this document downloaded from Terms and Conditions of Use:

vulcanhammer.net
All of the information, data and computer software
(“information”) presented on this web site is for general
information only. While every effort will be made to insure
Since 1997, your complete its accuracy, this information should not be used or relied on
for any specific application without independent, competent
online resource for professional examination and verification of its accuracy,
suitability and applicability by a licensed professional. Anyone
information geotecnical making use of this information does so at his or her own risk
and assumes any and all liability resulting from such use.
engineering and deep The entire risk as to quality or usability of the information

foundations: contained within is with the reader. In no event will this web
page or webmaster be held liable, nor does this web page
or its webmaster provide insurance against liability, for
The Wave Equation Page for any damages including lost profits, lost savings or any
other incidental or consequential damages arising from
Piling the use or inability to use the information contained
within.
Online books on all aspects of
This site is not an official site of Prentice-Hall,
soil mechanics, foundations and Pile Buck, the University of Tennessee at
marine construction Chattanooga, or Vulcan Foundation
Equipment. All references to sources of
Free general engineering and software, equipment, parts, service
or repairs do not constitute an
geotechnical software endorsement.

And much more...

Visit our
companion site
http://www.vulcanhammer.org
 /0
X~'A
UILU-ENG-91-2001

5S1CIVIL ENGINEERING STUDIES

STRUCTURAL RESEARCH SERIES NO. 558

CIlPV ,
ISSN: 0069-4274

A LARGE STRAIN PLASTICITY MODEL

FOR IMPLICIT FINITE ELEMENT ANALYSES

By
ROBERT H. DODDS, JR.
and
BRIAN E. HEALY

A Report on a Research Project

Sponsored by the
DAVID TAYLOR RESEARCH CENTER
METALS AND WELDING DIVISION
ANNAPOLIS, MARYLAND
Research Contract: N61533-88-C-0035

DEPARTMENT OF CIVIL ENGINEERING

UNIVERSITY OF ILLINOIS AT
URBANA-CHAMPAIGN
URBANA, ILLINOIS
JANUARY 1991
 50272-101
1. REPORT NO. 2. 3. Recipient'. Acceaalon No.
REPORT DOCUMENTATION
PAGE UILU-ENG-91-2001
4. lltJe and SUbtItle 5. Report Date

A Large Strain Plasticity Model for Implicit January 199'1

Finite Element Analyses 8.

7. Author{s) 8. Perfor'rNng OrganIzation Report No.

R.H. Dodds, Jr. and B.E. Healy SRS 558

9. PerfOl'l11k1g OrganIzatIon Name.net Addreaa 10. ProfectfTaak/Wortt Unit No.

University of Illinois at Urbana-Champaign

Department of Civil Engineering 11. ContrIlCt(C) or GnInt(Q) No.

205 N. Mathews Avenue N61533-88-C-0035

Urbana, Illinois 61801
12.Sponaorlng OrganIzation Name.net Addreaa 13. Type of Report & Pertod Covered

David Taylor Research Center Annual:11-1-89 to 12-31-90

Metal and Welding Division, Code 281
14.
Annapolis, Maryland 21542
15. SUpplementary Note.

18.AbstrKt (limit: 200 words)

The theoretical basis and numerical implementation of a plasticity model suitable for finite strains and ro-
tations are described. The constitutive equations governing J2 flow the0!y' are formulated using strains-
stresses and their rates defined on the unrotated frame of reference. Unhke models based on the classical
J aumann (or corotational) stress rate, the present model predicts physically acceptable responses for homo-
geneous deformations of exceedingly large magnitude. The associated numerical algorithms accommodate
the large strain increments that arise in finite-element formulations employing an implicit solution of the
global equilibrium equations. The resulting computational framework divorces the finite rotation effects on
. . +--=- ,. . .. _-,...,... -_ . . __ +_"'..."..... :_+rt. . . . . _n+~" ...... 1"'"tI..f +).....0
:Sll(:1111-:SUt:::s:s Id.Lt:::s HU1l1 ll1Lt::t;1a.L.1Ull U~
"'1"\+.0",
~al.t::..,
L.llt::
+".....
ut'ual.~
L.U
.An+o +h.o ~f"'\+o ...;nl ....0"" . . . . .,. .,. . .
'1 ........
1.11\.;
,("1.0
111al.t::~lal ~\';"''pUll'''~ uv\.;~
n 1.1"'- .... ....:1 (+';,...........0\ C"I+a~
.I"\.'1T.o.,.-
a luau \Ull1~J ..,I.t::p.
Consequently, all of the numerical refinements developed previously for small-strain plasticity (radial re-
turn with subincrementation, plane stress modifications, kinematic hardening, consistent tangent operators)
are utilized without modificatIOn. Details of the numerical algorithms are provided including the necessary
transformation matrices and additional techniques re9.uired for finite deformations in plane stress. Several
numerical examples are presented to illustrate the realIstic responses predicted by the model and the ro-
bustness of the numerical procedures.

17. Document AnatyaIa •• Descripton

Large Strains, Large Rotations, Plasticity, Finite Elements, Numerical Methods, Implicit, Total Lagrangian

b. IdentIftera/Open-Ended Terms

c. COSATI field/Group

18. Avallabllty Statement 19.5ecurtty ctass (ThIs Report) 21. No. of Pages

UNCLASSIFIED 34
Release Unlimited 20. Securtty Class (ThIs Page) 22.Pr1ce

UNCLASSIFIED
(See ANSI-Z39.18) OPTIONAL FORM 272 (4-77)
Department of Commerce
 A LARGE STRAIN PLASTICITY MODEL
FOR IMPLICIT FINITE ELEMENT ANALYSES

By

Robert H. Dodds, Jr.

and

Brian E. Healy

Department of Civil Engineering

University of Illinois

A Report on a Research Project Sponsored by the:

DAVID TAYLOR RESEARCH CENTER

METALS AND WELDING DIVISION
Annapolis, Maryland 21402

University of Illinois
Urbana, Illinois
January 1991
 ABSTRACT

The theoretical basis and numerical implementation of a plasticity model suitable for fi-
nite strains and rotations are described. The constitutive equations governing J2 flow theory
are formulated using strains-stresses and their rates defined on the unrotated frame of refer-
ence. Unlike models based on the classical Jaumann (or corotational) stress rate, the present
model predicts physically acceptable responses for homogeneous deformations of exceeding-
ly large magnitude. The associated numerical algorithms accommodate the large strain incre-
ments that arise in finite-element formulations employing an implicit solution of the global
equilibrium equations. The resulting computational framework divorces the finite rotation
effects on strain-stress rates from integration of the rates to update the material response
over a load (time) step. Consequently, all of the numerical refinements developed previously
for small-strain plasticity (radial return with subincrementation, plane stress modifications,
kinematic hardening, consistent tangent operators) are utilized without modification. Details
of the numerical algorithms are provided including the necessary transformation matrices
and additional techniques required for finite deformations in plane stress. Several numerical
examples are presented to illustrate the realistic responses predicted by the model and the
robustness of the numerical procedures.

- ii -
 ACKNOWLEDGMENTS

The first author (R.H. Dodds) was supported by the David Taylor Research Center Metals
and Welding Division (Code 281), Annapolis, Maryland under contract N61533-88-C-0035
to the University of Illinois. The second author (B.E. Healy) acknowledges support provided
by the Center for Supercomputing Research and Development under grant SCCA 90-82144
from the illinois Dept. of Commerce. Computations reported here were performed on an
Apollo DN-10000 workstation operated by the Department of Civil Engineering. Acquisi-
tion of this computer was made possible by a grant from Hewlett-Packard, Inc.

- iii -
 TABLE OF CONTENTS

Section No. Page

1. INTRODUCTION. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 1

2. KINEMATICS, STRAIN-STRESS MEASURES AND THEIR RATES .... 3

3. ELASTO-PLASTIC CONSTITUTIVE FRAMEWORK . . . . . . . . . . . . . . . .. 7

3.1 Selection of Stress-Strain Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 7
3.2 Plasticity Rate Equations ................................... 8
3.3 Stress Updating on the Unrotated Configuration . . . . . . . . . . . . . .. 10

4. NUMERICAL PROCEDURES FOR FINITE STRAINS. . . . . . . . . . . . . .. 10

4.1 Stress Updating Procedure ................................. 11
4.2 Tangent Modulus for Stiffness Updating ...................... 12
4.3 Plane-Stress Idealization .................................. 14
4.4 Polar Decomposition ......... 0 ••••••• 0 •••••••••••••••••••• 15

5. NUMERICAL EXAMPLES ...................................... 16

5.1 Homogeneous Finite Extension ......... . . . . . . . . . . . . . . . . . . .. 17
5.2 Homogeneous Finite Shear ................................ 19
5.3 Crack-Tip Blunting In Small-Scale Yielding .................. 23

6. SUMMARY AND CONCLUSIONS ............................... 26

7. REFERENCES ................................................ 28

Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 31

- iv -
 LIST OF TABLES

Table No. Page

1 Relative Computational Effort Required for Polar Decomposition ....... 16

- v -
 LIST OF FIGURES

Figure No. Page

1 Initial and deformed configurations illustrating two methods of

decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 4

2 Homogeneous finite extension. Comparison of finite-element

results with analytical solutions. Incremental, linear-elastic
material: E= 1, v=0.3 ............................................ 18

3a Homogeneous finite simple shear. Comparison of finite-element

and analytical solutions for the Cauchy normalstress. Incremental,
linear-elastic material: E = 1, v = 0.3 ................................. 20

3b Homogeneous finite simple shear. Comparison of finite-element

and analytical solutions for the Cauchy shear stress. Incremental,
linear-elastic material: E = 1, v = 0.3 ...................... _. . . . . . . . . .. 22

4 Finite-element model (plane-strain) for boundary layer idealization

of the small-scale yielding problem ................................. 24

5 Crack-tip deformations for increasing applied load in the SSY model 26

6 Comparison of finite-element results for initially blunt tip with

asymptotic HRR fields and with sharp-tip models (r and f) are
coordinates in initial configuration) ................................ 27

-vi-
A LARGE STRAIN PLASTICITY MODEL
FOR IMPLICIT FINITE ELEMENT
ANALYSES

1. INTRODUCTION

High-end workstations and mini-supercomputers are making feasible the routine consid-
eration of plasticity with large strain and rotation effects in finite-element analyses. Such
diverse phenomenon as post-buckling deformations, metal forming, contact/indentation
and the micromechanics of ductile fracture may be realistically modeled. Finite rotations
of material axes (those attached to material points that rotate with the continuum in a
local sense) complicate the definition of strain-stress rates and their numerical integra-
tion to advance the material response over a load (or time) step. Traditionally, constitutive
models for large strain plasticity in finite-element codes are cast in a spatial setting which
mandates use of an objective stress rate to remove that part of the total stress rate due to
simple rigid rotation of the material. In a spatial setting, the components of Cauchy (true)
stress are defined relative to a fixed, Cartesian system; thus rigid rotati-on alone alters the
stress components. While numerous objective stress rates may be constructed [2] with
each leading to a potentially different material response, the Jaumann stress rate has been
implemented universally in both explicit and implicit codes given its apparent simplicity,
for example [17] and [1].
Over the past ten years, serious objections to constitutive models employing the Jau-
mann rate have developed as more complex material behavior is considered (e.g., kine-
matic hardening and viscoplasticity) and as the magnitude of deformations experienced
in the applications has increased (plastic strains exceeding 50-100%). The first objection
addresses the increased complexity of numerical algorithms to accommodate the spatial
setting; tensorial state variables within the plasticity model, for example, the back-stress
in kinematic hardening, must also be expressed using an objective rate and modified to
reflect finite rotations. Processing of the purely kinematic effects due to finite rotations
is thus interwoven with integration of evolution equations for the internal state variables.
Consequently, development of each new material model requires potentially individual
treatment of finite rotations. The second objection to use of the Jaumann rate concerns
the physically unacceptable stresses predicted at large strains under certain conditions.
The problem of simple finite shear illustrates the deficiency [5]. An incremental, linear-
elastic material law is used to relate the J aumann stress rate to the rate of deformation
expressed in a fixed Cartesian system. The predicted Cauchy stresses oscillate in an un-
realistic manner (a12 actually reverses sign). Nagtegaal and de Jong [22] noted such stress

1
oscillations with kinematic hardening in elasto-plasticityfor a material which strain hard-
ens monotonically in tension. Atluri [2] later showed that similar oscillations exist for iso-
tropic hardening unless the elastic strains are vanishingly small. The oscillatory response
derives from the constant spin rate tensor characteristic of simple shear while the actual
rigid-body rotation diminishes with increasing deformation, approaching Tf/2 in the limit.
The Cauchy stress obtained using the constant spin tensor becomes erroneous once the
logarithmic shear strain, Y12, exceeds 100%.
Atluri [2] demonstrated that removal of the oscillatory response in simple shear may
be accomplished through definition of alternate stress rates or through a more general
construction of the hypo-elastic material law. In a desire to retain the simplest hypo-elas-
tic material law as a direct generalization of the conventional small-strain forms, Green
and Naghdi [9] introduced an objective stress rate that has been discussed extensively by
Dienes [5], Johnson and Bammann [15] and Atluri [2]. A Cauchy stress measure and its
objective rate are defined on an unrotated orthogonal reference frame established
through polar decomposition of the total deformation gradient at each material point.
This constitutive model predicts monotonically increasing stresses in simple shear for in-
cremental, linear-elasticity. Using this concept of an unrotated reference frame for con-
stitutive modeling, Flanagan and Thylor [8] developed the PRONTO 2-D and 3-D [32]
codes for transient dynamic analysis with explicit time integration. An impressive collec-
tion of material behaviors and contact algorithms are included in these codes. Constitu-
tive computations are performed using strains, stresses and their objective rates defined
on the unrotated reference frame. Effects of finite rotations are thus transparent to inte-
gration algorithms for stresses and the material state variables. The numerical architec-
ture of existing small-strain plasticity models is fully retained. Flanagan and Taylor note
their computational challenge was development of an exceptionally efficient algorithm
for evolution of the polar decomposition with time in the globally explicit solution.
The present paper describes the implementation and performance of the unrotated
reference frame concept for finite-element solutions that use implicit methods to resolve
the global equilibrium equations and as such represents an extension of Flanagan and
Taylor's work. While efficient methods for polar decomposition remain an issue, two addi-
tional challenges face the developer of an implicit code: 1) efficient and accurate schemes
to integrate the plasticity rate equations over the very large strain increments characteris-
tic of implicit methods, and 2) consistent tangent operators to maintain quadratic rates
of convergence for global equilibrium iterations. Accordingly, the contents of the paper
are as follows: (i) a description of the kinematics of finite deformation and development
of the strain-stress rates, (ii) a brief development of the rate independent plasticity theory
for finite strains and a discussion of numerical techniques to integrate the plasticity rate
equations, (iii) details of the computational steps to process finite rotations at a material
point, development of a consistent tangent operator, and complications arising for the
plane-stress idealization. The paper concludes with the solution of several example prob-

2
lems that illustrate the physically acceptable responses predicted by the material model
and the robustness of the numerical implementation. Finite simple extension and shear
are examined with comparisons made between analytical and numerical solutions. Severe
blunting at a crack-tip is analyzed in the final example with specific attention given to the
global convergence rate.

2. KINEMATICS, STRAIN-STRESS MEASURES AND THEIR RATES

Development of the finite strain plasticity model begins with consideration of the defor-
mation gradient

F = ax/ax, det(F) = J> 0 (1)

where X denotes the Cartesian position vectors for material points defined on the refer-
ence (undeformed) configuration. Position vectors for material points at time t are de-
noted x (configuration B in Fig. 1, after Flanagan and Taylor [8]). The displacements of
material points are thus given by u = x-X. In static analyses we associate the time-like
parameter t with a specified level of loading imposed on the model. Stress and deforma-
tion rates are thus defined with respect to the applied loading rather than with time.
The polar decomposition of F yields

F = VR= RU (2)

where Vand U are the left- and right-symmetric, positive definite stretch tensors, respec-
tively; R is a orthogonal rotation tensor. The principal values of V and U are the stretch
ratios, 'N, of the deformation. These two methods for decomposing the motion of a materi-
al point are illustrated in Fig. 1. In the initial configuration, Bo , we define an orthogonal
reference frame at each material point such that the motion relative to these axes is only
deformation throughout the loading history. With the RU decomposition, for example,
these axes are "spatial" during the motion from Bo to Bu; they are not altered by deforma-
tion of the material. However, during the motion from Bu to B these axes are "material";
they rotate with the body in a local average sense at each material point. Strain-stress
tensors and their rates referred to these axes are said to be defined in the unrotated config-
uration.
The material derivative of displacement with respect to an applied loading parameter
is written as v = x (i.e., the material point velocity in dynamic analyses). The spatial gradi-
ent of this material derivative with respect to the current configuration is given by

L =~= av ax = FF- 1 (3)

ax ax ax

3
 Rigid Axes Attached To A
Material Point
Axes Are Material:
Follow the Rotation

r
Axes Are Spatial: Do Not
Follow Deformation

Fixed, Global
Axes

F
+--

Axes Are Material:

Axes Are Spatial: Do Not Follow the Rotation
Follow Deformation

Fig. 1. Initial and deformed configurations illustrating two methods of decomposition.

The symmetric part of L is the spatial rate of the deformation tensor, denoted D; the skew-
symmetric part, denoted W, is the spin rate or the vorticity tensor. Thus,

L=D+W (4)

where

(5)

W represents the rate of rotation of the principal axes of the spatial rate of deformation
D. When integrated over the loading history, the principal values of D are recognized as
the logarithmic (true) strains of infinitesimal fibers oriented in the principal directions if
the principal directions do not rotate. It is important to note that D and W have no sense
of the deformation history; they are instantaneous rates.
Using the RU decomposition of F, the spatial gradient L may be also written in the
form

(6)

in which the following relations are used

:if = RV +RV (7)

and

(8)

The first term in egn (6) is the rate of rigid-body rotation at a material point and is de-
noted Q. The spin rate Wand Q are identical when the principal axes of D coincide with
the principal axes of the current stretch V. Simple extension and pure rotation satisfy this
condition. The symmetric part of the second term in eqn (6) is called the unrotated defor-
mation rate tensor and is denoted d

d = 1- (00- 1 + V-IV) . (9)

2

The unrotated rate of deformation defines a material strain rate relative to the orthogonal
reference frame indicated on configuration B in Fig. l.
Using the orthogonality property of R that d(R TR)/dt = 0

(10)

the unrotated deformation rate may be expressed in the simpler form as

5
 d = RTDR . (11)

The principle of virtual displacements applied in the current (B) configuration readily
demonstrates the work conjugacy of the the spatial rate of deformation, D, and the sym-
metric Cauchy (true) stress, T. Since components of both D and T are defined relative to
the fixed, global axes, the conjugate stress measure for d on the unrotated configuration
is given simply by

(12)

where t is termed the unrotated Cauchy stress, i.e., T is the tensor t expressed on the fixed
global axes.
In subsequent sections, finite-element solutions are considered which employ a Total
Lagrangian (T.L.) description of the motion [4]. Constitutive quantities d and t computed
on the unrotated configuration must be transformed into stress-strain measures required
within the T.L. framework. The T.L. deformation measure is the Green strain given by

(13)

which has the rate

E= ~ CFTF + FTy) = FTDF (14)

...

By using the RU decomposition for F in eqn (14) and the transformation of eqn (11), the
deformation rates are related by

(15)

Upon equating the stress work rates per unit volume in the current(B) and reference (Bo)
configurations. the stress measure conjugate to the Green strain rate is determined to be
the second Piola-Kirchoff stress, denoted S,

(16)

By substituting the RU decomposition of F into eqn (16) and using the transformation in
eqn (12), we find

(17)

Finally, we need the relationship between the rates of t and S. The derivative of S yields

s = m-liv- I + tr(D)S - V-IUS - SUV- I (18)

6
where tr( ... ) denotes the trace of the symmetric tensor D.

3. ELASTO-PLASTIC CONSTITUTIVE FRAMEWORK

3.1 Selection of Strain-Stress Rate
The simplest form of a hypo-elastic constitutive relation is adopted to couple a materially
objective stress rate with a work conjugate deformation rate. The Jaumann and Green-
Naghdi objective stress rates are

TJ = T- Wf+TW = c: D (19a)

(19b)

where the modulus tensor C may depend linearly on the current stress tensor and on histo-
ry dependent state variables. Once the objective stress rate is evaluated using C : D, the
needed spatial rate of Cauchy stress, T, is found by computing W or Q and transposing
the above equations. In a finite-element setting, these rate expressions are numerically
integrated to provide incremental values of the Cauchy stress corresponding to load
(time) steps.
When D vanishes both the Jaumann and Green-Naghdi rates predIcted by the consti-
tutive models also vanish; however, the two stress rates lead to different spatial rates of
Cauchv stress since Wand Q are Qenerallv not identical. Use of the soin tensor W in eon
- -- - - -' '-'" J . . . . . .

(19a) causes the physically unreasonable response predicted for the finite shear problem;
the Green-Naghdi rate leads to a realistic response (see numerical examples below).
The Jaumann rate is adopted extensively in finite-element codes -- the quantity W
is readily available as a by-product of computing D whereas computation ofQ requires
polar decompositions of F. Hughes and Winget [12] recognized that a constant spin rate
W (or rotation rate Q) limits the acceptable step sizes for implicit codes. They developed
a numerical integration scheme for eqn (19a) that retains objectivity of the Jaumann rate
for rotation increments exceeding 30°. Such refinements, however, do not remove the fun-
damental cause (W) of the oscillatory response in simple shear. Roy, et al. [27] recently
implemented a 2-D, implicit finite-element code based on the Green-Naghdi rate as ex-
pressed in eqn (19b). They employed the Hughes-Winget procedure to integrate TGN us-
ing Q computed from polar decompositions of F at the start and end of each load incre-
ment.
The Green-Naghdi rate may be written alternatively as the rate of unrotated Cauchy
stress, i, expressed on the fixed, Cartesian axes

(20)

7
Transformation of the spatial deformation rate D in this expression to the unrotated defor-
mation rate d yields

(21)

Constitutive computations, equivalent to the Green-Naghdi rate in eqn (19b), there-

fore can be performed using stress-strain rates defined on the unrotated configuration.
Updated values of t are rotated via R to obtain the updated Cauchy stress at the end of
a load increment. The numerical problems of integrating the rotation rates in eqns (19a)
and (19b) are thus avoided. Moreover, internal state variables of the plasticity model, e.g.,
the back-stress for kinematic hardening, are also defined and maintained on the unro-
tated configuration and thus never require correction for finite rotation effects. The sim-
plicity derived from this constitutive framework is very appealing and it is adopted in sub-
sequent developments of the finite strain plasticity model. The potential disadvantage of
this constitutive framework is the numerical effort to compute R for use in eqns (11,12)
from the polar decomposition F = RU at thousands of material points for each of many
load steps. This issue is discussed in the section on numerical procedures.
3.2 Plasticity Rate Equations
The incremental plasticity theory considered here assumes initial isotropy of the material
and neglects strain-rate effects. A von Mises yield surface and associated flow rule are
adopted. A mixed isotropic-kinematic hardening model defines subsequent yield sur-
faces. The Mises yield surface is given by

(22)

where ts' is the deviatoric part of the shifted stress vector t s , R is the radius of the yield
surface in deviatoric stress space, and Ep is the effective plastic strain. R is related to the
effective tensile stress Y by

(23)

The shifted stress ts is given by

(24)

where t is the current Cauchy stress on the unrotated configuration and tb is the back-
stress on the unrotated configuration which locates the center of the yield surface (for iso-
tropic hardening, tb = 0).
Further developments require kinematic decomposition of the total strain rate d into
elastic and plastic components. The multiplicative decomposition of the deformation gra-
dient

8
 (25)

appears most compatible with the physical basis of elastic-plastic deformation in crystal-
line metals (see, for example, [3]). FPrepresents plastic flow (dislocations) while Ferepre-
sents lattice distortion; rigid rotation of the material structure may be considered in either
term. Substitution of this decomposition into the spatial rate of the displacement gradient
eqn (3) yields

(26)

We now impose the restriction that elastic strains remain vanishingly small compared
to the unrecoverable plastic strains; a behavior closely followed by ductile metals having
an elastic modulus orders of magnitude greater than the flow stress. Consequently, FP and
Fe are uniquely determined by unloading from a plastic state. This considerably simplifies
the above expression and permits separate treatment of material elasticity and plasticity.
U sing the left polar decomposition and writing the stretch as the product of elastic and
plastic parts yields

(27)

Identifying the elastic deformation as

(28)

and using the small elastic strain assumption, we have

(29)

Consequently, the expression for L is approximated by

L = L e + II . (30)

As in eqn (5), the symmetric part of this approximation for L is taken as D with the result
that

D = De + I)P . (31)

Given the restriction of vanishingly small elastic strains, the multiplicative decomposi-
tion of the deformation gradient in eqn (25) leads to the familiar additive decomposition
of the spatial deformation rate D into elastic and plastic components. The conversion of
D to the unrotated configuration using eqn (11) provides the decomposition scheme need-
ed for d as

9
 (32)

Once the above transformation of elastic and plastic strain rates onto the unrotated con-
figuration is accomplished, remaining steps in development of the finite-strain plasticity
theory are identical to those for classical small-strain theory.
If the elastic strains are not vanishingly small, the incrementally linear form of this hy-
po-elastic material model predicts hysteretic dissipation and residual stresses for some
closed loading paths, for example, the path defined by finite extension -+ finite
shear -+ tension unloading -+ shear unloading [18]. Uncoupled loading-unloading for ex-
tension and shear produces no residual stresses. For finite-strain plasticity of ductile met-
als having large modulus-to-yield stress ratios this situation is not a serious concern since
plastic strains are commonly 50-100 times greater than the elastic strains.
3.3 Stress Updating on the Unrotated Configuration
The plasticity rate equations are numerically integrated over a finite time (load) incre-
mentusing the elastic predictor-radial return algorithm [7,16,19,29,30]. Because integra-
tion of the stress rate occurs with all quantities cast onto the unrotated configuration, al-
gorithmic details of the integration procedure are identical to those developed for con-
ventional small-strain plasticity models. The elastic predictor-radial return method pro-
vides the most accuracy for both single step and subincrementation schemes (the strain
increment is divided into m subincrements with the plasticity integration procedure
applied successively over each subincrement). Moreover, the procedure is unconditional-
ly stable and mixed isotropic-kinematic hardening is easily included.
The plane-stress idealization introduces additional complexities at two levels. First,
the (u, v) nodal displacements do not provide a means to compute the through-thickness
strain increment, tl.d 33 . The updated stress t33 must be zero yet non-zero values of the
back stress, t33(b), and the shifted stress, t33(s), are required to match the Bauschinger ef-
fect predicted by a corresponding 3-D model defined with plane-stress boundary condi-
tions. The elastic predictor-radial return algorithm, for example, can be executed itera-
tively in a 3-D setting to compute simultaneously tl.d 33 and the updated stresses under the
constraint that t33 -+ 0 . Simo and Taylor [31] and Keppel and Dodds [16] provide details
of two such schemes. The second complication introduced by plane-stress involves the
F33 term of the deformation gradient which accounts for finite changes of material thick-
ness due to loading. This term must be constructed from increments of tl.d 33 determined
by the stress update procedure. Computation of F33 is described in the next section.

4. NUMERICAL PROCEDURES FOR FINITE STRAINS

The numerical algorithms in this section are developed for a Total Lagrangian setting.
Only minor modifications are required for use of these same algorithms in an Updated

10
Lagrangian setting. The global solution is advanced from time (load) tn to tn+l using an
incremental-iterative Newton method. Iterations at tn+l to remove unbalanced nodal
forces are conducted under fixed external loading and no change in the prescribed dis-
placements for displacement controlled loading. Each such iteration, denoted i, provides
a revised estimate for the total displacements at t n +l, denoted U~?l. Fully converged dis-
placements at tn are denoted Un. Following Pinsky, et al. [23] a mid-increment scheme
is adopted in which deformation rates are evaluated on the intermediate configuration
at 1/z(un + U~?l) = U~i~lj2 • The choice of 0.5 represents a specific form of the generalized
trapezoidal rule that is unconditionally stable and second-order accurate. Key and Krieg
[17] have demonstrated the optimality of the mid-point configuration for integrating the
rate of deformation and the resulting correspondence with logarithmic strain.
The following sections describe the computational processes performed at each mate-
rial (Gauss) point to: 1) update stresses, 2) provide a consistent tangent matrix for updat-
ing the global stiffness matrix and 3) resolve complications arising from the plane-stress
idealization. A brief discussion of the procedure to compute the polar decomposition of
the deformation gradient is also provided. The organization of a particular finite-element
code dictates which operations are performed in the element dependent routines and
which are performed in the material models; thus no particular distinction is made here.
Standard finite-element procedures to compute deformation gradients at Gauss points
in a T.L. setting are also omitted.
4.1 Stress Updating Procedure
The computational steps are:
Step 1. Compute the deformation gradients at n + V2 and n + 1
F(i) _
!lex + U n(i)+l )
U (i) (i)
n+l - ax I n+1 = det(F n+1) (33)

F(i) _
!lex + U n(i)+112
(1

n+ll2 - ax (34)

Step 2. Compute polar decompositions at n + V2 and n + 1

(35)

(36)

Step 3. Compute the i th estimate for the spatial gradient of the displacement incre-
ment over the step

(37)

where

11
 a(~ 0»)
~F(i) = a( U n(i)+1 - Un
)
= U
(38)
ax ax
Step 4. Compute the i th estimate for the spatial deformation increment over the step

(39)

Step 5. Rotate the increment of spatial deformation to the unrotated configuration

AdO) -- RO)T . A DO) . R O) (40)
il n+lh il n+lh

Step 6. The terms of ~d (i) define the strain increments for use in a conventional
small-strain plasticity model. Invoke the small-strain plasticity model to pro-
vide the i th estimate for the unrotated Cauchy stress at n + 1

(41)

where C denotes the small-strain integration process using the elastic-predic-

tor, radial return algorithm. The integration process requires state variables
at n: the unrotated Cauchy stress, the equivalent plastic strain, and the back-
stresses on the unrotated configuration.
Step 7. The unrotated Cauchy stress at n + lis transformed to the 2nd Piola-Kirchoff
stress at n + 1 as required for the T.L. setting -
U) - JU) .
Sn+1 U(i)-l . t(i) . U O)-l
(42)
- n+l n+l n+1 n+l

Key advantages of the above steps are the absence of half-angle rotations applied to
stresses (and back-stresses) found in co-rotational rate formulations, eqn (19), and most
importantly, the ability to use an existing small-strain plasticity model for Step 6 without
modification since all quantities are referred to the unrotated configuration. The disad-
vantage is the need to perform two polar decompositions for each stress update.
Finally, converged deformation increments ~D are summed over k load steps to de-
fine the logarithmic strains for output
n=k

eij = I !::J)ij (43)

n=l

4.2 Tangent Modulus for Stiffness Updating

A tangent modulus matrix, denoted [C ep ] , is needed to form the element-structure stif-
fness matrix in implicit codes. The moduli couple increments of Green strain with incre-
ments of 2nd Piola-Kirchoff stress required by the IL. formulation. To maintain a qua-
dratic convergence rate of the global Newton iterations, the tangent operator must be
consistent with the numerical algorithm employed to integrate the stress rate just de-
scribed. Consistency implies that the finite stress increment predicted by the tangent oper-

12
ator acting on a strain increment matches, to first order, the stress increment predicted
by the integration procedure.

The small-strain plasticity model provides the consistent tangent modulus [30] that
relates the unrotated stress increments and unrotated deformation increments (in matrix-
vector form)

(44)

The needed form of the above relation for the global T.L. approach is

(45)

where {~EG} is the incremental Green strain. To transform [C;p] ~ [C ep ] , the incremental
forms of the rate transformations in eqns (15) and (18) are employed

(46)

where tr( .. ) denotes the trace of a tensor and

~d = U- I ~EG U- I . (47)

Attempts to combine eqns (46) and (47) into a transformation operator yield a non-
symmetric [C tp ] even though [C;p] is symmetric. Moreover, the resulting expression is un-
necessarily complex and very difficult to express in the matrix form of eqn (45). To pre-
serve the symmetry of [C ep ] , two assumptions are made to develop an approximate trans-
formation operator: (1) the material is incompressible such that tr(~D) ~ 0 and (2) the
term ~U may be neglected in comparison to U and S. With these two assumptions, the
approximate transformation of tangent moduli may be written in matrix form as

(48)

Terms of the 6x6 matrix [T] (for 3-D) are derived from the symmetric, positive definite
matrix U computed from the polar decomposition F = RU at the current configuration.
The 3-D form of [T] is given below. Axisymmetric and 2-D specializations are derived
by omitting the appropriate rows and columns. The row-column ordering of [T] is: x, y,
z, xy, yz, xz. To shorten the notation, we introduce the following terms:

U1 rr-l.
= U11,
U rr1;
2 = u21
U rr-l.
3 = u22,
U
4 =
U-I.
31,
U
5 = rr-l.
u32,
U rr-1
6 = u33 . (49)

13
With this notation, [T] is given by

u 21 u 22 u~ 2u 1U2 2u2U4 2ulU4

u 22 u 32 us2 2u 2U 3 2u3US 2u 2U s

[T] = u42 u S2 U~ 2u4US 2uSU6 2u 4U 6 (50)

UIU2 U2 U 3 U4 US + U2
UIU3
2
U4 U 3+ U2US UIUS + U2U4
U2 U4 U3 U S USU6 U2U S + U4 U 3 U3U6 + U~ U2 U 6 + U4 U S

UIU4 U2 U S U4 U6 UIUS + U2 U4 U2U 6 + U4 US UIU6 + U~

Numerical tests demonstrate that this approximate transformation of tangent moduli

maintains the convergence rate of the global Newton iterations (subsequently discussed
example problems show this). Use of the consistent moduli for the unrotated configuration
in eqn (44) appears more important for good convergence rates than the purely geometric
transformation approximated by eqn (48).

4.3 Plane-Stress Idealization

The F 31 , F 32 , F 13 , and F23 terms ofF vanish for motion restricted to the 1-2 plane (plane-
stress, plane-strain, and axisymmetric idealizations). The FIb F 12, F 21 , and F22 terms are
determined from the in-plane displacements. The F33 term is necessary for computation
of J = det(F); J appears in the stress and tangent moduli transformations, eqns (42) and
(43). For plane-strain analyses F33 = 1; for axisymmetric analyses F33 = (Ro + u )/Ro where
Ro is the undeformed radius of the material point.
For plane-stress conditions, F33 is simply the current thickness, T, divided by the unde-
formed thickness, To,

F33 = ~ = To + ~T (51)
To To
The change in thickness is obtained by integrating the unrotated deformation rate over
the loading history to define the through-thickness logarithmic strain

InCl + ~~) = e33 = I I d 33 = D33 = I tJ.d 33 (52)

where equivalence of the (3,3) spatial deformation and (3,3) unrotated deformation terms
is noted for motion in the 1-2 plane. The solution of the above expression for ilT and
the substitution into eqn (51) provides the needed expression for F33 as

(53)

The term I~d33 is maintained as a history dependent quantity at each Gauss point in
the same manner as the accumulated plastic strain fp .

14
4.4 Polar Decomposition
The polar decomposition F = RU is a key step in the stress-updating algorithm and must
be performed twice for each Gauss point for each stress update, i.e., at n + 1;1 and n + 1.
The computational effort required for the polar decomposition should be insignificant
relative to the element stiffness computation and the equation solving effort. For their
explicit code, Flanagan and Taylor [8] developed an algorithm for the integration of
R = QR that maintains orthogonality of R for the very small displacement increments
characteristic of explicit solutions. Numerical tests readily show their procedure fails for
large displacement increments experienced with implicit global solutions. The following
algorithm removes such approximations and yet remains computationally very efficient
with the framework of an implicit solution.
Step 1. Compute the right Cauchy-Green tensor

(54)

and its square

(55)

where only the upper-triangular form of the symmetric products (6 terms) are
actually computed and stored.
Step 2. Compute the eigenvaluesl! ,l~ andl~ of C. A Jacobi transformation proce-
dure specifically designed for 3x3 matrices is used to extract the eigenvalues.
Do-loops are eliminated by explicitly coding each off-diagonal rotation form.
Two or, at most, three sweeps are needed to obtained eigenvalues converged
to a 10-6 tolerance.
Step 3. Compute invariants of U and the det(F)

(56a)

(56b)

(56c)

Step 4. Form the upper triangle of the symmetric, right stretch, U, and it's symmetric
inverse, V-I (see [11])

(57a)

where I denotes a unit tensor with the f3 coefficients defined by

f31 = l/(/ullv - IIIv) , {32 = Iulllu, {33 = I~ - IIu (57b)

15
 Similarly, the inverse of U may be formed directly as

(57c)

where the y coefficients defined by

Yl = l/IIIu(/ullu-Illu), Y2 = Iu1f2u- 111u(/t;+IIu), (57d)

Y3 = - IIIu - IuCPu - 2IIu) , Y4 = Iu

Step 5. Form R as the product

R= FU- 1 (58)

The FORTRAN code listing for the above procedure is given in the appendix. Table 1
summarizes the relative computational effort required for (1) generation of the element
tangent stiffness matrix, (2) stress updating at all Gauss points of the element (including
two polar decompositions at each Gauss point), and (3) the relative time required for a
single polar decomposition. Results are given for an axisymmetric, 8-node isoparametric
element and a 3-D, 20-node isoparametric brick element. Both elements employ reduced
integration rules; 2x2 for the axisymmetric element and 2x2x2 for the 3~D element. Com-
putations were performed on a Unix workstation. The CPU time required for generation
of the element tangent stiffness is assigned a unit value for each case. The results clearly
demonstrate that polar decompositions are not a computational issue in an implicit code.

Table 1. Relative Computational Effort Required for Polar Decompositions

Computation 8-Node Axisymmetric 20-Node 3-D
(2x2 Gauss Rule) (2x2x2 Gauss Rule)
Element [KT ] 1.0 1.0
Element Stress Updating 0.46 0.15
Single Polar Decomposition 0.01 0.0005

5. NUMERICAL EXAMPLES

Numerical results for three example problems are presented in this section. The examples
demonstrate the excellent performance of the finite strain model for both 2-D and 3-D

16
configurations. The first two examples consider finite, homogeneous deformation in un-
coupled extension and simple shear. By adopting an incrementally-linear material, ana-
lytical solutions may be constructed for these two problems to assess the accuracy of the
finite-element solutions. The third example considers the plane-strain, Mode I small-
scale yielding problem for a material that follows the rate independent, incremental plas-
ticity theory. An initially blunt notch tip is opened to several times the initial width in a
boundary layer model that approximates the conditions at a crack tip in an infinite body
The finite-strain model is implemented as a pre- and post- processor for the existing
small-strain plasticity model in our research finite-element system (POLO-FINITE [6]).
The small~strain model required no changes; we consider this to be a significant advan-
tage of adopting the unrotated configuration to perform constitutive computations.
5.1 Homogeneous Finite Extension
Consider a unit block of material aligned with edges parallel to the coordinate axes (see
Fig. 2). The block is constrained and loaded consistent with uniaxial tension in the Xl
direction. The displacement field is given by

(59)

where a, k are constants and t is a time-like loading parameter that increases monotoni-
cally from zero. For a unit cube, a may be taken as unity. The coordinate stretch ratios
are then

(60)

In the absence of rotation, the unrotated Cauchy, Jaumann, and Green-Naghdi stress
rates are identical, i.e., R = I, F = U, d = D, and W = O. The hypo-elastic relations all have
the form

t = AItr(d) + 2Jld (61)

where A and Jl are the Lame constants. Direct integration of these relations yields

(62)

for the Cauchy stresses. The stretch ratios are related by

(63)

from which the corresponding axial force is found to be (for a cube with unit initial edge
lengths)

(64)

17
 1.0 :::l 2.0

0.9 ~
• Axial Force, Finite-Element (3-~ & Plane-Stress)
A Axial Strain, Finite-Element

O.B ...
Exact
~ 1 ~
.~

cd
1.5 ~
~
0.7 rrJ
~ ~

..... 0.6 cd

~
0)
U
~
0.5 1.0
X2
~ 1 Deformed
U
.~

~
~
0.4 S
~
f-'
(X)

~ 0.3
I1 ~
.~
~
0.5 ~
bI)
0.2 ... / / L 0

0.1 J-/' Thickness = 1 b= 1 ---I

Al J
~

0.0 , I 0.0
1 2 3 4 5 6 7

Stretch Ratio, A1
Fig. 2. Homogeneous finite extension. Comparison of finite-element results with analytical solutions.
Incremental, linear-elastic material: E = 1, v = 0.3.
where E and v are Young's modulus and Poisson's ratio, respectively. This load-stretch
response, plotted in Fig. 2, exhibits a maximum load effect ap1/ aAI = 0 at a critical stretch
ratio of Al = 5.29.
Plane-stress and 3-D finite-element models are analyzed for this finite extension
problem. The plane-stress model contains four, linear quadrilateral elements; the 3-D
model contains eight, linear brick elements. The integration order for the plane-stress
elements is 2x2 with a 2x2x2 order used for the 3-D elements. Constitutive computations
are performed by the incremental plasticity model with a yield stress large enough to pre-
vent plastic deformation. Numerical results for these two models should be identical pro-
vided the through-thickness strain computations described in eqns (51) and (53) are per-
formed during solution of the plane-stress model. Both models are loaded by displace-
ments imposed on the face Xl = 1. Thirty (30) equal size increments are imposed to reach
the deformation Al = 7. Iterations at each load step are performed until the convergence
test given by

I R II ~ I p II * 10-4 (65)

is satisfied, where I I denotes the Euclidean norm, R is the residual force vector, and P
is the vector of total reactions at the constrained nodes. For both the plane-stress and
3-D solutions, a total of 60 iterations are performed for the 30 load steps; two iterations
are needed for convergence at each load step. The plane-stress and 3-D solutions are
identical. Figure 2 compares the analytical and computed axial forces as a function of the
stretch ratio. The figure also compares the exact axial strain, In AI, with the finite-element
approximation obtained by summing deformation increments as in eqn (43). The maxi-
mum error in predicted axial force is 0.3 % while the maximum error in the predicted loga-
rithmic strain is 0.1 %.
5.2 Homogeneous Finite Shear
In the finite simple shear problem discussed by Dienes [5], material undergoes simulta-
neous stretching and large rotation. Analytical solutions for each stress rate are now avail-
able for assessment of numerical implementations. A unit cube of material is again
employed as shown in Fig. 3. The displacement field is given by

(66)

where a is a constant and t is a time-like loading parameter that increases monotonically

from zero. The deformation gradient and material displacement derivative are simply

1 at 0 o a 0
F = 0 1 0 L= 0 0 0 (67)
o o 1 o 0 0

19
 3.0

2.5
1X 2

~ Deformed Shape
/--------;#

2.0
I1 /

/
/
/
//
/

/
/
Unrotated Cauchy
1.5
L //
/

X1 Rate, Eqn (73)

l.-1--.i
1.0 Finite-Element
T12
• 8 Increments

N
o
0.5 - ......
......
......
,, /
/
A 16 Increments

0.0 ,, /
/

,, /
/
/

...... /
...... /

~)y -
-0.5
Jaumann Rate, Eqn (71
-1.0
o 1 2 3 4 5 6 7 8 9 10

Shear Strain, Yxy

Fig. 3a. Homogeneous finite simple shear. Comparison of finite-element and analytical solutions
for the Cauchy shear stress. Incremental, linear-elastic material: E = 1, v = 0.3.
F shows that the deformation is isochoricwithl = 1. Since motion is restricted to theX1-X2
plane, R must have the form

cos f3 sin f3 0
R = - sinf3 cosf3 0 (68)

o o 1
where the angle f3 is given by

f3 = tan- 1 (at/2) . (69)

The rate of deformation in the fixed Cartesian system, eqn (5), and the unrotated configu-
ration, eqn (11), are

o 1 0 - sin2fJ cos 2f3 0

a a (70)
D =- 1 0 0 d = - cos2f3 sin 2f3 0
2 2
o 0 0 o 0 0
Dienes [5] adopted the incrementally-linear, eqn (61), for the Jaumann and unrotated
stress rates. The constitutive models are thus

TJ = Altr(D) + 2.£lD (71a)

i = Altr( d) + 2.£ld (71b)

The integration of eqn (71a) yields the following solution for the Cauchy stresses

TIl = - T22 = .£l(1 - cos at), T12 =.£l sin at (72a)

while the integration of eqn (71b) yields the unrotated Cauchy stresses as

t11 = - t22 = 4.£l1n(cosfJ), t12 = 2.£l(2fJ - tanfJ) (72b)

These stresses are rotated to the fixed Cartesian system using R from eqn (68) to yield

TIl = - T22 = 4.£l[cos2fJln(cosfJ) +fJsin2fJ- sin2 fJ],

T12 = 2.£l cos 2fJ[2fJ - 2 tan 2fJ In (cosfJ) - tanfJ] (73)

Figure 3 compares the analytical solutions for the Cauchy stresses obtained using the
Jaumann rate and the unrotated Cauchy rate. The solution for the Jaumann rate exhibits
a physically unacceptable harmonic oscillation while the solution for the unrotated

21
Fig. 3b. Homogeneous finite simple shear. Comparison of finite-element and analytical solutions
for the Cauchy normal stress. Incremental, linear-elastic material: E = 1, v = 0.3.
Cauchy rate increases monotonically with increasing deformation. The Green-Naghdi
rate yields the same solution for the stresses as eqn (73).
The finite-element model for the finite shear problem contains four, linear quadrilat-
eral elements with an integration order of 2x2. To maintain the isochoric deformation de-
scribed by eqn (66), the model must be loaded by prescribing the displacements at all
nodes. Thus no iterations are necessary. The logarithmic shear strain, Yxy, is increased to
a magnitude of 8 in separate analyses using 8 and 16 equal size displacement increments.
For a unit cube, the shear strain equals the imposed displacement along the edge X 2 = l.
The computed Cauchy stresses for these analyses are compared with the exact solutions
in Fig. 4. The finite-element stresses very closely match the exact solution and show very
minor dependence on the load-step magnitude. In the first of 8 increments, the error is
3%; as Yxy ~ 8 the error decreases to less than 0.1 %.
5.3 Crack-Tip Blunting In Small-Scale Yielding
Small-scale yielding (SSY) in Mode I plane-strain characterizes the deformation near a
crack tip in an infinite body. Rice and Tracey [26] and McMeeking [20,21] developed a
boundary-layer approximation for the infinite body model that is suitable for finite-ele-
ment analysis. The SSY model consists of an annular region containing either a sharp or
smoothly blunt crack tip which is subjected to increasing displacements of the elastic
(Mode I) singular field on the outer circular boundary. SSY mod~ls are commonly
employed in fracture mechanics studies to investigate continuum based, micromechanics
parameters that describe the initiation of ductile crack growth. Very efficient finite-ele-
ment models that include the effects of large strains and large material rotation are essen-
tial for studies that investigate such parameters.
Figure 4 shows the inner portion of the plane-strain finite-element model developed
to solve the SSY problem. The crack is modeled as a notch of initial width bo having a
semi-circular tip. The mesh extends to a radius R = 500Obo and contains 2328 nodes, 737
8-node isoparametric elements. The smallest element at the notch tip has length bo /12.
The use of reduced (2x2) integration eliminates locking due to the incompressible plastic
deformation. The uniaxial, true stress-logarithmic strain curve follows a power-law form

(74)

where the material constants selected for analysis are: fo = 0.002, ao = 60, n = 10, a = 1,
and Poisson's ratio v = 0.3. Symmetrical boundary conditions are applied on the crack
plane (Xl ~ bo , X 2 = 0). The notch surface remains traction free. Displacement incre-
ments of the elastic K]- field for Mode I are imposed on the outer circular boundary. The
boundary displacements are increased monotonically in 40 equal increments to the level
KJ / (cTa R) = 1.11 at which point the plastic zone extends = RI10. Iterations at fixed KJ are
performed until the convergence test, eqn (65), with a tolerance of 5 x 10-4 is satisfied.
This tolerance insures 0.1 % convergence of the strains for this problem.

23
 R = SOOOb o
X2
2328 Nodes
737 Elements
t KI Displacement Field Imposed
on Remote Boundary

N
+:--

' - Symmetry Conditions

.
X1

Fig. 4. Finite-element model (plane-strain) for boundary layer idealization of the small-scale yielding problem.
 Figure 5 shows the deformed near-tip region at increasing levels of the notch opening.
Equivalent plastic strains in the notch-tip element exceed 2 at the maximum load. Figure
6 shows the opening mode Cauchy stress, aee , and equivalent plastic strain ahead of the
notch tip on the crack plane. The radial distance is normalized by J/ ao where J is the value
of Rice'sJ-integral [24]. A domain integral method [28] is used to extract J from the nu-
merical solution; for SSY conditions J = K1(1 - v 2) / E. The notch opening b can be esti-
mated as 0.5J/ao ; the horizontal axis thus spans 0 ~ lOb. Outside the 'blunting' zone
of size r = 3b, the present stresses achieve a steady-state condition which scales with
J/ao ' Once b/b o > 3, excellent agreement is observed between the present stresses and
those of a conventional small-strain model containing a sharp crack tip (modeled by sin-
gularity elements). The plastic strain distribution, shown for maximum load in Fig. 6, re-
veals that the zone of finite strains extends to r = 3b beyond which the strains are a only
a few multiples of the yield strain Eo. The small-strain, asymptotic stresses (HRR) of
Hutchinson [13] and Rice and Rosengren [25] are shown for comparison. The present fi-
nite strain results for an initially blunt notch and those for a small-strain model with an
initially sharp notch both fall below the HRR solution at increasing distances from the
tip, i.e., the HRR solution contains only the leading term of the full SSY field and is correct
only for r ~ 0 .
Each of the 40 loading increments required an average of3 iterations for convergence.
Elements at the notch-tip sustain strain increments of about 4.3 % or 21 x Eo. To gauge
the convergence rate, this problem was analyzed using a conventional small-strain plastic-
ity model-the same number of iterations were required for convergence. The finite-
strain solution required 10% more CPU time than the corresponding small-strain analy-
sis. A solution using 200 increments produced essentially no difference in the stresses and
a maximum 1.5 % difference in notch-tip strains. A solution with 20 increments converged
without difficulty but with a loss of accuracy in notch-tip strains.

6. SUMMARY AND CONCLUSIONS

The numerical implementation of a flow theory plasticity model suitable for large strains
and large rotations has been presented. The constitutive computations are formulated us-
ing strains. stresses. and their rates cast on the unrotated reference frame thereby remov-
ing many of the complicating details of previous finite strain models. Polar decomposition
of the deformation gradients is employed to establish the unrotated reference frame; the
resulting constitutive model is equivalent to one based on the Green-Naghdi stress rate
but much simpler to implement numerically. The stress updating process is developed for
an implicit, Total Lagrangian formulation of the finite-element method. Details of the
various transformations and the consistent tangent modulus are given including a very ef-
ficient algorithm for 3-D polar decomposition. This finite-strain material model retains

25
 (a) (b)
Note: Deformations to Scale
N
~

b/b o Epl @ Notch-Tip

b/21 (a) 2.13 1.00

wtt1l (b) I 3.36 I 1.58
\... Notch-Tip
(C) (c) I 4.96 I 2.08

Fig. 5. Crack-tip deformations for increasing applied load in the SSY model.
 5 0.5

o Small-Strain Solution
',I HRR (Sharp-Tip Model)
0.4
4- " .......
.....

1 I. )II '-J
-------------------
-~~ . . no
-I 0.3
a()()

aO
J r X7 u- au~ Epl
b/b o KJ /(eJ2oR) -I 0.2
N
'-.J I JTl r
• 2.13 0.278
1')
~.
I- 4 • 3.36 I 0.625 ~ ~ 0.1
• 4.96 I 1.11
Results for ()= 0

1 0.0
o 1 2 J 4 5

rj(J jao )
Fig. 6 Comparison of finite-element results for initially blunt tip with asymptotic HRR fields and with sharp-tip
models (r and 0 are coordinates in initial configuration).
the full numerical architecture of a conventional small-strain plasticity model with iso-
topic-kinematic hardening and has been implemented as a pre- and post- processor for
such a model in our finite-element code.
Numerical tests demonstrate that for an implicit global solution, the computational
effort required for the polar decompositions is insignificant relative to the effort required
for updating element stiffnesses. Three numerical examples illustrate the acceptable re-
sponses predicted by the material model for simple homogeneous deformation of an in-
cremental, linear-elastic material and for ductile fracture analyses of a material following
incremental plasticity. Large step sizes are accommodated without undue loss of solution
accuracy or convergence rate of the global equilibrium iterations.

7. REFERENCES

1. ABAQUS User's Manual, Version 4.8, Hibbitt, Karlsson & Sorensen, Inc., Providence R.I.,
1990.

2. Atluri, S. N., "On constitutive relations at finite strain: hypo-elasticity and elasto-plasticity
with isotropic and kinematic hardening," Computer Methods in Applied Mechanics and Engi-
neering, Vol. 43, 1984, pp. 137-171.

3. Asaro, R. J., "Crystal Plasticity," Journal of Applied Mechanics, 50, 1984, pp. 1-12.

4. Bathe, K. J., "Finite element procedures in engineering analysis," Prentice-Hall, 1982.

5. Dienes, J. K., "On the analysis of rotation and stress rate in deforming bodies," Acta Mechan-
ica, Vol. 32, 1979, pp. 217-232.

6. Dodds, R. H., and Lopez, L. A., "Software virtual machines for development of finite-ele-
ment systems," International J. for Engineering with Computers, Vol. 13, 1985, pp. 18-26.

7. Dodds, R. H., "Numerical techniques for plasticity computations in finite-element analysis,"

Computers & Structures, Vol. 26, No.5, 1987, pp. 767-779.

8. Flanagan, D. P., and Taylor, L. M., ':.c\.n accurate numerical algorithm for stress integration
with finite rotations," Computer Methods in Applied Mechanics and Engineering, Vol. 62, 1987,
pp. 305-320.

9. Green, A. E., and Naghdi, P. M., '~general theory of an elastic-plastic continuum," Arch.
Rat. Mech. Anal., Vol. 18, No. 1965, pp. 251.

10. Hibbitt, H. D., Marcal, P. V, and Rice, J. R., ':.c\. finite element formulation for problems of
large strain and displacement," International Journal of Solids and Structures, Vol. 6, 1970, pp
1069-1086.

11. Hoger, A., and Carlson, D. E.,"Determination of the stretch and rotation in the polar decom-
position of the deformation gradient," Quarterly of Applied Mathematics, Vol. 10, 1984.

28
12. Hughes, T. J. and Winget, J., "Finite rotation effects in numerical integration of rate constitu-
tive equations arising in large-deformation analysis," International 1. for Numerical Methods
in Engineering, Vol. 15 (12), 1980, pp. 1862-1867.
13. Hutchinson, J. W, "Singular behavior at the end of a tensile crack in a hardening material,"
Journal of the Mechanics and Physics of Solids, Vol. 16, 1968, pp. 13-31.
14. Jaumann, G., "Geschlossenes system physikalischer und chemisher differentialgesefze," Sitz
Zer. Akad. Wiss. Wein, (IIa), Vol. 120, 1911, pp. 385.
15. Johnson, G. C. and Bammann, D. J., '~discussion of stress rates in finite deformation prob-
lems," International 1. for Solids and Structures, Vol. 20 (8), 1984, pp. 725-737.
16. Keppel, M. and Dodds, R. H., " Improved numerical techniques for plasticity computations
in finite-element analysis," Computers & Structures, Vol. 36, No.1, 1990, pp. 183-185.
17. Key, S. Wand Krieg, R. D., "On the numerical implementation of inelastic time dependent
and time independent, finite strain constitutive equations in structural mechanics," Comput-
er Methods in Applied Mechanics and Engineering, Vol. 33, 1982, pp. 439-452.
18. Kojic, M., and Bathe, K. 1., "Studies of finite-element procedures: stress solution of a closed
elastic strain path with stretching and shearing using the Updated Lagrangian J aumann for-
mulation," Computer and Structures, Vol. 26, No. 1/2, 1987, pp. 175-179.
19. Krieg, R. D., and Key, S. W "Implementation of a time independent plasticity theory into
structural computer programs. In Constitutive Equations in Viscoplasticity: Computational
and Engineering Aspects, AMD-20 (Edited by J. A. Stricklin and K. J. Saczalski), ASME, New
York, 1976, pp. 125-138.
20. McMeeking, R. M., and Rice, 1. R., "Finite-element formulations for problems of large elas-
tic-plastic deformation," International Journal of Solids and Structures, Vol. 11, 1975, pp
601-616.
21. McMeeking, R. M., "Finite deformation analysis of crack-tip opening in elastic-plastic ma-
terials and implications for fracture/' Journal of the Mechanics and Physics of Solids, Vol. 25,
1977, pp. 357-381.
22. Nagtegaal, 1. C., and de long, 1. E., "Some computational aspects of elastic-plastic, large
strain analysis," International J. for Numerical Methods in Engineering, Vol. 12, 1981, pp.
15-41.
23. Pinsky, PM., Ortiz, M., and Pister, K. S., "Numerical integration of rate constitutive equa-
tions in finite deformation analysis," Computer Methods in Applied Mechanics and Engineer-
ing, Vol. 40, 1983, pp. 137-158.
24. Rice, J. R.,'~ path independent integral and the approximate analysis of strain concentra-
tions by notches and cracks," Journal of Applied Mechanics, Vol. 35, 1968, pp. 379-386.
25. Rice, J. R., and Rosengren, G. E, "Plane strain deformation near a crack tip in a power law
hardening material," Journal o/the Mechanics and Physics of Solids, Vol. 16, 1968, pp. 1-12.
26. Rice, J. R., and Tracey, D. M., "Computational fracture mechanics," in Numerical and Com-
puter Methods in Structural Mechanics, eds. SJ. Fenves, et al., Academic Press, New York,
1968, pp. 585-623.

29
27. Roy, S., Fossum, A. F., and Dexter, R. J., "On the use of polar decomposition in the integra-
tion of hypo--elastic constitutive laws," submitted for publication.

28. Shih, C. F., Moran, B., and Nakamura, T., "Energy release rate along a three-dimensional
crack front in a thermally stressed body," Internationall of Fracture, Vol. 30, 1986, pp.
79-102.

29. Simo, 1. C., and Ortiz, M., ''A unified approach to finite deformation elasto-plastic analysis
based on the use of hyper-elastic constitutive equations," Computer Methods in Applied Me-
chanics and Engineering, Vol. 49, 1985, pp. 221-245.

30. Simo, J. C., and Taylor, R. S., "Consistent tangent operators for rate-independent elasto-
plasticity," Computer Methods in Applied Mechanics and Engineering, Vol. 48, 1985, pp.
101-118.

31. Simo, J. C., and Taylor, R. S., ''A return mapping algorithm for plane stress elasto-plasticity,"
Internationall of Numerical Methods in Engineering, Vol. 22,1986, pp. 649-670.

32. Taylor, L. M. and Flanagan, D. P., "PRONTO 2D, a two-dimensional transient solid dynam-
ics program," SAND86-0594, Sandia National Laboratories, Albuquerque, NM, 1987.

30
 APPENDIX A

FORTRAN CODE FOR POLAR DECOMPOSITION

31
 c ****************************************************** ev(3) sqrt(ev(3»
c * * c
c * mtmplr -- polar decomposition of the deformation * iu ev(1)+ev(2)+ev(3)
c * gradient into the rotation tensor and * iiu ev(1)*ev(2)+ev(2)*ev(3)+ev(1)*ev(3)
c * the right stretch tensor. (3x3) * iiiu ev(1)*ev(2)*ev(3)
c * * detf iiiu
c ****************************************************** c
c c compute the right stretch tensor.
c c
subroutine mtmplr( f. r, u, ui, datf ) a1 1.0/(iu*iiu-iiiu)
implicit intarer (a-l) b1 iu*iiiu
real ( 3 , 3 ) . r(3,3), u(e), ui(6), e(6), ee(6), c1 iu*iu-iiu
& ev(3), et(e), iu, iiu, iiiu, a!. bl, cl, a2, c
& b2, e2, d2, date u (1) a1 * b1 + c1*c(1) - cc(l)
c u(2) a1 * c1*c(2) - cc(2)
c C,CC,u and ui are in symmetric u(3) al * bl + c1*c(3) - cc(3)
c upper triangular form. u(4) a1 * cl*C(4) - cc(4)
c u(5) a1 * c1*c(5) - cc(5)
c compute the metric tensor. u (6) a1 * b1 + c1*c(6) - CC(6)
c c
c(l) f(l,l)*f(l,l)+f(2,l)*f(2,l)+f(3,l)*f(3,l) c compute the inverse of the right
c(2) f(l,l)*f(l,2)+f(2,l)*f(2,2)+f(3,l)*f(3,2) c stretch tensor.
c(3) f(l,2)*f(l,2)+f(2,2)*f(2,2)+f(3,2)*f(3,2) c
c(4) f(1,1)*f(l,3)+f(2,l)*f(2,3)+f(3,1)*f(3,3) a2 1.0/(iiiu*(iu*iiu-ii u»
c(5) f(1,2)*f(l,3)+f(2,2)*f(2,3)+f(3,2)*f(3,3) b2 iu*iiu*iiu-iiiu*(iu* u+iiu)
c (6) f(1,3)*f(1,3)+f(2,3)*f(2,3)+f(3,3)*f(3,3) c2 -iiiu-iu*(iu*iu-2.0* iu)
c d2 iu
W ct(l) c(l) c
N
ct(2) c (3) ui(l) a2 * b2 + c2*c(1) + d2*cC(1)
ct(3) c (6) ui (2) a2 * c2*c(2) + d2*cC(2)
ct(4) c (2) ui(3) a2 * b2 + c2*c(3) + d2*cc(3)
ct(5) c(5) ui(4) a2 * c2*C(4) + d2*cc(4)
ct(6) c(4) ui(5) a2 * c2*C(5) + d2*cc(5)
c ui(6) a2 * b2 + c2*c(6) + d2*cc(B)
c compute the square of the metric tensor. c
c c compute the rotation tensor.
cc(l) c(1)*C(1)+c(2)*c(2)+C(4)*c(4) c
cc (2) c(1)*c(2)+c(2)*c(3)+c(4)*c(5) r(l,l) f(1,1)*ui(1)+f(1,2)*ui(2)+f(1,3)*ui(4)
cc(3) c(2)*C(2)+c(3)*c(3)+C(5)*c(5) r (1,2) f(1,1)*ui(2)+f(1,2)*ui(3)+f(1,3)*ui(5)
cc(4) c(1)*C(4)+c(2)*c(5)+c(4)*c(6) r(1,3) f(1,1)*ui(1)+f(1,2)*ui(5)+f(1,3)*ui(B)
cc(5) c(2)*C(4)+c(3)*c(5)+c(5)*c(6) r(2,1) f(2,1)*ui(1)+f(2,2)*ui(2)+f(2,3)*ui(4)
cc(6) c(4)*C(4)+c(5)*c(5)+c(6)*c(6) r(2,2) f(2,1)*ui(2)+f(2,2)*ui(3)+f(2,3)*ui(5)
c r(2,3) f(2,l)*ui(4)+f(2,2)*ui(5)+f(2,3)*ui(6)
c compute the principal values of the r(3,1) f(3,1)*ui(1)+f(3,2)*ui(2)+f(3,3)*ui(4)
c metric tensor. r (3,2) f(3,1)*ui(2)+f(3,2)*ui(3)+f(3,3)*ui(5)
c r(3,3) f(3,1)*ui(4)+f(3,2)*ui(5)+f(3,3)*ui(B)
call mtmevd( ct, ev c
c return
c compute the invariants of the right end
c stretch tensor. the determinant of c ******************************************************
c deformation tensor is product of c *
c right stretch eigenvalues. c * mtmevd -- compute eigenvalues of 3x3 symmetric
*
c c * matrix stored in packed format *
ev(l) sqrt(ev(l» c * *
ev (2) sqrt(ev(2» c
*
******************************************************
 c c
10 swpnum = swpnum + 1
c thold = 0.0001 ** swpnum
subroutine mtmevd( k, lamda ) sqtol = jactol * jactol
implicit integer (a-z) if ( thold .It. sqtol ) thold sqtol
real k(l) ,lamda(l) ,kbari,kbarj,kbar,ki, c
& kj,mi,mj,scale,alpha,gamma,x,xsign,jactol, c enter sweep loop -- work on lower triangle
& thold,sqtol,ratiok,rad,errork,swap, c only. rows are done from top to bottom
& m1,m2,m3,k1,k2,k3,k4,ks,k6 c columns are done from left to right.
logical cvgtst c skip when already within tolerance.
data maxswp/15/.jactol/1.0e-041 c
c ratiok = (k4*k4) I ( k2*k1 )
c initialize lamda. m, sweep parameters. if ( ratiok .It. thold ) go to 15
c kbari -m2 * k4
m1 1.0 kbarj -m1 * k4
m2 1.0 kbar k2 * m1 - k1 * m2
m3 1.0 rad ( kbar * kbar I 4.0 ) + kbari * kbarj
k1 k(l) xsign 1. 0
k2 k(2) x kbar I 2.0 + sign(xsign, kbar) * sqrt(rad)
k3 k(3) if ( (abs (x) .It. j actol*abs (kbarj» . or .
. k4 k(4) & (abs(x).lt.jactol*abs(kbari») then
ks k(5) alpha 0.0
k6 k (6) gamma = -k4 I k2
lamda(l) k1 else
lamda(2) k2 alpha kbarj I x
lamda(3) k3 gamma -kbari I x
swpnum 0 end if
w c ki k5
w c scale [k] to avoid potential problems kj k6
c with exponential overflow and underflow. k5 ki + gamma * kj
c ka kj + alpha * ki
kj k1 kj k1
kj mine k2,kj mj m1
kj mine k3,kj ki k2
ki k1 mi m2
ki max( k2,ki k1 kj + alpha * alpha * ki + 2.0 * alpha * k4
ki max( k3,ki m1 mj + alpha * alpha * mi
mj 1.0 k2 ki + gamma * gamma * kj + 2.0 * gamma * k4
mi 1.0 m2 mi + gamma * gamma * mj
c k4 0.0
c compute the scale factor and do the scaling c
c c row 3 and column 1
iexp int ( ( 10g10(kj)+log10(ki) ) * 0.2S c ------------------
scale 1. 0 I ( 10.0 ** iexp ) c
m1 m1 * scale 15 ratiok = ( k6*k6 ) I ( k3*k1 )
m2 m2 * scale if ( ratiok .It. thold ) go to 20
m3 m3 * scale kbari -m3 * ka
k1 k1 * scale kbarj -m1 * ka
k4 k4 * scale kbar k3 * m1 - k1 * m3
k2 k2 * scale rad ( kbar * kbar I 4.0 ) + kbari * kbarj
k6 ka * scale xsign 1.0
k5 k5 * scale x kbar I 2.0 + sign(xsign, kbar) * sqrt(rad)
k3 k3 * scale if ( (abs(x).lt.jactol*abs(kbarj» .or.
c & (abs(x) .It.jactol*abs(kbari)) ) then
c begin a new sweep alpha = 0.0
 gamma -k6 / k3 c
else c end of sweep loop
alpha kbarj / x c
gamma -kbari I x c
end if c update eigenvalue vector -- lamda
ki k5 c
kj k4 25 lamda(l) kl / ml
k5 ki + gamma • kj larnda(2) k2 / m2
k4 kj + alpha • ki larnda(3) k3 / rn3
kj kl c
mj ml c check off-diagonal elements for convergence
ki k3 c
mi m3 c
k1 kj + alpha • alpha • ki + 2.0 • alpha * k6 cvgtst = .true.
ml mj + alpha * alpha * mi c
k3 ki + gamma * gamma * kj + 2.0 * gamma * k6 errork = k4 * k4 / ( k2 * kl )
m3 mi + gamma * gamma * mj if ( errork .gt. sqtol ) cvgtst .false.
k6 0.0 c
c
errork = k6 * k6 / ( k3 * kl )
c row 3 and column 2 if ( errork .gt. sqtol ) cvgtst
c .false.
c
c
20 ratiok = ( k5*k5 ) / ( k3*k2 ) errork = k5 * k5 / ( k3 * k2 )
if ( ratiok .It. thold ) go to 25 if ( errork .gt. sqtol ) cvgtst .false.
kbari -m3 * k5 c
kbarj -m2 * k5 if cvgtst) go to 30
kbar k3 * m2 - k2 * m3 if swpnum .It. maxswp ) go to 10
w rad ( kbar * kbar / 4.0 ) + kbari * kbarj c
~ c
xsign 1.0 eigenvalues have converged. reorder and exit.
x kbar / 2.0 + sign(xsign, kbar) * sqrt(rad) c
if ( (abs(x).lt.jactol*abs(kbarj» .or. 30 if (lamda(2) .It.lamda(I)) then
& (abs(x) .It.jactol*abs(kbari)) ) then swap lamda(l)
alpha 0.0 lamda(l) lamda(2)
gamma -k5 / k3 lamda(2) swap
else end if
alpha kbarj / x c
gamma -kbari / x if ( lamda(3) .It. lamda(l) ) then
end if swap lamda(l)
ki k6 lamda(l) lamda(3)
kj k4 lamda(3) swap
k6 ki + gamma * kj end if
k4 kj + alpha * ki c
kj k2 if ( lamda(3) .It . lamda(2) ) then
mj m2 swap lamda(2)
ki k3 lamda(2) lamda(3)
mi m3 lamda(3) swap
k2 kj + alpha * alpha * ki + 2.0 * alpha * k5 end if
m2 mj + alpha * alpha * mi c
k3 ki + gamma * gamma * kj + 2.0 * gamma * k5 c
m3 mi + gamma * gamma * mj return
k5 0.0 end