W13-79M/831030!47-!?%Oi0010 Perpamon Prw Ltd ANALYTICAL SOLUTION FOR EMBEDDED ELLIPTICAL CRACKS, AND FINITE ELEMENT ALTERNATING METHOD FOR ELLIPTICAL SURFACE CRACKS, SUBJECTED TO ARBITRARY LOADINGS Center for the Advancement T. NISHIOKA and S. N. ATLURI of Computational Mechanics, Georgia Institute of Technology, School of Civil Engineering, Atlanta, GA 30332, U.S.A. Abstract-The complete solution for an embedded elliptical crack in an infinitesolid and subjected to arbitrary tractions on the crack surface is rederived from Vijayakumar and Atluri’s general solution procedure. The general procedure for evaluating the necessary elliptic integrals in the generalized solution for elliptical crack is also derived in this paper. The generalized solution is employed in the Schwartz alternating technique in conjunction with the finite element method. This finite element-alternating method gives an inexpensive way to evaluate accurate stress intensity factors for embedded or elliptical cracks in engineering structural components. 1. INTRODUCTION THEPROBLEMS of embedded and surface elliptical cracks have received much attention due to the fact that the actual flaw from which fracture is initiated in a structural component, can be approximated, often by an ellipse or a part of an ellipse. S&e analytical solutions to these problems have been limited to the case of an embedded crack in an infinite solid with a relatively simple loads, many numerical methods have been developed to analyze fracture susceptibility of embedded or surface flawed, three-dimensional engineering structures. However the agreement on the stress intensity factor solution at the deepest point of semi-elliptical surface flaw was extremely poor. The discrepancies obtained by early methods were sometimes 250%. Recently, a critical evaluation of numerical solutions to the benchmark surface flaw problem has been made in Ref. [I]. In this study, three typical numerical methods such as the Schwartz alternating method, the finite element method and the boundary integral equation method were compared in terms of the stress intensity factors along flaw border. It was concluded in Ref. [ 11that the alternating method was very inexpensive, but it gave about 210% precision which was rather poor comparing with 25% accuracy of the 3-D hybrid crack elementf2-41 and “boundary integral equation” methods[5]. In the alternating method[6,7] the analytical solution, which is a basic solution required in the alternating technique, has been limited to a cubic pressure variation on the crack surface[8]. This limitation is thought to be one of the major reasons for the inaccurate solution through the alternating method. Since 1971, no work has appeared in literature to generalize the solution in Ref.[8] to an arbitrary pressure variation on the crack surface, due to the seemingly insurmountable mathematical and algebraic difficulties. Recently, Vijayakumar and Atluri[9] have derived a general solution procedure for an embedded elliptical crack, subject to arbitrary crack-face tractions, in an infinite solid. The first part of the present paper, a more detailed solution, as well as a general procedure for the evaluation of the required elliptic integrals, is derived. The solutions in Ref. [9] and the present paper represent a generalization, hitherto thought to be unachievable, of the potential representation of Segedin [ lo] and Shah and Kobayashi [S] to solve both the problems of arbitrary normal and shear tractions on the crack face. The major objective of the latter part of the present paper is to show that when this generalized solution is implemented in the alternating method in conjunction with the finite element method, the alternating method becomes a very inexpensive procedure for routine evaluation of accurate stress intensity factors for elliptical cracks in various structural components. In order to save computational time, several special techniques are developed in the present paper. Results obtained by the present finite element alternating method are compared with the benchmark estimate[l] and other results available in literature. 241 248 T. NISHIOKA and S. N. ATLURI 2.POTENTIALFUNCTIONSFORANELLIPTICALCRACK IN AN INFINITE SOLIDWITH ARBITRARY CRACK-FACETRACTIONS Using the well-known Trefftz’s formulation[ll], the problem is reduced to finding the appropriate potential functions [9]. Suppose that x1 and x2 are Cartesian coordinates in the plane of the elliptical crack and x1 is normal to the crack-plane as shown in Fig. 1, such that: (x,/a,)z+ (Xl/tlJ” = 1, a, > a? (1) describes the border of the elliptical crack of aspect ratio (a,/~). The ellipsoidal coordinates &(a. = 1.23) are defined as the roots of the cubic equation +-(g&-(&)-($=0 (21 where we write eqn (2) in an alternate form: w(s) = P(s)lQ(s) (3) P(s) = (s - 5Xs - MS - 5J; Q(s) = s(s + a$(s+ a:) (4) and where The elliptic boundary (1) in the plane x3 = 0 corresponds to the curve & = & = 0. The crack surface itself, namely, the region inside ellipse (1) in the plane x3 = 0 is given in a simple manner by the surface & = 0. The three potential functions f,(a = 1,2,3) of the Trefftz’s formulation[9,11] are taken to be: (5) where k+l+l [w(s)1 ds -\/acs>’ (61 In ew (5) Cu,k,lare undetermined coefficients and the commas are introduced for convenience only. Fig. I. Elliptical crack in an infinite solid. 249 Solution for embedded elliptical cracks Denoting by fu,a the partial derivative of f, with respect to x,(/3 = 1,2, 3) we write for the first, second, and third partial derivatives of f, as: The components of displacement ui and stress “ii in terms of J”~(CX = 1,2,3) are given by UI= (1 - 2Wl.j +A,,) - (3 - 4V)fl.j -t x3(0,, (loa) &?= (I - 2yN2.3 + f3.2)- (3 - 4v)f*,3+ %(V.AQ (lob) (1Oc) and (IN A @_ wf3.22+ 2vf3.,, - 2f2.32 - 2vf1,31+ X3tV.fM (Ilb) 2/N - 2v)f3,,2 - u - VW,,*3 +.fxJ f x3tmu1 (llc) 72 = ~I-2 = (Ild) c33 = 2/-d-f3.33+ x3(vh,331 @3, = 2Y[ - (1 - VIf1.33 + dfl,ll +f2,2,1 f .@a31 UW C3?= b.4 - (1 - v)f2,33+ d_fr,,*i- f2.22)-I-xm,231 (1W and w v.f^ = f,,, +f*,*+f,,rc where Jo and v are the shear modulus and Poission’s ratio. By successive differentiation, it can be shown from eqn (6) that, since ~(5~) = 0, Fir,= I cc 8k+‘mk+‘+’ Ei axfad -= ds -- dQW = ak,al,am,gk+i+l f c3 1 ’ 3 ds ~Q(s) (13) wherein k, = k; 1, = 1; m, = 0. (14) In eqn (13), we have used the additional notation that ah implies the jth partial derivatives with respect to x,. Similarly, the first order partial derivatives of Fk, with respect to xa (/3 = 1,2, 3) can be expressed by I Fkl.0= = where (1.9 Ez k, = k + S,,; I, = I+ &; m, = S38; where &,a etc., are the well-known Kronecker deltas. In the case of the second and third order partial derivatives, we derive: (16) T.NISHIOKAandS.N. ATLURI 250 where F;lPy = (k + I + l)! (18) (19) a% Pn=z=-2/(a;+s) (20) (a=1,2,3) in which a3 = 0, and (21) where GO = (k+ I+ M'QW p;,p~p3”,x~,x~x,“, ko(ko-21) (53 - 5A53 - 52) I lo(lo-;) 2PlXl I mo(mo-1) 2P2X2 .\=t3 ko=k+8,p+8,y;10=i+S~p+S2y;mO=S38+S3~ k, = k,+ S,,; I, = I,+ Szs; m, = m,+ i& . (22) . The partial derivatives of F’& in eqn (21) are given in Appendix. It is noted that the above derivatives are needed (i) in satisfying the boundary conditions on the crack-face and (ii) in evaluating the far-field stresses in the solid containing the elliptical crack which is subject to arbitrary tractions. It is now seen from eqns (13) to (21) that we need to evaluate a generic integral of the type: 11 a,kl a,a;lh k+l+l ds ec’ To accomplish this, we expand Ok+“’ in terms of x2 and carry out term by term differentiations. Then we get 2q-zr-I, x,2pmZq-k, (r)! (2p-2q-k,)!(2q*22r-I,)!(2~1m,)! . (2p - 2q)! (2q - 2r)! (2r)! (p-q)! (q-r)! 2r-m, where (.) denotes a multiplication, and Thus one of the key algebraic steps in the successful application of the analytical work in Ref.[9], in conjunction with an alternating method, is the evaluation of generic elliptic integrals in eqn (25). 3. SYSTEMATIC PROCEDURE OF EVALUATION OF THE ELLIPTIC INTEGRALS -Ip-+q_,,, In general, the integrals of eqn (25), for a given set of parameters p, q, and r, can be evaluated in terms of incomplete elliptic integrals of the first and second kinds, and Jacobian elliptic functions[l2]. Although the closed-form expressions for lower order components of .l,_,, q-r,r were given in Ref. [7,8], as pointed out in Ref. [7,8], deriving the closed-form expressions of the elliptic integrals involves 251 Solution for embedded elliptical cracks exorbitant, if not unpleasant, algebraic work. Therefore, it is important to develop a systematic generic procedure to evaluate these integrals for arbitrary values of p, 4 and r. Thus, we rewrite eqn (25) using Jacobian efliptic functions[l2] as: Jp-q,q-r.r= F = ;y MI (sn”Pu)(nd2q-“u)(nc”U) du I0 2 2 Lp.q-r.r (26) where SFI’Uj = L&f+ 53). (27) The following identities for Jacobian elliptic functions are used. sn’u + cn’u = 1; k2sn2u + dn’u = 1; dn’u - k2cn2u = k12; ki2sn2u i- cn2u = dn’u 1 1 ndu = dnlr_ncu=G:sdu=e (28) dnu where k? = (0: - a;)/& k’?= 1 - k*. (29) We may derive by using integration by parts in eqn (26) that: L,, 4-r, I = C2r_Ll)k,2 {(sn2p+‘u)(nc2r-i u)(nd2q-“-‘u)l$ + [2(-p f r - 1)+ 2(p - q - r + 2)kZIL,,_, r-j t k’(-2p + 2q - 3)L,,q-,,-,I. (30) Thus, we need the starting values of L,,,_,.,_, and Lp,y-r,r_2 to evaluate Lp,U_r.r.The lowest order starting values are: &.,-I L, q, -2 = = U’(sPr2Pu)(nd2qu)(nc-2u)du I0 u’(sn2PU)(nd’4U)(nc-4U)du.. (314 @lb) The above integrals can be reduced as follows: j+Y+2/p(2-Y~p r.,.,.-2=&x& c2 ((p- 1)- jf!j!(2j_o y_o y-J y)!~!~~(q_j-y~ (3%) where I,, = 1?m+2 = u’nd% du 2m(2- k’)J,, + (1 - 2n1)1~,,,_~-kZmulcnulnd2mi’ul (2m + l)kf2 (33) (34) T. NISHIOKA and S. N. ATLURI - Pfesent qo,o Closed-Form Systematic Procedure Expressions 17,81 Fig. i. Elliptic integrals obtained by the present systematic procedure. For 2(p - j - y) < 0 in eqns (32a) and (32b), we find I_*, = Gzm,where G2mct _ - ulsnu,cnu, k2dnZm-’ -I (1 - 2m)k”G~~_~+ 2mC2 - k21Gz, (2m + 1) (35) Thus, finally we see that one needs the following starting values for evaluating the general terms of I 2m+2Jand G2m+2: I, = G,, = F(u,) = u1 I,=~IE(u,)-kZsnu,cdu,J Gz = E(u,) (36) where Ffu,) and E(u,) are incomplete elliptic integrals of the first and second kinds respectively, The procedure shown in eqns (2&o@ is implemented in a computer program to evaluate the elliptic integrals numerically. The variation of numerical values of the elliptic integrals JI,, with the parameters 1, m, and n is shown in Fig. 2. To check the validity of the present procedure and computer program, the integral values are compared with those obtained by the closed-form expressions in Ref.[7,8]. For the lower order components of J,,,,, the present integral values are exactly the same as those from the closed-form expressions. For the higher order components of Jr,,,,, the integral values change monotonically with the parameters. 4. RELATION BETWEEN CRACK-FACE TRACTIONS AND POTENTIAL, FUNCTIONS Let the tractions along the crack surface be expressed in the form (37) so that the values of (i, j) specify the symmetries of the load with respect to the axes of the ellipse. The solution in terms of the potential function is assumed in the form (38) 253 Solution for embedded elliptical cracks The potential functions expressed by eqn (5) are to be understood as that k and I replace (2k - 21+ i) and (21-t j), respectively. In the Trefftz formulation[ll] the boundary conditions can be expressed in terms of the potential functions: t39d in which the boundary condition for fj is uncoupled from f, and fi. However, if eqns (39a) and (39b) are directly used as in Ref. [9], there will be singular terms in the equations relating the coefficients C of eqn (37) to coefficients A of eqn (38). To overcome this difficulty, we use alternative forms for the boundary conditions, Since f,(a = I, 2,3) are harmonic functions, f&33= -L, LI - f,.zz(o = 1,273). (40) Then, eqns (39a) and (39b) can be rewritten as follows: Substituting eqns (37) and (38) into eqns (41a, b), we obtain the follo~~ing linear algebraic equations upon comparing coefficients of like powers in the polynomial series. For the Mode I problem, m=O,l ,..., M n=O, l,..., m. WW For the mixed problem of Modes II and III, n=O,l,...,m (4%) where L,=(m-n+k-I+i),L,=nt/+j, I”.j~=(2k+i+jt1)!(2L,+2S,,i!(2L,?+26,”)!(lY_1 a (k-m)! (IL, + S,,)! z”.” I2 = (2ki-itj-tl)! (k-m+i+j-l)! (43) 2) (Lz + 8&J! (2L,+2-2i)!(2L,t2-2j)! (L,tl-i)! (&+1-j)! ’ (44 (45) and the elliptic integrals J,,,,, LZ,0(O), etc are generally defined in eqn (25). It should be noted that the elliptic integrals Jt_,+$.L,.O(O),etc. in eqns (42a) and (42b) involve no singularities. 2% T. NISHIOKA and S. N. ATLURI The relation between the parameters A and parameters C in eqn (42) can be summarized in a matrir form: 14 = IN Nxl ICI (46 NxN Nxl where N is the total number of coefficients A or C. For a complete polynomial expressed by eqn (37) the maximum degree of the polynomial M, and the number of coefficients N can be expressed respectively, by M, = 2M + 1 and N = (&f + 1)(2&f+ 31x3. For an incomplete polynomial, the maximum degree of poIynomia1 and the number of coefficients depend on not only the parameter M but alsc parameters i and j in eqn (37). 5. STRESS INTENSITY FACTORS Once the coefficients C are determined by solving eqn (46) for given loadings on the crack surface, the stress intensity factors corresponding to these loads are computed from the following equationr9]. For the Mode I problem, AlI4 2i 5 i i=Q k=() ,=,) j=” (-2)2k+i+i(2k + i + j + I)! (471 where 6 is the elliptic angle and A = a: sin2 $ + ai cos2 8. (48) For the mixed mode problem of Modes II and III, in which H, =: i 232 (_2)2k+i+j (2k + i + j .+ I)! i=Oj=Ok=Ol=O H, = (~)2k-2’+~(~)2’+i Cy;iLl., f.sl) ‘. i 1: $ i ix0 j-0 k=O I=” (_2)2kih-J(2k + 3 _ i _ j)!(~)*k~*‘+‘-i(~)zi+‘~i c~!;“:,;J) ($2) 6. FINITE ELE~~T ~TE~AT~G METHODS In the three-dimensional Schwartz alternating method for elliptical crack problem, which was originally developed by Kobayashi et a/.[& 71, two analytical solutions were required as follows: The elliptical crack subjected to cubic normal loading on the crack surface, in an infinite solid. This limitation was imposed due to the non-availabihty, until the present work, of a general solution for a general polynomial type normal loading on the crack face. I’A similar method can be found in Ref.[19].However, in Ref. [19], the analytical solution for the elliptical crack subjected to cubic normal loading[7,81 was used, while the generalized solution described in the present paper is employed in the present method. Additional information on finite element alternating method may be found in Ref. [l9J. 255 Solution for embedded elliptical cracks Solution 2 A semi-infinite body subjected to uniform normal and shear stress over a rectangular portion of the surface. However, as mentioned earlier, Solution 1 has been limited only to cubic polynomial variation along the crack surface. This is one of the reasons for causing error in the classical alternating method[6,7]. Moreover, the use of Solution 2 has restricted the application of the classical alternating method to only media bounded by straight surfaces. To overcome these difficulties, Solutions 1 and 2 can be replaced by Solution 1 The elliptical crack subjected to arbitrary normal and shear loading on the crack surface, in an infinite solid, as presently discussed; Solution 2 A general numerical solution technique such as the finite element method or boundary element method. Solution 1 is explained in the earlier part of the present paper. In the present paper the finite element method is used to generate Solution 2 because of its simplicity. The present finite element alternating method requires the following steps as shown in Table 1. (1) Solve the untracked body under the given external loads by using finite element method. The untracked body has the same geometry with the given problem except the crack. To save computation time in solving the finite element equations, a special solution technique is implemented. This will be explained later. (2) Using finite element solution, we compute the stresses at the location of the original crack. (3) Compare the residual stresses calculated in Step (2) with a permissible stress magnitude. In the present study one percent of the maximum external applied stress is used for the permissible stress magnitude. (4) To satisfy the stress boundary condition, reverse the residual stresses. Then determine Table 1. Flow chart for finite element-alternating technique Solve the untracked body under external by using finite loads element method (FEM) Using FEMsolutions compute stresses t fitting crack face at NO stresses in step (2) I t step 5 Step 6 step 7 ,crac!i. Determine coefficients C in the potential functions c Calculate the k-factors current iteration for ‘+ Calculate residual stresses an external surfaces of the body due to the loaded Reverse them and calculate equivalent nodal farces. i step 8 the Consider nodal forces in step (7) as external applied theloads acting on the untracked body I Add the k-factor solutions iterations of I 1 256 T. NISHIOKA and S. N. ATLURI coefficients A of eqn (37) by using the following least squares method: I,= f s. (&- &)2dS (a = I, 2,3) (53) where &a is the reversed residual stresses calculated by the finite element solution, ai:! is defined by eqn (37), S, is the region of the crack, and I, are the functionals to be minimized. Rewriting eqn (37) in a matrix form: and substituting eqn (54) into eqn (53), we obtain the relation between the coefficients A and the residual stresses: where (5) Determine the coefficients C of eqn (38) in the potential functions by solving eqn (46) = [K’U]). (6) Calculate the stress intensity factors for the current iteration by substituting coe~cients C in eqns (472-02). (7) Calculate the residual stresses on external surfaces of the body due to the loads in Step (4). To satisfy the stress condition, reverse the residual stresses and calculate equivalent nodal forces. These nodal forces {Q} can be expressed in terms of coefficients C: (0 lQh,, = -IGlmCCl (58) [Gl, = j- I~lT~~lbI dS (59) and S “8 where m denotes the number for finite element, [IV] is the matrix of isoparametric element shape functions, [n] is the matrix of the normal direction cosines and [p] is the basis function matrix for stresses and can be derived from eqn (11). In order to save computational time, the matrix [G], is calculated prior to the start of iteration shown in Table 1. (8) Consider the nodal forces in Step (7) as externai applied loads acting on the untracked body. Repeat a11steps in the iteration process until the residual stresses on the crack surface become negligible (Step 3). To obtain the final solution, add the stress intensity factors of all iterations. As shown in the flow chart for the finite element alternating method, we need to solve the following type of finite element equations: [Us”, q’, . . . t q’7 = lQ”, Q’, . . . , Q”1 (60) and Qi = Qi(qi-‘): i = I, 2,. . . , n (61) in which the superscript denotes the cycle of iteration, [K] is the global (assembled~ stiffness matrix of Solution 257 for embeddedellipticalcracks body and remains the same during the iteration process, and q’ is the nodal displacement vector for ith iteration. Q’ is the nodal force vector for ith iteration and depends on the solution for the previous iteration qi-’ as expressed by eqn (61). An efficient equation solver OPTBLOK developed by Mondkar and Powell[l31 is used to save computational time in solving eqn (60). The solution algorithm is devided into three parts, i.e. (i) reduction of stiffness matrix, (ii) reduction of load vector, and (iii) back substitution. In OPTBLOK, the reduction of stiffness matrix is done only once, although the reduction of load vector and back substitution may be repeated for any number of load cases. Thus, denoting CPU time for each part by T,, T2, and T3 respectively. The total CPU time T in solving eqn (60) using OPTBLOK can be expressed the untracked by T=T,+(nt1)(TztT3)=(T,tTztT7)tn(TZtT-J (62) where n is the total number of iterations. Since T, is much larger than (T2 t T3), a great amount of computational time can be expected to be saved by comparing with the case in which eqn (60) is solved for each iteration (T* = (n t l)(T, t Tz t TJ). To illustrate this situation, we consider the example given in Ref.[13]. For a set of linear equations with the number of equations of 1960, and half band width of 200, the CPU time for reduction of load vector and back substitution was about 5.6% of the total CPU time (T, t T3 = 0.056T). Since for a typical problem the present alternating method needs 3 iterations (n = 3), the additional cost in this case is only about 16.8%, which is considerably smaller than 300% in the case when eqn (60) is solved for each iteration. 7. ANALYSES OF EMBEDDED CRACKS All numerical analyses given in the present paper concern the Mode I crack problem. A linear elastic material with Poisson’s ratio 0.3 is used for all analyses. All calculations were made by using a CDC CYBER 74. 7.1 Circular crack in an infinite solid We consider an infinite solid containing a penny-shaped crack, the faces of which are subject to normal stresses of the type (i) ai? = -pox: t pox:, and (ii) o-1-2= -3p,x:x2 t pox:. The known analytical solution@] for these problems are (i) KI = 16p, cos e/(&r), and (ii) KI = 32p, sin 38/35&) respectively. The analytical solution for the circular crack (~/a, = 1) for arbitrary loadings can be rederived from the solution presented in the present paper by taking the proper limits as k + 0 and rearranging the equations extensively. In the present analysis for convenience, the above crack was modeled as an ellipse, with (&/a~) = 0.982. It is seen from the results presented in Figs. 3(a, b) that the present 10, =33 = -3P,x, 2 x* + p,x: 0: 5.. a0 \I Y oc ,- -0 5 (a) Fig. 3. Stress intensity (b) factor for a circular crack in an infinite solid and subjected CT’:,’ = - P,xT t P,,x;, and (b) CT\!)= - 3P,,x:x2 t P,,x:. to pressure: (a) 758 T. NISHfOKA and S. N. ATLURI analytical solution for the nearly circular crack with (ar/a,) = 0.982 is in excellent accord with the analytical solution[8] for the circular crack (aJar = 1). Especially, both the solutions are practically the same around 8 = 0”. 7.2 Circular crack in a finite ~0~~~ bar Consider a round bar with an embedded circular crack and subjected to pure tension or pure bending at the ends of the bar. The crack is located at the center of the bar as shown in Fig. 4. To satisfy the boundary conditions on the external surfaces of the bar, the finite element alternating method was used with the present analytical solution for the nearly circular crack (u2/uI = 0.982). Figure 5 shows the finite element breakdown for the untracked round bar. Only one-eighth of the round bar is modeled due either to symmetry or to antisymmetry. The prescribed displacements imposed on the finite element model for the untracked bar are also shown in Fig. 5. The analytical solution used as Solution 1 in the present alternating method was that corresponding to eqns (37) and (38) with (A4 = 2, i = 0, j = 0) for the tension problem and with (M = 2, i = 1, j = 0) for the bending problem. The matrix [G] given in eqns (58) and (59) is calculated on the surfaces of r = R and x3 = L, prior to the start of iteration as shown in Table 1. Normalized stress intensity factors for various geometries under pure tension as well as pure bending are summarized in Table 2. For the tension problems, the present results differ less than 0.6% from other results obtained by an axisymmetric finite element method[14], and integral equation method [15]. For the bending problem, the present result differs by - 1.l% from an approximate solution obtained by Benthem and Koiter[tft]. Since Benthem’s approximate solution differs by 0.8% from Sneddon’s solution, the difference between the present result and true solution may be expected to be much smaller. Figure 6 shows the variation of residual stress on the crack surface with each cycle of iteration in the alternating technique (see Step 2 in flow chart). It is seen that the residual stress decreases rapidly and monotonically with the number of iterations. I 0982R Fig. 4. Fig. 5. Fig. 4. Circular crack in a round bar subjected to remote tension or remote bending. Fig. 5. Finite element breakdown of an untracked round bar Solution for embedded elliptical cracks Table ?. Comparison of stress intensitv factors for a circular crack in a round bar / I / LOAD I a/R Others i---- I ITfXlSiOtl o.5 i 1.114 (Yamamoto) 1.402 (Yamamoto) 1.072 1.081 (f0.8%) (Sneddon) (Benthem) 0.6801 (Benthem) I L/R=- L/R=- Tension a/R=05 L/R=1 0 I,0 09 0 I23 4 Cycle of lterotlons Fig. 7. Fig. 6 Fig. 6. Variation of residual stress on the crack surface. Fig. 7. Elliptical crack in a finite-thickness plate. 7.3 Elliptical crack in a finite-thickness plate The geometry of a finite-thickness plate with an elliptical crack, subjected to tension loading at the end of the plate is shown in Fig. 7. Shah and Kobayashi[17] have solved a similar problem, in which the plate has the same crack lengths and thickness as here, but has an infinite length and width. Figure 8 shows the finite element breakdown for the untracked plate. Due to the symmetries of the plate, only one-eighth of the plate was model by finite elements. The matrix [G] given in eqns (58) and (59) is calculated on the surfaces of xl = W, x2 = h, and x3 = L, prior to the start of iteration as shown in Table 1. The variation of the stress intensity factor obtained by the present method is shown in Fig. 9 and compared with the result obtained by Shah and Kobayashi[l7]. The stress intensity factors are normalized by that of the crack in an infinite solid. The present result agrees excellently with Shah and Kobayashi’s result especially near the surface of the plate (6 = 90”) where there is more of the effect of finite plate thickness. It is noted that in the present finite element alternating method six terms (1, x:, x:, XT, x:x:, x$ for &)(A4 = 2, i = 0, j = 0) were used to satisfy the boundary condition of the crack, while in the classical alternating method [ 171only three terms (1, xi, xi) for a$: were used. The CPU time for the analysis of an embedded crack in a finite body was about 800 sec. with a CDC CY BER 74. T. NISHIOKA and S. N. ATLURI 260 30 Elliptml 60 72 Angle 0 (Degree) Fig. 9 Fig. 8. Fig. 8. Finite element breakdown for an untracked finite-thickness plate. Fig. 9. Stress intensity factor for an elliptical crack in a finite-thickness plate; a:/a~ = 0.4. az/h = 0.X 8. ANALYSES OF SURFACE CRACKS Next, we consider a semi-elliptical surface flaw in a plate under remote tension and bending as shown in Figs. lO(a, b). This problem is one of the “benchmark” problems proposed by the Three Dimensional Fracture Analysis Workshop at Battelle Columbus Laboratories in 1976[18]. A critical evaluation of various numerical solutions to the benchmark problem has been recently made in Ref. [l], to establish a “best estimate” of the stress intensity factor variation along the flaw border. The derived best estimate curve for the stress intensity factor is believed within F 3% of the actual value along the crack front. fb) Fig. 10. Benchmark semi-elliptical surface flaw problem; (a) remote tension and lb) remote bending; a2i2al = 0.25, H/ W B 2.0, a!/ W S 0.2. 261 Solution for embedded elliptical cracks x3t - r H/w=20 !3Jw =02 a,/20,:025 0,/t =o 25 ” =03 80 459 elements nodes - Fig. 1I. Finite element breakdown for an untracked plate. The results of the present analyses are compared with the best estimate of magnification factors defined by the following equations: for the tension problem, _ F, = E& I-J We) (63) a: sin” 0 t a: cos2 0)“4 for the bending problem, where E(k) is the complete elliptic integral of the second kind, and stresses u,, and C* are defined in Figs. IO(a, b) respectiveiy. The typical finite element model used for the untracked plate is shown in Fig. 11, which consists of 1377 total degrees of freedom (before imposition of boundary condition) and total number of 80 finite elements. The matrix [G] in eqn (58) is calculated on the surfaces of x1 = W, x2 = 0, x2 = t, and x1 = H, prior to the start of iteration process. 8.1 Residual stress distribution on semi-ellipticul crack In fitting polynomials in a bounded region using the least squares method, it is well known that accuracy of fitting in the fitting region can be increased with the increasing number of polynomial terms; however, in the region outside of the fitting region the fitted curve may change drastically. In the present alternating method, as explained earlier, the solution for the entire elliptical crack in an infinite solid is impIemented as Solution 1. 262 T. NISHIOKA and S. N. ATLURI x,=0 TYPE C 1._ r____.._ ji___^^__ 02 0 0, x2 08 i orA ‘I 50 Elliptical Fig. 12. Ankle B (Degreel Fig. 13 Fig. 12. Distributions of Residual stress over the entire elliptical crack surface. Fig. 13. Magnification factors with various prescribed fictitious residual stress distributions; 42~1, = 0.25, a>/t = 0.25, remote tension. For these reasons, in Step 4 of the alternating method, it is necessary to define stress over the entire crack plane including the portion of the crack which lies outside of the finite body. Numerical experimentation is done for arriving at an optimum pressure distribution on the crack surface extended into the fictitious region. Three types of fictitious stress distributions are prescribed as shown in Fig. 12. The results obtained by the present alternating method are compared in terms of the magnification factors as shown in Fig. 13 for the tension problem and Fig. 14 for the bending problem. Twelve terms of the fifth order polynomia1 (M = 2, i = 0, j = 0,l) in eqn (37) were used for c$. All the present results agree excellently with the benchmark estimates [ I]. In fact, the differences of the present results from the benchmark estimates are within the error band of the benchmark estimate as shown in Figs. 13 and 14. The variations of residual stress on the crack surface with each cycle of iteration in these analyses are compared in Fig. 15. It is seen from the figure that Type A gives unstable convergency while Types B and C give monotonic convergencies. Moreover, Type C is the easiest way to prescribe the ~ctitious stress distribution especially for a part-elliptical crack or corner quadrant-elliptical crack. Therefore Type C is used for the following analyses. To check the effect of the degree of polynomials for the applied stress a\? in eqn (37), the stress intensity factors are also obtained by using the cubic order polynomial (M = 1, i = 0, j = 0, 1). The results are compared with those obtained by the fifth order polynomial in Fig. 16 for the tension problem and in Fig. 17 for the bending problem. It is clearly seen that the stress intensity factors become closer to the benchmark estimate [ll when the higher order polynomial is used for the fitting of the residual stress in Step (4), (see Table 1). 8.2 Analysis of the “benchmark” problem For the geometry of the plate containing a semi-elliptical crack as shown in Fig. 10, further analyses are made only changing the ratio a,/t from 0.25 to 0.5 and 0.75. The variation of the ma~ification factor 263 Solution for embedded elliptical cracks TYPE A BendIng - Present ‘--K Bwchmarh Estimate (1pRo, =0.25 0,/t =0.25 o,/w=0.2 H/W=20 with Error Ba7d (I ) 14 13 12 I I I TYPE B L ‘... ..,,_ IO 09 08 07 -.. ..,, .‘.. ‘....,:‘y... .._....y- ...._..___ __ -L -‘-..__.__ -.---.._._.._...... 0 30 60 Elliptical 90 Angle B (Degree) Fig. 14. Magnification factors with various prescribed fictitious residual stress distributions; az/t = il.25, remote bending. from IO 0,6 08 o TensIon; d”‘=~ A &“‘=a, Bendlng; m 0,7 0.6 0.6 TYPE B o.6 TYPE 05 00, 000 a2/2a, = 0.25, L 012345 0 Cycle Fig. 15. Comparision I of 2 3 4 Iterations of residual stress variations. 01234 C 264 T. NISBIOKA and S. N. ATLURI IS a,&=025 t a,/t=0.25 “--DOP=3 - Present 0 Benchmark o<o+ 30 Elliptical 60 Angle f If Estimate 90 8 (Degree) Fig. 16. Fig. 16. Magnification factors Elliptical Angle @ (Degree) Fig. 17. with various degree of polynomials in the fitting of the residual stress; az/2al = 0.25, a& = 0.25, remote tension. Fig. 17. Magnification factors with various degree of polynomials in the fitting of the residuai stress; a2/2al = 0.25, aa/: = 0.25, remote bending. for az/t = 0.5 is shown in Fig. 18 for the tension problem and in Fig. 19 for the bending problem. For the tension problem the present result agrees well with the benchmark estimate [ I] at the deepest point of the crack (0 = 90”),while the present result is higher than the upper error bound of the benchmark estimate at the surface of the plate (8 = 0)o. For the bending problem the present result is close to the upper error bound of the benchmark estimate at both the points of B = 0” and t?= 90”. The variation of magni~cation factor for a,/r = 0.75 is shown in Fig. 20 for the tension problem and in Fig. 21 for the bending problem. As seen from the figures all the present results agree well with the benchmark estimate except near the plate surface. Figure 22 shows the variation of the magnification factor F,,, for the tension probtem with the fractional crack depth. At the deepest point of the crack (e = 90”) the present results agree excellently with the benchmark estimate[l] while at the surface of the plate (0 = 0“) the present results are on the upper error bound of the benchmark estimate. Figure 23 shows the variation of the magnification factor Fh for the bending problem with the fractional crack depth. As seen from the figure the present results agree well with the benchmark estimate[l] at the deepest point (6 = 900). At the plate surface (0 = 00) the present results give higher values of Fb for a&t = 0.5 and 0.75 than the upper limit of error band shown in Ref. [l]. It is noted that the scatter bounds of the numerical solutions compared in Ref. [ 11were several times bigger than the estimation error bands. Therefore it may be concluded that the present finite element alternating method gives a comparable precission to the 3-D hybrid crack element[2,41 and the boundary integral equation method[5]. The CPU time for the analysis of surface crack was about 850 sec. with a CDC CYBER 74. 9. CONCLUSION The complete solution for an embedded efliptical crack in an infinite elastic solid and subjected to arbitrary tractions on the crack surface is rederived from Ref.[9]. While the solution can be reduced to a 1‘0 ” 0 . 30 Angle 60 90 ) e (Degree) Benchmark Estimate with Error Band ( I Present Ellipttcol .. .. . .. ...0 ..I - _... - .I --- - - -. Fig. 18. Magnificationfactor for a surface flaw in a plate: a2/2al = 0.25, az/t = OS, remote tension. ‘;; .-s t 2 b Lz a,/t=O.5 - I 30 Angle f?encl#nark with Error Resent Elliptical --o-- - /t =0.5 60 ) 90 e (Degree) Estimate Band ( i _.. - - Fig. 19. Magnification factor for a surface flaw in a plate; 0212~ = 0.25, azlt = 0.5, remote bending. 0 I.5 a2 Elliptical 30 with 60 90 B (Degree) Band (I 1 Angle Error Benchmark Estimate Present a2/2al = 0.25, al/t = 0.75, remote tension. Fig. 20. Magnification factor for a surface flaw in a plate; 0 - : : 30 Angle Present Elliptical : - =0,75 8 (Degree) 60 90 Fig. 21. Magnification factor for a surface flaw in a plate: a2/2al = 0.25, ap/t = 0.75, remote bending. 0 ’ I 0.0 15 0,/t 001 0.0 .’ 1.0’. = U 04 Depth Crack Error 02 with Benchmark Present a,/t 06 0.8 (I) Estimate Band Fig. 22. Variation of magnification factor with the fractional az/t remote tension, u2/2a, = 0.25. 0” I c ;Ff 5_ E b 0 2 LLY 1.5 2.0 crack depth IO Depth 0.4 : factor a,/t Q6 with i + I.0 the fractional 0.8 ) depth a:/t; remote bending, aJ2a1 = 0.25. of magnification Crack : 0.2 Fig. 23. Variation Preseni Benchmark Estmate ..._... with Error Band ( I 00 t 0 001 : ’5 i crack Solution for embedded elliptical cracks 267 closed-form solution for a relatively simple loading such as constant or linear variation of the tractions, for a high order polynomial variation of the tractions the solution procedure requires a digital computer. The genera] evaluation procedure for the necessary elliptic integrals is also derived. The integral values obtained by the present procedure agree exactly with those obtained by the closed-form expressionj7,Sl. The alternating method in conjunction with the presently derived analytical solution and the finite element method is developed, to analyse an embedded or surface elliptical crack in an arbitrary-shaped finite solid. The present finite element aIternating method Leads to accurate evaluation of stress intensity factors and is about one order of magnitude inexpensive in computing costs as compared to those with the hybrid finite element method [2,4]. It is also demonstrated that the stress intensity factors obtained by the alternating method can be improved when the degree of polynomials in the applied stress for the analytical solution is increased, However, the advantage of using the presently developed higher order solutions can be more dramatically demonstrated in problems wherein the stress solution for the untracked solid, at the location of the crack, is more complex than for problems solved here. Such problems, as for instance of pressurized cylinders with internal or external surface cracks, or of cracks at pressure-vessei-nozzle junctions, will be discussed in a forthcoming companion paper. Acknowledgements-The authors gratefully acknowledge the financial supportfar this work provided by U.S.A.F.O.S.R. under grant 81-0057to Georgia Institute of Technology. They appreciate the timely encouragement of Dr. Anthony Amos, the responsible AFOSR program official. They thank Ms. Margarete Eiteman for her care and patience in preparing this manuscript. REFERENCES [I] J. J. McGowan (Ed.), A critical evaluation of numerical solutions to the “benchmark” surface problem. SESA Monogruph (1980). 121S. N. Atluri and K. Kathiresan, Stress analysis of typical flaws in aerospace structural components using three-dimensional hybrid displacement finite element methods. AIAA Paper78-513, Proc. A~AA-ASPS 19th SMD Conf.Bethesda MDpp.340-350.(Aug. 1978). 131S. N. Atluri and K. Kathiresan, Three dimensional analysis of surface flaws in thick walled reactor pressure vessels using displacement-hybrid finite element methods. N~cfear ~~gi~g~e~ig~51,163-176(1979). [4] S. N. Atluri and K. Kathiresan, Stress intensity factor solutions for arbitrarily shaped surface flaws in reactor pressure vessel nozzle corners. Itlt. J. Pressure Vessels Piping 8, 3 13-332(1980). [S] J. Heliot, R. Labbens, A. Peiiissier-Tanon, Benchmark problem no. l-semi-elliptical surface crack-results of computation. Znt.J. Fracture 15, R197-R202(1979). 161A. S. Kobayashi. A. N. Enetanyaa and R. C. Shah, Stress intensity factors of ellipticai cracks. P~spec~s oi Fracture ~er~~u~ics (Edited by G. C. Sih, H. C. Van Elst and D. Broek), pp. 525-544.Noordhoff International, Leyden (1975). [7] R. C. Shah and A. S. Kobayashi, On the surface flaw problem. The Surface Crack: Physical Probfetns and ~um~utati(~~uf Solutions (Edited by .I. L. Swedlow), pp. 79-124. The American Society of Mechanical Engineers (1972). 181R. C. Shah and A. S. Kobayashi, Stress intensity factor for an elliptical crack under arbitrary normal loading. Engng Fracture Mech. 3, 71-96 (1971). [9] K. Vijayakumar and S. N. Atluri. An embedded elliptical flaw in an infinite solid, subject to arbitrary crack-face tractions. J. Appl. Mech. 48, 88-96 (1981). IlO] C. M. Segedin. Some three-dimensional mixed boundary-value problems in elasticity. Report 67-3. Department of Aeronautics and Astronautics College of Engineering, University of Washington, June 1967. (111E. Trefftz, Kandbueh der Pk~sik, Vol. 6, u. 92. Snrin~er-Verbal. Rerlin (192%. i 12j R. F. Byrd and M. D. Friedman, Handbook of Eifipts Infegr& for En&&s and Scientists, Springer-Verlag, Berlin (1971). (131D. P. Mondkar and G. H. Powell, Large capacity equation solver for structural analysis. C~~pff~.Structures 4,699-728 (1974). [ 141 Y. Yamamoto and Y. Sumi. Stress intensity Factors in a cracked axisymmetric body calculated by the finite element methcd J. Socl Naval Architerts of Japan 133, 179-187(1973). [W I. N. Sneddon and R. J. Ta& The effect of a penny-shaped crack on the dist~bution of stress in a long circular cylinder. 18~ J. Enninn Sri. I, 391-409f 1963). [16] J. P. Benthem and W. T. Koiter, Asymptotic approximations to crack problems. ~e~~o~~ of A~afysis und Sof~fiuns of Cmk P~/)b~e~~(Edited by G. C. Sib), pp. 131-178.Noordhoff International, Leyden (1973) [ 171 R. C. Shah, and A. S. Kobayashi, Ellipticalcrack in a finite-thickness plate subjected to tensile and bending loading. L Pressure Vessel rerk~ofl~~~: pp. 47-54 (Feb. 1974). [IS] L. E. Hulbert. Benchmark problems for three-dimensional fracture analysis. lnt. J; Fracture 13, 87-91 (1977). [I91 F. W. Smith and T. E. Kullgren, Theoretical and experimental analysis of surface cracks emanating from fastener holes .~FFDI_-TR-7~1~~Air Force Flight Dynamics Laboratory (Feb. 1977). (R&oed 2 &ember 1981;receiaed far ~ubl~&otion t February 1982) APPENDIX & ~~ fi To evaluate the rtress components at a given point (xl, x2, x2) by using a computer, a general evaluation procedure for obtaining the partial derivatives of F&, is also one of the key algebraic steps in the successful application of the present analytical solution. We Spstematic procedure of evaluation of the partial d&&es 268 T. NISHIOKA and S. N. ATLURI rewrite eqn (18) in the following form: F&, = (k + 1+ I)! T, Tz Ts iAl! where 7-c= x$“‘:Tc= p&); The parameter k,, I, and m, are given by eqn (1% T; = p:‘(&); TX= p?‘(<,); Ts = d/Q!&). (A?) Then, using the chain rule of differentiation, the derivatives of F&,, can be evaluated by aFirar_ --,k+l+l)!“&.R. ax6 t ,431 where R, = T, Tz.. T,_, T,+, Denoting 2 6 .Tc,. IA4r by T,J and 5dx by l.s where the required partial derivatives & are given by 2xaQM Q’(s) and P’(s) indicate, respectively, the derivative of Q(s) and P(s) expressed by eqn (4) with respect to s. b.44)