Computers and Structures 80 (2002) 1399–1408

www.elsevier.com/locate/compstruc

Some algorithms to correct a geometry in order to create

a finite element mesh

, G. Bugeda *, E. On
R. Ribo ~ate
International Center for Numerical Methods in Engineering (CIMNE), Universidad Polit

ecnica de Catalu~
na, Campus Norte UPC,
08034 Barcelona, Spain
Received 16 February 2001; accepted 4 May 2002

Abstract
One of the major difficulties in meshing 3D complex geometries is to deal with non-proper geometrical definitions
coming from CAD systems. Typically, CAD systems do not take care of the proper definition of the geometries for the
analysis purposes. In addition, the use of standard CAD files (IGES, VDA, . . .) for the transfer of geometries between
different systems introduce some additional difficulties.
In this work, a collection of algorithms to repair and/or to improve the geometry definitions are provided. The aim
of these algorithms is to make as easy as possible the generation of a mesh over complex geometries given some
minimum requirements of quality and correctness. The geometrical model will be considered as composed of a set of
NURBS lines and trimmed surfaces.
Some examples of application of the algorithms and of the meshes generated from the corrected geometry are also
presented in this work.

2002 Elsevier Science Ltd. All rights reserved.

Keywords: Mesh generation; Geometry correction; NURBS

1. Introduction able for other uses like visualization or CAM but, in

many cases, they are not good enough for meshing op-
Mesh generation using structured and unstructured erations. In these cases, it is necessary to adapt/repair
grids is still nowadays one of the bottlenecks in the the geometrical entities by changing their mathematical
practical application of the finite element method (see description while maintaining the same geometrical
[4,9,11,14]). shape. The final entities should be better suited for the
Geometrical models created for finite element mesh mesh generation operations (see [5,10]).
generation purposes are typically created by the mesh The algorithms proposed in this paper are considered
generation software or imported from an external CAD as a set of filters that will be applied to a geometry
system. In both cases, the geometrical model must sat- definition just after being imported or created. The al-
isfy a series of quality constraints. gorithms are:
Normally, geometrical models are constructed as a
set of NURBS lines and trimmed surfaces (see [1,3,6–8]).
These entities can be mathematically correct and suit- • collapse of entities,
• correction of the list of knots,
• reparametrization,
• conversion to similar cubic entity,
*
Corresponding author. Fax: +34-93-401-6517. • union of curves,
E-mail address: bugeda@cimne.upc.es (G. Bugeda). • reorientation of the boundary of the surfaces,
URL: http://cimne.upc.es/. • collapse of small angles.

0045-7949/02/$ - see front matter
2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 4 5 - 7 9 4 9 ( 0 2 ) 0 0 1 0 5 - 0
1400 R. Rib
o et al. / Computers and Structures 80 (2002) 1399–1408

With the help of these algorithms, it is possible to the list of knots that contains the value of u ðu 2 ½uI ;
automate the process of importing geometry from CAD uIþ1 Þ. Next, the following recursive expression can be
and mesh on it while maintaining an acceptable level applied:
of quality criteria. This process should be performed ~k1 k k1~k1
~ ð1  aki Þxk1
i1 Pi1 ðuÞ þ ai xi Pi ðuÞ
without the need of manual intervention. Pik ðuÞ ¼ k
To understand better what follows, a short descrip- xi


tion of the NURBS lines and surfaces entities is given. A
k ¼ 1; . . . ; n  r

ð1Þ
deeper description can be found in [2,12].
i ¼ I  n þ k þ 1; . . . ; I þ 1


with
0

~ ~

Pi ðuÞ ¼ Pi

x0 ¼ x
i i
1.1. NURBS lines
u  ui1
A NURBS line is a geometrical entity, described as a aki ¼
uiþnk  ui1
parametric line in the 3D space, that is defined with a set
of knots, a set of control points, and a set of weights if it
xki ¼ ð1  aki Þxk1 k k1
i1 þ ai xi
is rational (Fig. 1).
The knots are a list of non-decreasing numbers then:
u0 ; . . . ; uLþn that begin in the lower range of the pa-
rameter space (usually 0.0) and finish in the high range sðuÞ ¼ ~
~ nr
PIþ1 ðuÞ
of it (usually 1.0). A knot has multiplicity r if its value is
repeated r times inside the list of knots. The initial and
1.2. NURBS surfaces
the end knots must have multiplicity r ¼ n þ 1, where n
is the degree of the curve.
A NURBS surface is the extension of the NURBS
The control points ~ P1 ; . . . ; ~
PL are a list of points in the
line to one additional dimension in the parametric space.
3D space that are part of the NURBS definition. The
Most of the properties of the NURBS curves applies
line will interpolate the first and the last point and will
here. There is a list of knots for every parametric direc-
smoothly approximate the other points.
tion u and v. The control points are a set of Pij points
The weights x1 ; . . . ; xL are a set of non-negative real
with i 2 ½1; . . . ; Lu  and j 2 ½1; . . . ; Lv  (Fig. 2).
numbers, one for every control point. They allow the
The evaluation of the surface can be made in different
shape of the curve to change and also to have an exact
ways:
representation of conic curves like circles or ellipses.
The number of knots, Nk must be equal to Nk ¼ Np þ
• To evaluate first the NURBS curve corresponding to
n þ 1, where Np is the number of control points.
one of the parametric directions u (maintaining v
The evaluation of a NURBS curve for a given value
constant) and obtain a NURBS line. Then, to evalu-
of the parameter u can be done in a recursive manner.
ate the resulting NURBS curve in the second direc-
First, it is necessary to identify the interval ½uI ; uIþ1  in
tion v.

Fig. 1. Example of a quadratic NURBS line with four control Fig. 2. NURBS surface with the corresponding control poly-
points evaluated for a given value of u. gon.
 R. Rib
o et al. / Computers and Structures 80 (2002) 1399–1408 1401

• Recursive evaluation: similar to the expression (1) • Two curves are collapsed if they share their end
given for the NURBS curve. points and the maximum distance between them is
• Evaluation by means of the b-spline base. This tech- smaller than . In addition, they are also collapsed
nique is also usable for curves. if a curve belongs to the interior of another one. This
last property means that, in addition to the first crite-
The evaluation with the b-spline base is done by de- rion, its end points ~
Pi accomplish
fining a base of b-splines, Ni , as a recursive function in
the following recursive way: distð~
Pi ; LÞ < 

1 if ui1 6 u < ui
Ni0 ðuÞ ¼ where ~LðtÞ is the second curve. In the latter case, the
0 if not:
final result of the collapse operation is not a single
n u  ui1 uiþn  u n1
Ni ðuÞ ¼ Nin1 ðuÞ þ N ðuÞ curve but a group of curves as shown in Fig. 3.
uin1  ui1 uiþn  ui iþ1
• Two patch surfaces are collapsed if they share their
Then, the following expression is applied: boundary curves and a similar maximum distance
P P criterion is accomplished.
~ m n
i j xij Pij Ni ðuÞNj ðvÞ • Two volumes are considered as equivalent if they
sðu; vÞ ¼ P P
~ ð2Þ
i
m n
j xij Ni ðuÞNj ðvÞ
share their boundary surfaces.

The computation of the maximum distance between

2. Collapse of entities
different curves or surfaces is made in an approximated
way as follows: a fixed number of points ~Pi are chosen in
The standard CAD exchange formats (IGES,
the interior of the curve or surface depending on their
VDA, . . .) do not contain the topological connection
geometrical characteristics. The distance point-curve or
between the different surface patches defining a closed
point-surface di is computed for each of the selected
geometry. These files just contain the mathematical de-
points. Then, dmax is approximated as dmax ¼ maxðdi Þ.
scription of the patches. Nevertheless, in order to pro-
In general, a big number of control points and a high
ceed with a correct mesh generation it is necessary to
degree of the line or surface formulations will imply a
check the correctness of the 3D geometry by ensuring,
bigger geometrical complexity. For this reason, the
for example, that it is defined by a completely closed
number of selected points will be related to the two
surface. In addition, many mesh generation algorithms
mentioned values.
need to know the neighboring relation between patches
In the computation of the di values the ~ Pi0 point is
in order to advance in their work. For this reason, it is
obtained as the mapping of the original point over the
necessary to obtain the topological connection between
the different surface patches.
An important additional difficulty is that, very often,
the boundary curves defining neighboring patches are
very similar, but not identical. Normally, the corre-
sponding pairs of curves are close enough for the dif-
ferent visualization or CAM operations, but not for the
necessary operations for the analysis. This makes the
task of identifying neighbor patches very difficult.
In order to solve this difficulty it is necessary to define
a geometrical tolerance. When two different geometrical
entities are separated by a distance smaller than the
prescribed tolerance they are considered to be identical
and they are collapsed. This tolerance can be specified
by the user or can be obtained in an automatic way as a
certain percentage of the total size of the geometrical
model.
Given a geometrical tolerance , the collapse criteria
for the different geometrical entities are the following:

• Two points are collapsed into a single one if:

k~
P1  ~
P2 k < 
Fig. 3. The collapse procedure reduces the initial four curves to
where ~
Pi is the vector of coordinates of point i. only three.
1402 R. Rib
o et al. / Computers and Structures 80 (2002) 1399–1408

~
P1 ; . . . ; ~
PL ð4Þ

x1 ; . . . ; xL ð5Þ

the list (3) will be reduced to:

k0 ; . . . ; ki ; kiþ1 ; . . . ; kiþn ; kiþnþpþ1 ; . . . ; kLþn

and the list of control points (4) and weights (5) to:
P1 ; . . . ; ~
~ Piþ1 ; ~
Piþpþ2 ; . . . ; ~
PL ð6Þ

x1 ; . . . ; xiþ1 ; xiþpþ2 ; . . . ; xL ð7Þ

Note that the new list will have L  p control points and
that the resulting curve will be an approximation to the
original one, being the original one a mathematically
incorrect representation of a curve.
Fig. 4. Saving of elements when unnecessary details are elimi-
nated.
4. Reparametrization

geometrical entity and di ¼ k~ Pi  ~

Pi0 k. The mapping op- A very desiderable property for the mesh generation
eration is the transformation of an original point to the algorithm is to use the arc length as the defining pa-
closest point contained into a geometrical entity like a rameter over the curve or surface. This property is not
curve or surface. This procedure is described in [12]. normally available in the NURBS curves or surfaces.
The collapse operation is also useful to eliminate This can be partially corrected by using a constant dis-
small details of the geometry that can be of no relevance tance between each pair of consecutive control points.
for the analysis. If some lines or surface patches are Nevertheless, in order to obtain the best possible quality
smaller than the tolerance, they can disappear and their during the mesh generation process it is convenient to
neighbor entities are reconnected. This operation is improve the parametrization from the very beginning.
normally called feature reduction. From the practical The corresponding changes that are produced in the
point of view, many details that are crucial for the design curves or surfaces during the improvement of their
task, can be totally irrelevant for the analysis. This op- parametric definition are named reparametrization.
eration can save a lot of elements and degrees of free- A major problem arises when there are big discrep-
dom in the final finite element mesh. In Fig. 4 a ancies between the spacing of distances between the
graphical example of this possibility is shown. control points and the spacing between the corre-
sponding associated points of the curves or surfaces.
This problem can be detected by a comparison between
3. Correction of the list of knots the modulus of the derivatives in the following way:


 d~L



The list of knots is a set of non-decreasing real values 1  dt t¼ki 
that belong to the parametric space and are used for the <
<F ð8Þ
F   d~L




definition of a NURBS. A more detailed definition of  dt 
t¼kiþ
knot can be obtained in [2].
In the formulation of a NURBS curve or surface the where the ki in (8) represents an interior knot and F is a
list of knots cannot contain any knot with a multiplicity maximum acceptable parameter that, for acceptable
higher than the degree of the mathematical entity. Nev- parametrizations, can have values between 4 and 5. In
ertheless, in many cases the standard exchange files this formula, ~
LðtÞ represent the curve.
contain this type of error. In these cases, it is necessary The first step of the correcting process consists of
to correct the list of knots by eliminating the overdefined decomposing the curve into the addition of the set of the
knot and the associated weights and points. In this way, equivalent B
ezier curves. This can be done by inserting
if we have a n degree NURBS curve defined by L points multiple knots until a multiplicity equal to the order of
with the following list of knots: the mathematical entity is reached. The process of in-
serting knots is described in [2].
k0 ; . . . ; ki ; kiþ1 ; . . . ; kiþnþp ; . . . ; kLþn with From the geometrical and the parametrization points
ki ¼ kiþ1 ¼    ¼ kiþnþp ð3Þ of view, the defined curve is completely equivalent to the
 R. Rib
o et al. / Computers and Structures 80 (2002) 1399–1408 1403

original one. If we have a n degree curve and L defining

points with the following values:
k0 ; . . . ; ki ; . . . ; kLþn
ð9Þ
P1 ; . . . ; ~
~ PL

new knots are inserted with the same value of the already
existing knots into (9) until they all have a multiplicity of
the same order as the curve
k0 ; . . . ; k0 ; . . . ; ki ; . . . ; ki ; . . . ; kLþn ; . . . ; kLþn ð10Þ
|fflfflfflfflffl{zfflfflfflfflffl} |fflfflfflfflffl{zfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
nþ1 nþ1 nþ1

After inserting these knots, the number of control

points of the curve will increase. The new number of
points will B
ezier equal to the sum of the difference
between the original multiplicities and the order of each
of the original knots.
Taking the list of points and grouping them in sets of
n þ 1 points, it is possible to build successive B
ezier
curves with the same degree than the original curve and
Fig. 5. Comparison between the derivatives of the curve before
the same continuity between successive B
ezier curves. and after the reparametrization.
These new curves will not depend on the list of knots
and, therefore, their shape will not change if any of the
knots is moved. Consequently, the increments between exchange of CAD files is always carried out with different
each groups of knots can be recomputed in order to geometrical tolerances. Hence, it seem logical to allow
obtain a better parametrization. A possible modification geometrical changes below the limits of this tolerance.
based in the chord length parametrization consists in The process consists of computing a set of points ~ Pi
creating the following new list: belonging to the interior of the curve. These points will
be the base of an interpolation algorithm that will be
k0 ; . . . ; kn ; . . . ; ki ; . . . ; kiþn ; . . . ; kL0 ; . . . ; kL0 þn ð11Þ used for the definition of the new curve with the re-
quired approximation to the original one. The criteria
k0 ¼    ¼ kn ki ¼    ¼ kiþn kL0 ¼    ¼ kL0 þn ð12Þ for the selection of the number of points LI is obtained
by correlating the number of control points L with the
lc accepted tolerance. A new cubic curve can be interpo-
kiþnþ1  kiþn ¼ ð13Þ lated from the new set of points using different proce-
ltot
dures like the one described in [2]. Fig. 6 shows an
where lc in (13) is the length of the corresponding B
ezier
curve and ltot is the total length.
The new curve parametrized in this way will have the
same geometrical shape than the original one but with a
different parametric definition. Fig. 5 shows the typical
improvement than can be obtained with the described
algorithm.

5. Conversion to similar cubic entity

In some cases, the correction described in the last

section is not enough for ensuring a good parametriza-
tion. Typically, these cases arise when one or more
control points are repeated, or else when the orders of
magnitude of the distances between points have big
discrepancies between them.
In these cases it is acceptable to use a new curve or
surface not identical to the original one but with a good Fig. 6. Conversion of a NURBS to an approximated interpo-
enough approximation. We should keep in mind that the lant cubic NURBS.
1404 R. Rib
o et al. / Computers and Structures 80 (2002) 1399–1408

example of this type of conversion. In this case, the The new list will be:
discrepancy between both curves is high due to a point
akA;0 ; . . . ; akA;LAþn ; akA;LAþn þ bkB;0 ; . . . ; akA;LAþn
with C 0 continuity.
þ bkB;LBþn ð16Þ

6. Union of curves lA lB
a¼ b¼ ð17Þ
lA þ lB lA þ lB
It is possible to convert a set of connected NURBS
curves into a single curve. Normally, it is advantageous where lA and lB in (17) are the respective lengths of the
to keep the number of curves as small as possible be- curves.
cause it can make the mesh generation task easier.
Typically, the presence of extremely small segment lines
can increase the complexity of the mesh generation. 7. Subdivision of curves
The criteria for joining different curves are the fol-
lowing: Some of the algorithms related with the mesh gen-
eration task solve small non-linear systems of equations
1. Enough degree of continuity between the different that involve the derivatives of the analytical expression
segments must exist. Typically, a C 1 continuity is con- defining the curves. The presence of strong changes of
sidered as enough. these derivatives inside the curve can produce conver-
2. None of the segments can support any individual sur- gence difficulties in the solution of the mentioned sys-
face patch. The set of segments to be joined must be- tems. Due to this reason, it is very convenient to
long to the boundaries of the same surface patches. maintain a C 1 continuity over the whole curve (in
practice, small angle discontinuities can also be al-
Fig. 7 shows different examples of these situations. lowed). Hence, it is convenient to subdivide (cut) the
Before the union can proceed, the different curves original curves at points not satisfying the required
must have the same degree. Hence, the degree of all the continuity.
segments will be increased until they reach the maximum This subdivision is made through the insertion of
value n. Next, the different control points will be joined additional knots at the required cutting points until a
and a new list of knots will be computed starting from multiplicity equal to the order of the curve is reached.
the original ones in the following way: The new curves will be:
kA;0 ; . . . ; kA;LAþn curve A ð14Þ
k0 ; . . . ; ki ; . . . ; ki ; . . . ; kLþn
|fflfflfflfflffl{zfflfflfflfflffl}
kB;0 ; . . . ; kB;LBþn curve B ð15Þ nþ1

Fig. 7. Different criteria for accepting (or not) the union of curves.
 R. Rib
o et al. / Computers and Structures 80 (2002) 1399–1408 1405

k0 ki where ~
S ðu; vÞ is the parametric surface. If the curve lays
;...; curve A
ki ki on u ¼ 0 or 1 the corresponding similar expressions will
be used.
kiþnþ1  kiþnþ1 kLþn  kiþnþ1 An additional convenient check consists of comput-
;...; curve B
kLþn  ki kLþn  ki ing the normal vector to the surface in its center and
to compute the approximated normal vector to the
The control points will be simply spread depending on boundary curves. The approximate normal vector can
the number of knots corresponding to each curve. be computed as:
X
~L ¼
N Lðti Þ  ~
~ C ~Lðti þ Dti Þ  ~
C ð20Þ
8. Reorientation of the boundary of the surfaces i

where ~C is the center of the surface. The boundary

Some of the mesh generation algorithms require a
curves will be well oriented if:
proper orientation of all the boundary curves. A curve
~
LðtÞ that belongs to the boundary of a surface with a ~N
N ~L > 0
normal vector N~ is considered as well oriented if the
~
vector V defined by
~
~  dL
V~ ¼ N 9. Collapse of small angles
dt
always points towards the interior of the surface. A very common problem in files imported from CAD
In some cases, the information imported from a systems is that some of the surfaces contain almost null
CAD system does not satisfy the mentioned criterion. angles that make impossible the generation of a proper
It is then convenient to check this possibility and to mesh. In those cases, it is convenient to collapse part of
change the corresponding orientation when necessary the lines that define these angles until converting the
(Fig. 8). angles into bigger ones allowing to place acceptable
Nevertheless, some times it is not easy to identify the mesh elements.
interior of a surface. In this work, we use the fact that The collapsing angle will be computed depending on
for non-trimmed surfaces the curves must belong to the the given tolerance . In general, the resulting criterion
boundary of a NURBS surface. For trimmed surfaces, it should guarantee that the size of the resulting curve is in
is necessary to look for any of the trimming curves agreement with the rest of the contiguous curves (see
placed on the surface. Fig. 9).
A boundary line of a NURBS surface patch is well
oriented if:
10. Examples
LðtÞ d~
d~ S ðu; vÞ
 >0 ~
L2v¼0 ð18Þ
dt du The algorithms presented in previous sections have
been implemented into the GiD pre/postprocessing sys-
LðtÞ d~
d~ S ðu; vÞ tem [13] developed at CIMNE. The following exam-
 <0 ~
L2v¼1 ð19Þ
dt du ples are a set of representative applications of these

Fig. 8. Criterion of signs and orientations of the trimming lines Fig. 9. Collapse of a too small angle. The collapse will be ac-
of a trimmed surface. cepted if L is bigger than a minimum value.
1406 R. Rib
o et al. / Computers and Structures 80 (2002) 1399–1408

Fig. 10. Geometry of a set of casting teeth for construction Fig. 11. Fotorrealistic render of the previous geometry.
machines. This set is the one used in the casting process.

algorithms for the preparation of different geometries in model of a casting set of teeth. The number of surface
order to be meshed. Typically, the geometry has been patches used for the geometry definition is around 400.
defined using a CAD system and the corresponding in- In this particular case, before the mesh generation it has
formation has been introduced into GiD using standard been necessary to correct the geometry in order to create
CAD files (IGES, VDA, . . .). In all cases, the use of the a closed volume (without gaps). The final obtained mesh
presented algorithms has carried out the improvements/ has around 40,000 elements.
reparations needed to allow meshing of the geometry
without the necessity of any additional operation. 10.2. Tarazona cathedral

10.1. Geometry of a set of casting teeth Figs. 12 and 13 represent a structural analysis model
with a non-linear damage material model of the Tara-
This example (see Figs. 10 and 11), corresponds with zona cathedral. In this case, it has been necessary to
the generation of a finite element mesh over the solid correct many of the surfaces used to describe the fine

Fig. 12. Geometry of the Tarazona cathedral (Spain).

 R. Rib
o et al. / Computers and Structures 80 (2002) 1399–1408 1407

Fig. 13. Mesh of the cathedral.

details of the shape in some parts of the geometry. The

final mesh has around 40,000 elements.

10.3. Sheet stamping processes

This is a typical case corresponding with a sheet

stamping analysis (see Fig. 14). The geometry has been
modeled using traditional CAD systems and before
proceeding to the classical mesh generation operations it
has been necessary to use all the correction algorithms
presented in these pages. Some of the problems that Fig. 14. Analysis of sheet stamping processes.
usually appear in this type of geometries are:

• some surfaces are too small and must disappear,

• other surfaces have a bad mathematical description,
• some of them are correctly defined but with some
properties that are not suitable for meshing on them,
• some of the geometrical details are very small and
produce extremely complex geometries.

Fig. 14 shows a typical mesh for this type of cases

with around 400,000 finite elements.

10.4. Aluminium casting

These type of geometrical models have the same

problems as the previous one. However, they are more
complicated to improve due to the inherent complexity
of their surface patches (see Figs. 15 and 16).

11. Conclusions

CAD models are not always best suited as starting

points for mesh generation. CAD models and the Fig. 15. Mesh of an aluminium casting model.
1408 R. Rib
o et al. / Computers and Structures 80 (2002) 1399–1408

human interaction to a minimum and facilitating a lot

the mesh generation task.
An academic version of the GiD pre/postprocessing
system, where the algorithms presented in this paper
have been implemented, can be freely downloaded at
http://gid.cimne.upc.es.

References

[1] Knuth DN. In: The art of computer programming, vol. 3.

Addison-Wesley; 1973.
[2] Farin G. Curves and surfaces for CADG. 3rd ed. New
York: Academic Press Inc; 1993.
[3] Foley JD, van Dam A, Feiner SK, Hughes JF. Computer
graphics. Principles and Practice. 2nd ed. Addison-Wesley;
1993.
[4] George PL. Automatic mesh generation. Application to
finite element methods. New York: Wiley; 1991.
Fig. 16. Detail of the previous mesh. [5] Le Gouez JM, Boheas MA, Miles CJ. Data require-
ments for analysis modelling. Technical report, NAFEMS,
corresponding exchange files are more suited for design 1995.
and CAM operations than for the analysis task. For this [6] Hoschek J, Lasser D. Fundamentals of computer aided
reason, it is normally necessary to repair the CAD geometric design. Addison-Wesley; 1993.
[7] Lancaster P, Salkauskas K. Curve and surface fitting. 1st
models in order to make them suitable for mesh gener-
ed. Academic Press; 1988.
ation. [8] Levin A. Interpolating nets of curves by smooth subdivi-
The proposed algorithms have shown to be efficient sion surfaces. Computer Graphics Proceedings (SIG-
in dealing with three different problems: GRAPH) 1999:57–64.
[9] L€ohner R, Parikh P. Generation of three-dimensional
• The elimination of some small features that are nec- unstructured grids by the advancing front method. Int J
essary for the design but are not important for the Numer Meth Fluids 1988;(8):1135–49.
numerical analysis. [10] NAFEMS. Integrate cad and analysis, 1996.
• The correction of some mistakes and accuracies that [11] O~nate E. C
alculo de estructuras por el m
etodo de los
are found in many geometrical interchange files. elementos finitos. CIMNE 1995.
[12] Rib
o R. Development of an integrated system for the
• The change in the mathematical definition of some
geometry dealing, mesh generation and data for the finite
entities in order to make them better suited for mesh element analysis. Ph.D. Thesis, Universitat Polit
ecnica de
generation. Catalunya, 2000.
[13] Rib
o R et al. GiD reference manual. CIMNE, 1998.
All these algorithms can be executed automatically as [14] Zienkiewicz OC, Taylor R. In: The finite element method
a filter to the imported or created geometry reducing the 5th ed, vol. I. John Wiley and Sons; 2000.