Journal of Materials Processing Technology 134 (2003) 108114

Boundary controlled automatic remeshing for 2D simulation of metal forming

Abstract During nite element simulation of metal forming processes, the mesh undergoes large distortions leading to inaccurate results and numerical instability in the solution process. In order to overcome this problem, the domain is to be remeshed whenever the quality of the mesh or geometrical representation of the contacting boundary violates the desired criteria. This paper describes a strategy, wherein the mesh density during remeshing is controlled by regenerating the boundary segments dynamically, using a background mesh when mesh distorts above allowable limits. This, in conjunction with an advancing front technique is used for generating the nite element mesh. The background mesh species the desired length of boundary segments at different locations as per the geometrical conguration of the deforming body and the die. This gives a possibility of remeshing with one background mesh xed in space and/or multiple background meshes each attached to one of the dies. This also allows easier user intervention if required. The procedure is applied and demonstrated for 2D metal forming problems with quadrilateral mesh generation based on paving technique. # 2002 Elsevier Science B.V. All rights reserved.
Keywords: Remeshing; Background mesh; Quadrilateral mesh; Paving; Metal forming; Advancing front

1. Introduction Finite element simulation for metal forming processes has reached a stage of wide industrial application, where the simulator is used as a tool in the process design. Various types of formulations for simulation of diverse processes are well reported and documented [1,2]. The updated Lagrangian method is the most widely used framework for the ow of metal in the dies. However, due to the large deformation in the process and inherent nature of Lagrangian formulation with mesh nodes attached to material points, the nite element mesh distorts considerably. This leads to (a) inaccuracies in variable computation due to distorted elements, (b) numerical instability and divergence and (c) inaccurate description of contact boundary. Remeshing the domain at intermediate stages of deformation and mapping the variables from the old mesh to the new one can considerably alleviate this problem. Remeshing requires (1) a decision-making mechanism for the need to remesh, (2) an element size denition for the generation of new mesh and (3) an efcient algorithm for mapping of variables.
*

Corresponding author.

The most commonly used nite elements for simulation of 2D problems are linear triangular or quadrilateral elements. The application of constant stress triangle is limited by its accuracy, though the generation of mesh is considerably easier. The most popularly used element for metal forming simulations is the four-noded linear quadrilateral. However, the remeshing procedures for quadrilateral elements have not received much attention. This could be due to the difculties in generating all-quadrilateral meshes. Various methods have been proposed for remeshing in metal forming simulations. One way is to manually remesh the region by extracting the boundary of the deforming body at any given stage. However, this is a laborious procedure and not suited for industrial application. The h-adaptive meshing based on an error estimate is widely used in many nite element applications [3]. However, this is computationally expensive and does not take the geometry of dies into consideration. Joun and Lee [4] have applied a method that enhances quality of the elements by the optimal placement of the nodes without changing the element connectivity. They demonstrated this procedure by applying to metal forming problems. This could be an effective way of reducing the number of times a new mesh is generated, but cannot replace the need for complete remeshing.

0924-0136/02/$ see front matter # 2002 Elsevier Science B.V. All rights reserved. PII: S 0 9 2 4 - 0 1 3 6 ( 0 2 ) 0 1 0 1 2 - 9

B.P. Gautham et al. / Journal of Materials Processing Technology 134 (2003) 108114

109

Petersen et al. [5] have proposed a method wherein, the mesh is locally rened in zones where plastic deformation is expected to be the largest. This condition typically occurs close to free die edges or in vicinity of contact points. The renement is carried out after generation of an initial mesh. Petersen and Martins [6] have discussed remeshing strategy of rening the mesh based on strain rate and plastic strain gradient. They used a grid-based approach, linking the boundary of the grid with the contour of the component. Besides the criterion on element size distribution during remeshing which is based on eld variable distribution, it would also be necessary to have criteria based on geometry of the deforming body and the dies. In many a case it would be sufcient to regenerate the mesh only on the basis of geometrical considerations as these give indicators to the size of the mesh required. However, methods for carrying out efcient boundary rening is not widely discussed in literature. This paper describes a method for generating and rening the mesh of the deforming body based on contact and geometrical considerations for 2D simulations. A mesh generation scheme based on paving algorithm [7] is used to generate an all-quadrilateral mesh. In the paving algorithm, the element size and quality is controlled by the lengths of the boundary segments. The criteria required for the creation of the boundary segments in various zones in the new mesh is dened by introducing a controlling reference mesh (background mesh). This method helps rening the mesh where it is required, while coarsening it if the material is not undergoing considerable deformation. 2. Quadrilateral mesh generation A number of methods exist for the generation of allquadrilateral elements. The most common is to generate sub-domains of parametric spaces and mesh them. However, this technique requires user-identication of the suitable sub-spaces. Another popular technique is based on triangulation, which generates triangles and merges them to form quadrilaterals. Paving is an advancing front method proposed by Blacker and Stephenson [7] for generation of all-quadrilateral mesh. In this method, elements are added row by row on the paving boundary after appropriate identication of the boundary segments. In the process, the boundary advances until closure. White and Kinney [8] have enhanced this method by redesigning the algorithm to generate and rene one element at a time. In the paving technique, the element size near the boundary is governed by the length of the boundary segments. The elements created in the interior also depend on the previously created elements as element size changes gradually with the paving boundary moving away from the domain boundary. Hence it is strongly inuenced by the starting boundary discretisation.

In the current work, a mesh generation technique based on the paving method discussed by White and Kinney [8] is used with further enhancements for robustness. 3. Remeshing procedure The quality of the mesh is to be continuously monitored as the simulation progresses in order to make a decision on the need for remesh. Fig. 1 shows the owchart showing various steps involved in the remeshing procedure. In a typical metal forming situation, it is possible that a few elements may get excessively distorted in a small zone even while the average quality of the mesh is within limits, requiring a remesh. Hence, the quality check is to be taken as a combination of average mesh quality and quality of maximum distorted element. One such combination used by the authors is given below:  D p1  p2 d p3 amax p4 dmax 1 (1) a where D is an index representing mesh distortion. The variable a is the ratio of the largest and the smallest sides of the quadrilateral element, while d is the ratio of the larger to smaller diagonals. Over-bar indicates the average value, while subscript max indicates the maximum value over all the elements. The values used in the present analysis for p1p4 are 0.4, 0.4, 0.1, and 0.1, respectively. Remeshing is done when the value of D exceeds 0.5. 4. Background mesh The boundary of the deforming body comes in contact with the dies and this in turn governs the progress of deformation. The most signicant changes in the strain and stress occur near the boundary, either near a contact zone with the die or close to a fold or lap formation. Along with this, the boundary segments undergo the severest elongation or shortening, leading to distortion of the mesh and inaccurate representation of contact status. Hence it would be most prudent to concentrate rst on rening the representation of the boundary, which can be subsequently followed up with rening the mesh based on error estimates. We have proposed the use of a background mesh encompassing the domain of analysis that can be used for dening the segment length on the boundary. When the rened boundary is used with paving technique for generation of the mesh, a desired new mesh can be obtained. Background mesh consists of rectangles with sides aligned with global axes. Though this condition is not essential, it will be the best method for reducing time required for locating a point's placement in the background mesh. The area covered by all the background mesh rectangles should encompass boundaries at all times. It is not necessary that the background mesh should have common nodes. A node of any of the rectangles in the background mesh may lie on the side of its adjoining rectangle. Each rectangle in the background mesh is assigned unique segment length.

110

B.P. Gautham et al. / Journal of Materials Processing Technology 134 (2003) 108114

Fig. 1. Flowchart showing the remeshing sequence.

During remeshing, the boundary of the deforming body is extracted from the element connectivity of the current mesh. In the case of mesh consisting of linear elements such as four-noded quadrilaterals, the boundary is a polygon consisting of linear segments. The sides of this polygon, referred here as the boundary mesh, are to be rened as per the requirements dened by the background mesh. The mesh boundary is modied by a method, which takes the desired segment length from the background mesh and forms the new boundary discretisation. The rst step is to identify the rectangles of background mesh in which segments of the boundary under consideration fall in. The ratio of the boundary segment length to the desired segment length obtained from the background mesh is computed. Based on this ratio a decision of merge or divide is made. Various possible cases to be considered are: 1. If the length ratio of a segment exceeds a certain value then it will be divided into two or more segments. 2. If the length ratios of two consecutive segments are less than certain value, the segments are merged to one linear segment subject to criteria discussed below.

3. Another possibility is to join two or more successive segments and divide them into a number of new segments to get desired length elements. For case 1 the original boundary segment is divided when the length ratio exceeds 3/2. The number of segments into which the original segment is divided is the integer value of the length ratio multiplied by 4/3, such that the segment is divided into two for the length ratio of 3/2. Case 2 will lead to violation of volume conservation unless the segments to be merged are collinear. While merging two non-collinear segments, the common node will be lost and the new segment joins the other two nodes. In such a situation, a decision needs to be made depending on

Fig. 2. Schematic illustration for joining of the segments.

B.P. Gautham et al. / Journal of Materials Processing Technology 134 (2003) 108114

111

Fig. 3. Variation of maximum area loss with e.

the impact of such a merger. The decision is based on the ratio of the new segment length to the sum of lengths of the two original segments. The limits for this should be based on the amount of violation of volume conservation constraint. The criteria for merger of segments of length l1 and l2 is dened as l !e l1 l2 (2)

the dies. This concept can be further extended by specifying multiple background meshes, each attached to different dies. In such a case, the minimum segment length specication computed from all the background meshes is to be taken. 5. Mapping of field variables and contact conditions Once the domain is remeshed, it will be necessary to map eld variables such as stresses and strains from the old mesh to the new one. In the current implementation, rst the nodal values of eld variables of the old mesh are computed by carrying out global smoothing of the values available at the

where l is the length of the segment resulting from the merging operation as shown in Fig. 2, and e the limit for the decision to merge. The volume gained or lost will be equal to the volume of the triangle formed by the original segments and the merged segment. Fig. 3 shows the variation of A/l2 with e where A is the area of the triangle. The perturbation will lead to a departure d 2A=l from the original surface. In the present work the value of e is taken as 0.9 for which A/ l2 is approximately 0.1. In case 3, it would be appropriate to t a new local boundary curve from the nodes and make divisions. While using a background mesh, the user can make segment length specications based on the geometry of the dies and expected deformation. The background mesh based control also facilitates coarsening of the mesh when smaller boundary segments move into background mesh rectangle with larger length specication. As, it is not critical to obtain the desired length of boundary segments exactly, upper and lower bounds are provided over the desired segment length. A background mesh should be dened at the beginning of the simulation. The same background mesh is used for the generation of the initial mesh as well as subsequent remeshing. It is also possible to have user intervention by stopping the simulation at an intermediate stage and changing the background mesh. The background mesh can be xed in space or can move with one of the dies. In case of simulations such as extrusion, it will be advisable to have a xed background mesh. For forging problems, the mesh can be made to move with one of

Fig. 4. (a) A typical background mesh (thick line shows the domain to be meshed; numbers in background mesh rectangles indicate the desired length of boundary segments); (b) all-quadrilateral mesh generated using the background mesh.

112

B.P. Gautham et al. / Journal of Materials Processing Technology 134 (2003) 108114

element gauss points. In this process, the eld variables at nodes are computed by solving Mg C (3) where X M me ; X ce ; C
e e

me ij ce i

Z Z
dVe

Nie Nje dVe ;

dVe

Nie g dVe

and g is the eld variable to be computed at the nodes.

Next step is to compute the nodal eld variable values for the new mesh. This is done by identifying the element of the old mesh in which the current node falls and then computing the value using standard shape functions and nodal values of the old mesh. This information is further transferred to the gauss points of the new mesh. This procedure gives a smooth variation of eld variables on the mesh. In order to assess the error introduced due to the mapping, an error norm g e corresponding to variable g is dened as given in the following equation: s P n o 2 e g g dVe g (4) e P R o 2 e g dVe where the superscript o represents the value of variables as per the variable distribution in the original mesh and superscript n refers to the value of the variable at the same coordinate but as per the new mesh and mapped variable distribution. The summation is carried out over all elements corresponding to the original mesh. As the problem considered is elasto-plastic, it is possible that the components of the stress eld and the plastic strain eld after mapping may not satisfy the yield surface. In such a case, the following criteria are used: (a) If the equivalent stress is greater than the flow stress at that plastic strain, the stress components are scaled to yield surface. (b) If the equivalent stress is less than the flow stress value, the gauss point is set to unloading condition. Another important issue during migrating to a new mesh is the transfer of contact node status and contact reactions.

Fig. 5. Workpiece, initial and final die position for example 1.

Fig. 6. Mesh before and after remeshing for example 1 in the neck region.

B.P. Gautham et al. / Journal of Materials Processing Technology 134 (2003) 108114

113

The primary criterion is to satisfy equilibrium boundary conditions. In the current implementation, the internal forces in the remeshed domain are computed using the stress eld obtained after mapping. The boundary nodes of the new mesh are identied and checked for possible contact with the dies. Using this information and the internal nodal forces, the decision on the contact state is made. It is observed that this procedure gives smooth transition of the forces acting on the die. Any possible unbalance in the force introduced during this process gets eliminated during the subsequent increment.

6. Results A typical background mesh around a square domain is shown in Fig. 4(a). Here the square region of 50 50 is meshed using the boundary segment lengths specied on the background mesh. Fig. 4(b) shows the all-quadrilateral mesh generated using the background mesh and the paving algorithm. The concept of background mesh is implemented and tested for various 2D metal forming applications. Two examples are presented here. Fig. 5 shows the initial and nal positions of the dies and the initial position of the workpiece during extrusion of an axisymmetric body. Fig. 6 shows the mesh before and after remeshing using a background mesh at last remesh stage near the neck region. It can be seen that the mesh is distorted near the neck region before remeshing. After remeshing, the segments in the neck region

Fig. 8. Mesh before and after remeshing in the flash region of forging for example 2.

Fig. 7. Workpiece, initial and final die position for example 2.

Fig. 9. Cumulative loss of volume.

114

B.P. Gautham et al. / Journal of Materials Processing Technology 134 (2003) 108114

Fig. 10. Error norm for plastic strain due to mapping.

are smaller in length and the nal mesh produced is considerably improved. The second example shows renement of the mesh in the ash region of a hot forging. The initial and nal die locations and the initial workpiece location are shown in Fig. 7. The mesh before and after remeshing is shown in Fig. 8. It can be seen that mesh takes a more regular shape with ner elements near the curved boundary. The number of remesh steps required in these two cases are 14 and 23 for the two examples discussed above. The cumulative net volume loss as a function of remesh number is shown in Fig. 9. It can be seen that the total volume loss due to remesh is less than 0.05%. Fig. 10 shows the variation of error norm for plastic strain while mapping from old to new mesh with remesh number for both the examples discussed above. It is seen that the maximum norm is approximately 1.5%. 7. Conclusion This paper proposes a method of using a background mesh that controls the boundary segment lengths of a deforming body. The knowledge of the workpiece, die geometry and the expected deformations is necessary in dening the boundary mesh. Use of background mesh along with advancing front based mesh generation gives a very simple, robust and useful automatic remeshing scheme, which can be used for a number of metal forming applications.

The effectiveness of the method is demonstrated by applying to 2D bulk metal forming cases, using paving algorithm for all-quadrilateral mesh generation. However, the application of this method requires some knowledge of how the metal might ow. References
[1] S. Kobayashi, S.-I. Oh, S.-I. Altan, Metal Forming and the Finite Element Method, Oxford University Press, Oxford, 1989. [2] P. Hartley, I. Pillinger, C. Sturgess (Eds.), Numerical Modeling of Material Deformation Processes, Springer, Berlin, 1992. [3] O.C. Zienkewicz, R.L. Taylor, The Finite Element Method, vol. 1, Basic Formulation and Linear Problems, McGraw-Hill, Singapore, 1989. [4] M.S. Joun, M.C. Lee, Quadrilateral finite-element generation and mesh quality control for metal forming simulation, Int. J. Numer. Meth. Eng. 40 (1997) 40594075. [5] S.B. Petersen, B.P.P.A. Gouveia, J.M.C. Rodrigues, P.A.F. Martins, A metal-forming approach to automatic generation of graded initial quadrilateral finite element meshes, Eng. Comput. 15 (1998) 577 587. [6] S.B. Petersen, P.A.F. Martins, Finite element remeshing: a metal forming approach for quadrilateral mesh generation and refinement, Int. J. Numer. Meth. Eng. 40 (1997) 14491464. [7] T.D. Blacker, M.B. Stephenson, Paving: a new approach to automated quadrilateral mesh generation, Int. J. Numer. Meth. Eng. 32 (1991) 811847. [8] D.R. White, P. Kinney, Redesign of paving algorithm: robustness enhancements through element by element meshing, in: Proceedings of the Sixth International Meshing Roundtable, Sandia National Laboratories, October 1997, pp. 323335.