Computers and Structures 74 (2000) 601±628 www.elsevier.com/locate/compstruc
Robust integration schemes for generalized viscoplasticity with internal-state variables
A.F. Saleeb*, T.E. Wilt, W. Li
Department of Civil Engineering, The University of Akron, Akron, OH 44325, USA Received 11 July 1998; accepted 31 August 1998
Abstract
This paper is concerned with the development of a general framework for the implicit time-stepping integrators for the ¯ow and evolution equations in complex viscoplastic models. The primary goal is to present a complete theoretical formulation, and to address in detail the algorithmic and numerical analysis aspects involved in its ®nite element implementation, as well as to critically assess the numerical performance of the developed schemes in a comprehensive set of test cases. On the theoretical side, we use the unconditionally stable, backward Euler dierence scheme. Its mathematical structure is of sucient generality to allow a uni®ed treatment of dierent classes of viscoplastic models with internal variables. Two speci®c models of this type, which are representatives of the present state-of-the-art in metal viscoplasticity, are considered in the applications reported here; i.e., generalized viscoplasticity with complete potential structure fully associative (GVIPS) and non-associative (NAV) models. The matrix forms developed for both these models are directly applicable for both initially isotropic and anisotropic materials, in three-dimensions as well as subspace applications (i.e., plane stress/strain, axisymmetric, generalized plane stress in shells). On the computational side, issues related to eciency and robustness are emphasized in developing the (local) iterative algorithm. In particular, closed-form expressions for residual vectors and (consistent) material tangent stiness arrays are given explicitly for the GVIPS model, with the maximum matrix sizes `optimized' to depend only on the number of independent stress components (but independent of the number of viscoplastic internal state parameters). Signi®cant robustness of the local iterative solution is provided by complementing the basic Newton±Raphson scheme with a local line search strategy for convergence. # 2000 Elsevier Science Ltd. All rights reserved.
Keywords: Viscoplasticity; Numerical integration; Implicit; Finite element; Line search; Anisotropic; Isotropic; Algorithm; Multiax- ial; Tangent stiness
1. Introduction performance of inelastic structural components operat- ing under complex thermomechanical and multiaxial At present, it is widely accepted that large-scale nu- loading conditions. This is largely due to the signi®cant merical simulations, based on the ®nite element progress made over the years in the development of method, provide a viable approach for assessing the theories of plasticity and viscoplasticity for the phe- nomenological representations of inelastic constitutive properties. In particular, the present state-of-the-art in * Corresponding author. Tel.: +1-330-972-7286; fax: +1- mathematical modeling of metal viscoplasticity is very 330-972-6020. well developed, i.e., based on the so-called internal
0045-7949/00/$ - see front matter # 2000 Elsevier Science Ltd. All rights reserved. PII: S0045-7949(99)00020-6 602 A.F. Saleeb et al. / Computers and Structures 74 (2000) 601±628 variable formalism in the thermodynamics of irrevers- Motivated by these practical considerations, the ible processes [1±6]. A large number of specialized development of stress integration algorithms with forms of these modern uni®ed viscoplastic models improved performance has received considerable atten- (e.g., isotropic or anisotropic, fully associative or non- tion in the recent literature on computational plasticity associative, isothermal or nonisothermal, etc.) have and viscoplasticity. In this, implicit (e.g. backward been successfully applied to dierent metals and com- Euler, generalized midpoint and trapezoidal rules) posite material systems [4±12]. In these, account is methods of time integration have been particularly made for a host of complicated physical phenomena emphasized, mainly because of their superior stability, and deformation mechanisms; e.g., load-rate depen- and wider range of applicability, compared to the ear- dency, creep-plasticity interactions, stress relaxation, lier explicit counterparts. More speci®cally, a number nonlinear kinematic and isotropic hardening, of well-established implicit schemes are presently avail- Bauschinger and allied reversed-load eects, ratchetting able for inviscid plasticity (e.g. classical radial-return under cyclic loading, thermal and dynamic recovery and its generalizations in the form of operator-split/ mechanisms, material damage, etc.. Naturally, the bet- return-mapping algorithms) and viscoplasticity of ter accuracy and improved modeling capabilities in metals; see Refs. [13±27] for a sampling of these these representations have been often acquired at the works. We also refer to Ref. [25] for a recent survey expense of greater mathematical complexities, larger and summary of several proposals. In this regard, we number of internal variables and increased compu- note the very ecient implementations of the radial- tational demands. return schemes for metals corresponding to the special From the computational standpoint, the global sol- case of full isotropy (both elastic and inelastic re- utions for the discretized (e.g. standard displacement- sponses), enabled by the complete reductions of the based or stiness) ®nite element model are typically rate tensor equations to a ®nal single/scalar nonlinear achieved by an incremental/iterative (e.g. Newton or equation; e.g. in terms of the eective stress, or equiva- quasi-Newton) procedure. A common and important lently, the plastic multiplier or the `magnitude' of the problem in all these nonlinear material analyses con- inelastic strain vector. In particular, details of the cerns the temporal integration, in the standard strain- stress-updating in these algorithms have been worked driven formats, over a discrete sequence of time steps, out for inviscid plasticity, combined plasticity with of the rate evolutionary equations (i.e., sets of ®rst- time-/or strain-hardening creep, as well as classical vis- order dierential equations that are highly nonlinear, coplasticity with no (static/dynamic) recovery e.g. coupled and mathematically sti) in the underlying [13,14,17,26,27]. Furthermore, the importance of using material constitutive models. For a given integration the consistent tangent stinesses in these cases has scheme, the result is a set of local nonlinear response been also clearly demonstrated [13,14,16,21,25,]. functions which de®ne the stress tensor (and other Finally, we also note that in contrast to their time- state variables) as a function of the strain history up to independent elastoplastic counterparts, applications of the current time step, at each material/integration the implicit integration to viscoplasticity with nonlinear point. This local stress update constitutes the main kinematic hardening and (static/dynamic) recovery subject of the present paper. mechanisms have been rather limited, especially for The mathematical forms, accuracy and stability general classes of anisotropic materials. Models of this properties, as well as, implementation details, for the latter type are the focus in the present study. selected integrator, will directly impact both the accu- racy and eciency of the overall ®nite element sol- ution. For instance, concerning accuracy we note that 2. Objectives and outline the global (driving) residual force vectors are directly formed from the updated local stress ®elds. On the In this paper, we are concerned with the develop- other hand, the linearization of these latter stresses will ment of a general framework for the implicit time-step- also provide the material tangent stiness (iteration ping integrators for the ¯ow and evolution equations Jacobian matrix) used to form and subsequently in generalized viscoplastic models. The primary goal is assemble element stinesses that govern the conver- to present a complete theoretical formulation, and to gence rate of the global solution; e.g. the well known address in detail the algorithmic and numerical analy- property of the (asymptotic) quadratic rate of conver- sis aspects involved in its ®nite element implemen- gence for the Newton scheme can be preserved only if tations, as well as to critically assess the numerical a consistent (algorithmic) material stiness is utilized performance of the developed schemes in a comprehen- [13±15]. Several other important factors here include sive set of test cases. time- step controls for conditional stability of explicit On the theoretical side, the general framework is integration, or the design of appropriate local iterative developed in the speci®c context of the fully implicit schemes ensuring convergence in implicit algorithms. (backward Euler) dierence scheme. In addition to its A.F. Saleeb et al. / Computers and Structures 74 (2000) 601±628 603 desirable feature of unconditional stability, this scheme the evolution of each following the Armstrong± is also known to exhibit excellent convergence/ accu- Frederick rule [4,5], i.e., the continuous `(linear) hard- racy characteristics in large scale computations that ening minus dynamic recovery' format. Dierent, more typically involve large time steps, especially with com- elaborate, functional forms are also often utilized in plex anisotropic inelastic models [17,19,25]. the ¯ow equations, e.g., exponential or combined Furthermore, compared to other members of its class; hyperbolic sine-power type [5,12]. e.g., semi-implicit, generalized midpoint integration The majority of viscoplastic models currently in use rules, the fully implicit scheme leads to very ecient belong to the NAV class [4,5,12]. Several developments and direct implementation, i.e., entirely bypassing com- in the GVIPS class are currently underway [7±9] as plications associated with the so- called `continuation these models are very appealing in view of their theor- of the time step' calculations [15,17,21,24,]. The latter etically sound basis from the thermodynamics stand- are required for the semi-implicit methods to advance point, and the symmetries exhibited in the the solution from the intermediate (generalized mid- corresponding linearization of the integrated forms of point) time station (where all the viscoplastic rate their ¯ow and evolution equations. Both types of equations are evaluated) to the end of the step. For models have been actually characterized fro real example, in order to preserve the stability and conver- metals, and these are used in the applications reported gence properties of the method, this will necessitate the here. Speci®cally, as representative of the NAV class, enforcement of global balance (equilibrium) conditions we utilize the isotropic form given by Freed and (and correspondingly evaluating the ®nite element re- Walker in Ref. [12]. For GVIPS, we use both isotropic sidual vectors) at the generalized mid-time step [15]. and transversely isotropic forms [6,7,25], with This problem is even more critical in the context of decoupled Gibb's function, in the numerical simu- `nonclassical' implementation of the above semi-im- lations. plicit midpoint rules; e.g. in the form of approximately The rather complex nature of the evolution discretized, asymptotic integration method [24] where a equations in the above models provides for an ideal nonlinear completion is employed. setting to test any stress integration algorithm. This On the constitutive modeling side, the generality of will become evident in several comparisons reported the present framework allows a uni®ed treatment of latter in the Section 5. dierent classes of viscoplastic models with internal In summary, for each of the viscoplastic models con- variables. To this end, two speci®c models which are sidered, and the associated implicit backward Euler in- representatives of the present state-of-the-art in metal tegration method, the following constitutes the major viscoplasticity (of the generalized J2 or von Mises tasks required to achieve the above stated primary type) are considered here. For brevity, these will be goals: subsequently referred to as GVIPS models (for 1. detailed study of the underlying mathematical struc- Generalized Viscoplasticity with Potential Structure) ture of the viscoplastic equations, and the corre- and NAV models (for Non-Associative sponding integrated ®elds of stress and internal state Viscoplasticity). Each of these models include a num- variables. ber of distinctive features, and collectively they span a 2. development and implementation of the implicit, very wide spectrum of characteristic and predictive backward-Euler stress- updating algorithm, and the capabilities that are shared by many other viscoplastic associated (Newton type) nonlinear/iterative models. equation solver, and its further enhancements (e.g., For example, the GVIPS class [6±9] is unique in its in the form of line searches and time step subincre- utilization of a nonlinear kinematic hardening (back menting) stress) mechanism resulting from an assumed thermo- 3. testing of the convergence, stability, and accuracy dynamic potential (Gibb's) function for the `equation- properties of the algorithms. of-state' de®ning the associated (conjugate/dual) in- ternal strain variable, and dissipation function O- Our primary objective in item (1) is to identify the per- form), including thermal (static) recovery contribution, tinent matrix operators needed in the algorithmic for its rate of evolution. A characteristic partitioning, development; i.e., concise forms of the constitutive re- with several regions of discontinuities of the state sidual vectors for the integrated ®eld equations. The (stress-internal variable) space is also used to account implementation in item (2) utilizes closed-form ex- for stress-reversal/cyclic loading. For the ¯ow pressions of the consistently-derived material tangent equations, simple power functions are typically stiness; i.e. model/integrator-dependent. Here a cru- employed. cial consideration, for computational eciency, con- On the other hand, the NAV models are typically cerns a systematic elimination procedure to reduce to a characterized by the use of several internal state vari- minimum the number of unknowns governing the local ables, particularly for the tensorial back stress, with iterations. To this end, with the stress tensor as the pri- 604 A.F. Saleeb et al. / Computers and Structures 74 (2000) 601±628 mary unknown, analytical reduction was utilized lead- present terminology (e.g. phrases such as `fully associ- ing to `optimized' sizes of all the arrays involved with ative', or the equivalent `complete potential structure') maximum sizes equalling the number of independent to become more meaningful in the present context of stress components i.e., six for the full three-dimen- model classi®cation, we make the following remarks at sional (3D) space, or less in subspaces (e.g. four in the outset. plane-strain/axisymmetric; three for plane stress; and Following the recent work in Ref. [6±9], in the gen- ®ve for plate/shell problems). This is in contrast to eral form for fully associative GVIPS, there are two straightforward `standard' implementation in which key ingredients; (1) the thermodynamic potential, or the unknown solution vector includes stress plus all Gibb's function (or conjugate free energy, Helmholtz tensorial and scalar internal variables. Furthermore, function) de®ning the equations of state, and (2) the the line search enhancement of the basic Newton iter- viscoplastic dissipation function O-form) for the ensu- ation will signi®cantly add to the robustness of the ing ¯ow and evolution laws of inelastic strain and in- overall integration algorithm. ternal state parameters. However, this is dierent from In conjunction with (3), an extensive number of dis- usage by other researchers [10±12], of such notions as criminating test problems will be utilized. The objective the general structure of the `thermodynamically based' is to enable ®rm conclusions to be made regarding the constitutive equations where only the O-form (¯ow or `relative' merits and/or limitations of various schemes. Finally, an outline of the remainder of the paper is dissipation potential) was utilized. These latter forms given as follows. In Section 3, we discuss the math- are introduced only for convenience in satisfying the ematical structure underlying both GVIPS and NAV thermodynamic admissibility restriction (based on classes of viscoplastic models, pointing to several poss- simple properties of non- negativeness and convexity ible extensions and generalizations, particularly per- of these functions) in the form of the well-known taining to the use of a number of internal state Clausius±Duhem or dissipative inequality O-form) [2]. parameters and the eect of initial material anisotro- However, such nonlinear kinematic hardening forms pies for both elastic and viscoplastic response com- do not presuppose or automatically imply the existence ponents. We also outline the practically- used forms of the total (integrated) forms of the associated ther- and speci®c expressions in the ¯ow/evolution equations modynamic potentials, e.g. the Helmholtz free energy, for each. or Gibb's function. For distinction, i.e., when using The details of the implicit integrator are given in only the O-forms, the term `incomplete' potential form Section 4. These include the general discrete forms of seems more appropriate. Since the opposite situation the integrated equations and the local Newton method, typically occurs in NAV models (i.e., implicit or expli- together with speci®c forms for the constitutive re- cit assumption for the existence of linear equation of siduals and consistent (algorithmic) material stiness. state for the conjugate internal variables, but no O- The outline of the line search employed will also be form), we prefer the description nonassociative for given here. In Section 5, we give the results of a large both of these latter formulations. number of numerical simulations using both GVIPS The complete potential-based class of inelastic con- and NAV models. These range from simple uniaxial stitutive equations possess a number of distinct and validation tests under varying conditions (creep, relax- important attributes from both a theoretical and a ation, cyclic, etc.) to multiaxial testing under nonpro- computational standpoint. First, they constitute the portional loading, as well as a structural problem cornerstone of numerous regularity properties and involving nonhomogeneous states of stresses and de- bounding (limit) theorems in plasticity and viscoplasti- formations. The emphasis is placed on evaluating con- city. Second, they result in a suciently general vari- vergence properties, robustness, and computation ational structure, whose properties can be exploited to eciency. For comparison, we also show some of the results obtained using other, explicit, integration derive a number of useful material conservation laws. methods. Third, on the numerical side, the discrete form of this same variational structure is of great advantage in the development of ecient algorithms for ®nite element 3. Mathematical formulations of viscoplastic models implementation, for example, symmetry- preserving material tangent stiness operators are easily obtained 3.1. General in implicit solution schemes (such as those given later). Finally, this complete potential-based framework con- In this section, we describe the mathematical struc- veniently lends itself to intelligent application of sym- ture and basic governing equations, corresponding to bolic manipulation systems that facilitate the the two viscoplastic models considered; i.e., GVIPS construction, implementation, and analysis of new de- and NAV. However, for clarity, and in order for the formation and damage models [6]. A.F. Saleeb et al. / Computers and Structures 74 (2000) 601±628 605
@C 3.2. Generalized viscoplasticity with potential structure gd 6 D d (GVIPS) @D i where eij is the total strain and eij is the inelastic strain The scope of the following discussions is limited to tensor, respectively. cases of small deformations and isothermal conditions, The above relations are known as the equations of in which the initial state is assumed to be stress-free b d state, in which, sij,aT,D , are the `force-like' thermo- throughout. For generally, the initial state of material b d dynamic state variables while eij,gT,gD, are the dual/ isotropy or anisotropy is left unspeci®ed here; i.e., conjugate `displacement-like' variables (anities). except for the explicit appearance of the corresponding Now assume the existence of a dissipation potential material-directionally tensors in the arguments list for of the form, the potential functions employed, all derivations remain generally applicable. O O s ,ab ,Dd 7 Here, and for the remainder of the paper, a ij T Cartesian reference frame is utilized, and both matrix which, when written in its decoupled form [6], including and index notations (with summation implied on `¯ow' as well as `static (thermal) recovery' terms, gives, repeated subscripts) are used interchangeably, with the same letters indicating tensor components of variables, Xnb Xnd b d b b d d and their vector matrix `counterparts' (e.g. using the O O1 sij,aT,D O2 aT O3 D 8 familiar contracted Voigt representations in vector b1 d1 forms for components of the corresponding symmetric, Adopting `standard' thermodynamic arguments [1± second-order, tensors. 3,6], and the de®nition of the inelastic strain rate e_i , Consider the following general form for the Gibb's ij one can easily derive the Clausius±Duhem or dissipa- potential: tion inequality. That is, C C s ,ab ,Dd 1 b d ij T O sij,aT,D
n n where ab denotes the Tth-order tensorial internal state Xb Xd T s e_i ab :g_b Ddg_d r0 9 b ijij T T D variables (where b 1, ...,nb; T 1, ...,nT), aT can be b1 d1 a second-order tensor (back stress) or higher order tenors, b denotes the number of internal state vari- where `:' indicate scalar multiplications (complete d b b ables; D is the scalar internal state variable (where contraction) of the (identical-order) tensors aT and g_T, d 1, ...,nd and sij the Cauchy stress tensor. This is a and the overdot _ signi®es a rate or time derivative. typical general form for the Gibb's potential, in which The rate of change of the anities may then be as many internal variables as desired may be speci®ed expressed in terms of their corresponding internal state in Eq. (1). The potential may be decoupled into elastic variables. This results in the ¯ow or kinetic laws and inelastic parts; @O e_i 10 ij @s e i b d ij C C sij C aT,D 2 @O or, upon further decoupling of inelastic mechanisms for g_b 11 T @ab practically used forms, T
n n Xb Xd @O C Ce s Ci ab Ci Dd 3 g_d 12 ij a T D D @Dd b1 d1 From Eqs. (3) and (6), It then follows from Eq. (3) that: d @C @ 2C g_b Á a_b 13 @C T b b b T i dt @aT @aT@aT eij eij 4 @sij d @C @ 2C g_d D_ d 14 D d d d @C dt @D @D @D gb 5 T b @aT where `Á' in Eq. (13) signi®es a `T-times' contraction 606 A.F. Saleeb et al. / Computers and Structures 74 (2000) 601±628
e (i.e., summation over all tensor indices in the Tth- sor C ijkl can be derived in the following convenient form; b order variable aT see Refs. [25,28] for details " # " # 1 1 Ce 2G P aR 2 G G Q @ 2C @ 2C @O t l t a_b Ág_b Á 15 T b b T b b b 20 @aT@aT @aT@aT @aT a 2 l G d d m V V 3 3 t l 2 1 2 1 d @ C @ C @O where the symbol indicates tensor product, and D_ g_d 16 @Dd@Dd D @Dd@Dd @Dd m1 3a 4 Gl Gt b 21 Eqs. (11)±(14) which de®ne the evolution of the in- ternal state. Together, the equations of the state, Eqs. The a,b,l,Gl and Gt are the ®ve generalized Lames (4)±(6) and evolutionary laws, Eqs. (10), (15) and (16), constants with the subscripts `l' and `t' identifying the constitute the ®nal general forms of GVIPS. longitudinal and transverse directions, respectively, From Eqs. (15) and (16), the (generally-nonlinear) with respect to the ®ber direction (normal to plane of compliance operators (generalized hardening moduli) isotropy) for the shear moduli, Gl and Gt: The fourth- are de®ned as follows: order symmetric tensors P,Q and R are the same as those de®ned later in Section 3.3.2 (see Eq. (29)). @ 2C @ 2C Other, more general forms of anisotropy can be writ- Lab , LDd 17 T b b d d @aT@aT @D @D ten similarly. which relate the rates of `force-like' state variables b _ d b d a_T,D to the `displacement-like' variables g_T,g_D), and provide a link between the assumed Gibb's and comp- lementary dissipation potential.
3.3.2. The viscoplastic response 3.3. Speci®c form for the GVIPS model For the viscoplastic response, we take T 2,nb 1,nd 0, so that only one second-order in- In this section, a speci®c form of the GVIPS model ternal variable, in this case the back (internal) stress, is is outlined. For completeness a transversely-isotropic speci®ed. We write the plastic dissipation function, O, elasto-viscoplastic material, i.e., a solid reinforced by a in terms of the `anisotropic' invariants of the stress single family of ®bers as in Refs. [10,11,25,26], is tensor, sij, the internal variable, aij, including separate assumed. ¯ow/hardening and static-recovery contributions, O1 and O2, respectively, 3.3.1. Elastic response Á Á Á The hyperelastic response is assumed to be linear. O O sij,aij,Vij O1 sij,sij,Vij O2 aij,Vij 22 That is, a quadratic strain±energy±density function, U e, is assumed to exist such that the Cauchy (true) O O1 F O2 G 23 stress components sij are given by @U e e e @O @O1 sij C e 18 _i @ee ijkl kl eij 24 ij @sij @sij where a superscript `e' denotes the elastic component, and the following expression, @O g_ij 25 @aij e i eij eij eij 19 where the parentheses are used to indicate functional holds true in accordance with the basic hypothesis of dependence on the arguments inside ( ); and the ma- the additive decomposition of total strain tensor, eij, terial directionality tensor Vij is de®ned as Vij vivj, e i into its elastic, eij, and inelastic components, eij: In the where v is the unit vector de®ning the material ®ber e above, C ijkl denotes the fourth-order symmetric elastic direction. moduli tensor. The F and G are the quadratic functions in terms of the invariants of the deviatoric components of the Remark 1. In terms of its ®ve independent elastic eective stress, sij aij , and the back stress, aij, ten- moduli, the present transversely-isotropic elasticity ten- sors, respectively. A.F. Saleeb et al. / Computers and Structures 74 (2000) 601±628 607
1 F s a M s a 1 where H and Kt (stress units), and b (dimensionless) 2 ij ij ijkl kl kl 2K t are (positive) material constants. The function G G aij was de®ned previously in Eq. (26). Together 1 with the strain vector decomposition of Eq. (19), the G a M a 26 2 ij ijkl kl above Eqs. (30) and (31) de®ne the equations of state, 2K t e @C Á 1 Mijkl is a symmetric fourth-order tensor which is a e ei C e s , ij ij @s ijkl kl function of the ®ber orientation tensor Vij (material- ij directionality) and is de®ned as, 32 @Ci 1 g p 1 ij @a h ij M P xQ zR 27 ij ijkl ijkl ijkl 2 ijkl in which h and pij are de®ned as, In the above, 0RxR1 and 0RzR1 are material- H strength ratios de®ned as, h , p M a 33 G b ij ijkl kl Z2 1 4 o2 1 x , z De®ning the free energy C ee,g through ordinary Z2 4o2 1 Legener-type transformation (e.g. 2), i.e., C C e C i, where C e Ce for linear elasticity here, K Y and C i a Á g Ci a the dissipation function O s ,s Z l , o l 28 ij ij Kt Yt is now introduced as follows, where the constants Kt (and Kl and Yt (and Yl are _ O sije_ij Cr0 34 threshold strengths of the composite material in trans- _ verse (and longitudinal) shear and tension/com- Substituting for C in terms of s_ ij and a_ij from Eqs. pression, respectively. The fourth-order tensors Pijkl, (30)±(32), one obtains, Qijkl and Rijkl are de®ned as, ! @O @O @ 2Ci 1 1 Á e_i , g_ a_ 35 P I d d , I d d d d @s ij @a ij @a @a kl ijkl ijkl 3 ij kl ijkl 2 ik jl il jk ij ij ij kl
The above rate form connects a_ij and g_ij through the 1 Á Q V d V d V d 2V V hardening-moduli tensor which is the Hessian matrix ijkl 2 ik jl il jk jl ik ij kl of Ci: Using Eq. (35), and the second of Eq. (31), yields Á 1 R 3V V V d V d d d 29 h 0 Â Ã ijkl ij kl ij kl kl ij 3 ij kl a_ h Z m a a g_ L 1g_ 36 ij ijkl h 1 2b ij kl kl ijkl kl
For the isotropic case, x z 0 and Mijkl reduces to where L is the present, single, compliance operator, Pijkl, i.e. simply the classical von Mises-type forms in ijkl 0 functions F and G. and h is a scaled derivative, For the case considered here, nb 1, nd 0, the pre- 1 dh sent purely mechanical theory, (Eq. (1)) has the follow- h 0 37 K 2 dG ing decoupled form of a Gibb's complementary free- t energy density function in terms of stress and internal m In addition, Z ijkl above is de®ned as the `generalized state parameters, i.e., with superscripts `e' and `i' indi- inverse' of the deviatoric, positive semi-de®nite, ma- cating elastic and inelastic parts, m 1 terial tensor operator Mijkl, i.e. Z ijkl Mijkl: We refer to Ref. [25] for details of dierent conditions (three- C s ,a Ce s Ci a 30 ij ij ij ij dimensional, plane-strain/axisymmetric, generalized plane stress, etc.). The terms Ce and Ci are de®ned to have the following forms, Remark 2. Relative to the conjugate (dual) variables ee and s of the elastic response, one de®nes the free-energy e 1 e Á 1 C sij sij C skl 2 ijkl and (complementary-energy) `natural' norms, through 31 the positive-de®nite material-dependent moduli tensor 2 i K t 1b Ce, and its inverse, respectively. Analogously, for the vis- C aij G 1 bH coplastic counterpart, we can de®ne the semi-norms for 608 A.F. Saleeb et al. / Computers and Structures 74 (2000) 601±628 stress, ksk, and (conjugate) inelastic strains, keik, Remark 3. The above Eq. (44) provides the simplest through the M and its generalized inverse Zm; i.e., form of internal loading±unloading criterion; i.e., with r the evolution for the internal strain variables, g_ij, during 1 p loading according to Eq. (40), but with a dierent dissi- ksk s Á M Á s, keik 2ei Á Z Á ei 38 2 m pation function during unloading giving a `magni®ed' u u 1 rate of evolution, g_ij Lijkl Lklmn g_mn, to account for These generalized `norms' will be used repeatedly in the `stier' response upon unloading, with the moduli the sequel, thus allowing a uni®ed, and very eective, Lu taken in the same form as for loading in Eq. (36) treatment of both GVIPS and NAV models (e.g. see klmn with G0 replacing G. However, alternative criterion are Section 3.4.2) also possible to handle these observed readjustment of in- ternal state [8], and may actually have direct impact on such cyclic-loading phenomena as ratchetting.
3.3.3. The viscoplastic response: ¯ow and evolution laws From Eqs. (31) and (35), the ®nal expressions for the inelastic ¯ow and evolution equations for the 3.4. Generalized nonassociative viscoplasticity (NAV) speci®c selection of power functional forms for O1 F and O G , Eq. (23) can be straightforwardly derived; 2 Considering the nonassociative models, their popu- i.e., larity in the literature is mainly due to the added mod- eling ¯exibility to account for experimental phenomena e_i f FG , G M s a 39 ij ij ij ijkl kl kl by using arbitrary rate forms in the ¯ow and evolution equations. In particular, the Freed±Walker viscoplastic l model [12] is used, as an example, to discuss the theor- g_ e_i p , p M a 40 ij ij h ij ij ijkl kl etical basis of the NAV class of models. This model was formulated to account for both  à l short and long-term material behavior, by admitting a_ L 1 e_i p 41 two tensorial internal variables (short and long term) ij ijkl kl h kl in the additive decomposition of the back stress, and where f,h, and l are the ¯ow, hardening, and recovery one scalar internal variable (a drag stress). As with all functions, respectively, de®ned as, `uni®ed' viscoplastic models, creep-plasticity interaction is also accounted for, but the model was shown to ana- F n H f F , h , l RG m b 42 lytically reduce to the limiting cases of pure creep the- 2m G b ory at high temperature (steady-state condition) and to plasticity theory at low temperature (rate-independent in which n,m,H,b,R, and m are (positive) material con- condition). The existence of these creep and plasticity stants with mrb: bounds was made use of in simplifying the process of Several regions of discontinuities in the state space model parameter characterization. were introduced to account for stress-reversal/cyclic- To date, characterizations and applications of this loading/dynamic-recovery eects. This simply amounts model have been limited to the case of isotropy. For to using dierent functional forms for f,h, and l in easy reference to the original presentation in Ref. [12], dierent regions; i.e., this will also be assumed here. Unlike the more general f f if F > 0 nonisothermal treatment in Ref. [12], we restrict our- selves, as before, to the isothermal case. Tests reported later may be given at dierent temperatures, but the f 0 otherwise 43 given temperature does not change during a test. For clarity, we ®rst summarize the original equations in the above reference before transforming into similar forms G G if G > G0 and dev sij aij pij > 0 (and notations) as those used for GVIPS.
G G0otherwise 44 3.4.1. Elastic behavior Recalling the above isotropy restriction, the linear The `small' constant (cut-o value) G is an `adjusted' 0 elastic response is decomposed into the deviatoric and parameter selected to ®t the experimental data, and hydrostatic components; i.e. prevents singularity in h when G40 and dev( ) is the deviatoric part of ( ). i i Sij 2m0 eij eij ekk 0 45 A.F. Saleeb et al. / Computers and Structures 74 (2000) 601±628 609 s 3k e 46 i kk 0 kk e_ij f J,DGij, Gij Mijkl skl akl 54 In the above we de®ne, the elastic shear, m , and bulk, 0 where f is a function of invariant J and the drag stress k0, moduli, as well as, D, i.e., 1 1 p Sij sij skkdij, sij eij ekkdij 47 1 J 3 3 f yA psinh 55 2 J D for the deviatoric stress and strain, respectively. In keeping with the previous framework presented in and J, analogous to the previous de®nitions in F, Eq. Section 3.3, the above equations of the elastic behavior (26), is a quadratic function in terms sij aij), i.e., can of course be generalized to other material-sym- metry groups, e.g. see Remark 1 above. 2 1 J ksij aijk sij aij Mijkl skl akl 56 2 3.4.2. Viscoplastic behavior where Mijkl is the fourth-order tensor as de®ned in Eq. The original inelastic ¯ow equation was postulated (27). The special case of isotropy assumed in the orig- in the form [12], inal model [12], is now trivially obtained by simply taking x z 0, and Mijkl reduces to Pijkl (see Eq. i i 1 sij aij e_ij ke_ knij, nij 48 (27)). The inelastic strain tensor modulus (magnitude) 2 ksij aijk can be de®ned analogously by using Zm (the general- in which, ized inverse of M as a metric in the de®nition of the second of Eq. (38); see also Remark 2 above. i ke_ kyZ if ksij aijk < K 49 The (deviatoric, tensor-valued) back stress, aij, accounts for kinematic (¯ow-induced, anisotropic) where, y > 0 is a thermal function, Zr0 is the so- hardening eects, and is taken to be a ®nite sum of the called Zener parameter and K > 0 is the yield stress. In s l s individual back stresses, i.e., aij aij aij, where aij addition, for isotropic viscoplasticity, the `convention- l and aij are the short-term and long-term back stresses, al' eective-stress von Mises norm ksij aijk, is de®ned respectively. Their evolutions are described by the lin- as, ear `hardening-minus-dynamic recovery' forms; r 1 as ks a k s a s a 50 s i ij i ij ij ij ij ij ij a_ij 2Hs e_ij ke_ k 2 2Ls Note that the norms pertaining to the deviatoric ten- sors of this section are de®ned as, ! al l i ij i r a_ij 2Hl e_ij ke_ k 57 p 1 2Ll kIk 2I I and kPk P P 51 ij ij 2 ij ij The above evolution of the back stress accounts auto- Qwith Iij being any deviatoric `strain-like' tensor, and matically, and in a continuous fashion, for the rapid ij the `dual' deviatoric `stress-like' tensor. change in stiness that is observed during the tran- The temperature-dependent function, y, was taken in sition from elastic to plastic behavior upon reversed the following (T is temperature) Arrhenius form, loading. The constants Hs, and Hl denote the short- and long-term hardening moduli, and Ls and Ll are Q y exp when T RT < T the corresponding thermal limiting states, which are KT t m given as m0 is the elastic shear modulus, and H0,f0,C,D0 and are material constants), Q T y exp ln t 1 when 0 < TRT 52 C D D D KT T t H m , L f 0 t s 0 s 0 dC and the Zener parameter, Z,as m C D D D ksij aijk 0 0 Z A sinh 53 Hl , Ll 1 f0 58 D H0 dC where Q,Tt,Tm,A and n are material constants. Now, As was done for the ¯ow law, the above original evol- Eq. (48) may be recast into the following more general ution equations with hardening/dynamic recovery form, mechanisms may be recast in the following form, 610 A.F. Saleeb et al. / Computers and Structures 74 (2000) 601±628 h i s i s a_ij 2 Hse_ij gspij de®ned by under-curved symbols and vectors by underlined symbols. h i l i l a_ij 2 Hle_ij glpij 59 4.1. General form of Newton iterative scheme where gs and gl are de®ned as; p In the typical nonlinear ®nite element analysis, the Hs J solution to the speci®ed load history is performed gs yA sinh 2Ls D incrementally. This involves projecting a known sol- ution at step n (at time tn), where the current stress p H J and internal variables are known, to time tn1 where g l yA sinh 60 the above quantities are unknown. The implementation l 2L D l is given here following the `standard' deformation-dri- The scalar internal variable D, the drag strength, ven format; i.e., for speci®ed total strain history accounts for isotropic hardening eects with its evol- en 4en1: ution expressed as, In general, within the context of the generalized mid- Á point scheme the update expression for the vector of _ i D h ke_ k yg D0RDRDmax 61 state variables, S, may be expressed as 0RjR1): Â Ã The drag strength is bounded in the interval Sn1 Sn 1 fDSn fDSn1 64 D0RDRDmax, where Dmax is de®ned below; see Eq. (63) For example, when using the explicit, forward Euler Because of its scalar nature, the strain hardening method, the update Sn1 depends only on the known and dynamic recovery terms are combined into a quantities at n, i.e., DSn (with f 0). For the fully im- single, nonlinear, hardening function, h, the hardening plicit backward Euler scheme considered here, the modulus, update is based upon the unknown quantities at tn1, ! i.e., DSn1 (with f 1). Thus the update expression m for the fully implicit backward Euler has the form, D D0 = dC h hD Â Ã 62 sinh D D0 = dC Sn1 Sn DSn1 65 and the other material constants are given as, where S is the vector of state variables and DS is the increment in state variables. The dimension of the S D D g A sinh 0 and DS vectors depend on the particular constitutive dC model, for example, for the GVIPS model, 1 s D C D S 66 max 2 0 a
If the 3D space is considered, s,a and are 6 Â 1 vec- h h when T RT < T D 1 t tors. Thus, S would have the dimension of 12 Â 1: The
expression for the state variable increment DSn1, h0 h1 based upon a ®rst-order expansion, can easily be hD h0 when 0 < TRTt 63 Tt shown to be,