EngOpt 2012 3rd International Conference on Engineering Optimization

Rio de Janeiro, Brazil, 01 - 05 July 2012.

A Finite Element approach using an Augmented Lagrangian Method to simulate impact problems under large 3D elastoplastic deformation
Alex Alves Bandeira1, Paulo de Mattos Pimenta2
1

Department of Construction and Structures, Polytechnic School at Federal University of Bahia, Salvador, Brazil, E-mail: alexbandeira@ufba.br 2 Department of Structural Engineering, Polytechnic School at University of So Paulo, So Paulo, Brazil, E-mail: ppimenta@usp.br

1 Abstract In this work the Augmented Lagrangian Method is used to solve impact problems with friction in solid mechanics. The purpose of this paper is to present the Augmented Lagrangian algorithmic to simulate numerical examples of impact problems using the formulations of contact mechanics, non-linear dynamics and elastoplasticity. The contact mechanics formulations, contact node-tosurface, node-to-edge and node-to-node, are presented in detail in [1]. The Newmark method is used to solve the dynamic analysis, for more details see [2] and [3]. The elastoplasticity algorithm is based on an associated von Mises yield function formulated in principal axes, considering the possible sliding of slave node from one surface to the adjacent one. The modified algorithmic developed in this research is derived from Laursen and Maker [4]. Here, the elastoplasticity and the nonlinear dynamics are considered in all impact examples. The contact formulations are derived based on exact linearization. The bodies in contact undergo finite deformation within an elastoplastic range. For the contact formulation within the finite element method, the matrix formulation for a node-to-surface element consisting of a master surface with four nodes and a contacting node is derived. The discretised contact surfaces are not smooth, i.e. there is no continuity of the normal vector between the adjacent surfaces. At the edge between the surfaces the normal is not uniquely defined, that needs a special algorithmic treatment, see [1] for more details. A numerical example of impact problem in elastoplastic range is presented to show the ability of these algorithmic. This example is modelled by using brick elements. 2 Keywords: Contact mechanics, dynamic, impact, elastoplasticity 3 Introduction The numerous contact formulations have been presented in the literature. Three-dimensional node-to-surface contact formulation can be found in [1], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15] and [16]. The non-penetration condition as a purely geometrical constraint is used to impose contact in normal direction. For the frictional response, the constitutive equation of the classical Coulombs law, like in [17], [5], [18] and [19] is used. Frictional phenomena have been considered within the framework of plasticity theory. The response in tangential direction can be divided into two different actions. First, no tangential relative displacement of the two bodies occurs which is so-called stick condition. The second action is associated with a relative tangential movement in the contact interface, so-called slip condition. In this work a three-dimensional brick element with eight nodes is used for the treatment of finite elastic-plastic deformation of the contacting surfaces. Here, an associative von Mises plasticity of the asperities is considered by formulating an incremental stress integration scheme on the principal axes of the incremental stretch tensor, see [20] and [21] for more details. An augmented Lagrangian method is applied to solve the frictional contact problems. The mathematical theories concerning augmented Lagrangian method in the context of mathematical programming problems subjected to equality and inequality constraints are well established by [22], [23] and [24]. Recently, in the context of finite element methods, augmented Lagrangian techniques have been successfully applied to frictional contact problems in solid mechanics by [8]. When large deformations occur in the contact mechanics, a large amount of sliding can happen within the contact interface. Two methods can be followed to discretise the contact interfaces. One, which applied here, is to discretise the contact interface by the isoparametric interpolation, where there is no continuity in the normal vector between adjacent surfaces. Thus, in the edge between two adjacent surfaces the normal is not uniquely defined and a special algorithmic treatment is needed. The other one is the smooth contact discretization, which allows for a smooth sliding of contacting nodes on the master surface. Within this approach, a Hermitian, Spline or Bzier interpolation is used to discretise the master surface [25]. This leads to a C1 or even C2 interpolation of the surface. In three-dimensions it is difficult to develop one general formulation for smooth contact. The node-to-surface, node-to-edge and node-to-node contact formulations used to solve contact problems with large sliding is coded in CMAP program (Contact Mechanics Analysis Program) developed by Alex Alves Bandeira since 1995. This program has been developed in object-oriented programming using C++. Information of this language can be easily founded in the literature. Several numerical recipes for numerical methods are presented in [26]. These formulations are characterized by algorithmic stability, short evaluation time, high performance and quadratic rate of convergence within a Newton equation-solving strategy, owing to the exact linearization employed. The most general approach for the solution of the dynamic response of structural systems is the direct numerical integration of the dynamic equilibrium equations. This involves, after the solution is defined at time zero, the attempt to satisfy dynamic equilibrium at discrete points in time. Most methods use equal time intervals at , 2 , 3 ... . Many different numerical techniques have previously been presented; however, all approaches can fundamentally be classified as either explicit or implicit integration methods. Explicit methods do not involve the solution of a set of linear equations at each step. Basically, these methods use the differential

equation at time to predict a solution at time + . For most real structures, which contain stiff elements, a very small time step is required in order to obtain a stable solution. Therefore, all explicit methods are conditionally stable with respect to the size of the time step. Implicit methods attempt to satisfy the differential equation at time after the solution at time + is found. These methods require the solution of a set of linear equations at each time step; however, larger time steps may be used. Implicit methods can be conditionally or unconditionally stable. There exist a large number of accurate, higher-order, multi-step methods that have been developed for the numerical solution of differential equations. These multistep methods assume that the solution is a smooth function in which the higher derivatives are continuous. The exact solution of many nonlinear structures requires that the accelerations, the second derivative of the displacements, are not smooth functions. This discontinuity of the acceleration is caused by the nonlinear hysteresis of most structural materials, contact between parts of the structure, and buckling of elements. Therefore, only single-step of Newmark methods will be shortly presented in this paper. Based on a significant amount of experience in the literature, is possible to say that the single-step, implicit, unconditional stable methods are necessaries conditions for the step-by-step dynamic analysis of practical structures. A brief remark of the Newmark integration scheme is also presented. This algorithmic is applied to calculated impact problems undergoing large 3D elastoplastic deformation. In the end of this paper, a numerical example of impact mechanics undergoing large three-dimensional deformation is presented to show the ability of these theories and the algorithm presented in Annex A. 4 Frictional Contact Mechanics Formulation This section summarises the continuum formulations applied to solve multibody frictional contact problems undergoing large 3D deformation. The node-to-surface, node-to-edge and node-to-node contact formulations are presented in details in [1]. The node-tosurface formulation was also presented in [5]. Within this reason, it will be not present in this paper. Only the new contributions of the contact formulation will be presented. 4.1 Contact Kinematics The normal and tangential gaps define contact contribution. The impenetrability condition can be formulated as gN = ( xs x m ) n c 0 and furthermore, the penetration function can be formulated in following manner (1)

( x s x m ) nc g = N 0

, if ( x s x m ) nc 0
(2)

, if ( x s x m ) nc < 0

where the bar denotes the closest distance of a point xs to the surfaces of Bm, see also (3). The graphical interpretation is illustrated in Figure 1.

B x
m m s

gN x
_m _ m

a1
_

a2

nc

Figure 1 Normal convention The sliding path of the slave node xs on the contact surface m is described by the total tangential gap as gT =

||

& a

|| dt,

(3)

where = 1, 2. The parameters 1 or 2 are introduced to describe the surface of the solids and can be interpreted as convective coordinates.

are the tangents at the solution points

and

of the distance function minimization

d (1,2,t)
These tangential vectors are obtained from

= minimum || xs

x m(1,2,t) ||.

(4)

x m ( , t)

x m,( ,t).

(5)

The tangential relative slip between two bodies is related to the change of the solution point ( , ) of the minimum distance problem (4). The time derivatives of the parameters normal vector. This yields

from (3) are obtained using the orthogonality condition of both tangent and

&
where

1 g N b
b
=

{ [ vs

v m ] a

g N n c v m, },

(6)

a a

is the metric and

x ,m n c is the curvature of the contact surface. When the frictional contact

problem is solved in an exact way, the slave node moves on the master surface. With this condition equation (6) can be simplified as vs

gN

can be neglected and the

vm

& a .

The relative tangential velocity at the contact point is defined by

& gT

= vs

& v m. Then g T

& a .

4.2 Contact Contribution to the Weak Form The contact contribution to the weak form can be formulated in following manner Wc(u,v) =

Boc
where tN is the contact pressure, tT is the tangential traction vector, n c is the internal normal at the slave node xs, vs the velocity at the slave node and v m the velocity at the master nodes. The variation of normal gap is defined by gN = ( vs

( tN

nc

+ tT ) ( vs

v m ) dA

(7)

& v m a ) n c

+ ( xs

xm) n c

= ( vs

v m ) n c.
n c

(8)

In equation (8), the orthogonality condition

a n c = 0
s

and the simple relationship

m

nc nc=1

nc=0

are used.

Following the same idea, gN can be written as gN = ( u u Wc(u,v) =

Boc
Wc(u,v) =

tN

n c ( vs

n c. With these results, equation (7) can be rewritten as v m ) dA + tT ( vs v m ) dA

s Boc

Boc
Wc(u,v) =

tN gN dA +

Boc
tN gN dA +

tT gT dA tT dA

Boc
with tT = tT

Boc

(9)

oc oc The terms in equation (10) will be defined in the following sections. The non-symmetry has been encountered in the linearization of
Wc for the stick case stemming only from the term tT in (10). For the slip case the tangential stiffness matrix is clearly nonsymmetrical. The algorithm used to integrate the frictional equations is completely developed in [5]. In this work, a simple nonassociated Coulomb friction law is used. 5 Elastic-plastic Material The Von Mises elastic-plastic constitutive law is based on the following multiplicative decomposition of the deformation gradient F = F eF p , (11) where the superscript e and p describe the elastic and the plastic part, respectively. The elastic logarithmic strain tensor is given by
1 eT 2 F

a . The linearization of the virtual work (9) is defined by Wc u = [tN gN + tN (gN)] dA + u B s B s

[tT + tT ( )] dA.

(10)

E = lnV , (E e ) =
where

where

V = (F

(12)

is the elastic left stretching tensor. The elasticity is described by the following strain energy function

1 2 + ( E e : E e ) , 2

(13)

 = trE e
that leads to the following Kirchhoff-Treftz stress tensor

and

E e = DevE e .

(14)

= D eE e ,

where

D e = I I + 2 I I I .

1 3

(15)

The logarithmic isotropic linear elastic material simplifies the volumetric-isochoric splitting. Note also that (15) is similar to the expression of the small strain Hookes Law. The fourth-order tangent tensor is obtained from (15) as shown in detail in [20]. For computational purposes the classical radial return algorithm along with the von Mises plasticity with linear isotropic hardening is summarized in [20] and [21]. The assumed linear isotropic hardening used in this paper is presented in the Figure 2.

Figure 2 Constitutive equation for steel material (Linear) Its function behaviour is defined as following

y ( ) = yo + h p .
where

(16)

yo

is the initial yield stress,

is the equivalent plastic strain and

is the linear hardening parameter.

6 Solution of equilibrium equation for dynamic analysis The Newmark integration scheme can also understood to be an extension if the linear acceleration method [2]. The following assumptions are used. see [3]: = + 1 + (17) = + + + (18)

where and are parameters that can be determined to obtain integration accuracy and stability. When = e = , relations (17) and (18) correspond to the linear acceleration method. Newmark originally proposed as an unconditionally stable scheme the constant-avarege-acceleration method (also called trapezoidal rule), in which case = e = . In addition to (17) and (18), for solution of the displacement, velocities, and accelerations, the equilibrium equations for nonlinear analysis at time + are defined by: + + = (19) where = for linear elasticity. Solving from (19) for in therms of and then substituting for into (44), we obtain equations for and , each in terms of the unknown displacement only.These two relations for and are substituted into (19) to solve for , after which, using (17) and (18), and can be calculated as following: = + 1 + 1 (20) , and

The variables rewritten by

are known in the step + . Substituting (20) and (21) in equation (19), the equilibrium equation is (21) 1 ,

+ + 1 The stiffness matrix is defined by deriving equation (22):

= = + + (23) , , , The complete implementation scheme using the Newmark method is given in annex A. This algorithmic presented in Annex A is used to calculated frictional impact problems using augmented Lagrangian method. It is important to mention that this algorithmic is an evolution of the one presented in [4].

1 + 1

1 1 2 + , ,

(22)

7 Numerical Example: Impact with two pipes Here, one example is select to be analysed using the formulations presented in this paper. This numerical example covers frictional elastoplastic impact problems. A finite element code was written by Bandeira using C++, entitled CMAP Contact Mechanics Analysis Program to model these examples. This problem consists of two steel pipes both having a thickness of 0.01 meter, external diameter of 0.05 meter and length of 1.26 meter, see Figure 3. One pipe is oriented horizontally and completely fixed at both ends. The other pipe is above to the horizontal one and completely fixed at one end in a plane normal to the horizontal pipe. Both pipes have yield strength of 250 MPa and hardening modulus of 250 MPa. A impact force of 1 kN is applied at all nodes belonging the last 0.112 meter of the other end of the superior pipe. This load is applied in 10 times increments of 0.15 microseconds. Brick elements were used in the mesh. Six-hundred steps were used with a termination time of 90 microseconds.

Figure 3 Initial and final configuration In the initial configuration the pipe rests on the block without generating contact forces. The contact area in the initial configuration is exactly a point. The problem involves 27.300 degrees of freedom. The penalty parameters used are N = 10 and T = 10. The friction coefficient between the contact surfaces is set to = 0,4. Results from the pipe impact are shown in Figure 4.

Figure 4a Equilibrium Configuration

Figure 4b Parameter

Figure 4 Numerical results at time 90 ms (Step 600) 8 Conclusions The algorithmic presented in this paper can be applied to analyse contact problems and impact problems in three-dimensions undergoing large elastoplastic deformations. The theories developed in this paper work perfectly. The overall performance of the developed algorithms was remarkably efficient yielding good results and quadratic rate of convergence. The contact formulations have been usually developed in two dimensions. For the three-dimensional analysis the formulation of contact is much more complicated. One of the reasons, obviously, is the complexity involving space geometry when the projection of the slave node on the master surface is calculated. This requires a sophisticated algorithm. [5] developed the node-to-surface contact element considering the contact between a slave node and a surface element. In that formulation the sliding of the slave node from a master surface to the adjacent one is not allowed. The contact formulation presented in [1] allows for the general sliding of a slave node on the master surfaces. Therefore, this formulation constitutes a general and consistent contribution to the contact mechanics. To achieve this, the node-to-edge and the node-to-node contact formulations are employed. An outstanding feature is the quadratic rate of convergence within a Newton equation-solving strategy, owing to the exact linearization employed. The algorithmic presented in this paper had a good performance to solve the numerical example of impact problem.

9 Acknowledgements The authors wish to express sincere appreciation to the FAPESB1 and CNPq2 for the support obtained in a project approved at Edictal FAPESB/CNPq 25/2010. Part of this work is a result of this research project. The authors would like to thank also the Master of Science Program of Structures Engineering of UFBA3 for the support to present this article in this congress. 10 References [1] A. A. Bandeira, P. Wriggers and P. d. M. Pimenta, "Homogenization Methods Leading to Interface Laws of Contact Mechanics A Finite Element Approach for Large 3D Deformation using Augmented Lagrangian Method," in International Journal of Numerical Methods in Engineering, London, England, 2001. [2] K.-J. Bathe, Finite Element Procedures, First ed., New Jersey: Prentice-Hall, Englewood Cliffs, 1996. [3] N. M. Newmark, "A Method of Computational for Structural Dynamics," ASCE Journal of Engineering Mechanics Division, vol. 85, pp. 67-94, 1959. [4] T. A. Laursen and B. N. Maker, An augmented Lagrangian quasi-Newton solver for constrained nonlinear finite element applications, International Journal for Numerical Methods in Engineering, vol. 38, pp. 3571-3590, 1995. [5] T. A. Laursen and J. C. Simo, A Continuum-Based Finite Element Formulation for the Implicit Solution of Multibody, Large Deformation Frictional Contact Problems, Int. J. Num. Meth. Engng., vol. 36, pp. 3451-3485, 1993a. [6] J. O. Hallquist, G. L. Goudreau and D. J. Benson, "Sliding interfaces with contact-impact in large-scale lagrangian computations," Computer Methods in Applied Mechanics and Engineering, vol. 51, pp. 107-137, 1985. [7] J. H. Heegaard and A. Curnier, An Augmented Lagrangian Method for Discrete Large-Slip Contact Problems, International Journal for Numerical Methods in Engineering, vol. 36, pp. 569-593, 1993. [8] J. C. Simo and T. A. Laursen, "An augmented Lagrangian Treatment of Contact Problems Involving Friction," Computers & Structures, vol. 42, pp. 97-116, 1992. [9] A. A. Bandeira, P. Wriggers and P. d. M. Pimenta, "Computational Analysis of Contact Mechanics undergoing Large 3D Deformation," in 2rd European Congress of Computational Mechanics, Cracow, Poland, 2001. [10] J. T. Oden and E. B. Pires, "Algorithms and Numerical Results for Finite Element Approximations of Contact Problems with Non-Classical Friction Laws," Computer & Structures, vol. 19, pp. 137-147, 1983. [11] H. Parisch, A Consistent Tangent Stiffness Matrix for Three-Dimensional Non-Linear Contact Analysis, International Journal for Numerical Methods in Engineering, vol. 28, pp. 1803-1812, 1989. [12] P. Alart and A. Curnier, "A Mixed Formulation for Frictional Contact Problems prone to Newton like Solution Methods," Computational Methods in Applied Mechanics and Engineering, vol. 92, pp. 353-375, 1991. [13] P. Wriggers, "Finite Element Algorithms for Contact Problems," Archives of Computational Methods in Engineering, vol. 2, pp. 1-49, 1995. [14] P.Wriggers and J. C. Simo, A Note on Tangent Stiffness for Fully Nonlinear Contact Problems, Communications in Applied Numerical Methods, vol. 1, pp. 199-203, 1985. [15] P. Wriggers, J. C. Simo and R. L. Taylor, "Penalty and Augmented Lagrangian Formulations for Contact Problems," in Proceedings of NUMETA 85 Conference, Balkema, Rotterdam, 1985. [16] P. Wriggers, T. V. Van and E. Stein, "Finite Element Formulation of Large Deformation Impact-Contact Problems with Friction," Computers & Structures, vol. 37, pp. 319-331, 1990. [17] A. Curnier, "A Theory of Friction," International Journal for Solids Structures, pp. 637-647, 1984. [18] T. A. Laursen and J. C. Simo, "Algorithmic Symmetrization of Coulomb Frictional Problems using Augmented Lagrangians," Computer Methods in Applied Mechanics and Engineering, vol. 108, pp. 133-146, 1993b. [19] D. Tabor, "Friction The Present State of Our Understanding," Journal of Lubrication Technology, vol. 103, pp. 169-116, 1981. [20] P. M. Pimenta, "Finite Deformation Soil Plasticity on Principal Axes," in III Complas, Barcelona, Espanha, 1992. [21] A. A. Bandeira, P. d. M. Pimenta and P. Wriggers, "A 3D study of the contact interface behavior using elastic-plastic constitutive equations," Lecture Notes In Applied And Computational Mechanics, vol. 27, pp. 313-324, 2006. [22] R. Fletcher, Practical methods of optimization, vol. 2, Chichester: John Wiley & Sons, 1980. [23] D. G. Luenberger, Linear and Nonlinear Programming, 2. ed. Reading ed., Massachusetts: Addison-Wesley Publishing Company, 1984. [24] D. P. Bertsekas, Nonlinear programming, Belmont: Athena Scientific, 1995. [25] G. Farin, Curves and Surfaces for Computer Aided Geometric Design: A Practical Guide, Third ed., Department of Computer
1

FAPESB (Bahia Research Foundation): Is an independent public foundation with the mission to foster research and the scientific and technological development of the State of Bahia. 2 CNPq (National Counsel of Technological and Scientific Development): Is an independent public foundation with the mission to foster research and the scientific and technological development of all States of Brazil. 3 UFBA (Federal University of Bahia).

Science, Arizona State University: Academic Press, 1993. [26] W. H. PRESS, Numerical recipes in C: the art of scientific computing, Second ed., Cambridge: University Press, 1995.

11 Annex A ALTERNATE ALGORITHM OF AUGMENTED LAGRANGIAN FOR IMPACT WITH FRICTION I) Initialize the iteration process: a) Set initial displacement 0,25 0,5 + time increment and the penalty values , ,initial velocity , initial acceleration e . Make . The time increment suggested is t = = , Newmark coefficients

.The Newmark coefficients must meet , where

= = 0,25 e

= = 0,5,

finite element, is the Young coefficient, is the Poisson coefficient, is the volumetric weight of the material and = = , = + , = and make = e is the wave speed in a dimension. Set the Lagrangian multipliers = . Set the frictional coefficient = + 0,2 + , where is the adhesion, is the plowing and 5 10. b) Let the vector of equality constraints, the vector of inequality constraints (normal contact force) and the inequality constraints (tangential contact force). Compute the exact linearization: , , , = , , , + + + + = + + , + , + 1 , , , = 1 1 1 2
, , , , ,

is the size of the smallest side of

0,5 and

c) Solve the initial search direction = = . , = . d) Initialize energies: e) Set iteration counters: j (BFGS), i (Newton Method) and k (Augmented Lagrangian Method): i = 0, j = 0, k = 0. Make = 0 and go to step IIa. II) Loop on i until equilibrium: a) Line search and solution vector update: i) Initialize: = 1,0. = . = ii) Evaluate: iii) Perform line search if required: IF > or THEN itere para 0,1 at b) Update: = c) Divergence check: IF , , , , , . , <0, . , , , , , , .

+ 1

1 + + + by solving this linear system:

1 +

THEN go to Table 2 (Newton Method). d) Equilibrium check: IF , THEN go to step III. e) Line search stepsize check: IF < 0,001 , THEN go to Table 2 (Newton Method). f) BFGS update limit check: IF j = max_bfgs , THEN go to Table 2 (Newton Method). g) Go to Table 3 (BFGS Method). III ) Augmented a) Compute new multipliers: for all constraints. Equality constraints: = + Inequality constraints: Normal force.

>

Inequality constraints: Tangential force IF THEN ELSE IF:

b) Make k = k + 1 and actualize the Augmented Lagrangian multipliers: c) Check for multiplier convergence: IF 1, 2, , for all for all

= =

+ ,

. . , ,

+ 1 + 1 2 1 1 1 = 1 2 Make = + 1 IF there is a new time increment of Newmark method, THEN go to step IIa. ELSE IF, the system converged. ELSE IF, go to step IIId. d) Re-initialize with new multipliers: ii) Solve iii) Reset: i) Compute new residual: = = , , using this linear system: , , , , , , , , , , , , . .

1+ 3 + , for all , THEN stop. The system converged. Update the following vectors by using Newmark Method: = and =

, such that,

. .

<

e) Repeat equilibrium search: go to step IIa.

Table 1: Alternate algorithm of Augmented Lagrangian for impact with friction a) Update: i = i + 1. b) Determine the exactly linearization: , , = , , + , , Newton Method ,

= =

+ ,

c) Determine the direction d) Actualize the energy: = ,

where

1 + + + + , , , ,

+ 1 + ,

. + by solving the linear equation system: , , , , . ,

+ +

+ ,

1 =

1 1 2

e) Make j = 0 and go to step IIa of the algorithm described in Table 1.

Table 2: Newton Method

a) Update the BFGS counters: i = i + 1, j = j + 1. b) Determine the BFGS vectors e , where: = , , , , , ,

MTODO DO BFGS ,

, , , ,

, , , ,

, , ,

, , ,

, ,

, , ,

, ,

, ,

, ,

, ,

d) Evaluate: = , , , , e) Go to step IIa of the algorithm presented on Table 1.

c) Solve

using this linear system:

, ,

, ,

, ,

Table 3: BFGS Method