Taylor Plasticity Notes

-13- SECTION 2. INTRODUCTION TO VARIATIONAL THEOREMS, 2.1, Derivatives of Functionals - The Variation ‘The weak form of a differential equation is also celled a variational equation. ‘The notion of a variation is associated with the concept of a derivative of a functional (a functional is a function of functions). In order to construct a derivative of a functional, it is necessary to introduce a scalar parameter which may be used as the limiting parameter in the derivative. This may be done by introducing the parameter 7 and defining the family of functions given by T(x) = T(x) +nt(x) 2.1) The function + is an arbitrary function and is similar to the arbitrary function W introduced in the construction of the weak form, The function 7 is called the variation of the function T' and often written as 67’ (some authors call r(x) alone the variation of the function). Introducing the family of functions 7” into the functional we obtain, using the steady state heat equation as an example, the result G" = G(W,T) = [WzkT}da-[wQda+ [Waar (2.2) a a t ‘The derivative of the functional with respect to 7 may be constructed using conventional methods of calculus. Thus, 0 8 stim - 2.3) dy 0 on where G? is the value of G" for 7 equal to 0. The construction of the derivative of the functional requires the computation of a dq With this result in hand, the derivative of the functional with respect to 7 is given by (2.4) dG ne [Wstesdn (25) ‘The limit of the derivative as 7 goes to zero is called the variation of the functional equation. For the linear steady state heat equation the derivative with respect to 7 is constant, hence the derivative is the variation of G. We shall define the deriva tive of the functional representing the weak form of a differential equation as 96. AWWyr) 26) dq-14- ‘This is the notation commonly used to define inner products. We shall also observe that symmetry of such relations is fundamental to the derivation of variational theorems. 2.2. Symmetry of Inner Products ‘An inner product functional is symmetric if A(W,r) = A(t,W) an Symmetry of the inner product resulting from the variation of a weak form is a sufficient condition for the existence of a variational theorem to generate the weak form, Symmetry of the functional A also implies that the tangent matrix of the Bubnov-Galerkin finite element method will be symmetric. ‘A variational theorem, given by a functional I(T), has a first variation which is identical to the weak form. ‘Thus, given I(T) we can construct G(W ,T') as lim. d(T") a0 dn Note that use of (2.1) leads to a result where ¢ replaces W in the weak form. Thus, the r is an arbitrary function with the same restrictions as we established in defining the weak form with W. Variational theorems are quite common for several problem classes; however, at the present time we have the functional G and desire to know if a variational theorem exists. In practice we seldom need to have the variational theorem, but knowledge that it exists is helpful since it implies properties of the discrete problem which are beneficial (e.g., symmetry of the tangent matrices, etc.). The construction of a variational theorem from a weak form is performed as follows: (a) Check symmetry of the functional A(W,r). Ifsymmetric then (b) Perform the following substitutions in G(W ,T) W(x) > T(x,t) (2.9) T (x,t) 9 nT (x,t) (2.10) = G(r,T) (2.8) to define G(T nT) (©) Integrate the functional result from (b) with respect to 7 over the interval 0 to L The result of the above process gives 1 1(T) = [Q(T nT )da (2.11) ° Performing the variation of I and setting to zero gives m2) « @(e,7) = 0 (2.12) mo dn-15- and a problem commonly referred to as a variational theorem. A variational theorem is a functional whose first variation, when set to zero, yields the governing differential equations and boundary conditions associated with some problem. For the steady state heat equation we have G(T nT) = JTsknT,da- [rears Jrqar (2.18) a a 5 ‘The integral is trivial and gives UT) = i 7,kT,d0-[Teda+ [Tq,ar (2.14) a a 5 ‘The first variation of the variational theorem generates a variational equation which is the weak form of the partial differential equation. The first variation is defined by replacing T by T= T+nt (2.15) and performing the derivative defined by (2.12). The second variation of the theorem generates the inner product A(r,t) (2.16) If the second variation is strictly positive (.e., A is positive for all r), the variational theorem is called a minimum principle and the discrete tangent matrix is positive definite. If the second variation can have either positive or negative values the variational theorem is a stationary principle and the discrete tangent matrix is indefi- nite. 2.8. Variational Notation A formalism for constructing a variation of a functional may be identified and is similar to constructing the differential of a function. The differential of the function (x;) may be written as (2.17) where x; are the set of independent variables. Similarly, we may write a first variation as ar, | ar BT = 5 8ut 5 bust (2.18) where u, u, are the dependent variables of the functional and the derivatives, 6u is the variation of the variable (ie, it is formally the q(x), and 67 is called the first, variation of the functional. This construction is a formal process as the indicated partial derivatives have no direct definition (indeed the result of the derivative is-16- obtained from (2.3). However, applying the formal construction can be formally performed using usual constructions for a derivative of a function. For the functional (2.14), we obtain the result a a On él = Af 5p ub tvensan- [ap (T@er dns | op(74,)47 ane) Performing the derivatives leads to 61 = 3] (kT ,+7,k)8T,d0- | Qerda+ fq, oT ar (2.20) a a Te Collecting terms we have 41 = f oTskT,d0-[QeTan+ [q,oTar (2.21) a a i which is identical to (2.2) with 6 replacing W, ete. This formal construction is easy to apply but masks the meaning of a variation. We may also use the above process to perform linearizations of variational equations in order to construct solution processes based on Newton’s method. We shall address this aspect at a later time. 2.4, References for variational methods & applied functional analysis [2.11MM, Vainberg, “Variational Methods for the Study of Nonlinear Operators", Holden-Day Inc., San Francisco, (1964). [2.2]F.B. Hildebrand, "Methods of Applied Mathematics", 2nd ed., Prentice-Hall, Englewood Cliffs, (1965). [2.3]J.T. Oden, “Applied Functional Analysis", Prentice-Hall, Englewood Cliffs, (1979). [2.4]K. Washizu, "Variational Methods in Elasticity & Plasticity’, 3rd ed., Pergamon Press, New York, (1982). [2.5)J.T. Oden and J.N. Reddy, "Variational Methods in Theoretical Mechanics", 2nd ed., Springer-Verlag, Berlin, (1983). [2.6]J.N. Reddy, “Applied Functional Analysis and Variational Methods in Engineer- ing", McGraw-Hill, New York, (1986). [2.71G. Strang, “Introduction to Applied Mathematics", Wellesley Cambridge Press, Wellesley, MA, (1986).-17- SECTION 3. VARIATIONAL THEOREMS FOR LINEAR ELASTICITY 3.1 Equations for Small Deformations A summary of the governing equations for linear elasticity is given below. The equations are presented using direct notation. For a presentation using indicial notation see Chapter 6, Z&T, Vol. 1. The presentation below assumes small (infinitesimal) deformations and general three dimensional behavior in a Cartesian coordinate system, x, where the domain of analysis is Q with boundary I. The dependent variables are given in terms of the displacement vector, u, the stress tensor, ¢, and the strain tensor, e. The basic governing equations are: (a) Balance of linear momentum expressed as V-o+pb, = 0 1) where p is the mass density, b, is the body force per unit mass, and V is the gradient operator. (b) Balance of angular momentum, which leads to symmetry of the stress tensor o=oT (3.2) (©) Deformation measures based upon the gradient of the displacement vector, Vu, which may be split as follows Vu = Vu+ Vu (3.3) where the symmetric part is VOu = 2[Vu+(Vu)"] (3.4) and the skew symmetric part is vou = 3[Vu-(Vu)"] (8.5) Based upon this split, the symmetric part defines the strain = Vu 6) and the skew symmetric part defines the spin, or small rotation, o = Vu (3.7) In a three dimensional setting the above tensors have 9 components. However, if the tensor is symmetric only 6 are independent and if the tensor is skew symmetric only 3 are independent. The component ordering for each of the tensors is given by On On 1s o >| on O22 05 (3.8) Os Oz 933,-18- which from the balance of angular momentum must be symmetric, hence oy = 4 9) ‘The gradient of the displacement has the components ordered as (with no symme- tries) Uy U2 Urs Vu | ue: wap uae (8.10) Usi Usa Usa ‘The strain tensor is the symmetric part with components fu éa e9 olen ee es (11) £31 &32 39 and symmetry condition ey = ey @.12) ‘The spin tensor is skew symmetric,thus, oy = — 0% (3.13) which implies 3, = 22 = 033 = 0. Accordingly, oy Om os 0 op a @ | Om O2|=|-o12 0 as (3.12) sy, 2 3) |-a3 —O 0 ‘The basic equations which are independent of material constitution are completed by specifying the boundary conditions. For this purpose the boundary, I, is split into two parts: (@ Specified displacements on the part I,, given ai u=t (3.15) where @ is a specified quantity; and (e) specified tractions on the part I, given as: t=on where €is a specified quantity. In the balance of momentum, the body force was specified per unit of mass. ‘This may be converted to a body force per unit volume (i.e., unit weight/volume) using (8.16) pb, = by (8.17)-19- In addition, transient problems may be considered by introducing the inertial forces through the d'Alembert principle b, > ba -t (3.18) ‘The inclusion of inertial forces also requires the specification of the initial conditions u(x,0) = do(x) (3.19) W(x,0) = vo(x) (8.20) where dy is the initial displacement field, and vo is the initial velccity field. 3.2 Constitutive Equations for Linear Elasticity The linear theory for quasi-static analysis is completed by specifying the constitutive behavior for the materiel. In small deformation analysis the strain is expressed as an additive sum of parts, We shall consider several alternatives for splits during the course; however, we begin by considering a linear elastic material with an additional known strain, Accordingly, e=e"te? (3.21) where e” is the strain caused by stresses and is called the mechanical part, e° is a second part which we assume is a specified strain. For example, e° as a thermal strain is given by e =e = a(T-T)) (8.22) where T is temperature and T is a stress free temperature. The constitutive equations relating stress to mechanical strain may be written (in matrix notation, which is also called Voigt notation) as o = De® = D(e-e*) (3.23) where the matrix of stresses is ordered as the vector a o=[on on on cu om on] (3.24) the matrix of strains is ordered as the vector (note factors of 2 are used to make shearing components the engineering strains, 7) ee[eu cm en 202 2en Bony (3.25) and D is the matrix of elastic constants given by Du Du Dy Du Dy Dis Du Dn Diy Du Dis Dos _|Du Da Dis Du Dis Ds ] Pl Da Da Da Du Dis Dis Di Da Dis Dis Des Des=20- Assuming the existence of a strain energy density, W(e”), from which stresses are computed as aw ob = So (3.27) the elastic modulus matrix is symmetric and satisfies Dy = Dy (3.28) ‘Using tensor quantities, the constitutive equation for linear elasticity is written in indicial notation as: as = Cabot (E0a~ £84) (3.29) ‘The transformation from the tensor to the matrix (Voigt) form is accomplished by the index transformations shown in Table 3.1 Matrix Index aie || aie ‘ab ir | 22 | 83 | 12 | 28 21 | 32 Table 3.1 Transformation of indices from tensor to matrix form. Thus, using this table, we have Cun > Dy ; Cy233 > Djs, 5 ete. (8.30) ‘The above set of equations defines the governing equations for use in solving linear elastic boundary value problems in which the inertial forces may be ignored. We next discuss some variational theorems which include the elasticity equations in a form amenable for finite element developments. For the present, the subsequent discussions will assume that inertial forces may be ignored. The inclusion of inertial forces precludes the development of varis- tional theorems in a simple form. 8.8 Hu-Washizu Variational Theorem for Linear Elasticity Instead of constructing the weak form of the equations and then deducing the existence of a variational theorem, as done for the thermal problem, a variational theorem which includes ali the equations for the linear theory of elasticity (without inertial forces) will be stated, The variational theorem is a result of the work of the Chinese scholar, Hu, and the Japanese scholar, K. Washizu (3.1, and, thus, is known as the Hu-Washizu variational theorem. The theorem may be written as-21- I(u,o,e) = ff e™DedQ-fe™Deda+fo™(Vu-e)d2 (8.81) a a a - fut, do- Juttar- [t?(u-a)ar = Stationary a 8 4 Note that the integral defining the variational theorem is a scalar; hence, a trans- pose may be introduced into each term without changing the meaning. For example, T= faTbaa = [(a"b)" a0 = [b?ada (3.32) a a a A variational theorem is stationary when the arguments (e.g., u, ¢, 2) satisfy the conditions where the first variation vanishes. ‘To construct the first variation, we proceed as in Section 2.2. Accordingly, we introduce the variations to the displacement, U, the stress, S, and the strain, E, uw = ut+7U (3.33) o'=o+n8 (3.34) and e" = e+nE (8.35) and define the single parameter functional 7’ = I(u",o",e") (3.36) ‘The first variation is then defined as the derivative of J” with respect to 7 and evaluated at 7 =0. For the Hu-Washizu theorem the first variation defining the stationary condition is given by a fe'Deda-[E"Deda dq ano a + [87 (Vu- e)da+ [o7(VU-E)da- [U"b, do (8.378) a a a - fottar- [a7 s(u-ayar- fe" Udr = 0 Fs BR i ‘The first variation may also be constucted using (2.17) for each of the variables. The result is 61 = [e™Deda- [se"De?dQ a a + [407 (Vu-e)dr+ fo V%Eu-se)d0- [ 5u"b, dO (3.37b) a a a~22- ~ Jou? tar- fa? so(u-a)ar- ft? ouar qT Ty Te and the two forms lead to identical results. In order to show that the theorem in form (3.37a) is equivalent to the equations for linear elasticity, we need to group all the terms together which multiply each variation function (e.g., the U, 8, E). To accomplish the grouping it is necessary to integrate by parts the term involving VU, Accordingly, fet veuda - -futv-cdns ft uar+ fuer (3.38) a a t i Grouping all the terms we obtain oe = JET(D(e-2°)-o1d2 ano + [87 (Vu-2)d2-fU"(V-0+b,)da (3.39) a a + Jo%@-bar- fn?s(u-aar = 0 Ty Tw The fundamental lemma of the calculus of variations states that each expression multiplying an arbitrary function in each integral type must vanish at each point in the domain of the integral. The lemma is easy to prove. Suppose that an expression does not vanish at a point, then, since the variation is arbitrary, we can assume that it is equal to the value of the non-vanishing expression, This results in the integral of the square of a function, which must then be positive, and hence the integral will not be zero. This leads to a contradiction, and thus the only possibility is that the assumption of a non-vanishing expression is false. The expression which multiplies each variation function is called an Euler equation of the variational theorem. For the Hu-Washizu theorem, the variations multiply the constitutive equation, the strain-displacement equation, the balance of linear momentum, the traction boundary condition, and the displacement boundary condition, Indeed, the only equation not contained is the balance of angular momentum. ‘The Hu-Washizu variational principle will serve as the basis for most of what we need in the course. There are other variational principles which can be deduced directly from the principle. Two of these, the Hellinger-Reissner principle and the principle of minimum potential energy are presented below since they are also often used in constructing finite element formulations in linear elasticity.=23- 8.4 Hellinger-Reissner Variational Theorem for Linear Elasticity ‘The Hellinger-Reissner principle eliminates the strain as a primary dependent variable; consequently, only the displacement, u, and the stress, o, remain as arguments in the functional for which variations are constructed. The strains are eliminated by developing an expression in terms of the stresses. For linear elasticity this leads to . e=e+D a (8.40) ‘The need to develop an expression for strains in terms of stresses limits the applica- tion of the Hellinger-Reissner principle. For example, in finite deformation elasticity the development of a relation similar to (3.40) is not possible in general. On the other hand, the Hellinger-Reissner principle is an important limiting case when considering problems with constraints (e.g., linear elastic incompressible problems, thin plates as a limit case of the thick Mindlin-Reissner theory). Thus, we shall on occasion use the principle in our studies. Introducing (3.40) into the Hu-Washizu principle leads to the result T(u,o) = -} fe" DePda-} [oD ada- fot eda a a a + J oT Vuda- fu"b, da (3.41) a a - futéar- fe(u-war y Ty ‘The Euler equations for this principle are vy = 2 +Dto (3.42) together with (3.1), (3.15) and (8.16). The strain-displacement equations are deduced by either directly stating (3.6) or comparing (3.40) to (3.42). The first term in (8.41) may be omitted since its first variation is zero. 3.5 Minimum Potential Energy Theorem for Linear Elasticity The principle of minimum potential energy eliminates both the stress, o, and the strain, ¢, as arguments of the functional. In addition, the displacement boundary conditions are assumed to be imposed as a constraint on the principle. The MPE theorem may be deduced by assuming e= Vu (8.43) and usa (3.44)=24- are satisfied at each point of © and I, respectively. Thus, the variational theorem is given by the integral functional Ta) = 3 f (Vu) D(Vu)da- f(Vu)? Deda (3.45) a a - Jutb,do- fuéar a i Since stress does not appear explicitly in the theorem, the constitutive equation must be given. Accordingly, in addition to (3.43) and (3.44) the relation o = D(e-e*) (8.46) ‘is given. ‘The principle of minimum potential energy is often used as the basis for developing a displacement finite element method. [3.1]K. Washizu, “Variational Methods in Elasticity & Plasticity”, 3rd ed., Pergamon Press, New York, (1982).=25- SECTION 4. FINITE ELEMENT APPROXIMATIONS 4.1 Introduction A variational equation or theorem may be solved using the direct method of the calculus of variations [4.1]. In the direct method of the calculus of variations the dependent variables are expressed as a set of trial functions multiplying parameters, This reduces a steady state problem to an algebraic process and a transient problem to a set of ordinary differential equations. In the finite element method we divide the region into elements and perform the approximations on each element. As indicated in Section 1.3 the region is divided as Q2=%= ‘Fo 4 and integrals are defined as f-f-E5 (4.2) 20 eG A similar construction is performed for the boundaries. With this construction the parts of the variational equation or theorem are evaluated element by element. In the next sections we consider the computation of the external force (from applied loads) and the internal force (from stresses) by the finite element process. 4.2 External Force Computation In our study we will normally satisfy the displacement boundary conditions u = @ by setting nodal values of the displacement to the values of a evaluated at nodes. Thus, by this approach we shall assume the integral over I, vanishes. This step is not necessary but is common in most applications. The remaing terms involving specified applied loads are due to the body forces, b,, and the applied surface tractions, f, The variational terms are Tye Jute, aos furtar (43) a te ‘The finite element approximation for displacements is introduced as w= N,(g)ul (44) where N is the shape function at node J, ¢ are natural coordinates for the element, and w! are the values of the displacement vector at node I. Using (4.4) in (4.3) yields=26- 1, = (a!) [Nyb,da+ friar) = (oF, (4.5) a, ty where F; is the applied nodal force vector at node I. 4.3 Internal Force Computation ‘The stress divergence term in the Hu-Washizu variational principle is generated from the variation with respect to the displacements, u, of the term 1, = [Vu oda = ¥ [(Vuyr oda (46) a "a Using the finite element approximation for displacement, the symmetric part of the strains defined by the symmetric part of the deformation gradient in each element is given by Vu = e(u) = Bru! (4.7) where B, is the strain displacement matrix for the element. If the components of the strain for 3-dimensional problems are ordered as oF = [en senrtan Pe 26m 26a] 48) and related to the displacement derivatives by ae [sa tea stnastae + Ugg stag + 4320431 + “41a] (4.9) the strain-displacement matrix is expressed as: Nia 0 0 0 Nip 0 oo By = (4.10) 1") Nig Nia 0 0 Nis Nis Nis 0) Mia where any Ny = Bt For a 2-dimensional plane strain problem the non-zero strains reduce to ge [ew séas60s 2612] 4.)-27- and are expressed in terms of the displacement derivatives as fe [ua Maa +0 tye + uaa] (4.12) thus, B; becomes: Nia 0 _| 0 Ma Bela oo (4.13) Nia Nin Finally, for a 2-dimensional axisymmetric problem (with no’ torsional loading) the strains are [ex éenseu0 2200] (4.14) and are expressed in terms of the displacements as ce [ort ht + (4.15) 1 ‘The strain-displacement matrix for axisymmetry, By, becomes: Nn 0 0 Nig B; = (4.16) fo | se x Nig Nia where x, x» now denote the axisymmetric coordinates r, z, respectively. (4.17) ‘The stress divergence term for each element may be written as 1, = (al? [(ByF oda (4.18) a In the sequel we define the variation with respect to the nodal displacements, u’, as the internal force for the stress divergence which is expressed by ‘4.1, For axiaymmetry it fs aloo necessary to replace the volume element by dO 4 ndndn ‘and the surface clement by ar + 2,08 ‘where dS is an boundary diforential in the «1-3 plane.=28- Pio) = [Boda (4.19) a which gives = (a Po) (4.20) ‘The stress divergence term is a basic finite element quantity and must produce a response which is free of spurious modes or locking tendencies. Locking is generally associated with poor performance at or near the incompressible limit. ‘To study the locking problem, in the next sections we split the formulation into deviatoric and volumetric terms. 4.2 Split into Deviatoric and Spherical Parts For problems in mechanics it is common to split the stress and strain tensors into their deviatoric and spherical parts. For stress the spherical part is the mean stress defined by 1 1 p= Giro) = 50m (4.21) For strain the spherical part is the volume change defined by = tre) = on (4.22) ‘The deviatoric part, s, is defined so that its trace is zero. The stress may be written in terms of the deviatoric and pressure parts as o=stpl (4.23) where, 1 is the rank two identity tensor, which in matrix notation is given by the vector 4? 1” = [1,1,1,0,0,0] (4.24) In matrix form the pressure is given by = dit page (4.25) thus, deviatoric part of the stresses now may be computed as eect «{1-21%" s= 0-511 o=(t fu'}e (4.26) where Lis a 6x6 identity matrix. We note that the trace of the stress gives Io =3p=i"s+pll = 175+3p (4.27) “42. In Zionkiewics and Taylor, Vol. 1 (4.2), the identity 1 is denoted by m.-29- and hence _ 1"s [ be . as required. ae 1 a | We define . ne 4 Taw = I- : i” (4.29) as the deviatoric projector. Similarly, the volumetric projector is defined by La = dart (4.30) ‘These operators obey the following properties T= Lethe 431) Taco = Taco Yeu = Caco)” (4.32) Toa = Toa Toot = Ghat)” (4.33) and TatTaw = TieoTor = 0 (4.34) In the above m is any positive integer power. Utilizing the above properties for defining a devistor we can operate on the strain to define its deviatoric and volumetric parts. Accordingly, the deviatoric and volumetric parts are given by es erto1 (4.38) where e is the strain deviator and @ is the change in volume. Using o=ite (4.36) we obtain e@=lime ; Te=0 (4.37) The strain-displacement matrix may now also be written as a deviatoric and volumetric form, Accordingly, we use the strain split e(a) = Bru! = (Bay)! +(Byg pa! (4.38) where Baw = lio B (4.39) Bw = LaB = qib (4.40)-30- where b=17B ; ITB =0 4.41) For 3-dimensional problems by = [aa NraNia] (4.42) is the volumetric strain-displacement matrix for a node I, in its simplest form. In 2-dimensional plane problems the volumetric strain-displacement matrix is given by b= [71 Nia] (4.43) and for 2-dimensional axisymmetric problems by = [mas Bt ar] (4.44) 1 The deviatoric matrix By, is constructed from (4.35) and yields for the 3-dimensional problem 2Nr. -Ni2 -Nis 3Ni3 2Nig -Nis 2Nia 0 3Nig 3Nia (4.45) 2Ni -Nig “Nia 2Nig “Nin -Nia 8Nia 3Ni1 ‘The construction of the deviatoric matrix for the 2-dimensional axisymmetric problem is left as an exercise for the reader. (4.46) 4.8 Internal Force - Deviatoric and Volumetric Parts ‘The sbove split of terms is useful in writing the internal force calculations in terms of deviatoric and volumetric parts. Accordingly, P, = [Bfoda = fBr@+ pda 447 a, a,-31- which after rearrangement gives Py = fBreda+ [Bri pda (4.48) a a If we introduce B = Buy + Buy = Bag + 5 1b 4.49) and use the properties defined above for products of the deviatoric and volumetric terms, then P, = [(BL,)sda+ [b7 pda (4.50) a a and, thus, the internal force is composed of the sum of deviatoric and volumetric parts. 4.4 Constitutive Equations for Isotropic Linear Elasticity ‘The constitutive equation for isotropic linear elasticity may be expressed as o = Altr(e)+2ue (4.51) L. : ven where 4 and y are the Lamé parameters which are related to Young's modulus, E, and Poisson's ratio, v, by vE E rm Conia For different values of v, the Lamé parameters have the following ranges Osvs} 5; OSASoo (4.53) and E E Osvs} ; yeeesz (4.54) For an incompressible material v is 3; and 4 is a parameter which causes difficulties since it is infinite. Another parameter which is related to 4 and x is the bulk modulus, K, which is defined by eae 2 E ee Sots gage ae ‘We note that K also tends to infinity as v approaches }. & : In index form the constitutive equation is given by 7 2d oy = Abyeu+2uey (hur) € (458) 2 ey lrg Some-32- For a general linear elastic material we have oy = Cyan asp where cj are the elastic moduli. For an isotropic material the elastic moduli are given by cm = Ady Sut n(Su5p + 65) (4.58) ‘We note that the above definition for the moduli satisfies all the necessary symmetry conditions; that is Cys = Cay = pnt = Coe (4.59) ‘The relations may be transformed to matrix (Voigt) notation following Table 3.1 and expressed as o=De (4.60) where the elastic moduli are split into D= (4.61) with 111000 111000 ee 00000 0/722 = aha (4.62) 000000 000000 and 200000 020000 002000 Deals oa G00 (4.63) 000010 000001 used as non-dimensional matrices to split of the moduli.** If the moduli matrices are premultiplied by I,. and Ij. the following results are obtained TaD, = Dy (4.64) TawD, = 0 (4.65) “43, Note that in D, the terms multiplying shears have unit values since engincering shear strain are ‘waod (Ley ry Bey).LaD, = 511 = 3p, (4.66) 422000 2 42000 1]-2 2 400 0|_A4 Dylaw = TwPs = 3] 9 9 09 3 0 0] 7 De 4.67) 000030 000008 Once Dag has been computed it may be noted that, Lie Deus = Daw La = Dew (4.68) and, thus, it is a deviatoric quantity. In the following section, the computation of the element stiffness matrix for a displacement approach is given and is based upon the above representations for the moduli. 4.5 Stiffness for Displacement Formulation ‘The displacement formulation is accomplished for a linear elastic material by noting that the constitutive equation is given by (for simplicity e° is assumed to be zero) o=De (4.69) and the strains for a displacement approach are given by e = Bu (4.70) where u’ are the displacements at node J. Constructing the deviatoric and volumetric parts may be accomplished by writing 8 = Tayo = IuyDe = Tuy (4D, + uD, (4.71) and pl =WgDe = Ta(AD, +uD,)e (4.72) If we use the properties of the moduli multiplied by the projectors, the above equations reduce to 8 = “Daye = uD,e = uD, Bay) 0” (4.78) and 24 yp,e= KDe = KU(1"e) = K10 (4.74) pls (a+=34- Thus, the pressure constitutive equation is p=Ke (4.75) Noting that the volumetric strain may be computed from 6 = byw (4.76) the pressure for the displacement model may be computed from : p=Kbjw a7 We recalll from Section 4.3 that P, = [(BZ,)8d0+ Jer pao (4.78) a a, ‘Using the above definitions and identities the internal force vector may be written as P, = J #5.) Dy Ba)s d2u7 + J Kb;b3 dow (4.79) a, a and, thus, for isotropic linear elasticity, the stiffness matrix may be deduced as the sum of the deviatoric and volumetric parts Ky = Kacodis + Boos (4.80) where Coody = [xBi1D, Bay d2 = [xBFDsyBsd2 (4.81) a a, and Beda = f Kbybjdo = fxB7D,B’ da (4.82) a a 4.6 Numerical Integration Generally the computation of integrals for the finite element arrays is performed using numerical integration (i.e., quadrature). The use of the same quadrature for each part of the stress divergence terms given above (in P and K) leads to the conventional displacement approach for numerically integrated finite element developments. The minimum order quadrature which produces a stiffness with the correct rank (j.e., number of element degree-of-freedoms less the number of rigid body modes) will be called a standard or full quadrature (or integration). The next lowest order of quadrature is called a reduced quadrature. Alternatively, use of standard quadrature on one term and reduced quadrature on another leads to a method called selective reduced quadrature.-35- A typical integral is evaluated by first transforming the integral onto a natural coordinate space roman = fren irs (4.83) Qe a where J denotes integration over the natural coordinates £, dg denotes déydé, in a 2-dimensions, and j(¢) is the determinant of the jacobian transformation ax IE) = % (4.84) ‘Thus G(G) = detd(E) (4.85) ‘The integrals over O are approximated using a quadrature formula, thus L Jfx(@nsi@rds = EIEN IG (4.86) a where £, and w; are quadrature points and quadrature weights. For brick elements in three dimensions and quadrilateral elements in two dimensions, the integration is generally carried out as a product of one-dimensional Gaussian quadrature. Thus, for 2-dimensions, 11 facerds = | facerder ate as a Aa and for 3-dimensions ait Jaceras = Jf Jecerderder des (4.88) 3 Aad Using quadrature, the stress divergence is given by Py = BBE)" OGG (489) and the stiffness matrix is computed by quadrature as By = EBNF DEBE) III (4.90) Similar expressions may be deduced for each of the terms defined by the deviatoric/volumetric splits. The use of quadrature reduces the development of finite element arrays to an algebraic process involving matrix operations. The basic algorithm is given by:-36- 1, Initialize the array 2. Loop over the quadrature points, / a. Compute j(€:)w=¢ b. Compute the matrix part of the integrand, (e.g., By ( 1)" 0; = Ay). ¢. Accumulate the array, e.g., P, © Py+Aye 3. Repeat step 2 until all quadrature points in element are considered. Additional steps are involved in computing the entries in each array. For example, the determination of B; requires computation of the derivatives of the shape functions, N,,, and computation of o; requires an evaluation of the constitutive equation at the quadrature point. The evaluation of the shape functions is performed using a shape function subprogram. In FEAP, the shape function routine for 2 dimensions is called shape and is accessed by the call call shape( xil, xi2, xl, shp, xsj, ndm, nel, ix, flag) where xil - natural coordinate ¢, at quadrature point (input) xi2 - natural coordinate & at quadrature point (input) x1 - array of nodal coordinates for element (x1 (ndm, nen) ) (input) shp - array of shape functions and derivatives (shp (3 , nen) ) (output) xsj - jacobian determinant at quadrature point (output) ndm ~ spatial dimension of problems (input) nel - number of nodes on element (between 3 and 9) (input) ix - array of global node numbers on element (ix (nen) ) (input) flag - flag, if false derivatives returned with respect to x (input); - if true derivatives returned with respect to ¢. ‘The array of shape functions has the following meanings: shp(1,1) is Ny, shp(2,1) is yg shp (3,1) is Ny ‘The quadrature points may be obtained by a call to pgauss: call pgauss( 1, lint, xil, xi2, w) where 1 ~ number of quadrature points in each direction (input). Lint - total number of quadrature points (output).-37- xi - array of natural coordinates ¢, (output). xi2 - array of natural coordinates é, (output). w _ ~ array of quadrature weights (output) Using the above two utility subprograms a 2-dimensional formulation for displacement (or mixed) finite element method can be easily developed for FEAP. An example, is element e1mt 01 which is shown below. References [4.1]F.B, Hildebrand, “Methods of Applied Mathematics", 2nd ed., Prentice-Hall, Englewood Cliffs, (1965). [4.2]0.C. Zienkiewicz and R.L. Taylor, "The Finite Element Method, Vol 1.”, 4th ed., ‘McGraw-Hill, London, (1989).-38- SECTION 5. MIXED FINITE ELEMENT METHODS 5.1 Solutions using the Hu-Washizu Variational Theorem A finite element formulation which is free from locking at the incompressible or nearly incompressible limit may be developed from a mixed variational approach. In the work considered here we use the Hu-Washizu variational principle, which we recall may be written as Iu,o,e) = ffeTDeda-fe™Deda+ fo7(Vu-e)daQ (6) a a a - Jutb,da- Juttar- fe cu-war = Stationary a i & In the principle, displacements appear up to first derivatives, while the stresses and strains appear without any derivatives. Accordingly, the continuity conditions we may use in finite element approximations are C° for the displacements and C”* for the stresses and strains (a C~? function is one whose first integral will be continuous). Appropriate interpolations for each element are thus u(g) = N(E)u(t) 2) CE) = 0(E)0%t) 6.3) and eS) = wal et) (6.4) where ¢,(€) and y,(€) are interpolations which are continuous in each element but may be discontinuous across element boundaries.*? The parameters o” and e* are not necessarily nodal values and, thus, may have no direct physical meaning. If, for the present, we ignore the integral for the body force, and the traction and displacement boundary integrals and consider an isotropic linear elastic material, the remaining terms may be split into deviatoric and volumetric parts as I(u,o,e) = Bf xe™Daqed0- J ue" Dip edQ+ fs" Le(u)-elda a a a +} [xeda-[Ke0°da+ | plau)-o1da 65) a a a where e(u) = Tay Vu (6.6) A Strict, 9, and y, need only be piocewise continuous in exch element; however, this makes the ‘evaluation of integrals over each element more difficult and to date is rarely used.-39- and ou) = 17 Vu 6.7) are the strain-displacement relations for the deviatoric and volumetric parts, respectively. Constructing the variation for the above split leads to the following Euler equations which hold in the domain 0: () Balance of Momentum V-(st1p)+b, = 0 8) which is also written as div(s+1p)+b, = 0 (b) Strain-Displacement equations e(u)-e =0 (6.98) ou)-@ = 0 (6.9b) (©) Constitutive equations BDg 8 = 0 (6.10) Ko-p=0 (6.10b) In addition the boundary conditions for I, and I, are obtained. Using the interpolations described above, the Hu-Washizu variational theorem may be approximated by summing the integrals over each element. Accordingly, Iu, o,e) = Iiu,o,e) = D1.(a,0,€) an If the deviatoric part is approximated by taking e=e(u) (6.12) for each point of @, this part of the problem is given as a displacement model. The variational expression (5.5) becomes Tu, p,6) = Bf weTw)Dgy eC) d0- f 16"(a)Day €° dO (6.13) a A a +} [Ketda-[Kee°da+ | plau)-e]d0 ta 4 a a a which may be split into integrals over the elements as Ku, p,6) = Inu, p,0) = L1u, p,6) (6.14) ‘A mixed approximation may now be used to describe the pressure and the volume change in each element. Accordingly, we assume-40- PCE) = gal E) P(t) = ED (6.15) OE) = Gal E Ot) = WEE (5.16) where it is noted that the same approximating functions are used for both p and 0. If the material is isotropic linear elastic, the use of the same functions will permit an exact satisfaction of the constitutive equation, equation (5.10b) at each point of the domain of an element. For other situations, the constitutive equation may be approximately satisfied. Recall that the strain-displacement equations for a finite element approximation are given by e(u) = Bru! (6.17) ‘Thus, the finite element approximation for the mixed formulation may be written as T.(a,p,@) = (u!)7| 2 J #BT Daw By dow? - J #BT Dice e° do a a veer ane Jara 6.18) a a al fe? by daw’ - [reas] a a If we define the following matrices: k= fe" ean (6.19) a, wos [xe ean (6.19b) a he= fe" eda (5.19) a, gr = [e*b,do (6.194) 4, and recall that the deviatoric stiffness is defined as (Kaw iz = [BF DieBy do (6.20) a and denote the effects of initial deviatoric strains as (Phar = [ABP Day ed = [ uBFD, edo (6.21) a aae where e° are the deviatoric initial strains. The mixed variational terms become 1,9.) = (01)? # (Bag) 0? ~ (Po )r] (6.22) +0"[ }ko-2°]+p"[g,u’-ho] If we denote the variations of pressure and volume change as Py = p+al (6.23) 6, = 6+n® (6.24) the first variation of (5.22) may be written in the matrix form al (Kaos BF Of] [CPL dr a = [cont nt,e7] g Oo -hiip|-|] o |} 25a) a o -h kjle Py or in variational notation as (Kew) sf 0 ]fu" (Peer él, = [cou op? 207] g 0 -hilp|-| o | {5.25») 0 -h k ile iz We note that the parameters p and 6 (and their variations II and @) are associated with a single element, consequently, from the stationarity condition, the last two rows of the above matrix expression must vanish and may be solved at the element level. The requirement for a solution to exist is that *? Ny 2 Np (6.26) where n, and n, are the number of parameters associated with the volume change and pressure approximations, respectively. We have satisfied this requirement by taking an equal number for the two approximations. Also, since we used the same functions for the two approximations, the matrix h is square and positive definite (provided our approximating functions are linearly independent), consequently, we may perform the element solutions by inverting only h. The enltions to ‘the second row of (5.25) is = 18. q : (5.27) and the solution to the third rowis obey Ko an p=h"(ke-x°) (6.28) = a Mee | a . le J B2 This is a mixed potch test requirement. See Zienkiewics and’ r, Vol. 1., Chapter 12 (6.1). Pek’ May yr = Pa,-42- the above simplifies to au! rf dy + BF by Ju! ~ (Phy -BF “| an ‘The volumetric stiffness for the mixed formulation is given as (Key = Br kb, (6.32) and the volumetric initial force by (Paar = By? (6.33) ‘The stress divergence term for the mixed model formulation is computed from P, = [BI (s+ p1)d0 (6.34) a, where the deviatoric stress is expressed by the displacement approximation as 8 = uDéq(Byu’ - e°) (6.35) and the pressure is expressed by the mixed approximation as p= ¢é)h "(ke - 2°) 6.36) 5.2 Finite Element Solution for Mixed Formulation ‘The mixed finite element solution for the linear elastic problem requires selecting a set of approximating functions for ¢. The number of ¢ functions will affect the rank of the volumetric terms. The modified volumetric stiffness has a rank which is given by rank (Ky) = min(rank(b), rank (Ik)) (637) Provided the approximations for ¢ are linearly independent, and the number is small compared to the number of degrees-of-freedom on the element, the rank will normally be that of k. For example, 4-node quadrilateral or 8-node brick elements can use a single function f=1 (5.38) for the approximating space. This gives a rank of 1 for the volumetric stiffness. The requirement for the approximation is guided by the principle that: (1) we use the minimum number of functions which make K have correct rank for a single element, and (2) the functions produce an element which is invariant with respect to the input data. For example, if we show that two functions are sufficient for a 2-dimensional element, use of tals a=-43- L would not be good since the element is not invariant with respect to a permutation in the definition of £, and £,. Several alternatives are possible, one being gre 15 a= Gites another is to use 3 functions with Or =15 b= 61 5 Os = oo ‘The actual functions selected. must be subjected to further evaluations to decide which best meets the objectives of the problem solution. An algorithm to implement the above mixed model for linear elasticity where D is constant in each element may be summarized as: 1, Initialize arrays: g;, h, k, x°. FEAP will initialize K and the element residual. 2. Loop over quadrature points, ! a. Compute shape functions: In 2-d problems FEAP uses, N;(&) = shp(3,1,1) Nis(4) = shpG,1,D b. Compute the volume element times the quadrature weight jru, = dvQ) Cc jm 3. Loop over quadrature points, ! a. Compute the volumetric strain matrices, gy and h. 4. Invert h and compute by by = hg, 5. Loop over quadrature points, a. Compute strain-displacement matrix, B, and strains, ¢ : 7 : \ a= [Law Br( £1) + 5 1 6(6) B10! b. Compute quadrature stre id D(e-e°) ec. Compute the residual rer Fo eats) ~*\\,\ d. Compute the deviatoric tangent, Ky C e. Compute the volumetric local tangent, k and x° int on : R= Fy ZBI) IG)-44- 6. Compute the tangent, K K = Ky +b" kb References [5.1]0.C. Zienkiewicz and R.L. Taylor, “The Finite Element Method, Vol 1.", 4th ed., ‘Hill, London, (1989).SECTION 5. MIXED FINITE ELEMENT METHODS (Cont.) 5.8 Mixed Solutions for Anisotropic Linear Elastic Materials. ‘A more general form of the Hu-Washizu principle is needed to consider either anisotropic linear elastic materials or inelastic materials in which there is coupling between volumetric and deviatoric effects. In this section we construct the form of the functional for an anisotropic linear elastic material. Accordingly, we have o = Die-e" 6.39) where D is a symmetric matrix in which there may be coupling between the deviatoric and volumetric strain effects. It is now assumed that a finite element solution will be constructed in which deviatoric strains, e, are computed directly from the displacements but the volumetric strain, 6, is computed from a mixed form. Accord- ingly, 2 = Tye(u)+ 210 (6.40) Actress may be computed from # as & = Dilan (e(u)=e) +5 1(0-6")] 6.41) where 6° = 17e°, The stress may be split into deviatoric and pressure parts as @=5+1p (6.42) where 8 = Lap DL Tay (eC) ~e8)+ 51(6-6"D] 6.43a) and B= EI" DUTae(e(u)~e°)+ 3100-69] 6.43b) Ifwe define De. ? Dace = Tso D Lew a= F4aeD1then the stress may be written as 2 IDage(u)+do+ 21(€%e(u)+dy50)+6°+ 1p" 6.44) This form of the stress may be multiplied by the virtual Z and integrated over the domain to obtain part of the variational equation associated with the strain energy. Subsequently, adding the terms associated with the mixed volumetric pressure and volume change Vainberg’s theorem may be used to obtain a variational theorem. Alternatively, the stress and strain splits may be substituted into (5.1). The result is : Dan a Tecu) a ‘des 2 I(u,p,@) bftecw, lf al ‘ 2 45) + f(e(u)s +0 p°)da+ J plo(m)-01d2+ Lex a a ‘This form of the variational principle is equivalent to (5.13) which was deduced for isotropic materials. The added terms in (6.45) are all associated with d which defines a coupling between deviatoric and volumetric strains. For isotropy d is zero. If we introduce finite element interpolations using standard displacement interpolation together with the pressure and volume interpolations given by (5.15) and (5.16), the first variation of (5.45) for a single element is r BID,,.By Bldg [a], [Brs' I, = (507,86 ul +a"B, me? + ae dQ (6.46) r Ls 1 sat, | Bie +6p' Jette eral 4 |ocoat 1f[ MF, Jones ote The variational equation (5.46) may be expressed in terms of stresses by substituting the interpolations into (5.43) resulting in 61, = [60,36 risa 641 +587 [#0 -ora0| §]sco0f 168 ray 2, Jno 2t— a Q,| Since the interpolations for the pressure and volume change are associated with a single element it is possible to solve for their parameters at the element level. Accordingly, the multiple of 5p yields JeTb, aaa, = JePednd = hd (6.48) a a "which yields 6 = bya; = hg a; * (6.49) where h and g; are as defined in (6.19c) and (5.194), respectively. similarly, the equation multiplying 56 yields the equation Jo? pan = [oP eanp = hp (6.50) a a Using these results, the first integral in the variational equation defines the stress divergence terms Ig = sof | [BPada+b; fo” pda) 6.51) a a which upon use of the definitions for the mixed pressure, p, and the mixed volumetric strain displacement equation, by, yields 41, = 60 [BT[é+1p]da (6.52) a ‘The stress of the mixed method is defined as o=8+1p (6.53) and, in general, is not equal to &. The stress @, however, is the stress which is computed from the constitutive equation for each material. Thus, when we later consider other material models (e.g., viscoelasticity, plasticity, etc.) the effective material moduli are the ones computed by linearizing the constitutive equation expressed in terms of the & stresses. The residual for a finite element formulation is most efficiently computed from the mixed stress and we note the result is identical to the form of the standard displacement model except for the stress expression used. ‘The tangent matrix may be expressed in terms of the displacements alone by writing the variational equation (5.46) as op at ap | Begs Kr gy Yr 81, = (807 58" ,507)/ 5 yr —b |] 6 [+ 5I0+5Tex (6.54) gf -h oll ‘Using the solutions to (5.48) and (5.50) the dependence of (5.54) on 6 and p may be eliminated to give 61, = 807 Ky Gy + 51o+ STex where Ky = Bay + bibs + B7k5 + by kab,-4- The algorithm for the development of a mixed element based upon the above may be summarized as: Step 1: Numerical integration of strain matrices ” a.) Compute ¢=[1, £1, £2,--] (for the 4-node element ¢=1 b.) Compute arrays h= ean 4, g = fo" do a Step 2: Mixed volumetric strain displacement matrix a.) Compute b; = hg, Step 3: Constitution computation for each quadrature point a.) Compute e= Bay @ = o6)b,a; z= Leet 310 & =Dlz-e"] p= 17a/3 = feraan a Step 4: Mixed Pressure a.) Compute p = 9(¢)h?# Step 5: Residual and Stiffness Integrals a.) Compute mixed stress o = 1,,,+1p b.) Compute =e J BT odQ Bows = fr Bi Ds By dQ a k= J Bl dgdQ a Kea = f "dyes dO a 7)Step 6: Stiffness assembly a.) Compute : Ky = Bandy + krby + by k3 + By kya by Assignment Problem 1. Consider a linear elastic isotropic material model where D is given by At2n a 4 000 A Atty 4 000 4a A At2n 0:0 0 0 0 © 200 0 0 0 OO 0 0 0 00 7 show that the principle given by (5.45) reduces to the one given by (6.13). Problem 2. For a 4-node isoparametric quadrilateral element in two dimensions, compute the number of operations required to form the deviatoric stiffness matrix at each quadrature point, using: (a) Ky = JB? Daw By d2 a ) Ky = J BY, DBany dQ cy where Beet = Teo Br is computed explicitly, Note sparsity in each form of the B arrays but assume D is fully populated. Sepa- rate your operations into two groups: (a) additions and subtractions and (b) multiplies. Do not include costs of computing shape functions, their derivatives, and the jacobian determinant needed for numerical integration. Problem 3. Modify the 4-node quadrilateral, plane strain, mixed element, e1mt03 (which currently considers only isotropic linear elasticity) to permit solution of prob-_ Jems with a fully populated D matrix. It is suggested you develop a new element-6- (e.g., elmt13) as your modified element to facilitate the testing phase. The modifications consist of: a.) Permit input of a fully populated D elasticity matrix (4x4 for plane strain); b.) Modification of the subroutine which sets up the elasticity matrix from the input parameters; and ©.) Add coupling effects to the stiffness matrix (i.e., the terms associated with d). ‘Test your element for an isotropic material. Conduct patch tests and a solution to a simple problem. Then specify properties for an anisotropic material which renders he D matrix ful One option is to use an orthotropic material with elastic axes different from the coordinate axes, Rotate to the coordinate directions to get the fully Populated matrix. Perform constant strain patch tests (use a simple patch with a rectangular outer shape). Be sure bo itis set.to he consistent with coupled deformations from stress inputs. Problem 4. Consider the problem of an infinitely long thick walled cylinder which is loaded by a constant internal pressure. The problem may be considered as a plane strain problem for annular slices, For a typical annulus use symmetry conditions Gand isolate one row of elements to model the problem. The boundaries of the row of elements are: (a) the inner radius, (b) the outer radius, (c) a boundary where the polar angle, 6, is zero, and (d) a boundary where the polar angle @ a small value (do not make the angle too small or solution will become sensitive, generally an angle of 5 to 10 degrees will be adequate), Analyze the mesh using either elmt 03 or your modified element using 5, 10, and 20 elements in the radial direction and ‘isotropic linear elastic properties with Poisson ratios of 0.3, 0.4, 0.45, and 0.499995. Repeat the analysis using elmt01 (be sure the parameters are set to get plane strain). Note using elmt 01 you can also do an axisymmetric analysis to identify the influence of yoursmall 6 angle. Submit a short report on your findings. ~-41- SECTION 6. MIXED FINITE ELEMENT METHODS 6.1 Solutions using Hu-Washizu Variational Theorem: General Problems The finite element approximation for the mixed formulation of a general linear elastic material (i.e., anistotropic behavior) may be written for a typical element as 1,(a,p,0) = } [e™Deda- fe™De°dN+ p(V-u-4)d2 (6.1) Dy a Using the approximations introduced for the isotropic model for the displacement and mixed volume change gives € = Tae Byul + 51960 62) which when introduced into the variational theorem gives 1.00,p,0) = } [Uae Brul + £14(5)0F Dla Bru! + 31 6(6)01d0 a ~ [Uae Bru’ + 510601" Deda 6a) a +p| [bj daw’ - [¢7edne a, a For symmetric D , we can define the following matrices: (Baw ig = [BP Hes DYieeBy dQ (64a) a (Kady = 3 fo" 1 DIB, a0 (6b) a k= 2 [¢"i"Digda (6.40) 9 4 os 1 fyryp.e x af¢ 1De°dd (6.44) h= feToaa (6.4e) a g = fe"b,da 64h a-42- and denote the effects of initial deviatoric strains as (Por = [BT TagDe°do (64g) 4, The mixed variational terms become T,(a,p,6) = 3(Cul)? (Kay yy a7 + (ul (KZ); 6 + 07 (K,.)7 w + 07k 0) (6.5) (ul)? (Pi, - 67 2° +p" gpa! - pho ‘The first variation of (6.5) may be written in the matrix form a, (Baw g BF CRB] w] Pode &: = [cory 1", 67] go -h lip|-| o (66) : (Ky -h ok Sle . Recall that the terms which multiply the variations in pressure, I, and the variation in the volume change, ®, are associated with individual elements, and, thus, the second row of (6.6) may be solved at the element level to give the parameters for the volume change, 8, as z 6 =h'g,u 67 and the solution to the third row is p= h"[(K,,)ju7+ko-2°] (6.8) Defining a modified volumetric strain-displacement matrix as by; = hg, (6.9) Substitution of the above results into the first equation gives & (OF (aed #BF dy HKG YB, +B RB Jw” (6.10 = (Py) - BP *) ‘Thus, the stiffness matrix for the general anisotropic linear elastic formulation is given by Ky = (Kew iy + By (Keo )g + (KS sy +B; ky (11) This operation may be performed after all the integrals over the element are evaluated. ‘The matrices which involve the elastic moduli may be simplified by defining some reduced terms. Accordingly, we let ®* a= tpi (6.12)Also, define dig = gip1 = RY (6.13) ‘Then DI... = D-d1" (6.14a) or TayD = D-14" (6.14b) which gives FheD1 = (D-14")1 = ands = daw (6.15) Finally, the deviatoric part of the modulus is now defined in terms of the above as Dao = Tey DIaey = D-d17 - 1d" + yy 117 (6.16) For isotropy, the above expressions reduce to: dy = K (6.172) d =[K,K,K,0,0,0!" (6.17b) day = 0 (6.17¢) and Om) 00 ; : (6.174) 30 03 ‘The matrices for the mixed treatment of the symmetric D anisotropic case are computed as: (Bae) = [BF Daw By a0 (6.182) a (Ka)s = [97 dZ,By do (6.18b) a A. ICD is not aymmetrie, equations (6.12) through (6.16) must be modified. Essentially, this requires ‘scomputation of two d terms as dp 2 Di; a =D% ‘and using these in the remaining equations instead of d (note, when D is aymmetrie the dp and d terms are equal)-44- and : k= fdme™edQ ~ (6.18c) a ‘The matrix for the initial strains is computed as = [ord erda “(@.184) a which is a 1x6 vector. Example: Tri-linear interpolation for u, constant ¢. As an example, we consider the case where the set of shape functions for the displacements is the tri-linear interpolation Wie) = Zt ede) 14 eh ss) where ¢] are the values of the natural coordinates at the J-node. The interpolation for the pressure (and volume change) is constant g=1 This element is often called B1-PO (order 1 interpolations for the brick element, order 0 for the pressure/volume change). In 2-dimensions the element is Q1-PO, for the order 1 quadrilateral. Higher order elements are also defined, for example, the Q2-P1 element uses quadratic interpolation for displacements (Lagrange interpolations) and linear for the pressure with (8) = [1.41.21 Alternatively, itis possible to use the interpolations CE) = (1,2(8), 2089] ‘The matrices for the B1-P0 (or Q1-PO) element reduce to (Bos = [dioBy da a, which is a column vector (of size 1x24). For isotropy, this matrix is zero. The vol- ‘ume stiffness becomes: k= fdwda a which is a 1x1 matrix and for constant d,.; becomes K = hot = dui Qe ~ ~-45- where Q, is the volume of the element. For isotropy hy is the bulk modulus times the element volume, The other matrices in the stiffness are h=h= fan=9, a g = fbrda a which gives the modified volumetric strain-displacement equation The initial strain term is computed as = fa7eda De For the mixed element the internal force is computed using P, = JBI (s(u)+ p1)da a where the deviatoric part of the stress is computed from the displacement form, and the pressure is computed from the mixed form P= 2 (CO + habsIv’— =) .-46- SECTION 7. SHAPE FUNCTION REPRESENTATIONS FOR BILINEAR ELEMENT 7.1 Conventional Representation The shape functions for the bilinear quadrilateral isoparametric element are given by NE) = W144 be) 7.) Using these shape functions, the derivatives with respect to the natural coordinates are computed to be oNr Be, = MHO+aE) (7.22) and ON; Fe 7 hada) (7.2b) Using the shape functions, the interpolation for the global Cartesian coordinates may be expressed in each element as x = N;(g)x! (7.8) where x! are the values of coordinates at the nodes of the element and the repeated index J implies summation over the 4 nodes describing the quadrilateral element. ‘The derivatives of the shape functions with respect to the global coordinates, x, are computed using the chain rule. Accordingly, a ee aa which may be written in direct (matrix) notation as VpNr = VeNrS 2) ‘When solved for the derivatives with respect to the global coordinates we obtain VaN1 = VyN1S* (78) In the above V.Nr = (= x (7.78) = (2M. a VNr = [ WU | (7.7)-47- ag) =| 3 oe (116) ‘Using the shape functions (7.1) for the 4-node element, the terms in J have the structure ax, 4 "3 WE ded al eee (78) nr B = mod(a,2)+1 The constant part of J is evaluated at the point ¢ = 0 (commonly named the element center), and is given by any one aM = Yel a9) thus Th) = 2] Sat el (7.10) ae, ia describe the derivatives of the oe at the element center. We denote the jacobian at the center as Jo, that is = J(0) (7.11) ‘The global derivatives of the shape functions at the element center become V,N}(0) = V,Ny(0)So" (7.12) In subsequent developments we use the notation ba = a le (7.13) to denote the derivatives of the shape functions at the element center. In subsequent descriptions we will define Aduag = VB at Leh = ME ata = Ads 714) 12, Note that mod(d,j)=i=(5)* where j {is evaluated in integer arithmetic. Thus, mod(1,2) ‘and mod (3,2) are both evaluated be 1, while hod (2,2) and mad(4,2) are 0,-48- which is the coefficient to the spatially varying part of the jacobian transformation. ‘That is, the jacobian determinant may be expressed as , IulE) IulE)] [So Sods}, | ATaz be Adar or Jal) Tax) Coder Woda} [Adag2 Adza $1 which in matrix notation may be written as se) = Jo tMJE (7.16) | (7.15) _ sk Ye at A Leys AS z -([é 2 (7.178) and bd, A] FS * as = [is ral jo - 1.2 Alternative Representation An alternative representation for the shape functions has been proposed by Belytschko. In the development of stabilized elements he introduced the representation NyCE) = Vy +S bali a4 TV ACE (7.18) where x; are global cartesian coordinates, we ZalNno= Wabesdeadeat) (7.19) are the values of the global coordinates at the element center, ACE) = fi b0 (7.20) and 6; and T are constant parameters associated with node I. These parameters may be evaluated by defining the shape functions at each node and using the fact that NiEs) = bu 7.21) where 5,y is the Kronecker delta function for the nodes. Evaluating the alternative shape function expression at each node gives NylEs) = by +B ba af -29)4+T ACE) (7.22)-49- Introducing the notation w=[1,1,1,1] ” h” =[1,-1,1,-1) xf = Cat, x2, a}, xf] BT = [bir ,bin bis bie] and the parameter vectors 8 = (8, 82,53, 54) YP = (1,020.0) ‘The shape functions at the nodes may be written in the matrix form 2 I= %517 + Db, (x;- 291)? +h” ft (7.23a) (7.23b) (7.23c) (7.234) (7.24a) (7.24b) (7.25) Note that the rows in the expression are associated with the J in the N; shape functions, while the columns are associated with the J where the ¢, are evaluated. The Tis a 4x4 identity matrix for the element. Using this form, the parameters 5 andT” may be easily computed. First by multiplying (from the right) by 1, we obtain Is1=6 In obtaining this result we note that 1s and xP = 4x? which gives (x-af1)?1 = 0 Finally, we note that nT1=0 Next by multiplying (again from the right) by h, we get b’h=4 Ih=h= Eatbear where"? sbeath 72, The factor} fx sometimes called an hour glse shape, and when the coordinate x, i replaced by the dplecement ute fatr Uf, defines the magnitude ofthe hourglass mode. (7.26) (7.27a) (7.27) (7.28) (7.29) (7.80) (7.31) (7.32)ae ‘Thus, the parameters for I are computed as 2 : re wth -¥ xhbj] 2 (7.33) It remains to compute the b;. 1.3 Derivatives with Alternative Formulation Using the alternative expression for the shape functions, the derivatives with respect to the global coordinates, z;, are given by aN; Oa where the by are constant over the entire element and are computed by the conventional expressions at the center of the element, The derivatives of the function h may also be computed using the chain rule and ere given by Vk = Veho? (7.35) For the specific functional expression for h, the gradient with respect to the natural coordinates is given by = bg tT & (7.34) 2 Veh = (7.36) #-[] Furthermore, the inverse for the jacobian matrix is given by Oxy _ Oxy 7_ i | % 36, = I ae an, (7.37) hs where j(£) is the determinant of the jacobian transformation matrix, J. Recall that the derivative of a global coordinate with respect to a natural coordinate has a constant and a linear part. For the specific form of the h( £) function the product of the linear part vanishes and the relationship forthe gradient simplifies to Veh = ey Vehaa! (7.38) where jo is the value of the jacobian determinant evaluated at the element center. ‘The jacobian determinant at the center of the element is computed to be Jo = Jo) Jo dex - oder Joe (7.39) We note also that the jacobian determinant at any location in the element may be expressed as HE) = Jot jb jade (7.40)Ci -51- where - ha = So Adm2- Soar bdie (7.41a) da = AT, Joon - AS Jods (7.41b) With the above definitions and : =| Ou by = [ | (7.42) the gradient of the displacement may be written as V,u = V,Nyu! = [us pyar st (7.48) ‘The structure of this representation is useful knowledge when we consider the construction of the enhanced part of the strains in the next section.-52- SECTION 8, ENHANCED STRAIN MIXED FINITE ELEMENT METHODS 8.1 Hu-Washizu Variational Theorem for Linear Elasticity An alternative to the mixed finite element method discussed previously is given by the enhanced strain method [8.1]. The enhanced strain method is related to earlier works which utilized incompatible displacement modes; however, the method does not have the deficiencies which are present in the earlier works, Enhanced strains provide great flexibility in designing accurate finite element models for prob- Jems which have constraints or other similar types of difficulties. In the enhanced strain method we again use the Hu-Washizu variational principle, which we recall may be written for linear elasticity as Iu,o,e) = pf e™Deda-fe™De'da+ fo(Vu-e)d2 —B.1) a a a -Jutb, da- fuTtar- [t?(u-a)dr = Stationary a Re h ‘The strain tensor is expressed as an additive sum of the symmetric gradient of the displacement vector, Vu, and the enhanced strains, é, and written as e(u,2) = Vus aE) (8.2) If we again ignore the integral for the body force and the traction and displacement. boundary integrals, upon use of (8.2) the remaining terms become Tu,o,8) = 4 [(VuseF D(Vus sda (8.3) a - [(vuse)Deda-fo™eda a a Introducing the variations for each function as, a, = +70 (8.42) o, = o+78 (8.4b) and &, = é+nB (8.40) the variation for the above enhanced principle is given by 2. [evourp(veas e-e°)d0 (85) an) + [8 w(veuse-e*) -olaa- fs" eda a a-53- After integration by parts of the variation of the displacement gradient term (and (also considering the body force term), the following Euler equations are obtained for the domain 0: (a) Balance of momentum div[D(Vu+é-e°)]+b, = 0 (8.6) (b) Strain-displacement equations on the enhanced modes é=0 (8.7) (©) Constitutive equations D(Vutz-2°)-o = 0 (8.8) In addition the boundary conditions for I, and T, are obtained. We note that (8.7) implies that, at the solution, the enhanced strains must vanish. Substitution of this result into the remaining equations yields the appropriate displacement equations of equilibrium and constitutive equation for linear elasticity, from (8.6) and (8.8), respectively. While the enhanced strains vanish pointwise at a solution, in an approximate scheme based upon the enhanced strain method this is not the case. The enhanced strains will only vanish in some integral sense over each element, just as the balance of momentum and constitutive equations are approximated by finite element solutions. w In the enhanced strain principle, displacements appear up to first derivatives, while the stresses and enhanced strains appear without any derivatives. Accord- ingly, the continuity conditions we may use in finite element approximations again are C° for the displacements and C™! for the stresses and enhanced strains. Appro- priate interpolations for displacements and stresses are the same as given previously for each element, and are thus ug) = Ny(E)ul(e) (8.9) and OE) = GCF) 0%) (8.10) reapectively. The choice of appropriate approximating functions ¢, will be affected by the strain approximation, as will be shown below. From (8.2), the strain approximations are now given by e(u,é) = Vut ae) (8.1) where the approximations for the enhanced strains are assumed as HE) = vole rele) (8.12) ~ It should be noted that different aan are introduced for the stress and the & enhanced strain terms.-54- Using the interpolations described above, the Hu-Washizu variational theorem may be approximated by summing the integrals over each element. Accordingly, Mu,o,e) = Iu,o,e) = D1u,o,e) (8.13) ‘The variational expression in each element becomes 1.(u,0,8) = } [(Vus 2 D(Vus 2)da (8.14) a ~ [cvuse?Derda- foeda a a, Substituting the approximations for displacements, stresses, and enhanced strains and replacing with wew +n! (8.15a) of = of +98 (8.15) and & = +n (8.150) gives the first variation in each element as Ky fr 0 [w) [pe wy? By SF | fy Hes Q2, || * |-| 22 (8.16) 0 Q, 0 jlo?] Lo where Ky = J BIDB,dQ (8.178) a fy = Jv pB,do (8.17) a Bia, = JviDdy,da (8.17c) a Qu = fet v,an (8.174) a Pos [BrDperda (8.17e) a B= futDeda 8.17 a ~-55- The discrete Euler equation generated by the third equation of (8.16) is given in each element by Qupe? = 0 (8.18) There are at least three possible ways this may be used: (a) The e* are zero, which is not a useful result. (b) The interpolations for ¢, are orthogonal to the interpolations y, which means that Q.5 = 0 (8.19) which is the solution to be followed here. This is not perfect since we will not obtain a method to compute the o* directly from the variational formulation. (c) Acombination of options (a) and (b). For a formulation which satisfies (8.19), the variational equations in each element reduce to fe He) Since the interpolations for the enhanced strains are assumed for each element independently, the second of (8.20) may be solved at the element level giving Pe ie9*[ -By w] (8.21) which may be substituted into the first equation to give d, . 50 | (UY Kyu’ -B; (8.22) where Ry = Ky-19, (Bap Tas (8.23) and By = PP- £57 (H,,)7 #2 (8.24) 8.2 Stresses in the Enhanced Method Since the stresses based upon the mixed approximation are no longer available an alternative is needed for computations. Simo and Rifai suggest using a least square projection technique to obtain the stresses; however, the stresses which are directly utilized in the variational equation (8.5) may be deduced as & = D(Vn+e-e) (8.25)-56- In subsequent development we shall use these stresses for all calculations of arrays, as well as, for outputs and stress projections to nodes. Thus, the variation in each element may be written a, dn ce It is noted that the orthogonality condition fs"edn =0 (8.27) 8 = fivouyreaas [Beaa (8.26) a a ‘has been incorporated in the above variation. With the above description, the residual in each element becomes: R, = F,- [Breda (8.28) a for the contribution in the element to the global residual. Similarly, the residual for the enhanced modes is computed from R=- viedo (8.29) a Note that the residual for the enhanced modes will vanish at a solution since it belongs to a single element. 8.3 Construction of Enhanced Modes The construction of the enhanced modes depends crucially on the orthogonality requirement being satisfied for each element. Based upon the study of the shape functions using the alternative representation we recall that the gradient of the displacement involves a constant part and a part which depends only on the determinant of the jacobian matrix, j(¢), the constant part of the jacobian matrix, Jo, and gradients of local coordinates, £. Accordingly, it is useful to express the enhanced strains in a similar form. Using tensor notation we introduce the representations Jo P 1 = FwH™ E(E)Tp) (8.30) which represents a transformation of the local enhanced strains, E, expressed on the bi-unit square to the global strains, 2, using the transformation defined at the element center. The weighting by the jacobian determinant terms is motivated by the gradient of the shape functions. Similarly, a transformation of the local stresses, Z, on the bi-unit square element to the global stresses, 0, is given by o = JME) : (8.31) ~-57- ‘These transformations have the property that : _ do a tr(oé) = Fee BY (8.32) ‘The transformations may also be written in matrix form as, P| €= FHP RE) (8.33) o= (8.34) where for 2-dimensional problems Ey £= = (8.358) 2B» Zn Lag re (8.35b) Xsa Zi c and and o have similar ordering. The matrix Fy is given by Bu Jonson 0 2WJoudow Todor 22 O 2TorJor2 F ; oe 7 (8.36) Joudon Soro 0 (JorJo22+ Jordo21). In matrix form (8.32) may be written as rz = 05st oe = ve (8.37) ‘The integral over the element becomes foteaa = jo [27Edo = 0 (8.38) a o ‘Thus, the satisfaction of the orthogonality condition may be accomplished by constructing the interpolations in the natural coordinate system and transforming to the global frame using (8.33) and (8.34), A number of alternatives are discussed in the paper by Simo and Rifai. Here we consider the simplest form, which indeed is identical to the modified incompatible mode formulation [8.2]. It should be noted however, that no ad-hoc assumptions are required in the enhanced formulation, con- (Bary to whatis necessary when using incompatible modes.-58- For the simplest form, the interpolations 10 0 os): 010 off Z=19 9 ane : (8.39) 0 0 0 alls for the stress and $0 0 OTE =| & 0 O|| E= 0 0 0 ole (8.40) 00 & &]LE for the enhanced strains are used. The integrals of natural coordinates over the bilinear (2-dimensional) element obey the following properties 0 =; podd Jeo = (8.41) a pel > peven Thus, the interpolations given by (8.39) and (8.40) contain only linear polynomials in and, thus, satisfy the orthogonality condition (8.98). These interpolations have been incorporated into the element routine elmt04 which has been developed for a linear elastic-viscoelastic material, as well as, for non-linear materials. 8.1. J.C. Simo and M.S. Rifai, “A class of mixed assumed strain methods and the method of incompatible modes,” Int. J. Num. Meth. Engr. 29, pp 1595-1638, 1990. 8.2. R.L, Taylor, P.J. Beresford, and E.L. Wilson, "A Non-conforming element for Stress Analysis," Int. J. Num. Meth. Engr., 10, pp 1211-1219, 1976.-59- SECTION 9. ENHANCED STRAIN METHODS; NON-LINEAR MATERIALS 9.1 Non-Linear Elasticity. For a non-linear, hyperelastic material the stresses are computed — a strain energy density function, W (¢), through on W on de ‘The partial derivative is understood in terms of components, where oy = (9.2) i ‘We note that for the linear material model discussed previously that We) = de7De-e" De® (9.3) For the enhanced formulation the computation of stresses is given by ow (94) - ate In subsequent development we shall use these stresses for all calculations of arrays, as well as, for outputs and stress projections to nodes. Thus, for the enhanced formulation the variation in each element may be written as (see equations (8.26) to (8.29) of previous notes on enhanced strain) Ea = fcvouy eda+ [Beda (5) ts 4, In a manner identical to the linear elastic material, the residual in each element becomes: R, = F,- [B7ada (9.6) oo Similarly, the residual for the enhanced modes is computed from R, =- [veda -0 (9.7) 4, We note above that at a solution the residual, R,, should vanish independently in each element.-60- 9.2 Solution Strategy - Newton's Method The solution to a non-linear problem is commonly computed using a sequence of linear approximations. A popular scheme is Newton's method, which may be summarized a.) Given the set of equations : f(x) =0 (9.8a) where x are the dependent variables. b.) Construct the linear part of f about a current point x” as =0 (9.8b) where dx” is an increment of x. c.) Solve the linear problem dx) = ~(FOy1g) 5 PO (9.8¢) and update the solution as xD = x4 dx .8a) In the above, F®, is the Jacobian or tangent matrix for the equations. Repeat steps b.) and c.) until the solution converges to within a tolerance, tol. Convergence may be assessed from Idx) 1 < tol ix? | (9.8e) where | x | is the length of the vector, x. ‘Using Newton's method on the eet of equations defined by (9.6) and (9.7) above gives the problem Re] ree] [eg 27 awe] _ pao? A we] [Ro [ | a Lae |" | 0 . In the above, the terms in the Jacobian are defined as o a. RQ = = (9.10) which expands to : a6|° ae fuio? KQ = [B=] — da = (B,D, B, dQ 11) J de| Sut Hi-61- where a) ai) _ a6 Dee ag (9.12) define the tangent moduli for the material. For the non-linear elastic material o ao &W. Y= Seog (9.13) Similarly, Ju? BB, da (@.14a) a and . AY, = [v2 DP y,do (9.14b) a Since the second equation in (9.9) is complete at the element level, we may perform a partial solution by static condensation, Accordingly, a = (Hy RO £9, dw] (9.15) which may be substituted into the first equation to give RO? = ROR awn (9.16) where RP = RO RO (9.17a) and | RO = KO Are, (9.17) . The reduced first equations may be assembled into the global equations. Thus after adding any nodal forces, F;, the assembled equations become TK dul = SRP + Fy (9.18) which may be solved for the incremental nodal displacements, du“, After the solve, the new nodal displacements are updated WD Sy 4 duty (9.19) ‘The incremental displacements also may be substituted back into (9.15) to compute the increments to the enhanced modes; these may then be used for the update-62- POD = ZB 4 ggnin (9.20) It should be noted that these last steps may not be pérformed until after the element arrays are assembled and the resulting global problem is solved for the incremental nodal displacements. Consequently, for this algorithm, it is necessary to save the arrays used in (9.15) for the later update of the enhanced modes. In elmt04, the enhanced element for 2-dimensional plane strain applications, the arrays are moved into history arrays using a pmove routine. This requires additional storage for the enhanced formulation with respect to that needed for a displacement or a mixed B-bar type of formulation. It is possible to modify the above algorithm such that the additional storage is reduced to saving only the current values of the enhanced mode parameters, £”), The alternate algorithm is given by linearizing the residual, ft,, with respect to # only. Accordingly, with u” known we enter each element calculation with the enhanced strain parameters at the values "and perform the following steps. 1. Fork=Oset ae gi = gr) etc : (9.212) where a single superscript i denotes the value of £ computed in the last global iteration. yee 2. Compute the linear part of ft, as oe Rew, 200) HP genn gy” * (9.21b) where now \ i (9.210) \n with a (9.214) 3. Solve for the increment aetinn . (AY aR 216) 4. Update the solution grb) = gH, qghiaen 9.21 5. Set k — k+1 and repeat Steps 2, to 4, until convergence achieved (or a set number of k-steps is completed). 6 Set as ghtbel) (9.21g) i)w -63- and save for the next global iteration, as well as use for subsequent steps for the global i-iterations or to compute stresses. ‘The only information to be stored is the 2”, The algorithm requires repeated computation of R“™ and H¢}); however, using only 2 or 3 iterations generally suffices (even though convergence may not be achieved for the first few values of the i-global iterations). Once the -iteration is completed, linearization with respect to both u’ and é” is performed, leading to (9.16) to (9.18) for the global steps. If the & iteration is converged, the R® is zero in (9.16) to (9.18) thus simplifying slightly the steps involved. : While the above process has been illustrated for the non-linear elastic material, it may be directly extended to any material for which we can iteratively compute the stresses, 6, and the tangent moduli, D{’. In subsequent presentations we shall discuss the construction of these steps for linear viscoelastic materials, elasto- plastic materials, and a class of viscoplastic materials. In FEAP, the vector of nodal displacements, u”, together with the vectors Au” and du”, are retained in global arrays. The global arrays are passed to each element in a local array, ul (nd£,nen, i). The definitions of the vectors is as follows: Vector Definition Description Current solution value at each node, J. bu! =u w(t, ) Difference between current and previous solution dy = yy Increment from last iteration ‘The array ul contains information for the current element according to the following: Array ul (ndf,nen,1) Description Tocal uw ul(ndf,nen,2) | local Au%® All ul(ndf,nen,3) | local dw’ All ul(ndf,nen,4) | local a ‘Transient ul(nd£,nen,5) | local a” ‘Transient ul(ndf,nen,6) | usedforb.c.onu%™ | Alle? 19.4. This part ofthe array should not be used in elements or changed while in an element.-64- SECTION 10. LINEAR VISCOELASTICITY 10.1 Isotropic Model The representation of a constitutive equation for linear viscoelasticity may be in the form of either a differential equation or an integral equation form. In the discussion to be presented here we assume the material is linear and isotropic. Accordingly, the stress and strain may be split as o=s+lp (10.1) and 1 eretsle (10.2) where o is the Cauchy stress, s is the stress deviator, and p is the mean (pressure) stress defined in matrix form as p= gr oe (10.3) « is strain, e is the strain deviator, and @ is the volume change defined in matrix form as o=1"e (10.4) In the presentation given here we assume that the pressure-volume parts of the behavior are governed by a linear elastic model p=Ke (10.5) where K is the bulk elastic modulus defined in terms of Young’s modulus and Pois- son’s ratio as E © 3-20) ‘The deviatoric parts are assumed to satisfy a linear viscoelastic model. Linear viscoelastic behavior may be stated in the form of differential equation models or in the form of integral equations. In the differential equation model the constitutive equation may be written as K (10.6) P(s) = 24Q(e) ao.7 where P and Q are differential operators expressed as a om P = Pm Smt Puget t °° + Po (20.82) yn ” Q= ange tea gert 6 +%0 (10.8b)8 5 * 30+¥) is identical to the elastic shear modulus. Alternatively, the operator may be written as G (10.9) x 8 = 26 (wer Ema’) (10.108) dsig=e (10.10b) 4 ‘This form of the representation is equivalent to a generalized Maxwell model (a set of Maxwell models in parallel). The set of first order differential equations may be integrated for specified strains, e. The integral for each term is given by the homo- geneous differential equation solution, ai, a(t) = Ce (10.11) and variation of parameters on C to give t a(t) = feerme(rdr (20.12) An advantage to the differential equation form is that it may be easily extended to include aging effects by making the parameters time dependent. ‘An alternative to the linear viscoelastic model in differential form is to use an integral equation form. The integral form equivalent to the above is expressed in terms of the relaxation modulus function. The relaxation modulus function is defined in terms of an idealized experiment in which, at time labeled zero (t= 0), a specimen is subjected to a constant strain, eo, and the stress response, s(t), is mea- sured. For a linear material a unique relation is obtained which is independent of the magnitude of the applied strain. This relation may be written as s(t) = 2G(t)eq (10.13) where G(t) is defined as the shear relaxation modulus function. Using linearity and superposition for an arbitrary state of strain yields an integral equation specified as ‘ s(t) = fG(t-r)e(r)dr (10.14) It is noted that the above form is a generalization of the Maxwell material. Indeed the integral equation form may be defined as a generalized Maxwell model by assuming the shear relaxation modulus function in the Prony series form-66- » G(t)= Got B Ge (20.15) or the alternate form x G(t) = Got Due) (0.168) a where x Hot Dy = 1 (10.16b) ai With this form the integral equation form is identical to the differential equation model for the generalized Maxwell material. In the subsequent discussion we will consider the generalized Maxwell material and let N be 1 (.., the standard linear solid). The addition of more terms may be easily accommodated based upon the one term representation. Accordingly, G(t) = Got me) (10.174) where Hotwy = 1 (10.17) In applications involving a linear viscoelastic model, it is usually assumed that the material is undisturbed until a time identified as zero. At time zero a strain may be suddenly applied and then varied over subsequent time. The integral representation for the the model may be simplified by dividing the integral into ' « a t Joodr = Joordes foodes frar (20.18) ~ — e ® ‘The first term is zero, the second term includes the jump term associated with e at time zero, and the last term covers the subsequent history of strain. The result of this separation when applied to (10.14) gives ‘ 8(t) = 2G(t)e+2 | G(t-1)8(r)de (10.19) 3 where subsequently the 0 limit on the integral is understood as 0*. Substitution of (10.17) into (10.19) gives t s(t) = 2GLupe(t)+ mre (ey fe eC )de)] (10.20) 3 It remains to evaluate the remaining integral. Accordingly, we divide the integral a-67- ‘ ty ‘ foods = [Odes foods 0.21) 2 2 fa Ifwe define the integral as ‘ H(t) = femecrrdr (10.22) a the above separation gives : ‘ H(t) = Ct) 4 feMe(sdde (10.23) i ‘The strain rate is now approximated as constant over each time increment ¢, to t, thus a(t) = e(t)-e, | See | ae (20.24) where e, denotes the value of the strain at time ¢, and At denotes the time increment t-t,. With this approximation, the integral may be evaluated exactly [10.1]. ‘Anumerical approximation may also be employed and one proposal uses a midpoint (one-point) approximation for the integral as [10.2] t fede = tent (10.25) i where t,,, denotes the time }(t,+1). An alternative is to use an exact integral Thus, the integral may be approximated as i(t) = B+ es (e-e,) (10.26) where i} denotes i'(t,). A more efficient and stable form of the process is to define a recursion formula which includes the negative exponential multiplier term. Accordingly, we define hy = etna (20.27) and then Yt) = eit) = eM} + et” (e-e,)] (10.28) The above may be simplified to bYt) = et [edn +(e-e,)] (10.29) ‘This form requires only one evaluation of an exponential term. Furthermore, a zero value of the time step produces a correct answer, as well as, a very large value of the time step producing a zero value. Thus, the form is doubly asymptotically accurate.-68- The use of finite difference approximations on the differential equation form directly does not produce this property. = ‘The constitutive equation now has the simple form s(t) = 2GLupe(t)+ ph (t)] (10.30) The inclusion of more terms in the series reduces to evaluation of additional h'(t) integral recursions. The required storage is increased by a need to preserve the h’ for each quadrature point in the problem and each term in the series. The implementation of the viscoelastic model into a Newton solution process requires the computation of the tangent tensor. Accordingly, we need to compute SS ee (10.31) where Iz is the deviatoric operator identified previously. The partial derivative with respect to the deviatoric stress is given by as ant Fe = 24m +n ZO) (10.82) The derivative of the last term is computed from (10.29) and becomes s = ebay (10.33) Thus, the tangent tensor is given by % = 2G po + web Tey (10.84) ‘The only modification from a linear elastic material is the substitution of the factor Goixe = GL po + wy ete] (10.35) for the elastic shear modulus. Again we note that for zero At the full elastic modulus is recovered, whereas for very large increments the equilibrium modulus jy G is used. The above formulation has been incorporated into elmt 04 in the subroutine viscoe. Note the simplicity of the additional coding needed to include the linear viscoelastic formulation. Since the material is linear, use of the consistently derived tangent modulus terms leads to convergence in one iteration (the second iteration produces a zero residual). 10.LR.L, Taylor, KS. Pister and G.L. Goudreau, "Thermomechanical Analysis of ‘Viscoelastic Solids," International Journal for Numerical Methods in Engineer- ing, Vol. 2, 45-79, 1970. 10.2.LR. Herrmann and FE, Peterson, "A Numerical Procedure for Viscoelastic Stress Analysis," Proceedings 7th ICRPG Mechanical Behavior Working Group, Orlando, FL, 1968.-69- SECTION 11. PLASTICITY TYPE FORMULATIONS 11.1 Constitutive Equations for Plasticity Models ‘The constitutive equations for a material which behaves according to a plasticity type formulation for deformation states which exceed the elastic limit may be expressed by assuming that the strains are decomposed according to excite? an where e* are the elastic strains and ¢? are the inelastic strains. If the material is non-linear hyper-elastic we may deduce the stress from the expression for the elastic strain energy as o-@ a2) td ae where W is the strain energy density and is expressed as a function of the elastic drains and o and e* are stress and strain energy conjugates. For a linear hyper- clastic material the stress to elastic strain relation is given by o = Det = D(e-e?-2") 1.3) In the following discussion we limit our comments to linear elastic materials and ‘also set £° zero. The inelastic component of the strain rate is related to the gradient of'a loading function with respect to stress. Accordingly, ee it foe 4) where f is a loading function and 7 is a sealar rate term eglled the plastic consistency parameter. The plastic consistency parameter, 7, is zero for elastic behavior fand positive for plastic behavior. A back stress is defined as a which is related to the plastic strain rate through a = 2 Hine = er as) where Hy, is a kinematic hardening modulus. The yield surface is defined in an Tesodiative manner, using the same function as the loading function, and is expressed as ¢ flo,a,é) = F(E)-¥(e) (se) where the stress and back stress appear in the form Eso-a an and Y(@) = Yo+ Hime? (1.8) ee-10- is a function which measures the size of the current yield surface. Commonly, Yo is related to g,, the yield stress in uniaxial tension. The isotropic hardening bebavior of the material is included in ¥ through an effective or accumulated plastic strain defined by t(2 es jee }a a9) In (1.8) Hig is an isotropic hardening modulus. In the present study both the isotropic and the kinematic hardening moduli will be assumed as constants. Using the definition of the plastic strain rate the effective plastic strain may also be writ ten as ‘ 2 af af P= a é fG oe (1.10) thus it is evident that é? is dependent on the integral of 7 and the particular loading/yield function used to describe the material. Generally, the model is completed by describing a scalar function, g, which describes the limit behavior of the model. Different limit equations may be written for rate independent plasticity, rate dependent plasticity, and generalized plasticity models, The simplest relation is for classical, associative, rate independent plasticity where : g=flo,a,e)s0 (ip is used. Later alternative forms will be introduced to represent other types of mate- tial behavior. 11.2 Solution Algorithm for the Constitutive Equations ‘The solution of the above set of equations may be effected numerically using a variety of algorithms. A very effective method to integrate plasticity equations is the operator split method with a return map concept [11.1,11.2]. the algorithm may be extended to include a variety of modeling concepts for the limit behavior; however, for the present we restrict our attention to plasticity as defined by (1.11) above. Accordingly, a discrete solution at time t, is defined in terms of the state o,, dy, £2, and é?. The solution is then advanced to time fn»: by specifying the strain, éy41- ‘To initiate the solution at ft, a trial state is computed assuming the step is entirely elastic. Recall that a step is elastic when 7 is zero. This implies that there will be no change to ”, a, or é” during an elastic increment. The step is initiated by taking the trial values for plastic quantities efTR = oP (11.12a) ath = Ge (11.12)-11- : enn = é8 (11.126) and ma = 0 (11.124) ‘Thus for linear elasticity oR = Dement) (11.18) ‘The trial stress given by (11.13) is checked in (1.6) and (1.11) to determine if the step is elastic or whether inelastic terms should be active. If the state at t,,: is elastic the stresses (as well as other state variables) are set equal to the trial value; oth- erwise, a correction is required to include the inelastic terms. For an inelastic step the stresses must satisfy (1.2) for the time f,.; which requires the rate equations for ¢? and a to be integrated over the time increment. Accordingly, integrating non-linear terms using a backward Euler implicit method between f, and t,41, the plastic strain is given by at 3 eh = ER + Ane i 1.14) ea nel and the back stress by 2 4) aby = a8 + Bain dent 5 (11.15) ae | 4 where H;, is a constant kinematic hardening parameter and the integral of the consistency parameter is given by Inet = Tova 1.16) Similarly, evaluating (1.11) at tas gives of Bnet = for = 0 ate BOG air Finally, integration of (14) produces y' ha = Rt anne oh 3h (11.18) a ‘The set of equations (1'3), (11.15) and (11.17) constitute a set of non-linear equations in terms of the values of oys1» @o1 and Aye Which must be solved for each stress point and each time step of interest. A Newton method may be used to solve the equations. To simplify the notation the subscripts on n+1 are omitted. The iteration counter is shown as a superscript (j) and initial iterate values are taken as the trial stress and zero 4, The iterative solution is continued until the norm the . residuals are within acceptable tolerance values (e.g., normally, half machine precision relative to the initial iterate values are used since Newton's method then-72- guarantees that machine precision is achieved if the next iteration is checked). Before proceeding with Newton’s method we note that the following relations hold af _ af ax _ af 2o ~ E30 DE (11.19a) af _ af ae _ af o- oe --F (11.196) ‘Thus, treating the ar and (11.17) as residual equations, in the form RY = e-e2 -D1G9-2 2 (11.20a) RY = a+ Arind of -a? (11.206) and RP = -f(o%,a eX) (11.20¢) we may linearize the equations to obtain (note the iteration counter j is omitted in the coefficient array for simplicity) é ae 2 at) ~arae, a dot) TRY oe 1+ Bind f —Biin( ' a dad | =| T? (11.21) a Et aa en ) di RY ote ee 7 v ax? az where oY ae = 56 Or (11.22a) and 1 ef a =D’ Ear (11.22b) The solutions to (11.21) are computed and added to obtain the next iterates. oF) = G+ do (11.288) at) = a+ dad? (11.23b) AFD 5 a4 aged (11.28¢) ~-13- define the next iterates. The solution is terminated whenever the norms of the residuals are smaller than a selected small tolerance. Once convergence is achieved for each stress point evaluation (j.e., to compute the stress at each Gauss point for a given strain), the stresses may be used in the finite element to compute each element residual, In addition it is necessary to compute the tangent moduli, D,, for use in the element stiffhess matrix (if one is used) for the next iteration on the momentum balance equation. That is we need to perform a new solution to see if the strains we used to compute the stresses are correct, This is accomplished, as before, by solving DKyw’ = F,+ DR, (11.24) where Ky, and R; are the element stiffness and residual, respectively. The computation of the tangent moduli may be obtained by noting that the computation of the last stress increment in the Newton solution of (11.21) may be written as 4 vr a, af doi? . . Re, a jam RY [se Hwa ee T+ Hm oe A Za [Eh aa ss ot ° RY ant x At convergence for the given strain, e, the residuals will vanish; however, if we now consider a linearization with respect to strain only R, contributes to the change. ‘The linearization of the residuals with respect to an increment of strain yields R,=de ; R,=0 ; Rp =0 (11.26) Denoting the inverse matrix as vf a, Di Dy D A ayaE, gee Du Dy Dis xf oF eo cof, Bn Ba Bas |=|-Bandssde I+ Hand seag ~ Han (gy (455 127 Bn be Bs ar ar = ar “oF ‘The final result for the linearization with respect to strain gives doF?] [Dude da | =| Dade (11.28) ax?) [Dade thus, the tangent moduli for the next momentum iteration is D, = by (11.29)-14- Except for giving the form of f this completes the specification of the general algorithm. Multiple yield surfaces may be included by modifying (1.4) to & , ofs ray é Ge (11.80) with each part of the yield surface described by a separate function Be = fy(o,a,e) 50 (11.31) ‘An active condition for each surface is denoted when j, 20. Thus, usually only 1 or 2 of the surfaces are active at any time. As constitutive equations become complex the specification of the parameters is more difficult, A systematic procedure for determining the parameters from experi- mental data is given by Ju, et. al. °, The method provides the best estimates for the parameters and their sensit ies to errors or inconsistencies in the data. 11.8 Isotropic Plasticity - the J; Model As in previous developments, the strain is again split into deviatoric, e, and volumetric (spherical), 9, parts and expressed as e=eriie (11.82) where o=17 & (11.83) For our study on inelastic behavior, the decomposition into elastic and plastic parts may now be expressed as e+e? (11.34a) and o= 6+eP (11.34b) The stress also is split into the deviatoric, s, and pressure (spherical) parts as o=st+lp (11.35) where pe dito (11.86) With the above splits the isotropic linear elastic constitutive equations are given by p= K(@-6?) : (11.37a)-15- and s = 2G(e-e?) (11.87b) In the developments below we restrict plasticity to the deviatoric part only. Thus 6? vanishes and the yield function can depend only on the deviatoric part of the stress. For isotropic materials the yield and loading function may be expressed in terms of the invariants of stress and back stress. The invariants of s are denoted as J, Ja, and J; and given by J, = 178 =0 (11.388) Jy = 38's (11.38b) and Jy = det(s) (11.380) ‘The simplest formulation is where the function depends only on Jz. We write this model using _ V2d; = (ssh = ist (11.39) and including the back stress, the limit equation as g = f(s,a,é) = Ils-all-¥(e?) 0 (11.40) where Y is the radius of the yield function which is related to the uniaxial yield stress, o,, through Ye af to+ Hee”) _ 1.41) and, thus, includes the effects of linear isotropic hardening. The back stress adjusted value Eis given by Ees-@ 5. 74,2 (11.42) A simple calculation shows that Ca ee ae ao ~ 3s + as" as” OF (where we recall that Jag =1- 2117), and et ost ‘Thus, the evolution of the back stress satisfies-76- Noting that at the initial state « is zero, we can conclude that the back stress evolves such that it is a purely deviatoric quantity. thus, 17a =0 (11.46) With this fact we then have the following important properties : Z=0 ; nso ; nm=1 a1.47) Based upon the above all aspects of the J plasticity model are restricted to deviatoric components only. Thus, our model is completed by giving the evolution equations for plastic strain and effective plastic strain in the form =7n (11.482) P f 2, es J gid (11.48b) a ‘The discrete form of the isotropic J, model is given by the equations Pua = KOq Pt (11.498) C. = 26 (ea —eP) a (11.496) (a ant A (11.496) % One = Ont 2 Hin Ane Baa . (11.498) (has at +E ne : (11.49) Boe = WEa Ya $0 ay SE (1149p Ene = Snet ~ Get x (11.492) and Yor = Vz (oy + Hi Ba) = Yat 3 Huw Ansa (11.49) ‘The solution of the Jz model is straight forward and may be accomplished by solving only a scalar equation in a ‘The solution is performed using a trial state . based upon the assumption that 47%, is zero. Accordingly, ave =e 5 ama 5 en Hee (11.50) This yields the trial deviator stresses STB = 26 (em — eR) = 2G (ens —eF) ais) which may be used to check the limit equation, g,,:. If the limit equation is satisfied then the trial values define the solution at f,.:. If the trial values violate the limit equation, it is necessary to perform the second part of the algorithm. The-17- second part of the algorithm solves the discrete rate equations using the trial values as initial conditions. Accordingly, using (11.49c), (11.50) and (11.51) in (11.49b) gives Sper = Bye — 2G Ang Daa (11.52) Next subtracting (11.494) from (11.52) gives Znet = Bt Gnas = FR -0-2(G+ 2 Han )Aner Doar (11.63) Noting that 2,41 = II Zns1 Il m4: (11.53) gives also that s?% - a, is in the direction 2,1, and may be written as the scalar equation CN Za Il - Wa W 4 20G+ 3 Hass) anette =0 (11.54) Jeu that is, the coefficient must vanish to obtain a solution, In the above EA = Sha ~ Gn (11.55) Combining (11.54) with (11.49f) and (11.49h) yields the scalar equation, W2PEW AY, = 21642 (Hu + Hind (11.56) Once dny3 is known it may be combined with the result Be = Oe = TT (11.57) Tse to give the stress as Baer = BE 2G Ane Dna (11.58) as well as the plastic strain and the back stress. In practice 4,,; is reduced slightly that 8,,: is always slightly outside the limit yield condition. Accordingly, the solution to (11.5%) is perturbed as my C4 ee ner = REBA VACA 2 NL. Way : 21G + 5 (Huo + Hai) 2G where tol is a small value, say 10°. The solution of the problem, as shown above, does not require a linearization or an iteration process. If non-linear isotropic hardening is included or alternative forms for the limit equation are employed, the equation equivalent to (11.56) will be nonlinear in 4,,, and a linearization and iteration process are required to obtain a solution. Similarly, if non-linear kinematic hardening is introduced the reduction to a salar equation may also be complicated.-78- Once the converged value for 4,4; is known and the stress state determined, a check on satisfaction of the momentum equation must be made. If the momentum equation is not satisfied for the current time, t,,1, another iteration is necessary to improve the estimate for the state of strain, ¢,,1. If a Newton type solution method is used it is necessary to compute an appropriate tangent modulus matrix for each stress point in the analysis. For the solution process developed here, this may be achieved by selecting as primary dependent variables the stress, 8,,,, the back stress, @,,1, and the consistency parameter, 4,,:. Writing appropriate residual equations as R, = neha an=0 (11.608) R, = an a- FAHD =0 (11.601) and Ry = ¥, +3 Hyd - EI =0 (11.600) In the above we have deleted specific reference to the values at f,,1, to avoid added complexity in the linearization performed below. It should be understood that 4 denotes the value of the solution in the f,,1 step, ie., previously given as Jy,1, and etc. for all the other variables. We note that for the current strain, e, the above equations are satisfied; however, to proceed to the next iteration of the momentum equation we consider a linearization of the above equations with respect to a change in the strain also, which we a by de. eae the linearization n becomes an =o er ao (I-nn") = THN (11.61) Using this result, the linearization of (11.60) yields the set of equations soivrtn -_4 2 3G yee a : Gal) (de hin hin . STEN +a gitN -pAmnllde|=| 0] are 2 da} Lo at =n? gle ‘The inverse to the coefficient matrix may be computed by first solving the first two equation for ds and da in terms of da and de, and then substituting the result into the third equation to obtain a final expression for 4 in terms of de. This also per- mits the substitution of alternative limit equations without changing the solution to the first part. Accordingly, we consider a. a *qemN ~TaqN |[as]_| ge-naa | 24H iin 4 24H bin |] da] ~ | -2 Hyams : siz 31st 3& -19- The solution to this equation is™ ds}_|2G(I-BN) BN de-nda | -| 2GCN eel eee ee where B and C are given by 264 2G4 BHund — 2d Beqrip- Te } OC" enema” siamo where we have noted that . Hi a Mee D= 14+2(G+= Tet = as (11.65b) ‘This result may be substituted into the third equation in (11.62) to obtain 2Gn" de = 21G +3 (Huo + Hin Ida (11.66) Substituting this result back into the first of equation (11.64) yields the incremental equation which yields the tangent modulus matrix for the algorithm. Thus, we obtain ds = 2G{1- B(1-nn")- Ann? ]de (11.67) 4 where Cat athe — (11.68) G+ gO Hino + Hain) Finally, for the differential strains, de, the tangent becomes ds = 2G (lg - BUI —nn?)- Ann? }de (11.69) dds doe Pde : ads ¢t0T to &s D4 KIL 11.4, ISOTROPIC VISCOPLASTICITY - THE J; MODEL ‘The previous section presented the formulation and solution algorithm for a J Classical plasticity model. In this section we show how such a formulation may be easily extended to include rate effects in the inelastic behavior. The model selected for exposition is classical viscoplasticity as introduced by Prager for one-dimension and extended to full three-dimensional form by Perzyna [11.4]. For the viscoplastic model considered here, the only modification to the formulation is the replacement of the limit equation for g. Indeed, other models representing the problems of generalized plasticity and generalized viscoplasticity can 11.1. Bee Appendix fora discussion on the inveree of this type of matrix-80- also be developed by such replacement, In viscoplasticity, the relationship for g becomes a constitutive equation describing the evolution for the consistency parameter, 7. Accordingly, we write & = Olf(s,a,é)]-f7 <0 (11.70) where the yield condition, f, still is given by f(s,a,é") = Ils-all-Y(é?) @L7) with Y the radius of the yield function which is related to a uniaxial yield stress, o,, through ri and, thus, includes the effects of linear isotropic hardening. For viscoplasticity, ¥ and gy relate to the inelastic behavior in the limit as 7 tends to zero (the rate independent limit). For loading rates which are finite, the stress state may lie outside the yield surface. The function © together with the parameter ¢ determine the rate dependency of the model. Perzyna considers alternatives for representing ©; however, here we restrict our attention to the simple case given by O(f) =(f" (11.73) where m is a positive integer power. Other functional forms for @ may be considered without conceptual difficulty. All the other equations for the model remain as given in Section 3. For trial stress values for which the yield function defined by (11.71) exceeds zero, the behavior is inelastic and the return map solution for the viscoplastic model is given by (11.49a) to (11.49e), (11.49g) and (11.49h). The formulation is completed by integration of the constitutive equation (11.70) for the time interval t, to tnt (ie., At) using a backward Euler evaluation of the integrals to obtain AOS (Snr s @nsi»Epn I~ Sane = 0 a1.74) The discrete consistency parameter may be obtained by combining the scalar coefficient from (11.54) with (11.74) to obtain a single nonlinear equation in 2,,;. Accord- ‘ingly, we obtain Oy + Hig &?) (11.72) ALOL NE? Il ~¥n- 2G +E H+ Hata )) Ayes) 6 =0 (1178) For the simple model used here for ©, the above becomes asus " ~¥,-2164 3 lie Handa ) Edo =0 (11.76) In general, the above equation is nonlinear and must be solved numerically. For the case where m = 1 the equation is linear and has the solution-81- TR yy dg, = BECUZ MY (14 tol) a7 C+ RAG + (Huo + His) Comparing (11.77) to (11.59) we can observe that the limit solution for ¢ zero is identical to the classical plasticity problem. The stress, back stress, plastic strain, and effective plastic strain are computed using the same expressions as for the classical plasticity model. For nonzero ¢, the presence of At in (11.77) implies a rate dependency, with results depending on time durations for applying and changing loads on the body. The extension to higher powers of m may be constructed using a Newton solution scheme to solve the non-linear scalar equation. 11.1J.C. Simo and RL. Taylor, "Consistent Tangent Operators for Rate- Independent Elastoplasticity," Computer Methods in Applied Mechanics and Engineering, Vol. 48, pp. 101-118, 1985. 11.2..C. Simo and R.L. Taylor, "A Return Mapping Algorithm for Plane Stress Elastoplasticity,” Int. J. Numerical Methods in Engineering, Vol. 22, pp. 649-670, 1986. 11.3.3.W. Ju, J.C. Simo, K.S. Pister and R.L. Taylor, “A Parameter Estimation Algo- rithm for Inelastic Material Behavior,” Constitutive Equations for Engineering Materials, C. Desai, editor, January 1987. 11.4.P. Perzyna, “Fundamental Problems in Viscoplasticity", Adv. Appl. Mech., 9, 243-377, 1966.-82- Appendix: Properties for J; plasticity models The solution of the J plasticity model leads to derivatives of the yield and loading functions in the form of _ 7 (A) where = == TET and E=s-a (A.3) ‘We note that n has the properties n=0 ; n’n=1 (AA) In the derivation of the tangent the derivative of n leads to oie 2 ant) (As) daz OS TEN which appears in several location in the tangent matrices. The inversion of the tangent matrices may be simplified using the Sherman-Morrison-Woodbury formula which is described on page 51 in [11.A1]. (A+ UV")? = At-ATUWV" AT (A6) where W=(1+V'AtUyt (A.7) In the above A is an nxn matrix, U,V are nxk matrices, where & $n, and Wis a kxk matrix. The inverse may be proved by multiplying the results together to recover the identity matrix, In the case of the deviatoric model A is diagonal and U and V are proportional to m which is rank 1, thus leading to a scalar W (ie., a 1x1 matrix), ‘There are some properties which need to be noted: nn™(nn") = nn? (A.8a) (I-nn")n = 0 (Ab) and (I-nn?)(I-nn?) = I-nn? (A.8c) Example 1: Consider the matrix Hy = AI+Bnn™-83- Using the Sherman-Morrison-Woodward formula the inverse is given by noting that U is equal to Bn and Vis equal to n, thus eee we By = F1-(G (Ba) Wa" (SD) where . Byi_l_A UA ee The above simplifies to al ige v Bl a aearBee Example 2: Consider the matrix H, = CI+D(I-nn") which may be rewritten as H, = (C+D)1-Dnn™ for which the solution from crample 1 ure 1 HP = gpl gan) Recollecting into the original tea of matrices gives (I-nn)) 2 - als ou OB A slightly more general form for an inverse results in considering the case with kinematic hardening. In this case we encounter a matrix of the form _[Al+BU-nn?) — C(I-nn7) x-| D(I-nn?) E1+F(I-nn") ow ‘The inverse may be written as a _[aI+b(I-nn™) — c(I-nn") 10 = -| d(i-nn") el+f(I-nn") a where 1 1 o=Z i erg (A.1a) and the remaining coefficients obtained by solving the small matrix problem A+B C l[b c B Cl[a 0 [ D Pale f “3 rls ‘| ca‘The solution to (A.11b) is given by be ifz+F -c ][B c][a 0 (3 s]--a -D aval lls | ome G = (A+B)(E+F)-CD (A.11d) The inverse may be proved by multiplying the two matrices together and show that the result is an identity matrix. 11.A1, G.H. Golub and CF. Van Loan, Matrix Computations, 2nd ed., The Johns Hop- kins University Press, Baltimore, 1989. where= 85 - SECTION 12. AUGMENTED LAGRANGIAN FORMULATIONS 12.1. Constraint Equations - Introduction ‘The solution of many problems requires imposition of constraints as part of the formulation. For example, if it is desired to solve the incompressible equations for linear elasticity it is necessary to impose the constraint tr(e) = 17e =0 2.1) Another type of constraint is to impose boundary conditions on a node, where we wish to impose the condition for node I that w =a (12.2) in which @ denotes a specified value. This type of constraint can be made more general by letting the degrees-of-freedom be associated with a rotated local coordinate system (e.g., a spherical coordinate frame) where now a? (12.3) in which T; is an orthogonal rotation matrix which transforms the degrees-of freedom from the global Cartesian to the prime system. Many other conditions could be given; however, the above suffice for the present, The inclusion of the constraints into the finite element problem may be performed by several different approaches. For constraints of the type (12.2) it is easy to directly eliminate the variables associated with u’, as is done in FEAP. On the other hand the inclusion of (12.1) or (12.3) presents more difficulty to implement. Thus, an alternative method is needed to implement general types of constraints. A common approach is to use penalty methods; however, these are sensitive to the value of the penalty parameter selected. A better approach, which is numerically superior, is to use an augmented Lagrangian approach. This method is an extension to the penalty method and uses values for the penalty parameter which lead to a better conditioned numerical prob- Jem. In the sequel we first consider penalty methods, based upon a mixed formulation. Subsequently, we show how to extend the mixed penalty treatment to the augmented Lagrangian algorithm which is based on an iterative update procedure generally attributed to Uzawa [12.1]. ul = Tyu! = 12.2, Mixed Penalty Methods for Constraints Consider a general constraint equation expressed as, g(u) =0 (12.4) which is to be imposed for some part of the domain, Q,. The constraint may be included as part of the problem formulation by supplementing the variational problem, 7 (w), with the term . T(u,a) = faTg(u)da (12.5) a,-86- Define the variations as u+nU (42.6) and Ay = AtnA (12.6b) ‘The variation of the integral gives the added terms — fat ecurans furetaaa 27 dn 4? ned a, - where \ an a7 a (12.8) au ani ‘The Euler equation for the first integral leads to the constraint equation. g(u) =0 (12.9) for each point in Q,, and the second equation leads to a term which is combined with the variation of the original variational theorem to generate revised Euler equations for the problem. In 2 finite element matrix setting we can approximate the 4 in each element as A= Ni(x)a" ‘ (12.10) and use the usual isoparametric interpolations for u. Thus, (12.7) generates the matrix problem Pp Gen) _Tcqeyr cytyr]| Pa 2. a= & Ee [can (uty IF] (12.1) where 4 = Ef Nig(uda (12.122) “Oy and Pied J N,G7adQ (12.12b) «gS For non-linear constraint equations it is necessary to linearize this expression for combination with the remaining part of the problem. Performing the linearization leads to the problem @ ryr]| © Kay |[da*] ~ [arr wr], Ki, || du? fee-87- where Ki, == J N,G™ N}dQ (12.14a) . a K,=d J NiGNydQ = (Ki, )* (12.14b) "Oy Ki == Jurat 28 wyan -EfuatSn,do cane a ‘a. ‘The difficulty with the above formulation lies in the fact that there are no terms in (12.13) which are associated with the diagonals for the 4 degrees-of-freedom, More- over, if the constraints are linear there are no terms on the diagonals for any of the degrees-of-freedom. This greatly, complicates a solution process since for a direct solution the equations must be ordered to eliminate the displacement equations prior to the Lagrange multiplier equations. Furthermore, iterative methods are even more difficult to consider. The deficiency associated with the diagonals for the Lagrange multiplier equations may be removed by adding a regularization term to (12.5). The modification to the variational term considered takes the form I.(a,a) = J ecmaa- fF ataaa (12.15) a a, where k is a penalty parameter introduced such that in the limit as & goes to infinity the original problem is recovered. = far(gcn)-Zardas [U" Gada 12.16) a, Oy ‘The Euler equation for the first integral now gives the constraint equation. (12.17) for each point in Q,. It is evident that the solution converges to satisfy the constraint only in the limit when & is infinity. The linearization of (12.16) gives the matrix problem oo er cyt ]| Kee Kes || €a? [cary wor] ea (12.18) where Bi, = E J MitINjdo 12.19) ed := 88 - Many cases for constraints permit the elimination of the equations for 4° at a local level. Thus, if a Newton solution scheme is employed the residual equations may be written as Ri a[—Ps+Keed"] _[-Ki, Kes |[ da] _[O]0 09, R,(u)+R}|~ | Re(u)-P} || Kip Ky +Ki, |] du7|~ joy” 2 ‘This gives the set of equations to solve for the increment as -Ki, Key |[ aa’) _[-Po+Ki,a? [x wart ||| : [ roe | tae Solving the first row of (12.21) gives da? = (Ki, "(Pi +Kisdu’]- 4° (12.22a) Since the residual equation for 4” is linear it may be solved to give af = (Ki,)? PL (12.22b) and this simplifies (12.22a) to da? = (Bi, {Ki du’) (12.226) which when substituted into the second of (12.21) once again yields a displacement model for the problem which is expressed as Kj du! = Ry(u)-P} (12.28) where 2 A Ky + BG +Kip(RGy"Kly > a2 The above solution process defines a perturbed Lagrangian form of the penalty solution process. In order to yield a solution which provides an adequate satisfaction of the constraint equation, fairly large values for the penalty parameter should be used (generally on the order of about half machine precision, e.g., 10° or 10"). The values used then yield stiffhess modifications for the second term on the right hand side of (12.24) which are several orders larger than components appearing in the stiffness, Kyy. If the values are too large, ill conditioning for the solution to the linear equations will result; if too small, the constraint may be violated by an unac- ceptable amount. Furthermore, iterative solutions become very difficult for these large penalty values. Consequently, an alternative approach is needed. In the next section, the augmented Lagrangian method is introduced as an alternative. 12,8, Augmented Lagrangian Method for Constraints ‘The augmented Lagrangian strategy presented is a simple modification to the perturbed Lagrangian form which now becomes-89- 1.444) = [A+ ag g(a) d2- [pataao (12.25) a De where 2, is the augmented term. The variation to (12.25) gives Ha} = fat(g(u)-Layda+ [UG"(a+a,dQ 12.26) dn z n=O Qe a, + fatg(uaa a ‘The Euler equation for the variation of 4 gives the equation gcuy-Za =0 (12.27) which may be used to compute 4. The variation for 24 gives the constraint equation g(u) =0 (12.28) and, thus, the constraint equation is satisfied independently of the value of the penalty parameter, k, and we also conclude that 4 must vanish at the solution. Using, these facts we also note that the algorithm merely reduces to the original Lagrange multiplier method, but with 2, used as the multiplier. The method may be made computationally viable by making the determination of 44 an iterative algorithm. ‘The Uzawa algorithm is the simplest algorithm which may be considered. In the Uzawa algorithm we introduce an outer iteration loop for the augmentation. For each step in the analysis we assume: 1. Let j be the augmentation iteration counter. For each time, tn.1, set j to zero and take the initial value of the augmented multiplier as AK? = aK(tn) where the dependence on the n+1 step on the left side is implied. Let a4(0)=0. 2. Solve the non-linear problem Ky da) _[ -Pi+Ki,a’ |_ 0 Kj, Ky+Kiy|[ du’ | [Ri(u)-(Pat] ” [Rr(u)-(Pad where ay wa? = (Bag Y*Pe $ Wve a omabt) hlekr) eh (obi) ae aee(Palh = E J NGA? + af Jan : Pee eee vt bot 1 \ in rey 2-90- In the above the iteration aspects for the incremental problem are not shown. 8. After the iteration for the incremental problem converges update the augmented parameters using 4 #9 ARID ABO 4 gh where af = (Ki,)7 PA is computed using the converged solution from step 2. 4. Check convergence for the augmented step. If the constraint is satisfied to within a specified tolerance, or the change in the 4” is less than some tolerance times 24*” proceed to the next time and repeat steps 1 to 3. Ifnot converged increase the j counter and repeat steps 2 and 3. ‘To perform the above algorithm it is necessary for the penalty parameter k to be large enough for the iteration to converge. All that is required is that the terms in the added stiffness be somewhat larger than the original stiffness terms. The convergence rate for the augmented iteration is generally linear, not quadratic as in a Newton solution. The larger k is made the more rapid the convergence. Thus, it is desirable for the value to be at least one or two orders in magnitude larger than the conventional stiffness terms (as compared with the six or seven used in a penalty approach). Use of values with this range in magnitudes leads to 3-6 augmented steps for most problems. The number of non-linear iterations will decrease for the later augmented steps since the violation in the constraint is becoming less and le Using the above augmented Lagrangian approach to satisfy the incompressibil- ity constraint leads to a particularly simple update. For the constant pressure/volume element there is only one equation for each element. Thus the equations to be solved are scalar. For the enhanced element there is one equation at each Gauss point 50 it is also easy to modify. For more complex situations, involving multi-point constraints, the situation is slightly more complex, Augmented approaches have been used to solve a variety of problems in finite element methods. In some cases (for example, frictional contact problems) it is possible to augment in a way which renders a problem which originally has an un-symmetric tangent matrices to one which is symmetric. In general, the method is the one of current choice since, as a special case, it also includes an option of penalty solution through the perturbed Lagrangian approach (merely omit all augmented steps!). 12.1K.J. Arrow, L. Hurwiez and H. Uzawa, Studies in Non-Linear Programming, Stanford University Press, 1958. (see also pp. 357 ff, Zienkiewize and Taylor, ‘Vol 1. 4th ed., McGraw-Hill, London, 1989).-91- SECTION 13. TRANSIENT ANALYSIS OF NON-LINEAR PROBLEMS 13.1, Adding the transient terms ‘The variational equation for a quasi-static problem solved by the finite element method is expressed as | = Duly [Bfede-[wyb,da-[Njtde=0 a3.) Theo a a n where & is computed for a displacement, mixed, or enhanced method as described in previous sections of the notes. In order to extend the variational equation to accom- modate transient analysis, the body force vector, b,, is replaced by be by- pa (13.2) in which ii is the acceleration vector. With this replacement the variational equation becomes ay dq = yu ye [Breda+ JNrpada- Jrrb,da- frrtany (18.3) nso f a, a a t which leads to the residual equation for each node R, = F;-5 [BPeda-y [N;pada } numerical dissipation is introduced. The values of # control primarily the stability but also influence the form of the matrix problem. A f of zero leads to a-93- formulation which is called explicit, where for no damping, the solution for the acceleration, a,,1, involves only the mass matrix. For a diagonal mass this solution step is very efficient; however, in general the method is only conditionally stable and very small time steps are needed. For f non-zero, the method is implicit and a solution step normally involves linearization of the momentum equation and an iterative solution process based on Newton’s method. The advantage of implicit solutions is improved stability, where quite large time steps may usually be taken. For example, for #=0.25, the method for linear problems is unconditionally stable. This method is commonly called trapezoidal rule or constant average acceleration. ‘Values of f less than 0.25 should not be used since they are only conditionally stable with allowable time steps not much larger than the explicit scheme. ‘The advancement of a solution from one step to the next is completed by combining (18.15) with the momentum equation written at time t,,.. Accordingly, Raa = Fra -P(Gn)-Mana = 0 (43.16) In order to advance the solution to the next time it is necessary to recast the prob- Jem in an iterative form. This involves selecting appropriate values for the variables to initiate the step, linearization of the equations, solution of the linearized equations, and updating of the variables. Since the Newmark formulas are linear and have scalar coefficients they may be directly used in the momentum equation to reduce the primary unknowns to a single vector. This vector may be the displacements, u,,1, the velocities, Vn.1, or the accelerations, a,,1. For the explicit case the only viable choice is accelerations. In the sequel we will address the implicit case and use the displacements, u,,1, as the primary unknowns. For an implicit solution it is best to select the initial value for the iterate as 2, =u, (13.17) Any other choice may perturb the displacements in such a way to cause false inelastic values, especially near boundaries, which impede convergence of the Newton method. With the choice (13.17), the values of the initial state which satisfy the ‘Newmark formulas are given by (ae ae 1, = (a Lv, +as(a ze (13.18a) and o . 1 ( eal ath =~ gavet (to gg Je (13.18b) Linearizing the Newmark formulas leads to the results duh? = pada? (18.194) and dvit? = yatdalt? (13.198)-94- ‘Thus the appropriate update formulas (which also satisfy the Newmark formulas) are given by us? = uf, +dult? (13.20a) {2D = vO + a dup (13.20) and ; fae alt? = alts + og dul? (13.20) ‘The linearization of the momentum equation leads to KO du? = RB, 3.21) where KM = | (13.22) or K® = K,+-Lo+—4.M 13.28) ‘ pu Fae In (13.23), K, is the tangent stiffness matrix as computed for the quasi-static problem, C, is a tangent damping matrix, and Mis the mass matrix introduced above. 18.1.N. Newmark, Proc. ASCE, vol 85, EM3, pp 67-94, 1959.-95- SECTION 14. FINITE DEFORMATION FORMULATIONS 14.1, Kinematics and Deformation ‘The basic equations for solid mechanics may be found in standard references on the subject (e.g., see Chadwick [14.1] or Gurtin [14.2]). Here only a summary of the basic equations is presented. A body B has material points whose positions are given by the vector X in a fixed reference configuration, 9, in a three dimensional space. In Cartesian coordinates the position vector may be described in terms of its components as: X=X,E, 3; A=13 (14.1) where E, are unit base vectors. After the body is loaded each material point is described by its position vector, x, in the current configuration, Q. The position vector in the current configuration may be given in terms of its components as X= xe, 3 @=1,3 (14.2) where e, are unit base vectors for the current time. In our discussion, common origins and directions are used for the reference and current coordinates. The position vector at the current time is related to the reference configuration position vector through the mapping x = 9(X,t) (14.3) Whe i origins and directions for the coordinate frames are used, a displace- meat Yéetdr, u, imay'be intréduéed as the change between the two frames. Accord- ingly," - as x=1Xt+u (14.4) is used. In the above 1 is a rank two shifter tensor between the two coordinate frames, and is given by °° : 15 640E, ; @A=1,3 (14.5) where 6,4 is a Kronecker delta quantity such that 1; a=A ' San = {0 | aw : (14.6) In,component form we then have tq = Sa XatUe aan "461, As tnuch as possible we adopt the notation that uppercase letlers refer to quantities defined in {the refdRénce configuration and lower case letters to quantities defined in the current configuration. Ex- ‘ceptions occur when quantities are related to both the reference nd current configurations.-96- A fundamental measure of deformation is described by the deformation gradient relative to X given by ag x (14.8) subject to the constraint J =detF >0 (14.9) to ensure that material volume elements remain positive. The determinant of the deformation gradient maps a volume element in the reference configuration into one in the reference configuration, that is dv = detFdV = JdV (14.10) where dV is a volume element in the reference configuration and du its correspond- ing form in the current configuration. The deformation gradient relates the current configuration to the reference configuration, consequently it has components defined as F= FyueE, (14.11) ‘The deformation gradient may be expressed in terms of the displacement as au Felty (14.12) Using F directly complicates the development of constitutive equations and it is common to introduce deformation measures which are related completely to either the reference or the current configurations. Accordingly, for the reference configuration, the right Cauchy-Green deformation tensor, C, is introduced as C=F'F (14.13) Alternatively;he,Green strain tensor, E (do not confuse with the base vectors), is introduced as. E = }(C-1)) (14.14) where 1p is the rank two identity tensor with respect to the reference configuration and is given by, 1p = b4nE,ES (14.15) and S,p is a Kronecker delta for the reference configuration. The Green strain may be expressed-in terms of the displacements as art, MB yry, (Uy oe Be a[s 5x tg 1+ ag og : (14.16)-97- Defining the displacement vector for the reference configuration as U=lu (14.17) with components Un = baa tte (14.18) the components of the Green strain may be written in the familiar form Wy , Uy, Ue We wh ete By = (z *3e, tox at) (14.19) In the current configuration a common deformation measure is the left Cauchy- Green deformation tensor, b, expressed as b= FF" (14.20) ‘The Almansi strain tensor, e, may be expressed in terms of b as e = 3(1,-b") (14.21) where 1, is the rank two identity tensor with respect to the current configuration and is given by 1, = bpeaeh (14.22) and 6,, is a Kronecker delta for the current configuration. vad 10 14.2, Stress and Traction Measures 9 Stress measures the amount of force per unit of area. In finite deformation problems care must be taken to describe the configuration to which stress is mea- sured. The Cauchy stress, o, and the Kirchhoff stress, r, are measures téfined with respect to the current configuration. They are related through the determinant of the deformation gradient as T= Tpegep = Jo = Joules (14.28) The second Piola-Kirchhoff stress, S, is a stress measure with respect to the reference configuration and has components S = SapE,E (14.24) ‘The second Piola-Kirchhoff stress is related to the Kirchhoff stress throu 1 = FSF" (14.25) Finally, the first Piola-Kirchhoff stress, P, is related to S through P= FS = Pe, Eh (14.26)which gives t= PF" (14.27) For the current configuration traction is given by teo7n (14.28) where n is an unit outward pointing normal to a surface defined in the current configuration. This form of the traction may be related to reference surface quantity through force relations defined as tds = tydS (14.29) where ds and dS are surface elements in the current and reference configurations, respectively, and ty is traction on the reference configuration. Note that the direction of the traction component is preserved during the transformation and, thus, remains related to the current configuration forces. The reference configuration traction is deduced from the first Piola-Kirchhoff stress through t, = PN (14.30) where N is an unit outward pointing normal to the reference surface. Using the definition for traction and stresses we obtain F'nds = JNdS (14.31) and ds = J[N-(C*N)} dS (14.32) to relate changes in the surface area and transformation of the normals. 14.8. Balance of Momentum. ‘The balance of momentum for a solid body consists of two parts: balance of linear momentum; and balance of angular momentum. The balance of linear momentum may be expressed by integrating the surface and body loads over the body. Accordingly, for a body force per unit mass, b,, the resultant force, R, acting on a body is given by Joby do+ Jtds =R (14.24) a ia where p is the mass density per unit volume and a0 is the surface area of the body, both for the current configuration. The mass density in the current configuration is related to the reference configuration mass density, go, through po=Jp (14.25)-99- The total linear momentum of the body is given by p= J pvdv (14.26) a The balance of linear momentum describes the translational equilibrium of a body (or any part of a body) and is obtained by equating the resultant force, R, to the rate of change of the body momentum, p. Accordingly, pb, du+ | tds = | pwdv (14.27) J feds = J a a a Introducing the relationship between traction and stress and using the divergence principal leads to the balance of linear momentum relation Jldive+ p(by-w)1dv = 0 (14.28) a where div is the divergence with respect to the current configuration, that is, | ag dive = = e% (14.29) Since the above result must hold for any part of a body a local form for balance of linear momentum may be deduced as diva +pb,, = pv (14.30) This relation is also called the local equilibrium equation for a body. Similar relations may be constructed for the balance of angular momentum and lead to the requirement a=" (14.31) that is, the Cauchy stress tensor is symmetric and, thus, has only 6 independent components. ‘The balance of momentum may also be written for the reference configuration using results deduced above. Accordingly, we may write the integrals with respect to the reference body as frobudV+ [ todS = fpovdV (14.32) De BD Me where the definitions for p, o and nds in terms of reference configuration quantities have been used. Using the divergence principle on the traction term leads to the result, JUDivP + po(b,-W1dV =O. (14.33) a-100- which has the local form DivP + pobm = pov (14.34) In these relations Div is the divergence with respect to the reference configuration coordinates oo oPa Due = Sete. (14.35) We also note that the symmetry of the Cauchy stress tensor, o, leads to the corre- sponding requirement on P FP" = PF” (14.36) and subsequently to the symmetry of the second Piola-Kirchhoff stress tensor s=s? (14.37) 14.4. Boundary Conditions The basic boundary conditions for a solid region consist of two types: displacement boundary conditions and traction boundary conditions. Boundary conditions are defined on each part of the boundary for which a component or components of a vector may be specified without solution of any auxiliary problem. The conditions are usually given in terms of their components with respect to a local coordinate system defined by the orthogonal basis, e’,, a=1,2,3. At each point on the boundary one (and only one) boundary condition must be specified for all three directions of the basis e’,. Generally, these conditions may be a mixture of displacement and traction boundary conditions. For displacement boundary conditions, components of the position vector, x, may be expressed with respect to the basis as x= x',e, (14.38) Boundary conditions may now be given for each component by requiring “? fg = a (14.38) for each point on the displacement boundary, an,. The boundary condition may also be expressed in terms of components of the displacement vector, u. Accordingly, u=x-1X= ue, (14.39) x! define components of the displacement with respect to the prime coordinates, Thus, boundary conditions may now be given for each displacement component by requiring specified quantity is indicated by a superposed bar").w= ie (14.40) In general, the boundary condition is non-linear unless points in the reference configuration can be identified easily (such as fixed points). ‘The second type of boundary condition is a traction boundary condition. Using the orthogonal basis described above, the traction vector t may be written as t= tes (14.41) ‘Traction boundary conditions may be given for each component by requiring (14.42) for each point on the boundary, aQ,. 14.5, Initial Conditions Initial conditions describe the state of a body at the start of an analysis. The conditions describe the initial kinematic state with respect to the reference configuration used to define the body and the initial state of stress in this position. In addition, for constitutive equations the initial values for internal variables which evolve in time must be given. ‘The initial conditions for the kinematic state consist of specifying the position and velocity at time zero. Accordingly, x(0) = 9(X,0) = do(X) (14.43) 0 and v(0) = @(X,0) = %(X) (14.44) are specified at each point in the body. The initial conditions for stresses are specified as. o(0) = & (14.45) at each point in the body. 14.6, Material Constitution - Finite Elasticity ‘The equations are completed by specifying a material constitution. As an example, we consider a finite deformation form for hyperelasticity. Thus, we postu- late the existence of a strain energy density function, W, from which stresses are computed by taking a derivative with respect to a deformation measure. For a strain energy density expressed in terms of the right Cauchy-Green deformation tensor, C, the second Piola-Kirchhoff stress tensor is computed as (14.46)-102- For an isotropic material the strain energy density depends only on the three invariants of the deformation. Here we consider the three invariants as Io = trC = Cxx (14.472) Ue = 3(13-trC?) = 3 (Cre Cu - Cu Cix) (14.47b) and Ig = det = J? (14.476) and write the strain energy density as W(C) = WUe,He,J) (14.48) We select J instead of I as the measure of the volume change. Thus, the stress is computed as ow ale , ow ate , aw ad 8 = 257530 +7 5m, ac +7 a7 96 449) The derivatives of the invariants may be evaluated as we =h (14.50a) a = Igl)-C (14.50b) a =Jct (14.50a) Thus, the stress is computed to be S= age 194237 Uete- o+ tact 4.51) The second Piola-Kirchhoff stress may be transformed to the Kirchhoff stress using (14.25), and gives 22 Ww T= 2aTG => b+ Fyplleb- +4 a1, (14.52) ‘As an example, we consider the case of a Neo-Hookean material which includes a compressibility effect. The strain energy density is expressed as W(Ie,d) = w(Ie-3-2Ind)+2a( J -1F (14.53) ‘The material constants 4 and have béen selected to give the same response in small deformations as a linear elastic material using the Lame’ moduli. Substitu- tion into (14.51) gives S = 2y(1)-C1)+ad(J-1)07 (14.54)-103- which may be transformed to give the Kirchhoff stress = 2y(b-1,)+4d(J- DL (14.55) ‘The Cauchy stress is then obtained from Some formulations require computation of the elastic moduli for the finite elasticity model. The elastic moduli with repect to the reference configuration are deduced from -42W 3Cac ‘The spatial elasticities related to the Cauchy stress, ¢, are obtained by the push forward ¢ 1 cw = 5 FuFy Pix Fu Com, For the Neo-Hookean model the material modul with respect to the reference configuration are given as Cry = AT (2T-1) CCH ~2(u-AIT(T-1)) CHC ‘Transforming to spatial quantities gives cy = A(2T~1) 6y8u - 202 - AT -1)) bade Other forms of constitutive equations may be introduced using appropriate expan- sions of the strain energy density function. As an alternative, an elastic formulation may also be expressed in terms of the principal stretches (which are the sqare root of the eigenvalues of C); however, the computations are quite delicate (see (14.7). 14.6, Variational Description ‘A variational theorem for finite elasticity may be written in the reference configuration as Ta) = JW(CCw))aV ~ fur pobnaV— J winds (14.56) Dy My where % denotes the specified tractions in the reference configuration and Qo, is the traction boundary for the reference configuration. In a finite element formulation, the basic element arrays evolve from the balance of linear momentum equations written as a variational equation. Accordingly, in the reference configuration a variational equation is defined from (14.56) as** s1= fo scav-foutb.dV- f suTds=0 (14.57) ae D9 Dy Woy‘woyyenbe rouoste}zea oy seaudxe oy ofqjod we yoru e poem af WoRyeju xsyonr ‘axOUNION IN agen © On * POT-105- where du is a variation of the displacement (often called a virtual displacement) which is arbitrary except at the kinematic boundary condition locations, 2Q,, where, for convenience, it vanishes. Since a virtual displacement is an arbitrary function, satisfaction of the variational equation implies satisfaction of the balance of linear momentum at each point in the body as well as the traction boundary conditions. We note that using (14.26), (14.46) and constructing the variation of C, the first term reduces to OW 50 = issc = oF 50 OC = }SSC = SFP (14.58) Furthermore, by introducing the inertial forces through the body force as Dn > Da -¥ = by -% (14.59) where v is the velocity vector, the variational equation may be written as 61 = fou? povav+ jer"Pav (14.60) cy Ny - [ou obn dV - f du7 dS = 0 DA Dy This is the variational equation form of the equations which is used for subsequent development of the finite element arrays. The first term side represents the inertial terms. For static and quasi-static problems this term may be neglected. The second term is the stress divergence term which also may be given in terms of the second Piola-Kirchhoff stress as 6F"P = 6F"FS = }5CS = 6ES (14.61) where symmetry of the second Piola-Kirchhoff stress is noted. The third and fourth terms of the variational equation represent the effects of body and surface traction loadings. ‘The above variational equation may be transformed to the current configuration as 61 = J au? pvdv+ fv(suy oldv (14.62) a a - J su? pb do- f ou Eds = 0 a aD, ‘The last result is identical to the conventional, small deformation formulation found in earlier sections of the notes as well as in finite element texts (e.g., see Hughes [14.3] or Zienkiewicz and Taylor [14.4,14.5]) except that integrals are performed over the deformed current configuration. Representations with respect to a fixed reference configuration are introduced to simplify the development of the basic relations. Some authors refer to the case-106- where the reference configuration is the initial description of the body (at time zero) as a total Lagrangian description and to one which is referred to the previous computed configuration as an updated Lagrangian description. For the development considered here it is not important which is selected since ultimately all integra- tions are carried out over the current configuration; and, either a total or an updated description can be transformed to the current state. 14.7, Linearized Equations ‘The stress divergence term may be written in many forms, as shown above. To solve a boundary value problem the nonlinear equations may be linearized and solved as a sequence of linear problems. The linearization should be considered in a reference configuration representation. In this section it is expedient to again use a tensor form of the equations instead of the matrix form used above. Accordingly, a formulation based upon the second Piola-Kirchhoff stress and written in tensor form is considered for the linearization step. Jer(oF"Fs)av = r(orsF")av (14.63) D9 Dy which are equivalent forms. In the above, the trace operation denotes the following step (reference configuration tensors are used as an example, but other forms also hold) tr(AB) = Ay By (14.64) Note that in the reference configuration the domain, Q% is fixed (i.e, does not change) which is not true for a formulation considered directly in the current configuration. Consequently, a linearization may be written as AC [tr(oFSF" dV) = [tr(sF SAF" )dV + [tr(SFASF")dV (14.65) Qq Qy 2 We also note that for a continuum problem A(6F) vanishes, which is not true for problems in beams, plates and shells and, thus, additional terms are necessary. The linearization may be transformed to the current configuration and expressed in terms of quantities associated with the Cauchy stress. Accordingly, using oF = V(6u)F (14.66) and a similar expression for AF gives AC [ tr(SF SET aV) = [tr(V(su)oV(Au)? )do + [ tr(V(du)Ao)dé14.67) 2 a a where Ao = Sas? (14.68)-107- ‘The first term on the right hand side leads to the geometric stiffness term in a finite element formulation, whereas, the second term depends on the material constitution and leads to the material part of the stiffness. ‘The material part involves AS which is computed for each particular constitutive relation. This will be discussed later for a particular constitutive equation; however, in general we seek an expression of the form AS = CAE (14.69) where @ are the material moduli for the material constitution expressed in the reference configuration. When used with the definition of Ac this may be transformed to the current configuration as Ao = che (14.70) where ¢ are the material moduli expressed in the current configuration. The moduli are related through Jo = FFCF'F’ (14.71) In the above Ae is the symmetric part of the gradient of the incremental displacement. It is expressed as de = 3[VAu+(VAu)?] (14.72) Substitution of the above into the term for the material part of the stiffness yields fer(v(su)do)dv = fir(deede) )dv (14.73) a a which we note is also identical to the form of the linear problem. 14.8, Element Technology A finite element discretization may be constructed by dividing the body into finite elements. Accordingly, we have Q=% = 59, (14.74) where 0, is the domain of an individual element, e, and ®, is the domain covered by all the elements. We note that in general ©, is an approximation to the domain of the real body since the elements only have mapped polygonal shapes. With this approximation the integrals in the variational equation may be approximated as ford = food =z fC dav (14.75) a Oy “a ‘Using this approximation the variational equation become z fou? pvdu+ fv(suy ody =) fou” pb, do + J surias (14.76) “a a *|a De- 108 - ‘An approximate variational solution may be developed by writing trial solutions and test functions for the motions and virtual displacements, respectively. Adopting an isoparametric formulation (e.g., see Hughes [14.3] or Zienkiewiez and Taylor (14.4,14.5]) we may write for a typical element X=Ny(E)K! ; [=1,2, ,nen (4.77) where nen is the number of nodes defining an element, I are node labels for the element, N;(€) are shape functions for node I which maintain suitable continuity between contiguous elements and X! are the coordinates for node I. Similarly, we may write approximations for the current configuration as x = N;(g)x!(t) (14.78) the displacements as w= N,(E)ul(t) (14.79) the incremental displacements as Au = N;()du(t) (14.80) and the virtual displacements as = Ny(§)su! (14.81) ‘Time dependence is included in the nodal parameters for the current position and displacements. 14.9. Consistent and Lumped Mass Matrices Using the above approximations we may discretize the terms in the variational equation for each element. Accordingly, the first term becomes Jou? pvdv = (su")! [Ny pNs dvs? (14.82) Dy DX where summation convention is implied for the a and b indices. The integral for the shape functions defines the consistent mass matrix for the analysis which may be written as = J pNydv, (14.83) 4 For procedures to construct a lumped mass see either Appendix 8 in [14.4] or [14.6].-109- 14,10, Stress Divergence Matrix The stress divergence term may be expanded by noting symmetry of o to give fertv(swoldy = ftrtseoldv a a where 6¢ is given by be = 3{V(6u)+(V(du))"] Introducing matrix notation for o and de as a o= [o11,612,098 212,022,001] and be the stress divergence term may be written as . [ sen der Beu0 2812 28630 256n | Joe" odv 4 Expressing the de in terms of the virtual displacements gives Osu; Fn Odu2 i diy asuy Oe abu, Oxy” aus rm Oxs Obs Oxy abuts Oe abu Bas (14.84) (14.85) (14.86) (14.87) (14.88) (14.89) Using the interpolations for the virtual displacements in each element leads to the matrix relation-110- Ni 0 0 0 Nig 0 be aye ge=| 0 9 Nea] out! - Bou! (14.90) Mia Nu 0 | ub 0 Nis Nia b Nis 0 Mia In the above, the notation aN; —_— 14.91) Nn = Fe 491 has been used for conciseness. The B, matrix describes the transformation from the virtual displacements, du! to the de. The stress divergence term may now be written as Jee odv = (sul)? [BP ode (14.92) a Dy The above expressions are identical to results obtained for the linear problem except that all calculations are based upon coordinates in the current (deformed) configuration. 14.11. Geometric stiffness The geometric stiffness for a finite element formulation is obtained by substituting the interpolations described above into the geometric term for a single element. Accordingly, fercv (suo V(Aw)™ dv = (sul)? Jer (VNT o VN, ) dv du? (14.93) a & Evaluation of the right hand side of the above expression leads to the geometric stiffness matrix which is given by (K)y = fer(WNP oVNy)dv1, (44.94) a, In component form the expression for the geometric stiffness is given as aN; aNy 8) = AY ay 5 (K~y = I5 oH See dv 6; (14.95)-ll- 14.12, Material tangent matrix - standard B matrix formulation. ‘The material tangent matrix is deduced from the term Je (v(su)do)dv = ftr(deede)dv (14.96) Q Dy which is evaluated for a typical element. In matrix notation the right hand side becomes Jer(seede)dv = [ 5e"DAedv 14.97) a Q where D denotes the material moduli in the current configuration given in the matrix representation introduced for the linear problem. Furthermore, substitution of the finite element interpolations into the incremental strain term leads to the result in matrix form Ae = ByAu’ (14.98) ‘Thus, the material tangent is computed from Jee" Dae dv = (sul)? JBIDB,dvaw (14.99) a Q and the material tangent matrix is given by (K")y = f BT DB, dv (14.100) a, which again is identical to the linear problem except that all steps are performed for the current configuration. 14.13. Loading terms The right hand side terms may be discretized by introducing the interpolations for the virtual displacement. Accordingly, the body force may be given as Jou" ody dv = (su!)? [Ny pb, dv (14.101) a Qe and the boundary loading is J ouTéas = (ul)? i Nytds (14.102) a, ae-112- 14.14, Basic finite element formulation Accumulating all terms together, the variational equation may be written as pioty| aes aioe = (14.103) 7 a, where f; is the sum of the body and surface traction terms. fy = [Nipbyadv+ [ Nytds (14.104) Qe Wy Since Su’ is arbitrary, the variational equation leads to the discrete balance of linear momentum equations ¥|Myx’+ [BP odv-f;|=0 (14.105) a, which may be written in the compact matrix form M&+N(o) =f (14.106) where N(o>) is the stress divergence vector. Solution of this set of equations together with satisfying the material constitution and the displacement boundary conditions, yields the solution to a problem. A common solution procedure is to use a Newton type solution method and solve a sequence of linear problems. Accordingly, in a Newton Method we write the momentum equation as R = f-M&-N(o) = 0 (14.107) A linearization of this set of equations gives the result MaAi+K,Au = R (14.108) where K, = K"+Ké (14.109) ‘The above description is for a standard displacement type formulation. We refer to the method as the standard B-matrix formulation. 14.15. Mixed formulation In the mixed formulation used, a modified deformation gradient, (as described in Simo and Taylor [14.7)), is used. The mixed formulation is used to permit solution of incompressible and nearly incompressible materials, as well as, compressible solutions which can be treated by a standard B matrix formulations. Thus, the-113- modified deformation gradient is based upon a separation into volumetric and deviatoric parts as F = Fa Few (14.110) where F,,; measures volumetric part and F,,, the deviatoric part of deformation. Since det F measures the volumetric part we have J = detF = detPyy detFn (14.111) which leads to the result detFiy = J (14.112) and det Fy = 1 (14.118) This may be accomplished by using Fy = JY 1 (14.114) for the volumetric part which gives Fug = IMF (14.115) for the deviatoric part. The modified deformation gradient may then be constructed by replacing the volumetric part by a mized treatment. Accordingly, we define - (ey el= 4.116) F ( 5 ) F (14.116) as the modified tensor. In the above expression, @ is a mixed representation for the determinant of the deformation gradient. The modified right Cauchy-Green deformation tensor is then computed from qa4.u17) with similar definitions for E and 6. The virtual modified deformation gradient is now given by 56 1). rr 1,4 (vou-jaivsut,)]}F (14.118) A three field variational statement of the problem is completed by adding to the motion, ¢, and mixed determinant of the modified deformation gradient, 6, the mixed pressure, p. 1(u,0,p) = [W(E(a,0))dV+ | p(J-0)dVv (14.119) D9 Dy - J? rob, dV ~ J ut dS Dy Wo,-14- A variational equation including the effects of inertia may be constructed following steps above and written as (see, [14.7]) fou" pvdv+ [ V6U" (in + pLs)dv = [6u"pby do+ J Su” Eds (14.120) a a a ey for the linear momentum equation, (14.121) for the relationship between the mixed pressure and the trace of the stress, and Jovca-4 dv = 0 (14.122) } 7 for the relation between the mixed and the determinant of the deformation gradient. In the above, the modified Cauchy stress, ¢, and the modified Kirchhoff stress, 7, are related to the modified second Piola-Kirchhoff stress by Jo=t= FSF (14.128) where § is computed using € as the deformation measure. The deviatoric part of the stress, 6, is then computed using Step = A FUP )S = Taw 6 (14.124) where I is a rank four identity tensor. The spherical part of the stress is given by the mixed pressure, p, not tré. The mixed pressure p is computed from éré using the variational equation given above. Thus, the stress in this approach is computed using o = pit Saw (14.125) A finite element implementation for the above may be deduced using the isoparametric interpolations given above for X, x, u, and 5u. In addition interpolations for 6, 50, p, and Sp must be given. In the low order elements the above functions are all taken as constant in each element. Discretization of the modified momentum equation gives M&+N(6i0+P1,) = F (14.126) where the stress divergence vector for a typical node is given by Ni (Gao PL) = E [BF (Sa0+ Pav (14.127) “a, ‘The pressure, p, appearing in the above relations may be obtained by first computing the mixed volume, 6, using the third variational equation, Accordingly, for each element (with the constant interpolations for 9 and p) integration of the third-115- variational equation yields a solution 2, oe as (14,128) for each element, where ©, is the volume of the element in the current configuration and 2,9 is the volume in the undeformed reference configuration. The @ may now be used to define the modified deformation quantity and the modified stress state, may be determined in each element. Finally, use of the second variational equation yields the mixed pressure as 1 jez 36 dv (14.129) 0 a in each element. This may be combined with the deviatoric part of & to define the mixed stress, ¢, in each element, A tangent matrix may be computed for the mixed formulation. Details for the construction are included in [14.7].-116- REFERENCES 014.11 Chadwick, P,, Continuum Mechanics, John Wiley and Sons, New York, 1976. (14.2) Gurtin, Mortin E., An Introduction to Continuum Mechanics, Academic Press, New York, 1981. (14.3) Hughes, Thomas J.R., The Finite Element Method, Prentice-Hall, Englewood Cliffs, 1987. [14.4] Zienkiewicz, 0.C., and R. L. Taylor, The Finite Element Method, 4th ed., Volume 1, Basic Formulation and Linear Problems, McGraw-Hill, London, 1989. 14.5] : Zienkiewicz, 0.C., and R. L. Taylor, The Finite Element Method, 4th ed., Volume 2, Solid and Fluid Mechanics, Dynamics and Nonlinearity, McGraw-Hill, Lon- don, 1991. 14.6) . Hinton, E., T. Rock, and O.C. Zienkiewicz, "A note on mass lumping and related Processes in the finite element method," Earthquake Engineering and Struc- tural Dynamics, 4, 1976, pp 245-249. 14.71 Simo, J.C. and R.L. Taylor, "Quasi-incompressible finite elasticity in principal stretches. Continuum basis and numerical algorithms", Computer Methods in Applied Mechanics and Engineering, 85, 1991, pp 273-310. [14.8] Simo, J.C. and F. Armero, "Geometrically non-linear enhanced strain mixed methods and the method of incompatible modes,” International Journal for Numerical Methods in Engineering, xx, 1992, pp xx-xx. [14.9] Handbook of Mathematical Functions, Abramowitz, M., and I.A. Stegun (eds.), Dover Publications, New York, 1965. 114.10] Simo, J.C, and TJ.R. Hughes, Elastoplasticity and Viscoplasticity, Computa- tional Aspects, Springer (to appear). ~CE 238 Notes - REPRESENTATIONS FOR STRESS AND STRAIN IN 2-DIMENSIONS ‘The stress divergence term in the weak form of the balance of linear momentum is written as 6a"R, = J be7odQ where, in two-dimensions, the virtual strain in terms of all nonzero components @ie., ignoring symmetry) is given by be = b= [ Se, 5eu0, de, 5ex2,5en | similarly the strain is a [ensemsena €12 en | and the stress is o = 05 = [onsonso 012,00 | Using the fact that o and ¢ are obtained from symmetric tensor quantities, the stress and strain arrays may be reduced in size to four independent quantities. To accomplish this reduction we may introduce two projector quantities. There are alternatives which may be employed for the projection and two are considered below. 7 ection form 1. w ‘The first projection form is given as follows #5 = Prey where Pi = eooor eoorne eonoco i a [entsere ean 20x] ‘Similarly,: os = Pao where eee ee P,=]0010 ooo1 oe and a= [eu .2n,0%,0. | It easily verified that PIP, =] where I, is a 4x4 identity matrix. With these definitions and the above identity, it also may be shown that o, = Plos and = Phe, With the above constructions we have the alternative representations for the stress-divergence term SUR, = | Sef 500 = foe? oda a a For a linear elastic isotropic material the elastic moduli for the constitutive equation a = Dyes is given by At2y a a0 A at2p 24 0 a A At2u 0 0 0 o 4 Although this is the traditional method for reducing the problem there are a number of difficulties which arise. One is the representation of the invariants of the deviator. The traces is conveniently expressed in matrix form leading to the vanishing of the first invariant of the deviatoric stress, J, being written as J, =1%s=0 D=og: ‘The second invariant of the deviator in tensor form is given as Tq = Ptrace(s*) = } sy sy ‘The matrix expression for this is more difficult since }sTs # Jy : Additional problems arise when dealing with the coordinate transformation of stress and strain by a rotation. In tensor form a transformation is given by ecalea where a is any admissible transformation. In particular if a is a rotation then it is an orthonormal transformation for which \ alsa Using the transformations to matrix representations in terms of 4 components the desired transformation may be written as = Te where T is constructed from the a using the reductions previously defined for transforming tensor stresses to matrix stresses. The difficulty is that for rotations T is not an orthonormal transformation. Consequently, Tat While appropriate transformations may be defined, there are other aspects which may be important. For example transformation of the elastic moduli by Di, = T°D,T generates moduli for which the spectral transformation leads to different eigenvalues, Generally this is of no consequence, however, if such a computation is desired we have non-physical results from the transformation selected. Projection form 2. It is possible to modify the projections slightly so that the undesirable results cited above are avoided. Accordingly, we introduce the modified projections 6 = Pa where 100 0 0100 0010 000 VF 000 43.and ene sos, Bere | Similarly, o, = Pa, where a= [on.o%2,04 Bou | It easily verified that pe With these definitions and the above identity, it also may be shown that & = Bo, and & = Bey With the above constructions we have the alternative representations for the stress-divergence term 6U"R, = Jee ose = [oat o,a0 a a For a linear elastic isotropic material the elastic moduli for the constitutive equation is given by a. « a atte A Ol aaa Dy a a asae 0 [= AEE t2eh 0 0 0 24 The modified projection affects only the shearing components, hence the first invariant is still expressed in matrix form as Jy=Va=0 while the matrix expression for the second invariant is now given as-5- ‘The coordinate transformation of stress and strain is given by &,=T4 where f is constructed from the a using the reductions previously defined for transforming tensor stresses to matrix stresses. The orthonormality of the transformation is preserved, consequently =f leading to =f 2, Furthermore transformation of the elastic moduli by B= TDF generates moduli for which the spectral transformation has identical eigenvalues. Finally, we note that the results for the two projections may be interchanged using the fact that a = Ih and a= Pe where 1000 p=|9 19% ° 7/0 0 1.0 000 WZ. Thus, the relationship between the elastic moduli is given as p=fbd ‘The selection of the projector to be used leads to subtle but important differ- ences. There is no one choice which is clearly superior to the other for all parts of a computation. For computation of the finite element arrays, all existing elements generally use the projection method 1, thus if a constitutive model is developed ‘using projection method 2 it is necessary to transform the results returned to the equivalent of projection method 1 (ie., the returned stresses must be o, and the moduli D,). At the constitutive level it is easier to deal with the shearing components using the projection method 2 (especially for plasticity formulations involving the deviatoric invariant J). Transformations are also more consistent.

Taylor Plasticity Notes

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Taylor Plasticity Notes

Uploaded by

Copyright:

Available Formats

You might also like