Sie sind auf Seite 1von 13

COMMUNICATIONS IN NUMERICAL METHODS IN ENGINEERING Commun. Numer. Meth.

Engng 2007; 23:227239 Published online 7 September 2006 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/cnm.894

Enhanced remeshing from STL les with applications to surface grid generation
Desheng Wang1, , Oubay Hassan2, , , Kenneth Morgan2, and Nigel Weatherill2,
1 School 2 Civil

of Physical and Mathematical Sciences, Nanyang Technological University, Singapore and Computational Engineering Centre, School of Engineering, University of Wales Swansea, Singleton Park, Swansea SA2 8PP, U.K.

SUMMARY In this paper, we present a robust and efcient technique for surface meshing based on Stereolithography (STL) surface denition. Ridges are correctly identied using checks on dihedral angle and comparisons of the ratio of triangle edge lengths. Robust edge swapping, splitting and contraction are performed with stringent element validity checks. The improved remeshing technique is applied directly to CAD models and it is also used to improve surface triangulations which are reconstructed from sectional contours extracted from medical images. All the application examples, including the circulatory system and the urinary system, demonstrate that the enhanced remeshing procedure is robust and efcient for surface grid generation. Copyright q 2006 John Wiley & Sons, Ltd.
Received 11 July 2005; Revised 13 May 2006; Accepted 15 May 2006 KEY WORDS:

surface remeshing; edge swapping; edge splitting; edge contraction

1. INTRODUCTION Most boundary representations for engineering simulations such as computational uid dynamics (CFD) originate directly from computer-aided design (CAD) systems. IGES and STL are the two most common types of output format from CAD systems. The generation of an STL triangulation is very efcient and it can approximate most geometries very precisely. Recently, various papers have addressed the generation of surface grids from STL surface representations [15]. Surface
Correspondence

to: Oubay Hassan, Civil and Computational Engineering Centre, School of Engineering, University of Wales Swansea, Singleton Park, Swansea SA2 8PP, U.K. E-mail: o.hassan@swansea.ac.uk E-mail: desheng@ntu.edu.sg E-mail: K.Morgan@Swansea.ac.uk E-mail: N.P.Weatherill@swansea.ac.uk Contract/grant sponsor: Nanyang Technological University Start-up Grant; contract/grant number: M58110000 Copyright q 2006 John Wiley & Sons, Ltd.

228

D. WANG ET AL.

remeshing is the most widely used technique for the generation of high-quality surface meshes from an STL surface denition [613]. This procedure includes, edge splitting/contraction together with edge swapping and nodal smoothing. For a successful implementation of the remeshing technique, ridges and corners have to be identied. However, automation of this identication procedure is a non-trivial task and may result in a degradation of the surface conformity whilst manual intervention results in a degradation of the efciency and robustness of the procedure. In this paper, surface remeshing from STL surface representation is enhanced by addressing various issues, which have not previously been addressed and normally contribute to the loss of robustness and efciency of this technique. First of all, the surface features such as ridges and corners are correctly identied using dihedral angle and the ratio of edge lengths in adjacent triangles. Secondly, the robustness of edge splitting/contraction procedures are improved by the continuous application of element validity checks. In addition, in order to stabilize the renement and coarsening of the edges, grouping of edges according to their length is carried out and the group with the longest edges is considered rst. Various complex congurations generated from CAD models or reconstructed from medical images are tested to demonstrate the capabilities of the improved remeshing technique. The remaining part of the paper is organized as follows. Section 2 discusses the characteristics of STL les and the preprocessing which is required for feature identication, Section 3 provides the details of the enhanced surface remeshing procedure and Section 4 describes the applications to surface grid generation from CAD models and medical images.

2. CHARACTERISTICS OF STL FILES AND PREPROCESSING Each facet (or triangle) in a STL triangulation is described by the co-ordinates of the three nodes and the three components of the normal vector. The facets may exhibit very large aspect ratios which match the ratio of the two principal surface curvatures. Hence, many thin or small triangles exist which do not have the required size and quality for use in numerical simulations. Several papers proposed approaches for the generation of surface meshes suitable for FEM or CFD simulations directly from STL triangulations [15]. Also, some methods have been developed for the repair of STL les, which have overlapping surfaces, gaps, and intersections [4, 14]. In the present paper, the STL triangulation is always assumed to be valid and watertight. The following preprocessing procedure, which consists of four steps, is performed to identify all geometrical features to be preserved and to obtain an initial surface triangulation suitable for the remeshing. (I) Merging of duplicate points. An STL le is composed of the vertex co-ordinates and the associated vertex normals of each triangle independently. Therefore, to ensure consistency, a connectivity array has to be created. Nodes which are shared by multiple elements must be identied and uniquely numbered. In the present implementation, a digital binary tree is utilized for this purpose [4, 5]. (II) Surface orientation. Many of the operations required for feature identication and the remeshing process depend on the dihedral angle between adjacent facets. Hence, consistent orientation of the facet has to be maintained. This is achieved by ensuring that the orientation of the two elements connected to an edge is consistent. Any triangle on the surface can be considered as the starting triangle and placed in a stack and marked as a used triangle. Then a loop over the triangles in
Copyright q 2006 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2007; 23:227239 DOI: 10.1002/cnm

ENHANCED SURFACE REMESHING

229

the stack is performed and a check on the orientation of its unmarked neighbouring elements is conducted. If the orientation is consistent, the element is marked as used and added to the stack. Otherwise, the orientation is reversed before the element is marked and added to the stack. The procedure continues until the stack is empty. The normal of the triangles is computed according to the orientation. If any one of the three dihedral angles, which are formed by the normal of a triangle with the normals of its three vertices, is greater than a right angle, the orientation is reversed by a simple exchange operation. In the case of several closed surfaces, i.e. one outside and several inside surfaces, the above procedure is carried out for each closed loop and the orientation is determined based on the volume enclosed by each loop. (III) Feature identication. An STL le does not contain any information about geometric features such as ridges and corners. This information is essential for the correct placement of new nodes or the relocation of existing nodes using G1 interpolation [1517]. In these cases, multiple normals at ridges are normally required to avoid any loss of geometric accuracy. Hence, these features need to be derived rst such that they remain intact during the remeshing process. Several criteria for the identication of these features are applied. First, for each edge, the dihedral angle between two adjacent triangles is computed. Edges with a dihedral angle larger than a specied threshold are considered to be part of a ridge. The threshold value can be automatically determined by analysing the principal curvatures of the given geometry. In the present examples, an angle of 3045 has resulted from such analysis. This method can identify most of the ridges which are the intersections of two different parts of the geometry. However, in the case of a smooth blending of two shapes, the above method will fail to identify the full ridges, as shown in Figure 1. Here the dihedral angle for edge AB is less than 10 and therefore, the edge will not be identied as a ridge. To overcome this problem a second criterion based on the aspect ratio of the two adjacent triangles of each edge is used. An edge is always considered as a ridge if one of the two neighbouring elements has a large aspect ratio and the edge is the smallest of that triangle. The ratio criterion can be a user-specied value, however, a value of 1:10 is used for all the present examples. It is found that the second treatment can help to identify all the ridges as required. However, this may also result in over-identifying the number of ridges. Nevertheless, this over-identication will not result in a loss of accuracy in the geometry approximation when using G1 interpolation. The restriction imposed by this constraint will be overcome later by the use of ridge relaxation. The proposed technique can be applied to any given triangulation. In many such cases, a userspecication of ridges can exist, and hence it can be included in the feature identication procedure. Finally, when all ridges are correctly marked, any point which connects more than two ridges is considered as a corner point. In the following remeshing, a ridge is not allowed to be swapped, a corner point is not allowed to move, but a ridge point can be allowed to move along the ridge.

A B

Figure 1. Ridge identication: AB connects two triangles with large differences in edge lengths.
Copyright q 2006 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2007; 23:227239 DOI: 10.1002/cnm

230

D. WANG ET AL.

(IV) Multi-normal and curvature computation. For any internal point, i.e. non-ridge point or corner point, its normal is computed using the normals of its connecting triangles. Area weighted average is employed for this calculation. For the computation of the curvatures the method described in Reference [6] is applied. This method utilizes quadratic surface interpolation through the connecting points of a vertex. Using the curvatures, an isotropic or anisotropic sizing can be derived easily. But for a ridge point or a corner point, multi-normals and curvatures must be computed and stored. The triangles connecting the point are divided into several patches separated by the ridges. Using the triangles forming each patch, a normal and curvature is computed. For isotropic meshing, if curvature is to be used to determine the spacing required at a given point, the minimum of the two spacings computed using the two principal directions of curvature is used. For ridge points, the spacing is taken to be the minimum spacing obtained on each of the surrounding patches. These multiple treatments of normals and curvatures ensure that the renement process will produce meshes which conform accurately to the given geometry.

3. REMESHING PROCEDURE WITH ENHANCED ROBUSTNESS AND EFFICIENCY For the remeshing of the initial surface triangulation, local operations which include edge splitting/ contraction, edge swapping and point smoothing are performed [6, 8]. Here, only some unaddressed issues which are concerned about the robustness and efciency of the procedure are discussed in detail. The edge splitting and contraction are related to normalized edge lengths with respect to the computed spacing. The spacing can be dened by the use of sources [17], background mesh, octree-based interpolation, or some other kind of controls [18]. Also, a curvature-controlled metric can be used for computation of the spacing to ensure that geometric accuracy is achieved. Sizing derivation from curvature has been well developed in several papers [6, 1921]. The main radii of the curvature are used to construct a geometric metric and this metric prescribes mesh sizes as well as element stretching directions at mesh vertices [6]. For isotropic meshing, the largest radius is used, while for an anisotropic case, two radii are used. For the computation of the curvatures, a local quadric surface is constructed, which is based on a least-square t of neighbouring mesh vertices to the point concerned [6]. Based on this quadric surface, the curvatures can be derived easily by computing the fundamental forms. Using the computed spacing, the edge normalized length is computed [6, 20]. A normalized edge length between Cs (which is set to be 2 here) and Cc (which is set to be 1/ 2 here) is accepted whilst edges having a normalized length larger that Cs are split and edges with a normalized length smaller that Cc are collapsed. Before the remeshing, the sizing given or computed from curvatures can be smoothed with gradation, such as the H-correction proposed in Reference [22]. This will reduce the ratio of neighbouring element edge lengths. The smaller the gradation parameter, the smoother the nal mesh. Also, the gradation can be performed before the mesh optimization. In the following, four elementary mesh modication operators will be presented, followed by a complete procedure of the mesh renement and coarsening. 3.1. Edge splitting Suppose an edge BD, shown in Figure 2, has a normalized edge length larger than the given splitting criterion Cs , then a new point P, on the surface, will be required to split the edge into
Copyright q 2006 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2007; 23:227239 DOI: 10.1002/cnm

ENHANCED SURFACE REMESHING

231
D

P A M C A

C A

(a)

(b)

(c)

Figure 2. Edge splitting using G1 interpolation.

M B

(a)

(b)

(c)

Figure 3. Two edge contraction schemes can be used: (a) AB is to be contracted; (b) the G1 interpolation of the midpoint is chosen; and (c) B is chosen to be kept unchanged.

two smaller edges BP and PD. The triangulation will then be updated and the two adjacent triangles to edge BD will be replaced by four new triangles and two new edges will be created. Initially, the midpoint M of the edge BD is chosen and is projected onto the surface using the G1 interpolation [16, 17]. Figure 2 shows the complete process of edge splitting. Usually, after the point insertion, neighbouring edge swapping is locally performed to improve the geometry approximation as well as the element shape quality. If the edge to be split is a ridge, the multi-normals are used to project the point onto the two adjacent surfaces using the G1 interpolation. This is achieved by rst computing two physical co-ordinates of the new points using the two adjacent surfaces to the ridge. The nal physical coordinate is computed using a weighted average based on the curvature of the two adjacent surfaces. The weighting ensures that the points will be moved toward the surface with high curvature. 3.2. Edge contraction When the metric-controlled edge length of an edge is less than the collapse criterion Cc , the edge will be removed by merging the two points forming the edge under consideration. The merged point can be either one of the two end points or a new point located along the edge to be removed. This is
Copyright q 2006 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2007; 23:227239 DOI: 10.1002/cnm

232

D. WANG ET AL.

illustrated in Figure 3. When an edge is contracted the element data structure must be updated and the two triangles neighbouring the edge have to be removed. In this paper, the merging technique shown in Figure 3(c) is applied and it is proven to be adequate for the geometry approximation. Special care is needed when collapsing edges which have one or more of their nodes on a ridge. In this case, the internal point will be moved to coincide with the ridge point. Furthermore, edges which have their two points on a ridge are only considered if the two points are two consecutive points on the ridge. The difculty in the edge collapse procedure for surfaces dened using the STL-type format is highlighted in many publications [4, 5]. Continuous edge collapsing may result in self-intersecting surfaces. Therefore, a rigorous validity check of the new local conguration has to be carried out before the collapse is accepted. Here, the following procedure is carried out to ensure the consistency of the resultant triangulation: 1. Suppose AB in Figure 4(a) is the edge to be collapsed and that vertex A is to be moved to B. The local conguration is checked initially to determine if there is a 3-division conguration neighbouring the edge, as illustrated in Figure 4(a). In this case, contraction is not allowed since a negative element will be created as shown in Figure 4(b). 2. The signed areas of the triangles in the new conguration are calculated in the tangent plane at point B. Any negative area indicates that overlapping has occurred and, in this case, contraction is not allowed. 3. A further check is performed in the physical space to ensure that the new conguration will not contain triangles with minimum angle tending to zero. In addition, the dihedral angles between all the neighbouring triangles in the new conguration are determined and any conguration resulting in an angle larger than a given criterion (here 45 is used) is rejected. After contraction, neighbouring edge swapping is applied to improve the shape of the new elements and the geometry approximation. 3.3. Edge swapping The edge swapping process is illustrated in Figure 5. The edge BD in Figure 5(a) is identied as a non-ridge edge that connects the surface triangles ABD and BCD. The objective of edge swapping is to replace the connection BD by the diagonal edge AC. The conguration after the swapping is shown in Figure 5(b). If the new conguration has better quality than the original one, the

D A E C B B

E C

(a)

(b)

Figure 4. An overlap is generated if the edge is contracted.


Copyright q 2006 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2007; 23:227239 DOI: 10.1002/cnm

ENHANCED SURFACE REMESHING

233

D N1 N2

D N1

N2

(a)

(b)

Figure 5. Edge swapping: (a) the initial conguration; and (b) the alternative conguration.

swapping operation is performed. Different measures may be adopted to analyse the quality of a conguration. Mesh quality measures can be classied as being either measures of element shape [18, 23] or measures of geometric approximation. The rst geometric approximation measure [6, 8] is the gap between the surface and the triangle. This is dened as the maximum angle between the triangle normal and the normals evaluated at the vertices of the triangles. The second measure is the smoothness criterion. This is dened as the angle between the normal of two adjacent triangles. Although other criteria can be used to measure the geometry approximation, the above criteria have proven sufcient for all the examples attempted. Various element quality measurements can be used for the element shape analysis such as minimum angles and number of triangles surrounding a point [18, 23]. In the swapping procedure, the validity of the surface triangulation must be maintained. This is accomplished by using procedures similar to those employed when edge contraction is performed. Normally, in the initial renement and coarsening, more attention is paid to the geometry approximation, while element quality becomes the main concern in the nal optimization. 3.4. Point smoothing Point smoothing, or relocation, modies the position of a vertex without changing the topology of the triangles meeting at the vertex. The objective is to nd an optimal position on the surface for the vertex, in the sense that this alternative conguration should have better quality, according to criteria such as minimal angle, shape quality, size conformity or others [18]. Here, only the element shape quality is considered and various smoothing methods can be used [18, 19, 24]. The smoothing takes place in the tangent plane constructed at the point under consideration [8]. Using the spring analogy method, a new point is located at the centroid of all the projected neighbouring triangles. The triangle which contains the new node is identied and the node is projected onto the surface using G1 interpolation. Finally, the triangulation which will result from replacing the initial point by the new point is evaluated using the same geometric and shape measures used in the edge swapping. In order to ensure robust implementation of the smoothing procedure, a relaxation technique is implemented. A point located on a ridge is only smoothed using the two adjacent points on the same ridge. 3.5. Renement/coarsening procedure The renement/coarsening procedure modies, iteratively, an existing surface triangulation through edge splitting or contraction. The objective is to ensure that elements are in better conformity with
Copyright q 2006 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2007; 23:227239 DOI: 10.1002/cnm

234

D. WANG ET AL.

the size specication. Edge lengths computed with respect to size or metric are compared with the given splitting parameter Cs and contraction parameter Cc . If the initial mesh is almost isotropic, sequential processing of the edges of the current mesh is feasible and efcient. However, the initial surface triangulation obtained from the STL le often contains very large aspect ratio triangles. In this case, the result is frequently to produce distorted local congurations with clusterings of many stretched triangles. Performing further operations, such as edge swapping and edge collapsing, while maintaining the surface integrity, will not always be possible. To overcome this problem, a grouping technique is applied, in which the edges are sorted into groups according to their lengths. The number of groups is determined automatically. The rst group contains all edges which are larger than half the maximum edge length present in the STL le. The subsequent groups include the edges which are smaller than the edges in the previous group but larger than half the maximum edge length in the previous group. This is continued until the remaining edges are less than the minimum splitting criterion. Within each group, there is no strict order. The group which contains the longest edges is processed rst and each group is updated dynamically. This guarantees that, in global terms, the longest edges are rened rst. The grouping technique results in a more stable and robust procedure for renement/coarsening of the initial mesh. During the remeshing procedure, it is possible for a split edge to appear again after a series of edge collapses and swapping of adjacent edges. This can occur in regions of high curvature where the split conguration may require swapping to improve the quality of the triangles, and hence reproducing the edge which was split before. In this case, the splitting of the edge will result in a new point which is coincident with an existing point. To overcome this problem, edge splitting and collapse have to be performed at the same time to avoid the creation of a zero length edge. As both splitting and coarsening are local operations, the computational time required increases linearly with the number of nodes added and edges merged. With the grouping technique, it is found that the total number of contraction operations is far less than the total number of splitting operations. Hence, the complete procedure is very efcient. When all normalized edges are bound by Cs and Cc ridge relaxation procedures are employed. In this method, the dihedral angle of the two adjacent triangles is checked for all ridge edges and, if the angle is less than the previously prescribed threshold, the edge is reconsidered as a non-ridge. Classical G1 interpolation, based on edge splitting/contraction, together with edge swapping, is applied to the reclassied edges and to their area of inuence. Finally, the surface triangulation is enhanced via a combination of edge swapping, point smoothing and node connectivity optimization. The combined application of these two post processing techniques is found to be very effective in improving the regularity and smoothness of the triangulated surface.

4. APPLICATIONS TO SURFACE GRID GENERATION The enhanced surface remeshing technique is rst applied to two geometries generated using standard CAD software and represented by STL triangulations. The geometries represent models of a circulatory system and a urinary system. In both cases, the mesh spacing is governed by surface curvatures. Table I provides statistical information about the initial and nal mesh generated using the proposed method. Here, for a surface triangle e, its quality (in the isotropic case) is dened as
Copyright q 2006 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2007; 23:227239 DOI: 10.1002/cnm

ENHANCED SURFACE REMESHING

235

Table I. Statistics of surface remeshing examples from STL les.


Exam. Circ.Sys Urin.Sys In.NP 21 536 3444 In.NE 43 080 6884 Fin.NP 76 962 121 549 Fin.NE 153 932 243 094 Re.Time (s) 52.6s 84.1s Q.Avg 0.977 0.976 Q.Min 0.467 0.534 G.Per 99.99 99.68

(a)

(b)

(c)

Figure 6. Surface remeshing from STL les for a circulatory system: (a) initial STL triangulation; and (b) and (c) local views of the nal surface mesh.
Copyright q 2006 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2007; 23:227239 DOI: 10.1002/cnm

236

D. WANG ET AL.

(a)

(b)

Figure 7. Surface remeshing from STL les for a urinary system: (a) initial surface triangulation (STL); and (b) local view of nal surface mesh.

3 2 Q e = 4 3.0Ae / i=1 L i , where Ae is the area of the triangle and L i is the length of the ith edge. Elements with Q e larger than 0.7 are considered to be of a good quality while elements with a value lower that 0.3 indicate poor-quality elements. For each geometry, the following information is provided: the initial number of points, In.NP, the initial number of elements, In.NE, the nal number of nodes after remeshing, Fin.NP, the nal number of elements after remeshing, Fin.NE, the CPU time (using a Pentium IV PC with a clock speed of 2 GHz) required for the remeshing, Re.Time, average mesh quality, Q.Avg, minimal element quality, Q.Min, the percentage of elements with good quality measure, Q e >0.7, G.Per. Figure 6 shows the original STL surface and details of the nal remeshed surface for the rst CAD model. The original surface mesh for the second model, together with the nal remeshed surface, are shown in Figure 7. It can be seen that the remeshing procedure is highly efcient and produces high-quality surface meshes. The time required to generate the surface meshes which comply with the required accuracy to represent a given surface is less that 20 s per 100 000 elements. Furthermore, in both cases, all the generated elements have a quality measure Q e higher that 0.4 which is a clear demonstration of the suitability of the meshes for use in analysis.
Copyright q 2006 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2007; 23:227239 DOI: 10.1002/cnm

ENHANCED SURFACE REMESHING

237

(a)

(b)

(c)

(d)

Figure 8. Remeshing examples of surface grid generation from medical images: (a) and (b) initial and nal surface triangulations of a hip; and (c) and (d) initial and nal surface triangulations of a heart.

Copyright q

2006 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2007; 23:227239 DOI: 10.1002/cnm

238

D. WANG ET AL.

Table II. Statistics of surface triangulation optimization examples.


Exam. Hip Heart In.NP 18 489 1801 In.NE 36 982 3618 Fin.NP 7756 2897 Fin.NE 15 516 5810 Re.Time (s) 11.7s 6.5s Q.Avg 0.978 0.974 Q.Min 0.825 0.632 G.Per 100.0 99.49

The second application of the improved remeshing technique is to optimize surface triangulations which are reconstructed from sectional contours or medical images directly. Two examples are presented in Figure 8. The rst example is a constructed geometry of a hip and the second example is a constructed geometry of a heart. In both cases a uniform mesh spacing was used as a nal target for the remeshing. Figure 8 shows the initial and the nal surface meshes of the two geometries. Table II provides statistical information about the initial and nal meshes generated using the proposed method.

5. CONCLUSIONS AND FUTURE WORK The enhanced procedure of remeshing from STL triangulation is both robust and efcient. It has been applied successfully to two kinds of surface grid generation: meshing for geometries from CAD systems with STL representations and mesh optimization for surface triangulation constructed from medical images or sectional contours. Numerical examples demonstrate that high-quality surface meshes can be generated in a robust and efcient manner. Future work will include addressing the issues of anisotropy and the automation of the detection of ridges and corners, in particular, for meshes extracted from scanned images.
REFERENCES 1. Ito Y, Nakahashi K. Surface triangulation for polygonal models based on CAD data. International Journal for Numerical Methods in Fluids 2002; 39:7596. 2. Stereolithography Interface Specication. 3D Systems, Inc.: CA, 1989. 3. Nakahashi K, Sharov D. Direct surface triangulation using the advancing front method. AIAA Paper 95-1686-CP, 1995. 4. Ito Y, Nakahashi K. Direct surface triangulation using stereolithography (STL) data. AIAA Paper 2000-0924, 2000. 5. B chet C, Cuilliere JC, Trochu F. Generation of a nite element MESH from stereolithography (STL) les. e Computer Aided Design 2002; 34:117. 6. Frey PJ. About surface remeshing. Proceedings of the 9th International Meshing Roundtable, 2000; 123156. 7. L hner R. Regridding surface triangulation. Journal of Computational Physics 1996; 126:110. o 8. Frey P, Borouchaki H. Geometric surface mesh optimization. Computing and Visualization in Science 1998; 1:113121. 9. Suzazhsky V, Gotsman C. Explicit surface remeshing. Proceedings of the Eurographics Symposium on Geometry Processing, 2003; 1728. 10. Hormann K, Labsik U, Greinor G. Remeshing triangulated surfaces with optimal parameterizations. Computer Aided Design 2001; 33:779788. 11. Rassineux A, Villon P, Savignat JM, Stab O. Surface remeshing by local Hermite diffuse interpolation. International Journal for Numerical Methods in Engineering 2000; 49:3149. 12. Frey PJ. Generation and adaptation of computational surface meshes from discrete anatomical data. International Journal for Numerical Methods in Engineering 2004; 60:10471049. 13. De Cougny HL. Renement and coarsening of surface meshes. Engineering with Computers 1998; 14:214222. Copyright q 2006 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2007; 23:227239 DOI: 10.1002/cnm

ENHANCED SURFACE REMESHING

239

14. Butlin G, Stops C. CAD data repair. Proceedings of the 5th International Meshing Roundtable, Pennsylvania, U.S.A., 1996; 712. 15. Farin G. Curves and Surfaces for Computer Aided Geometric Design. Academic Press: New York, 1988. 16. Walton DJ, Meek DS. A triangular G 1 patch from boundary curves. Computer Aided Design 1996; 28:113123. 17. Weatherill NP, Marchant MJ, Hassan O, Marcum DL. Grid adaptation using a distribution of sources applied to inviscid compressible ow simulations. International Journal for Numerical Methods in Engineering 1994; 19:739764. 18. Thompson J, Soni BK, Weatherill NP. Handbook of Grid Generation. CRC Press: LLC, Boca Raton, FL, 1999. 19. Borouchaki H, George PL. Parametric surface meshing using a combined advancing-front generalized Delaunay approach. International Journal for Numerical Methods in Engineering 2000; 49:233259. 20. George PL, Borouchaki H. Delaunay Triangulation and Meshing. Application to Finite Elements Methods. Herm` s: e Paris, 1998. 21. Lee CK. On curvature element-size control in metric surface mesh generation. International Journal for Numerical Methods in Engineering 2001; 50:787807. 22. Borouchaki H, Hecht F, Frey PJ. Mesh gradation control. International Journal for Numerical Methods in Engineering 1998; 43:11431157. 23. Frey PJ, Borouchaki H. Surface mesh quality evaluation. International Journal for Numerical Methods in Engineering 1998; 45:101118. 24. Knupp PM. Achieving nite element mesh quality via optimization of the Jacobian matrix norm and associated quantities. Part Ia framework for surface mesh optimization. International Journal for Numerical Methods in Engineering 2000; 48:401420.

Copyright q

2006 John Wiley & Sons, Ltd.

Commun. Numer. Meth. Engng 2007; 23:227239 DOI: 10.1002/cnm

Das könnte Ihnen auch gefallen