On a Class of Models for the

W. D. I W A N Yielding Behavior of Continuous

and Composite Systems
Associate Professor,
D e p a r t m e n t o f A p p l i e d Mechanics,
C a l i f o r n i a Institute o f T e c h n o l o g y
Pasadena, Calif.

A class of one-dimensional models for the yielding behavior of materials and structures
is presented. This class of models leads to stress-strain relations which exhibit a
Bauschinger effect of the Massing type, and both the steady-state and nonsteady-state
cyclic behavior are completely specified if the initial monotonic loading behavior is known.
The concepts of the one-dimensional class of models are extended to three-dimensions
and lead to a subsequent generalisation of the customary concepts of the incremental
theory of plasticity.

Introduction teresis loops should have this form wasfirstsuggested by G. Mass-

ing [4] and is often referred to as Massing's hypothesis. The
As MORE and more attention is given to the static and recent development of the concept of kinematic hardening by
dynamic analysis of rate independent yielding systems, it is in- Prager [5] has provided a means for introducing a Bauschinger
creasingly desirable that this behavior be expressed mathemati- effect into the incremental theory of plasticity but, even with
cally in a manner which is as straightforward as possible and this extension, stress-strain behavior satisfying Massing's hy-
is at the same time in essential agreement with known experi- pothesis has only been obtained for the special case of a linear
mental data. Ideally, this mathematical formulation would be work-hardening law [6], When the work-hardening law is not
based on a consideration of the atomic or crystalline structure linear, it would appear that some additional assumptions need to
in the case of continuous systems1 and on a consideration of the be added to the kinematic hardening hypothesis but as yet these
behavior of individual constituent elements in the case of com- assumptions have not been set forth.
posite systems. However, such an approach can be very com- In one-dimensional analysis, resort is often made to formula-
plex and at present it has only been fruitful in certain relatively tions of a strictly empirical nature in an attempt to overcome the
simple, well defined composite systems. difficulties mentioned previously. It is assumed that the be-
Another possible starting point for the development of a havior of the system can be accurately described in terms of a
mathematical formulation for continuous systems would be a certain class of mathematical functions, and empirical data,
mathematical theory of plasticity and indeed, considerable effort usually from a monotonically increasing load test, are used to de-
has been expended in this direction (see references [l] 2 and [2] termine particular function parameters. This procedure gen-
for a summary of the major work in this field). One such theory, erally works well when restricted to describing the noncyclic be-
the incremental theory of plasticity, describes the three-dimen- havior of a system. The difficulty again arises when the method
sional yielding behavior of a material in terms of a yield surface is extended in order to describe the nature of the hysteretic be-
in stress space, along with a flow rule and a work-hardening law. havior for cyclic loading. Since it is impractical to test every
To date, this theory has been applied quite successfully to the conceivable case of cyclic loading, certain assumptions must be
rather large class of problems which have to do with the introduced which will enable extension of the empirical mono-
monotonic loading behavior of materials. However, when the tonic or stabilized initial loading data to these cases. For
theory is extended to the analysis of cyclic and hysteretic be- steady-state cyclic behavior or loading between fixed limits,
havior, certain difficulties arise. For the most part, the difficulties Massing's hypothesis will suffice. However, for cases of transient
result from the requirement that the theory be capable of account- loading or loading between variable limits, this hypothesis is of
ing for a Bauschinger effect. Furthermore, experimental results no help. In these latter cases, one is generally forced into using
indicate that this Bauschinger effect should have a specific form; some arbitrary mathematical rule which may or may not be
based on the physical nature of the system.
namely, that the stress-strain curves associated with one-dimen-
sional symmetrical closed hysteresis loops should be of the same The present paper considers a class of phj-sically motivated
form as those of stabilized initial loading curve (or cyclic stress- models for the rate independent hysteretic behavior of both con-
strain curve for cyclic hardening or softening systems) except tinuous and composite systems. It will be shown that this class of
for an enlargement by a factor of two [3], The idea that hys- models provides a bridge between empirical initial loading data
and the analysis of the steady-state and transient cyclic behavior
1 For the purposes of this paper, yielding systems will be divided
of hysteretic systems and, at the same time, provides a basis for
into two general types: Continuous systems used in the classical
comparing many of the current purely empirical formulations.
sense to denote solid bodies such as metals, and composite systems Since the models in their basic form are one-dimensional, they
or structures which consist of a collection of macroscopic components will lead to considerable simplification for problems of this type.
some or all of which may themselves be hysteretic or yielding. However, the concepts of the models can easily be incorporated
2 Numbers in brackets designate References at end of paper.
into the general three-dimensional incremental theory of plas-
Contributed by the Applied Mechanics Division for publication ticity, and it will be shown that this leads to a very straightfor-
(without presentation) in the JOURNAL OF APPLIED MECHANICS.
Discussion of this paper should be addressed to the Editorial De-
ward method of accounting for a realistic Bauschinger effect.
partment, ASME, United Engineering Center, 345 East 47th Street,
New York, N. Y. 10017, and will be accepted until October 15, 1967. Models
Discussion received after the closing date will be returned. Manu-
script received by ASME Applied Mechanics Division, August 26, The class of models which will be considered consists of a col-
1966; final draft, February 7, 1967. Paper No. 67—APM-S. lection of perfectly elastic and rigid-plastic or slip elements ar-

612 / SEPTEMBER 1 9 6 7 Transactions of the A S M E

Copyright © 1967 by ASME
Downloaded 03 Sep 2012 to 125.236.238.108. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 ranged in either a series-parallel or a parallel-series combina- consideration will onlj' be given to the case where the slip stresses
tion. It is assumed that the models contain a very large number cr* are distributed while the elastic moduli remain constant and
of elements and that the properties of the elements are distributed equal to some value E. Furthermore, let the number of elements
in some fashion. The distribution of these properties will then become very large so that a * can be described in terms of a dis-
dictate the particular form that the hysteretic behavior will take. tribution function <p(<r*), where <p(<T*)d<r* is the fraction of the
In continuous systems, the models might be thought of in terms total slip elements having a slip stress between a* and a* + da*.
of the distribution of slip-planes or dislocations throughout the Then equation (1) becomes
material, while, in the case of composite systems, the models can /»co /*Ee
best be thought of as depicting the distributed }'ield characteristics a — Et I <p(<j*)d<j* + o-VCo-" )du*; e > 0 (2)
of the individual components of the system. J Et Jo
The parallel-series model has been suggested by several authors
It will be noted that in this case E is just the slope of the stress-
as a simple model for the yielding behavior of materials but rela-
strain curve when e —> 0. Moreover, if the first term vanishes as
tively few authors have actually employed the model in studies of
6 —»• co, the sj'stem will have an "ultimate yield stress" given by
such behavior. Massing [4] used this model as early as 1926 to
make some general statements about the behavior of materials,
and Prandtl [7] used a physically different but mathematically <r„ = <r*ip(a*)d(T* (3)
equivalent model as a vehicle for the application of kinetic theory
to a rather wide range of problems associated with rate effects. If the stress is not bounded, this corresponds to the case where a
More recently, Duwez [S] applied the parallel-series model to finite fraction of the yielding elements has an infinite yield level.
single crystals by way of investigating the hypothesis of a The selection of the distribution function tp(<r*) could be ac-
secondary structure in crystals and showed that the model could complished ex nihilo or could be based on experimental data.
be made to give stress-strain curves and hysteretic energy loss One possible method of selection based on experimental data
results which were in close agreement with experimental data. woidd be to make use of an experimental^' determined initial
Drucker [9] has also discussed this model and has indicated some loading stress-strain curve. From equation (2), it is seen that the
of its advantages and shortcomings. distribution function would be related to the curvature of this
The series-parallel model has received only little attention in stress-strain curve by the formula
the literature. Ivlev [10] has discussed an n-element one and 1 d2<r
two-dimensional series-parallel model with viscosity and has ^ = * r
used this model to obtain a set of stress-strain relations for a
three-element system, but has not considered the general dis- A method of selection which is based on equation (4) and which
tributed element model. Prager [11] has given an interesting results in a very simple form for the hysteresis relations is dis-
extension of Ivlev's work by suggesting a kinematic analog for cussed in reference [12].
Ivlev's two-dimensional model. The parallel-series formulation will always lead to stress-strain
relations in a form which gives stress as a function of strain. This
The purpose of the present paper is to both complement and
is probably the most convenient form for the dynamicist since
extend the work of earlier authors by bringing together the salient
strains or displacements are normally taken as the independent
features of the two basic models into a consistent framework, while
at the same time examining certain new areas of application.
Before considering the specific models, it should be noted that
the formulation of the hysteretic behavior can be accomplished
either in terms of force-deflection relations or stress-strain rela- A A ^
tions. The former is more natural for composite systems while the
latter is more natural for continuous systems. As the two formu- A2,E2
lations are parallel in eveiy way, only the stress-strain approach
will be used here. This will make it somewhat easier to extend
the models to three dimensions later.

Parallel-Series Model
The parallel-series model consists of a collection of elastic and
slip elements arranged as indicated in Fig. 1, where cr,*, Eit and
A t are the critical slipping stress, the elastic modulus, and the • A.i ,E.i
area of the ith element. Let the slip elements be of negligible
F = <xA
cr•
size and let it be assumed that each element is initially in an un-
stressed state.3 Then it is easily shown that the initial loading A A A -
behavior will be described by
Fy oV*
<r = F/A = Y. E (1)
!= 1 N i = n + l -V
where the summation from 1 to n includes all of those elements • A ,E
which remain elastic after loading to a strain e, and the summa- N N
tion from « + 1 to iV includes all of those elements which have N
slipped or yielded. A A A
In general, both a* and E{ could be distributed parameters.
However, as illustrative of the general situation, in what follows,
3 The latter assumption is not essential and could be relaxed if de-

sired. However, this would necessitate the introduction of another X =€ L

parameter into the problem which would specify the distribution of
preload stresses in the model and this, in turn, would lead to con-
siderable complication of the ensuing equations. Furthermore, it
would be quite difficult to determine what form this distribution A. = A / N
should take on the basis of experimental stress-strain data. The i
usual assumption is to neglect this factor. Fig. 1 Parallel-series model

Journal of Applied Mechanics SEPTEMBER 1 9 6 7 / 613

Downloaded 03 Sep 2012 to 125.236.238.108. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 variables in the equations of motion. On the other hand, most of For the series-parallel model, the lower loop curve will be
the equations of the incremental theory of plasticity are based
, a r«
upon the idea that the strain should be separable into an elastic (a + <r*)y (a*)d<r*
and a plastic part, the total strain being just the sum of these two. * B Jo
The series-parallel model will be more acceptable from this point r
of view. + - a*)y(<7*)da* (10)
E J (ldm - <r)/2
Series-Parallel Model In order to demonstrate that this class of systems satisfies
The series-parallel model consists of a collection of elastic and
Massing's hypothesis, it need only be noted that transforming
slip elements arranged as indicated in Fig. 2 where a * is the
either equation (9) or (10) by means of the equations e' = (e,„ —
critical slipping stress of the itli element and each elastic element
e)/2 and a' = (<r„, — tr)/2 gives equations in e' and a' which are
has an equal elastic modulus E' and nominal length l,-
identical to the corresponding initial loading equations. This is
= l/N. For conceptual purposes, let each elastic element have
not at all surprising, as Massing originally based his hypothesis
an area A. Then, upon initial loading, the strain of each set of
on a similar consideration of the behavior of a collection of slip
elastic and plastic elements will be
elements.
e, = 0; 0 < a < tT;* It is often easier to obtain experimental data on the cyclic
er — <J, i = 1, . . ., Ar (5) energy loss per unit volume of a hysteretic material than it is to
<T > <T,* obtain data on the initial loading or cyclic stress-strain relations.
E'
For the present class of systems, this specific energy loss per cycle
and the total strain will be is related to the yield distribution functions and thereby to the
I Ar
k ii stress-strain relations in a veiy simple manner. For the parallel-
1+hE'
+
' / + /„ ; t lE Ar
(6)
series model, it may be shown that
When the number of elements becomes very large, the distribu- dW(tm)
tion of the slip stresses a * may be described by a distribution <p(Etm) = (11)
4 emE°- dt„
function 7(cr*), where y(a*)da* is the fraction of the total ele-
ments having a slip stress between a* and a* + d<r*. The initial where IF(e„,) is the energy loss per cycle per unit volume ex-

r
loading stress-strain relation then becomes pressed as a function of strain amplitude. For the series-parallel
1 model, the equivalent expression is
£ = <r + a (a - <r*)y(<r*)d<r* & > 0 (7)
¥ Jo <xy(<rj = J_ dnV(aJ
(12)
where E = E'(l -f l<,)/lo is the small strain modulus of the system, E 4trm d<r„,
and a may be considered an arbitrary constant. In this case, it
is seen that the stress is not bounded for e —»- co. In fact, the where IF(<rm) is the specific energy loss expressed as a function of
modulus of the system approaches a constant value of E/{ 1 + a). the stress amplitude. Thus the distribution function is com-
As in the case of the parallel-series model, the distribution func- pletely specified if the specific energy loss is known. Moreover,
tion y(<r*) could be determined from experimental monotonic the distribution functions are related to the stress-strain relations
stress-strain data by making use of the fact that, from (7), through equations (4) and (8). Hence knowledge of the specific
energy loss is also sufficient to determine the stress-strain relation
«7(<r) = (8) for this class of systems. For example, in the case of the parallel-
do-
Cyclic Loading-Fixed Limits
In general, for cyclic loading, the functions <p(cr*) and 7(cr*)
could themselves be functions of the maximum stress or strain J. ,E -A..E'
- V
amplitude, the number of stress reversals, or some other factors.
Thus it would be possible in principle to account for such ob- K v y H
served behavior as cyclic softening and hardening. However, for
the purposes of the present analysis, it will be assumed that the v J—

distribution functions are independent of such factors. Then, if

the system is initially loaded to some state of stress and strain t; i/U
denoted by <r„, and 6„„ "unloading" to a state —em or —<r,„ and Fig. 2 Series-parallel model
then "reloading" to a,„ will result in a symmetrical hysteresis loop
as indicated in Fig. 3.
For the class of systems considered here, the form of the stress-
strain relation for the lower loop curve will result from the action
of three groups of slip elements; those elements which were not
yielded (or caused to slip) upon initial loading and therefore re-
main in an unyielded state, those elements which were yielded in a
positive sense upon initial loading but have stopped slipping, and
those elements which were yielded in a positive sense upon initial
loading but have now yielded in a negative sense. Thus, for the
parallel-series model, the stress-strain relation for the lower loop
curve becomes

a = — | <T*<p(<T*)da*
- JFo

J
+
>Eem

E(Cm-t)
,/2
(Ee - Ee,„ + a*)<p(a*)da*

+ Et J <p(<r*)d<j* (<J)
J Etm Fig. 3 Cyclic loading—fixed limits

614 / SEPTEMBER 1 9 6 7 Transactions of the A S M E

Downloaded 03 Sep 2012 to 125.236.238.108. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 NORMALIZED STRESS AMPLITUOE
Fig. 4 Cyclic loading—variable limits
Fig. 5 Cyclic energy dissipation comparisons

series model, the simultaneous solution of equations (4) and (11)

gives (ab - o*)y(<r*)da*
= E + E U(«-„t>/:
. . f i - J ' i j dem (13) / * ( « — <rc)/2
ie dt
Jo -n- L m <ltm L=oJ + {<rc + <r*)y(a*)d<r*
J (a-«,)/2

f
which is the equation of the stabilized initial loading curve. It is
interesting to note that this equation indicates that there will be (<r - a*)y(a*)da* (lo)
no hysteresis loss unless the slope of the specific energj'loss func- +
Jo
tion changes from its value at e,„ = 0.
The previous process could be carried on indefinitely with little
difficulty merel}' by keeping track of the fraction of elements in
Cyclic Loading—Variable Limits each "yielding category" after every loading reversal.
In order to perform a dynamic analysis of the transient re-
sponse of a yielding system, it is necessary to be able to describe
the stress-strain behavior of the system for cyclic loading between Specification of Parameters—Some Comparisons
variable limits. It is this requirement that leads to difficulty in The parameters describing the present class of models may be
purely empirical formulations. However, for the present class of determined directly from experimental observations of the .system
systems, transient behavior is well defined. Consider, for ex- in question or may be deduced from empirical formulations based
ample, the situation indicated in Fig. 4. The system is initially upon such observations. Several empirical formulations for the
loaded to some state denoted by b, unloaded to c, reloaded to d, uniaxial stress-strain behavior of continuous systems have been
and so on. The portion of the loop from b to c will be the same proposed based upon the results of experimental testing. Un-
as for the steady-state cyclic case so long as |e(.| < |e,,|. If fortunately, these empirical formulations are expressed in a
tal > Mi the system will follow the steady-state loop curve until variety of different forms so that comparisons are often difficult.
c = — e4 and then continue along an extension of the initial load- However, within the framework of the present class of models,
ing curve. If the direction of loading is changed when tal > tali these different formulations can easily be reduced to a common
the new loading curve will have the same form as the steady-state form for comparison. This has been done for three of the more
loop curve, equation (9), except that em would be replaced by popular formulations, and the results are shown in Fig. 5, where
— ec, and the sign of all a and e-terms would be changed. If the the specific energy loss per cycle is given as a function of stress
direction of loading is changed when tal < |e6|, the situation is a amplitude.
bit more complex but not unreasonably so. The fraction of ele- Pisarenko [13] proposed a formulation reportedly due to
ments in any given yielded or unyielded state is known at c so one Davidenkov in which the equation of the lower loop curve is
need only determine what happens to each of these separate
groups as the loading is reversed. In the case of the parallel-
<r = E { e + - [(e„, e)» - 2«->e,„»] (16)
series model, this will lead to an expression for the stress-strain n
curve between r and d of the form
and his results indicated that n = 2 for St 20 steel and n = 3 for
brass. .Morrow [3], on the other hand, proposes a relation of the
(T =
. | F CT*<p(lT*)(la* form
Jo

+
f I

r*Ea
|
e-ft)/2
(Ee -

(Ee -
Eec -

Etb +
a*)(p(a*)ila

a*)<p(c*)da*
e = a/E + Ka"

and finds that /n ~ 0.15 for a large number of materials of engi-

neering interest. Lazan [14] has expressed the specific energy
loss per cycle empirically and has found that the mean variation
(17)

J JC(a-
-u)!2 for many materials not selected for high damping can be ex-
pressed by a two-segment line as indicated in the figure.
+ Ee
?e i ^(o* )da* (14) Only the general shape of the different specific energy loss
J Ka
functions is shown, as their relative positions would depend upon
delation (14) will be valid for ec < e < eb. When e > et, the the specific valuesof empirical parameters for particular materials.
system again follows the initial loading curve, equation (2). Thus the stress has been arbitrarily normalized by the stress at
For the series-parallel model, the corresponding relation would which the slope of the stress-strain curve becomes horizontal in
be Pisarenko's formulation.

Journal of Applied Mechanics SEPTEMBER 1967 / 615

Downloaded 03 Sep 2012 to 125.236.238.108. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 It is seen that the results of Lazan and Morrow are in almost the "origin" of the ?ith surface (n used here as a superscript
perfect agreement for high stress levels. This is the region of always denotes the rath yield surface). It will be further as-
interest for low-cycle-fatigue studies, and it is generally agreed sumed that the tensor <x(sn is related to the plastic strain associated
that the energy loss in this region results primarily from rate with the nth surface (e,/p)n through the equation
independent hysteresis. The results based on Pisarenko's formu- oCij" = e(€,/)" (19)
lation differ noticeably from both those of Lazan and Morrow,
where c is a constant independent of the surface. Thus the yield
one of the major causes being the boundedness of the stress in
Pisarenko's formulation. However, the differences in energy dis- surface translates in a direction parallel to the increment of plastic
sipation between the three formulations for high stress levels in- strain at any point in the stress history of the material. In order
dicated here are, in general, no larger than the variations in dis- to complete the formulation, it will be assumed that the plastic-
sipation associated with other factors such as cyclic softening and strain increment associated with each surface obeys the flow rule
hardening which are not considered here. For low stress levels, of von Mises and, therefore, that
the slope of the Pisarenko (St 20) curve approaches a value of
P)n = -^JL dX„; d\„ > 0
3:1, which is near the highest values observed by Lazan. The (20)
slope of the Pisarenko (brass) curve approaches a value of 4:1,
which is appreciably higher than the maximum slope observed by = 0; d\„ < 0
Lnzan. Furthermore, all of the surfaces with d\„ > 0 must pass through
Although the energy comparison of Fig. 5 is interesting in its the point p in stress space. Thus it may easily be shown that
own right, within the context of the present discussion it serves a W n
more important role as an indicator for the selection of yield dis- da ij
tribution functions for the different distributed element models. d\ - ^ (21)
These yield distributions are related to the specific energy loss c dF,
O ) / j > M

through equations (11) and (12). Thus, for example, it can be dov
h:l j \ <><Iu /
shown that the Pisarenko formulation for St 20 corresponds to a for all n such that d\n > 0. When c/X„ < 0, the material behaves
distribution function for the parallel-series model <p(a*), which is elastically.
a constant, while that for brass corresponds to a <p(a*), which is Since the total plastic strain will be the sum of the plastic
proportional to <r*. Similarly, Lazan's average dissipation curve strains associated with each individual surface, the overall work-
for high stress leads to a distribution function for the series- hardening behavior of the material will depend on the combined
parallel model y(a*) which is proportional to a* to the fifth action of a number of separate surfaces. Furthermore, since the
power. work-hardening law of each individual surface has been assumed
Very little experimental data on the hysteretic behavior of com- to be linear, the form of the overall work-hardening behavior will
posite systems and structures are available. However, in one only be a function of the distribution of these surfaces. If there is
case where such data were available, the yield distribution func- a finite number of individual surfaces, the overall work-hardening
tion ip(o*) was fairly well approximated by a constant value over relation will be piecewise linear. On the other hand, if the yield
a large stress range [12], surfaces are distributed continuously throughout some region of
stress space, the overall work-hardening behavior will also be
Extension to Three Dimensions continuous and nonlinear. After an initial loading of the ma-
terial, the distribution of the yield surfaces will be changed and,
The concepts of the present class of models, especially those
therefore, so will the overall work-hardening behavior for subse-
of the series-parallel model, may be carried over into the frame-
quent loading or unloading. It is this feature that leads to a
work of the incremental theory of plasticity with little dif-
Bauschinger effect of the Massing type in the present formulation.
ficulty. In the incremental theory, the behavior of a yielding
material is described in terms of an initial yield condition or yield In order to demonstrate more precisely what the previous
surface which specifies the stress state for which yielding first formulation implies, consider the special case where the yield
occurs, a work-hardening rule which specifies how the yield surfaces are given by the von Mises condition. Then these sur-
surface is changed during plastic flow, and a flow rule which re- faces will be cylinders in principal stress space, and their projec-
lates the plastic strain increment to the state of stress and the tions on the 7r-plane of this space will be circles. Let Fig. 6(a)
stress increment. The only difference between the present represent the distribution of the yield surfaces (in this case, a
formulation and the classical approach will be the introduction of finite number) for the virgin material. Then, as the element is
a collection of yield surfaces in place of the usual single surface. initially loaded, the stress point p contacts first one and then
Each of these surfaces will individually obey a linear work-hard- another yield surface. As a result of the kinematic hardening,
ening law of the type developed by Prager [5] but their combined each surface is carried along with p so long as d\n > 0, as indicated
action will, in general, give rise to a nonlinear work-hardening law schematically in Fig. 6(6). Now the original distribution of the
for the material as a whole. This approach will lead to a realistic yield surfaces will be altered by the loading to point A. Thus,
Bauschinger effect of a type which could not be obtained by the even if the element is "unloaded" along the same path as the
usual technique of using a single yield surface and a nonlinear initial loading, the stress-strain relation and the overall work-
work-hardening law even with kinematic hardening. hardening behavior will be changed. This is indicated sche-
matically in Fig. 6(c). On loading to A, the stress point p would
Let the state of stress of an element of the yielding material be
have encountered four separate yield surfaces resulting in four
represented by a point p in the nine-space <r,y, and let the corre-
slope discontinuities in the stress-strain relation. However, on
sponding total strain consist of the sum of an elastic strain e,/
unloading, the stress point would only have encountered two
which obeys Hooke's law and a plastic strain e,y which is to be
surfaces with only two resulting discontinuities. It can readily be
deduced from the incremental theory. Then, following the sug-
seen that this would lead to a piecewise linear work-hardening
gestion of Prager, a particular yield surface in this space can be
behavior exhibiting a Bauschinger effect.
represented as
When the yield surfaces are distributed continuously, it may be
F„(a,j - a , - / ) = kn" (18) shown that the total incremental plastic strain at any point along
where kn is a constant for each n, and a 1; n is a tensor specifying the loading path of the element will be

rff,(» = { /»i* ^<1-,, -

Jo
a,j(k) - - i (o- - a ( f c ) ) 5 0 J j^o-,,,, a,m(k) - (o- - a(k))5lm
T(k)dk\ dabr (22)

616 / SEPTEMBER 1 9 6 7 Transactions ot the A S M E

Downloaded 03 Sep 2012 to 125.236.238.108. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 manner as indicated here. However, this idea was never ex-
tended to incorporate the concept of kinematic hardening as a
means of obtaining a more reasonable nonlinear work-hardening
behavior. The present formulation might therefore be thought of
as an extension of Koiter's idea of singular yield surfaces even
though it is stated in somewhat different terms.

Summary and Conclusions

The results of the preceding analysis maj' be summarized as
follows:
1 A class of one-dimensional models for the yielding behavior
(a) (b)
of materials and structures has been examined. This class con-
sists of a collection of ideal slip and elastic elements.
2 It has been shown that this class of models leads to stress-
strain relations which exhibit a Bauschinger effect of the Massing
type.
3 Both the steady-state and uonsteady-state cyclic behavior
are completely specified if the initial monotonic loading behavior
is known.
4 Relations between the cyclic specific energy dissipation
and the stress-strain relations have been presented, and these
have been used as a basis for the comparison of several empirical
hysteretic formulations. The empirical results were in turn used
to indicate the possible form of the major model parameters.
5 The concepts of this class of models have been extended to
three dimensions with a resulting generalization of the customary
formulation of the incremental theory of plasticity.
(c) (d)

Fig. 6 Schematic behavior of yield surfaces in 7T-plane References

1 Drucker, D . C., "Plasticity," Structural Mechanics: Proceedings
of the First Symposium on Naval Structural Mechanics, Goodier and
where ) is the a,-,-" for all of those surfaces with kn — k, a = Iloff, eds., Pergamon Press, New York, 1960.
a(k) = «,-,•(&), and T(k)dk is the fraction of the total number 2 Naghdi, P. M., "Stress-Strain Relations and Thermoplas-
of yield surfaces with a value of k between k and k + dk. The ticity," Plasticity: Proceedings of the Second Symposium on Naval
Structural Mechanics, Lee and Symonds eds., Pergamon Press, New
integral is carried out over all elements for which d\ > 0. Ex- York, 1960.
pression (22) along with knowledge of the dependence of a(k) on 3 Morrow, J., "Cyclic Plastic Strain Energy and Fatigue of
e,-/' would in principle allow the solution of a general multiaxial Metals," Internal Friction Damping, and Cyclic Plasticity, American
stress problem. Society for Testing Materials, 1965.
4 Massing, G., "Eigenspannungen und Verfestigung beim
If the initial loading is one-dimensional, as indicated in Fig. 6(d), Messing," Proceedings of the Second International Congress of Applied
the general stress-strain behavior is simplified considerably and Mechanics, 1926, pp. 332-335.
it is possible to relate the three-dimensional formulation to that 5 Prager, W., " T h e Theory of Plasticity: A Survey of Recent
of the one-dimensional model. Let the loading take place along Achievements," (James Clayton Lecture), Proeedings of The Insti-
tution of Mechanical Engineering, London, Vol. 169, 1955, p. 41.
the (Ti-axis, then, if the surfaces are imagined to be continuously 6 Shield, R. T., and Ziegler, H., "On Prager's Hardening Rule,"
distributed, Zeitschrift filr Angeicandte Mathematik und Pliysik, vd. IXa, Fast-. 3,
1958, pp. 260-276.
7 Prandtl, L., "Ein Gedankenmodell zur kinetischen Theorie
T(k)dk > da (23) der festen Korper," Zeitschrift fur angewandle mathematik und
mechanik, Vol. 8, No. 2, Apr. 1928, pp. 85-106.
8 Duwez, P., "On the Plasticitv of Crystals," Physical Review,
Integrating gives Vol. 47, 1935, pp. 494-501.
9 Drucker, D . C., " O n the Continuum as an Assemblage of
V2<r,/3
[ffi - 3k/\/2]V(k)dk (24)
Homogeneous Elements or States," Brown University Technical Re-
3c I
port No. 50,1966.
10 Ivlev, D. E)., " T h e Theory of Complex M e d i a , " Soviet
Physics—Doklady, V o l . 8 , No. 1, July 1963, pp. 28-30.
But this equation is identical to that obtained for the one-dimen-
11 Prager, W., "Models of Plastic Behavior," Proceedings of the
sional series-parallel model if Fifth U. S. A ational Congress of Applied Mechanics, A S M E , 1966, pp.
T

447-448.
c = 2E/3a 12 Iwan, W . D., " A Distributed-Element Model for Hysteresis
(25) and Its Steady-State Dynamic Response," JOURNAL OF A P P L I E D
MECHANICS, V o l . 33, N o . 4, TRANS. A S M E , V o l . 88, Series E , D e c .
1966, pp. 893-900.
13 Pisarenko, G. S., "Vibrations of Elastic Systems Taking
Therefore, all of the features of the series-parallel model, including Account of Energy Dissipation in the Materials," W A D D T R 60-5S2,
its characteristic Bauschinger effect, carry over into the three- Feb. 1962.
14 Lazan, B. J., "Energy Dissipation Mechanisms in Structures,
dimensional analj'sis with the distribution of the yield surfaces With Particular Reference to Material Damping," Structural Damp-
determined by the distribution of yielding elements in the one- ing, A S M E , Dec. 1959.
dimensional model. 15 Batford, S. B. and Budiansky, B., " A Mathematical Theory of
The approach presented here is similar in some respects to the Plasticity Based on the Concept of Slip," N A C A T N 1871, 1949.
16 Koiter, W . T., "Stress-Strain Relations, Uniqueness, and
slip theory of Batdorf and Budiansky [15] as interpreted by Variational Theorems for Elastic-Plastic Materials With a Singular
Ivoiter. Koiter [16] showed that the slip theory could be visual- Yield Surface," Quarterly of Applied Mathematics, Vol. 11, No. 3, pp.
ized in terms of a collection of yield surfaces in much the same 350-354.

Journal of Applied Mechanics SEPTEMBER 1967 / 617

Downloaded 03 Sep 2012 to 125.236.238.108. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm