A MULTI-BLOCK ORTHOGONAL GRID GENERATION

USING CAD SYSTEM

Vladimir V. Chudanov1, Anna E.Aksenova, Valerii A. Pervichko,
Alexander G.Churbanov, Irina G.Plotnikova, Vladimir V. Varenkov
and Petr N. Vabishchevich
1
Nuclear Safety Institute, Moscow, Russia, [email protected]

ABSTRACT

In this work there is presented a non-traditional approach to generation of block-structured orthogonal grids on the basis of up-to-
date CAD systems. The features of the algorithm of constructing orthogonal grids are explained. Examples of constructed grids
are provided.
Keywords: mesh generation, computational geometry

and unstructured grids called as hybrid grids and grids

1. INTRODUCTION consisting of structured grids with zones of overlapping
(referred to as Chimera, composite or patched grids).
Generation of grids and their processing become more and From all listed above grids block-structured grids appear to
more important fields of scientific researches in connection be optimal in sense of compromise between possibilities to
with increasing needs of industry. An improvement of handle complex geometrical objects and possibilities to
processes of grid generation and processing depends on employ a wide variety of developed early programs of
improvement of mathematical, algorithmic and program numerical analysis. The application of block-structured grids
concepts. to modeling of complex geometrical configurations consists
Grid generation — the process allowing to obtain discrete in representation of the initial complicated domain via a set
models of computational domains from the common of more simple subregions — blocks. In this case a grid is
geometrical configurations. The geometrical configuration generated in each subregion and the link between subregions
can be discretized into grids of two principal types: structured is controlled according to the connectivity matrix for
and unstructured. subregions. As the node connectivity in such grids is defined
in a very simple way via indices, it makes possible to employ
Structured grids consist of quadrilateral in the 2D or a lot of well developed and easy checked algorithms.
hexahedral in 3D case cells. Connectivity of structured grid is Moreover, block-structured grids indicated some advantages
provided by the trivial identification of adjacent nodes with in implementation on parallel computing systems.
increasing of coordinate indices.
The overlapped grid approach has great versatility [1][2][3]
Unstructured grids are characterized by cells of an arbitrary and demonstrates both merits and demerits. The primary
form (usually triangles, quadrilateral in the 2D and bottleneck lies here in the necessity to connect components of
tetrahedrons, hexahedrons in the 3D case) and have not any a grid composition via interpolation that can yield the lack of
trivial identification of neighbours with increasing of an conservation. Different variants of hyperbolic and elliptic
index. For unstructured grids the set of nodes and the generators are used in this approach.
corresponding connectivity matrix are generated to define
irregular forms of cells as well as neighbour nodes. In Experience of numerical simulation shows that quality of
practice sometimes there are used combinations of structured computational grids makes an essential impact on accuracy of
numerical solutions. In this sense orthogonal or quasi- on its basis a new orthogonal grid using the developed in this
orthogonal grids are more preferable and provide significant work solver.
advantages in solution of systems of differential partial
equations: Below there are presented peculiarities of constructing initial
multi-block grids on the basis of CAD systems as well as the
∑ Transformation of governing equations performed in algorithm of transformation of initial grids into orthogonal
order to use such grids leads to appearing a small ones. Some examples of constructed grids are also provided
enough number of the additional terms, along with predictions of flows derived on them.

∑ Orthogonal grids provides the highest accuracy of

calculations, 2. APPLICATION OF CAD SYSTEMS TO
CONSTRUCTION OF MULTI-BLOCK GRIDS
∑ Boundary conditions of various type can be
implemented in a very simple way.
For construction of an initial multi-block grid we use modern
Elliptic generators are in common use to construct orthogonal CAD system — Mechanical Desktop (version 2.0). This
grids (see, for example, [4][5]). The basic difficulties of such choice is based on a number of reasons, and the primary is
an approach are connected with non-standard nonlinear the fact that the given product satisfies to the majority of the
boundary conditions and fulfillment of orthogonality on requirements that can be imposed upon to grid generation
boundaries of domains where the grid is designed. The systems.
primary merit is in the fact that the problem of grid
The basic requirements to grid generators can be formulated
construction is solved in a regular domain.
as the following list of necessary possibilities:
The alternative approach is connected with searching
∑ Construction of the initial geometry;
functions G ∈ R 3 (orthogonal set). For these functions the
classical mixed boundary value problem is formulated on the ∑ Editing and completion of the configuration;
basis of the selfadjoint elliptical equations of second order.
∑ Development of a block structure;
However, this well-posed problem is solved in an irregular
domain which, as a rule, is curvilinear and multiply ∑ Production of the final computational grid with a high
connected. Precisely this approach is considered in the level of a user's control on grid quality involving
present work. smoothness, orthogonality, spacing etc.
The problem of solution of the mixed boundary value Mechanical Desktop meets all the above requirements except
problem in an irregular computational domain for deriving the last.
mapping G ∈ R 3 is resolved here in a very simple fashion. A very important element for the efficient process of grid
First, in the computational domain there is constructed a non- generation of a grid is the development and implementation
orthogonal fine enough grid. Further, on the basis of the of a non-conflicting graphic user interface (GUI) [6].
finite-volume method (or variational-difference method) the Performance of GUI operating depends on the following
difference scheme is created. And finally, the corresponding requirements:
selfadjoint grid elliptic problem is solved using modern high-
performance iterative linear solvers. A general technology of ∑ Compatibility — possibility of information exchange
generation of orthogonal grids is based on this computational with other systems. For example, through the format
algorithm. It is very convenient to use this technology of grid IGES;
generation in a combination with a multi-block representation
∑ Consistency of logical organization of technology of
of a geometrical configuration.
supporting and making decisions: decomposition onto
Development of a complete system of grid generation is a subdomains, distribution of the topological information
very complex problem. Any grid generation system should etc.;
have all necessary attributes existing in modern well-known
systems of generation of block-structured grids, such as NGP, ∑ Easy in usage: convenient organization of the dialogue
ICEM-CFD, EAGLE, namely: a graphic user interface, a with a user;
model of grid topology, a built-in CAD system based on ∑ Existence of an expert subsystem: analysis of possibility
NURBS- splines, tools for analysis of constructed grids etc. of splitting onto subdomains, check of curve
On the other hand, the most part of modern CAD systems orientation etc.
provide many functions from the above list, such as a
powerfull graphic interface, a rich toolkit for construction of Structural possibilities of grid generation in most codes are
the complex geometrical configurations based on NURBS based on GTM data structure. Usage of topological objects of
representation, a possibility to design grids on domain the system (edges, surfaces and blocks) to control the grid
boundaries, analysis of constructed grid and so on. Besides, information for each subdomain allows to specify grid nodes
CAD systems allows to present a complex configuration as a only once and to use them in all adjacent objects where these
combination of simple domains. Thus, practically all nodes are common. These common nodes are defined by the
operations of grid generation are included and so, the only topological specification of surfaces and blocks in the
necessary operation which should be implemented is to take domain. Mechanical Desktop does have built-in GTM data
some multi-block grid from a CAD system and to construct structure and allows to prescribe topological information
through specification of text attributes to edges or surfaces of Ω whereas edges of G - into edges of square Ω . Suppose
the geometrical configuration. Further, this CAD system ∂ (x, y , z )
in addition that Jacobian of this mapping ℑ = >0.
∂ (α , β )
provides possibility to construct multi-block representation of
the domain, to check direction of curve orientation etc.
In domain Ω let us construct uniform grid with respect to
Geometry description in the most of modern commercial grid α, β with spacing hα , h β which serves as an image of some
generators is based on NURBS splines [7][8][9][10][11]. The
same way of presentation is used in Mechanical Desktop for curvilinear grid in domain G .
the geometrical database. Omitting intermediate transformations, let us write the
Generation of surface grids immediately on NURBS surfaces following representation for gradient ( T - additional
allows surface grids to preserve a high degree of accuracy variable):
with respect to the initial geometrical data. However, it also
requires high quality of models in CAD systems in order to
produce good grids. High quality here means that the model  lβ ∂T 
 
should not contain any undesirable intervals, overlaps or  grad α T   1 − cos ϕ  S ∂α 
 = 
intersections between surfaces. Many grid generators have  grad β T   − cos ϕ 1  lα ∂T 
    
internal CAD system in order to create, modify and/or restore ∂β
an imported geometry at preparation for grid generation.  S 

Production of the final computational grid with a high level

of user's control on grid quality including smoothness,
orthogonality and spacing was incorporated in our approach
(xα x β + yα y β + zα z β ) = lα l β cos ϕ (1)
into the code for constructing orthogonal grids. Thus, the
process of grid generation consists of the following four
items: l ⊥ = lα l β sin ϕ = S
∑ Construction, editing of topology of block structure
using facilities of Mechanical Desktop; Define functions θ = θ (α , β ) è η = η (α , β ) on a smooth 3D
r r
∑ Saving of derived results in the IGES; surface so that u = gradθ and w = gradη .

∑ Conversion of the saved information by means of

converter in the format of our external code for
r r
Then in local contravariant basis n α , n β
r
{ } vectors r
u and
construction of orthogonal grids; w can be written as follows:

∑ Construction of the final computational grid with high-

level control of a user on the grid quality (smoothness, u α = grad α θ wα = grad α η
è .
orthogonality and spacing). The given item can be u β = grad β θ w β = grad β η
performed several times. In other words, if quality of
the already obtained orthogonal grid seems to be
unsatisfactory, this grid can be passed on the input of Let us require that orthogonality constraint holds for vectors
r r
our code again. The repetition of this step allows to u ,w:
obtain an orthogonal grid of desirable quality.

(u )
This way of grid generation makes possible to design block- cos ϕ αβ
α  1  w α 
structured orthogonal and quasi-orthogonal grids. In the next u β  
 w β
=0 (2)

section the features of the developed numerical algorithm for  cos ϕ αβ 1  
constructing orthogonal grids are demonstrated on an
example of so-called 2.5D geometry. Taking into account that

3. ALGORITHM OF CONSTRUCTING
 lβ ∂η 
ORTHOGONAL GRIDS  
 wα   1 − cos ϕ αβ  S ∂α ,
 = 
wβ   − cos ϕ αβ 1   lα ∂η 
Let us consider the algorithm of constructing orthogonal     
 S ∂β 
curvilinear grid in domain G ∈ R 3 having the form of
curvilinear rectangular and lying on a smooth 3D surface.
expression (2) can be rewritten like this
Assume that in domain G ∈ R 3 there is specified curvilinear
coordinate system {α (x, y , z ), β (x , y, z )} and does exist non-
 lβ lα β ηα 
degenerated mapping which transforms curvilinear domain k  u α u  =0 .

G into unit square Ω so, that vertices of curvilinear  S S η β 
rectangular G are transformed into vertices of unit square
and that is equivalent to the following relations: α
∂ 1 lβ w ∂ 1 lα w β
+ =0
∂α k S ∂β k S
l β uα  lβ ∂η 
k = ηβ   (7)
S  wα   1 − cosϕ αβ  S ∂α 
, (3)  = 
l uβ  wβ   − cosϕαβ 1  lα ∂η 
k α = −ηα     
S  S ∂β 

k = k (α , β ) = sin 2 ϕ is non-zero function. Mixed boundary value problem is considered at the domain
boundary:
Differentiation of the first equation of system (3) with respect
to α and the second - to β and summation of the derived r
equations results in: η = 0,(α , β ) ∈ ä2 ,η = 1,(α , β ) ∈ ä4 ,w = 0,(α , β ) ∈ ä1 U ä3 , (8)

where n is the normal to the corresponding boundary.

α
∂ lβ u ∂ lα u β
k + k =0 The developed algorithm can be easy generalized to the 3D
∂α S ∂β S case.
 lβ ∂θ 
  (4)
 uα   1 − cosϕ αβ  S ∂α 
 =  4. VERIFICATION OF THE ALGORITHM
uβ   − cosϕαβ 1  lα ∂θ 
    
 S ∂β  In this section there are presented results of investigation on
influence of size of the initial grid on quality of derived
Mixed boundary value problem is imposed at the boundary: orthogonal grids. The results are derived for the surface
depicted in Fig.1.
r
θ = 0,(α , β ) ∈ ä1,θ = 1,(α , β ) ∈ ä3 ,u = 0,(α , β ) ∈ ä2 U ä4 , (5)
Z

X Y

Transform scalar product (2) to the form:

(w α
)
 1
w β 
cosϕαβ  u α

 u β

=0

0.3

 cosϕαβ 1  
0.2

0.1

0
0 0

or 0.2

0.4 0.4
0.2

y1 (
( 0.6 0.6 x1
0.8 0.8

1 1

 lβ α lα β   θα 
 w
 S w k   = 0 ,
 θ β 
Figure 1. Surface for testing of algorithm
 S

Two sequences of calculations have been conducted. In the

that is equivalent to system of equations: first one the following sequence of refining initial grids
11 × 11,21 × 21,41 × 41,81 × 81,161 × 161,321 × 321 has been
α used in order to derive the orthogonal grid of fixed size. In
1 lβ w
= θβ contrast, the second series is connected with construction of
k S (6) orthogonal grids of different size
β
1 lα w
= −θα 11 × 11,21 × 21,41 × 41,81 × 81,161 × 161,321 × 321 using the
k S initial grid of fixed size.
To estimate quality of derived orthogonal grids, we used in
and after similar differentiation and addition we obtain from
both cases grid function cos ϕ defining in curvilinear
system (6):
coordinates in accordance with expression (1). The results of
calculations are presented in Table 1. Values of this 2D
function cos ϕ are shown in this Table as a fraction where
the numerator shows the minimal, whereas the denominator –
the maximal values of this grid function. It is easy to see from
Table 1 that we have the direct dependence of quality of
derived orthogonal grids on increasing of size of the initial
grid. In other words, the higher size of the initial grid 5. EXAMPLES OF BLOCK-STRUCTURED
provides the better quality of the derived orthogonal grid. GRIDS

Table 1 Below there are presented examples of constructing block-

structured grids on the basis of the presented above
technology.
Base \ Orto 11 × 11 21 × 21 41 × 41
11 × 11 −1.360e −1 −1.640 e −1 −1.871e −1 Figures 2 and 3 demonstrate methods of specifying
− 7.041e −1 1.556e − 2 6.05e − 2 7.009 e − 2 topological data and boundary conditions in the 2D case
0.00
21 × 21 −1.303e −1 −6.195e − 2 −8.669 e − 2 using Mechanical Desktop (version 2.0).
−7.167 e −1 −8.077 e − 4 2.635e − 2 4.641e − 2
1.290 e − 2
41 × 41 −1.279 e −1 −5.417 e − 2 −3.991e − 2
−7.230 e −1 − 9.007 e − 4 −1.696 e − 4 1.736 e − 2
2.250 e − 2
81 × 81 −1.273e −1 −5.295e − 2 − 3.635e − 2
−7.262 e −1 −3.437 e −5 3.694 e − 4 4.596e − 4
2.721e − 2
161 × 161 −1.271e −1 −5.260 e − 2 − 3.507 e − 2
−7.274 e −1 9.105e − 4 8.509 e − 4 1.224 e −3
2.957 e − 2
321 × 321 −1.271e −1 −5.249 e − 2 − 3.489 e − 2
−7.280 e −1 1.272 e − 3 9.371e − 4 8.897 e − 4
3.076e − 2
+ + + +
Base \ Orto 81 × 81 161 × 161 321 × 321
11 × 11 −1.984 e −1 − 2.041e −1 − 2.069 e −1
− 7.041e −1 1.093e −1 1.398e −1 1.590 e −1
0.00
21 × 21 −9.892 e − 2 −1.050e −1 −1.080 e −1
−7.167 e −1 5.740 e − 2 6.696 e − 2 7.792 e − 2
1.290 e − 2
41 × 41 − 4.410 e − 2 −5.027 e − 2 −5.335e − 2
− 7230e −1 2.740 e − 2 3.446 e − 2 4.096e − 2
2.250 e − 2 Figure 2. Implicit specification of topological data
81 × 81 − 3.728e − 2 − 3.833e − 2 − 4.082 e − 2
(attribute with names B_0)
−7.262 e −1 9.890 e −3 1.715e − 2 2.001e − 2
2.721e − 2
161 × 161 − 3.533e − 2 −3.575e − 2 − 3.646e − 2
−7.274 e −1 1.334 e −3 6.087 e −3 9.229 e − 3
2.957 e − 2
321 × 321 −3.474 e − 2 −3.491e − 2 −3.515e − 2
−7.280 e −1 9.682 e − 4 1.081e −3 3.455e − 3
3.076e − 2

Besides, we estimate quality of the derived orthogonal grids

in dependence of recurrent employment of our algorithm (see
Table 2). At the beginning we used as the initial grid the
orthogonal 81 × 81 derived from the initial grid of the same
size, with the following minimal and maximal values of the
grid function −13.334
.533e − 2 (see Table 1).
e −3

Table 2 Figure 3. Explicit definition of topological data

Such a representation is constructed by a user via possibilities

Base \ Orto 81 × 81 of Mechanical Desktop.
81 × 81 − 3.728e − 2
− 7.262e −1 9.890e − 3 Then the automatic stage of constructing an orthogonal grid
2.721e − 2 is started.
81 × 81 −1.266e − 2
− 3.728e − 2 7.011e − 3 The visible part of a display containing the geometry
9.890e − 3
81 × 81 −1.057e − 2
presented in the form of blocks, topological data and
−1.266e − 2 6.341e − 3 boundary conditions is saved as a IGES file.
7.011e − 3
After that this saved data is transformed through the
After processing of this grid via our orthogonalizing code we developed converter into the format of our code for
obtain new orthogonal grid with a more higher quality in constructing orthogonal grids.
sense of the grid cosine. This operation was repeated several
times considering the derived at the previous step orthogonal This code performs generation of the final orthogonal grid
grid as the initial one at the current step. using the developed solver which allows a high-level control
of a user on quality of the grid. Some examples of the derived
grids are presented below.
Figure 4 shows the surface between side blades of a turbine.
Figure 5 demonstrates an aircraft airfoil with a part of the
fuselage.
Stages of constructing a multi-block structure between side
surfaces of blades are depicted in Fig. 6.
Finally, an automobile surface and a display body are shown
in Fig. 7 and 8, respectively.

6. CONCLUSIONS

• A new recursive-iterative method for constructing

orthogonal conformal grid is developed here. It
employs solution of elliptic equations on the initial
grid.
• This approach can be easy used to construct multi-block
structures with the following properties:
• Continuity of the first and second derivatives at Figure 4. A pair of of turbomachinery blades at the
boundaries, multiblock grid of support

• Efficient control on grid spacing.

• Here there is proposed a non-traditional approach to use
CAD systems as the system for geometrical modeling
which does have all attributes of commercial grid
generators:
• Improved GUI,
• Build-in model of the grid topology,
• NURBS — as the geometrical base of data,
• IGES as the format for data exchange with other
CAD systems,
• Powerful toolkit for analysis of constructed grids.
Thus, there is proposed a new method for generation of large-
scale multi-block orthogonal grids for applied and industrial
CFD applications. This method is automatized enough (no
less 90%).
As it was mentioned by Thompson in [12] grid generation
tools must be designed to be applied by design engineers
rather than grid generation specialists.
There is a clear need for interaction CFD with commercial
CAD venders.

Figure 5. A wing with a part of fuselage at the

multiblock grid of support
 a) Upper and lower surfaces

Figure 7. A multiblock grid of support for an

automobile surface

b) Left surfaces

c) Common form Figure 8. A part body for a display

Figure 6. A multiblock structure between lateral

areas of blades
 REFERENCES

[1] M.M. Rai, "A conservative treatment of zonal

boundaries for Euler equation calculations", J. Comput.
Phys., Vol. 62 pp.472-503 (1986)
[2] J.F.Thompson, "A composite grid generation code for
general 3D regions — the Eagle code", AIAA J., Vol. 26
(3) pp.271-272 (1988)
[3] J.L. Steger, F.C. Dougherty, J.A. Benek, "A chimera
grid scheme, Advances in Grid Generation", Ghia K.N
and Ghia U., ASME FED, Vol. 5 (1983).
[4] G. Ryskin, L.G. Leal, "Orthogonal mapping", J.
Comput. Phys., Vol.50 pp.71-100 (1983)
[5] R. Duraiswami, A. Prosperetti, "Orthogonal mapping in
two dimensions", J. Comput. Phys., Vol.98 pp.254-268
(1992)
[6] A. Gaither and et al. The National Grid Project: a system
overview. NASA CP 3291, 1995
[7] G. Farin, Curves and Surfaces for Computer Aided
Geometric Design. Thrid Edition, Academic Press, San
Diego, CA, 1992
[8] M.E. Mortensen, Geometric Modeling. Wiley & Sons,
New York, NY, 1985
[9] L. Piegl, "On NURBS: A Survey". IEEE Computer
Graphics & Applications, pp. 55. January (1991)
[10] F. Yamaguchi, Curves and Surfaces in Computer Aided
Geometric Design. Springer-Verlag, New York, NY,
1988
[11] J. Hoschek, D. Lasser, Fundamentals of Computer
Aided Geometric Design. A K Peters, Ltd., Wellesley,
MA, 1993
[12] J.F. Thompson, B. Hamman, "A survey of grid
generation techniques and systems with emphasis on
recent developments", Surveys on Mathematics for
Industry, Vol.6 pp.289-310 (1997)