COMPUTATIONAL MECHANICS New Trends and Applications S. Idelsohn, E. Oate and E. Dvorkin (Eds.

) n c CIMNE, Barcelona, Spain 1998

ADAPTIVE ANALYSIS OF SOFTENING SOLIDS USING A RESIDUAL-TYPE ERROR ESTIMATOR

P. D ez, M. Arroyo and A. Huerta
Departamento Matemtica Aplicada III a E.T.S. de Ingenieros de Caminos, Canales y Puertos Universidad Politcnica de Catalua e n Campus Norte UPC, 08034 Barcelona, Spain e-mail: diez@etseccpb.upc.es, web page:http://www.upc.es/ma3/lacan.html

Key words: Adaptivity, Error estimation, softening solids, regularized models Abstract. This paper focuses on the numerical simulation of strain softening mechanical problems. Two problems arise: (1) the constitutive model has to be regular and (2) the numerical technique must be able to capture the two scales of the problem (the macroscopic geometrical representation and the microscopic behavior in the localization bands). The Perzyna viscoplastic model is used in order to obtain a regularized softening model allowing to simulate strain localization phenomena. This model is applied to quasistatic examples. The viscous regularization of quasistatic processes is also discussed: in quasistatics, the internal length associated with the obtained band width is no longer only a function of the material parameters but also depends on the bound value problem (geometry and loads, specially loading velocity). An adaptive computation is applied to softening viscoplastic materials showing strain localization. As the key ingredient of the adaptive strategy, a residual type error estimator is generalized to deal with such highly nonlinear material model. In several numerical examples the adaptive process is able to detect complex collapse modes that are not captured by a rst, even if ne, mesh. Consequently, adaptive strategies are found to be essential to detect the collapse mechanism and to assess the optimal location of the elements in the mesh.

P. D M. Arroyo and A. Huerta ez,

1 INTRODUCTION Adaptive mesh renement in strain softening problems has received important attention in the last decade. The presence of two well-dierentiated length scales in such problems seems to indicate that adaptive remeshing strategies, in a general sense, are the natural approach. Recall that the spatial interpolation of the primitive variables must describe both the macroscopic scale associated to the solid geometry and the micro-scale related to the shear band. For instance, [13] shows for two dierent models, that in the localization area element sizes must be one order of magnitude smaller than the internal length if errors under 5% are desired. Moreover, a priori knowledge of the location of the localization area is sometimes not possible. Adaptivity in nite element computations requires three main ingredients. The rst one is an algorithm for increasing/decreasing the richness of the interpolation in a particular area of the computational domain. For instance, a good mesh generator if h-renement [21,38] is used. Second, an error estimator or error indicator must be employed to locate where there is a need for renement/de-renement. And, third, a remeshing criterion must be used to translate the output of the error analysis into the input of the mesh generator, for instance, the distribution of desired mesh sizes. These three steps are fundamental in adaptivity. However, only the second one will be discussed here because it is, probably, the critical one in softening problems. In fact, [23] presented an excellent discussion on the diculties that have deterred an extensive application of adaptive methods in the context of strain localization: 1) path dependent constitutive equations, and 2) error estimation relying, in statics, on the ellipticity of the equations which is lost at the inception of localization. Here, these two diculties are overcomed because: 1) an error estimator mathematically sound for path dependent constitutive equations is employed, and 2) a well known regularized model which precludes the loss of ellipticity is used. The main goal here, is to show that adaptive mesh renement based on estimating the actual error is now possible for regularized problems. Any model maintaining ellipticity after the inception of the localization could be used. Moreover, the numerical examples show that adaptivity based on the error estimation is essential for accurate computations and, in certain problems, for capturing a realistic physical behavior. The study of localization in solids has received much attention and a number of dierent approaches have been devised to overcome the diculties encountered during its analysis. A possible approach is to consider the limit problem and consequently, jumps in the displacement eld across surfaces (strong discontinuity approach). In fact, it emanates from classical perfect plasticity where discontinuous displacements are understood in a distributional sense, see for instance [30,2,17]. Another possible approach is to regularize the problem precluding any discontinuity in the displacement eld. These models bear the same fundamental property: an internal length is introduced to limit the thickness of the localization band. Among these are the micropolar constitutive models [10], non-local models [7,27], gradient dependent models [11,18],

P. D M. Arroyo and A. Huerta ez,

and rate-dependent models [32,8]. The present study is developed in the context of regularized models. Here, a simple rate-dependent model is employed [26] because the main focus is on adaptivity. The issue of a mathematically sound error estimator for path dependent constitutive equations is crucial in this study. Adaptivity in softening problems has been associated to error indicators [25,37,28]. Error indicators are based on heuristic considerations while error estimators approximate a measure of the actual error in a given norm. In this work, a tool for assessing the error measured in the energy norm is proposed. The obtained approximation to the error is asymptotically exact, that is, tends to the actual error if the element size tends to zero [4,5]. In that sense, this tool is an error estimator. 2 PERZYNA VISCOPLASTICITY MODEL As noted previously, the Perzyna viscoplastic model is used for regularization [26,31]. Thus, regularization is associated to viscous eects in the inelastic range. For clarity, this model is briey reviewed in this section. In small-strain viscoplasticity, the strain rate is decomposed into an elastic strain rate e and a viscoplastic strain rate vp , = e + vp , and the stress rate is obtained as = C : ( vp ), where C is the elastic moduli tensor. Eq. (1) and (2) are very similar to those of elastoplasticity, with plastic strain replaced by viscoplastic strain vp . An expression for vp is needed. For associative ow, the Perzyna model [26] takes vp = f 0
N

f ,

where and N are the material parameters of the model, f is the Von Mises yield function which depends on the yield stress [15] This yield stress is assumed to be a linear function of the equivalent viscoplastic strain , according to = 0 + h , where 0 is the initial yield stress and h is the hardening (for h > 0) or softening (for h < 0, used here) parameter. The symbol f means that f = f 0 if f 0 . if f 0

P. D M. Arroyo and A. Huerta ez,

In contrast to elastoplasticity, stress states outside the yield surface (i.e., with f > 0) are admissible in viscoplasticity. Combining Eq. (3) and (5) shows that the viscoplastic strain rate vp is, in general, non-zero. On the other hand, for stress states inside or on the yield surface, vp = 0 and the strain rate is purely elastic.

" = "e + "vp _ _ _

E

"e

"vp = _

0

,

0

0

"vp

_ = E " , "vp  _ _  = 0 +

5 6

h "vp

Figure 1. One dimensional rheologic scheme of Perzyna type viscoplastic model (N = 1).

Figure 1 shows the one-dimensional rheologic scheme of the Perzyna model for the case N = 1 [30]. The viscous eect is represented by the damper in the right, with viscosity 0 /, and softening is represented by the element on the left, which symbolizes the linear evolution of the yield stress, Eq. (4).

The Perzyna viscoplastic model has been used as a regularization technique in transient dynamic processes [31,8]. Here, it will be employed to regularize a quasistatic problem.

For dynamic analysis (i.e. inertia eects are considered) the width of the localization band can be predicted a priori as a function of material parameters [31]. It is independent of the loading velocity. The rationale is that the stress waves (those producing loading and unloading paths) travel at a celerity which depends only on material parameters. Thus, in dynamic analysis, the band width depends on the stress wave celerity and the viscous eects.

P. D M. Arroyo and A. Huerta ez,

X1.E6 8.00

7.00

6.00

5.00

4.00

3.00

2.00

Fast loading Slow loading Medium loading

0.00 0.20 0.40 0.60 0.80 1.00 1.20 X1.E-4 1.40

1.00

0.00

Figure 2. Inuence of the loading velocity in the response of the Perzyna model in a quasistatic case (reaction versus imposed displacement)

For quasistatic analysis processes (i.e. inertia terms are neglected) time does not appear explicitly in the momentum balance. However time is still an independent variable of the problem because it is present in the constitutive relations, Eqs. (1)-(3). Time has still its physical meaning and it is not, as in rate-independent elastoplasticity, a loading parameter (i.e. a pseudo-time). Notice that for quasistatic problems the loading velocity has a crucial inuence on the material response, see Figure 2 for a one-dimensional example. Thus, from a physical point of view, viscoplasticity in quasistatics is still a sound regularization technique (i.e. viscous eects and the associated regularization are still present). Section 4 illustrates this issue with several examples. In fact, viscoplasticity is an obviously sound regularization technique in a quasistatic analysis because a quasistatic problem is simply a dynamic problem where inertia terms are below the threshold of computational accuracy. Viscoplasticity is not employed here as a numerical strategy to avoid imposing the plastic consistency condition. It is assumed to be the real material behavior. It is not used as a numerical relaxation technique [24] designed to get the elastoplastic solution. In fact, for softening models, the steady state limit (after zero loading velocity is reached) of a viscoplastic analysis does not coincide with an elastoplastic solution. Nevertheless, in quasistatics, the velocity governing the loading and unloading is no longer the stress wave celerity but the external load velocity. Thus, the band width in quasistatics is not governed exclusively by material parameters, it depends also on the load velocity. In fact, in [20] it is demonstrated that for one-dimensional problems the load velocity plays a crucial role in the resulting band width.

P. D M. Arroyo and A. Huerta ez,

Another dierence between dynamic and quasistatic viscoplastic analysis is related with the need and the inuence of imperfections. It is well known that dynamic viscoplastic problems do not need an imperfection to induce localization, reecting or incoming stress waves are usually employed to initiate localization. Quasistatic viscoplastic problems in simple domains, however, do need of an imperfection to induce localization. In any case, both in dynamics and quasistatics, if an imperfection exists, the resulting band width may be aected by the imperfection in its vicinity, see [31,36] for dynamic cases and [3] or Section 4 for quasistatics. 3 ADAPTIVITY BASED ON ERROR ESTIMATION The key feature of an adaptive strategies is the error assessment. Once the approximate solution is computed, it has to be decided if this solution is accurate enough or if a new mesh must be generated in order to obtain an improved solution. The error assessment fournishes the global error measure (a number) and its distribution in the domain (a function or a set of numbers). The error information is used to decide if remeshing must be performed and, if it must be, to determine the element size of the new mesh in each part of the domain. The error assessment can be done using error estimators or error indicators. While error estimators fournish objective error measures in a suitable error norm, error indicators use heuristic criteria, mainly based on interpolation error estimates, to obtain qualitative information about the error. Error indicators are usually explicitely computed, they may be used to identify the zones where the mesh must be ner and they are cheaper than error estimators which typically require to solve a set of local problems. However, the information given by error indicators cannot be used as an objective error assessment tool because the selection of the error indicator is very sensitive to the analysed problem. Here, adaptivity is driven by the residual tipe a posterior error estimator introduced in [12]. This estimator has been generalized to nonlinear problems, precludes the usual drawbacks of residual type estimators (computation of the ux jumps and ux spliting procedure) and gives good results in the analysed examples. Once the error is estimated, the adaptive procedure requires a remeshing criterion in order to generate the input for the mesh generator. The error estimator furnishes local measures of the error in each element, that is, eh k for k = 1, . . . This set of numbers describes the spatial distribution of the error. The input of a mesh generator is a distribution of desired element size in the computational domain. Generally, this is described by the desired element size in each element of the current mesh, that is, hk . for k = 1, . . . Thus, a remeshing criterion is required to translate eh k into hk Dierent remeshing criteria have been dened [39,22,19,12] leading to quite dierent optimal meshes. This is because the underlying optimality criteria are dierent. In fact, all these remeshing criteria tend to equidistribute the error in some sense. The choice of the error function that has to be uniform is related with the underlying optimality

P. D M. Arroyo and A. Huerta ez,

criterion. In the examples presented in section 4, two remeshing criteria are used and it is shown how the obtained meshes are very dierent. In the following, the quadrilateral mesh generator developed by Sarrate [29] is used. This mesh generator supplies excellent unstructured meshes, both ensuring the prescribed element size and the regularity of the elements. 4 NUMERICAL EXAMPLES

2 ? t

6
50 mm

,,

uy

= t = v
t

?
R = 2:5

mm

25 mm

,,

ux

=0

= 2
1011 Pa  = 0:3 N=1 0 = 2
108 Pa = 500 s,1 h = ,2
1010 Pa R = 2:5 mm v = 1 m
s,1 max = 0:13 mm
E

-x

, uy = 0 ,
Computational domain

Two dimensional specimen with a circular opening

Figure 3. Rectangular specimen with one centered imperfection

Two examples are presented in this section. Both of them reproduce the compression of a plane strain rectangular specimen. In order to induce the strain localization in the specimen, some imperfection must be introduced. Typically, these imperfections are introduced by either a weaker element or a geometric imperfection [31,8]. Here, geometric imperfections are used: circular openings inside the material create a weaker zone in the specimen. The dierence between the two presented examples is the number and the location of these circular openings. In the rst example the specimen has one

P. D M. Arroyo and A. Huerta ez,

centered circular opening and, consequently, the two axes of symmetry allow to study only one fourth of the specimen (see Figure 3). In the second example, the specimen has two circular openings symmetric with respect to the center. That allows to study only one half of the specimen (see Figure 4). In both examples the tests are driven by imposing the velocity at the top of the specimen.

10 mm = 2 mm

5 6 5 mm ?
:

5 6 5 mm ?
:

5 mm

= 2 mm

Example 2a: Specimen with two distant circular openings

Example 2b: Specimen with two close circular openings

Figure 4. Rectangular specimen with two symmetric openings

4.1 Rectangular specimen with one imperfection As shown in Figure 5, the collapse mechanism in this example is formed by two strain localization bands and softening behavior is observed. In Figures 6 and 7 cross sections of equivalent inelastic strain and Von Mises stress are shown. It can be seen that strain grows in the localization band along the loading process, but Von Mises stress remains almost constant once the failure mechanism is developed. Recall that all the deformation is localized along the band. Once the band is developed, the specimen behaves like two sliding rigid blocks: the material inside the band is practically owing.

P. D M. Arroyo and A. Huerta ez,

X1.E6 6.00

5.00

Reaction N

4.00

3.00

2.00

1.00

0.00 0.00 0.20 0.40 0.60

m

0.80

1.00

1.20 X1.E-4

1.

0.00E+00 7.00E-04 1.40E-03 2.10E-03 2.80E-03 3.50E-03 4.20E-03 4.90E-03 5.60E-03 6.30E-03 7.00E-03 7.70E-03 8.40E-03 9.10E-03

Mesh deformation ampli ed 15 times

Equivalent inelastic strain

Figure 5. Description of example 1

X1.E-2 1.60

1.40

1.20

1.00

0.80

0.60

0.40

0.20

B
0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 X1.E-2 4.0

0.00

B'

A'

Figure 6. Evolution of the prole of the equivalent inelastic strain across the shear band along the normal direction (A-A)

P. D M. Arroyo and A. Huerta ez,

X1.E8 2.25 2.00 1.75 1.50 1.25 1.00 0.75 0.50 0.25 0.00 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 X1.E-2 4.00

X1.E8 2.25 2.00 1.75 1.50 1.25 1.00 0.75 0.50 0.25 0.00 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 X1.E-2 4

= 0:04 mm
X1.E8 2.25 2.00 1.75 1.50 1.25 1.00 0.75 0.50 0.25 0.00 0.00 0.50 1.00 1.50 2.00 2.50 3.00

= 0:06 mm
X1.E8 2.25 2.00 1.75 1.50 1.25 1.00 0.75 0.50 0.25 0.00

3.50 X1.E-2

4.00

0.00

0.50

1.00

1.50

2.00

2.50

3.00

3.50 X1.E-2

= 0:08 mm
X1.E8 2.25 2.00 1.75 1.50 1.25 1.00 0.75 0.50 0.25 0.00 0.00 0.50 1.00 1.50 2.00 2.50 3.00

= 0:10 mm
X1.E8 2.25 2.00 1.75 1.50 1.25 1.00 0.75 0.50 0.25 0.00

3.50 X1.E-2

4.00

0.00

0.50

1.00

1.50

2.00

2.50

3.00

3.50 X1.E-2

= 0:12 mm Pro les along A-A'

= 0:13 mm

Figure 7. Evolution of the Von Mises equivalent stress (solid line) and the yield stress (dashed line) in the loading process

10

P. D M. Arroyo and A. Huerta ez,

Figure 8. Inuence of the imperfection size: contours and proles of the inelastic strain.

11

P. D M. Arroyo and A. Huerta ez,

Mesh 0 110 elements Global accuracy: 3.99

Mesh 1 590 elements Global accuracy: 1.05

Mesh 2 611 elements Global accuracy: 0.64

Mesh 3 561 elements Global accuracy: 0.56

Mesh 4 564 elements Global accuracy: 0.49

Final mesh with pollution errors 601 elements Global accuracy: 0.49

Figure 9. Remeshing process using Li-Bettesss criterion, without pollution errors, for a prescribed accuracy of 0.5%

12

P. D M. Arroyo and A. Huerta ez,

Figure 10. Remeshing process using Li-Bettesss and considering pollution errors for a prescribed accuracy of 0.5%: succession of meshes and estimated error distributions

13

P. D M. Arroyo and A. Huerta ez,

Figure 11. Remeshing process using ULA criterion and considering pollution errors for a prescribed accuracy of 0.5%: succession of meshes, accuracy distributions and comparison with LB nal mesh

14

P. D M. Arroyo and A. Huerta ez,

Figure 12. Remeshing process using Li-Bettesss for a prescribed accuracy of 1.5%: succession of meshes and estimated error distributions

15

P. D M. Arroyo and A. Huerta ez,

This simple example has been used to study the inuence of the imperfection size and the loading velocity in the resulting localization band. First, for a given load velocity, three problems have been examined. The reference problem has an imperfection of radius R = 2.5mm as shown in Figure 3. Two other problems have been studied with radii Rsmall = 1.25mm and Rlarge = 5mm. As shown in Figure 8, the specimen with the large imperfection has the same band width than the reference one. On the contrary, the specimen with the small imperfection has a narrower band in the neighborhood of the imperfection. However, far enough from the imperfection, the width of the band is practically the same as in the other cases. In fact, if the loading process with the small imperfection continues, when the imposed displacement is large enough (e.g. = 0.19mm, see Figure 8) the band width is practically the same as the one obtained with larger imperfections. Thus, the imperfection size plays a role in the determination of the band width only if it is small. Moreover, this inuence is only local and disappears at large load levels. Second, the specimen with R = 2.5mm has been studied with dierent loading velocities. As expected, when the load velocity decreases, the viscoplastic solution tends to the elastoplastic one. Thus, if loading velocity is very small the band width is very narrow and, from a practical point of view, the viscoplastic solution has the same problems as the elastoplastic: 1) pathological mesh dependence and 2) an almost vertical softening branch with the associated convergence diculties. An adaptive procedure has been used. First the computations are carried out with a coarse almost uniform mesh. Then the error is estimated at the end of the loading process. Using the estimated error distribution, a remeshing criterion and a mesh generator, a new mesh is created and the computations are carried out from scratch. This is repeated until the estimated error is below some acceptability requirements. Firstly, two series of adapted meshes following the Li and Bettess [19] optimality criterion are presented. This remeshing criterion seeks uniform error distribution ( e k constant for every k). In the adaptive procedure of Figure 9 the error is estimated only locally (interior and patch estimate), in the series shown in Figure 10 pollution errors are also taken into account. The goal in both examples is to obtain an error below the 0.5%. Discretizations corresponding to the local estimate have less elements than the ones obtained considering the pollution error. This is because the local error is lower and, consequently, a mesh with less elements suces. However, the nal distributions of elements are very similar. In fact, the pollution error is only relevant in the rst mesh which is coarse and roughly uniform. Once the discretization is rened where it is needed (in particular in the vicinity of the singularities) the pollution eects are attenuated and become negligible. Considering pollution error does not make a big dierence in the nal results. This result was expected given the strong ellipticity of the problem. Figure 10 shows also the evolution of the error distribution along the adaptive procedure. The distribution of the error tends to be uniform, as expected. In Figure 11 the adaptive procedure is carried out using a dierent remeshing cri-

16

P. D M. Arroyo and A. Huerta ez,

terion. Here, the Relative Local (RL) optimality criterion introduced in [12] which is based on obtaining a uniform distribution of local relative errors ( e k / u k constant for every k) is used. In Figure 11 the accuracy maps (local relative error distribution) for the nal mesh in Figure 10 is also shown. Using the RL criterion the accuracy is almost uniform over the domain but it requires more elements than the Li and Bettess criterion. This result was expected because the Li and Bettess criterion is optimal in the sense that it produces the mesh with less elements given a globally acceptable error. The RL criterion guarantees uniform accuracy, even locally, but it has a larger cost. Note that the numerical examples conrm that the problem has been regularized because the behavior of the solution does not vary along the remeshing process. The error estimation is robust in the sense that the adaptive procedure converges (the nal mesh has the desired error).

4.2 Rectangular specimen with two imperfections

The mechanism of failure in this case is much more complex. In fact, it depends strongly on the position of the circular openings. Two cases are examined with dierent horizontal gaps between the openings (see Figure 4). In this case only the Li & Bettess remeshing criterion has been used [19]. Example 2.a (distant openings) If the horizontal distance between the circular openings is large enough, the behavior is similar to the previous case. One shear band is developed aligned with the two openings. The remeshing process (see Figure 12) leads to a mesh with a large number of elements concentrated along a single shear band. Figure 13 shows the general behavior of the solution: the softening force-displacement curve is similar to the previous case and the equivalent inelastic strain is concentrated along the shear band, both in the original and the nal meshes of the remeshing process. That is, the captured collapse mechanism is the same in both meshes. Example 2.b (close openings) On the contrary, if the circular openings are closer, the behavior of the solution is much more complex and the original mesh is not able to reproduce such a mechanism. Figure 14 shows the succession of meshes in this case. It is worth noting that, in the nal mesh, according to the concentration of elements, two bands are developed. In fact, the resulting bands are not aligned with the imperfections (as in example 2.a), but have opposite inclination.

17

P. D M. Arroyo and A. Huerta ez,

X1.E7 1.20

1.00

R. N

0.80

0.60

0.40

0.20

0.00 0.00 0.20 0.40 0.60

m

0.80

1.00 X1.E-4

1.2

Mesh deformation ampli ed 10 times

0.00E+00 7.00E-04 1.40E-03 2.10E-03 2.80E-03 3.50E-03 4.20E-03 4.90E-03 5.60E-03 6.30E-03 7.00E-03 7.70E-03 8.40E-03 9.10E-03

Equivalent inelastic strain in mesh 0

Equivalent inelastic strain in mesh 5

Figure 13. General solution for example 2a: Reaction versus imposed displacement, deformation of mesh 5 and inelastic strain contours for meshes 0 and 5

18

P. D M. Arroyo and A. Huerta ez,

Figure 14. Remeshing process using Li-Bettesss for a prescribed accuracy of 1.5%: succession of meshes and estimated error distributions

19

P. D M. Arroyo and A. Huerta ez,

Figure 15. Numerical bifurcation in the rst meshes: mesh deformation amplied 40 times and equivalent inelastic strain contours

20

P. D M. Arroyo and A. Huerta ez,

Figure 16. Two consecutive failure mechanisms for the nal mesh: mesh deformation amplied 40 times and equivalent inelastic strain contours at two moments of the loading history

21

P. D M. Arroyo and A. Huerta ez,

Indeed Figure 15 shows how the computed equivalent inelastic strain and the deformation evolve along the remeshing process. Only after two remeshing steps the mesh captures two bands. In the previous meshes the discretization is not accurate enough and only one band is completely developed. Since large deformations are considered, once the rst band evolves enough, the kinematic mechanism associated with this band locks: then a second band appears as a new deformation mode with less energy. Figure 15 shows also how the force-displacement curves for meshes 0 and 1 are qualitatively dierent from meshes 2 to 5. In fact the shape of the force-displacement curve for meshes 2,3,4 and 5 are practically identical and have two inections in the descending branch. In Figure 16 it is shown how the inections correspond to the formation of a new failure mechanism. The solution given by the last mesh is obviously more accurate than the original one because the energy of deformation (area under the force-displacement curve) is lower. In fact, since the error is controlled in energy norm, one can be sure that the actual curve, associated with the exact solution, is not too far from the obtained curve (the error in energy norm is less than 1.5% and, consequently the dierence of the area under the curves is less than 1.5%). The rst meshes are not able to reproduce the behavior of the actual solution because the elements along the later band (which develops in a further stage of the loading process) are too large and, consequently, the discretization is too sti. Then, the size of the elements in this zone does not allow the inception of softening. On the contrary, once the remeshing process introduces small enough elements along the second band, the second mechanism can be captured. Thus, this example demonstrates that adaptivity based on error estimation is an essential tool for the determination of a priori unpredictable nal solutions. Without this adaptive strategy, the initial mesh (mesh 0 in gure 14) and the resulting solution could be regarded as correct, and the second mechanism would not be detected. 5 CONCLUDING REMARKS The Perzyna viscoplastic model has been used in order to obtain a regularized softening model allowing to simulate strain localization phenomena. This model has been applied to quasistatic examples, where inertia terms are negligible. The viscous regularization of quasistatic processes has been discussed: the rate eects are still present and regularize the problem. However a dierence between the dynamic and the quasistatic cases must be mentioned: in quasistatics, the internal length associated with the obtained band width is no longer only a function of the material parameters but also depends on the boundary value problem (geometry and loads, specially loading velocity). An adaptive computation has been successfully applied to softening viscoplastic materials showing strain localization. As the key ingredient of the adaptive strategy, a residual type error estimator has been generalized to deal with such highly nonlinear material model. Moreover, this estimator has been designed in order to account for pollution errors. However, as expected, the pollution errors have been found to be neg-

22

P. D M. Arroyo and A. Huerta ez,

ligible, specially in the rened meshes, with elements concentrated in the vicinity of the singularities. In the numerical examples, the adaptive process is shown to be able to detect complex collapse modes that are not captured by a rst, even if ne, mesh. This is specially interesting for softening localization problems, where small variations in the geometry of the problem may induce very dierent mechanical behavior. In this situation the location of the localization band cannot be predicted a priori. For instance, in one of the examples, the rst mesh is not able to reproduce the two consecutive mechanisms captured with the nal mesh. The second mechanism is associated with a second shear band appearing in a further stage of the loading process. However, if adaptivity is not used, the rst mesh would be regarded as correct and one never would detect the second mechanism. In fact, even if some heuristic remeshing is done, based on the solution obtained with the rst mesh, the mesh would not be rened along the second shear band, that is, where it is needed to capture the second mechanism. On the contrary, if a remeshing strategy based on the error distribution is used, the elements are concentrated along this second band and the new mesh is able to reproduce the two mechanisms. Consequently, adaptive strategies based on error estimation are essential to detect the collapse mechanism and to assess the location of the elements in an optimal mesh. REFERENCES [1] [2] Ainsworth, M. and J.T. Oden (1993), A unied approach to a posteriori error estimation using element residual methods, Numerische Mathematik, 65, 23-50. Armero, F. and Garikipati, K., (1996), An analysis of strong discontinuities in the multiplicative nite strain plasticity and their relation with the numerical simulation of strain localization in solids, Int. J. Solids Structures, 33, No. 2022, pp. 2863 2885. Arroyo, A., P. D and A. Huerta (1997) Error estimation and adaptivity in strain ez localization problems with softening materials, Monograph CIMNE n. 40, Barcelona. Babuka, I., R. Durn & R. Rodr s a guez (1992), Analysis of the eciency of an a posteriori error estimator for linear triangular nite elements, SIAM (Society for Industrial and Applied Mathematics) Journal of Numerical Analysis, 29 947964 Babuka, I. & R. Rodr s guez, (1993) The problem of the selection of an a posteriori error indicator based on smoothening techniques, International Journal for Numerical Methods in Engineering, 36 539567 Bank, R.E. and A. Weiser (1985), Some a posteriori error estimators for elliptic partial dierential equations, Mathematics of Computation, 44, 283-301. Belytschko, T., Baant, Z.P., Hyun, Y. and Chang, T.P., (1986), Strain Softening z Materials and Finite Element Solutions, Computers and Structures, Vol. 23, pp. 163-180.

[3] [4]

[5]

[6] [7]

23

P. D M. Arroyo and A. Huerta ez,

[8] [9]

[10] [11]

[12] [13]

[14] [15] [16] [17] [18] [19]

[20] [21]

[22] [23]

[24] [25]

Belytschko, T., B. Moran and M. Kulkarni (1991), On the crucial role of imperfections in quasi-static viscoplastic solutions, J. Appl. Mech. ASME, 58, 658-665. Cirak, F. & E. Ramm (1997) A-posteriori error estimation and adaptivity for linear elasticity using the reciprocal theorem, submited to Computer Methods in Applied Mechanics and Engineering de Borst, R., (1991), Simulation of Strain Localization: a Reappraisal of the Cosserat Continuum, Engineering Computations, Vol. 8, pp. 317-332. de Borst, R. and Mhlhaus, H.B., (1992), Gradient Dependent Plasticity: Foru mulation and Algorithmic Aspects, International Journal of Numerical Methods in Engineering, Vol. 35, pp. 521-540. D P. (1996), Un nuevo estimador de error para el mtodo de los elementos nitos, ez, e Doctoral Thesis, Universitat Polit`cnica de Catalunya, Barcelona. e Huerta, A., and Pijaudier-Cabot, G., (1994), Discretization Inuence on the Regularization by Two Localization Limiters, Journal of Engineering Mechanics, ASCE, Vol. 120, No. 6, pp. 1198-1218. Hughes, T.J.R. (1987), The nite element method, Prentice Hall International, Stanford Khan and Huang (1995), Continuum theory of plasticity John Wiley, New York Ladev`ze, P., J.P. Pelle and Ph. Rougeot (1991), Error estimation and mesh e optimization for classical nite elements, Engineering Computations, 8, 69-80. Larsson, R. and Runesson, K., (1996), Element embedded localization band based on regularized displacement discontinuity, J. Eng. Mech., ASCE, 122, pp. 402411. Lasry, D. and Belytschko, T., (1988), Localization Limiters in Transient Problems, International Journal of Solids and Structures, Vol. 24, pp. 581-597. Li, L.Y. and P. Bettess (1995), Notes on mesh optimal criteria in adaptive nite element computations, Communications in numerical methods in engineering, 11, 911-915. Needleman, A. (1988), Material rate dependence and mesh sensitivity in localization problems, Computer Methods in Applied Mechanics and Engineering, 67, 69-85. Oden, J.T., L. Demkowicz, W. Rachowicz & T.A. Westermann (1989) Toward a universal h-p adaptive nite element strategy, part 2. A posteriori error estimation, Computer Methods in Applied Mechanics and Engineering, 77, 113-180. Oate, E. and G. Bugeda (1993), A study of mesh optimality criteria in adaptive n nite element analysis, Engineering Computations, 10, 307-321. Ortiz, M., and Quigley IV, J.J., (1991), Adaptive Mesh Renement in Strain Localization Problems, Computer Methods in Applied Mechanics and Engineering, Vol. 90, pp. 781-804. Owen & Hinton (1980), Finite elements in plasticity, Pineridge Press, Swansea Pastor, M., Peraire, J., and Zienkiewicz, O.C., (1991), Adaptive Remeshing for Shear Band Localisation Problems, Archive of Applied Mechanics, Vol. 61, pp. 30-39.

24

P. D M. Arroyo and A. Huerta ez,

[26] [27] [28]

[29]

[30]

[31] [32]

[33]

[34]

[35]

[36]

[37]

[38]

[39]

Perzyna, P. (1966), Fundamental problems in viscoplasticity, Recent Advances in Applied Mechanics, Academic Press, New York, 9, 243-377. Pijaudier-Cabot, G., and Baant, Z.P., (1987), Nonlocal Damage Theory, Journal z of Engineering Mechanics, ASCE, Vol. 113, pp. 1512-1533. Pijaudier-Cabot, G., Bod, L., and Huerta, A., (1995), Arbitrary Lagrangiane Eulerian nite element analysis of strain localization in transient problems Int. J. for Numerical Methods in Engng., 38, pp. 4171-4191. Sarrate, J. (1996), Modelizacin numrica de la interaccin uido-slido r o e o o igido: desarrollo de algoritmos, generacin de mallas y adaptividad, Doctoral Thesis, Unio versitat Polit`cnica de Catalunya, Barcelona. e Simo, J.C., Oliver, J. and Armero, F., (1993), An analysis of strong discontinuities induced by strain-softening in rate-independent inelastic solids, Computational Mechanics, 12, pp. 277296. Sluys, L.J. (1992), Wave propagation, localisation and dispersion in softening solids, Doctoral Thesis, Delft University of Technology. Sluys, L.J., and de Borst, R., (1992), Wave Propagation and Localization in RateDependent Cracked Medium. Model Formulation and One Dimensional Examples, International Journal of Solid and Structures, Vol. 29, pp. 2945-2958. Strouboulis, T. & K.A. Haque (1992), Recent experiences with error estimation and adaptivity, Part I: Review of error estimators for scalar elliptic problems, Computer Methods in Applied Mechanics and Engineering, 97, 399436 Strouboulis, T. & K.A. Haque (1992), Recent experiences with error estimation and adaptivity, Part II: Error estimation for h-adaptive approximations on grids of triangles and quadrilaterals, Computer Methods in Applied Mechanics and Engineering, 100, 359430 Wahlbin, L.B. (1991), Local behavior in nite element methods in P.G. Ciarlet & J.L. Lions (eds.) Handbook of numerical analysis, Vol. II, North-Holland, Amsterdam, 357-522 Wang, W.M., L.J. Sluys and R. De Borst (1996), Interaction between material length scale and imperfection size for localization phenomena in viscoplastic media, Eur. J. Mech., A/Solids, 15(3), 447-464. Yu, J., Peric, D., and Owen, D.R.J., (1992), Adaptive Finite Element Analysis of a Strain Localisation Problem for the Elastoplastic Cosserat Continuum, Proceeding of the 3rd International Conference on Computational Plasticity, COMPLAS III, Eds. D.R.J. Owen et al., Pineridge Press, pp. 551-566. Zhu, J.Z., E. Hinton & O.C. Zienkiewicz (1993), Mesh enrichment against mesh regeneration using quadrilateral elements, Communications in numerical methods in engineering, 9, 547-554 Zienkiewicz, O.C. and J.Z. Zhu (1987), A simple error estimator and adaptive procedure for practical engineering analysis, International Journal for Numerical Methods in Engineering, 24, 337-357.

25