 7 1.41, 1.35 1.93, 1.88
6 16.44, 16.44 30 7 1.42, 1.42 1.94, 1.94
7 17.40, 16.99 30 7 1.51, 1.47 2.02, 1.99
8 18.35, 19.16 30 7 1.59, 1.66 2.09, 2.15
9 20.20, 8.10 30 7 1.75, 0.70 2.23, 1.21
10 14.0, 18.10 30 7 1.21, 1.57 1.74, 2.08
11 19.71, 19.30 30 7 1.71, 1.67 2.20, 2.00
Fig. 3 Stress wave on the front face of PVDF ﬁlm
Fig. 4 Bidetail fatigue specimen of 2024T62 AL alloy. All dimensions
are in millimeters.
Fig. 2 Sketch of PVDF transducer
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 105
residual compressive stress state in the LSP area is very useful to
resist the initiation and propagation of the fatigue crack.
3.5 SEM Analyses of Fatigue Fractures. The fatigue frac
tures were analyzed by a scanning electron microscope (SEM).
There is no obvious fatigue origin on the fracture surface and
sometimes there are multiple fatigue origins after LSP. In general,
the fatigue crack is initiated in the center of the inside surface of
the hole and sometimes in the subsurface of the LSP area (as
shown in Fig. 7(a)). But there is a clear fatigue origin in the
fracture of the unshocked specimen and the crack is initiated
generally in a square corner of the hole (as shown in Fig. 7(b)).
The fatigue fracture of the LSP’d specimen is smoother than that
of the unshocked specimen, and does not exhibit fatigue striations.
In contrast, the fatigue fracture of unshocked specimen is rather
rough and has clear fatigue striations.
LSP results in surface hardening, an increase in dislocation
density, and an increase in the compressive residual stress in the
surface layer. Among these factors, the compressive residual stress
is the major factor; the surface hardening and dislocation density
increases are relevant to the creation of this compressive residual
stress.
4 Conclusions
(1) LSP with a high power density (GW/cm
2
) and a short
duration (tens of ns) can induce a stress wave in metals, which can
be estimated from the theory of explosive gas dynamics and
measured with a PVDF transducer. When the peak pressure of the
stress wave is larger than the Hugoniot Elastic Limit of the
material, the metal can be fatigue strengthened.
(2) The fatigue life of 2024T62 aluminum alloy is improved
greatly after LSP. With 95% conﬁdence, the mean fatigue life of
LSP specimens is 4.5–9.8 times than that of unshocked specimens.
(3) LSP results in surface hardening, increased dislocation
density and a compressive residual stress in the strengthened layer.
The compressive residual stress is the major factor for improving
the fatigue life of the aluminum alloy.
Acknowledgment
The authors wish to thank the reviewers for their kindness in
reviewing and editing the paper.
References
1 Vaccari, J. A., “Laser shocking extends fatigue life,” American Machinist,
1992, 7, pp. 62–64.
2 Peyre, P., Merrien, P., and Lieurade, H. P., et al., “Renforcementd’alliages
d’aluminum moul’esparondes dechocaser,” Mat’ eriaux & techniques, 1993, Vol.
6–7, pp. 7–12.
Fig. 5 Hardness distribution in the LSP area of 2024T62
Fig. 6 The microstructure comparision of 2024T62 aluminum alloy be
fore and after LSP. (a) Microscopic photo of 2024T62 before LSP ؋ 20 K;
(b) Microscopic photo of 2024T62 after LSP ؋ 20 K.
Fig. 7 The fatigue fracture of 2024T62. (a) Fracture of LSP specimen ؋
200; (b) fracture of unshocked specimen ؋ 200
Table 2 Results of 2024T62 fatigue test
Specimen
no.
Max. load
(kN)
Min. load
(kN)
Stress radio
(R)
Test
frequency
Cycles of
unshocked
specimens
Cycles of LSP
specimens
Increase
(%)
1 4.2 0.42 0.1 13 96 580 356 930 269.6
2 4.2 0.42 0.1 13 77 710 755 870 872.7
3 4.2 0.42 0.1 13 73 850 297 750 303.2
4 4.2 0.42 0.1 13 88 630 719 640 711.9
5 4.2 0.42 0.1 13 93 580 Ͼ523 080 Ͼ459
6 4.2 0.42 0.1 13 73 700 567 610 670.2
7 4.2 0.42 0.1 13 98 000 414 270 322.7
8 4.2 0.42 0.1 13 95 340 2 355 900 2 371
9 4.2 0.42 0.1 13 81 480 259 360 218.3
10 4.2 0.42 0.1 13 107 320 799 840 645.3
11 4.2 0.42 0.1 13 96 440 823 210 753.6
Average 690.8
106 / Vol. 122, JANUARY 2000 Transactions of the ASME
3 Grevey, D., Maiffredy, L., and Vannes, A. B., “Laser shock on a TRIP alloy:
mechanical and metalurgical consequences,” Journal of Materials Science, 1992, Vol.
27, pp. 2110–2116.
4 Puig, T., Decamps, B., and Bourda, C., et al., “Deformation of a ␥/␥Ј
waspaloy after laser shock,” Material Science and Engineering, 1992, Vol. A154, pp.
183–191.
5 Zhang Hong, Tang Yaxin, and Yu Chengye, et al., “Effects of laser shock
processing on the fatigue life of fastener holes,” Chinese Journal of Lasers, Vol. A23,
No. 12, pp. 1112–1116.
6 Zeldovich, Y. B., and Raizer, Y. P., Physics of shock wave and high
temperature hydrodynamics of explosive gas, New York, Academic Press, 1966, pp.
154–167.
7 Henrych, J., The dynamics of explosion and its use, Elsevier, NY, 1979, pp.
63–72.
8 Romain, J. P., Bauer, F., and Zanouri, D., “Measurement of laser
induced shock pressures using PVDF gauges,” High Pressure Science and Tech
nology, 1993, S. C. Schmidf, ed., American Institute of Physics, 1994, pp.
1915–1919.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 107
Masahito Mochizuki
1
Department of Manufacturing Science,
Osaka University,
Suita, Osaka 5650871,
Japan
email: mmochi@mapse.eng.osakau.ac.jp
Toshio Hattori
Hitachi, Ltd.,
Tsuchiura, Ibaraki 3000013,
Japan
Kimiaki Nakakado
Hitachi Construction Machinery Co., Ltd.,
Tsuchiura, Ibaraki 3000013,
Japan
Residual Stress Reduction and
Fatigue Strength Improvement
by Controlling Welding Pass
Sequences
The effects of residual stress on fatigue strength at a weld toe in a multipass ﬁllet weld
joint were evaluated. The residual stresses in the weld joints were varied by controlling
the sequence of welding passes. The residual stress at the weld toe was 80 MPa in the
specimen whose last welding pass was on the main plate side, but it was 170 MPa in the
specimen whose last pass was on the attachment side. The fatigue strength was nearly the
same at high stress amplitude for both specimens, but the fatigue strength of the specimen
whose last weld pass on the main plate was higher than that of the other specimen at low
stress amplitude. This difference is due to the magnitude of the initial residual stress and
the relaxation of the residual stress under fatigue cycling. The effects of the residual stress
were shown in a modiﬁed Goodman diagram, in which residual stress is treated as a mean
stress.
Introduction
In designing a welded structure, it is important to know the
relation between fatigue strength of welds and factors which affect
the fatigue strength (e.g., Almen and Black, 1963). The factors that
affect the fatigue strength are residual stress, stress concentration,
mechanical properties of the material, and macro/microstructure.
Residual stress is one of the most important factors, and it is well
known that residual stress is more effective on highcycle fatigue
than the other factors.
There are two methods for evaluating fatigue strength under
residual stress ﬁelds. They are fracture mechanics and fatigue
strength curve. In the method of fracture mechanics, which uses
the principle of elastic stress superposition, stress intensity factor
due to initial residual stress is analyzed and the effects of residual
stress are evaluated as the change of stress ratio (Buchalet and
Bamford, 1976; Todoroki and Kobayashi, 1988). There are two
subsidiary methods that uses fracture mechanics to evaluate fa
tigue strength: one uses the redistribution of residual stress caused
by crack propagation (Glinka, 1979; Nelson, 1982) and the other
uses effective stress intensity factor range due to crack closure
(Parker, 1982; Beghini et al., 1994). On the other hand, a method
using a fatigue strength curve is generally used by treating residual
stress as a mean stress (Maddox, 1991; Yamashita et al., 1996).
The effects of residual stress on fatigue strength are evaluated by
using a fatigue limit diagram like a modiﬁed Goodman diagram.
Most of these evaluations were conducted by using a specimen
which was made speciﬁcally for the evaluation. That is, residual
stress distribution or crack propagation path are intentionally de
termined according to the evaluation conditions. It is very impor
tant for engineers and researchers to obtain a fundamental princi
ple regarding the effects of residual stress on fatigue strength from
these evaluation results. On the other hand, the evaluations of the
relation between residual stress and fatigue strength in an actual
weld joint were difﬁcult especially at the weld toe or the weld root
where fatigue failure easily occurs (e.g., Hudak et al., 1985;
Yamashita et al., 1996). One of the reasons is considered that the
precision of residual stress analysis and measurement at the weld
toe or the weld root in an actual weld joint had not been enough to
accurately evaluate the effect on the fatigue strength (Mochizuki et
al., 1995). Furthermore, it is difﬁcult to vary residual stress by only
changing the welding process without treatment after welding such
as stressrelief heat treatment or shot peening (Mochizuki et al.,
1996).
However, it is possible to vary residual stress value by changing
welding pass sequences at a multipass ﬁllet weld joint. In this
case, the other factors which affect fatigue strength, such as
conﬁguration or mechanical properties, can be kept the same. This
method does not need any treatment after welding for varying
residual stress. Therefore, fatigue strength can be evaluated for two
different residual stress revel. In this paper, the effects of the
residual stress on fatigue strength at a weld toe in a multipass ﬁllet
weld joint were clariﬁed experimentally and theoretically. The
applicability of improving fatigue strength by optimizing welding
pass sequence was studied. A method using a fatigue strength
curve was applied to these evaluations because this method is easy
to apply to the strength design of welded structures.
MultiPass Fillet Weld Joint
A weld joint was fabricated for evaluating residual stress and
fatigue strength. Two attachments were ﬁlletwelded on both sides
of a main plate as shown in Fig. 1. External loading during the
fatigue test was added along the main plate. The base material was
a carbon steel (JIS SS400), and the welding material is nearly the
same. Chemical compositions of the materials, their mechanical
properties, and the welding conditions are listed in Tables 1, 2, and
3, respectively. Two types of weld joints were fabricated by
changing welding pass sequences, as shown in Fig. 2. One was a
weld joint whose ﬁnal welding pass in each quadrant was depos
ited at the attachment side, and the other was a joint whose ﬁnal
pass was deposited on the main plate side. The weld joints were
made to have the same shape in the weld toe which is subjected to
stress concentration.
Stress concentration factor ␣ was calculated using an elastic
ﬁniteelement analysis. The shape of the weld toe after welding
was traced and the ﬁnite elements were divided. The stress con
centration factor of the weld joint whose ﬁnal welding pass was
deposited at the main plate side was ␣ ϭ 3.76 to 4.12, and the
factor of the joint whose ﬁnal pass was at the attachment side was
1
Formerly, Hitachi, Ltd., Ibaraki, Japan.
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
September 17, 1998; revised manuscript received June 28, 1999. Associate Technical
Editor: K. S. Chan.
108 / Vol. 122, JANUARY 2000 Transactions of the ASME Copyright © 2000 by ASME
␣ ϭ 3.68 to 4.10. In the multipass ﬁllet weld joint, the effect of
the shape of the weld toe on fatigue strength can be negligible.
Residual stress and fatigue strength of a similar weld joint heat
treated after welding (two hours at 625°C) were also measured.
Relation Between Welding Sequence and Residual
Stress
Method of Residual Stress Analysis and Measurement. Re
sidual stresses in the two multipass ﬁllet weld joints were mea
sured by using straingauges (Lu, 1996) and Xray diffraction (The
Society of Materials Science, Japan, 1981). First, residual stresses
on the center of the weld toe and the base metal were non
destructively measured by Xray diffraction. Irradiation range of
Xrays was 5mm width for the welding longitudinal direction and
1.5mm length for the transverse direction (Park et al., 1994).
Next, strain gauges were pasted on the specimen surface in the
same position as that for Xray diffraction. Residual stress was
calculated from the released elastic strains measured when the
weld joint was cut into small pieces from around the strain gauges.
Welding residual stresses were also calculated by thermal
elasticplastic analysis using the ﬁnite element method for the two
welding pass sequences in a multipass ﬁllet weld joint. The
temperature distribution at each time resulting from the heat input
of welding was computed by transient heat conduction analysis
during the welding procedure. Transient stress during a welding
step was successively calculated from thermal loads obtained by
heat conduction analysis. This calculation involved thermal
elasticplastic analysis with nonlinearity of materials. Residual
stress can be determined as the stress when the material is cooled
down to room temperature after welding. The details of the ana
lytical method and material properties are described in the previ
ous papers by the authors (1993, 1995).
Results of Residual Stress Analysis and Measurement.
Transverse residual stresses
Ry
, which is perpendicular from
longitudinal welding direction, are compared in Figs. 3 and 4 for
two kinds of specimens, respectively. These stresses were obtained
by measurements and numerical analysis. The measurement is the
average stress of the four measured values from each quadrant.
Measurements and analytical distributions of residual stress agree
well for two multipass ﬁllet weld joints. Therefore, the result from
thermal elasticplastic analysis was used as a standard for residual
stress used for evaluating fatigue strength.
Transverse residual stresses
Ry
in the weld toe on the main
plate are 170 MPa, for the weld joint whose ﬁnal welding pass was
deposited on the attachment side, and 80 MPa for the joint whose
ﬁnal pass was on the main plate side. Transverse residual stress
Ry
Fig. 1 Conﬁgurations of a multipass ﬁllet weld joint (Units: mm)
Table 1 Chemical composition of the materials used
Table 2 Mechanical properties of the materials used
Table 3 Welding conditions of a multipass ﬁllet weld joint
Fig. 2 Sequences of welding pass in multipass ﬁllet weld joints
Fig. 3 Comparison of residual stress in a multipass ﬁllet weld joint (last
pass on the attachment)
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 109
on the weld toe in a multipass ﬁllet weld joint can be changed by
controlling the laminating location and sequence of the welding
pass.
The measured value of transverse residual stress at the weld toe
was Ϫ5 MPa in a ﬁllet weld joint, which was heattreated for stress
relief after welding, and in this paper it was treated as zero MPa.
Production of Initial Residual Stress Caused by Laminating
of Welding Pass. The generation of residual stress on the weld
toe of the main plate side is shown in Figs. 5 and 6 for the two
types of weld specimens, respectively. These histories of residual
stress at the end of each welding pass were obtained from thermal
elasticplastic analysis. Numbers on the horizontal axis show the
welding pass sequence. Each transverse residual stress
Ry
on the
weld toe of the main plate is considered to generate by the
following mechanism.
In the weld joint whose ﬁnal welding pass was deposited on the
attachment side,
Ry
slightly increases from the ﬁrst to the third
welding passes. Then, higher
Ry
is added by the fourth welding
pass because the deposit on the fourth welding pass shrinks itself
and the weld toe of the main plate is pulled by the fourth pass. On
the other hand, in the weld joint whose ﬁnal pass was on the main
plate side,
Ry
also increases from the ﬁrst to the third welding
passes. Then, residual stress present up to the third welding pass is
completely released because the weld toe is heated over the melt
ing temperature. Residual stress due to the fourth welding pass is
regenerated after the stress relaxation. As a result,
Ry
on the weld
toe of the main plate becomes smaller than that of the attachment.
Maximum residual stress of welding longitudinal direction
Rx
is
generated during the last welding pass because longitudinal resid
ual stress is caused by the shrinkage itself. Therefore,
Rx
on the
weld toe of the main plate is larger when the ﬁnal pass was
deposited on the attachment side than when the ﬁnal pass was
deposited on the main plate side.
Residual stresses in two kinds of multipass ﬁllet weld joints
become different as a results of the different production of residual
stress. And the sequence of welding passes affects residual stress
in a multipass weld joint.
Effect of Residual Stress on Fatigue Strength
Method of Fatigue Test. The fatigue test was done by apply
ing a tensile loading to the main plate with a stress ratio R ϭ 0 for
two multipass ﬁllet weld joints. The maximum force of the test
machine was Ϯ3 MN and the frequency was 8 to 9.5 Hz.
Results of Fatigue Test. The fatigue strength curves in two
multipass ﬁllet weld joint are compared in Fig. 7. The vertical axis
shows the nominal stress range ⌬
y
along the loading direction
and the horizontal axis shows the number of cycles to failure N
f
.
It was conﬁrmed from the observation during the fatigue test that
the initial surface crack nucleated at the center of the weld toe and
it propagated as a semielliptical crack.
The fatigue strength of the two kinds of multipass ﬁllet weld
joints are nearly the same in the high stress range. A clear differ
ence appears around the fatigue cycle of 1 ϫ 10
5
cycles. These
results show that high cycle fatigue strength can be improved by
varying the sequence of welding passes.
The results of a fatigue test using the ﬁllet weld joints after heat
treatment for stress relief are also shown in Fig. 7. The fatigue
strength at 1 ϫ 10
7
cycles for the weld joint in which residual
stress can be negligible is higher than the joint in which residual
stress exists. That is, the fatigue strength of the ﬁllet weld joint
without residual stress is higher than that of the joint with residual
stress.
Relation Between ResidualStress Relaxation Behavior and
Fatigue Strength. The effect of the residual stress relaxation on
the fatigue strength in a multipass ﬁllet weld joint is discussed.
The change of residual stress in the weld joint was measured by
Xray diffraction for every fatigue cycles of 10
n
(n ϭ 0, 1,
Fig. 4 Comparison of residual stress in a multipass ﬁllet weld joint (last
pass on the main plate)
Fig. 5 History of residual stress at the weld toe in a multipass ﬁllet weld
joints by multipass welding (last pass on the attachment)
Fig. 6 History of residual stress at the weld toe in a multipass ﬁllet weld
joint by multipass welding (last pass on the main plate)
110 / Vol. 122, JANUARY 2000 Transactions of the ASME
2, . . . , 7). The relaxation behavior of transverse residual stress
Ry
at the weld toe of the main plate was evaluated.
Stress relaxation behaviors of
Ry
at the weld toe of the main
plate are shown in Figs. 8 and 9 for the two weld specimens,
respectively. These ﬁgures show that the weld joint whose last
welding pass was deposited on the attachment side and the joint
whose last pass was on the main plate side exhibit similar tenden
cies for stress relaxation. Initial transverse residual stress
Ry
is
tensile, and gradually decreases when the stress range is large.
Residual stress becomes zero near the 1 ϫ10
4
cycles. On the other
hand, residual stress hardly changes when the stress range is small.
In addition, residual stress rapidly decreases after crack initiation
and propagation at high fatigue cycles.
The presence of plastic deformation in the weld toe is discussed
next. The maximum transverse stress at the weld toe
y ⅐ max
is
assumed to be elastically equal to the sum of
Ry
and ␣ ⅐ ⌬
y
,
where
Ry
is the transverse residual stress at the weld toe of the
main plate, ␣ is stress concentration factor, and ⌬
y
is nominal
stress range. The yield stress of the material and
y ⅐ max
are com
pared in the following. Plastic deformation at the weld toe is
greatly generated when the stress range is large and residual stress
is gradually released. On the other hand, plastic deformation does
not occur in the small stress range in which residual stress is hardly
relaxed. It is thought that the relaxation behavior of residual stress
is determined by the presence of plastic deformation in the weld
toe during fatigue cycling.
From these results of residual stress relaxation, it can be con
cluded that at the high stress range, the magnitude of residual stress
does not affect the fatigue strength because residual stress is
relaxed at the low fatigue cycles. At the low stress range, fatigue
strength depends on the initial residual stress because residual
stress hardly changes by fatigue loading. The results of the fatigue
test support these conclusions on residual stress relaxation during
fatigue cycling.
Discussion
Figure 10 shows a fatigue limit diagram, which relates mean
stress and stress amplitude, obtained from three types of multipass
ﬁllet weld joints. Residual stress was treated as a part of mean
stress and was added to the stress by external loading. The fatigue
strength at 1 ϫ10
7
cycles was used for the fatigue limit. A straight
line in the ﬁgure represents the modiﬁed Goodman diagram. The
effects of residual stress on fatigue strength in a ﬁlletweld joint
are shown by the modiﬁed Goodman diagram when residual stress
is treated as mean stress.
The fatigue limit diagram for a simple rectangular plate made of
the same material is also shown in Fig. 10. The fatigue limit was
obtained from the fatigue strength at 1 ϫ 10
7
cycles when the
mean stress is zero. From this ﬁgure, fatigue strength reduction
factor  ϭ 3.53 is calculated. Factor  is smaller than stress
concentration factor ␣.
It is commonly considered that the effect of residual stress on
fatigue strength can be evaluated by treating the residual stress as
a mean stress. From the results of this paper, the commonly held
Fig. 7 Relation between nominal stress range and fatigue life in multi
pass ﬁllet weld joints
Fig. 8 Residual stress relaxation by stress cycles at the weld toe in a
multipass ﬁllet weld joint (last pass on the attachment)
Fig. 9 Residual stress relaxation by stress cycles at the weld toe in a
multipass ﬁllet weld joint (last pass on the main plate)
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 111
view can be validated even when the object to be evaluated is
complicated such as a multipass ﬁllet weld joint.
The method for improving fatigue strength was developed by
optimizing the sequence of welding passes and by reducing resid
ual stress at the critical position. Rapid evaluation of welding
residual stress will help expand the application of the proposed
method.
Summary
The effects of residual stress on fatigue strength at a weld toe in
a multipass ﬁllet weld joint were evaluated. The residual stresses
in the specimen were varied by controlling the sequence of weld
ing passes. Residual stress calculated by thermal elasticplastic
analysis agreed well with the measurements obtained by strain
gauge and Xray diffraction methods. The residual stress at the
weld toe was 80 MPa in the specimen whose last welding pass was
on the main plate, but it was 170 MPa in the specimen whose last
pass was on the attachment. The fatigue strength of the specimens
was nearly the same at high stress amplitude, but the fatigue
strength of the specimen whose last welding pass on the main plate
was higher than that of the other specimen at low stress amplitude.
This difference is due to the magnitude of the initial residual stress
and the relaxation of the residual stress under the fatigue cycling.
The effects of the residual stress were shown in the modiﬁed
Goodman diagram, in which residual stress is treated as a mean
stress. The method for improving fatigue strength was proposed by
optimizing the sequence of welding passes and reducing residual
stress.
References
Almen, J. O., and Black, P. H., 1963, Residual Stresses and Fatigue in Metals,
McGrawHill, NY.
Beghini, M., Bertini, L., and Vitale, E., 1994, “Fatigue Crack Growth in Residual
Stress Fields, Experimental Results and Modelling,” Fatigue and Fracture for Engi
neering Materials and Structure, Vol. 17, pp. 1433–1444.
Buchalet, C. B., and Bamford, W. H., 1976, “Stress Intensity Factor Solutions for
Continuous Surface Flaws in Reactor Pressure Vessels,” Mechanics of Crack Growth,
ASTM STP, Vol. 590, pp. 385–402.
Hudak, Jr. S. J., Burnside, O. H., and Chan, K. S., 1985, “Analysis of Corrosion
Fatigue Crack Growth in Welded Tubular Joints,” ASME Journal of Energy Re
sources Technology, Vol. 107, pp. 212–219.
Glinka, G., 1979, “Effect of Residual Stress on Fatigue Crack Growth in Steel
Weldments under Constant and Variable Amplitude Loads,” Fracture Mechanics,
ASTM STP, Vol. 677, pp. 198–214.
Lu, J., Edited, 1996, Handbook of Measurement of Residual Stresses, The Fairmont
Press, Inc.
Maddox, S. J., 1991, Fatigue Strength of Weld Structures, Abington Publishing,
Cambridge, pp. 152–154.
Mochizuki, M., Enomoto, K., Okamoto, N., Saito, H., and Hayashi, E., 1993,
“Welding Residual Stresses at the Intersection of a Small Diameter Pipe Penetrating
a Thick Plate,” Nuclear Engineering and Design, Vol. 144, No. 3, pp. 439–447.
Mochizuki, M., Saito, N., Enomoto, K., Sakata, S., and Saito, H., 1995, “A Study
on Residual Stress of ButtWelded Plate Joint Using Inherent Strain Analysis,”
Transactions of the 13th International Conference on Structural Mechanics in Reac
tor Technology, Porto Alegre, Brazil, Vol. 2, Div. F, pp. 243–248.
Mochizuki, M., Hayashi, M., and Hattori, T., 1996, “Effect of Welding Sequence
on Residual Stress in MultiPass ButtWelded Pipe Joints,” Transactions of the Japan
Society of Mechanical Engineers (in Japanese), Vol. 62A, No. 604, pp. 2719–2725.
Nelson, D. V., 1982, “Effects of Residual Stress on Fatigue Crack Propagation,”
Residual Stress Effects in Fatigue, ASTM STP, Vol. 776, pp. 172–194.
Park, W., Kobayashi, H., Nakamura, H., and Koide, T., 1994, “Measurement and
Evaluation of Welding Residual Stresses for Austenitic Stainless Steel by Xray
Diffraction Method,” Proceedings of the Annual Meeting of JSME/MMD (in Japa
nese), Vol. B, No. 94037, pp. 70–71.
Parker, A. P., 1982, “Stress Intensity Factors, Crack Proﬁles, and Fatigue Crack
Growth Rates in Residual Stress Fields,” Residual Stress Effects in Fatigue, ASTM
STP, Vol. 776, pp. 13–31.
The Society of Materials Science, Japan, Edited., 1981, Standard Method of Xray
Stress Measurement (in Japanese), Yokendo Ltd.
Todoroki, A., and Kobayashi, H., 1988, “Prediction of Fatigue Crack Growth Rate
in Residual Stress Field (Application of Superposition Technique),” Transactions of
the Japan Society of Mechanical Engineers (in Japanese), Vol. 54A, No. 497, pp.
30–37.
Yamashita, T., Hattori, T., Iida, K., Nomoto, T., and Sato, M., 1997, “Effects of
Residual Stress on Fatigue Strength of Small Diameter Welded Pipe Joint,” ASME
Journal of Pressure Vessel Technology, Vol. 119, pp. 428–434.
Fig. 10 Effect of residual stress on fatigue strength in multipass ﬁllet
weld joints using a modiﬁed Goodman diagram
112 / Vol. 122, JANUARY 2000 Transactions of the ASME
Jian Cao
Brad Kinsey
Department of Mechanical Engineering,
Northwestern University,
Evanston, IL 60208
Sara A. Solla
Department of Physics and Astronomy,
Northwestern University,
Evanston, IL 60208;
Department of Physiology,
Northwestern University Medical School,
Chicago, IL 60611
Consistent and Minimal
Springback Using a Stepped
Binder Force Trajectory and
Neural Network Control
One of the greatest challenges of manufacturing sheet metal parts is to obtain consistent
part dimensions. Springback, the elastic material recovery when the tooling is removed,
is the major cause of variations and inconsistencies in the ﬁnal part geometry. Obtaining
a consistent and desirable amount of springback is extremely difﬁcult due to the nonlinear
effects and interactions between process and material parameters. In this paper, the
exceptional ability of a neural network along with a stepped binder force trajectory to
control springback angle and maximum principal strain in a simulated channel forming
process is demonstrated. When faced with even large variations in material properties,
sheet thickness, and friction condition, our control system produces a robust ﬁnal part
shape.
Introduction
Obtaining consistent and accurate part dimensions is crucial
in today’s competitive manufacturing industry. Inconsistencies
in part dimensions slow new product launches, increase
changeover times, create difﬁculties in downstream processes,
require extra quality assurance efforts, and decrease customer
satisfaction and loyalty for the ﬁnal product. In the sheet metal
forming process, a major factor preventing accurate ﬁnal part
dimensions is springback in the material. Springback is the
geometric difference between the part in its fully loaded con
dition, i.e., conforming to the tooling geometry, and when the
part is in its unloaded, free state. For a complicated 3D part,
undesirable twist is another form of springback. The uneven
distribution of stress through the sheet thickness direction and
across the stamping in the loaded condition relaxes during
unloading, thus producing springback. Factors that affect the
amount of springback include variations in both process and
material parameters, such as friction condition, tooling geom
etry, material properties, sheet thickness, and die temperature.
Because controlling all of these variables in the manufacturing
process is nearly impossible, springback, in turn, cannot be
readily controlled. Adding to the difﬁculty is the fact that
springback is a highly nonlinear effect; therefore, simulations
and correcting methods are complicated. There has been a
tremendous amount of research interest related to springback in
recent years as is evident in proceedings of Society of Auto
motive Engineers, NUMIFORM, and NUMISHEET confer
ences.
Springback can be reduced through modiﬁcations to the forming
process. Several researchers have proposed to use a stepped binder
force trajectory to accomplish this objective (Ayres, 1984; Hishida
and Wagoner, 1993; Sunseri et al., 1996). A stepped binder force
trajectory is an instantaneous jump from a low binder force (LBF)
value to a high binder force (HBF) at a speciﬁed percentage of the
total punch displacement (PD%) (see Fig. 1). Sunseri et al. (1996)
investigated springback in the Aluminum channel forming process
shown in Fig. 2. Their work was conducted through experiments
and simulations at speciﬁc values for process and material param
eters. In a production environment, however, the amount of spring
back will deviate from the desired level due to variations in the
process. Therefore, a control system that accommodates variations
in process parameters is required.
In recent years, many research groups have investigated the use
of artiﬁcial neural networks to control sheet metal forming pro
cesses. Metal forming is an ideal candidate for neural network
control due to the nonlinear effects and interactions of the process
parameters. Cho et al. (1997) used a neural network to predict the
force in cold rolling, and Di and Thomson (1997) predicted the
wrinkling limit in square metal sheets under diagonal tension.
Among others (Elkins, 1994 and Yang et al., 1992), Forcellese et
al. (1997) used a neural network to control springback in a 60 deg
aluminum Vpunch air bending process. Their system was trained
using experimentally obtained examples consisting of ﬁve param
eters from the punch force trajectory, an offline measurement of
sheet thickness, and the target bend angle as the inputs into the
neural network and the punch displacement as the output. In
another research project, Rufﬁni and Cao (1998) proposed to use
a neural network to control springback angle in a channel forming
process with punch force trajectory as the sole source for identi
fying the process variations and adjusting the HBF used in Sunseri
et al. (1996). Preliminary results showed this approach to be
promising.
In this paper, the springback of an aluminum channel is con
trolled via a stepped binder force trajectory and neural network
control. The neural network determines the HBF and PD% of the
stepped binder force trajectory. Punch force trajectory was iden
tiﬁed as the key parameter that reﬂects variations in material
properties, sheet thickness, and friction coefﬁcient. Therefore, four
polynomial coefﬁcients from curve ﬁtting the punch force trajec
tory were used as inputs into the neural network. Figure 3 shows
a ﬂowchart of our proposed control system for this application.
Despite large variations in material properties, sheet thickness (t),
and friction coefﬁcient (), the springback angle () was main
tained between 0.2 and 0.6 deg and the maximum strain (⑀) was
limited to between 8% and 10%. Finally, a comparison with the
closedloop control method proposed by Sunseri et al. (1996) is
included to show the beneﬁts of our control method. While only
numerical simulation results are presented here, the control system
will be physically implementation in the future to verify improve
ment claims.
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division June
20, 1998; revised manuscript received June 14, 1999. Associate Technical Editor:
N. Chandra.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 113
Copyright © 2000 by ASME
Channel Forming Process
A simple geometry to investigate springback is a channel form
ing process. Therefore, the same aluminum channel forming pro
cess used by Sunseri et al. (1996) is employed here (see Fig. 2).
First, the effect of constant binder force (CBF) on springback was
evaluated. As the CBF was increased while all other process and
material parameters were held constant in our simulations, the
springback angle, , was reduced as shown graphically in Fig. 4
and physically in Fig. 5. However, the increased binder force
caused a subsequent increase in the maximum strain
1
in the ma
terial, solid line in Fig. 4, to levels that exceed the maximum
stretchability of aluminum (Graf and Hosford, 1993). By utilizing
a stepped binder force trajectory, moderate maximum strain levels
in the sidewall were obtained while reducing springback in the
process as demonstrated in the table of Fig. 1.
To produce a robust process when faced with deviations in the
friction coefﬁcient, Sunseri et al. (1996) implemented closedloop
variable binder force control to follow the punch force trajectory
obtained from the stepped binder force case with nominal process
conditions. This control method was able to produce consistent
springback levels when the friction coefﬁcient was varied from a
value of 0.1, for the nominal case, to 0.25. However, whether this
methodology could withstand variations of other parameters such
as material properties and sheet thickness was not determined.
Finite Element Model. A commercial Finite Element Analy
sis package (ABAQUS, 1997) was used for our numerical simu
lations of the channel forming process. Since the problem is close
to a planestrain condition and is symmetric, only one sixteenth of
the width and half of the length of the entire blank (220 mm ϫ 46
mm) was modelled. The binder, the die, and the punch were
modelled as three separate rigid surfaces. Each surface was mod
elled using fournode interface elements (ABAQUS type IRS4),
and a Coulomb friction law was assumed. Our blank mesh had an
uneven distribution of 40 fournode, reduced integration shell
elements (ABAQUS type S4R) with a more dense concentration of
elements where the blank contacted the punch and binder corner
radii. Boundary conditions were speciﬁed to create a planestrain
condition. The material was modelled to be isotropic, elastoplastic
following the von Mises yield criterion and isotropic strain hard
ening. The elastic properties were the Young’s modulus, E, of 70
GPa and Poisson’s ratio, , of 0.3. The plastic behavior of the sheet
material was modelled using a power law relation ( ϭ K⑀
n
). Our
nominal material, denoted Material 1, had a material strength
coefﬁcient, K, of 528 MPa and a strain hardening exponent, n, of
0.265.
Proposed Control System
In a channel forming process, springback in general is extremely
sensitive to variations in material and forming parameters. In this
work, we develop a methodology for controlling springback while
producing an acceptable amount of maximum strain in the material
through a combination of a stepped binder force trajectory and
neural network control. In a stepped binder force trajectory, two
critical values need to be determined, the magnitude of the HBF
and the PD% of the total punch displacement. These two param
eters are the outputs from our neural network. From our previous
research experience in sheet metal forming process control
(Rufﬁni and Cao, 1998; Kinsey and Cao, 1997; and Sunseri et al.,
1996), the punch force trajectory was selected as the parameter that
provides information about the current process. Therefore, poly
nomial coefﬁcients from curve ﬁtting the punch force trajectory
are used as inputs into the neural network.
A ﬂowchart of our proposed control system is shown in Fig. 3.
The forming process would proceed normally at a constant initial
binder force, 16 kN, to a depth of 10 mm. While the punch
displacement continues, the polynomial coefﬁcients from curve
ﬁtting of the punch force trajectory are calculated and fed into a
neural network. The outputs from the neural network, the HBF and
PD% for the stepped binder force trajectory, would be obtained in
time to make the appropriate HBF adjustment at the speciﬁed
punch location, 19 mm multiplied by the PD%.
The value of 10 mm, where the coefﬁcients from the punch
force trajectory are calculated, was chosen for two reasons. At this
distance, the punch force trajectory is well deﬁned and enough data
points are available for accurate curve ﬁtting to occur. Figure 6
shows the effect variations in t and have on the punch force
trajectory from 0 to 10 mm. Secondly, the distance of 10 mm
allows adequate CPU time for calculations to be conducted. As
suming a punch speed of 50 mm/sec and setting the minimum
PD% as 57.5% of the total punch displacement, 10.925 mm,
approximately 18.5 ms are available to compute inputs into the
neural network and predict the HBF and PD% of the stepped
binder force trajectory. Curve ﬁtting and neural network programs
on our Pentium II, 233 MHz computer required approximately 10
1
In this work, maximum strain refers to the maximum principal logarithmic strain
in the channel.
Nomencl at ure
CBF ϭ constant binder force
LBF ϭ low binder force
HBF ϭ high binder force
PD% ϭ punch displacement percentage
t ϭ sheet thickness
ϭ friction coefﬁcient
ϭ springback angle
⑀ ϭ maximum principal strain
u ϭ displacement of node in ﬁnite ele
ment model
ϭ angular rotation of node in ﬁnite
element model
E ϭ Young’s modulus
ϭ Poisson’s ratio
ϭ true stress
K ϭ material strength coefﬁcient
n ϭ strain hardening exponent
k
i
ϭ integral gain
k
p
ϭ proportional gain
Fig. 1 Stepped binder force trajectory
Fig. 2 Channel forming geometry
114 / Vol. 122, JANUARY 2000 Transactions of the ASME
ms to compute values. How close to a true step function is
produced will depend on the speed of the binder force varying
mechanism.
The punch force trajectory was broken up into three regions,
Region A, a transition region, and Region B (see Fig. 6). The
groups of lines in Region A correspond to varying values of sheet
thickness, and a seconddegree polynomial was used to ﬁt the data
in this region. Despite the fact that it was a second degree poly
nomial, only two input coefﬁcients were used to characterize this
region since the punch force trajectory nearly passed through the
origin. In Region B, the dissimilar slopes are mainly caused by
changes in the friction coefﬁcient. A linear interpolation provided
two additional inputs for the network. Thus, a total of four poly
nomial coefﬁcients were used as inputs into our neural network.
Note that the punch force trajectories are obtained from numer
ical simulations; therefore, they are smooth and have consistent
patterns due to variations in t and . In an actual forming process,
noise in the data acquisition equipment would produce disparities
in these curves. By using polynomial coefﬁcients from curve
ﬁtting the punch force trajectory, the inputs into the neural network
are in essence ﬁltered to account for these practical considerations.
To implement this control system in an actual forming opera
tion, training data would need to be produced. This information
could be derived from actual production of stampings during die
try out by varying process parameters. For instance, to account for
batch to batch variation in the material, material could be obtained
and formed which represents the full range of potential sheet
thicknesses and material properties. Also, the lubrication state
could be varied. Using various combinations of these process
parameters, the HBF and PD% values, which produce the desired
amount of springback, could be determined. In addition, numerical
simulations, calibrated to experimental results to assure accuracy,
could be used to expeditiously increase the amount of training
data. Once the neural network is well trained with the entire range
of potential process parameter values, actual values of the material
properties, sheet thickness, and friction state are not required since
the inputs into the neural network are polynomial coefﬁcients from
punch force trajectory curve ﬁtting.
Any increasing binder force trajectory could be implemented to
reduce springback. However, the simplicity of the stepped binder
force trajectory requires only two input parameters, HBF and
PD%. In addition, existing presses in industry have the ability to
produce a stepped binder force trajectory. Therefore, a stepped
binder force trajectory is an ideal choice.
Neural Networks. Artiﬁcial neural networks have been stud
ied for many years in the hope of mimicking the human brain’s
ability to solve problems that are ambiguous and require a large
amount of processing. Human brains accomplish this data process
ing by utilizing massive parallelism, with millions of neurons
working together to solve complicated problems. Similarly, artiﬁ
cial neural network models consist of many computational ele
ments, called “neurons” to correspond to their biological counter
parts, operating in parallel and connected by links with variable
weights. These weights are adapted during the training process,
most commonly through the backpropagation algorithm (Rumel
hart and McClelland, 1986), by presenting the neural network with
examples of inputoutput pairs exhibiting the relationship the
network is attempting to learn. The goal is for the neural network
to generalize, or extract, the pattern given in the inputoutput
examples. Further details on neural networks, in general, can be
found in Widrow and Lehr (1990). For our particular application,
the structure of the neural network was determined to be four input
parameters, ﬁve hidden neurons, and two outputs. A sigmoidal
activation function was used for the hidden neurons while the
outputs utilized a linear one. A more detailed discussion on how
this optimal structure was determined is given in Kinsey (1998).
To initially determine the feasibility of using a neural network to
control and minimize springback, the ability of the neural network
to handle large variations in the t and was investigated. Sheet
thickness values from 0.8 to 1.4 mm (in increments of 0.1 mm
between 0.8 and 1.2 mm) and friction coefﬁcient levels from 0.04
to 0.20 (in increments of 0.01 between 0.04 and 0.12 and after
wards in increments of 0.02) were considered. These variations in
t and values are obviously larger than what would be seen in an
actual mass production forming process but are used here for
Fig. 4 Springback angle and maximum strain versus constant binder
force
Fig. 5 Final part shape with increasing constant binder force
Fig. 6 Punch force trajectory for 0 to 10 mm of punch travel
Fig. 3 Flowchart of proposed neural network control system
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 115
demonstration purposes to show the capability of neural networks
to control even considerable changes in variables. Training data
was generated through trial and error simulations for 104 combi
nations of these two process parameters to determine the HBF and
PD% values which produced a springback angle, , in the range of
0.4 to 0.5 deg and a maximum strain, ⑀, in the range of 8% to 10%.
This extremely narrow springback range was intentionally chosen
with the realization that the predicted values for the HBF and PD%
from the trained network may not result in the springback amount
in the same tight range. A given HBF and PD% combination was
not necessarily the only one that would have provided the values
for and ⑀ in the speciﬁed ranges. However, by choosing the and
⑀ ranges intentionally small, we guaranteed that the network would
receive training data that was within a narrow window of possible
HBF and PD% values. These 104 simulation runs provided the
inputoutput pairs for the training data with four curve ﬁtting
polynomial coefﬁcients from the punch force trajectories serving
as the inputs and the HBF and PD% representing the desired
outputs.
Results
Once the neural network was well trained, four t and combi
nations left out of the training set as well as four additional t and
combinations not included in the previous range of t and
values were “fedforward” in the network to predict HBF and PD%
values for the stepped binder force trajectory following the proce
dure outlined in Fig. 3 and the previous section. The resulting
springback angle, , and maximum strain, ⑀, from the process were
then calculated. Table 1 shows the excellent results that were
obtained for these eight cases. All of the springback angles and
maximum strain values are within a range of 0.3 to 0.6 deg and 8%
to 10%, respectively.
The ability of the neural network to provide HBF and PD%
values to use when faced with large variations in t and demon
strates the potential of neural networks in this application. How
ever, there are other parameters that vary during the process, which
have a similar effect on ﬁnal part shape. Material properties, for
instance, have been shown to cause severe dimensional variations
in sheet metal stampings (Kinsey and Cao, 1997). When the
original trained network for was used to predict HBF and PD%
values with different material properties, unacceptable levels of
springback and maximum strain were produced. This is not sur
prising since the network was not trained to accommodate discrep
ancies in material properties. Therefore, additional training data
incorporating material variations was required. Deviations on the
true stressstrain curve for the nominal material, Material 1, were
created by varying the strength coefﬁcient, K, by Ϯ10% (Materials
2 and 3) and Ϯ20% (Materials 6 and 7) and the strain hardening
component, n, by approximately Ϯ16% (Materials 4 and 5).
Seven combinations of t and (t ϭ 0.9/ ϭ 0.04, t ϭ
1.1/ ϭ 0.06, t ϭ 1.4/ ϭ 0.08, t ϭ 1.0/ ϭ 0.10, t ϭ
0.8/ ϭ 0.12, t ϭ 1.0/ ϭ 0.16, and t ϭ 1.2/ ϭ 0.20) were
chosen from which training examples with the new materials were
created. Again through trial and error, 36 training examples with a
given material, t, and combination were produced that had the
same ranges for and ⑀ as were previously used, 0.4 to 0.5 deg and
8% to 10% respectively. One of the six new materials at each of
the t and combinations, was left out of the training set to be used
as a check to see if the neural network was able to predict accurate
values of HBF and PD%. After these additional training examples
with variations in material properties were added to the original
training set, the network was retrained with the same neural
network structure used previously.
Table 2 shows the results when the cases from the six combi
nations that were left out of the training set and four new material
thickness, t, and friction coefﬁcient, , combinations were fedfor
ward in the newly trained network. Once again, the neural network
was able to provide values of HBF and PD% that produced
acceptable levels of , 0.2 to 0.6 deg, and ⑀, 8% to 10%. However,
these simulation outputs were not always within the narrow spring
back angle range of 0.3 to 0.6 deg as was obtained in the t and
results. This indicates that the neural network has more difﬁculty
generalizing over variations in material. As previously stated, the
narrow range on during training was intentionally created to
allow for reasonable discrepancies in the feedforward process.
Furthermore, additional training sets or a more clever neural net
work structure, such as giving “hints” to the network about the
material properties, could improve the ability of the neural network
to handle variations in material properties.
Comparison of Results to ClosedLoop Control
The control system proposed by Sunseri et al. (1996) utilized
closedloop control of the binder force to follow the punch force
trajectory in order to control springback in the same channel
forming process. Regarding the friction coefﬁcient as possibly the
most signiﬁcant process parameter, the robustness of their control
system was tested against variations in the friction coefﬁcient, and
excellent results were obtained. However, variations in the mate
rial properties and sheet thickness were not investigated in their
work. Therefore, further closedloop control simulations with
these variations were conducted here in order to form a compari
son with the neural network control system.
First, the punch force trajectory from our nominal case, Material
1, t of 1.0 mm, and of 0.10, was created. A Proportional plus
Integral (PI) controller was used to adjust the binder force to allow
for following of the nominal punch force trajectory, as was done in
Sunseri et al. (1996). The maximum punch displacement between
adjustments of the binder force was 0.02 mm. Listed in Table 3 are
the proportional gains, k
p
, and the integral gains, k
i
, used by the
controller for the three cases investigated. Figure 7 shows how
well the punch force trajectory for these three cases (k
i
ϭ 0.8 for
Material 4) tracked the nominal punch force trajectory. Figure 8
shows how the binder force varied over the punch displacement to
Table 1 Results of variation in thickness and friction coefﬁcient Table 2 Results of variation in material properties, thickness, and fric
tion coefﬁcient
Table 3 Comparison of results from neural network and closedloop
control systems
116 / Vol. 122, JANUARY 2000 Transactions of the ASME
allow for following the nominal punch force trajectory (k
i
ϭ 0.8
for Material 4). Note that the binder force trajectories level off at
the end of the forming process similar to a stepped binder force
trajectory.
The and ⑀ values from these closedloop control experiments
are also shown in Table 3 along with the results from the neural
network control system. This table clearly indicates that the spring
back angle for the neural network control system was considerably
less and closer to the original range of 0.4 to 0.5 deg for all three
cases. Even for the case where the nominal t and values, 1.0 mm
and 0.10, respectively, were used with Material 5, the neural
network control system outperformed the closedloop control sys
tem.
Though a neural network system requires more up front work
producing a sufﬁcient number of examples to train the network, the
beneﬁts for controlling springback and maximum strain are unde
niable. Also, there are additional beneﬁts to the neural network
control system over this closedloop control strategy for this ap
plication. Note that the value of k
i
had to be adjusted for the
Material 4 case in Table 3, compared to the cases for Materials 3
and 5, in order for the punch force trajectory to be followed
accurately. That is, as the material and process parameters change,
the gains necessary to follow a punch force trajectory closely may
also change. Figure 7 also shows what would have been the actual
punch force trajectory for the Material 4 case if the k
i
from the
cases for Materials 3 and 5, 2.0, was used. Therefore, this closed
loop control system with preselected ﬁxed gains is not robust
when faced with large variations in process and material parame
ters. Furthermore, presses currently found in industry are capable
of readily producing a stepped binder force trajectory, as is used in
the neural network control system, while following a continuously
varying punch force trajectory would require a more vigorous
control system.
Conclusions
In this paper, a neural network system, along with a stepped
binder force trajectory, was proposed to control springback and
maximum strain in a simulated Aluminum channel forming pro
cess. A neural network was chosen due to its ability to handle the
highly nonlinear coupled effects that are found in metalforming
processes when variations in the material and process parameters
occur. The punch force trajectory was identiﬁed as the process
parameter that provides the greatest insight into deviations of
various material and process variables. Therefore, polynomial co
efﬁcients from curve ﬁtting of the punch force trajectory were used
as inputs into the neural network. The results show that the neural
network was successful at providing the high binder force (HBF)
and punch location (PD%) values for the stepped binder force
trajectory which produced acceptable values of springback (), 0.2
to 0.6 deg, and maximum strain (⑀), 8% to 10%, in the ﬁnal
product when faced with variations in the material strength (K) of
Ϯ20%, the strain hardening exponent (n) of Ϯ16%, the sheet
thickness (t) of Ϯ25%, and the friction coefﬁcient () of Ϯ65%.
Also when compared with the closedloop control strategy pro
posed by Sunseri et al. (1996) for the same process, the neural
network system was shown to be superior at minimizing spring
back in the presence of variations in material properties, as well as
being a more robust system to implement if adequate training data
is available.
While this work was conducted using simulations, the method
ology developed could be easily extended to an actual forming
process or experiments, which will be conduct in the future to
validate our claims. The only hardware requirements would be a
CPU, the ability to measure punch force trajectory, and the ability
to vary the binder force once during the process cycle. With this
neural network control system in place, a sheet metal forming
process would be robust to inevitable variations in material and
process parameters; thus, creating consistent ﬁnal part dimensions
that are critical to downstream processes and customer satisfaction.
Acknowledgment
This research was funded in part by NSF grants #CMS9622271
and #DMI9703249.
References
ABAQUS, 1997, User’s Manual, Theory Manual, Version 5.6, Hibbit, Karlsson,
and Sorensen, Providence, RI.
Ayres, R. A., 1984, “SHAPESET: A Process to Reduce Sidewall Curl Springback
in HighStrength Steel Rails,” Journal of Applied Metalworking, Vol. 3, No. 2, pp.
127–34.
Cho, S., Cho, Y., and Yoon, S., 1997, “Reliable Roll Force Prediction in Cold Mill
Using Multiple Neural Networks,” IEEE Transactions on Neural Networks, Vol. 8,
No. 4, pp. 874–82.
Di, S., and Thomson, P. F., 1997, “Neural Network Approach for Prediction of
Wrinkling Limit in Square Metal Sheet Under Diagonal Tension,” Journal of Testing
and Evaluation, Vol. 25, pp. 74–81.
Elkins, K. L., 1994, “OnLine Angle Control for Small Radius Air Bending,”
Carnegie Mellon University, UMI Dissertation Services, Ph.D. thesis.
Forcellese, A., Gabrielli F., and Rufﬁni, R., 1997, “Springback Control in a Air
Bending Process by Neural Network,” Submitted to Proc. III Convegno AITEM.
Graf, A., and Hosford, W., 1993, “Planestrain Tension Test of AL 2008T4
Sheets,” Proceedings SAE Congress Symposium on Sheet Forming, Detroit, MI.
Hishida, Y., and Wagoner, R., 1993, “Experimental Analysis of Blank Holding
Force Control in Sheet Forming,” Journal of Materials and Manufacturing, Vol. 2,
pp. 409–15.
Hornik, K., Stinchcombe, M., and White, H., 1989, “Multilayer Feedforward
Networks are Universal Approximators,” Neural Networks, Vol. 2, pp. 359–66.
Fig. 7 The target and actual punch force trajectories in closedloop
control
Fig. 8 Resulting binder force trajectories for closedloop control follow
ing punch force trajectory
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 117
Kinsey, B., 1998, “Process Control in Sheet Metal Forming,” Master’s thesis,
Department of Mechanical Engineering, Northwestern University.
Kinsey, B., and Cao, J., 1997, “An Experimental Study to Determine the Feasibility of
Implementing Process Control to Reduce Part Variation in a Stamping Plant,” Sheet Metal
Stamping: Development Applications, SAE Paper 970713, SP1221, pp. 107–12.
Rufﬁni, R., and Cao, J., 1998, “Using Neural Network for Springback Minimiza
tion in a Channel Forming Process,” Developments in Sheet Metal Stamping, SAE
Paper 98M154, SP1322, pp. 77–85.
Rumelhart, D. E. and McClelland, J. L., 1986, “Learning Internal Representation
by Error Propagation,” Parallel Distributed Processing, MIT Press, Cambridge, MA,
Vol. 1, Ch. 8.
Sunseri, M., Cao, J., Karaﬁllis, A. P., and Boyce, M. C., 1996, “Accommodation
of Springback Error in Channel Forming Using Active Binder Force Control: Nu
merical Simulations and Results,” ASME JOURNAL OF ENGINEERING MATERIALS AND
TECHNOLOGY, Vol. 118, No. 3, pp. 426–35.
Widrow, B. and Lher, M. A., 1990, “30 Years of Adaptive Neural Networks:
Perceptron, Madaline and Backpropagation,” Proceedings of the IEEE, Vol. 78, No.
9, pp. 1415–41.
Yang M., Shima, S., and Watanabe, T., 1992, “Development of Control System
Using Neural Network Combined with Deformation Model for an Intelligent
VBending Process of Sheet Metals,” Proceedings of Japan/USA Symposium on
Flexible Automation, ASME, Vol. 2, pp. 1485–90.
118 / Vol. 122, JANUARY 2000 Transactions of the ASME
K. T. Kim
Professor.
Assoc. Mem. ASME
email: korean@postech.edu
J. H. Cho
J. S. Kim
Department of Mechanical Engineering,
Pohang University of Science and Technology,
Pohang 790784, Korea
Cold Compaction of Composite
Powders
Densiﬁcation behavior of composite powders was investigated under cold compaction.
Experimental data were obtained for mixed copper and tungsten powders with various
volume fractions of tungsten powder under cold isostatic pressing and die compaction. A
model was also proposed for densiﬁcation of mixed—soft and hard—metal powders under
cold compaction. Theoretical predictions from the proposed model and models in the
literature were compared with experimental data. The agreements between experimental
data and theoretical predictions from the proposed model are very good for composite
powders at initial stage under cold isostatic pressing. Theoretical predictions, however,
underestimate experimental data under cold die compaction.
Introduction
P/M parts are typically produced in a sequence of cold compac
tion, sintering, and ﬁnishing. Nonuniform density distribution of
powder compacts under cold compaction has a great inﬂuence on
the subsequent process such as sintering, thus it is important to
predict densiﬁcation and density distribution of metal powder
under cold compaction. So far, the research on powder compaction
has been mainly focused on densiﬁcation of pure powders (Gur
son, 1977; Shima and Oyane, 1976; Arzt, 1982; Helle et al., 1985;
Fleck et al., 1992; Fleck, 1995; Larsson et al., 1996; Fleck et al.,
1996). Recently, it has been extended to mixed powders (Lange et
al., 1991; Besson and Evans, 1992; Gurson and McCabe, 1992;
Turner and Ashby, 1996; Jefferson et al., 1994; Besson, 1995),
since P/M has been considered as a major route for processing
composite materials.
Lange et al. (1991) investigated densiﬁcation behavior of mixed
aluminum and steel powders under cold compaction. Besson and
Evans (1992) studied densiﬁcation of copper powder reinforced
with ceramic powder such as SiC and alumina. Gurson and Mc
Cabe (1992) examined the yield function for mixed metal powders
by using data from triaxial compression test. By considering var
ious amounts, sizes and shapes of rigid inclusions, Turner and
Ashby (1996) investigated densiﬁcation behavior of mixed pow
ders under cold isostatic pressing. Jefferson et al. (1994) and
Besson (1995) theoretically studied densiﬁcation behavior of pow
ders surrounding an isolated rigid inclusion.
Recently, Bouvard (1993), Zavaliangos and Wen (1997) and
Storåkers et al. (1997) proposed theoretical models for densiﬁca
tion of mixed powders. Assuming that the matrix powder with
powerlaw creep response has a spherical shape with the same size
as the rigid inclusion, Bouvard (1993) proposed a model for
densiﬁcation of mixed powders under hydrostatic compression. By
employing the effect of the mixing quality on densiﬁcation, Zaval
iangos and Wen (1997) proposed a model for densiﬁcation of
mixed powders with perfectly plastic behaviors under cold iso
static pressing. Storåkers et al. (1997) obtained densiﬁcation mod
els for mixed spherical powders in different sizes with the same
hardening parameters and creep exponents under cold isostatic
pressing and die compaction. However, theoretical predictions
from these models did not agree well with experimental data for
composite powders under cold compaction.
The present paper reports on densiﬁcation behavior of mixed—
soft and hard—powders during cold compaction. Experimental
data were obtained for mixed—copper and tungsten—powders
under cold isostatic pressing and die compaction. Based on the
models in the literature, densiﬁcation equations were proposed for
composite powders under cold isostatic pressing and die compac
tion. In the model, soft and hard powders have a spherical shape
with the same size and different hardening parameters. Theoretical
predictions from the proposed model and models in the literature
were compared with experimental data for mixed—copper and
tungsten—powders under cold compaction.
Experiments
Materials and Specimens. Gas atomized copper powder (Py
ron, USA) with particle sizes between 150–200 m and hydrogen
reduced tungsten powder (Korea tungsten, Korea) with particle
sizes between 150–200 m were used. Figure 1 shows that the
shape of copper powder is spherical and that of tungsten powder is
polygonal.
The plastic ﬂow stress for the matrix material of copper powder
was obtained from solid copper produced in this work under
uniaxial compression test. That of tungsten powder was obtained
from experimental data by Yiu and Wang (1979).
Solid copper was produced by hot isostatic pressing as follows:
A hot isostatic press (HIP System 30T, Kobe steel, Japan) was
used to produce dense copper specimens. A copper tube was used
as a container of copper powder during hot isostatic pressing. The
container is 23 mm in inner diameter and 1 mm in thickness. The
powder in the container was degassed for 5 h at 400°C and vacuum
sealed. The sealed sample was hot isostatically pressed for 1 h
under 50 MPa at 800°C. After hot isostatic pressing, the container
was removed by machining. A specimen was machined into a
cylindrical shape with 12 mm in height and 10 mm in diameter.
Then, the specimen was annealed for 1 h at 600°C in a hydrogen
atmosphere. The ﬁnal relative density of the specimen was nearly
Ͼ 99%.
Uniaxial stressstrain response of the matrix material of copper
powder was obtained from uniaxial compression of solid copper at
room temperature by using an MTS servohydraulic testing ma
chine. To reduce the friction between the specimen and compres
sion platens, teﬂon sheets were inserted. The ﬂow stress of the
matrix material was obtained by measuring the height and diam
eter of the specimen by interrupting the test. Teﬂon sheets were
changed at each interruption.
Cold Isostatic Pressing. Copper powder was deoxidized for
45 min at 300°C in a hydrogen atmosphere. Mixtures of copper
and tungsten powders with various volume fractions of tungsten
powder were produced by using a gravity mixer for 1 h to obtain
uniform mixing. Composite powders were poured in a rubber
mold, then sealed with a latex envelope. The sealed sample was
compacted in a cold isostatic press (ABB Autoclave Engineers,
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division April
13, 1998; revised manuscript received July 16, 1999. Associate Technical Editor:
H. M. Zbib.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 119
Copyright © 2000 by ASME
USA) under hydrostatic pressures of 150–350 MPa. The compos
ite powder compact is about 13 mm in height and 10 mm in
diameter. Relative density of the compact was measured by
Archimedes’ method with parafﬁn treatment.
Cold Die Compaction. Densiﬁcation behaviors of composite
powders with various volume fractions of tungsten powder were
investigated under cold die compaction. Composite powders of
10 g was poured in a closed die and compacted under pressure of
150–650 MPa. To reduce the friction between the powder and die
wall, zinc stearate was sprayed on the die wall. Relative density of
the compact was measured by Archimedes’ method with parafﬁn
treatment after the compact was ejected from the die.
Analysis
This paper reports on densiﬁcation behavior of composite pow
ders under cold compaction. The densiﬁcation equations were
obtained for mixture of soft and hard powders under cold isostatic
pressing and cold die pressing. For simplicity, powders are as
sumed to have the same radius R with powerlaw hardening
parameters n
s
for soft powder and n
H
for hard powder. Also, both
powders are assumed to have a spherical shape. Thus, strain
hardening responses of the matrix material for soft and hard
powders can be written, respectively,
⑀
mS
ϭ
ͩ
mS
oS
ͪ
nS
, ⑀
mH
ϭ
ͩ
mH
oH
ͪ
nH
(1)
where
m
, ⑀
m
and
o
, respectively, are uniaxial stress, plastic
strain and a constant for the matrix material. The subscripts S and
H represent the soft and hard powder, respectively.
Cold Isostatic Pressing. The hydrostatic pressure P for den
siﬁcation of composite powders under cold isostatic pressing can
be written (Bouvard, 1993; Zavaliangos and Wen, 1997)
P ϭ
1
3V
contact
F
͑c͒
d
͑c͒
(2)
where F
(c)
are the interparticle contact forces, d
(c)
are the center to
center distances of the particles adjacent to each contact and the
superscript c denotes the contact type.
Assuming that the powder compact with relative density D
consists of N particles, the volume of the powder compact can be
written
V ϭ
ͩ
4
3
R
3
ͪ
N
D
. (3)
For composite powders with the volume fraction of soft powder
f
S
and that of hard powder f
H
, Eq. (2) becomes
P ϭ
D
4R
3 ͫ
f
S
Z
SS
F
SS
d
SS
2
ϩ f
S
Z
SH
F
SH
d
SH
2
ϩ f
H
Z
HS
F
HS
d
HS
2
ϩ f
H
Z
HH
F
HH
d
HH
2
ͬ
(4)
where Z denotes the number of contacts per particle. Z
␣
(␣,  ϭ
S, H) are the average coordination numbers of powder contact
ing with an ␣powder. F
␣
and d
␣
are the contact forces and the
center to center distances of powder contacting with an
␣powder, respectively.
For a single type powder compact, Arzt (1982) suggested that
the number of particle centers G within radius r around a reference
particle with radius R can be expressed by
G͑r͒ ϭ Z
o
ϩ C
ͩ
r
2R
Ϫ 1
ͪ
͑r Ն 2R͒ (5)
where the initial coordination number Z
o
ϭ 7.3 and the constant
C ϭ 15.5 for random dense packing. Densiﬁcation is modelled by
the concentric growth of the spherical particles. When the spheres
grow to the radius RЈ, some of them overlap. The number of these
overlap can be written from Eq. (5). Thus, if we take the original
radius R to be 1, Z(D) can be written
Z͑D͒ ϭ G͓2RЈ͑D͔͒ ϭ Z
o
ϩ C͑RЈ Ϫ 1͒ ͑r Ͻ RЈ͒ (6)
Applying Eq. (6) for densiﬁcation of composite powders, the
coordination numbers Z
␣
for various types of contact may be
written by a relationship between RЈ and d
␣
(Bouvard, 1993;
Zavaliangos and Wen, 1997). Thus,
Z
␣
ϭ f
ͫ
Z
o
ϩ C
ͩ
2R
d
␣
Ϫ 1
ͪͬ
͑␣,  ϭ S, H͒. (7)
When the same kind of spherical powders are in contact, the
relationship between the contact force and the center to center
Fig. 1 Scanning electron micrographs of (a) Cu powder and (b) W
powder
120 / Vol. 122, JANUARY 2000 Transactions of the ASME
distance of the particles can be obtained as follows: when both
powders have the same radius with workhardening response,
Matthews (1980) obtained the relationship between the contact
force F and the circle radius a of the contact area. Thus,
F ϭ
6n
o
2n ϩ 1
ͩ
8a
9R
ͪ
1/n
͑a
2
͒ (8)
where
o
and n denote the workhardening coefﬁcient and the
reciprocal of the usual workhardening index. Matthews (1980)
also obtained a relationship between the circle radius a of the
contact area and the center to center distance d when spherical
powders with the same radius are in contact. Thus,
d ϭ 2R
ͫ
1 Ϫ
ͩ
2n
2n ϩ 1
ͪ
2͑nϩ1͒
a
2
R
2ͬ
(9)
Hence, the relationship between the contact force F and the center
to center distance d can be written from Eqs. (8) and (9). Thus,
F ϭ
6n
1Ϫ1/n
2n ϩ 1
ͩ
8
9
ͪ
1/n
ͩ
1 Ϫ
d
2R
ͪ
͑2nϩ1͒/2n
ϫ
ͩ
2n
2n ϩ 1
ͪ
͑1Ϫn͒͑2nϩ1͒/n
o
R
2
(10)
When spherical powders of different materials, for instance, soft
and hard particles are in contact, we assume that only the soft
particle deforms. Considering the contact problem of two spheres
with different diameters, for instance, D
1
ϭ ␣D
2
with a positive
constant ␣, the relationship between the contact force F and the
circle radius a of the contact area can be written from Storåkers
(1997). Thus,
F
a
2
o
ͩ
6R
a
ͪ
1/n
ϭ 3͑1 ϩ ␣͒
1/n
(11)
where
o
and n are material constants of the soft particle. The
relationship between the center to center distance d and the circle
radius a of the contact area can be obtained from the combined
boundary value problem and the result by Biwa and Storåkers
(1995) for deformation of a specimen in hardness test. Thus,
d ϭ 2R Ϫ
͑1 ϩ ␣͒a
2
2R
(12)
where can be written as a function of hardening parameter n. For
instance, ϭ 1.08 when n ϭ 4.115 for copper powder. Hence,
when ␣ ϭ 1, i.e., two spheres have the equal size, the relationship
between the contact force F and the center to center distance d can
be written from Eqs. (11) and (12). Thus,
F ϭ 6 ͩ
ͱ2
3
ͪ
1/n
͑2nϩ1͒/2n
o
R
2
ͩ1 Ϫ
d
2R
ͪ
͑2nϩ1͒/2n
(13)
Bouvard (1993) and Storåkers et al. (1997) assumed that the
contact forces are the same regardless of various types of contact.
Thus,
F
HH
ϭ F
SH
ϭ F
HS
ϭ F
SS
(14)
Zavaliangos and Wen (1997), however, obtained the different
contact forces depending on the types of contact. Thus,
F
SH
ϭ F
HS
Ϸ ͑1 ϩ 0.6f
H
͒F
SS
,
F
HH
Ϸ S
R
fH
͑2 Ϫ exp͑1 Ϫ S
R
͒͒
1ϪfH
F
SS
(15)
where S
R
(ϭ
oH
/
oS
) is the ratio of yield strengths of the soft and
hard powders. By substituting Eqs. (10) and (13) into Eq. (14) or
(15), d
SH
(ϭd
HS
) and d
HH
can be written in terms of d
SS
as shown
in Eqs. (A7)–(A8) when the contact forces are the same and in Eqs.
(A9)–(A10) when they are different.
Hence, Z
SH
(ϭZ
HS
), Z
HH
, F
SH
(ϭF
HS
) and F
HH
can also be
written in terms of d
SS
. Thus,
Z
␣
ϭ Z
␣
͑d
SS
͒ ͑␣,  ϭ S, H͒ (16)
F
␣
ϭ F
␣
͑d
SS
͒ ͑␣,  ϭ S, H͒. (17)
By using the balance of energy, a relationship for the pressure
and densiﬁcation rate can be written (Bouvard, 1993; Zavaliangos
and Wen, 1997)
P
D
˙
D
ϭ Ϫ
1
V
contact
F
͑c͒
d
˙
͑c͒
. (18)
By substituting Eq. (4) and f
H
Z
HS
ϭ f
S
Z
SH
from Eq. (7) into Eq.
(18), we can write
dD
D
ϭ Ϫ3
L
1
M
1
dd
SS
(19)
where
L
1
ϭ f
Sͩ
Z
SS
F
SS
ϩ 2Z
SH
F
SH
dd
SH
dd
SS
ͪ
ϩ f
H
Z
HH
F
HH
dd
HH
dd
SS
(20)
M
1
ϭ f
S
͑Z
SS
F
SS
d
SS
ϩ 2Z
SH
F
SH
d
SH
͒ ϩ f
H
Z
HH
F
HH
d
HH
(21)
By substituting Eqs. (16) and (17) into Eqs. (19) and (4), we can
write
dD ϭ Ϫ3D
L
1
͑d
SS
͒
M
1
͑d
SS
͒
dd
SS
(22)
P ϭ
D
4R
3
L
2
͑d
SS
͒ (23)
where
L
2
͑d
SS
͒ ϭ f
S
Z
SS
F
SS
d
SS
2
ϩ f
S
Z
SH
͑d
SS
͒F
SH
͑d
SS
͒d
SH
͑d
SS
͒
ϩ f
H
Z
HH
͑d
SS
͒F
HH
͑d
SS
͒
d
HH
͑d
SS
͒
2
. (24)
Hence, the variation of relative density with pressure for composite
powders under cold isostatic pressing can be obtained from Eqs.
(22) and (23).
Cold Die Compaction. When powder aggregates of uniform
sphere are subjected to macroscopic stresses ⌺
ik
so that powders
with their centers at a distance r(ϭrn) from the center of a
reference powder are deformed by the stretch ratio (ϭd/ 2R),
Fleck et al. (1997) showed that macroscopic stresses ⌺
ik
are related
to the contact force F given by
⌺
ik
ϭ Ϫ
RZ
o
VA
͵
A
F
I
n
i
n
k
dA
Ϫ
1
2VA
͵
A
ͭ
͵
2R
2R/
ͩ
Fr
dZ
dr
dr
ͪͮ
n
i
n
k
dA (25)
where A is the surface area of reference powder and F
I
is the initial
contact force. V and n are the aggregate volume and the unit
vector, respectively.
For composite powders, we substitute the number of powders
touching an ␣powder initially f
␣
f

Z
o
, and the number of those
during densiﬁcation f
␣
Z
␣
into Z
o
and Z in Eq. (25), respectively.
Thus, Eq. (25) can be rewritten
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 121
⌺
ik
ϭ Ϫ
RZ
o
VA
͵
A
͕
␣,ϭS,H
f
␣
f

F
I␣
␣
͖n
i
n
k
dA
Ϫ
1
VA
͵
A
ͭ
␣,ϭS,H
f
␣
␣ ͵
2R
2R/␣
ͭ
F
␣
r
dZ
␣
dr
dr
ͮͮ
ϫ n
i
n
k
dA ͑␣,  ϭ S, H͒ (26)
where F
I␣
is the initial contact force between ␣ and powders.
Fleck et al. (1997) also showed that the relationship between F
I␣
and F
␣
can be written
F
␣
ϭ
ͩ
1 Ϫ
␣
r/2R
1 Ϫ
␣
ͪ
͑2n␣ϩ1͒/2n␣
F
I␣
͑␣,  ϭ S, H͒ (27)
where F
I␣
can be obtained from Eqs. (10) and (13). Thus,
F
I␣␣
ϭ
6n
␣
2n
␣
ϩ 1
ͩ
2n
␣
2n
␣
ϩ 1
ͪ
͑1Ϫn␣ ͒͑2n␣ϩ1͒/n␣
1Ϫ͑1/n␣ ͒
ϫ
ͩ
8
9
ͪ
1/n␣
o␣
R
2
͑1 Ϫ
␣␣
͒
͑2n␣ϩ1͒/2n␣
͑␣ ϭ S, H͒ (28)
F
ISH
ϭ 6ͩ
ͱ2
3
ͪ
1/nS
͑2nSϩ1͒/2nS
oS
R
2
͑1 Ϫ
SH
͒
͑2nSϩ1͒/2nS
(29)
We now assume that the average stretch ratio
av
of composite
powders can be written as the sum of the stretch ratio for each type
of contact multiplied by the volume fractions for the corresponding
type of contact. Thus,
av
ϭ f
S
2
SS
ϩ 2f
S
f
H
SH
ϩ f
H
2
HH
. (30)
During closed die compaction, the compaction ratio ⌳(ϭh/h
o
)
can be written as the ratio of compacted height h to initial height
h
o
of a powder compact. Assuming no friction exists between the
powder and die, the relationship between the compaction ratio ⌳
and the average stretch ratio
av
can be written
av
Ϫ2
ϭ 1 ϩ ͑⌳
Ϫ2
Ϫ 1͒ cos
2
(31)
where is the angle between a given orientation n and the zaxis,
i.e., the direction of die compaction (see Fig. 1 in Storåkers et al.,
1997).
By substituting Eqs. (27)–(31) into Eq. (26), macroscopic
stresses ⌺
ik
can be written at a given compaction ratio ⌳. The
pressure ⌺
33
under die compaction can be written as Eq. (B9) in
Appendix B. Thus,
⌺
33
ϭ Ϫ
RZ
o
VA
͵
0
/2
cos
2
R
2
sin d͓G
1
͔͑͒
Ϫ
1
2VA
͵
0
/2
cos
2
R
2
sin d͓G
2
͔͑͒ (32)
where G
1
() and G
2
() are written in Eq. (B5) and (B6), respec
tively. The relationship between the compaction ratio and relative
density for composite powders with initial relative density D
o
during die compaction can be written
D ϭ
D
o
⌳
. (33)
Hence, the variation of relative density with pressure P(ϭ⌺
33
) for
composite powders under die compaction can be obtained from
Eqs. (32) and (33).
Results and Discussion
Figures 2 and 3, respectively, show uniaxial stressplastic strain
responses of solid copper and solid tungsten under uniaxial com
pression. The circular points represent experimental data. The
solid curves are obtained from Ludwik’s equation to represent
experimental data. Thus,
⑀
mS
ϭ
ͩ
mS
480
ͪ
4.115
, ⑀
mH
ϭ
ͩ
mH
2900
ͪ
3.435
(34)
where subscripts S and H denote the soft powder (copper powder)
and the hard powder (tungsten powder), respectively. The yield
strength and Young’s modulus of copper powder are 149 MPa and
110.3 GPa and those of tungsten powder are 700 MPa and 344.8
GPa, respectively.
Cold Isostatic Pressing. Figure 4 shows optical micrographs
of composite powder compacts with 40 vol% W in Cu compacted
at various hydrostatic pressures during cold isostatic pressing. The
dark areas are pores in the compact. As pressure increases, copper
powders deform markedly, but tungsten powders do not deform as
much. From deformation between tungsten powders, it is observed
that hard powders also deform by contact and do not behave as
rigid inclusions. From contacts between copper and tungsten pow
ders, it is also observed that deformation in copper powder is very
large, but that in tungsten powder is negligibly small. Hence, the
assumption we made, the soft particle deforms only when soft and
hard particles are in contact, seems to be plausible.
Figure 5 shows optical micrographs of composite powder
Fig. 2 Uniaxial stressplastic strain relation for dense copper
Fig. 3 Uniaxial stressplastic strain relation for dense tungsten
122 / Vol. 122, JANUARY 2000 Transactions of the ASME
compacts with various volume fractions of tungsten powder
under hydrostatic pressure of 250 MPa during cold isostatic
pressing. Figure 5 show that more pores are observed in com
posite powder compacts as the volume fraction of tungsten
powder increases.
Figure 6 shows comparisons between experimental data and
theoretical predictions for the variation of relative density with
Fig. 4 Optical micrographs of 40 vol% of W in Cu composite powder
compacts at various hydrostatic pressures
Fig. 5 Optical micrographs of composite powder compacts with various
vol% of W in Cu under cold isostatic pressing at 250 MPa
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 123
pressure of composite powder compacts with various vol% of W
under cold isostatic pressing. The data points represent experimen
tal data. The dashed and solid curves were obtained by using Eqs.
(22) and (23) with Eq. (14) for the same contact forces and with
Eq. (15) for different contact forces, respectively. Figure 6 shows
that the dashed curves with the same contact forces lie higher than
the solid curves with different contact forces. The calculated
results from Eqs. (22) and (23) show that relative densities increase
almost linearly as pressure increases, not as observed in experi
mental data. Densiﬁcation behavior at ﬁnal stage may not be
properly modelled without including the interference between dif
ferent types of contacts at ﬁnal stage. However, theoretical pre
dictions from Eqs. (22) and (23) with Eq. (15) for different contact
forces show reasonably good agreements with experimental data at
initial stage of compaction.
Figures 7 show comparisons between experimental data and
theoretical predictions from the models by Bouvard (1993), Zaval
iangos and Wen (1997), Storåkers et al. (1997) and the present
model with Eq. (15) for the variations of relative density with
pressure of composite powders with various vol% of W. In the
model by Bouvard (1997), we assumed that the soft powder
behaves as a workhardening material in stead of a powerlaw
creep material. The calculated results from the model by Bouvard
(1997) underestimate experimental data of composite powders,
since the model assumed that the hard powder behaves as a rigid
inclusion. The model by Zavaliangos and Wen (1997) assumed
that the particles behave as perfectlyplastic materials. Thus, the
calculated results from the model by Zavaliangos and Wen (1997)
overestimate experimental data. The model by Storåkers et al.
(1997) assumed that the same workhardening parameter for two
Fig. 7(a)
Fig. 7(b)
Fig. 7(c) Fig. 7(d)
Fig. 7 Comparisons between experimental data and theoretical predictions from various models for the variation of relative density with
hydrostatic pressure of (a) 20 (b) 40 (c) 60 (d ) 80 vol% of W in Cu under cold isostatic pressing
Fig. 6 Comparisons between experimental data and theoretical predic
tions from the proposed model for the variation of relative density with
hydrostatic pressure of composite powders with various vol% of W in Cu
under cold isostatic pressing
124 / Vol. 122, JANUARY 2000 Transactions of the ASME
different types of particles. From Eq. (34), we obtained the similar
stressstrain relations for copper and tungsten powders with the
same workhardening parameter. Thus,
⑀
mS
ϭ
ͩ
mS
490
ͪ
3.75
, ⑀
mH
ϭ
ͩ
mH
2680
ͪ
3.75
(35)
Fig. 8 Optical micrographs of 40 vol% of W in Cu composite powder
compacts at various pressures
Fig. 9 Optical micrographs of composite powder compacts with various
vol% of W in Cu under cold die compaction at 300 MPa
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 125
The calculated results from the model by Storåkers et al. (1997)
also overestimate experimental data.
Cold Die Compaction. Figures 8 show optical micrographs of
composite powder compacts with 40 vol% of W in Cu compacted
at various pressures during cold die compaction. It is observed that
composite powder compacted by cold die compaction is less
homogeneous than that by cold isostatic pressing because of de
viatoric stress. The inhomogeneity during die compaction induces
the enlarged contact area of the powders. Thus, this leads the
proposed model to underestimate the experimental data.
Figures 9 show optical micrographs of composite powder com
pacts with various vol% of W in Cu during cold die compaction
under pressure of 300 MPa. Figure 10 shows comparisons between
experimental data and theoretical predictions for the variations of
relative density with pressure of composite powders with various
vol% of W in Cu under cold die compaction. The dashed and solid
curves were obtained from Eqs. (32) and (33) with Eq. (14) for the
same contact forces and with Eq. (15) for different contact forces,
respectively. Theoretical predictions from Eqs. (32) and (33) with
Eq. (14) for the same contact forces shows reasonably good
agreements with experimental data under cold die compaction,
although we assumed that only the soft material will deform when
the different particles are in contact.
Figure 11 shows comparisons between experimental data and
theoretical predictions from the model by Storåkers et al. (1997)
and the present model for the variations of relative density with
pressure of composite powders with various vol% of W in Cu
under cold die compaction. The dashed curves were obtained from
the model by Storåkers et al. (1997). The solid curves were
obtained from Eqs. (32) and (33) with Eq. (14) for the same
contact forces and the dashdotted curves from Eqs. (32) and (33)
with Eq. (15) for different contact forces. Theoretical predictions
for the same contact forces from Eqs. (32) and (33) with Eq. (14)
show reasonably good agreements with experimental data.
Conclusions
Densiﬁcation behaviors of composite powders during cold com
paction were investigated. Experimental data were obtained for mixed
copper and tungsten powders under cold isostatic pressing and die
compaction. Densiﬁcation equations were also proposed for compos
ite powders under cold isostatic pressing and die compaction.
Experimental data were compared with theoretical predictions
from the proposed model and models in the literature. At initial
stage of compaction, theoretical predictions from the assumption
of the different contact forces agreed well with experimental data
for mixed copper and tungsten powders under cold isostatic press
ing. Theoretical results from the assumption of the same contact
forces for various types of contact are higher than that from
different contact forces under cold isostatic pressing. Theoretical
predictions from models in the literature, however, did not agree
well with experimental data under cold isostatic pressing. In die
compaction, theoretical predictions from the assumption for the
same contact forces show reasonably good agreements with ex
perimental data.
Acknowledgments
This work was ﬁnancially supported by a grant from the Korean
Science and Engineering Foundation (KOSEF) under grant no.
97110070422. We are grateful for this support.
References
Arzt, E., 1982, “The Inﬂuence of an Increasing Particle Coordination on the
Densiﬁcation of Spherical Powders,” Acta Metall., Vol. 30, pp. 1883–1890.
Besson, J., and Evans, A. G., 1992, “The Effect of Reinforcements on the Densi
ﬁcation of A Metal Powder,” Acta Metall. Mater., Vol. 40, No. 9, pp. 2247–2255.
Besson, J., 1995, “Effect of Inclusion Shape and Volume Fraction on the Densi
ﬁcation of Particulate Composites,” Mech. Mater., Vol. 19, pp. 103–117.
Biwa, S., and Storåkers, B., 1995, “An Analysis of Fully Plastic Brinell Indenta
tion,” J. Mech. Phys. Solids, Vol. 43, No. 8, pp. 1303–1333.
Bouvard, D., 1993, “Modelling The Densiﬁcation of Powder Composites by Power
Law Creep,” Acta Metall. Mater., Vol. 41, No. 5, pp. 1413–1420.
Fleck, N. A., Kuhn, L. T., and McMeeking, R. M., 1992, “Yielding of Metal
Powder Bonded by Isolated Contacts,” J. Mech. Phys. Solids, Vol. 40, No. 5, pp.
1139–1162.
Fleck, N. A., 1995, “On the Cold Compaction of Powders,” J. Mech. Phys. Solids,
Vol. 43, No. 9, pp. 1409–1431.
Fleck, N. A., Storåkers, B., and McMeeking, R. M., 1997, “The Viscoplastic
Compaction of Powders,” IUTAM Symposium on Mechanics of Granular and Porous
Materials, 15–17 July 1996, Cambridge, England, eds. N. A. Fleck and A. C. F.
Cocks, Kluwer Academic Publishers, Amsterdam, pp. 1–10.
Gurson, A. L., 1977, “Continuum Theory of Ductile Rupture by Void Nucleation
and GrowthPart 1. Yield Criteria and Flow Rules for Porous Ductile Media,” ASME
J. ENGNG. MATER. TECH., Vol. 99, pp. 2–15.
Gurson, A., and McCabe, T., 1992, “Experimental Determination of Yield Func
tions for Compaction of Blended Metal Powders,” Proc. MPIF/APMI World Con
gress on Powder Metallurgy and Particular Materials, 21–26 June, 1992, San
Francisco, California, U.S.A.
Helle, A. S., Easterling, K. E., and Ashby, M. F., 1985, “HotIsostatic Pressing
Diagrams: New Developments,” Acta Metall., Vol. 33, No. 12, pp. 2163–2174.
Jefferson, G., Bassani, J. L., Xu, J., and McMeeking, R. M., 1994, “Densiﬁcation
by Plastic Deformation around Spherical Inclusions,” Int. J. Solids Structures, Vol.
31, No. 22, pp. 3059–3077.
Lange, F. F., Atteraas, L., Zok, F., and Porter, J. R., 1991, “Deformation Consol
idation of Metal Powders Containing Steel Inclusions,” Acta Metall. Mater., Vol. 39,
No. 2, pp. 209–219.
Larsson, P.L., Biwa, S., and Storåkers, B., 1996, “Analysis of Cold and Hot
Fig. 10 Comparisons between experimental data and theoretical predic
tions from the present model for the variations of relative density with
pressure of composite powders with various vol% of W in Cu under cold
die compaction
Fig. 11 Comparisons between experimental data and theoretical predic
tions from the present model and the model by Storåkers et al. (1997) for
the variations of relative density with pressure of composite powders
with various vol% of W in Cu under cold die compaction
126 / Vol. 122, JANUARY 2000 Transactions of the ASME
Isostatic Compaction of Spherical Particles,” Acta Mater., Vol. 44, No. 9, pp.
3655–3666.
Matthews, J. R., 1980, “Indentation Hardness and Hot Pressing,” Acta Metall., Vol.
28, pp. 311–318.
Shima, S., and Oyane, M., 1976, “Plasticity Theory for Porous Metals,” Int. J.
Mech. Sci., Vol. 18, pp. 285–291.
Storåkers, B., Fleck, N. A., and McMeeking, R. M., 1999, “The Viscoplastic
Compaction of Composite Powders,” J. Mech. Phys. Solids, Vol. 47, pp. 785–815.
Storåkers, B., 1997, “Local Contact Behavior of Viscoplastic Particles,” IUTAM
Symposium on Mechanics of Granular and Porous Materials, 15–17 July 1996,
Cambridge, England, eds. N. A. Fleck and A. C. F. Cocks, Kluwer Academic
Publishers, Amsterdam, pp. 173–184.
Turner, C. D., and Ashby, M. F., 1996, “The Cold Isostatic Pressing of Composite
Powders—I. Experimental Investigations Using Model Powders,” Acta Mater., Vol.
44, No. 11, pp. 4521–4530.
Yih, W. H., and Wang, C. T., 1979, Tungsten, Plenum Press, N.Y., pp. 282–289.
Zavaliangos, A., Lam, A., and Wen, A., 1996, Metal Matrix Composites, Advances
in Powder Metallurgy and Particulate Materials, 18. T. M. Cadle, K. S. Narasimhan,
eds., MPIF, Princeton, NJ, pp. 137–148.
Zavaliangos, A., and Wen, J., 1997, “The Effects of Mixing Quality on the
Densiﬁcation of Heterogeneous Powder Mixtures by Pressure,” Proc. International
Workshop on Modelling of Metal Powder Forming Process, 21–23 July 1997,
Grenoble, France, pp. 179–188.
A P P E N D I X A
When the contact forces are assumed to be the same or different
for various types of contact, the center to center distances of the
particles adjacent to each contact can be obtained as follows:
From Eqs. (10) and (13), the contact forces F
SS
, F
SH
and F
HH
,
respectively, can be written
F
SS
͑d
SS
͒ ϭ K
SSͩ
1 Ϫ
d
SS
2R
ͪ
͑2nSϩ1͒/2nS
(A1)
F
SH
͑d
SH
͒ ϭ K
SHͩ
1 Ϫ
d
SH
2R
ͪ
͑2nSϩ1͒/2nS
(A2)
F
HH
͑d
HH
͒ ϭ K
HHͩ
1 Ϫ
d
HH
2R
ͪ
͑2nHϩ1͒/2nH
(A3)
where K
SS
, K
SH
and K
HH
are
K
SS
ϭ
6n
S
2n
S
ϩ 1
ͩ
2n
S
2n
S
ϩ 1
ͪ
͑1ϪnS ͒͑2nSϩ1͒/nS
ϫ
1Ϫ͑1/nS ͒
ͩ
8
9
ͪ
1/nS
oS
R
2
(A4)
K
SH
ϭ 6ͩ
ͱ2
3
ͪ
1/nS
͑2nSϩ1͒/2nS
oS
R
2
(A5)
K
HH
ϭ
6n
H
2n
H
ϩ 1
ͩ
2n
H
2n
H
ϩ 1
ͪ
͑1ϪnH ͒͑2nHϩ1͒/nH
ϫ
1Ϫ͑1/nH ͒
ͩ
8
9
ͪ
1/nH
oH
R
2
(A6)
In Eqs. (A1)–(A6), n
S
and n
H
are hardening parameters for soft
and hard particles, respectively.
When the contact forces are assumed to be the same regardless
various types of contact, the center to center distances for each
type of contact can be written in terms of d
SS
from Eq. (14) and
Eqs. (A1)–(A6). Thus,
d
SH
͑ϭd
HS
͒ ϭ 2R
ͩ
1 Ϫ
ͩ
K
SS
K
SH
ͪ
2nS /͑2nSϩ1͒
ͩ
1 Ϫ
d
SS
2R
ͪͪ
(A7)
d
HH
ϭ 2R
ͩ
1 Ϫ
ͩ
K
SS
K
HH
ͪ
2nH /͑2nHϩ1͒
ͩ
1 Ϫ
d
SS
2R
ͪ
͑2nSϩ1͒nH /nS ͑2nHϩ1͒
ͪ
(A8)
When the contact forces are assumed to be different for different
types of contact, the center to center distances for each type of
contact can be written in terms of d
SS
from Eq. (15) and Eqs.
(A1)–(A6). Thus,
d
SH
͑ϭd
HS
͒
ϭ 2R
ͩ
1 Ϫ
ͩ
͑1 ϩ 0.6f
H
͒
K
SS
K
SH
ͪ
2nS /͑2nSϩ1͒
ͩ
1 Ϫ
d
SS
2R
ͪͪ
(A9)
d
HH
ϭ 2R
ͩ
1 Ϫ
ͩ
S
R
fH
͑2 Ϫ exp͑1 Ϫ S
R
͒͒
1ϪfH
K
SS
K
HH
ͪ
2nH /͑2nHϩ1͒
ϫ
ͩ
1 Ϫ
d
SS
2R
ͪ
͑2nSϩ1͒nH /nS ͑2nHϩ1͒
ͪ
(A10)
A P P E N D I X B
For densiﬁcation of composite powders under cold die compac
tion, the compacting pressure ⌺
33
can be obtained as follows:
When i ϭ 3, k ϭ 3, Eq. (26) becomes
⌺
33
ϭ Ϫ
RZ
o
VA
͵
A
n
3
n
3
dA͓ f
S
2
F
ISS
SS
ϩ 2f
S
f
H
F
ISH
SH
ϩ f
H
2
F
IHH
HH
͔ Ϫ
1
2VA
͵
A
n
3
n
3
dA
ϫ
ͫ
f
S
SS ͵
2R
2R/SS
F
SS
r
dZ
SS
dr
dr ϩ f
S
SH ͵
2R
2R/SH
F
SH
r
dZ
SH
dr
dr
ϩ f
H
HS ͵
2R
2R/HS
F
HS
r
dZ
HS
dr
dr
ϩ f
H
HH ͵
2R
2R/SH
F
HH
r
dZ
HH
dr
dr
ͬ
. (B1)
The relationship between
␣
and d
␣
can be written
␣
ϭ
d
␣
2R
͑␣,  ϭ S, H͒. (B2)
When the contact forces are assumed to be the same or different
for various types of contact, Eq. (B2) can be written
␣
ϭ
␣
͑
SS
͒ ͑␣,  ϭ S, H͒ (B3)
by using Eqs. (A7) and (A8) for the same contact forces and Eqs.
(A9) and (A10) for different contact forces.
By substituting Eqs. (B3) and (27) into (B1) with dA ϭ 2R sin
d, Eq. (B1) can be written
⌺
33
ϭ Ϫ
RZ
o
VA
͵
0
/2
cos
2
R
2
sin d͓G
1
͑
SS
͔͒
Ϫ
1
2VA
͵
0
/2
cos
2
R
2
sin d͓G
2
͑
SS
͔͒ (B4)
where G
1
(
SS
) and G
2
(
SS
), respectively, are
G
1
͑
SS
͒ ϭ f
S
2
F
ISS
SS
ϩ 2f
S
f
H
F
ISH
SH
ϩ f
H
2
F
IHH
HH
, (B5)
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 127
G
2
͑
SS
͒ ϭ C
ͭ
f
S
2
F
ISS
1 Ϫ
SS
SS
ͩ
n
S
4n
S
ϩ 1
Ϫ ͑1 Ϫ
SS
͒
n
S
6n
S
ϩ 1
ͪ
ϩ 2f
S
f
H
F
ISH
1 Ϫ
SH
SH
ͩ
n
S
4n
S
ϩ 1
Ϫ ͑1 Ϫ
SH
͒
n
S
6n
S
ϩ 1
ͪ
ϩ f
H
2
F
IHH
1 Ϫ
HH
HH
ͩ
n
H
4n
H
ϩ 1
Ϫ ͑1 Ϫ
HH
͒
n
H
6n
H
ϩ 1
ͪͮ
͑B6͒
From Eqs. (30) and (B3), the average stretch
av
for composite
powders can be written
av
ϭ
av
͑
SS
͒. (B7)
By substituting Eq. (31) into Eq. (B7), we may write
av
Ϫ2
͑
SS
͒ ϭ 1 ϩ ͑⌳
Ϫ2
Ϫ 1͒ cos
2
(B8)
Now, by substituting Eq. (B8) into Eqs. (B5) and (B6), Eq. (B4)
can be rewritten
⌺
33
ϭ Ϫ
RZ
o
VA
͵
0
/2
cos
2
R
2
sin d͓G
1
͔͑͒
Ϫ
1
2VA
͵
0
/2
cos
2
R
2
sin d͓G
2
͔͑͒ (B9)
Hence, the compacting pressure ⌺
33
can be obtained for a given
compaction ratio ⌳ by integrating Eq. (B9).
128 / Vol. 122, JANUARY 2000 Transactions of the ASME
I. Zarudi
L. C. Zhang
Mem. ASME
Department of Mechanical and
Mechatronic Engineering,
The University of Sydney,
NSW 2006, Australia
email: zhang@mech.eng.usyd.edu.au
On the Limit of Surface
Integrity of Alumina by
DuctileMode Grinding
This paper investigates both experimentally and theoretically the subsurface damage in
alumina by ductilemode grinding. It found that the distribution of the fractured area on
a ground mirror surface, with the Rms roughness in the range from 30 nm to 90 nm,
depends on not only the grinding conditions but also the pores in the bulk material.
Surface pit formation is the result of interaction of abrasive grains of the grinding wheel
with pores. Thus the surface quality achievable by ductilemode grinding is limited by the
initial microstructure of a material. The investigation shows that median and radial
cracks do not appear and hence are not the cause of fracture as usually thought.
1 Introduction
High surface integrity of hard, brittle materials with perfect
mirrorlike and fracturefree surfaces of monocrystalline materials
can be achieved by ultraprecision grinding (Zarudi and Zhang,
1997a, Suzuki, 1997). It has been shown theoretically and exper
imentally that the mechanism of material removal can be purely
ductile when ﬁve independent slip and twin systems are activated
(Zarudi and Zhang, 1997b). However, in the case of grinding
polycrystals a certain amount of fractured surface areas always
appear causing the degradation of its surface integrity (Zarudi and
Zhang, 1997b, Komanduri, 1996). This leads to controversial
opinions on the nature of these fractured surface areas. For exam
ple, one has considered that the cause of such surface fracture
could be the result of dislodgments of grains (Komanduri, 1996),
or of the median and radial cracks developed in the subsurface
(Bifano et al., 1991). Recently, Tele (Tele, 1995) admitted that
manufacturing defects such as pores or voids could also contribute
to the surface integrity in machining.
The present paper aims to investigate the mechanisms of surface
integrity in grinding alumina components.
2 Experiment
Polycrystalline alumina of 99.99% purity produced (Kyocera,
Japan) with grain size of 1 m and 25 m were ground with an
ultraprecision grinder, a modiﬁed Minini Junior 90 CF CNC
M286 (measured loop stiffness ϭ 80 N/m). The grinding param
eters used are listed in Tables 1 to 3. The horizontal and vertical
grinding forces, F
t
and F
n
, were measured with a threeaxis Kistler
dynamometer (Type 9257B). At least ﬁve samples were ground
under the same grinding conditions.
The properties of a ground surface were explored by means of
High Resolution Scanning Electron Microscope (HRSEM) JSM
6000F and Atomic Force Microscope (AFM). The structure of
pores in alumina after sintering and the subsurface structure of
ground specimens were studied by means of Transmission Elec
tron Microscope (TEM) EM 430. Plan view samples were used for
examinations of the alumina structure in the initial state. Five
samples were checked for every type of alumina. Twenty areas
were randomly recorded for every sample and the surface covered
by pores was quantiﬁed by means of the image analysis system
“MD30ϩ.” Parameters of pores were determined by the method
described by Glasbey (1994).
To prepare a crosssectional view sample a ground specimen
was ﬁrst sectioned into thin slices, of 1 mm thick, perpendicular to
the ground surface (Fig. 1(a)). After that two slices were glued
together in the manner shown in Fig. 1(b) and thinned down
mechanically and then by ion milling to a thickness of 40 nm
suitable for TEM study (more details can be found in Zarudi et al.,
1996). Five crosssectional view samples were examined for every
grinding regime. The density of dislocations was determined by
intersection method (Hirsh, 1971). At least 20 ﬁelds were checked
for every grinding regime. To determine the surface area covered
by pits, AFM records were quantiﬁed by means of “MD30ϩ.” At
least 10 records were examined for every ground regime.
3 Results and Discussion
3.1 Material Porosity Before Grinding. Pores were de
tected in both types of alumina after sintering. It showed that pores
in the 25 mgrained alumina with an average diameter of 2 m
covered 5–8%
Ϯ2%
of the whole ground surface and those in the 1
mgrained alumina featuring an average diameter of 0.3 m
covered 1–2%
Ϯ0.5%
of the ground surface.
3.2 Surface Topography After Grinding. Typical topog
raphy of ground surfaces is shown in Fig. 2. Mirror surfaces were
generated under all the table speeds listed in Table 1. Grooves
could be clearly observed under AFM Fig. 2(a, c). However, some
pits were also seen (Fig. 2(b, e)). The values of Rms roughness vs
table speed for both types of alumina are shown in Fig. 3. It is clear
that the Rms roughness increases with the increase of table speed.
For the 1 mgrained alumina the Rms roughness grows from 30
nm to 50 nm when the table speed changes from 0.02 m/min to 1
m/min. For the 25 mgrained alumina, it increased from 33 nm to
88 nm and thus the effect was more pronounced. Clearly, the
surface pits must have affected the measured surface roughness
values, because in the parts without pits Rms roughness was only
from 15 nm to 20 nm.
The surface areas covered by pits (SACP) are shown in Fig. 4.
It is obvious that the SACP decreases signiﬁcantly when the table
speed of grinding decreases. In other words, SACP becomes
smaller if the nominal chip thickness in grinding is smaller. Fur
thermore, the percentage of SACP on the ground surface of the 1
mgrained alumina is always much less than that on the 25
mgrained alumina. In the range of table speed from 0.2 m/min
to 0.5 m/min, the rate of increment of SACP on the 25 mgrained
alumina is quite high.
To study the nature of fractured surfaces of SACP we used
HRSEM. The detailed topography of a typical pit can be seen in
Fig. 5. In the central part of the pit no fractured facets could be
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
October 4, 1997; revised manuscript received April 27, 1999. Associate Technical
Editor: S. A. Meguid.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 129
Copyright © 2000 by ASME
seen and its inner structure could be attributed to the crosssection
of a pore. However, the edges of the pit show some fractured
fragments. Under higher magniﬁcations (Fig. 6), cracks were eas
ily located in the vicinity of the pore edges (Fig. 6(b)). The
topography of pits suggests that their formation not by grain
dislodgment but by the interaction of abrasives of the grinding
wheel with pores inside materials.
3.3 Subsurface Structure. Before ultraprecision grinding,
the damaged subsurface layer induced by specimen preparation
(conventional grinding) was determined by means of TEM on
crosssection view samples. These initially damaged zones were
ﬁrst removed completely during the subsequent ultraprecision
grinding. Thus in all the cases below, the characterisation of the
subsurface structure were purely caused by the ultraprecision
grinding.
Figure 7 shows two distinguished regions in the subsurface of
both types of alumina. The ﬁrst region, which is immediately
Table 1 Grinding conditions
Table 2 Wheel dressing conditions
Table 3 Wheel truing conditions
Fig. 1(a)
Fig. 2(a)
Fig. 1(b)
Fig. 1 Preparation of crosssection view samples for TEM investiga
tions: (a) cutting into thin slices; (b) gluing two slices together
Fig. 2(b)
Fig. 2(c)
Fig. 2(d)
Fig. 2 Topography of surfaces after ductileregime grinding: (a) and (b)
the 1 mgrained alumina; (c) and (d) the 25 mgrained alumina; (a) and
(c) 3d image of the surface; (b) and (d) general view
130 / Vol. 122, JANUARY 2000 Transactions of the ASME
under the ground surface, is characterised by an extremely high
density of dislocations. Individual dislocations can be hardly
recognised in this region. The dislocation density exceeds 10
12
cm
Ϫ2
. In the 25 mgrained alumina, the depth of the region is
from 0.2 m to 0.4 m, depending on the table speed used. In the
1 mgrained alumina it varies from 0.1 to 0.3 m. The density of
dislocations in the second region was 10
8
to 10
9
cm
Ϫ2
(Fig. 7). This
region has a thickness of 0.2 m to 0.5 m in the 25 mgrained
alumina and of 0.1 m to 0.2 m in the 1 mgrained alumina. No
microcracks were found in the subsurface areas without pores
although 10 areas were examined under each group of grinding
conditions. This proved to great extent that in the areas without
pores, material removal was under a real ductilemode.
The crackfree behavior in the ﬁrst region with high dislocation
density can be explained by the initiation of sufﬁcient number of
twin and slip systems (Chin, 1975), because in this region grinding
activated at least ﬁve independent such systems (Zarudi and
Znang, 1997). In the second zone, however, less than ﬁve inde
pendent slip and twin systems were activated and thus microcrack
ing should be highly possible there. However, this is inconsistent
with the above experimental observations. Such phenomenon may
be elucidated by the possibility of microcracking due to the pileup
of dislocations on an active slip plane in the second zone. The
threshold of an effective resolve shear stress (
s
) for cracking at
the leading edge of the pileup can be determined by (Hagan,
1979):
s
2
ϭ
3
8
ͫ
␥
͑1 Ϫ ͒L
ͬ
(1)
where ␥ is the fracture surface energy, is shear modulus, is
Poisson’s ratio, and L is the length of pileup. If the pileup length
Fig. 3 Effect of table speed on Rms roughness of alumina with different
grain sizes
Fig. 4 Effect of table speed on surface area covered by pits
Fig. 5(a)
Fig. 6(a)
Fig. 5(b)
Fig. 5 Topography of pits: (a) the 1 mgrained alumina, (b) the 25
mgrained alumina
Fig. 6(b)
Fig. 6 Subsurface structure in the vicinity of a pit: (a) general view; (b)
a detailed view of the pore edge with microcracks
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 131
decreases,
s
increases according to the above formula, and this
may eliminate crack initiation.
Based on this and considering that the length of the pileups in the
specimens (i.e., penetration depth of dislocations in the second area
was in the range from 150 to 500 nm, we can conclude that the
pileups are not sufﬁcient for microcracking. In addition, it is worth
noting that the interaction of different slip systems was a rare event in
the second region because of the extremely low density of disloca
tions. Hence, no microcracks can be created via interaction.
From the above discussion, it is clear that the initiation of
dislocations in subsurface does not create sufﬁcient stress concen
trators for microcracking. In addition, importantly, no median or
radial cracks were found, although ﬁve crosssectional view sam
ples were examined under each grinding condition and twenty
subsurface areas. These indicate that the existing explanations
about the formation of fractured surfaces are not suitable for
understanding the SACP in ultraprecision grinding of alumina.
Our following consideration, though incomplete and very approx
imate, provides additional information about the fracture around
pore edges and the variation of SACP on the ground surfaces of
different types of alumina.
3.4 Grinding Materials Containing Pores
3.4.1 Assumptions. For simplicity, we assume that forces in
grinding are distributed equally among active abrasive grains in the
grinding zone and that all grains have an equal diameter of 1 m for
the grinding wheel used in the present study. Experimentally deter
mined forces listed in Table 4 can then be used in calculations. Using
the method shown by Zhang and Howes (1995) the number of active
grains per micron in our experiment is found to be 2.5.
In grinding a material containing pores, as shown in Fig. 8, we
encounter two cases, an abrasive grain interacting with a closed
pore in the subsurface or with an open pore on the surface. The
details of the abrasivepore interaction can be very complex.
Nonuniform loading of bending, shear, and dynamic impact all
contribute to the local fracture. However, to facilitate the analysis,
consider the bendinginduced fracture, which, we think, is one of
the major causes of fracture around a pore. Following the approach
used for impact loads (Galiev, 1996) we assume further that the
stress state in the material induced by abrasive impact is similar to
that under an equivalent static load and that the shape of a pore is
axisymmetrical about the normal of the ground surface.
3.4.2 Closed Pore in the 1 mGrained Alumina. In this
case, the layer of alumina on the top of a pore is modeled by a
clamped circular plate with a variable thickness. The crosssection
of such a plate model is shown in Fig. 9(a). At the peripheral part,
the plate has a linearly varying thickness. The extension of line AB
intersects with the top surface at centre O. The ratio b/a was taken
as 0.8.
As the average radius of pores was only 0.3 m in the 1
mgrained alumina, interacting stress from single grit is roughly
approximated by a uniform pressure over the whole plate with a
diameter of 0.3 m (equal to the average diameter of pores). The
maximum bending moment of the plate is therefore (Timoshenko
and WoinowskyKrieger, 1976):
M
max
ϭ M
0
Ϫ
P
8
ϩ ␥
1
P, (2)
where
M
0
ϭ
P͑1 ϩ ͒a
2
16
, (3)
Fig. 7(a)
Fig. 8 Ductilemode grinding of alumina containing pores
Fig. 7(b)
Fig. 7 Subsurface structure of alumina after ultraprecision grinding: (a)
the 1 mgrained alumina, (b) the 25 mgrained alumina
Table 4 Measured grinding forces
132 / Vol. 122, JANUARY 2000 Transactions of the ASME
P is the total load applied on a pore through an active abrasive
grain, is Poisson’s ratio, and a is the radius of the plate. In Eq.
(2) ␥
1
is a function of b/a and can be taken as 0.05 for b/a ϭ 0.8
(Timoshenko and WoinowskyKrieger, 1976).
3.4.3 A Closed Pore in the 25 mGrained Alumina. As the
average radius of pores in the 25 mgrained alumina is 2 m, the
diameter of the plate model is also taken to be 2 m. The
interacting stress in this case distributes only partially on the plate
bounded by a circle of radius equal to 1 m, the average diameter
of abrasive grains, as shown in Fig. 9(b).
Under this loading condition, the maximum bending moment in
the plate is determined by (Timoshenko and WoinowskyKrieger,
1976):
M
max
ϭ M
0
Ϫ
P
4
ͩ
1 Ϫ
c
2
2a
2ͪ
ϩ ␥
1
P, (4)
where
M
0
ϭ
P
4
ͫ
͑1 ϩ ͒ log
a
c
ϩ 1 Ϫ
͑1 Ϫ ͒c
2
4a
2 ͬ
. (5)
3.4.3 An Open Pore. When a pore opens partially to the
grinding surface, the problem is considered as a plate with a central
hole subjected to an interacting stress from an abrasive grain, as
shown in Fig. 9(c). The maximum bending moment in this case is
(Timoshenko and WoinowskyKrieger, 1976):
M
max
ϭ
P
4
ͫ
͑1 ϩ ͒
a
2
d
2
ϩ 1 Ϫ
ͬ
ϫ
ͫ
͑1 Ϫ ͒
ͩ
a
2
d
2
Ϫ 1
ͪ
ϩ 2͑1 ϩ ͒
a
2
d
2
log
a
d
ͬ
, (6)
where d is radius of the hole.
On the other hand, the stress intensity factor of a plate with
voids can be expressed as (Rooke and Cartwrite, 1976):
K
1
ϭ
6M
͑h Ϫ g͒
3/2
f͑g/h͒, (7)
where M is the bending moment, h is the thickness of the plate in
the central part, and g is the length of a possible imperfections.
Some imperfection always exists in ceramics, which can be grain
boundaries, faults of structure etc. To be conservative, we took the
smallest imperfection size of 15 nm by assuming that the material
is almost perfect.
The stress intensity factors obtained above can then be used to
examine the fracture behavior of alumina around pore edges.
Assume that a fracture will take place when the stress intensity
factor exceeds the corresponding fracture toughness, which is 0.7
MPa ϫ m
1/2
for the 25 mgrained alumina and 2 MPa ϫ m
1/2
for
the 1 mgrained alumina (Xu et al., 1996).
Table 5 shows the critical thickness of the layer (h
c
) above a
pore, at which fracture occurs. For instance, in grinding the 25
mgrained alumina for a closed pore, h
c
is 200 nm under the table
speeds of 0.02 m/min and 0.1 m/min. Thus microcracking will take
place if the thickness of the layer above a pore is less than 200 nm.
Generally, h
c
is smaller under lower table speeds, indicating that
fracture initiation is harder. This means that under a lower table
speed, SACP will be smaller and the overall surface roughness
may also be smaller.
In the case with the 1 mgrained alumina, h
c
is considerably
smaller and is in the range from 20 nm to 50 nm when table speed
varies. This is consistent with the fact that the microfracturing
effect of pores in the 1 mgrained alumina is much lower than
that with the 25 mgrained alumina as presented by Fig. 4.
When a pore is open, edge fracture always exists since the edge
thickness of the open pore is normally variable and is often less
than h
c
. It is necessary to emphasize that the present model only
involves bending effect. In a real case, imperfections in material
and effect of shear and dynamic impact will make fracture easier
(h
c
larger). This will in turn intensify the process of the micro
crack initiation at a pore edge.
In short, the pore effect on microcracking makes SACP larger
and limits the surface integrity level achievable in the ultra
precision grinding of polyscrystalline alumina containing
pores. Microcracking around pore edges always occurs, al
though a pure ductilemode of material removal can be
Fig. 9(a)
Fig. 9(b)
Fig. 9(c)
Fig. 9 Model for interaction between an abrasive grain and material
containing pores: (a) and (b) closed pores, (c) an open pore; (a) the 1
mgrained alumina, (b) the 25 mgrained alumina
Table 5 Pore effect on microcracking in grinding
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 133
achieved by activating sufﬁcient slip and twin systems in the
areas without pores.
4 Conclusions
1. The nature of SACP on ground surfaces is related to pores
in the bulk material.
2. The subsurface structure of alumina after ultraprecision
grinding consists of a layer with a high density of disloca
tions (more than 10
12
cm
Ϫ2
) and another with lower density
of dislocations (about 10
9
cm
Ϫ2
).
3. A real ductilemode of material removal is possible due
to existence of more than ﬁve independent slip and twin
systems in the ﬁrst layer. However, the absence of mi
crocracks in the second layer is due to the small length of
pileups.
4. No radial and lateral cracks were found in the subsurface
area of the alumina after the ultraprecision grinding. This
suggests that the formation of SACP is not related to these
cracks. The bending models of open and closed pores,
though rough, showed acceptable explanation to the varia
tion of SACP.
5. The existence of pores in alumina has an effect on the level
of the surface integrity achievable by ultraprecision grind
ing. This seems to indicate that the ultimate surface integ
rity is determined not only by the depth of cut as usually
thought, but also to certain extent, by the percentage and
size of pores.
Acknowledgments
The authors wish to thank the Australian Research Council
(ARC) for continuing support of this project. We thank also the
Electron Microscope Unit for use of its facilities.
References
Bifano, T. G., Dow, T. A., and Scattergood, R. O., 1991, “DuctileRegime
Grinding: A New Technology for Machining Brittle Materials,” ASME JOURNAL
OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 113, pp. 184–189.
Chin, G. Y., 1975, “Slip and Twinning Systems in Ceramic Crystals,” Deformation
of Ceramic Materials, R. C. Brand and R. E. Tressler, eds., pp. 25–44.
Galiev, S. U., 1996, “Experimental Observations and Discussion of Counterintui
tive Behaviour of Plates and Shallow Shells Subjected to Blast Loading,” Int. J.
Impact Eng., Vol. 18, pp. 783–802.
Glasbey, C. A., 1995, Image Analysis for the Biological Science, Chichester, New
York.
Hagan, T., 1979, “Micromechanics of Crack Nucleating During Indentation,” J.
Mat. Sci., Vol. 14, pp. 2975–2980.
Hirsh, P. B., 1971, Electron Microscopy of Thin Crystals, Butterworth, London.
Komanduri, R., 1996, “On Material Removal Mechanisms in Finishing of Ad
vanced Ceramics and Glasses,” Cirp Annals, Vol. 45, pp. 09–514.
Rooke, D. P., and Cartwrite, D. J., 1976, Compendium of StressIntensity Factors,
Her Majesty’s Stationery Ofﬁce, London.
Suzuki, H., Wajima, N., Zahmaty, M. S., Kuriyagava, T., and Syoji, K., 1997,
“Precision Grinding of a Spherical Surface Accuracy Improving by OnMachine
Measurement,” Advances in Abrasive Technology, L. C. Zhang and N. Yasunaga,
eds., World Scientiﬁc, Singapore, pp. 116–121.
Tele, R., 1995, “Ceramography of High Performance Ceramics. Part IX: Pores and
Chips,” Prakt. Metallogr, Vol. 32, pp. 440–466.
Timoshenko, S. P., and WoinowskyKrieger, S., 1976, Theory of Plates and Shells,
McGrawHill, Singapore.
Zarudi, I., and Zhang, L. C., 1997a, “Subsurface Structure Change of Silicon After
UltraPrecision Grinding,” Advances in Abrasive Technology, L. C. Zhang and N.
Yasunaga, eds., Vol. 33, World Scientiﬁc, Singapore, pp. 33–38.
Zarudi, I., and Zhang, L. C., 1997b, “Surface and Subsurface Structure of
Alumina After Ductile Mode Grinding,” Proc. of 12th ASPE Annual Meeting, pp.
267–270.
Zarudi, I., 1996, “Plastic Zone Formation in Alumina Associated with Ultra
Precision Grinding,” Proc. of the 11th ASPE, 610–613.
Zarudi, I., Zhang, L. C., and Mai, Y.W., 1996, “Subsurface Damage in Alumina
Induced by SinglePoint Scratching,” J. of Mater. Sci., Vol. 31, pp. 905–914.
Zhang, Bi, and Howes, T. D., 1995, “Subsurface Evaluation of Ground Ceramics,”
Annals of CIRP, Vol. 44, pp. 263–266.
Xu, H. H. K., Wei, L., and Jahanmir, S., 1996, “Inﬂuence of Grain Size on the
Grinding Response of Alumina,” J. Am. Ceram. Soc., Vol. 79, pp. 1307–1313.
134 / Vol. 122, JANUARY 2000 Transactions of the ASME
Weili Cheng
Berkeley Engineering And Research, Inc.,
896 Seville Place,
Fremont, CA 94539
email: weilicheng@home.com
Measurement of the Axial
Residual Stresses Using the
Initial Strain Approach
Traditionally, the unknown variables to be estimated in residual stress measurement are
based on the approximation of the stresses. In this paper an alternative approach is
presented, in which the initial strain is used as the unknown variable to be estimated in
the measurement. A useful feature of the approach is that it does not require the
measurement to be made on the original part if the stress is uniform in the axial direction.
Instead, the original residual stress can be computed once the initial strains are obtained
from a specimen removed from the original part. This approach is incorporated into two
recently developed methods: the crack compliance method and the singleslice method for
measurement of the axial stress in plane strain. Experimental validation shows that the
approach based on initial strains leads to a result that agrees very well with the analytical
prediction.
Introduction
Residual stresses in many long structural parts can be idealized
as distributed uniformly in the axial direction with little outof
plane shear stresses. Examples are rods, beams, and long butt or
ﬁllet welds between two plates. In many cases the residual stresses
arise under conditions which are not easily simulated by numerical
computation. Figure 1(A) shows schematically a section of rod and
beam while Fig. 1(B) shows a section of ﬁllet weld between two
plates. In these examples the constraint is often such that residual
stress in the zdirection would be expected to be the most severe.
Therefore, an essential aspect in assessing the integrity of these
parts is to measure the distribution of axial ( zaxis) stress.
To obtain the residual stresses, the unknown quantities to be
estimated usually correspond to the stresses or average forces
released by removing or separating material. Thus, it is commonly
believed that the residual stresses are “lost” if any deformation due
to partial relieve of the stresses, such as fracture of a part, is not
sufﬁciently recorded. This is true for most methods based on an
approximation of stresses. One exception is the singleslice
method (Cheng and Finnie, 1998) developed for axisymmetric
bodies for which the original residual stress may be estimated from
a section removed from the original body. However, for an arbi
trary prismatic body the analysis based on an approximation of the
stress becomes much more complicated and it is not easy to ﬁnd a
series of functions that always satisfy the equilibrium condition.
To overcome these limitations, we turn to an alternative ap
proach based on the approximation of the incompatible initial
strain ﬁeld from which the residual stresses may be obtained. The
relation between residual stresses and initial strains has been
addressed in many classic textbooks on elasticity (Fung, 1965;
Timoshenko and Goodier, 1970). However, its use for residual
stress measurement is relatively new. Ueda and his colleagues ﬁrst
proposed their “inherent strain” method in the mid70’s. Instead of
solving the stresses alone, the inherent strain method aims at
establishing the incompatible initial strain ﬁeld that causes the
residual stresses. This involves measuring the deformation in dif
ferent directions due to cutting and/or sectioning the body into
strips and many small segments (Ueda et al., 1986; Ueda and
Fukuda, 1989; Ueda et al., 1996). With elaborate efforts in both
experiment and computation, they demonstrated that initial strains
can be successfully used to estimate residual stresses.
In the present approach initial strains are introduced only for
generating a series of stress distributions on the plane of measure
ment that are used to approximate the axial stresses released prior
to and/or during measurement. For this reason it is informative to
mention some distinctive features of the approach.
First, a residual axial stress ﬁeld is uniquely determined by a
given incompatible initial strain ﬁeld but the reverse is not true.
This implies that there exists more than one incompatible initial
strain ﬁeld that lead to the same axial residual stress distribution on
the plane of measurement.
Second, if the residual stress is altered by cuts made on planes
on which the outofplane shear stresses are small enough to be
neglected before cutting, the original residual axial stress may be
estimated using the initial strain ﬁeld measured from a section of
the part as long as little permanent deformation occurs during
cutting. Thus, the initial strain ﬁeld estimated from a fractured part
may be used to compute the original stress in the part before
fracture if the original axial stress was uniform over a short
distance in the direction normal to the plane of fracture.
Third, for residual stresses produced by local incompatible
deformation, such as due to welding, it is only necessary to deﬁne
the initial strains in a subregion containing the weld (Ueda et al.,
1996). This feature makes the approximation of the initial strains
due to welding easier than the approximation of the stress distri
bution over the entire region of the part. It also implies that the
original residual stress distribution can be estimated only if the
specimen contains the entire initial strain ﬁeld. For a welded part,
for instance, it means that the crosssection of the specimen should
be large enough to contain the entire crosssection of the weld.
It is worth of noting that the ﬁrst feature mentioned above is
unique to the present approach and, as demonstrated later, it makes
the construction of the initial strain distributions for numerical
computation relatively easier than the “inherent strain” approach.
In this paper an analysis based on the approximation of initial
strains is incorporated into the crack compliance method (Cheng
and Finnie, 1994; Prime, 1999) and the singleslice method (Cheng
and Finnie, 1998). A computational procedure is then presented to
obtain the change of strains due to stresses released for each given
initial strain function. After the initial strain ﬁeld is determined by
a least squares ﬁt over the experimentally measured strains, the
residual axial stress can be computed.
To demonstrate the use of initial strains for residual stress
measurement, an experiment using the crack compliance method
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division April
24, 1998; revised manuscript received July 16, 1999. Associate Technical Editor:
S. K. Datta.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 135
Copyright © 2000 by ASME
was carried out on a beam which had been cut apart in a previous
test. The original stress in the beam before it was cut apart was
then estimated using the initial strain approach. It is shown that the
result agrees very closely with the analytical solution.
Initial Strain Approach for the Crack Compliance
Method: Axial Stress in a Beam
The crack compliance method has been used to measure residual
stresses throughthethickness in parts of different conﬁgurations
(Cheng and Finnie, 1994; Prime, 1999). In this method the residual
stress distribution to be measured is approximated by a series of
continuous or piecewise functions with n unknown coefﬁcients.
From the linear superposition shown in Fig. 2, the deformation due
to introducing a progressively increasing depth of cut is the same
as that produced by applying the stress being released on the faces
of the cut with the sign reversed. This allows the deformation or
the “compliance function” due to each individual function in the
approximate stress distribution to be obtained. Since the number m
of measurements recorded during cutting is usually much larger
than the number n of unknown coefﬁcients, a least squares ﬁt is
commonly used to determine the unknowns.
To demonstrate how the initial strain approach works for the
crack compliance method, we consider the residual stress produced
by a beam subjected to pure bending beyond its elastic limit as
shown in Fig. 3. Since the residual stress is uniform along the
length between the two inner support pins, it is sufﬁcient to assume
that the initial strain ﬁeld is uniform along the entire length of the
beam. Also, since there exist more than one initial strain ﬁeld that
produce the same axial stress in the beam, we choose the one that
corresponds to the deformation caused by a temperature variation
e( y), which, when expressed in terms of a series of Legendre
polynomials L
i
( y) (Johnson, 1982), leads to
e͑y͒ ϭ e
x
͑y͒ ϭ e
y
͑y͒ ϭ e
z
͑y͒ ϭ
iϭ2
n
A
i
L
i
͑y͒
for Ϫ1 Յ y Յ 1 (1)
in which y is the normalized distance measured from the neutral
axis. It is seen that the uniform and linear terms are omitted from
Eq. (1) because they do not produce any residual stress for a beam
free of any constraint. The stress in the axial ( z) direction can
readily be obtained using the analytical solution available for
thermal stresses in a long strip (Timoshenko and Goodier, 1970).
That is,
z
͑y͒ ϭ ϪEe͑y͒ ϩ
E
2
͵
Ϫ1
1
e͑y͒dy ϩ 3E
y
2
͵
Ϫ1
1
e͑y͒ydy (2)
in which E is the elastic modulus. Since the integrals in Eq. (2)
vanish for all Legendre polynomials with i Ն 2, we arrive at a
very simple solution for the stress,
z
͑y͒ ϭ ϪEe͑y͒ ϭ ϪE
iϭ2
n
A
i
L
i
͑y͒ for Ϫ1 Յ y Յ 1 (3)
Now we introduce a cut of increasing depth to the beam. The
deformation due to release of the stress on the plane of cut can be
obtained as shown in Fig. 4(A). Thus, without specifying explicitly
the stress distribution on the faces of cut, the deformation due to
releasing the stress by cutting can be determined from the differ
ence between the bodies with and without the cut. This analysis
Fig. 1(A)
Fig. 1(B)
Fig. 1 Illustrations of (A) a beam and a rod and (B) a long butt weld
between two plates with residual axial stresses distributed uniformly
along the length except near the ends
Fig. 2 Linear superposition used for crack compliance method based
on an approximation of the stress
Fig. 3 A beam subjected to fourpoints bending to produce a uniform
axial stress distribution in the midsection between the two inner support
pins
136 / Vol. 122, JANUARY 2000 Transactions of the ASME
can be generalized to other situations, as shown in Fig. 4(B), in
which the shape of the beam is altered by material removal without
any plastic deformation. Although the stresses in latter cases are
different from the original ones, the deformation due to introduc
tion of a cut of increasing depth is caused by the same initial strain
or axial stress. Therefore, the original stress can still be computed
directly from Eq. (3) once the initial strain ﬁeld is estimated. On
the other hand, if the existing stresses on the plane of cut are to be
measured, they need to be computed using the estimated initial
strains for the exact geometry of the part shown in Fig. 4(B)
without the cut.
Denoting the strains produced by the initial strains with and
without the cut by ⑀
c
and ⑀
o
respectively, the change in strain due
to cutting is then given by
⑀͑a
j
͒ ϭ ⑀
c
͑a
j
͒ Ϫ ⑀
o
ϭ
iϭ2
n
A
i
͓⑀
ci
͑a
j
͒ Ϫ ⑀
i
o
͔ ϭ
iϭ2
n
A
i
⑀
i
e
͑a
j
͒ (4)
where a
j
is the jth depth of cut and ⑀
i
e
is the strain produced by the
ith order function L
i
( y) in Eq. (1). For a number m of depths Eq.
(4) can be written in a matrix form
͓⑀
e
͔ ⅐ A ϭ ⑀ (5)
where
͓⑀
e
͔ ϭ
΄
⑀
21
e
Ӈ ⑀
i1
e
Ӈ ⑀
n1
e
Ӈ Ӈ Ӈ
⑀
2j
e
Ӈ ⑀
ij
e
Ӈ ⑀
nj
e
Ӈ Ӈ Ӈ
⑀
2m
e
Ӈ ⑀
im
e
Ӈ ⑀
nm
e
΅
,
A ϭ
΄
A
2
Ӈ
A
i
Ӈ
A
n
΅
and ⑀ ϭ
΄
⑀
1
Ӈ
⑀
j
Ӈ
⑀
m
΅
(6)
For m Ͼ n a least squares ﬁt can be used to obtain the unknown
coefﬁcient vector A from the measured strain vector ⑀. This leads
to
A ϭ ͕͓⑀
e
͔
t
͓⑀
e
͔͖
Ϫ1
͓⑀
e
͔
t
⑀ (7)
where a superscript t denotes a transposed matrix. Substituting the
coefﬁcients A
i
computed from Eq. (7) into Eq. (3), the original
residual stress distribution can be obtained.
Initial Strains Approach for the SingleSlice Method:
Axial Stress in a Rod
The singleslice method, presented recently by Cheng and
Finnie (1998), is primarily for measuring the axial stress dis
tribution in the midsection of a part. Experimentally, it in
volves measuring the strain due to removing a slice from the
midsection. First, a complete cut is made to separate the part in
the midsection. On one of two sections strain gages are in
stalled on the surface exposed by the ﬁrst cut and a second cut
is made to remove a slice containing the strain gages while the
change of strains due to removing the slice is recorded. The
axial stress variation over the crosssection can then be esti
mated using the strain data.
Because the analysis developed by Cheng and Finnie (1998) for
axisymmetric problems is based on the approximation of the stress,
a rigorous solution is given here to show that the initial strain
approach provides an improved alternative. Consider a rod in plane
strain with an axisymmetric residual stress ﬁeld, shown in Fig.
5(A). As mentioned earlier, the choice of the initial strains is not
unique. To illustrate this point, an initial strain ﬁeld which is
uniform in the zdirection is chosen as
e
z
͑r͒ ϭ e͑r͒
e
r
͑r͒ ϭ e
͑r͒ ϭ e͑r͒ (8)
where r is the normalized radial distance and  is an arbitrary
constant. Obviously, a variety of initial strain ﬁelds may be pro
duced by using different values of . In this case the analytical
solution for the corresponding stress distribution in the zdirection
may be obtained as
z
͑r͒ ϭ
E͑ ϩ 1͒
1 Ϫ
2
ͩ
2
͵
0
1
erdr Ϫ e
ͪ
(9)
where E is the elastic modulus and the Poisson’s ratio. It is
seen from Eq. (9) that any initial strains deﬁned by Eq. (8) lead
to similar distributions of the axial stress even though they
produce different stresses in the r and directions. Thus, they
all can be used to estimate the same residual axial stress
because the deformation measured by cutting off the slice is
only dependent on the release of the axial stress. Particularly,
Fig. 5 Schematics of (A) a long cylinder with an axisymmetric residual
stress is separated at the midsection z ؍ 0; (B) a slice cut out along plane
z ؍ a while hoop and/or radial strains are recorded on plane z ؍ 0; (C)
linear superposition for the initial strain approach
Fig. 4 Schematics of (A) crack compliance method for a beam sub
jected to initial strains with and without a cut of increasing depth; (B)
after undergoing cutting or material removal the same beam is used to
measure the initial strains by introducing a cut of increasing depth
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 137
the axial stress can always be generated by specifying an initial
strain in the axial direction.
For axisymmetric problems, a complete polynomial series con
sists of only even orders 2i and the initial strain e may be
expressed as
e͑r͒ ϭ
iϭ1
n
A
i
r
2i
(10)
in which A
i
is the amplitude coefﬁcient for the 2ith order term.
From Eq. (9) the residual axial stress produced by the initial strain
given in Eq. (8) becomes
z
͑r͒ ϭ
E͑ ϩ 1͒
1 Ϫ
2
iϭ1
n
A
iͩ
1
i ϩ 1
Ϫ r
2i
ͪ
(11)
It can readily be shown that equilibrium condition is always
satisﬁed by Eq. (11). Since Eq. (10) is deﬁned by a complete even
order polynomial series, the corresponding axial stress distribution
is also a complete even order polynomial series. Therefore, the
axial stress distribution obtained earlier by Cheng and Finnie
(1998) can be constructed exactly by the polynomials given by Eq.
(11).
Now the strain or deformation produced by each term of the
stress given in Eq. (11) can be computed using each term of the
initial strains given in Eq. (10). Setting  ϭ 1, the assumed initial
strain ﬁeld becomes one produced by a temperature distribution,
which can be handled directly by most ﬁnite element programs.
Setting  ϭ 0, on the other hand, the only nonzero initial strain is
in the axial direction, which can be simulated by specifying a
temperature distribution while setting thermal expansion coefﬁ
cients in the r and directions to zero.
We now separate the rod shown in Fig. 5(A) by a complete cut
in the midsection. For elastic deformation the initial strain given
by Eq. (10) remains unchanged and the new stress state for either
half of the body can be determined from the same initial strain
ﬁeld. We now introduce another complete cut on plane z ϭ a to
remove a slice of thickness a as shown in Fig. 5(B). The change of
strain on the surface of the slice exposed by the ﬁrst cut is due to
release of the stresses on the plane z ϭ a. If the approach based
on the approximation of the stress were used (Cheng and Finnie,
1998), the normal and shear stress distributions on the plane z ϭ
a before the second cut would have to be computed ﬁrst. Then,
with a reversed sign, they would be used as the loading conditions
on the face exposed by the second cut shown in Fig. 5(B) to
compute the deformation of the slice due to the second cut. For the
approach based on initial strains, however, we do not need to
compute the normal and shear stresses existing on plane z ϭ a, nor
do we need to specify any additional loading conditions other than
the initial strains in the slice. A simple procedure based on linear
superposition shown in Fig. 5(C) may be used to obtain the
deformation due to cutting out the slice. It is seen that the defor
mation produced by loading on a slice of thickness a without any
initial strain, shown as (III), is equal to the difference in the
deformation obtained for the slice, shown as (II), subjected to only
initial strain and the deformation for one half of the rod, shown as
(I), subjected to only the same initial strain. Thus, the change of
strain on the face of the slice, z ϭ 0, due to cutting off the slice
can be obtained for each function given in Eq. (10) without
knowing the stresses on plane z ϭ a. This leads to a more
straightforward implementation than that based on the approxima
tion of the stress.
Using the linear superposition approach shown in Fig. 5(C), the
change in strain, say in the direction, due to cutting off the slice
may be expressed as
⑀
͑r
j
͒ ϭ ⑀
c
͑r
j
͒ Ϫ ⑀
o
͑r
j
͒ ϭ
iϭ2
n
A
i
͓⑀
i
c
͑r
j
͒ Ϫ ⑀
i
o
͑r
j
͔͒
ϭ
iϭ2
n
A
i
⑀
i
e
͑r
j
͒ (12)
where r
j
is the radial location of the jth strain gage and ⑀
i
e
is the
strain produced by the ith order function r
2i
in Eq. (10). For a
number m of strain gages Eq. (12) can be written in a matrix form
͓⑀
e
͔ ⅐ A ϭ ⑀
(13)
where matrix [⑀
e
], column vectors A and ⑀
have the same forms
as the counterparts in Eq. (6). Similarly, for m Ͼ n, the unknown
coefﬁcient vector A can be obtained by a least squares ﬁt as given
by Eq. (7).
Experimental Validation
The residual stress produced by fourpoint bending, as illus
trated in Fig. 3, has been used to validate the crack compliance
method in the past (Prime, 1991). This is because the residual
stress can be predicted accurately using the stressstrain relation
obtained from the same beam (Mayville and Finnie, 1982). Also,
it provides an ideal benchmark for checking the capability of the
method for measuring a rapidly varying residual stress ﬁeld
through the thickness since residual stresses with different magni
tude and gradient can be generated by controlling the extent of the
plastic deformation. As demonstrated by Finnie and Cheng (1996),
a ninth order residual stress distribution was measured successfully
by the crack compliance method using the compliance functions
computed either analytically or numerically by ﬁnite element
method. In this paper we present a more severe test. A residual
stress distribution was generated in another beam, which had a
steep gradient near the midplane of y/t ϭ 0.5. Both beams were
made of stainless steel with E ϭ 196 GPs or 28.5 ϫ 10
6
psi.
Because of different yield stresses in compression and tension, the
residual stress distributions produced by bending is not exactly
antisymmetric and the peak tensile stress is always a few percent
higher than the peak compressive stress. As a comparison, the two
stress distributions estimated analytically from two bending tests
denoted as A and B are shown in Fig. 6. The experimental results
given by Finnie and Cheng (1996) using crack compliance method
for beam A are also reproduced in the ﬁgure. It is seen that a ninth
order stress distribution is sufﬁcient to represent the entire stress
distribution for beam A. For beam B, however, the stress gradient
Fig. 6 Two different residual stresses (data lines) produced by four
points bending used for validation of the crack compliance method.
Dotted line—estimated using the compliance functions computed by
FEM.
138 / Vol. 122, JANUARY 2000 Transactions of the ASME
changes much more abruptly near the peak stress and deﬁnitely
requires a higherorder approximation.
The beam (B) was subsequently separated by electric discharge
wire machining (EDWM) near the midsection. Obviously, the
stress had been partially released within a region about one thick
ness from the end exposed by cutting.
In the present study one section of the beam was used to
measure the original residual stress by introducing a cut of pro
gressively increasing depth on a plane about one half (51%)
thickness away from the end, as shown in Fig. 7, while the change
in strain was recorded on the back face directly opposite the cut.
First, the residual stress on the plane of cut (about half thickness
away from the end) was estimated based on approximation of the
stress using Legendre polynomial series of orders from 10 to 19.
Convergent results were obtained when the order of approximation
was larger than 14. Figure 8 shows the estimated stress distribution
represented by a seventeenth order polynomial as a dashed line.
The estimated peak stress is seen to be about 27% lower than the
analytically predicted original residual stress. Then, the approach
presented in this paper was used to estimate the initial strain ﬁled
from which the original residual stress distribution was obtained
using Eq. (3). In this case initial strains were also represented by
Legendre polynomial series of orders 10 to 19. Figure 9 shows the
stress distributions estimated by polynomial series of orders 15 to
19. It is seen that the results stay remarkably stable for y/t Ͼ 0.2
and converge consistently for y/t Ͻ 0.2 with t being the thickness
of the beam. The average of the last three orders (17, 18, and 19)
is used as the ﬁnal result and also shown in Fig. 8 as a solid line.
Now the measured stress agrees closely with the predicted residual
stress. It is worthy of noting that the measured peak tensile stress
is also found to be a few percent higher than the peak compressive
stress, which is consistent with the analytically predicted stresses.
In both cases, the crack compliance functions were computed by
FEM. In the ﬁrst case, a ﬁne mesh was required only in the region
near the cut and a coarser mesh was used in the region away from
the cut. In the second case, the same ﬁne mesh was used for the
entire section of the beam to maintain a uniformly distributed
initial strain ﬁeld along the length of the beam, which required a
considerably larger amount of computation. Fortunately, the tre
mendous increase in computing power in the last few years has
made the computations feasible for most twodimensional prob
lems. In fact, all of the computation required for the present
measurement was carried out on a personal computer.
Discussion
An approach using initial strains for measurement of residual
axial stresses has been presented. Analytical solutions for the crack
Fig. 7 Conﬁguration of the test for measurement of the initial strains in
a section of the beam using the crack compliance method
Fig. 8 Residual stress distributions measured on the plane of cut
(dashed line) and in the original beam (solid line) compared with that
predicted from the bending test
Fig. 9 Results of convergence study using Legendre polynomials se
ries of orders 15–19 to approximate the initial strain ﬁeld in a beam due
to bending beyond elastic limit
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 139
compliance method and the single slice method were obtained for
beams and rods. Experimental validation was carried out by mea
suring the original residual stresses in a beam subjected to four
points bending and subsequently separated in the midsection. The
residual stress distribution to be measured had a gradient that is
considerably steeper than any of the residual stresses estimated
previously by a single continuous function. With the strains re
corded when a cut of progressively increasing depth was intro
duced on a plane about half thickness away from the end, the
present approach successfully estimated the original stress distri
bution which agrees well with the analytically predicted stress
distribution.
The steep stress gradient was approximated by a single contin
uous function using Legendre polynomials of different orders.
Convergent results were obtained when the order was larger than
14. As shown in Fig. 9 the estimated stress distribution stays
remarkably stable for normalized distance y/t Ͼ 0.2 and con
verges consistently for y/t Ͻ 0.2. The study of convergence also
demonstrates that the approach based on initial strains has virtually
identical capability as that based on stresses in representing rapidly
varying residual stresses for the crack compliance method.
The initial approach is also shown to provide a simpliﬁed
analysis for the single slice method for the measurement of axi
symmetric residual axial stresses, such as the waterquenched rod
described by Cheng and Finnie (1998). Using the same experi
mental data, the axial stress distribution in the waterquenched rod
was also obtained by the present approach and was found to agree
closely with that obtained earlier by the stressbased approach.
In summary, the paper shows that the original residual stress can
be measured even if the stress has been partially released by
cutting as long as the permanent deformation introduced by cutting
is negligible and the original stress was uniform over a short
distance along the length. This is true for both the crack compli
ance method and the single slice method.
Acknowledgment
This work was supported by the Electric Power Research Insti
tute. The author is grateful to the Project Manager, Dr. Raj Patha
nia, for his assistance. The author also thanks Professor Emeritus
Iain Finnie of University of California, Berkeley, for his support
and the many stimulating discussions that we shared for more than
a decade. The cutting of the specimen was performed at Precision
Technologies, Livermore, CA and the assistance of Mr. Mark
Bernhard and Mr. Erbin Burgonio with the experiment is greatly
appreciated.
References
Cheng, W., and Finnie, I., 1994, “An Overview of the Crack Compliance Method
for Residual Stress Measurement,” SEM, Proceedings of the Fourth International
Conference on Residual Stresses, pp. 449–458.
Cheng, W., and Finnie, I., 1998, “The SingleSlice Method for Measurement of
Axisymmetric Residual Stresses in Solid Rods or Hollow Cylinders in the Region of
Plane Strain,” ASME JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 120,
pp. 170–176.
Finnie, I., and Cheng, W., 1996, “Residual Stress Measurement by the Introduction
of Slots or Cracks,” Localized Damage IV, Nisitani et al., eds. Computation Mechan
ics Publications, pp. 37–51.
Fung, Y. C., 1965, Foundations of Solid Mechanics, PrenticeHall, Englewood
Cliffs, NJ.
Prime, M. B., 1991, “Experimental Veriﬁcation of the Crack Compliance Method,”
M.S. project report, University of California, Berkeley.
Prime, M. B., 1999, “Residual Stress Measurement by Successive Extension of a
Slot: The Crack Compliance Method,” Appl. Mech. Rev., Vol. 52, pp. 75–96.
Timoshenko, S. P., and Goodier, J. N., 1970, Theory of Elasticity, 3rd edition,
McGrawHill, New York.
Johnson, D., and Johnson, J., 1982, Mathematical Methods in Engineering and
Physics, PrenticeHall, Englewood Cliffs, NJ.
Mayville, R., and Finnie, I., 1982, “Uniaxial StressStrain Curves from a Bending
Test,” Experimental Mechanics, Vol. 22, p. 197.
Ueda, Y., Fukuda, K., and Kim, Y. C., 1986, “New Measuring Method of Axi
symmetric ThreeDimensional Residual Stresses Using Inherent Strains as Parame
ters,” ASME JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 108, pp.
328–334.
Ueda, Y., and Fukuda, K., 1989, “New Measuring Method of ThreeDimensional
Residual Stresses in Long Welded Joints Using Inherent Strains as Parameters—L
z
Method,” ASME JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 111, pp.
1–8.
Ueda, Y., Murakawa, H., and Ma, N. X., 1996, “Measuring Method for Residual
Stresses in Explosively Clad Plates and a Method of Residual Stress Reduction,”
ASME JOURNAL OF ENGINEERING MATERIALS AND TECHNOLOGY, Vol. 118, pp. 576–582.
140 / Vol. 122, JANUARY 2000 Transactions of the ASME
M. Chiarelli
Assistant Professor.
A. Lanciotti
Associate Professor.
Department of Aerospace Engineering,
University of Pisa,
Via Diotisalvi 2,
56126 Pisa, Italy
M. Sacchi
Engineer,
Breda Railway Constructions,
Via Ciliegiole 110b,
51100 Pistoia, Italy
Effect of Plasma Cutting on the
Fatigue Resistance of Fe510 D1
Steel
The paper describes the results of a research programme, carried out at the Department
of Aerospace Engineering of the University of Pisa, for the assessment of the inﬂuence of
plasma cutting on the physical and mechanical properties of Fe510 D1, a low carbon steel
widely used in carpentry. The activity started by observing that several industries rework
plasma cut edges, particularly in the case of fatigue structures, in spite of the good quality
of the plasma cut edges in a fully automatic process. Obviously, reworking is very
expensive and timeconsuming. Comparative fatigue tests demonstrated that the fatigue
resistance of plasma cut specimens in Fe510 steel was fully comparable to that of milled
specimens, as the consequence of the beneﬁcial residual stresses which formed in the
plasma cut edges.
1 Introduction
A research programme has been conducted at the Department of
Aerospace Engineering of the University of Pisa concerning the
effect of plasma cutting on the fatigue properties of Fe510 D1, a
steel typically used for the construction of railway vehicles and in
carpentry, thanks to its excellent quality in welding.
Different options exist to proﬁle a sheet or a plate; laser, plasma,
oxyfuel, waterjet and mechanical proﬁling are those most fre
quently used. Limiting our attention to railway constructions and
railway trucks in particular, they are typically welded structures
built by starting from plates with a thickness in the range of 6 to
12 mm. Plasma cutting in this case is cheaper and faster than laser
or waterjet cutting, and it provides better edge ﬁnish than oxy
fuel. A fundamental requirements in railway constructions is the
fatigue resistance of structures. In the case of free edges (edges
which are not subsequently welded), in spite of the good quality of
the plasma cut edges in a fully automatic process, reworking by
grinding or mechanical proﬁling is often performed. The heat
affected zone, the residual stresses, the possible defects and the
decarburized area are eliminated in this way, but obviously re
working is very expensive and timeconsuming. In the case of
edges which are subsequently incorporated in a welded joint this
problem is not likely to exist, as these effects are to be eliminated
during the subsequent fusion. Notwithstanding this, in many cases
milling is used to prepare the chamfers which are necessary to
weld thick plates. Reworking of free and occasionally of welded
edges, too, is not an uncommon practice; “where doubts exist” it is
performed in the case of welded bridges (Harris, 1996), even
though these structures are less critical from the fatigue point of
view.
An exhaustive review of the effects of gas cutting on the fatigue
resistance of mild and high tensile structural steels was carried out
by Gurney (1979). The main conclusion drawn is that the quality
of the cutting greatly inﬂuences the fatigue resistance of steels,
which can be as high as that of the parent metal up to one half of
this value.
In this context, it was interesting to quantify the fatigue resis
tance of plates cut by means of the equipment now on the market,
particularly since in recent years the technology of plasma cutting
has been greatly improved by the introduction of High Deﬁnition
Plasma Arc Cutting (HDPAC). This technique uses a small oriﬁce,
0.4 Ϭ 0.7 mm, instead of 3 Ϭ 4.5 mm for normal plasma cutting.
This allows us to obtain close tolerances and square edges, com
parable to the results that can be obtained by laser cutting, but at
a signiﬁcantly lower cost (Kirkpatrick, 1995; Harris and Lowery,
1996; Hoult et al., 1995). Therefore, it is possible that in the future
a system of this type will be accepted as a ﬁnishing process for
fatigue loaded structures.
2 Experimental
2.1 Base Material. Fe510 D1 in standard plate supply has a
ferritic structure; the chemical composition of this material is
given in Table 1, together with its mechanical properties (yield
stress,
y
, ultimate stress,
u
, and elongation to failure A), mea
sured in the longitudinal (L) and in the transversal (T) directions.
Dogbone specinens, 40 mm wide, were machined from plates
with thickness of 8 mm; plates of this thickness are typically used
in the construction of railway trucks. The external surfaces of the
specimens were not machined, so as to maintain, as in real con
structions, the “asreceived” condition of the plates. Specimens
were saw cut and milled in the longitudinal direction of the plates;
a notch was introduced in some of them, a central hole, 4 mm in
the diameter, to investigate the notch sensitivity of the material.
2.2 Plasma Cut Specimens. A group of specimens was
obtained by cutting them with a numerically controlled plasma
cutting machine. The torch was water cooled and had a nozzle with
an outlet diameter of 2.5 mm; the plasma gas was oxygen, 0.05
m
3
/s, at a pressure of 10.5 bar. Water was injected in the arc, 36
cc/s; the injection of water dilutes the plasma jet and improves the
quality of the cut edges. A current setting of 400 amps at 150 volts
was used. The distance between the torch and the plate was 3 mm;
the cutting speed was 92 mm/s. The plasma cut specimens were
also obtained in the longitudinal direction of the plates. The plasma
cut surfaces did not look as regular as the milled surfaces. The
plasma cut edges were not straight and the width of the plate on the
reverse side was about 0.8 mm smaller than that on the torch side,
Fig. 1. In addition, the width of the specimens ranged from 37.3 to
40.05 mm, while the nominal dimension was 40 mm. These
differences are generally meaningless in large structures, but can
be important in small structures, so that it can be concluded that
close tolerances cannot be obtained by standard plasma cutting.
Besides, small scratches were present on the cut surfaces. The
loads to be applied in the tests on plasma cut specimens were
evaluated by taking into account their actual dimensions.
Contributed by the Materials Division for publication in the JOURNAL OF ENGINEER
ING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials Division
December 11, 1998; revised manuscript received April 6, 1999. Associate Technical
Editor: A. E. Werner.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 141
Copyright © 2000 by ASME
Other important aspects concerning plasma arc cutting are local
structural modiﬁcation of the material and the presence of residual
stresses. The base material has a ferritic structure; the rapid cooling
after plasma cutting transforms the ferritic structure into ﬁne
perlite or martensite, like a thermal process of quenching. Figure 2
shows a transverse section of a plasma cut specimen (detail A in
Fig. 1); the heat affected zone, which is about 0.5 mm wide on the
plasma torch side and about 0.2 mm wide on the rear side, can be
clearly seen. It was composed of martensite (see the detail in Fig.
2), a harder structure than the basic ferrite. The hardness in the heat
affected zone was about 500 Vickers, more than twice the hardness
of the core material, Fig. 3.
As far as residual stresses are concerned, gas cutting always
produces thermal residual stresses: during the cutting process, the
area around the heat source is violently heated and locally fused.
The hot material tends to expand but the cooler surrounding area
partially prevents this expansion, producing a plastic deformation
of the hot area, also as a consequence of the low yield stress of the
hot material. So during the cooling phase, tensile residual stresses
are generated in the plastically compressed area. These stresses are
balanced by compressive residual stresses in the inner zones of the
plates. At the same time, as a function of the cooling rate, micro
structural transformations can take place in the material; they
produce changes in the volume and, as a consequence, residual
stresses. In particular, if the cooling rate is sufﬁciently high, the
transformation from ferrite to martensite takes place and this
produces an increase in volume with consequent compressive
residual stresses, which add to the thermal tensile residual stresses.
Measurements were carried out to establish the ﬁnal distribution of
the residual stresses. A plasma cut edge was instrumented with
strain gauges and dissected to allow the relaxation of the internal
stresses. A chain of nine strain gauges, located at 1 mm from each
other, was used for this measurement, together with a single strain
gauge placed on the plasma cut surface. Figure 4 shows the
position of the strain gauges together with the dissection scheme.
The results obtained are shown in Fig. 5. A very thin layer of
material near the plasma cut edge, about 0.2 mm thick, had been
subject to a compressive residual stress due to the microstructural
transformation from ferrite to martensite. The maximum compres
sive stress at the external surface was about 150 MPa. This stress
was superimposed on the thermal residual stress ﬁeld, which
involved the deeper layers of the specimen. The presence of a
compressive residual stress at the external surface partly explains
Table 1(b) Mechanical properties of Fe510 D1
Fibres
y
(MPa)
u
(MPa) A (%)
L 455 620 31.0
T 405 575 30.5
Table 1(a) Chemical composition of Fe510 D1
C: 0.193%
Mn: 1.520%
P: 0.015%
S: 0.0004%
Si: 0.360%
Al: 0.026%
Fig. 1 Aspect of the edges after plasma cutting
Fig. 2 Heat affected zone after plasma cutting
Fig. 3 Microhardness of the plasma cut heat affect zone
142 / Vol. 122, JANUARY 2000 Transactions of the ASME
the good fatigue behavior of plasma cut specimens, as will be
shown in the following. Note that the presence of a surface
compressive stress, instead of a thermal tensile residual stress, is
not a general result. It depends on the material, its thickness and on
the cutting conditions (gas, velocity, diameter of the oriﬁce and so
on). By changing these parameters a different stress ﬁeld can be
produced.
2.3 Test Equipment and Programme. Fatigue tests were
carried out at room temperature by means of a servohydraulic
fatigue machine with a maximum loading capacity of 200 KN.
Sinusoidal loading was applied. The load frequency ranged from
10 to 18 Hertz as a function of the load applied; the ratio between
the minimum and the maximum stresses in the load cycle cycle
was R ϭ 0. Plasma cut specimens were also tested at R ϭ 0.5.
Additional tests were also carried out to investigate the notch
sensitivity of the material, the effect of thermal relaxation on
plasma cut specimens, and the effect of the microscratches ob
served on the plasma cut surfaces. For this purpose, artiﬁcial
micronotches were introduced in some specimens.
3 Results and Discussion
Figure 6 shows the results concerning the milled specimens (the
arrow on the right of a symbol indicates a fatigue test suspended
without evidence of fatigue damage). The fatigue limit of un
notched specimens was about 370 MPa and shows a good result
compared with the yield stress of the material,
y
ϭ 455 MPa.
The fatigue limit decreased to 205 MPa in the case of the
notched specimens (central hole, 4 mm in diameter). By taking
into account the value of the stress concentration factor of these
specimens, K
t
ϭ 2.73 (Peterson, 1974), the notch sensitivity
factor was q ϭ 0.46. This value is in agreement with certain
experimental results available in the literature (Juvinall, 1967),
which allow the estimation of the notch sensitivity factor of a steel
on the basis of the radius of the notch and the yield stress of the
material.
The results of milled and of the plasma cut specimens are
compared in Fig. 7; as can be observed, they were fully compa
rable. Therefore, it can be concluded that plasma cutting in this
material, thickness 8 mm, is not detrimental from the fatigue
resistance point of view. This observation is conﬁrmed by the fact
that fatigue crack nucleation took place both in the plasma cut
edges and in the base material.
Prokopenko (1991) carried out tests on the fatigue behavior of
plasma cut edges. In this case, an increase of 60% in hardness and
comparable fatigue behavior was observed in plasma cut edges
when compared with the results of the initial nonhardened edges of
a structural steel.
The experimental activity was continued in order to use the
Fig. 4 Strain gauge map for residual strain evaluation in a plasma cut
edge
Fig. 5 Distribution of the longitudinal residual stress in a plasma cut
edge
Fig. 6 Results of the fatigue tests carried out on notched and un
notched specimens. Edge preparation: milling.
Fig. 7 Results of the fatigue tests carried out on milled and plasma cut
specimens
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 143
results obtained from the practical point of view. Two aspects were
examined in some detail: the ﬁrst was related to the possible
relaxation of the compressive residual stress, the second to the
possible brittle behavior of the martensite layer. With regard to the
ﬁrst aspect, complex welded structures were thermal relaxed to
eliminate the residual stresses introduced by the welding process;
it was important to verify that the fatigue resistance of a plasma cut
free edge included in a welded structure was preserved after the
thermal relaxation process. This operation was carried out in
accordance with the following procedure:
● maximum starting temperature: 200°C;
● heating gradient: 60°C/hr Ϭ 150°C/hr;
● reaching maximum temperature of 590°C Ϯ15°C and main
tenance for 1 hr;
● cooling gradient: 120°C/hr;
● exit temperature: Ͻ200°C;
● cooling in calm air.
Observations using an optical microscope showed that the sur
face layer was apparently unchanged after thermal relaxation. The
properties of the martensite layer were certainly changed after the
high temperature process, but there were no effects on the fatigue
resistance, so that, the fatigue results obtained from this group of
specimens were fully comparable with the previous results.
A second group of specimens was used for the analysis of the
notch sensitivity of the plasma cut edges. Small scratches are
always present on gas surfaces; the aim of these tests was to
investigate the effect of a bigger scratch produced during gas
cutting or accidentally introduced in the structure. Such damages
are not important in ductile materials, but it was important to
investigate their effects in the brittle martensite layer. Small sharp
notches were introduced in the specimens by means of a small
abrasive disk, thickness 0.25 mm. The notch consisted in a cut
along the whole thickness of the specimens, Fig. 8. Different
depths of the notch, “a” in Fig. 8, from 0.1 to 0.5 mm measured on
the plasma torch side, were tested. The results show that a notch
with a depth of 0.1 to 0.3 mm did not affect fatigue resistance; this
was conﬁrmed by the fact that the fatigue nucleation took place in
some specimens in a different location from the starting notch. All
the specimens cracked at the micronotch location when its depth
was increased to 0.5 mm. The results obtained in this case are
shown in Fig. 9. The micronotch decreased the fatigue stress by
about 5%, a relatively small amount. The conclusion of these tests
was that small scratches in a plasma cut edge are not detrimental.
Some tests on plasma cut specimens were carried out at a load
ratio R ϭ 0.5. This value is representative of several railway
applications in which the fatigue stresses superimposed the impor
tant static stress produced by the weight of the vehicle. Figure 10
shows the results of these tests, compared with the results relevant
to R ϭ 0. The scatter in the results was very high in this case and
several runouts were also recorded at the higher stress levels. On
the other hand, by increasing the stress ratio, the maximum stress
in the fatigue cycle is almost the yield stress of the material, so an
increase in the scatter is not surprising.
4 Conclusions
A research programme has been conducted at the Department of
Aerospace Engineering of the University of Pisa concerning the
effect of plasma cutting on the fatigue properties of Fe510 D1.
Plasma cutting is widely used in industry but in spite of the good
quality obtained by automatic processes, plasma cut edges in
critical fatigue structures are often reworked by grinding or me
chanical proﬁling.
Comparative fatigue tests were carried out on dogbone speci
mens obtained from plate by milling or plasma cutting. Obviously,
the plasma cut surfaces did not look as regular as the milled
surfaces; the edges were not straight and close tolerances are
difﬁcult to obtained. In addition, small scratches were present on
the cut surfaces.
Fig. 8 Specimen with micronotch
Fig. 9 Comparison between the fatigue results of notched and un
notched plasma cut specimens
Fig. 10 Comparison between the results of the fatigue tests carried out
on plasma cut specimens under different stress ratios
144 / Vol. 122, JANUARY 2000 Transactions of the ASME
The optical microscope showed that the heataffected zone of a
plasma cut edge was composed of martensite, a harder structure
than the basic ferrite. The hardness of this area was more than
twice the hardness of the parent material.
A compressive residual stress of about 150 MPa was measured
at the external surface of a plasma cut edge. This stress is super
imposed on the thermal residual stress ﬁeld, which involved the
deeper layers of the specimen.
The results of fatigue tests carried out on milled and plasma cut
specimens were fully comparable. So it can be concluded that
plasma cutting in this material, thickness 8 mm, is not detrimental
from the fatigue resistance point of view. It is important to note
that the extrapolation of the results obtained to other materials, to
different thickness or different cutting parameters, is not necessar
ily possible. Therefore, additional tests must be carried out when
one or more parameters are changed.
Additional fatigue tests conﬁrmed the good fatigue resistance of
plasma cut specimens after thermal relaxation even in the presence
of small artiﬁcial defects in the hard layer produced by plasma
cutting.
References
Gurney, T. R., 1979, Fatigue of Welded Structures, Second Edition, Cambridge
University Press.
Harris, I. D., 1996, “Plasma Arc Cutting of Bridge Steels,,” Proceedings of the
International Conference on Welded Sturctures in Particular Welded Bridges, Budap
est 2–3 September, pp. 363–364.
Harris, I. D., and Lowery, J., 1996, “High Tolerance Plasma Arc Cutting,” Welding
in the World, Vol. 37, No. 6, pp. 283–287.
Hoult, A. P., Pashby, I. R., and Chan, K., 1995, “Fine Plasma Cutting of Advanced
Aerospace Materials,” Journal of Materials Processing Technology, Vol. 48, Jan., pp.
825–831.
Juvinall, R. C., 1967, Stress, Strain, and Strength, McGrawHill, New York.
Kirkpatrick, I., 1995, “High Deﬁnition Plasma—An Alternative to Laser Technol
ogy,” Sheet Metal Industries, Vol. 72, pp. 18–19, Sept.
Peterson, R. E., 1974, Stress Concentration Factors, John Wiley, New York.
Prokopenko, A. V., et al., 1991, “Effect of Plasma Cutting on the Fatigue and
Cracking Resistance of Steel,” Strength of Materials (English translation of Problemy
Prochnosti), Vol. 22, No. 12; Aug., pp. 1738–1740.
Journal of Engineering Materials and Technology JANUARY 2000, Vol. 122 / 145