You are on page 1of 9

COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING lo (1977) 141-149

0 NORTH-HOLLAND PUBLISHING COMPANY

ON THE COMPUTATION OF PLASTIC STRESS-STRAIN RELATIONS


FOR POLYCRYSTALLINE METALS *
J. GULDENPFENNIG
Institut fiir Mechanik, Ruhr-Universittit, Bochum,

Germany

and
R. J. CLIFTON
Brown University, Providence,

R.I., U.S.A.

Received 15 June 1976


A computational procedure is presented for computing the macroscopic stress-strain behavior of polycrystalline
metals from consideration of slip on the slip systems of a large number of randomly oriented single crystals. This
procedure, based on the simplex method of linear programming, is illustrated by applying it to the case of simple
tension with no strain-hardening on individual slip systems. The method is shown to be efficient and accurate. Furthermore, the method can be extended to combined stress states and a variety of single-crystal hardening models.

1. Introduction
Many models have been proposed to predict qualitative and quantitative features of elasticplastic behavior of polycrystalline metals. One which is particularly attractive is the so-called
self-consistent model developed by Kroner [ 1 ] and Budiansky and Wu [ 21. This model incorporates slip on the slip systems of single crystals, viewed as spherical inclusions in an elastic-plastic
matrix which has the stress-strain characteristics of the bulk material. It has the advantage of
relating macroscopic plastic deformation to the underlying slip mechanisms in a consistent,
straightforward way. Unfortunately, it has the disadvantage that extensive computations are required for each load increment, even for cases of nominally homogeneous stress. The large computing effort arises from the large number of grains and slip systems which must be considered in
order to obtain satisfactory average behavior.
The objective of the present work has been to develop an efficient computational procedure
which would make it possible to employ self-consistent models in computations involving more
complex states of stress than have been considered so far. Herein a computational method based on
techniques used in linear programming is applied to the case of face-centered-cubic (FCC) metals.
Numerical results are given for the case of simple tension. The computed stress-strain curve is obtained in a remarkably short computing time and agrees fully with the results of previous computations [ 2,3].

* This paper is based on a thesis submitted by J. GtiIdenpfennig in partial fulfillment of the requirements for the Degree of Master
of Science in the Division of Engineering at Brown University, June, 1975.

J. Giildenpfennigand R. J. Clifton, On the computation of plastic stress-strainrelations

142

2. Governing equations
In self-consistent slip models of plastic deformation of polycrystalline metals the states of stress
and strain are characterized by nonhomogeneous stress and strain fields uii, eij which are constant
in any given grain, and by average stress and strain fields u$, E, which are the remote uniform
fields considered in the analysis of the interaction of a single grain with the surrounding matrix.
The total strains eij, Eij are assumed to be the sum of elastic and plastic strains where the respecp = 0, E:,= 0). The microscopic states uij, EC are
tive plastic strains EC, EG are isochoric (i.e. E,,
related to the macroscopic states c$, E$ by relations derived for the case of a homogeneous spherical inclusion embedded in an infinite homogeneous matrix having the elastic-plastic behavior of the
aggregate polycrystalline material and having remote uniform stress and plastic strain fields ut, E;.
For the case in which the individual grains are regarded as elastically isotropic the resulting relations are given by [ 2,3]
0..

11

bo = _
ij

2 (7 - 5~)G (pp.
15(1-V)

,+)

IJ

where V, G are Poissons ratio and the elastic shear modulus (superposed dots indicate time derivatives).
Because plastic strains are incompressible, it follows from (1) that the mean normal stress
uKK/3 is uniform. Then, eq. (1) can be rewritten as

where srj, S: are the stress deviators defined by Sij =


average of uij, i.e.

op.= 1
J

vv s

Utj - 5 uKK 6ij

uij dV,

etc. Also, since ug is the volume

G-d

it follows from (1) that for consistency EfJ. must be defined by

Gb)
For definiteness, in the further development of the governing equations we consider the special
case of FCC crystals. Such crystals (see fig. 1) have four slip planes (normal vectors m,) and six
possible slip directions (k ny)) on each plane for a total of 24 slip systems. The resolved shear
stress on the nth slip system is
r(n) =

uij

l$

= sij a[; )

where
,j;) = f (mp)+)

+ mf)n!)) .

(3)

J. Giildenpfennigand R. J. Clifton, On the computation of plastic stress-strainrelations

143

Fig. 1. Slip systems in FCC crystals.

The plastic strain rate 6; resulting from shear strain rates 9tn) on the various slip systems is

(4)

The remaining condition to be imposed is the constitutive equation which governs the elasticplastic behavior of the single crystals. In the present analysis we assume that the crystals are
ideally plastic. Then the constitutive conditions are
+I = 0
9

for r@) < rY

> 0

or for r@) = rY ,

for r@) = rY ,

i()

< 0 )

fC) = () )

where r,, is the yield stress.

3. Formualtion

as a nonlinear programming problem

Multiplication of eq. (1) by ~$1 and summation over indices i and j give, with the use of eqs.
(3) and (4), an equation of the following form for each slip system in each crystal:

(6)
where
G= 2G(7-5v)
15(1-v)

and

&,=s;+L%;.

(7)

144

J. Giildenpfennig and R.J. Clifton, On the computation

of plastic stress-strain relations

The quantity Q, is a measure of the macroscopic loading imposed on an element consisting of


many grains. If the material responds elastically, then Qij is determined by the macrosco@c stressrate St. On the other hand, if widespread plastic flow occurs, then t is small relative to Cl?;, and
Qij is determined by the macroscopic plastic strain-rate kc. From the computational viewpoint it
is convenient to regard Q, = Q,(S) as the prescribed loading history and compute the associated
histories s@) and Et(s).
For a prescribed history Qij(s), eq. (6) is, at each value of s and for each grain, a system of 24
linear equations in 48 unknowns (y@), i()). The additional conditions to be satisfied are the constitutive relations, eqs. (5). The resulting system of equations, inequalities, and bounds on the
variables r@), -j() suggests formulation of the equations as a linear or nonlinear programming
problem. To this end it is helpful to integrate eqs. (6) over an interval As and rewrite the equation
in terms of increments:

(6)

Then, since each slip system is counted twice, once for slip in each direction, the quantities AyCm),
#)(s + As) satisfy the inequalities
O< Aycrn) ,
0 <

+(s

@aI

+ As) < r,,

(8b)

Eqs. (6) and (8) have the form of constraint conditions for a standard linear programming problem (e.g. see [4]). To complete the formulation as a programming problem, it is necessary to prescribe an objective function which is minimized (or maximized) for values of r()(s + As), Aycrn)
which are consistent with the constitutive conditions (5). An appropriate objective function is the
function z defined by
z =

cn

(T,, -

.@)(s+ As)) A$)

(9)

The inequalities (8) ensure z > 0, and (for sufficiently small steps As) when z takes its minimal
value z = 0, the constitutive conditions (5) are satisfied in incremental form.
The objective function (9) is nonlinear, but it causes no difficulty, because of its simple form.
In fact, the objective function is so simple that its implications regarding which slip systems are
operative at each stage of the deformation are evident without performing calculations involving
the objective function.
The solution of (6) which satisfies (8) and minimizes (9) can be obtained by means of the
simplex method used in linear programming (see [4]). In this method one first finds an initial
basic feasible solution (i.e. a solution of (6) which satisfies the inequalities (8) and which has
fewer than 24 of the variables T()(s + As), A#) at limit values) and then proceeds systematically
from one basic feasible solution to another until the objective function is minimized. The variables which are set at limit values before solving eqs. (6) for the remaining 24 variables are called

J. Giildenpfennig and R. J. Clifton, On the computation of plastic stress-strainrelations

145

nonbasic variables; the variables which are free to assume other than limit values are called basic
variables; the set of basic variables at any stage of the solution process is called the basis of the
system (6) at that stage.
As long as the material remains elastic, the basic variables of the system (6) are T()(s + As) < T,,,
and minimization of (9) requires that all A+() must be zero. If the shear stress of the ith slip system @(s + As) reaches T,,, this slip system becomes active, and A#) replaces T@)(s+ As) = T,, in
the set of basic variables. Further loading causes other slip systems to become active and results
in a corresponding change in basis. As the deformation proceeds cases may arise in which unloading
of an active slip system occurs. If so, a change of basis also occurs, but this time a AT() leaves the
basis and the corresponding T(~)(s+ As) enters it. For each stage of loading, the stress-strain history
is linear as long as no new slip system becomes active, i.e. as long as the set of basic variables stays
constant. A limiting state of stress is reached when 5 independent slip systems become active in
each grain.
If the loading is such that stress reversal does not occur, then the number of slip systems to be
considered in each grain is reduced from 24 to 12. Furthermore, for the case of simple tension used
for the numerical example presented in the next section the only nonzero components of the macroscopic stress and plastic strain tensor are syl = -2st, = - 2& = 5 u and Ey, = - 2 E;, = - 2Efs, = EP,
where u is the imposed tensile stress and EP is the corresponding plastic strain. For simple tension
eq. (7) becomes

(loa)
(lob)
and the terms $)Qij

4. Computational

in (6) become c$)&, where 0 = i Q,, .

procedure

After introduction
fY =Go

c =y

rY

of dimensionless variables
x

AQ

--a11

(n)

n _

AY()
EY
>

7y -

d)(s + As)
>

y"=
TY

- T()(S)
jj"= rY
>
rY

(11)

rY

and
a mn = 2(7-5~)
15(1-V*)

a; a; )

m,n =

1,2 )... 12 )

the system (6) can be written in the matrix form


Ax+By=Bc,

(12)

146

J. Giildenpfennig and R. J. Clifton, On the computation of plastic stress-strainrelations

where X, y, c are 12-dimensional column vectors, and A, B are 12 X 12 matrices defined by


A = [-a]

= [A,,A,,

...A.,],

= [U,, U,, .. . U,,l,

B = [P]

in which A, and Un denote column vectors (U, is a unit vector with one in the nth row and zero
elsewhere).
In the early stages of loading when all c satisfy c > 0, the system (12) is in canonical form,
i.e. the basic variables yn have a unit coefficient in the nth equation and zero coefficients elsewhere.
The solution of (12) is simply y = c and x = 0. At each step AQ the right-hand side of (12) is
changed according to eq. (11). If after a step AQ one or more components of c become smaller
than zero, then the basis of ( 12) has to be changed. Transformation of (12) into the corresponding
canonical form is done by a pivot operation. This operation is carried out by starting with the
smallest component of Bc, say (Bc)~, and replacing the basic variables y* in the basis by x. To
make this change in basis, one considers the elementary matrix
D = [(II,

. .. A,

...

u,, ]

which has the inverse

II-=

[U 1 *.. Km... U,,],

where
Km =

-Ja

mm

{Ulrn ... -1 . .. aia,

}.

Multiplication of (12) by D- gives


D-Ax

+ D-'By

= D--l Bc ,

or

where B is identical with B except for column m, and column 1, becomes the unit vector U,. If
at least one component of & is smaller than zero, say (Bc), then the procedure is repeated using
the elementary matrix
D = (U, ... A,

... U,,)

provided that 2, is not equal to U,. If, however, A, = U,,, then the nth slip system is already
active and unloading must begin. In this case the pivot operation is carried out using the elementary matrix

J. Giildenpfennigand R.J. Clifton, On the computation of plastic stress-strainrelations

147

D = (V, .. . B, ... V,,) .


When all components of & are nonnegative, the system (12) is in canonical form for the given
increment AQ and pivoting is completed. The basic variables can then be evaluated by setting them
equal to the term in the corresponding row of &; the nonbasic variables have the value zero. The
plastic strain increment AeIi for the crystal being considered is evaluated from
(14)
The next step begins by considering again the same crystal subjected to a new loading increment
AQ. The vector c is changed according to eq. (11) with v taking the values of ym for the preceding step. Pivoting begins from the canonical form obtained at the end of the preceding step. This
process is repeated for the prescribed number of load increments, and then the whole procedure is
repeated for each crystal orientation.
In the numerical example presented herein, the loading consisted of eleven increments AQ for
each crystal orientation. The computations give the history of Q versus e;i for each of 9 1 different
crystal orientations. For symmetry reasons the crystal orientations are distributed over one of 48
similar triangles on the surface of a unit sphere. (see fig. 2). The c$) needed in the calculation of
the system (12) are obtained from

where Ziy) is a set of 12 matrices calculated for one crystal orientation, and Cij is the transformation matrix required to transform -cn)
CV,, to the orientation of interest. Written in terms of the two
Euler angles, the transformation matrix has the form

Fig. 2. Crystal orientations.

148

J. Giiidenpfennig and R. J. Clifton, On the computation of plastic stress-strain relations


-

Cii =

cos/3

cos 7j

sin /3
cos /3 sin n

-sin p

cos

sin n 1

- cos p

sin fl sin n

/.

cos q 1

The macroscopic tensile strain EP/cy is, from (2b), the average, over all orientations, of ~f,/e,.
This average is calculated by numerical integration of the integrals in the quotient

(19

over one of the 48 similar triangles. The integrals are evaluated numerically by a combination of
Simpsons rule and Newtons rule. Once EP is obtained from (1.5) for a given stage of loading Q(s),
the corresponding value of s:i can be obtained by integrating eq. (7).

5. Nume~cal results and concluding remarks


The computed stress and strain values are tabulated in table 1 and plotted in figs. 3 and 4 for
the case I/ I= l/3. Comparison with the results of Budiansky and Wu [2] shows that the difference
between the two solutions is not more than 0.3%. Such a discrepancy is explicable by the sensitiveness of two-dimensional numerical integration. A calculation using only 66 crystal directions
shows a difference of less than 1% from the calculation with 9 I directions.
The computing time for eleven steps AQ is only 24 set for 9 1 crystal orientations and only 17
set for 66 orientations (using an IBM 360/67 computer).
Overall, the computational procedure appears to work well. Whether or not it has advantages
over procedures used previously requires comparison of the various procedures for more difficult
problems involving combined stresses and nonproportional loading. As a next step the method
presented here has been applied recently [ 5 f to the determination of simple wave solutions for

Table 1. Stress-strainvalues(91 orientations, v = l/3)

1.1030
1.2105
1.2900
1.3775
1.4271
1.4577
1.4777
1.4899
1.5269
1.5331
1.5332

0.0783
0.3992
0.7667
1.3708
2.0382
2.7371
3.4538
4.1834
15.7885
164.1114
332.4446

J. Giildenpfennig and R. J. Clifton, On the computation of plastic stress-strain relations

IO

149

I5

EP/
EY
Fig. 3. Stress-strain curve.

Fig. 4. Stress-plastic strain curve.

combined longitudinal and torsional plastic waves. Solutions based on a self-consistent slip model
are shown to be in closer agreement with the experimental results [ 61 than are solutions [ 7,8]
based on assumed forms of smooth yield surfaces.

Acknowledgement

Scholarship support from the Deutscher Akademischer Austauschdienst is gratefully acknowledged as well as support from the U.S. Army Research Office through Contract No.
DAHC04-75-G-0174 with Brown University.

References
[ll E. Kroner, Fur plastischen Verformung des Viel-Kristalls, Acta Metal. 9 (1961) 155.

I21B. Budiansky

and T.T. Wu, Theoretical prediction of plastic strains of polycrystals, Proc. 4th Cong. Appl. Mech. (1962) 1175.
131 J.W. Hutchinson, Plastic stress-strain relations of FCC polycrystalline metals hardening according to Taylors rule, J. Mech.
Phys. Solids 12 (1964) 11.
141 G.B. Dantzig, Linear programming and extensions (Princeton Univ. Press, Princeton, 1963).
[51 J. Giildenpfennig and R.J. Clifton, An application of self-consistent slip models to plastic waves of combined stress
(unpublished manuscript).
[61 J. Lipkin and R. 3. Clifton, Plastic waves of combined stresses due to longitudinal impact of a pre-torqued tube, Part 1:
Experimental results, J. Appl. Mech. 37 (1970) 1107.
171 J. Lipkin and R. J. Clifton, Plastic waves of combined stresses due to longitudinal impact of a pretorqued tube, Part 2:
Comparison of theory with experiment, J. Appl. Mech. 37 (1970) 1113.
[gl R.P. Goel and L.E. Malvern, Biaxial plastic simple waves with combined kinematic and isotropic hardening, J. Appl. Mech. 37
(1970) 1100.

You might also like