You are on page 1of 14
Computer methods ‘engineering ELSEVIER Comput. Methods Appl. Mech, Engrg. 121 (1995) 15-28 On the accuracies of numerical integration algorithms for Gurson-based pressure-dependent elastoplastic constitutive models Z.L. Zhang Department of Mechanical Engineering, Lappeenranta University of Technology, 53851 Lappeenranta, Finland Received 30 July 1992; revised manuscript received 17 June 1994 Abstract A class of generalized mid-point algorithms for pressure-dependent elastoplastic models is formulated in the paper. The accuracies of the formulated generalized mid-point algorithms including the Euler backward algorithm and the one-step Euler forward algorithm are systematically analyzed against the exact solution for Gurson-based pressure-dependent elastoplastic model. The accuracies of the algorithms are assessed by means of iso-error maps. Results show that the formulated generalized mid-point algorithms are reasonably accurate in both small and large increment steps. It is found that in all the eases considered, the maximum ertors in the presence of volumetric strain increments are less than those without volumetric strain increments When the deviatoric strain increments are given in the radial direction, the true mid-point algorithm is the most accurate one. Furthermore, the optimal value « of the generalized mid-point algorithms, in terms of maximum errors, is observed between 0.5 and 1. For both small and large increment steps, the one-step Euler forward algorithm gives the worst accuracy. 1. Introduction Numerical integration algorithm plays a significant role in the solution of non-linear problems. Several algorithms for the integration of elastoplastic constutive equations have been proposed in the literature. Two families of algorithms have been presented by Ortiz and Popov [1], which generalize the trapezoidal and mid-point rules in a manner that facilitate satisfaction of plastic consistency condition. The accuracy and stability of a class of integration algorithms have been widely assessed for the classical J, problems [1-5]. For perfectly plastic von Mises model, Ortiz and Popov [1] found that for both trapezoidal and mid-point rules the choice « = 0.5 results in second-order accuracy, and in the presence of large strain increments larger values of « may result in improved accuracy. They also observed that the stability properties of the generalized trapezoidal rules are very sensitive to the degree of distortion of the loading surface, while the generalized mid-point rules are unconditionally stable for @ > 0.5, regardless of the choice of loading surface. Recently, Gratacos et al. [3] have investigated the generalized mid-point rules in detail for the case of von Mises criterion with isotropic work-hardening rule. They concluded that, in most cases, first-order a = 1 should be preferred to second-order @ = 0.5 for realistic time steps. Lee [6] has analyzed the accuracies of three numerical algorithms ~ the tangent-stiffness radial return (Euler forward algorithm), secant-stiffness radial-return (@ = 0.5) and simple radial-return (Euler backward algorithm) methods for a pressure-modified von Mises yield criterion with associated flow rule. Lee’s results show that, for both accuracy and efficiency, and for small and large time steps, the secant-stiffness radial return method (a = 0.5) is the best of all Recently, a pressure-dependent constitutive model, which was originally presented by Gurson [7, 8] and modified by Tvergaard and Needleman [9-11] is now becoming increasingly popular. In the (0045-7825 /95/S09.50 © 1995 Elsevier Science S.A. All rights reserved SSDI-0045-7825(94)00706-3 16 Z.L. Zhang | Comput. Methods Appl. Meck. Engrg. 121 (1995) 15-28 Gurson-based model, not only the hydrostatic pressure but the material damage parameter ~ void volume fraction f has been incorporated into the yield function: In comparison with classical J, von Mises model and pressure-modified von Mises model, according to the author's knowledge no similar assessments have been found concerning the accuracies of different numerical integration algorithms for the Gurson-based porous solid model. Upon recognizing the ever-increasing applications of the Gurson-based model, we think it is necessary and important to study the accuracy of the integration algorithms for the Gurson-based model. In the following, a class of generalized mid-point algorithms for the Gurson-based pressure dependent material model is formu- lated first. The one-step Euler forward algorithm is also described in order to compare the accuracy. Then, the Gurson-based model and its associated flow rule with isotropic hardening is briefly introduced. The numerical results are presented in the fourth section by means of iso-error maps. The Paper is closed with some conclusions. 2, Numerical algorithms 2.1, Generalized mid-point algorithms ‘The numerical integration problem in a displacement finite element analysis is to update the known state variables associated with an converged solution at time 1, into their corresponding values on the updated solution at time f+ A. Here and subsequently, the subindices and n + 1 are used to identify the value of the various state variables at time ¢, and f,,,. In the following, for simplicity of notation, bold face symbols are used for variables of tensorial character and the symbol (:) signifies doubly contracted tensor product, €.g. (D:e),, = Dyasfyys (S:S By applying the elastic-plastic splitting methodology, Ortiz and Popov [1] formulated a class of generalized mid-point algorithms which in the present context takes the form (41) (2.2) (2.3) (2.4) (25) (2.6) p44 = 6,10 Hest) = (Pasi Garis ers) 20 27 where one writes ase = (SB) = Hoty cas tava) 28) Figs = Mp0 Has a) 29) Fxq = (1— ale, + a0, (2.10) A,,, =(1— @)H, + aH, (2.11) In the above equations, o)., is the so-called trial stress or elastic predictor, a is the algorithmic parameter which takes values in the interval (0, 1], €, e* and e? donate the total, elastic and plastic strain tensors, respectively, @ the Cauchy stress tensor and HT signifies the internal state variables. In Eqs. (2. 3) and (2.4) D* is the clastic moduli. In the paper, the linear isotropic elasticity is assumed Das = (K —3G)6,8y) + 2G5,5 (2.12) Z.L. Zhang | Comput. Methods Appl. Mech. Engrg. 121 (1995) 15-28 ” where °s are Chroneker symbols, and G and K are the elastic shear and bulk moduli, respectively. In Eq. (2.7), p is the pressure, p=—Lor:I and q is the equivalent stress, q = \V3S:S, where I and $ are the second-order identity tensor and the deviatoric stress, respectively. The associated flow rule has been used in Eq. (2.8) It must be noted that in the present formulations, the equilibrium equations are satisfied exactly at the end of the time increment, in a full implicit way regardless of the value of a. For «= 1, one recovers the closest point return mapping algorithm, whereby the clastic predictor is relaxed onto the closest point on the updated yield surface. On the other hand, if one takes a =0, one obtains a semi-implicit algorithm in the sense that the plastic flow direction is known before the increment only with the value needs to be solved implicitly. Fig. 1 depicts a geometric interpretation of the algorithm (2.1)-(2.7). 2.2, Formulations of the generalized mid-point algorithms of pressure-dependent elastoplastic models ‘The same way as the Euler backward algorithm presented by Aravas [12], we can separate any stress tensor into its deviatoric and hydrostatic components = Pil + Spur = Pro +5 gMyer (2.13) s, where n = 3/(24)S is the unit vector in the deviatoric space normal to the yield surface. Corresponding- ly, the plastic strain increment can be decoupled into volumetric (Ae?) and deviatoric (Ae8) parts S, c= aet-+ dep =aa[ (2) y+ ( ins tase (2.14) ‘The above equation with @ = 1 was used by Aravas in the Euler backward formulations. However, direct application of the above equation to the generalized midpoint algorithms will cause some ifficulties, which will be indicated later. It is found that using the following rearrangement of Eq. (2.14) will pass these difficulties ae) (#) Snta , aer=aal-(2) 14 (2% ps |=} de Tt Ae, § 2.15) . sf @)_0 nie Guve diay | Peet Ae Ge ey) where. ab Fig. 1. Geometric interpretation of the generalized mid-point rules

You might also like