You are on page 1of 22

INTERNATIONALJ O U R N A LFOR NUMERICAL METHODS I N ENGINEERING, VOL.

29, 3 15-336 (1 990)

STUDIES IN ANISOTROPIC PLASTICITY WITH


REFERENCE TO THE HILL CRITERION

RENE DE BORST AND PETER H.FEENSTRA


Deyt Unioersity of Technology, Department of Civil EngineeringlTNO Institute f o r Building Materials and Structures, P.O.
Box 5048, 2600 G A Derf, The Netherlands

SUMMARY
Algorithms based upon the notion of return mapping have been developed for the Hill yield function of
anisotropic plasticity. The relative accuracy of two algorithms is assessed by means of iso-error maps. The
choice of the algorithm turns out to be much more critical for the orthotropic Hill criterion than for the
underlying isotropic von Mises plasticity model. A tangent operator that is consistent with the developed
integration algorithm is formulated and its efficiency is assessed compared with the classical continuum
tangent operator. The model has been applied to three shell/platt structures.

INTRODUCTION
Finite element and finite difference calculations of elasto-plastic solids have been carried out for
some twenty years. Many successful applications have been reported and elasto-plastic finite
element calculations are frequently used in the design process of critical metal structures.
Typically, J,-flow theory is used to model the non-linear response of the solid. As a consequence,
a vast majority of the research effort in computational plasticity has been invested in the
development and evaluation of algorithms which can integrate the elasto-plastic rate equations
for such a model in an accurate and efficient manner. The literature on this topic has increased
rapidly, and especially in the last few years a thorough understanding has been gained of the
characteristics of the various algorithms that have been proposed.' -
Much less attention has been devoted to theories other than J,-flow theory. Discussions of
pressure-sensitive yield functions such as the Drucker-Prager and Mohr-Coulomb criteria and
of cap models are only beginning to emerge.6- l o The accurate treatment of corners in yield
surfaces (e.g. Mohr-Coulomb and Tresca) is a much overlooked issue,' ' - I s and contributions
involving plastic anisotropy are virtually non-existent. l 6 - ''
Moreover, the initial stress
approach is often adopted in those contributions that have appeared, which seems somewhat
obsolete in view of the recent developments in J,-flow theory.
In this paper, algorithms for integrating the stress-strain law of the orthotropic Hill criterion"
and tangential operators that are consistent with the algorithms are discussed. The research is
motivated by the consideration that the strength properties of some important traditional
engineering materials such as wood, masonry and thin metal sheets do show anisotropy. But also
models of modern materials such as fibre-reinforced plastics cannot bypass the issue of
anisotropy in the inelastic regime.
Although implementation of the orthotropic Hill criterion already presents considerable
computational difficulties compared to the von Mises model, of which it is an extension, it is not

0029-598 1/90/0203 1 5-22s 1 1 .OO Received November 1988


0 1990 by John Wiley & Sons, Ltd. Revised M a y 1989
316 R. DE BORST AND P. H. FEENSTRA

suggested that the Hill model accurately captures the mechanical behaviour of wood, masonry,
fibre-reinforced plastics, etc. It is merely a first step in modelling these classes of materials
properly. Yet, much insight can be gained from the errors involved when an attempt is made to
extend the algorithms commonly used in von Mises plasticity to include anisotropy. This
experience can be very helpful when designing algorithms for other, more complicated yield
functions of anisotropic solids (e.g. the Tsai-Wu criterion”).
The arrangement of this study is as follows. First, the governing equations of the Hill plasticity
criterion are recapitulated and the determination of the parameters from experiments is
discussed. Next, two algorithms for integrating the constitutive rate equations are formulated.
The accuracy of the algorithms is compared using iso-error maps. Then, the tangential operator
that is consistent with the most accurate algorithm is established and some example problems are
solved and compared with other numerical solutions that are known in the literature.

GOVERNING EQUATIONS FOR THE HILL CRITERION


A fundamental notion of plasticity theory is the existence of a yield function, sayf, which depends
on the stress tensor u and possibly on a number of internal or state variables. We suppose tHat
these state variables, which govern the evolution of the yield surface, are collected in a vector q, so
thatf=f(u, 9). A stress point is defined to be in a plastic state whenf= 0 and in an elastic state
whenf< 0. Further plastic loading occurs when, in addition to the requirement thatf= 0, we
also have! = 0. During plastic loading the total strain rate & is composed of an elastic part 8‘ and
a plastic part Bp, as follows:
& = &’ + &P (1)
The elastic strain rate is related to the stress rate u via the elastic moduli as contained in the
matrix D,
&e = D-1&
(2)
while the plastic strain rates are assumed to be obtainable by differentiation of the yield function
f with respect to the stress tensor u,

x
where denotes a plastic multiplier, the magnitude of which can be determined by invoking the
consistency condition i= 0. Equation (3) represents an associated flow rule. This flow rule is
adopted because of the paucity of experimental data on the direction of the plastic flow when
anisotropic materials like wood, fibre-reinforced plastics and masonry are considered.
When the location of the yield locus is entirely governed by the stress tensor, i.e. when the yield
function is independent of the internal variables q, the expression for the yield function reduces to
f=f(u). In the present section we will limit the treatment to such an elastic-ideally plastic
material because of the difficulties that arise when it is attempted to define hardening rules for
anisotropic materials. The next section will discuss a simple, but effective solution to deal with
hardening behaviour.
A relatively simple yield function that can capture orthotropy in the strength properties has
been proposed by Hill” as an extension of the von Mises criterion. When the axes of the
123-co-ordinate system coincide with the axes of orthotropy, the yield function is given by
f = F ( o z 2 - a33)2+ G(a3, - 011)2 + H(ol, - + 2Laf3 + 2
022)2 M ~ +
3 2~N a : , - 1 (4)
STUDIES IN ANISOTROPIC PLASTICITY 317

with F, G, H, L, M and N material constants which can be determined experimentally. When Ol1,
OZ2 and designate the yield stresses respectively in the 1-, 2- and 3-directions1 F, G and
H follow from
1 1 1
2F = ?
622
+?-?
a33 a11

1. 1 1
2G =?
033
+ -2
a11
-7
a22

1 1 1
2H=:+?-? (7)
4 1 022 633

In a similar fashion, L, M and N can be related to the ultimate shear stresses 012,8 2 3 and 631:

1
2L = 7
a23

1
2M=- (9)
5.31
1
2N =-
5;2

When a reference yield stress ii is introduced, the Hill yield function can be represented by the
expression
f = a12(a11- 022)2+ a 2 3 ( 6 2 2 - a d 2 + % l ( b 3 3 -all)’
+ 6a4,af2 + 6 ~ ~ ~ +~ 66(66a$1
0 2 2 ~ - 20’ (1 1)
in which the material parameters F, G,H, L, M and N have been replaced by the dimensionless
constants a12,a 2 3 , ~ 3 1 a44,
, ass and a66. Please note that the underlying isotropic von Mises
yield function is recovered when all coefficients aij are set equal to 1.
A further simplification of notation occurs, when the matrix P,
&(a12 + %I) - 4a12 - 3% 0 0
- +a12 +(a23 + a12) - Sa23 0 0 O0 l

0
0
0
0 0
O 05 5
2a

is introduced, since the expression for the yield function can then be written as
2a66
O 1
f = :dPU -i? (13)
As will be discussed in greater detail in a subsequent section, equation (1 3) is less suitable for use
in algorithms that integrate elasto-plastic rate equations. The equivalent expression

f = (3 UTPU) 112 - 0 (14)


will prove to be more efficient. Accordingly, this form of the Hill yield function will be used in the
sequel of this paper. Again, note that the expression of equation (14) can also be used to represent
318 R. DE BORST AND P. H. FEENSTRA

the isotropic von Mises criterion, since only a matrix P must be substituted in which all the
coefficients aij have been set equal to 1.

MODELLING OF HARDENING BEHAVIOUR


Little is known in the literature on hardening rules for anisotropic materials. Probably, an
accurate hardening rule for such a material will be very complex and will involve a large number
of parameters. A single hardening parameter, for instance, is unlikely to be sufficient, since it is
hard to envisage that an initially anisotropic yield contour will retain its shape during the
hardening process. Furthermore, the number of anisotropic materials is quite large and it does
not seem realistic to suppose that all these materials obey the same hardening formalism.
In view of the above considerations and the current paucity of experimental data, it is proposed
to use Besseling’s fraction model” when hardening behaviour has to be modelled. For the sake of
completeness the fraction model will now be summarized briefly.
In the fraction or sublayer model it is assumed that the total stress cr in a material point is the
sum of a number of, say n, fraction stresses q,
n
G = 1 diai
i= 1

with 4ithe weight of fraction i. The sum of the weights 4, equals 1. Each fraction has its own yield
stress Ei and by properly selecting the weights and yield stresses in the fractions the hardening
behaviour can be approximated in a multilinear sense. Evidently, including more fractions results
in a higher accuracy, but is also more expensive from a computational point of view. Inasmuch as
each fraction has its own yield stress Ei, the flow rule (3) must be applied separately to each
fraction. Consequently, each fraction can have a different plastic strain, although the total strain
is equal for all fractions (parallel chain). It follows that the elastic strains of the fractions are also
different and so are the stresses.
The fraction model is basically a three-dimensional approach. T h s has some special
consequences for constrained stress situations such as occur in shells and in membranes, since the
stress in the constrained direction is taken into account at fraction level. Although the sum of the
fraction stresses in the constrained direction must vanish, the individual contributions of the
fractions need not be zero. A discussion of the difficulties that arise when implicit integration
algorithms are used, as is done in the next section, will be given in a separate paper.

INTEGRATION FOR FINITE LOAD INCREMENTS


A basic problem of computational plasticity is the integration of the set of equations (1)-(3) for
a finite time step. Here, the notion of time is used in a rather abstract sense, since time does not
enter the equations for the rate independent solid defined by equations (1)-(3). The concept of
time is merely used to order the sequence of events. For a finite step, equation (1) can be replaced
by the finite counterpart
AE = AE‘ i- A E ~ (16)
For solids which do not exhibit a degradation of the elastic properties in the course of the loading
process, the finite counterpart of the bijective relationship between stress rate & and elastic strain
rate be can also be stated directly,
AE‘ = D- AG (17)
STUDIES IN, ANISOTROPIC PLASTICITY 319

The difficulties arise with the evaluation of AcP:

A single-point integration rule (or generalized mid-point rule) yields the following expression for
the plastic strain increment:

with 0 < a < 1. Combination of equations (16), (17) and (19) results in the following expression for
the stress increment:

An = DAE- AAD-8.f
an
where the subscript t + aAt for the gradient to the yield surface has been dropped for
convenience. This expression can be interpreted as follow^.^*^ Starting from the initial stress state
no a trial stress increment An, = DAc is computed. If the trial stress n, = no + An, subsequently
appears to lie outside the yield surface, i.e. if f(a,)> 0, the stress is returned to the yield surface by
the mapping - AADaflan. The stress at the end of the loading step on then becomes
af
nn = a, - AAD -
an
A graphical explanation of this procedure is offered in Figure 1.
The requirement that the final stress is located on the yield surface can mathematically be
expressed as

f(a,-AADGaf) = O

Developing this expression in a Taylor series gives

afT af + -AA2-D-D-
f(a,) - AL-D-
1 af* a 2 1 a j + ... = 0
an am 2 an aa2 a 6

Neglecting second and higher order contributions in AA, we obtain the expression

This expression can be calculated explicitly if the gradient to the yield function aflae is evaluated
for n = a,.Because of the contributions of the higher order terms equation (24) will generally not
result in a rigorous return to the yield surface. Ortiz and Simo6 therefore propose to use the stress
after the first correction, i.e.
320 R. DE BORST AND P.H.FEENSTRA

Figure 1. Graphical interpretation of return mapping algorithms

as a new trial stress from which a new stress a2 can be computed in a fashion similar to that in
which al is calculated from at in equation (25). Repetition of this procedure, which is illustrated in
Figure 2, ultimately yields a stress state that is within some tolerance of the yield locus. The return
stress path approximately follows the steepest descent direction and the name 'tangent cutting
plane' algorithm has been coined by Ortiz and
There exist a few, yet important cases where the higher order contributions in equation (23)
happen to vanish. Obvious cases are the Tresca and the Mohr-Coulomb surfaces, since these
yield loci are linear in the principal stress space. The von Mises yield function as expressed
through equation (14) is another example. This can be shown most simply as follows. For the case
that the solid is elastically as well as plastically isotropic, P obeys the relation
PDP = 2pP (26)
with p the elastic shear modulus. Since it follows from equation (14) that

and

the second order term in equation (23) appears to vanish, for

In a similar fashion higher order terms can be shown to vanish. It is emphasized that this result
bears upon the particular definition of the yield function through equation (14). If equation (13)
had been selected, this useful result would not have been obtained. In this light it is worth
mentioning that Schreyer et aL2 have used equation (13) in their accuracy analysis of integration
algorithms for von Mises plasticity. In some of their error contours they observed a drift from the
yield surface for the elastic predictor-radial return method. Since this method is identical with the
application of equation (25) to a von Mises plasticity model (see Appendix), this drifting tendency
must be attributed to the choice of equation (13) to represent the von Mises model. Selection of
equation (14) does not produce any drift, even for a linear hardening rule and without adding
iterations.
STUDIES IN ANISOTROPIC PLASTICITY 321

Figure 2. Iterative procedure to return to the yield surface by the tangent cutting plane algorithm (Ortiz and Simo6)

For anisotropic media the derivation as presented in equations (26)-(29) is no longer valid and
iterations must be added so as to ensure that the yield condition is satisfied at the end of the
loading step.
As an alternative strategy to the procedure summarized above, which is basically of an explicit
nature, we can use a fully implicit (Euler backward) a l g ~ r i t h m , ~ * " ~in"which the gradient to the
yield function is evaluated at the end of the loading step. For this purpose we must express the
stress at the end of the step u, in terms of the trial stress state Q,. Substitution of equations (27) and
(14) into equation (21) results in
3A1
Q, = U~- -DPQ,
2a
or after rearrangement

where
3A1
A=I+-DP
25
Note that in contrast to the case of i ~ o t r o p y ~the
~ . 'matrices
~ D and P do not have the same
eigenvectors when anisotropy is taken into account. Substitution in the yield condition f (a,)= O
subsequently results in a non-linear equation in A1,

(33)

The solution of equation (33) is easily accomplished, e.g., using a Newton-Raphson method. The
disadvantage of a Newton type method is that the derivative d.fldA1 which is needed in the
iteration

Al? - (g) -1
.f (A].) (34)

is rather expensive to compute since

3- -9
'DP)a,l
dA1-8?i[f(Al)+ 03 [Q:PD(A-')~PA-' +A-'PA- (35)

Alternatively, one may resort to a secant-type method.


It is interesting to observe that in the case of a von Mises solid the Euler backward method also
reduces to an elastic predictor-radial return method (see again the derivations in the Appendix).
322 R. DE BORST AND P. H. FEENSTRA

Consequently, both methods give exactly identical results for isotropic von Mises plasticity. For
the Hill model, however, the results will be shown to be quite different.

ACCURACY ANALYSIS O F INTEGRATION ALGORITHMS


It has been no,ted in the preceding that the Hill criterion reduces to the von Mises criterion when
the coefficients aij are set equal to 1. Both integration algorithms then reduce to the well-known
elastic predictor-radial return method. For this case an iso-error map, in which the error with
regard to the exact solution is plotted as a function of the radial and tangential trial stress
increments Auradand Aa,,,, must consequently be identical to the iso-error contour given by
Krieg and Krieg' for the elastic predictor-radial return method. A comparison of Figure 3 with
the results by Krieg and Krieg' shows that this is indeed the case.
Some comments are in order with respect to the definition of the radial and tangential trial
stress increments. The radial trial stress increment is directed orthogonal to the yield surface at
the initial stress point while the tangential trial stress increment points along the yield surface
and is located in the deviatoric plane. For the initial stress being located at the point where the
yield surface is curved most strongly the unit magnitude of the radial trial stress increment is
taken equal to the stress increment that is needed to reach the point where the yield surface
has the strongest curvature (Figure q a ) ) . The unit magnitude of the tangential trial stress
increment is then taken equal to the stress increment that is needed to reach the point where
the yield surface has the smallest curvature, so that only for the isotropic case are the unit
magnitudes for the radial and tangential trial stress increment equal. For instance, for a I 2= 2,
a23 = a31 = 1 and the other coefficients being zero, Aarad= [ - 0.3333, - 03333, O.6667lT6 +
+
and hao, = [ 0.4472, 0.4472, 0.0IT5. For the 'flat' side of the yield contour converse
considerations hold, see Figure qb).
As alluded to in the preceding section, differences occur between the tangent cutting plane
algorithm and a straightforward application of the Euler backward method when the curvature is
not constant. We will demonstrate this for a Hill yield function in which the coefficients of
anisotropy have been assigned to the following values:
a12 = 2, a23 = 1, ilJl = 1, a44 = 0, a s s= 0, ab6 = 0
Equations (5)-(7) can be used to show that these parameter values result in a mild anisotropy,
since 5' I = tizz = 08165 5,633= 6.Furthermore, the shear stresses are left out of consideration,

Figure 3. Iso-error map and direction of the trial stress increment for the isotropic version of the Hill yield function
STUDIES IN ANlSOTROPlC PLASTICITY 323

s3 I

f=O f=O

Figure 4. Definition of unit magnitude of trial stress increments

since graphical representation of the results is scarcely possible when they are included. The
problem is that, strictly speaking, we can not represent the Hill yield contour in the
three-dimensional principal stress space, but only in the six-dimensional space of all stress
components. Nevertheless, a good insight can be gained of the characteristics of the algorithm by
considering only the normal stress components.
Owing to the fact that the algorithm enforces compliance with the yield condition at the end of
the loading step, the stress does not drift away from the yield locus. Furthermore, the Hill
criterion is independent of the hydrostatic pressure, which eliminates possible sources of error in
the hydrostatic direction of the stress space. Consequently, a representation in the deviatoric
plane completely defines the integration error. Figures 5 and 6 show that the angle 8 as defined by

is used to measure the difference between the exact stress oE and the numerically determined
stress location an,and also define the direction in which 8 is taken positive. By the term 'exact'
stress, the stress is denoted which is computed by dividing the imposed strain increment in loo0
sub-steps. Comparison with the analytical solution' for the von Mises yield contour showed that
this number of sub-steps is sufficient to keep the angle of the 'exact' solution within 0.1" of the
analytical solution. It would have been more elegant if the numerical solution could have been
compared with a proper analytical solution also for the case of anisotropic behaviour.
Unfortunately, extension of the Key solution' for von Mises plasticity to the Hill yield function
seems to be not possible because of the fact that the radius of the yield contour is no longer
constant in the deviatoric plane. It is further noted that the number of sub-steps has been taken
equal to loo0 irrespective of the magnitude and the direction of the trial stress increment. The
justification for this approach lies in the observation that for the isotropic case a comparison
between the analytical solution and the 'exact' solution could not give support to the idea that the
number of sub-steps needed 'for an accurate evaluation of oEmust be increased with an increasing
trial stress increment, as has been suggested by Schreyer et aL2
The results of an error analysis for the tangent cutting plane algorithm are presented in Figures
5 and 6. Owing to the fact that the curvature is not constant anymore, the location of the initial
stress becomes an additional factor that influences the error. The iso-error contour of Figure
5 represents the most critical case in the sense that the initial stress is positioned at the point of the
yield contour where the local curvature has a maximum value. The iso-error contour of Figure
324 R. DE BORST AND P. H. FEENSTRA

f=O
0

Figure 5. Iso-error map and direction of the trial stress increment for the anisotropic version of the Hill yield function
,,
with a,, = 081650, a, = a,, = 5. The initial stress point is located at the strongest curvature of the yield contour and
~

the results are for the tangent plane cutting algorithm6

......

f=O

Figure 6. Isoerror map and direction of the trial stress increment for the anisotropic version of the Hill yield function
with a,, = 0~81650,,, a,, = iiZ2= 0. The initial stress point is located at the smallest curvature of the yield contour and
the results are for the tangent plane cutting algorithm'

6 represents the case that the initial stress is positioned at the point of the yield contour where the
local curvature is smallest.
Comparison between the error contours of Figures 5 and 6 for the mildly anisotropic case and
the error contours of Figure 3 for the isotropic case shows that the error rapidly increases for this
algorithm. An even more disturbing observation is that the numerically determined stress can be
located on the side of the initial stress uo other than the side to which the trial stress increment,
and hence also the strain increment, was directed. This is shown in Figure 7, which gives the angle
4 between the initial stress u0 and the numerically computed final stress G,,. If the final stress
increment has the same direction in the deviatoric plane as the strain increment, 4 will be positive.
We observe that, in particular when the radial trial stress increment exceeds a value of twice the
unit magnitude, 4 is never positive. A graphical explanation for this phenomenon, which is
physically impossible under the assumption of associated flow and in the absence of
strain-softening, is offered in Figure 8. It appears that the representation of Figure 2 for the
tangent cutting plane algorithm does not meet the real return path very closely. As indicated by
STUDIES IN ANISOTROPIC PLASTICITY 325

f=O

0 1 2 3 4 5
t

Figure 7. Angle 4 between eoand eqas a function of the trial stress increment for the anisotropic version of the Hill yield
function with E33 = @8165a,,, 5 , I = E,, = E. The initial stress point is located at the strongest curvature of the yield
contour and the results are for the tangent plane cutting algorithm6

Figure 8. Realistic representation of the return path when the tangent cutting plane algorithm is used for the anisotropic
Hill criterion

Figure 8 the first iteration is usually very dominant. Subsequent iterations can apparently not
correct the directional error that is committed in the first iteration.
Most yield criteria that are known in the literature have a non-constant curvature in the stress
space similar to that of the Hill criterion. It is therefore believed that the observation for the Hill
yield contour that the tangent cutting plane algorithm can result in a final stress increment which
points in a n entirely different direction than the applied trial stress increment pertains to many
other yield criteria as well. The fact that this anomaly has not been noticed before must probably
be attributed to the manner in which the error has been represented in previous error analyses,
namely by plotting only the magnitude of the difference between the exact and the numerically
computed stress.
Considering the Euler backward algorithm for the same degree of anisotropy we observe not
only that the error 8 remains much smaller (Figure 9), but also that the final stress increment
always has the same direction as the trial stress increment. This is logical, since it is inherent in the
fact that the direction of the plastic flow is fully determined by the stress at the end of the loading
step.
326 R. DE BORST AND P. H. FEENSTRA

Q-

m-

(v-

--
a-’
O f 2 3 4 5
(a) t (b) t

Figure 9. Iso-error map and direction of the trial stress increment for the anisotropic version of the Hill yield function
with a, = 08165a,,, a,, = CZ2 = 6. The results are for the Euler backward algorithm: (a) the initial stress point is
located at the strongest curvature of the yield contour; and (b)the initial stress point is located at the smallest curvature of
the yield contour

0 1 2 3 + 5
(a) t (b) t
Figure 10. Iso-error maps for the Euler backward algorithm with u,, = 26,8,, = (rZ2= 5:(a) initial stress at point of the
yield contour with the strongest curvature; and (b) initial stress at point of the yield contour with the smallest curvature

To investigate whether the good behaviour of the Euler backward algorithm remains for
a more severe degree of anisotropy, two additional analysis were carried out. In the first analysis
the normal yield stresses in the material directions were taken as 61 = 4 E, az2= f E and i?33 = 8.
This results in the following set of material parameters:
a I 2 = 7, a23 = 1, a31= 1
The results as given in Figure 10 show that the maximum angular error increases slightly for the
initial stress being located at the strongly curved part of the yield surface, while it diminishes
somewhat for the ‘flat’ side of the yield contour. In line with a comparison between the results for
the isotropic yield function (Figure 3) and the mildly anisotropic yield function for the initial
stress being located at the strongly curved part of the yield surface [Figure 9(a)] we observe that
the region where the large angular errors occur becomes smaller and is more and more confined
to the case that the radial component of the trial stress is small and the tangential component is
large.
The second analysis considers a very high degree of anisotropy, since the normal yield stresses
in the material directions were taken as a l l = $5, CZz = 45 and CJ3 = 0. This results in the
STUDIES IN ANISOTROPIC PLASTICITY 327

following set of material parameters:


a12= 49, a23 = 1, a31 =1
The results of the error analysis are given in Figure 11 and confirm the tendency that was already
apparent from the preceding analyses. It seems remarkable that the error almost vanishes for
a large part of the domain, although the initial stress is positioned at the now very strongly curved
part of the yield surface. In fact, the error at this point tends to zero when the curvature increases
and more and more begins to resemble a genuine corner. Since a proper, fully implicit treatment
of corners as advocated by de B ~ r s t ' ~ also * ' ~ results in an exact integration, the observed
characteristics are explainable and logical. Since corners and sharply curved parts of yield
surfaces serve as attractors for the stress, the observation that only a small error is made at these
points is of particular importance.

CONSISTENT TANGENT OPERATOR


Within a loading step, equation (20) that sets the dependence of the stress increment on the
prescribed strain increment is basically a total stress-strain relation. Accordingly, we have
a deformation type plasticity theory within a loading step rather than a flow theory when a return
mapping algorithm is used. In this spirit the tangential relation between stress rate and strain rate
that is required when implicit solution strategies are used at a global (structural) level bears much
resemblance to the tangential operators that result from a deformation theory of plasticity. In
particular, equation (20) can be differentiated to yield

For the Hill model a2flaa2is given by the expression of equation (28). Introducing the auxiliary
matrix H,

(38)

or equivalently

(39)

Figure 11. Iso-error maps for the Euler backward algorithm with C3, = 5 0 . 6 , , = '522 = u: (a) initial stress at point of the
yield contour with the strongest curvature; and (b) initial stress at point of the yield contour with the smallest curvature
328 R. DE BORST AND P. H. FEENSTRA

equation (37) can be rewritten to give a familiar form of the rate equation:

For a perfectly plastic solid, f = f (a),and invoking the consistency condition f = 0 results in

Premultiplication of equation (40) with aflau and using equation (41) gives an explicit expression
for X ,
af
-H&

so that upon substitution of this result in equation (40) we obtain


af af
H--H
u=[H- -H- (43)
aa au
Thus, for perfect plasticity the so-called consistent tangent operator’ 1 * 1 4 8 2 2- ”as derived in
equations (37)-(43) can be obtained from the classical tangent operator simply by replacing the
elasticity matrix D by the matrix H as defined by equation (38).
For the fraction model, which will be used in one of the examples, the consistent tangent
operator as defined in equations (38) and (43) applies to each fraction separately. At integration
point level the tangential stress-strain relation is now given by

where Hi applies to fraction i and ui denotes the fraction stress.

EXAMPLES

A cylindrical shell
The first example problem is the cylindrical shell of Figure 12. The shell is simply supported at
both ends, while the displacements of the straight sides are free. Because of symmetry
considerations the calculations have been carried out for a quarter of the shell. Eight-noded
degenerated shell elements with four Gauss points in the plane and a five-point Simpson
integration through the thickness have been used. In all the calculations the Young’s moduius has
been taken as E = 21 x lo6 kN/mZ, the Poisson ratio has been set equal to zero while the
reference yield stress has been assigned a value 5 = 4200 kN/m2.
The shell has been selected for numerical simulation, since the uniformly distributed
(self-weight) load induces locally strongly varying membrane stresses as well as large bending
STUDIES IN ANISOTROPIC PLASTICITY 329

Figure 12. Geometry for cylindrical shell

moments both in the longitudinal and in the transversal direction. As with the other two
examples the shell has been analysed previously by Owen and Fig~eiras,'~. ' which makes
a comparison with their results possible.
Three different analyses have been undertaken. In the first analysis the yield function has been
assumed to be isotropic, so that all the coefficients orij are set equal to one. The result of Figure 13
shows the vertical displacement of the midside node along the straight side plotted against the
load intensity. The next analyses include anisotropy in the yield contour. First, the yield stress in
the 1-direction has been assigned a value of twice the reference yield stress: ifll = 25. Then, the
yield stress in the 2-direction has been assigned a value of twice the reference yield stress:
if.22 = 25. Figure 13 shows that the impact on the load-deflection curve is considerable. The
ultimate bearing capacity increases by about 30 per cent.
It is remarkable that, although the loadcleflection curves differ considerably, the limit load is
virtually identical for both analyses in which anisotropy is taken in account. This observation is
not in line with the suggestion that is made by the analyses by Owen and Fig~eiras.'~.''This is
because the calculations by Owen and Figueiras16*17have been terminated at a stage where the
loadcleflection curve still has a significant positive slope (see Figure 13). It is not clear why the
calculations of Owen and Figueiras have been terminated at that stage, but the present
observation underlines once more the importance of tracing the entire load-displacement curve
in order that a proper assessment of the limit load of a structure can be made.
All the calculations have been carried out under arc-length control with a linearized constraint
equation which is updated at every iteration. Furthermore, a full Newton-Raphson iterative
procedure has been used. In order to assess the impact of the use of the consistent tangent
operator on the convergence characteristics calculations have been carried out for the classical as
well as the consistent tangent operator. The results are summarized in Tables I, I1 and 111. In each
table, the calculated load intensity at the end of the step and the number of iterations needed to
achieve convergence (energy norm are given for the last ten load steps. The results
demonstrate that, while the improvements in convergence behaviour are already significant for
the case of isotropy, the difference becomes rather dramatic when anisotropy is taken into
account.

A clamped plate
The second example concerns a clamped plate which is loaded in the centre by a concentrated
load. Again, only a quarter of the plate has been modelled because of symmetry considerations
(Figure 14). Eight-noded elements with four-point integration in the plane and seven-point
330 R. DE BORST AND P. H. FEENSTRA

load q [ kN/m2 ]
4.0 ~

* limit load by Owen and Figueiras

displacement point A [ m 1

Figure 13. Load4isplacernent curves of point ‘A‘ of the cylindrical shell for isotropicand anisotropic strength properties

Table I. Convergence characteristics: cylindrical shell; isotropic


Step 1 2 3 4 5 6 7 8 9 10
Classical 2 5 6 8 9 5 6 6 7 8
Load q 1.498 1.729 1.954 2-163 2.379 2.468 2.561 2.647 2733 2.813
Consistent 2 3 4 4 5 4 4 4 3 3
Load q 1.498 1.729 1.954 2.164 2.380 2.470 2.562 2.649 2735 2.816

Table 11. Convergence characteristics: cylindrical shell; anisotropic 6,,= 26

Step 11 12 13 14 15 16 17 18 19 20
Classical 5 5 6 5 6 7 11 11 > 15 >15
Load q 3.016 3.107 3.199 3.293 3.380 3.459 3.545 3.628 3.711 3,779
Consistent 4 3 4 3 3 4 4 3 4 4
Load q 3.017 3.108 3.201 3.296 3.383 3.461 3.548 3.633 3.719 3.789

Table 111. Convergence characteristics: cylindrical shell; anisotropic a,, = 26


~~~~~~ ~ ~

Step 11 12 13 14 15 16 17 18 19 20
Classical 6 7 8 9 9 9 10 12 14 > 15
Load q 2.974 3.066 3.159 3.251 3.346 3.439 3.520 3.599 3.670 3.732
Consistent 4 4 3 4 4 3 3 4 4 4
Load q 2.977 3.070 3.163 3.256 3.352 3.357 3.450 3.535 3.614 3.685

integration throughout the depth have been used in all the calculations as well as in the
calculations of the spherical shell to be discussed next. Elastic isotropy has been assumed with
a Young’s modulus E = 30 x lo6 kN/m2 and a Poisson’s ratio v = 0.3.
Comparison with the results of Owen and Fig~eiras’~.’’is now more difficult, since they use
a hardening rule also for the case in which we have anisotropic strength properties. In a preceding
section our arguments have been laid out against the ad hoc development and implementation of
STUDIES IN ANISOTROPIC PLASTICITY 331

L = 6.0 [ml
d = 0.2 [m)

Figure 14. Geometry for clamped plate

5.0 -
, loadP[ W 1

4.0 -

3.0-

2.0 -

limit load by Owen and Figueiras

0.0
0.0
displacement centre [ m 1

Figure 15. Loaddisplacement curves of the centre of theclamped plate for isotropic and anisotropic strength properties

hardening rules for anisotropic plasticity. Instead, it has been proposed to use the fraction model.
Although hardening behaviour can be modelled to a high degree of accuracy using the fraction
model, the results usually deviate slightly from (isotropic) hardening rules, even for monotonic
loading. This is a possible cause of discrepancy between the present analysis and the results by
Owen and Figueiras.'6,17 In particular, the hardening behaviour with a hardening modulus of
h = 0.01~5has been approximated by a two-fraction model with 41 = 0.99 and 42 = 0.01. The
reference (isotropic) yield stress has been chosen as 5 = 30,000 kN/m2 for the first fraction. The
second fraction remains elastic throughout the loading process.
The material parameters are taken as 511= 30,000 kN/m2, tYZ2 = 40,000 kN/m2,
aI2= 20,200 kN/m2 and 545= 35,000 kN/m2, where 545is the yield stress which is measured in
a tensile test on a specimen with material axes under 45" with the direction of the uniaxial stress.
Using these data the coefficients ai,can be computed:
a 1 2 = 0.8294, t123 = 0.2956, a13 = 1.1706
a44 = 0.7352, a55 = l'oooo, 466 = 1'oooo
The load has been plotted as a function of the displacement of the centre of the plate in Figure
15. Again, a clear difference is observed between the results for the isotropic and the anisotropic
strength properties. The point where the calculations of Owen and F i g ~ e i r a s ' ~ ~have
' ' been
332 R. DE BORST AND P. H. FEENSTRA

terminated has also been plotted in Figure 15. The agreement with the present results is rather
close.
A comparison between the convergence characteristics of the consistent and the classical
tangent operator is given in Table IV for the anisotropic case. Again, the load magnitude at the
end of each of the nine load steps and the number of iterations needed to obtain a converged
solution (energy norm are presented.

A spherical shell
The last example is the spherical shell of Figure 16. Since the modelling as well as the material
properties are identical to those of the previous example, they are not discussed here. Figure 17
Table IV. Convergence characteristics: clamped plate; anisotropic
~ ~~

Step 4 5 6 7 8 9 10 11 12
Classical 4 6 6 8 10 15 > 15 > 15 > 15
Load P 1.714 1.956 2.200 2-419 2.662 2885 3.120 3.354 3.597
Consistent 4 4 2 4 4 4 4 5 6
Load P 1.714 1.958 2.204 2424 2.667 2.892 3.129 3.364 3.609

Figure 16. Geometry for spherical shell

4.0 i loadP'kN1

* lirnif load by Owen and Figtieifas

I.Ol/
olo 0.b2 0.64 0.k 0.08 o.io 0.i2 o h
displacement centre [ m ]

Figure 17. Loaddisplacement curves of the centre of the spherical shell for isotropic and anisotropic strength properties
STUDIES IN ANISOTROPIC PLASTICITY 333

Table V. Convergence characteristics: spherical shell; anisotropic


~~~~~~~~~~~~~

Step 4 5 6 7 8 9 10 11 12
Classical 6 5 7 10 13 15 > 15 > 15 > 15
Load P 2.851 3.099 3.325 3.564 3.818 4.039 4.303 4.549 4.834
Consistent 4 3 5 5 4 4 4 5 5
Load P 2.855 3.107 3.333 3.573 3.831 4.057 4.324 4589 4.943

shows the magnitude of the concentrated load in the centre of the shell as a function of the vertical
displacement in that point. Table V presents a comparison of the performance of the consistent
and the classical tangent operators for the anisotropic strength properties.

CONCLUSIONS
In this paper integration algorithms and consistent tangent operators have been derived and
tested for the anisotropic Hill plasticity model.
Two algorithms have been tested. The first, explicit, algorithm uses successive updates to reach
the yield surface (tangent cutting plane algorithm6). The second algorithm expresses the stress in
the final state as a function of the trial stress state, so that the problem is brought back to the
solution of a non-linear equation of a single scalar, namely the plastic multiplier AA, It is basically
a fully implicit (Euler backward) algorithm. For isotropic strength properties the Hill criterion
reduces to the von Mises function. Owing to the constant curvature of this yield contour in the
principal stress space both algorithms then give exactly the same result. In fact, they both reduce
to the well-known elastic predictor-radial corrector scheme. For anisotropic behaviour the
curvature of the yield surface can vary strongly along the yield contour. Iso-error maps reveal
that the Euler backward algorithm then performs in a manner superior to that of the tangent
cutting plane algorithm. It appears that the tangent cutting plane algorithm can produce final
stress increments that make an obtuse angle with the trial stress increment. This is physically
impossible under the assumption of associated flow and in the absence of softening. Since many
yield criteria that are known in the literature have a non-constant curvature in the principal stress
space, it is believed that this observation pertains to many other yield criteria.
A pleasant property of the Euler backward algorithm is that the error that is committed tends
to zero when the local curvature of the yield surface becomes stronger. This is particularly
important since such parts of the yield contour serve as stress attractors, i.e. the stress tends to
move to such a point in the course of the loading process. The observation that the error vanishes
when the local curvature tends to infinity is explainable for an Euler backward algorithm, since
a treatment of corners in the yield surface using such an algorithm results in an exact
solution.”. l 3
A consistent tangent operator has been derived and implemented for the Hill plasticity model
and its performance has been assessed. The improvement in convergence appeared to be even
more significant for anisotropic strength properties than when isotropy is considered.
In view of the many uncertainties and the complexities that are involved when hardening rules
have to be constructed for anisotropic materials, the use of the fraction model” seems attractive
in such circumstances. Its performance in two examples, one being a clamped plate, the other
being a spherical shell, turned out to be satisfactory.
334 R. DE BORST AND P.H. FEENSTRA

ACKNOWLEDGEMENTS

The contributions of the second author have been carried out as a part of his graduation project
at Delft University of Technology under the supervision of Prof. J. Blaauwendraad. The
calculations have been carried out with the D I A N A finite element code of the TNO Institute for
Building Materials and Structures.

APPENDIX
When the hydrostatic stress vector p is defined as
p = ftr(a)[1 1 1 0 0 01'
and the deviatoric stresses are given by
s=a-p
equation (14) can be used to derive that, for an isotropic solid,

where the subscript t refers to the trial stress state, and that

Using equations (A3), (A4) and (14), equation (25) can be shown to reduce to

Equation (AS) is exactly the formula for an elastic predictor-radial corrector scheme. This
completes the proof that the tangent cutting plane algorithm reduces to the elastic
predictor-radial return method when equation (14) is used as definition for the von Mises yield
criterion. Moreover, the stress point is returned to the yield surface in a single iteration. This
follows upon substitution of equation (A5) into equation (14). It is noted that the present
derivation pertains to an elastic-perfectly plastic solid, but that a similar derivation can be made
for hardening behaviour.'
I t has been asserted that the Euler backward algorithm also reduces to the elastic
predictor-radial return method. This is irrespective of the choice of equation (1 3) or equation (14)
as the definition for the von Mises yield criterion, and demonstrated using the definition of
equation (14). First, equation (30) is rewritten as

Since the solid is defined to be isotropic, equation (A6) reduces to


STUDIES IN ANISOTROPIC PLASTICITY 335

By virtue of the fact that the hydrostatic component p and the deviatoric s are linearly
independent, this results in the following set of uncoupled equations:
P, = P” (A8)

Next, the result of equation (A9) is substituted in equation (14). This gives

since Ail is always non-negative. Solving for A l gives


(+a:Pa,)”* - a
Al =
3P
Substitution of equation (A1 1) in equation (A9) and using equation (A2) gives for the final stress

We observe that the elastic predictor-radial corrector method is again obtained. Yet, there is
a major difference from the previous method, since the return to the yield surface is not obtained
in a single iteration. This is becausefas defined in equation (33) is a non-linear function in A l . In
the case of isotropic plasticity, convergence in a single iteration can still be achieved if equation
(A1 1) is used as a predictor for Ail. For anisotropic strength properties this choice accelerates the
convergence.

REFERENCES
1. R. D. Krieg and D. B. Krieg, ‘Accuracies of numerical solution methods for the elastic-perfectly plastic model’,
J . Pressure Vessel Technol. A S M E , 99, 510-515 (1977).
2. H. L. Schreyer, R. F. Kulak and J. M. Kramer, ‘Accurate numerical solution for elasto-plastic models’, J. Pressure
Vessel Technol. ASME, 101, 226-234 (1979).
3. P. J. Yoder and R. G. Whirley, ‘On the numerical implementation of elastoplastic models’, J. Appl. Mech. A S M E , 51,
283-288 (1984).
4. T. J. R. Hughes, ‘Numerical implementation of constitutive models, rate-independent deviatoric plasticity’. in S.
Nemat-Nasser et al. (eds.), Theoretical Foundations f o r Large-Scale Computations for Nonlinear Materiaf Behavior,
Martinus Nijhoff. Dordrecht, 1984. pp. 29-57.
5. M. Ortiz and E. P. Popov, ‘Accuracy and stability of integration algorithms for elastoplastic constitutive relations’,
Int. j. numer. methods eng., 21, 1561-1576 (1985).
6. M. Ortiz and J. C. Simo, ‘An analysis of a new class of integration algorithms for elastoplastic constitutive relations’,
Int. j. numer. methods eng., 23, 353-366 (1 986).
7. P. A. Vermeer, ‘A modified initial strain method for plasticity problems’, in W. Wittke (ed.), Proc. Third Int. ConJ
on Numerical Methods in Geomechanics, Balkema, Rotterdam, 1979. pp. 377-387.
8. B. Loret and J. H. PrCvost, ‘Accurate numerical solutions for Drucker-Prager elastic-plastic models’, Comp. Methods
Appl. Mech. Eng., 54, 259-277 (1986).
9. K. Runesson, ‘Implicit integration of elastoplastic relations with reference to soils’, I n [ . j. numer. anal. methods
geomech., 11, 315-321 (1987).
10. J. C. Simo, J.-W. Ju, R. L. Taylor and K. S. Pister. ‘Assessment of the cap model: consistent return algorithms and
rate-dependent extensions’, J. Eng. Mech. dic. ASCE, 114, 191-218 (1988).
1 I. J. C. Simo, J. G. Kennedy and S. Govindjee, “on-smooth multisurface plasticity and viscoplasticity. Loading/
unloading conditions and numerical algorithms’, I n ! . j. numer. methods eng., 26, 2161-2186 (1988).
12. R. de Borst, ‘Nonlinear analysis of frictional materials’, Dissertation, Delft University of Technology, Delft, 1986.
13. R. de Borst, ‘Integration of plasticity equations for singular yield functions’, Comp. Struct., 26, 823-829 (1987).
336 R. DE BORST A N D P. H. FEENSTRA

14. M. A. Crisfield, 'Consistent schemes for plasticity calculation with the Newton-Raphson method', in D. R. J. Owen et
al. (eds.), Computational Plasticity, Vol. I, Pineridge Press, Swansea, 1987, pp. 133-160.
15. S. W. Sloan and J. R. Booker, 'Removal of singularities in Tresca and M o h r x o u l o m b yield function', Commun. appl.
numer. methods. 2, 173-179 (1986).
16. D. R. J. Owen and J. A. Figueiras, 'Elasto-plastic analysis of anisotropic plates and shells by the semi-loof element',
lnr. j . nimer. methods eng., 19, 521-539 (1983).
17. D. R. J. Owen and J. A. Figueiras, 'Anisotropic elasto-plastic finite element analysis of thick and thin plates and shells',
Int. j . numer. methods eng., 19, 541-566 (1983).
18. K. Runesson. S. Sture and K. Willam, 'Integration in computational plasticity', Comp. Srruct., 30, 119-130 (1988).
19. R. Hill. 'A theory of the yielding and plastic flow of anisotropic materials', Proc. Roy. Soc. A193, 281-297 (1947).
20. S. W. Tsai and E. M. Wu, 'A general theory of strength of anisotropic materials', J . Compos. Mater.. 5, 58-80 (1971).
21. J . F. Besseling, 'A theory of elastic, plastic and creep deformations of an initially isotropic material showing
anisotropic strain-hardening, creep recovery and secondary creep', J. Appl. Mech. A S M E , 25, 529-536 (1958).
22. E. Ramm and A. Matzenmiller, 'Computational aspects of elasto-plasticity in shell analysis', in D. R. J. Owen et al.
(eds.). Computational Plasticity, Pineridge Press. Swansea, 1987. pp. 7 I 1 -735.
23. J. C. Simo and R. L. Taylor, 'A return mapping algorithm for plane stress elasto-plasticity', Inr. j. numer. methods eng.,
22, 649-670 (1986).
24. J. C. Simo and R. L. Taylor, 'Consistent tangent operators for rate-independent elasto-plasticity', Comp. Methods
Appl. Mcch. Eng., 48, 101-118 (1985).
25. K. Runesson, A. Samuelsson and L. Bernsprang, 'Numerical technique in plasticity including solution advancement
control'. I n f . j. numer. methods eny.. 22, 769-788 (1986).

You might also like