You are on page 1of 24

July 5, 2004 11:48 am

Rebec

DRAF
ca Br ann on

APPENDIX B. Further explanation of the plasticity equations.


This appendix first provides a qualitative newcomers introduction to plasticity by discussing the analog of plasticity with a spring and friction element connected in series. This simple model reproduces some essential features of non-hardening plasticity (existence of an upper limit on achievable stress and hysteresis) in a readily understandable onedimensional context. The model is extended to a two-dimensional system to introduce the concept of a yield surface, plastic normality, and vertex behavior). These simplified models are only thought experiments. We outline how to generalize what is learned in this simplified context to create a full 3D tensor plasticity model. This discussion will allow us to provide derivations or explanations of the governing equations that were stated without proof in Eqs. 3.9 through 3.12 of the main text.

Spring-friction models of plasticity


This section illustrates a classic 1D system and a new 2D model that can be used to introduce new students to plasticity modeling. The models described in this section share many features with general tensor plasticity theory, but the analogy is not perfect in every respect. Friction-spring element. Elementary plasticity texts [50] frequently introduce the concept of nonhardening plasticity by calling attention to its analog with a friction-spring element, which is simply a spring in series with a friction plate. The spring is governed by F = k e (B.1) where k is the spring constant and e is the elastic extension (change in length) of the spring under the applied force F . Incidentally, the friction-plate Jenkins element is conventionally depicted as k Y
Figure 2.1. Jenkins element (friction plate connected in series with a spring)

friction-spring JENKINS element (k,Y)

friction plates

linear spring

B-1
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

AFT DR
ann ca Br Rebec on

July 5, 2004 11:48 am

The parameters are the spring constant k and the critical sliding force Y . Please bear in mind that these sorts of models are of only limited use in the 3D context. Working with such models can help build your intuition and strategies, but must generally be abandoned in realistic contexts. The friction plate* is governed by the property that no extension (slipping) is allowed unless the force reaches a critical value Y . Once motion of the plates commences, the resistive force remains equal to the constant friction resistance force Y . Because the spring is connected in series with the friction plate, each one is loaded with force F . Consequently, during slip (when the force equals the constant Y), the spring extension must be given by e = Y k and will remain at this value until a load reversal occurs. Let denote the total extension of the element, measured from its initial state. Because the friction and spring elements are connected in series, the total extension must equal the sum of the individual extensions e from the spring element and p from the friction element:

= e + p (B.2) Upon removal of any load, only the elastic part of the total extension e is recovered, while a residual extension p , equal to the slip of the friction plates, remains unrecovered. We use the superscript p because this unrecovered displacement is in a sense permanent. Because p will be later seen to be analogous to plastic strain, we will refer to p as the plastic displacement. Behavior of the friction-spring element is history dependent in the sense that knowledge of the entire history of the total strain is generally needed to determine the force on the system at any given time. Consequently, as is done in general plasticity theory, the governing equations will be presented in rate form, which can be integrated through time from a known initial state to solve for the state at any later time.

* If the friction is accomplished by placing a block on the upper friction plate, then the critical yield force needed to induce sliding would be Y = W , where is the coefficient of friction and W is the weight of the block. To avoid issues associated with inertia associated with the mass of the block, the friction plate could be envisioned as wrapped tightly with wire. Of course, we are presuming here that the static and dynamic coefficients of friction are identical this is just a thought experiment after all. When the static and dynamic friction coefficients are not equal, the force-displacement curve exhibits a small peak similar to what is observed in actual materials. The difficulty in including it here, however, is that the nature of the peak depends on inertial effects, which in turn depend on loading rate.

B-2
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

July 5, 2004 11:48 am

Rebec

DRAF
ca Br ann on

To summarize, the friction-spring element is described by the following governing equations, now expressed in rate form: = e + p If F < Y , then the response is elastic If F = Y and F 0 , then the response is elastic. If F = Y and F > 0 , then the response is plastic. Here, elastic behavior is governed by p = 0 , and e = = F k and plastic behavior is governed by F = 0 e = 0 , and p =
(B.3a) (B.3b) (B.3c) (B.3d) (B.3e) (B.3f) (B.3g) (B.3h)

A superimposed dot denotes the time derivative. Of course, the friction-spring element is rate independent, so time is actually any monotonically increasing scalar used to parametrically describe the history each state variable. Equation B.3a is just the rate form of Eq. B.2. The inequality in Eq. B.3c is expressing the notion that, if the force is currently at the critical sliding value but the direction of motion is in the opposite direction, then subsequent motion will not produce frictional sliding. These equations have been expressed in rate form to better emphasize the analogy with the plasticity equations considered in the main text of this manuscript.* The equations governing the friction-spring element are occasionally presented more compactly (but less clearly) in Kuhn-Tucker form as ( F Y ) p = 0 (B.4)

* Even without history dependence, rate forms of governing equations are frequently useful. Suppose, for example, a system depends on a set of nonlinear functions such as u = f ( r, s ) and v = g ( r, s ) . The complete set of governing equations might not be solvable for r and s . However, in rate form, the equa tions u = f ,r r + f ,s s and v = g ,r r + g ,s s form a linear (and hence solvable) system with respect to rates.

B-3
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

AFT DR
ann ca Br Rebec on

July 5, 2004 11:48 am

Figure 2.2 shows an example of a force-displacement path for the friction-spring element corresponding to a prescribed displacement path that increases monotonically until a peak value is reached. At that point, the displacement is returned monotonically back to zero. Returning to a zero displacement requires pushing against the friction plate, so there is a residual stress when the displacement returns to zero. Upon release of this stress, there is a residual non-zero displacement residual . F DRIVING FUNCTION RESPONSE FUNCTION Y
residual

t Y

F Y

PARAMETRIC COMBINATION OF THE ABOVE GRAPHS


peak attainable tensile force

slope = k
residual
load released back to zero

peak attainable compressive force


strain returned to zero

Figure 2.2. A sample force-displacement curve. The total displacement is prescribed as a function of time (upper left), resulting in a force response (upper right). When F = F ( t ) is plotted parametrically versus = ( t ) , with the parameter being time, the result is the force displacement curve (bottom).

Each prescribed total displacement history will result in a unique force response history. The converse is not true. It is not appropriate for the force F to be regarded as a driving function through time. If, for example, the force history F = F ( t ) shown in the upper right of Fig. 2.2 were specified, then strain histories other than the one shown in the upper

B-4
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

July 5, 2004 11:48 am

Rebec

DRAF
ca Br ann on

left of Fig. 2.2 could result in the exact same force history. For example, the force could be brought up to the value Y and then simply held there for the amount of time elapsed on the plateau, without further increasing the displacement at all (i.e., the same force history could be attained by staying entirely in the elastic regime). Even though the force response function is determined uniquely by the displacement driving function, that does not mean that force is a function of displacement. If force were a true function of displacement, then each value of displacement would have to correspond to a unique force. Fig. 2.2 shows that this is not the case; for example, the half-way displacement, 2 , corresponds to two different forces on the force-displacement curve, F = +Y from the extension phase and F = Y from the recompression phase. The fact that force is not a function of displacement is important when considering energy and work. The work increment associated with motion of the friction-spring element equals the applied force times the increment in total displacement:
t

W(t) =

F dt
0

(B.5)

Its important to not write dt as simply d because doing so would (notationally) suggest that the force is a proper function of the displacement. If this were true, then we would have Fd = 0 for any closed displacement cycle (i.e., a cycle for which the dis placement varies from and then returns to its original starting value). However, the actual work is nonzero it equals the nonzero area enclosed by the stress strain curve. The elastic work is defined by
t

We( t ) =

F dt
0

(B.6)

and the plastic work is defined


t

Wp( t ) =

F
0

dt

(B.7)

Whenever plastic slip occurs, recall that Eqs. B.3d and B.3f require F > 0 and e = 0 . Consequently, F p 0 and therefore Wp 0
(B.8)

In words, the friction-spring element will never exhibit negative plastic work. It always requires work input to deform this system plastically. B-5
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

AFT DR
ann ca Br Rebec on

July 5, 2004 11:48 am

A two-degree-of-freedom friction system. The one-dimensional friction spring element nicely illustrates some basic features of plasticity. However, it fails to illustrate other essential and more complicated properties of general multi-dimensional plasticity. Below, a multiple degree of freedom toy problem is used to illustrate the notions of stress space, plastic normality, yield surfaces (with flats and vertices), and special stress and strain measures. Consider two identical friction elements (having spring constant k and flow force Y ) connected to a rigid T-shaped control arm, as sketched in Fig. 2.3. The individual frictionspring elements, numbered 1 and 2, have associated displacements and forces 1, 2, F 1, F 2 . Our goal is to analyze the T-bar system so that knowledge of these microscale or internal variables is not needed to describe the overall system response. In other words, we wish to figure out a way to compute the macroscale system loads P 1 and P 2 whenever the macroscale system displacements, 1 and 2 , are prescribed through time. We want to eliminate all mention of 1, 2, F 1, F 2 from the final equations. The geometry and displacement definitions shown for our T-bar system imply that 2 + 1 1 = ---------------2 2 1 2 = ---------------2 where a 2 2 -- L
(B.11)

<--> <-->

1 = 2 1 2 = 2 + 1

(B.9)

(B.10)

The forces, P 1 and P 2 , applied to the T-bar system in Fig. 2.3 result in friction-spring element loads of F 1 and F 2 respectively (positive in tension). Balance of force and torque gives us a way to express any function of the microscale internal forces, F 1 and F 2 , in terms of the macroscale loads, P 1 and P 2 : P1 = F2 + F1 P2 = F2 F1 where L P 2 P 2 -- a
(B.14)

<--> <-->

1 F 1 = -- ( P 1 P 2 ) 2 1 F 2 = -- ( P 1 + P 2 ) 2

(B.12)

(B.13)

Equations B.11 and B.14 introduced special alternative measures of displacement and force. In formal plasticity theory, the issue of best strain and stress measures becomes considerably more complicated, but this simple system illustrates the basic motivations nicely. By defining 2 as we have, we have absorbed a geometrical aspect of the system (namely, the ratio of the T-bars arm lengths) into our displacement and force measures in B-6
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

July 5, 2004 11:48 am

Rebec

DRAF
ca Br ann on

order to simplify the governing equations. Note that the displacement measure 2 equals the actual displacement times the ratio a L whereas the alternative force measure P 2 equals the actual force P 2 divided by a L . Consequently, these alternative displacementforce definitions are said to be conjugate to each other because P 2 2 = P 2 2 . The state of the T-bar system can be categorized into four possibilities, depending on whether each individual friction-spring element happens to be slipping plastically or deforming purely elastically. The first friction-spring element behaves elastically when F 1 < Y , or using Eq. B.12 to remove mention of the microscale force, this element is elastic if P 1 P 2 < 2Y . Likewise, the second friction-spring element is elastic if P 1 + P 2 < 2Y . The T-bar system as a whole is behaving elastically only if both friction-spring elements are behaving elastically. Consequently, there exists a range of allowable P 1 and P 2 loads for which the system will be elastic. Fig. 2.4a shows the load space of P 2 vs. P 1 , where the fully elastic region in the shaded square diamond. Regardless of whether real or scaled force definitions are used, the force space can broken into categories according to whether the individual friction-spring elements are behaving elastically or plastically. For each individual spring element, forces exceeding Y in magnitude are unachievable. Consequently system forces lying outside the shaded region are unattainable. Forces in the interior of the shaded region correspond to fully elastic system response. Whenever the T-bar system is experiencing frictional slip, the stress state will lie on the boundary of the diamond-shaped fully elastic region. This boundary is called the yield surface. This surface is a square diamond in P 2 vs. P 1 load space. However, recalling that P 2 = P 2 ( L a ) , the yield surface becomes a distorted (non-square) diamond in actual P 2 vs. P 1 load space [as shown in Fig. 2.4b for the case that L = 2a ].

B-7
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

AFT DR
ann ca Br Rebec on

July 5, 2004 11:48 am

Initial state
L

a P2

Loaded state
1 1 2

P1

p 1 p 1

Unloaded state
p 2

p 2

Figure 2.3. A two degree-of-freedom friction system. Depending on the nature of the applied loads, P 1 and P 2 , the friction elements may or may not undergo frictional sliding. After slip occurs, the T-bar will have permanent (unrecovered) deformation upon release of the loads.

B-8
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

July 5, 2004 11:48 am

Rebec

DRAF
ca Br ann on

P2

PP PE EP PE PP EE PP PP EP EE EP EP PE PP PP PE PP

P1

PP

P1

left-hand figure shows the yield surface in the scaled P 2 vs. P 1 force space, where P2 = P2 ( L a ) . The right figure shows the yield surface in real force space, where (for illustration purposes) we have used L = 2a . The yield surface contracts vertically because lower values of P2 are required to induce yield whenever P2 has a larger lever arm L in comparison to a . Note that the normals to the yield surfaces do not contract horizontally, not vertically.
Figure 2.4.

The elastic region and its boundary, the yield surface The

The fully elastic EE state corresponds to both friction-spring elements being elastic. In this case, p 1 = 0 , p 2 = 0 , and and e 1 = 1 = F2 k e 2 = 2 = F2 k
(B.15) (B.16)

from which it follows that P 1 = F 1 + F 2 = k 1 + k 2 = 2k 1 P 2 = F 1 F 2 = k 1 k 2 = 2k 1 This result can be expressed as a matrix equation, P1 P2 1 = 2k 0 0 2k 2
(B.18) (B.17) (2.17b)

B-9
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

AFT DR
ann ca Br Rebec on

July 5, 2004 11:48 am

The matrix in this equation is analogous to the elastic stiffness tensor in general plasticity theory. If we had selected springs with different spring constants, then the general form of the result would be unchanged, but the matrix in Eq. B.18 would become anisotropic. The EP state denotes loading situations where the first (top) friction-spring element is behaving elastically, while the second (bottom) one is behaving plastically. In this case, p 1 = 0 F2 = Y , and e 1 = 1 = F1 k F2 = 0 e 2 = 0 , and p 2 = 2
(B.19) (B.20)

In this EP case, the fact that F 2 = 0 can be used in Eq. B.12 to observe that P1 + P2 = 0 In other words, since F 2 = Y , we know that P 1 + P 2 = 2Y
(B.22) (B.21)

Again, we wish to infer from these equations how the applied forces P 1 and P 2 are related to their corresponding (conjugate) displacements, 1 and 2 . In other words, we wish to eliminate from the above equations any explicit mention of the microscale forces and displacements 1 , 2 , F 1 , and F 2 . From Eqs. B.2, B.9, and B.10, we know that
p p p 2 + 1 2 1 = ----------------- = ---2 2 p p p 2 1 2 2 = ---------------- = ---2 2

and and

e e e e F1 P1 P2 P1 2 + 1 1 1 = ----------------- = ----- = ----- = ----------------- = ----2 2 2k 2k 4k e e e e P 1 P 2 F1 1 2 1 ---- 2 = ---------------- = ----- = ----- = ----------------- = P 2 2 2k 2 2k 4k

(B.23)

(B.24)

e e In Eq. B.23, we first used Eq. B.19 to set 2 =0 . Then Eq. B.19 was used to express 1 in terms of F 1 . Next, Eq. B.12 was used to express the internal force F 1 in terms of the external, or macroscale, loads P 1 and P 2 . Finally, we made use of Eq. B.21 to replace P 2 by P 1 . In the last step of Eq. B.24, we again used Eq. B.21 to replace P 1 by P 2 . By doing this, we have shown that the elastic part of the response is given by the same equations as the fully elastic response cited in Eq. B.17, except that the system displacements now replaced by the elastic system displacements. In matrix form, the above EP results may be written p 2 1 = ---- 1 2 1 p 2 1e ----- 0 P 1 1 = 2k 12 0 ----- P 2 2k

(B.25)

B-10
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

July 5, 2004 11:48 am

Rebec

DRAF
ca Br ann on

Because the second friction-spring element is (by the EP assumption) slipping, we know that F 2 = Y . Referring to Eq. B.3d, we note that F 2 2 > 0 . Thus, if F 2 = Y , then 2 > 0 . If F 2 = Y , then 2 < 0 . Consequently, we see that the plastic displacement rate vector in Eq. B.25 always points in the direction of the outward normal to the yield surface whenever exactly one spring element is deforming plastically. Of course, this conclusion was reached by using the normalized force and displacement measures P 2 and 2 . However, because P 2 2 = P 2 2 , its straightforward to verify that the same conclusion holds for the real force and displacement measures as well. In general, normality of the plastic displacement rate holds for any conjugate load-displacement definitions. FULLY ACTIVE VERTEX LOADING. Lets now consider one of the PP states under fully active loading. By this, we mean that both friction-spring elements are deforming plastically (later we will consider the case that the stress state is at a PP state, but the displacement rate is such that only one friction-spring element is deforming plastically, while the other is unloading). When F 1 = F 2 = Y , the equations governing the friction-spring elements are given by Eq. B.3e: e p F1 = Y , F1 = 0 , 1 = 0 , and 1 = 1 (B.26) F2 = Y F2 = 0 e 2 = 0 , and p 1 =
(B.27)

From the first equations in the above set, Eqs. B.12 and B.13 require that 1 -- ( P 1 P 2 ) = Y and 1 ( P 1 + P 2 ) = Y -2 2 Both of these requirements can hold only if P1 = 0 and P 2 = 2Y
(B.29) (B.28)

In other words, when both friction plates are sliding in tension, the applied stresses on the T-bar system must be at the far-right vertex of the diamond-shaped yield surface. The fact that both elastic displacement rates in Eq. B.26 are zero means that the T-bar elastic dise e placement rates, 1 and 2 must both be zero (i.e, both springs are fully extended to a loading force equal to Y and therefore both elastic springs have reached their peak elastic e e displacements of 1 = 2 = Y k , which are constant and therefore have zero rates. The T-bar elastic displacements are therefore constant and given by Eq. B.18 with 1 and 2 e e replaced by 1 and 2 and P 1 and P 2 set equal to the values in Eq. B.29. The more interesting question for the PP state is: what is the direction of the plastic displacement rate? For the EP and PE states, we found that the plastic flow direction was normal to the yield surface. In this PP case, however, the stress happens to reside at a yield surface vertex, so the normal to the yield surface is multivalued. From Eq. B.26, we know that the plastic displacement rate is equal to the total displacement rate. Therefore, the T-bar plasp p tic displacement rates 1 and 2 must equal the total rates 1 and 2 . Thus, in matrix form B-11
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

AFT DR
ann ca Br Rebec on

July 5, 2004 11:48 am

p 1 = 1 p 2 2

(B.30)

For both friction-spring elements to be sliding, each one must satisfy the last condition cited in Eq. B.3d. Namely, 1 F 1 > 0 and 2 F 2 > 0 (B.31) or, since we are considering the case that F 1 = F 2 = Y , 1 > 0 and 2 > 0

(B.32)

Using Eqs. B.9 and B.10 to convert these requirements into T-bar quantities, shows that 2 1 > 0 and 2 + 1 > 0 (B.33) which means that the plastic displacement rate in Eq. B.30 must lie within the cone of possible outward normals that exist at the yield surface vertex. More specifically, recalling Eq. B.30, the plastic displacement rate will be precisely aligned with the total displacement rate. This result is quite different from the case where the stress resides on a non-

B-12
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

July 5, 2004 11:48 am

Rebec

DRAF
ca Br ann on

vertex point, where we found that the plastic displacement rate always pointed in the direction of the outward normal. Conversely, if the macroscale displacement rate points into the cone of limiting normals, then both spring elements will be actively deforming plastically. P
2

P2

PP PE EP PE PP EE PP PP EP EE EP EP PE PP PP PE PP

P1

PP

P1

a load state at a vertex, the nature of the plastic strain rate will depend on the nature of total strain rate. If the total strain rate points into the cone of limiting normals (shaded), then the plastic strain rate will be parallel to the total strain rate. If, on the other hand, the strain rate points outside the cone, then the plastic strain rate will be normal to the nearest yield surface (the other surface will become immediately inactive.)
Figure 2.5.

Cone of limiting normals For

PARTIALLY ACTIVE VERTEX LOADING: Now suppose that the stress state lies at one of the PP points on the yield surface, but the direction of loading is such that only one of the friction spring elements continues to deform plastically, while the other one unloads. In this case, the governing equations reduce to one of the EP or PE states. The plastic strain rate will be directed normal to the active portion of the yield surface. Overall, the purpose of the one- and two-dimensional toy problems in this section was to introduce some basic features of plasticity material models, the most important being normality of the plastic strain rate to the yield surface and non-negative plastic work. In plasticity theory, normality may or may not hold, depending on the experimental data, and positivity of the plastic work (or lack of it) must be assessed by examining the second law of thermodynamics. Now that some of the basic mathematical properties of plasticity have been illustrated in a simplified toy problem context, we are ready to begin discussing the tensor formulation of classical plasticity of real materials. B-13
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

AFT DR
ann ca Br Rebec on

July 5, 2004 11:48 am

Incorporating viscoelasticity. Viscoelasticity allows material response to be rate dependent and to exhibit a memory effect. Viscoelasticity is fundamentally different from non-hardening plasticity. A very simple viscoelastic 1D element is k Here, where we formerly had a friction plate in the plasticity model, now we have a dashpot (viscosity ). Numerous alternative configurations of springs and dashpots have been explored in the literature. Viscoelasticity models always involve combinations of springs and dashpots. Plasticity models always involve springs and friction plates. Visco-elasticplastic models always involve combinations of springs, friction-plates, and dashpots. For example, to incorporate viscous effects in the Jenkins element in Fig. 2.1, you could connect a dashpot in parallel with the friction element: k Y

Figure 2.6. Viscous Jenkins element (viscous friction in series with a spring)

With this sort of configuration, rate dependence would not commence until yield is reached. An alternative way to incorporate rate effects into a 1D plasticity model is to replace the linear-elastic springs in the rate-independent plasticity model with viscoelastic elements. Naturally, your choices for 1D approximate models should be guided by the qualitative nature of your experimental data for real materials. One-dimensional models serve only as thought experiments that can help you identify a reasonable general structure for your generalized 3D tensor models. Warning: whatever choices you make, be sure to verify that your model gives reasonable results for non-monotonic (load-unload) strain paths. There are many models out there that look great until tested under severe load reversals. Always push your model beyond the realm for which data exists to verify that its qualitative results are reasonable.

B-14
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

July 5, 2004 11:48 am

Rebec

DRAF
ca Br ann on

3D Tensor analogs of the 1D and 2D models.


The equation f ( ) = 0 describes the yield surface in 6D stress space. This yield sur face is recognized as nonhardening because it is strictly a function of stress it does not change size or shape with time.* For continued yield, the stress state must remain on the yield surface. In other words, f =0 for continued yield. Applying the chain rule gives Eq. 3.9. Namely, f =
3 3

i = 1j = 1

f --------- ij = ij

Bij ij
i = 1j = 1

= 0.

(B.34)

The geometrical interpretation of Eq. 3.9 is that the stress rate must be tangent to the yield surface. Since the gradient B of the yield function is normal to the yield surface, to B . This restriction follows from the nonharden the stress rate must be perpendicular ing assumption that the yield function depends only on the stress and not on any other internal state variables that could cause the yield surface to expand or contract. When internal variables are included, the stress rate has a normal component that makes the stress exactly keep up with the expanding or contracting surface. This effect is discussed separately on page 63 of the main text. Equation 3.10 in the main text expresses the typical decomposition of the strain rate into recoverable elastic part e plus a permanent plastic part p . Heuristically, the plastic permanent strain that would remain if all stresses were strain may be regarded as the released. Equation 3.11 of the main text states that the magnitude of the plastic strain rate p is not known, but its direction M is presumed known based on other physical arguments. Importantly, this direction depends only on the material state. In other words, all rate dependence (if any) is reflected through the magnitude of the plastic strain rate. The direction of the plastic strain rate is presumed rate independent. Whenever the plastic strain rate is normal to the yield surface (i.e., if M =B for some positive scalar ), then the flow rule is said to be associative. Equation 3.12 states that the stress rate is linearly related to the elastic part of the strain rate. The component form of Eq. 3.12 is ij =
3 3

Eijkl kl
k = 1l = 1

(B.35)

If the material is isotropic, this equation is merely the elastic expression of Hookes law of Eq. A.3 in rate form.

* Hardening and softening yield surfaces that do vary in time are discussed on page 63.

B-15
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

AFT DR
ann ca Br Rebec on

July 5, 2004 11:48 am

Importance of the yield function sign convention. If a material is plastically isotropic, its yield function satisfies the condition that f ( Q T Q ) = f ( ) for all orthogonal tensors Q . This implies that the yield function can be writtenas a function of invariants of stress. One of the simplest isotropic yield functions is that of Tresca which hypothesizes that yield commences when the largest shear stress reaches a critical value, k . If max and min are respectively the largest and smallest principal values of stress, then the largest -shear is max = 1- ( max min ) and occurs on a plane whose normal bisects two associ2 ated principal directions of stress. If the three principal stresses are known but have not yet been ordered, the Tresca criterion can be applied by simply considering the largest (in absolute value) of the three possible stress differences. In other words, the Tresca yield function may be written f ( ) = max ( ( 1 2 ) 2, ( 2 3 ) 2, ( 3 1 ) 2 ) 4k 2 (B.36) where { 1, 2, 3 } are the principal stresses. This is a properly defined yield function because it satisfies the essential sign convention properties: f=0 if and only if the stress is on the yield surface
(B.37a)

f < 0 if and only if the stress is inside the yield surface. (B.37b) The choice of the yield function is not unique. Many other functions can be constructed such that the above two conditions hold. Unfortunately, however, many authors* will wrongly claim that an alternative form for the Tresca yield function is f ( ) = [ ( 1 2 ) 2 4k 2 ] [ ( 2 3 ) 2 4k 2 ] [ ( 3 1 ) 2 4k 2 ] (B.38) It is straightforward to verify that this function may be written in terms of the standard -invariants, J 2 = 1- S and J 3 = detS , of the stress deviator S : 2 3 2 2 f ( ) = 4J 2 27J 3 36k 2 J 2 + 96k 4 J 2 64k 6 (B.39) Thus, the function f has an intoxicating appeal because its value may be computed without an eigenvalue analysis of the stress. The only problem is that f is not a valid yield function! The properties of Eq. B.37 are satisfied in only one direction, not both. The function f has properties If a stress is on the yield surface, then f =0 . If a stress is within the yield surface, then f < 0 .
(B.40a) (B.40b)

* and even respected plasticity texts [28].

B-16
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

July 5, 2004 11:48 am

Rebec

DRAF
ca Br ann on

However, the converses of both statements are false and therefore the sign requirements of Eqs. B.37 do not hold! The proof by counterexample is trivial. Consider 1 = 2 = 3k and 3 = 0 . Applying the valid yield function of Eq. B.36 gives f > 0 , indicating that this stress state lies outside the yield surface. Applying the invalid yield function of Eq. B.38 gives f = 20k 2 , which (being negative) falsely indicates that this stress state is inside the yield surface. Conclusion: the function f is inadmissible as the sole means of determining whether a stress state is below yield. The situation is not too hard to rectify. The basic idea is to first check whether the stress state lies near the yield surface by checking whether it lies within the inscribed and circumscribed Von Mises yield surfaces that bound the Tresca hexagon. Specifically, If J 2 < k 2 , then the stress is inside the yield surface, else if J 2 > 4k 2 3 , then the stress is outside the yield surface, else check the sign of Eq. B.39. This corresponds to the following yield function J 2 4k 2 3 if J 2 > 4k 2 3
3 2 2 4J 2 27J 3 36k 2 J 2 + 96k 4 J 2 64k 6 otherwise

(B.41)

f() =

(B.42a) (B.42b)

Another way to properly express the Tresca criterion is to use Lode coordinates. In terms of principal stresses, the Tresca yield function may be written f ( H, M, L ) = ( H L ) 2 4k 2 . The high and low stress eigenvalues, H and L , may be expressed analytically in terms of the Lode coordinates. Lode coordinates may themselves be easily expressed in terms of mechanics invariants ( I 1, J 2, J 3 ) , allowing the gradient of the yield function to be easily computed through chain rule differentiation.

Elastic potential
Quite often, the stress is presumed derivable from an elastic potential W = W ( e ) such that W ( p ) dW = ------------------------- = ------ d e The elastic stiffness is defined d d2W 2W E = ------- = ---------------- , or E ijkl = -----------------e e e d d e d e ij kl If the stiffness is constant, the elastic potential function is expressible as
-W = 1 e : E : e 2

(B.43)

(B.44)

(B.45)

B-17
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

AFT DR
ann ca Br Rebec on

July 5, 2004 11:48 am

ISV potentials
Hardening or softening yield functions are of the form f ( , 1, 2, ) . Simo and Hughes [50] define the flow rule to be associative if the plastic strain rate is normal to the yield surface in stress space f (B.46) where B ----- Typically, the plastic strain rate direction is presumed known at any moment in time (i.e., only its magnitude is unknown). Recalling the simplified 2D friction-spring model of plasticity, the direction of the plastic strain rate was always normal to the yield surface, but we will allow the possibility that some other direction might be desired. Specifically, we presume that p = B , (B.47) p = M , where M is a unit tensor ( M :M = 1 ) that can presumably be computed at any moment in time through knowledge ofonly the material state (not the rate of change of state). Stress space associativity is equivalent to assuming that B M = --------- and therefore = B (B.48) B Simo and Hughes extend the idea of associativity to include the following assumption about the evolution of the internal state variables. Specifically, we will say that the model is fully associative if there exist coefficients D kj (called plastic moduli) such that ff k = D kj ------- = d kj ------ j j
j=1 j=1 N isv N isv

(B.49)

Here the d kj coefficients are just an alternative set of plastic moduli defined in a manner compatible with . Namely, D kj d kj = --------B Equation B.49 can be written f -------- = D kj1 j k
j=1 N isv

(B.50)

(B.51)

It is natural to define conjugate internal state variables A k by


N isv A k = D kj1 j j=1

(B.52)

B-18
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

July 5, 2004 11:48 am

Rebec

DRAF
ca Br ann on

Thus, if the plastic moduli are constant, then Eq. B.51 can be written A k = k fwhere k = ------ j
(B.53)

Alternatively, using and d kj , we note that Eq. B.49 can be written f -------- = d kj1 j k
j=1 N isv

(B.54)

It is natural to define conjugate internal state variables a k by Ak a k = d kj1 j = --------B j=1 Thus, if the plastic moduli are constant, then Eq. B.51 can be written f a k = k , where k = ------ j
(B.56)
N isv

(B.55)

Note that this result is analogous to Eq. B.46. Simo and Hughes therefore postulate the existence of a plastic potential H such that H k = -------A k If the plastic moduli are constant, then
N isv N isv

(B.57)

H =

1 -2

Ak Dkj Aj
k = 1j = 1

(B.58)

Example: porosity. For a plastically incompressible matrix material, we know that the unstressed porosity must evolve according to = ( 1 )tr p (B.59) Alternatively, the porosity evolution equation may be written in terms of the commonlyused distention defined by = 1 ( 1 ) . Then the porosity evolution equation is = tr p We can alternatively define ln Then the evolution equation for this alternative porosity measure is = tr p B-19
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

(B.60)

(B.61)

(B.62)

AFT DR
ann ca Br Rebec on

July 5, 2004 11:48 am

All three porosity measures ( , , and ) characterize the same internal state variable. Knowing one gives values for the other. To analyze all three choices (as well as other unstated choices) for the porosity variable, let denote some unique measure of porosity. We will assume that there exists a function = ( )
(B.63)

that relates the porosity measure to the porosity measure . Then Eq. B.62 implies that is governed by d = ------ = ( )tr p , where = d ( ) -------------d d For example, if = , then ( ) = 1 e and = e = 1 if = , then ( ) = e and = e = if = , then ( ) = and = 1
(B.64)

(B.65)

The porosity may be regarded as an internal state variable. Now we are going to explore Simo and Hughes idea of the plastic modulus for our generalized porosity measure . An application will use one and only one measure of porosity. Our purpose here is to illustrate how the choice of porosity measure can have an impact on the simplicity of the calculations. Recall that the plastic strain rate may be written as f p = ----- Then the evolution equation becomes
f = tr -----

(B.66)

(B.67)

For this special case of a single internal state variable, the summation in Eq. B.49 now ranges over only one term. In other words, there is only one k , so the D kj matrix is just a 1 1 matrix, making it simply a single scalar D . Depending on the single choice for the porosity measure, Eq. B.49 becomes
f = D -----

(B.68)

Equating the last two equation gives


f f D ------ = tr -----

(B.69)

Using subscripts to explicitly show what is being held constant, this equation may be written
f f D ------ = tr ------

(B.70)

B-20
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

July 5, 2004 11:48 am

Rebec

DRAF
ca Br ann on

Since is the only internal state variable, we know that the yield function is of the form f = f ( , ) . Without loss in generality, we may alternatively assert that f = f ( S, p, ) , where S is the stress deviator and p is the pressure. Then, by the chain rule,*
f f tr ------ = ----- p S,

(B.71)

Now, observe that


f f ------ = ------ S, p

(B.72)

By a well-known identity (cyclic rule) from multivariable calculus, we note that


f ------ S, p p ------------------- = ------ f, S f ----- p S,

(B.73)

Substituting Eq. B.71 into Eq. B.70 and using the identity B.73 gives
p D ------ = f, S

(B.74)

When solving this partial differential equation, it is important to realize that any integration constants that appear will actually be functions of the constant stress deviator S . Sup pose that crush curve experiments are available that provide the porosity as a function of the applied pressure, for various (constant) values of the stress deviator. Such curves are often called p- curves. By the substitution = e , these curves can be readily converted into p- curves. An experimentally measured p = p ( , S ) function applies during crush (i.e., when f=0 ). Therefore, we can define p (B.75) g ( , S ) ----- S Thus, the function g = g ( , S ) may be regarded as a known property of the material. We have used the negative sign because a decrease in porosity normally results from an increase in pressure (though the opposite is true in tension). By the chain rule, we have p g ------ = p ------ = -------- d S S d So Eq. B.74 becomes
gD ---- =

(B.76)

(B.77)

or
* Incidentally, this operation is a good example of a danger with indicial notation. Specifically, tr ( f ) is not the same thing as f ( tr ) , so it should not be written in indicial notation as tr ( f kk ) because such an expression is ambiguous.

B-21
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

AFT DR
ann ca Br Rebec on

July 5, 2004 11:48 am

= Dg Showing the independent variables, ( ) =

(B.78)

Dg ( , S ) (B.79) For this to hold, the plastic modulus D must generally depend on the stress deviator. If the function g ( , S ) is approximated to be independent of the stress deviator, then the above equation may be integrated to obtain the natural measure of porosity for which the plastic modulus is constant. Alternatively, if is assumed to be the natural porosity measure, then D is constant and Eq. B.79 implies the crush curve. Namely
2 g ( , S ) = [ ( ) ] -------------------D

(B.80)

or 1 p p o = --- [ ( ) ] 2 d D
o

(B.81)

If, for example, itself is presumed to be the natural porosity measure, then 1 p p o = --- ( o ) D or, writing = ln , the crush curve becomes ----- = e D ( p p o ) o The integration constants o and p o depend on the shear stress.
(B.82)

(B.83)

B-22
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

July 5, 2004 11:48 am

Rebec

DRAF
ca Br ann on

Plastic dissipation
Simo and Hughes define the plastic dissipation to be
N isv

Dp = : p +

k Ak
k=1

(B.84)

where A k are the conjugate strains associated with the internal state variables; namely
N isv A k = D kj1 j j=1

(B.85)

Example: porosity. Consider a single porosity measure used in the previous example. Suppose that D is constant. Then where A is the conjugate strains associated with the internal state variables; namely A = --D Using Eq. B.79 gives

Dp = : p + A

(B.86)

(B.87)

= Therefore

Dg ( , S ) d

(B.88)

A = And

g ( , S ) ------------------- d D

(B.89)

g ( , S ) A = --- = ---------------D D B-23

(B.90)

Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

AFT DR
ann ca Br Rebec on

July 5, 2004 11:48 am

B-24
Copyright is reserved. Individual copies may be made for personal use. No part of this document may be reproduced without written permission from the author, Rebecca Brannon (rmbrann@me.unm.edu).

You might also like