Computational Mechanics 16 (1995) 266 271 9 Springer-Verlag 1995 An implementation of the Schwartz-Neumann Alternating Method for collinear multiple crackswith mixed type of boundary conditions L. H. Wang, S. N. Atluri 266 Abstract This paper presents an implementation of the Schwartz-Neumann Alternating Method for multiple collinear cracks with mixed types of boundary conditions. Numerical examples are also presented to show the effectiveness of the method. 2 The analytical solution The analytical solution used in the present implementation of the alternating method is based on Muskhelishvili's (1953) solution to n collinear multiple cracks in an infinite domain. For the sake of the completeness and the consistency, we include 1 here the fundamental complex variable solution to the collinear Introduction multiple cracks. The Schwartz-Neumann Alternating Method [Atluri (1986) ] has The n collinear multiple cracks, a~bi,i = 1,2 . . . . . n, are been proven to be a very efficient method in the computation assumed to be on the real axis. The x-coordinates of the left and of fracture parameters. Analytical solutions for cracks in the right crack tips of the i'th crack are ai and b~.The upper crack the infinite domain, with arbitrary crack-face loadings, are surfaces have stresses ~§ and rr+,ywhile the lower crack surfaces available in the literature. The Schwartz-Neumann Alternating have stresses ~ and a~r. Muskhelishvili's (1953) solution Method makes use of these solutions in solving fracture for complex potential functions is problems in finite domains with complicated geometries and boundary conditions. The use of the analytical solutions P~(z) simplifies the FEM models used in the analysis. Compared to the r (1) = Co(z) + Z ( ~ -- ~ traditional finite element method, the alternating method can reduce the computational cost significantly. Nishioka and Atluri (1983) developed the alternating method for elliptical Pn(z) ' ~ (2) n ( z ) = no(z) + X ~ Y -~ surface cracks. Park and Atluri (1992) used the alternating method to study the fatigue growth of multiple collinear cracks in a fuselage lap-joint. Nikishkov and Atluri (1994) extended where the alternating method to the elastic-plastic analysis. In this paper, we report the success in the implementation 1 ~X+(t)p(t) 1 q(t) dt of Schwartz-Neumann Alternating Method for mixed types of q)~ t----z d t + ~ i ! t - z boundary conditions. In the above cited previous implementations of the Schwartz-Neumann Alternating Method, it was difficult to handle the displacement boundary 1 ~X+(t)p(t) dt_l_~ q(t) dt conditions. Iterative schemes were needed to satisfy these [2~ tZz 2niLt-z displacement boundary conditions. In removing this restriction through the present work, we can solve problems with arbitrary where boundary conditions as effectively and efficiently as was done for traction boundary value problems in the above cited references. k=l Communicated by S. N. Atluri, 29 March 1995 P,(z)=fiGz k k=0 L. H. Wang, S. N. Atluri Computational Modeling Center, Georgia Institute of Technology, Atlanta, Georgia 30332 and Correspondenceto: S. N. Atluri p(t) = 89 i (t) + try (t)] - ~ [(r~(t) + a~(t)] The work presented herein was supported by a grant for the Federal Aviation Administration to the Center of Excellence for the Computational Modeling of Aircraft Structures at the GeorgiaInstitute q(t) = ~[~; (t) - o ; (t)] - ~i [ G ( t ) - G ( t ) ] of Technology. The integration path L is the union of all the cracks a;b;, i = 1, 2 . . . . . n. The + sign in X + (t) indicates that the upper surface value of the function X(t) is taken in the integration. c, and a are determined by the stress and the rigid body rotation at the infinity, q, k = 0, 1. . . . . n - 1 are determined by the uniqueness conditions of the displacements. in the evaluation of stress, Park (1993) used a set of approximate piecewise constant functions. Following the same philosophy, we use a set of approximate piecewise linear base functions. Actually, they are exactly piecewise linear base functions for a single crack. We can use the following functions to approximate the linear base functions with non-zero values on [d - e, d + e] on the K ~ q?(z) dz - ~ -Q(e) de = 0 crack aib i. Vl i = 1, 2 . . . . . n. (3) & p(t) = where { 3 - - tx/-~--a x / t - b X+(d~)(t-de) te[d-e,d+e]c[ai, 4v plane strain K= 1~ b ~] other plane stress (11) and Fr is the contour surrounding the i'th crack a~b~,i = 1,2 . . . . . n. The stresses and displacements are given by a~+ o-y= 2[~(z) + aS(z)] %-i%= (4) qS(z) + ~ ( ~ ) + ( z - i ) r 2#(u + iv) = K(9(Z) -- co(Y.) -- (z --s @(z) (5) (6) where r = ~P(z) and cd(z) = q~(z). and G must be zero if there are no stresses and no rigid body rotation at infinity. In the alternating method, q(t) =- O, because a y+ -= orand a xy+ -= rr-. Thus, the complex potentials y xy [Eq. (1) and Eq. (2)] can be simplified as P(z)+iP.(z) ~0(z) =$2(z) 2~iX(z) where a = a i, b = bi, de = d + e and do = d T- e. It can be easily verified that p (de) = 0 and p(d~) = 1. p (t) is the linear function of t in [d - e, d + el when the number of the cracks is one. The error of the approximation decreases as e ~ 0 for multiple cracks. Using this approximation, the Cauchy integral in Eq. (8) can be evaluated as in the following: r(z) = [F(d + e, z) - F ( d - e, z)] b (12) where k(t,z): [.x/[Z~-ax/-iCbt-dedt t--Z (7) :tx/~--ax/-t~[ (t-a)+(t-b)4 + (z - de)] where F(z) X+(d~) = -] x + p ( t ) dt (8) L t--z The uniqueness conditions of the displacements [Eq. (3)] can be simplified as e--'''~uJiz)az=O i = 1 , 2 . . . . . n. (9) 9in . . . . . + [(Z -- de)(2z - a - b) - ~ ( a - b) 2] < if the contours ~ , i = 1, 2,..., n are symmetric about the real axis. This set of equations leads to the following linear system 9ln (xft - a + x / t - b) The derivative of F(z) is (lO) Kqcj=r i i , j = 1,2 . . . . . n [F'(Z)=~~[ l+z-d~z_td where KO - ~ z)-~ ~x~dz and i ~ F(Z) d z ri= <X(z) -- (a + b + 2dr In ( V t -- a + x / ~ - - b) 2.1 The Cauchy integral The crack surface loads can be represented by a set of linearly independent base functions. Park and Atluri (1992) used a set of Delta functions as base functions. To improve the accuracy ,,/;G-a , / b - z - , / ; G , / a - z 267 The derivative of 12(z) is s = [ " ( z ) + t ,(z) 1 2 2zciX(z) k=l + Z -- a k 2.2 The strains, stresses, displacements and stress intensity factors The stresses, displacement gradients and displacements are given by 268 G + % = 2 [s (13) +/-2(z)] %-i%=O(z)+S2(e)+(z-e)O'(z) (14) 2/~(u + iv) = xo)(z) --o)(Y.) -- (z --~)l-2(z) (15) 2#(ux + iVx) = Kf2(z) - I2(~) -- (z - - s (16) 2~(vy--iuy) =K/'2(z) +.(2(2) + (z--i)g2'(z) in the infinite domain subjected to certain crack surface loading as described in the previous section. The other one has the same finite geometry as the original problem except that the cracks are ignored. The second problem is solved by using the finite element method. Since the cracks are ignored, the boundary of the FEM model is F = / ' ~ w F~. We discretize the uncracked finite domain using a FEM model. Let the prescribed displacements (including the zero displacements) on /'u be u = {lTi}, i = 1, 2. . . . . n~. Let the prescribed tractions (including zero tractions) on F~ be t = {~, i = 1, 2 . . . . . n,. Let the sampling points of tractions on the locations of the crack surfaces F~ be Pi, i = 1, 2 . . . . . np. The tractions at these sampling points are denoted as T = {Ti}, i = 1,2 . . . . . n r. We can solve for the crack surface tractions T for any given boundary loads u and t using the finite element method. Due to the linearity of the problem, we can denote the solution procedure as T = KUu + K t t --2/2(z) (17) (19) where K u and K t are linear operators of dimension {nr, nu} and where o'(z) = g2(z). The stress intensity factor is K~ -- iK~z = lim x~a k We can take the tractions T as the crack surface loads applied on the n collinear cracks in the infinite domain. Using the analytical solution, we can find the displacements u at the 2x/~k -- x) (a z -- ia v ) for the left crack tip ak location of Pu and the tractions t at the location of F t. Similarly, we denote these solutions as Kf - iK~ = lira ~ ( a y x - i a . ) for the right crack tip bk bk u=KUT m t = KtT They are 2 - )~X.b) (bk----ak) [F(Gb) + iP"(x"b)t (18) (20) (21) where K" and K t are linear operators of dimension (n,, nr) and (nt, nr). The boundary conditions for the original problem are u = u s and t = t ~ Since the crack surfaces are traction free, T = 0. We try to find the additional loads u a and t a such that where T = KU(u~ + u ~) + K t ( t ~ + t ~) for the left crack tip a k Zab = Nab = i bk for the right crack tip b k u" = K u T for the left crack tip a k t a = K tT for the right crack tip b k Subtracting the analytical solution from the FEM solution, we get the solution to the original problem. The existence of such loads u s and t a is shown in the following. Eliminating T in the above equations, we have and (I - A ) X = r l=l,lq:k where 3 Schwartz-Neumann Alternating Method Consider the case in which the n collinear cracks are in a finite domain. The crack surfaces are denoted collectively as/',. Let the boundary of the finite domain (not including the crack surface) be F. The boundary with prescribed tractions is F~. The boundary with prescribed displacements is Ft. We have e=Gur,. We make use of two simpler problems in order to solve the original problem. One of them is that of the n collinear cracks I K ~ K ~ ~-~Kt~ A = -~K ~ WK t] X= t~ (22) and I is the identity operator. The linear system Eq. (22) has an unique solution if I - A is not singular. If I - A is singular, there must exist a non-zero X such that ( I - A ) X = 0, which means that there exists a non-zero T such that T = K " ( u " ) + K t ( t a) u ~ = KUT m t ~= KtT In this case, if we subtract the analytical solution from the FEM solution, we obtain the solution to the following problem. The geometry is the same as the original problem, while the whole boundary F is free of external loadings. The crack surfaces are traction free. The FEM solution gives zero displacements for the crack surfaces, while the analytical solution gives non-zero displacements for the crack surfaces because of the non-zero T. Thus, the resulting solution has non-zero displacements at the crack surfaces. Since the cracks can not be opened without any external load, we have a contradiction. Therefore, I - A can not be singular. It is very expensive to find X by inverting I - A. An alternate iterative scheme can be devised as: Xi+l=AXi i=0,1,2 ..... ~ u = 0 and t = 0 on F than twice as much as it does in the infinite domain, the alternating method [Eq. (23)] described above converges. We can expect quick convergences for most of the practical applications. For any crack surface displacements, the displacement and stress at a point decay rapidly as the point moves away from the crack surfaces. Thus, the work done in the finite domain with the homogeneous boundary condition is very close to the work done in the infinite domain, which implies that the eigenvalues of A are very small. 269 4 Examples 4.1 Periodic collinear cracks First, we consider the problem of periodic collinear cracks in an infinite domain, subjected to the far field loadings. Only upper half of the body is modelled in the FEM analysis because of the symmetry. Mixed boundary conditions are specified on the left and right side of the block to ensure the periodic condition. The dimensions used in the FEM analysis are shown in Fig. 1. The single FEM mesh in the analysis is shown in Fig. 2. The results are compared to the analytical solutions [Anderson (23) G=I where X ~ {u ~ to} ~. This fixed point iteration scheme converges if all the eigenvalues of A are in the open interval ( - 1, 1). If this procedure converges, the solution is X= ~X _TTTTTITTTTTTT 2W=2 i i=1 The eigenvalues of A are smaller than 1. Let X~ be an eigenvector of A corresponding to the eigenvalue ), 2a "r " H=IO T = K " ( u ~) + K t ( # ) ,~u 2 = ~T )~t;~= K--~T Subtracting 2 times the FEM solution from the analytical solution, we have the following solutions, u = 0 and t = 0 on _F and the crack surface loading is ( 1 - 2 ) T , while the displacements at the crack surface are the same as those in the analytical solution. If the work done in opening the cracks in the infinite domain is W, the work done in opening the cracks in the finite domain (with the boundary condition u = 0 and t = 0) is (1 - .~) W, which is equal to the strain energy stored in the body. It must be positive. Thus, 2 < 1. It can be shown that 3~=>_0 if there is no boundary with prescribed displacements, i.e. there is no F~. In this case, the alternating method converges for cracks in finite domains with arbitrary shapes. In general, 2 can be negative. We have )~ > - 1 if (1 - )~) W < 2W, Thus, We h a v e the following form of convergence criterion. For an arbitrary distribution of crack surface displacements, if the crack surface loads do less work in the finite domain with the homogeneous boundary condition Fig. 1. Periodic collinear cracks in an infinite domain, subjected to the uniform far field loading lip Fig. 2. The FEM mesh used for the periodic cracks at different a/W 5.5 - - 5.0 ....... 4.5 / Periodic cracks Single crack // 4.0 3.5 Fig. 5. The FEM mesh for the hole cracks, modelling the upper half of the body 3.0 2.5 2.0 1.5 270 1.0 0.5 0 0 i I I I I I I I I 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 .0 Fig. 3. Normalised stress intensity factors K~versus the normalised crack lengths a/W Fig. 6. The FEM mesh for the hole cracks, modelling one quarter of the body (1991)1. 1.30 d 1.25 ! K1 = r~x / - ~ I~--a tari \2W,] ( 7~a ~]"2 ] (24) 1.20 where 2a is the crack length. Figure 3 shows the normalised stress intensity factor ~ = Ki/ax/-W. The solution of the alternating method agrees well with the analytical solution. The error increases as the a/W becomes very large. It is noticed that there are only one and half elements from the crack tip to the closest boundary for a/W = 0.85. The alternating method gives good result even for this case. 1.15 / 1.10 1.05 6 1.00 Using the mesh in Fig. 5 0.95 4.2 Cracks emanating from holes 0.90 Now, we consider the problem with cracks emanating from two holes. The details are shown in Fig. 4. The problem is bi-symmetric. First, we solve the problem by modelling only the upper half, in which we only deal with traction boundary conditions. The mesh is shown in Fig. 5. Then, we use the symmetric condition and model only the upper left quarter, in which we have to deal with mixed boundary conditions. The mesh used is shown in Fig. 6. Figure 7 shows the convergence T I I T T 2.00 0 r,l.OO 1 . . . 2 ; 3 4 , , 6 7 9 Fig. 7. Stress intensity factors K/versus the number of iterations of the stress intensity factors at the inner crack tip A for the two different approaches. The convergent rates are almost the same. Thus, the alternating method for the mixed boundary conditions enables us to substantially reduce the computational cost by using displacement boundary conditions. 5 Summary G= 1.00 T I . 0 ;. Fig. 4. Cracks emanating from two holes 1.00 , r, The implementation of the Schwartz-Neumann Alternating Method for collinear multiple cracks with mixed types of boundary conditions has been presented. The convergence rate is related to the ratio of the energies needed to have the same arbitrary crack surface displacements in the following two problems. The first is that of the cracks in an infinite domain. The second is the same as the original one except that all the prescribed tractions and displacements are set to zeroes. The closer the ratio is to the unity, the faster the procedure converges. If the ratio for any surface displacement is smaller than 1/2, this procedure will fail to converge. Fortunately, this will not occur in most of the applications in the practice. The efficiency and effectiveness are shown by the numerical examples. References Muskhelishvili, N. I. 1953: Some Basic Problems of the Mathematical Theory of Elasticity, Noordhoo, Groningen Nishioka, T.; Atluri, S. N. 1983: Analytical solution for embedded elliptical cracks, and finite element alternating method for elliptical surface cracks, subjected to arbitrary loadings. Engineering Fracture Mechanics. 17(3) 247-~268 Atluri, S. N. 1986: Computational methods in the mechanics of fracture. Amsterdam: North Holland, also translated in Russian, Mir Publishers, Moscow Murakami, Y. et al. 1987: Stress intensity factors handbook. Pergamon press, Oxford Anderson, T. L. 1991: Fracture Mechanics: fundamentals and applications. CRC Press Park, J. H.; Afluri, S. N. 1992: Fatigue Growth of Multiple-Cracks Near a Row of Fastener-Holes in a Fuselage Lap-loint, Durability of Metal Aircraft Structures, proceedings of the international workshop on structural integrity of aging airplanes, Atlanta. pp. 91-116. Also, Computational Mechanics. Vol. 13 No. 3 Dec, 1993. pp. 189-203 Park, J. H. 1993: Improvement of the Accuracy of Stress Fields in Multiple Hole Crack Problems. Internal Report for Computational Modeling Center, Georgia Institute of Technology Nikishkov, G. P.; Afluri, S. N. 1994: An analytical-numericalalternating method for elastic-plasticanalysis of cracks. Computational Mechanics. Vol. 13 No. 6 Mar pp. 427-442 271