OF SOLIDS George Z. Voyiadjis Department of Civil Engineering, University of Petroleum and Minerals, Dhahran, SaudiArabia ;\;J )I~.r ..)iJ ~W4.i..,.,'J ..I ~."s~;1I .;.J '1~WI I f J s,~ .:-jJ l (, ~) (5.r) J !l,..,..:.s ..) J ...,1)J .l (Finite Element) ~>~i ~I I ~ ..)J - ~~'11 Ifa~ .;.J ~)J .l W~I ;\;J ..ull . .s.f-l,1 ~~'11 ~l:-- ..) iJ ~WI~~.F- ..) ~l:J I ~~'1IJ 'J h-Ij'1\ The constitutive equations describing the behavior of large elasto-plastic defor- mations of solids havebeen proposed inaprevious paper [11. Theproblemof analysis of displacements, stresses and strains in elements made of this material subject to arbitrarily large deformations under the conditions of plane strain has been formu- lated in terms of the finite element method. 0377-9211/79/0104-0041$01.00 1979by The University of Petroleum and Minerals THE FINITE ELEMENT METHOD APPLIED TO LARGE ELASTO-PLASTIC DEFORMATIONS OF SOLIDS I nthis paper the finiteelement method is usedinthe case of problems involving large elasto-plastic de- formations of materials. Such materials display an initial linear elasticity which is followed, at increasing loading, by plastic strains with no rate, or viscous effects. The topics included inthis paper are: I . The finite element formulation of the analysis of two-dimensional problems. 2. Methods of solution of theequations resulting from the use of the finite element method. I n this paper, theproblem of analysis oflarge elastic- plastic deformations will beformulated interms of the finite element method. Since the basic procedures of the finite element method are described in numerous papers and monographs (see, for example Reference [2]), only the aspects related to the present problem will be emphasized in this paper. The fundamental equation in the following argu- ments is the principle oj virtual work (Equation (2.26), Reference [1]: L 8eT s dV - L 8uT fdV - I s 8uT P dS =0 (1) which is the condition of equilibrium in terms of the displacement fieldu(x, y). I nthefiniteelement method, the displacement fieldu is approx'imated by adiscrete model which contains a finite number of independent nodal displacements. For this purpose, the regions of integration in Equation (I ) are divided into a finite number of subregions, or elements. Within each ele- ment, the displacement fieldis approximated byknown functions which are continuous across the element boundaries. For anonlinear elastic body, for whichthe stress vector is aJunction of thedisplacement gradients, Equation (l) results in a systemof nonlinear algebraic equations. I n the present problem of anelastic-plastic body, the stress s is aJunctional of the displacement gradients, Equation (l) leads to a systemof Junctional equations for the nodal displacements. Figure 1shows atypical element meshinanarbitrary body with m elements and n nodal points. Thecompo- nents of displacement at a node iare Ui and Vi' The Figure 1, Typical Element Mesh; Notation for Global Nodal Displacements nodal displacement vector for the body, or the system, is the 2n X 1matrix qwhose elements are A typical element with its nodal displacements is shown in Figure 2. The nodal displacement vector Qk of the element k is Since, at the corresponding nodes, the nodal dis- placements of an element are identical with thenodal displacements of the body, it is evident that where the form of Tk is implied by the definitions of qandql stated byEquations (3) and(4), respectively. Thedisplacement fieldwithintheelementkis approx- imated by linear functions of x andy, Typical Triangular Element; Notation for Element Displacements 1 u =- [ ( a, +bl x +c,y) u, +( a2 +b2x +C2Y ) U2 2 , 1 + ( a3 +b3 x +C3 Y ) U3 ] 1 v =2,1 [ ( al +bl x +cly) v\ +( a2 +b2x +C2Y ) V2 + ( a3 +b3 x +C3 Y ) V3 ] where XI Y I 2,1 =2 X (area of triangle) =det X 2 Y 2 X 3 Y 3 al =X 2Y 3 - X 3 Y 2, bl =Y 2 - Y 3 , C, =X 3 - X 2 with a2, b2, C2, a3 , 3 , and C3 obtained by cyclic per- mutation of indices. I n matrix notation, with Uk =[~)b N, =(a, + bl X +c,y) /2,1, N2 =... etc. 1 =[6 ~J Nk =(I N" 1N 2, 1N 3)' Equation (6) becomes Uk=Nkqk' I n view of Equation (5), Uk=NkTkq The components of strain are computed from the components of displacement according to equation (8). [ 11. The result is where the matrix Bk corresponds to the linear terms in the strain-displacement relations, while BZ takes into account the nonlinear terms. I t can be easily verified that the matrices Bk and BZ are shown by Equations (10) and (11), respectively. 1 [bl 0 b 2 0 b 3 0 J 2,1 0 CI 0 C2 0 C3 CI bl C2 b2 C3 b) BZ=FkGkHk (10) (11) [ bOlO b,O b20 b20 b3 0 b3 0 ] C, 0 C, 0 C2 0 C2 0 C3 0 C3 C, bl CI b, C2 b2 C2 b2 C3 b3 C3 b3 u, 0 0 0 0 0 u, 0 0 VI 0 0 0 0 0 VI U2 0 0 0 [ ~ 0 b2 0 b) g J Gk= 0 0 U2 0 Hk= bl 0 b2 0 0 V2 0 0 CI 0 C2 0 C3 0 0 0 V2 0 C1 0 C2 0 C3 U3 0 0 0 0 0 U3 0 0 V3 0 0 0 0 0 V3 I n the following, the differential dek will be needed rather than the strain ek itself. Differentiation of (9) yields consequently, dek =(Bk+BZ) dqk =Bk dqk dek =BkT k dq. (13) From the stress-strain relation Equation (3.23), Reference [1]; Sk=L DkBkdqk Substitution of (8) and (13) into (1) results k=m 8qTI Tif Bis,dV k~ 1 Vk k~m -8qTI Tif NifdV k~ 1 Vk - 8qTI Ti f Nip dS =0 (16) k Sk The above equation is valid for any virtual 8qT;there- fore 'I mTi f BiskdV-1mTiFk- kfnTiPk=O k~ I Vk k~ I k- I and the last summation in Equations (16) and (17) extends onlyover theboundary elements. Equation (17) is anonlinear functional equation inq, becauseBk is afunction of qkand, thus, of q, and s is a functional of qas shown by (15). The first term onthe left-hand sideof Equation (17) will bedenoted byQ (q); thesecond andthethird terms combined, by R. Thus, Equation (17) can bewritten as Q;(q) =R i , i = 1,2, ... , 2n (18b) The numerical solution of (18) or (19) will require the derivatives 8Qi =' Kij i =1,2, ... , 2n 8q . 1 2 2 J J=" ... ,n Differentiation of Equation (17) yields k~m dQ =I TJ f (dBJ Sk +BJ ds k ) dV (20) k~ I V k From (12), (10) and (11), - 1 dB k =dB k "=- 4 , 1 2 d(:~) d(~;) 0 0 o 0 d ( ~ : )d ( : ; ) d(~:) d(:;) d(:~) dG:) [ b' 0 b 2 0 b 3 ~ ; ] . 0 b, o b 2 0 c 1 0 C2 0 c 3 0 c 1 0 c 2 0 C 3 and dlF Sk =C k dqk where [ bl 0 b 2 0 b 3 0J . 0 b l 0 b 2 0 b ] C 1 0 C2 0 C3 0 o C 1 0 C2 0 C 3 The first term in Equation (20) is thus, L k dB! Sk dV =K~dq (23) is the so-called initial stress, or geometric, stiffness matrix of the element k. The second terminEquation (20) can bewritten, with (14) as f BJ DkB k dqkdV =K k dqk Vk (25) where Kk =f B"TDkB"dV + f (B"TDkB;; Vk Vk + B;;T DkB" +B;;TDkB;;T)dV (26) The matrix K k will becalledthematerial stiffness matrix of the element k. (I n the problems of nonlinear elasti- city, the matrix K k is known as the elastic stiffness matrix.) With (23), (25), and (5), (20) becomes k=m dQ =I TJ (KZ +Kd T k dq (27) k~1 and, hence, the matrix K, whose elements are Kij = oQ) oqj, is k=m K =I TJ (K~ +K k ) T k (28) k~ I K k =K~+K k (21) is known as thetangent stiffness matrix of theelement k. The matrix K defined by Equation (28) is theglobal tan- gent stiffness matrix. b l 0 C 1 0 0 b l 0 C, C k =, b 2 0 C2 0 0 b 2 0 C2 b 3 0 C3 0 0 b 3 0 C3 The functional equations (18) aresolvedbyapplying incremental steps of loading and performing iterations within each increment. The computational procedure is then as follows: L et q(n)and R(n)be the nodal displacements andthe right-hand side of (18), respectively, at the end of the nth load increment, for which Q(q(n)) _ R(n) =O. L et LlR be the next load increment, and let Llq be the corresponding nodal displacement increment; they satisfy Q(q(n) +Llq) - (R(n) +LlR) =O. (30) The first approximation for Llq will be obtained by taking the first term of the Taylor expansion of Q at q(n), i.e., by replacing Equation (30) with Q(q(n)) +K o Llq - (R(n) +LlR) =0 or, in view of (29), KoLlq - LlR =0 K o =(~~) q~q(n) is the tangent stiffness matrix at q =q(n)(the tangent stiffness matrices have been discussed in Section 2; see Equations (19), (27), and (28)). From (31), the first approximation of Llq follows as Llq(l) =K.;-ILlR. I f the above approximate value is substituted into Equation (30), we have Q(q(n) +Llq(I)) - (R(n) +LlR) =-lJ '. (32) The next correction to Llq is obtained by expanding Q at q =q(n) +Llq(I)' which results in Q(q(n) +Llq(I)) + K(I)Llq - (R(n) +LlR) =0 or, with (32) in K(I)Llq =lJ '. (33) where K(j) is the tangent stiffness at q =q(n) +Llq(l)' From (33), Llq =Llq(l) +Llq(2)' (35) The ith correction to Llq follows, thus, from K U - I ) =K(q(n) +Llq(l) + +Llqu-I) (37) lJ 'U-l) =Q(q(n) +Llq(l) + +Llq(i-l)) - (R(n) +LlR) (38) The value of Llq after i corrections is Llq =Llq(l) +Llq(2) + ... +LlqU) (39) The above procedure is formally identical with the known Newton- Raphson method for the solution of a system of nonlinear algebraic equations. The convergence of this method for a system of nonlinear algebraic equations has been widely discussed (see, for example, [3]). I t should be kept in mind, however, that the equa- tions of the present problem (Equations (18)) are not algebraic equations in q. Consequently, many pro- perties of the Newton-Raphson method, especially those related to the nature ofthe approximation and to the convergence of the process, are not transferable to the case of Equations (18). The differences between the Newton-Raphson method applied to a system of nonlinear algebraic equations (e.g. of a nonlinear elastic problem) and the system offunctional equations (as in the present problem) is illustrated in Figures 3a,b. I n the case of algebraic equations (Figure 3a), after the first value Llq(l) has been found, the correspondiag (b) Q R (n +1) Figure 3. I teration Process for ( a) A Nonlinear Elastic Solid andfor ( b) An Elasto- Plastic solid value of Q(q(n) + ,1q(J )) (point A) and the slope K 1 for the next step can be computed exactly. The process is known to converge to the exact solution [31. I n the case of functional equations (Figure 3b), from the value ,1q(I), the value of Q(q(n) + ,1q(I)) can be determined only approximately, i.e., thepointAcannot belocated exactly onthetrue curveQ(q), andthe slope Kj for the next step is also only approximate. Con- sequently, the convergence theorems of theNewton- Raphson method, developed for systems of algebraic equations, are not directly applicable to the present problem. The same objections apply to the modified Newton-Raphson method (in which K i is replaced by the matrix Ko in every step of iteration) as well as to other iterative procedures developed for the finite element analysis of elasto-plastic solids, such as the 'initial strain' method [4) and the 'initial stress' method [51. The treatment of the incremental Equations (31) as asystemof ordinary differential equations (proposed in Reference [9)) and the application of higher-order numerical integration methods does not remove the difficulties outlined above. I n spite of the lack of aformal proof of convergence, the method described in this section has beenshownto yield results of remarkable accuracy. References [6-8) contain a detailed discussion and numerical results which prove, at least heuristically, the validity of the procedure. The problem of analysis of displacements, stresses, and strains in elements made of elasto-plastic metals, subjected to arbitrarily large deformations under the conditions of plane strain has been formulated in terms of the finite element method. The resulting system of integral equations for the nodal displace- ment can be solved by using acombination of Euler's forward integration with the Newton-Raphson itera- tion at each step. The work described in this paper seems to demon- strate the feasibility of the theory of plasticity oflarge deformations and the finiteelement technique insolv- ingcomplex problems (i.e., any shape andanydeform- ation) of the mechanics of elasto-plastic metals. At the same time, certain topics have been exposed as requir- ing further extensive investigations. They are 1. Efficient methods of solution of the integral equations resulting from the application of the finite element technique to the problems of stress and strain analysis. Specially, reduction of computer time and clarification of the nature of convergence appear to be of utmost urgency. 2. Criteria for selection of optimal types of finite elements. Clearly, the experience accumulated in linearly elastic problems is not directly transferable to the present problems. REFERENCES [I ] G. Z. Voyiadjis, 'Constitutive equations for work- hardening elasto-plastic materials', The Arabian Jour- nal for Science and Engineering, 3, No. I (1977), pp. 19-27. [2] O. C. Zienkiewicz, The Finite Element Method inEngin- eering Science. McGraw-Hill, 1971. [3] J . M. Ortega and W. C. Rheinboldt, I terative Solution of Nonlinear Equations in Several Variables. Academic Press, 1970. [4] J . H. Argyris, 'Elasto-plastic matrix displacement analysis of three-dimentional continua', Journal of the Royal Aeronautical Society, 69(1965), pp. 633-635. [5] O. C. Zienkiewicz, S. Valliappan and I . P. King, 'Elasto-- plastic solutions of engineering problems; "I nitial Stress"; Finite Element Approach', I nternationalJour- nal for Numerical Methods in Engineering, 1(1969) pp. 75-100. [6] P. V. Marcal, 'A comparative study of numerical method of elasto-plastic analysis', AI AA Journal, 6 (1968), pp. 157-158. [7] J . H. Argyris and D. W. Scharpf, 'Methods of elasto- plastic analysis', lSD, I SSe Symposium on Finite Ele- ment Technology, Stuttgart (1969). [8] O. C. Zienkiewicz and G. C. Nayak, 'Elasto-plastic stress analysis. A generalization for various constitutive relations including strain softening', I nternational Journal Numerical Methods in Engineering,S (1972), pp. 113-135. [9] Z. P. Bazant, 'Matrix differential equations andhigher- order numerical methods for problems of non-linear creep, viscoelasticity and elasto-plasticity', I nter- national Journal for Numerical Methods in Engineering, 4(1972), pp. 11-15.