Research Article • DOI: 10.

2478/amst-2019-0016 AMST • 2020 • AOP

Advances in Manufacturing Science and Technology

Approximating curve by a single segment of B-Spline

or Bézier curve directly in CAD environment
Mariusz Sobolak*, Piotr Połowniak, Adam Marciniec, Patrycja Ewa Jagiełowicz

Faculty of Mechanical Engineering and Aeronautics, Rzeszow University of Technology, Rzeszów, Poland

Abstract
The paper presents the method of approximating curves with a single segment of the B-Spline and Bézier curves. The
method for determining a single curve segment using the optimization methods in the CATIA environment is shown.
The algorithms of simulated annealing and design of experiment are used for optimization. For the same purpose, a
new original procedure for determining the distance between the given curves using explicit parameters in the CATIA
environment was also used. This approximation of the cyclic curves results in the curve oscillation as shown in the
examples. The results show that the approximation method with Bézier curve using control points as “free” points can
be applied to obtain the best results of approximation.

Keywords
Bézier curve, curve approximation, design of experiment, algorithms of simulated annealing

1. Introduction

Among all curves implemented in CAD environment, the Bézier simulated annealing algorithm, design of experiment, and the
curve is characterized by the best quality of division, hence it “mixed” method. The optimization consists of such matching of
is often used in CAD systems and in engineering applications Bézier curve that the distance between the curves is as small
[1]. This curve is recommended for CAD modeling when the as possible. The optimization uses the explicit procedure for
curve is divided into segments [2]. The sum of the segments is determining the distance of curves in the CATIA environment
geometrically equivalent to the curve before the division. This proposed by the authors because the procedure of curves
is especially important in industrial practice [3]. It happens that distance analysis implemented in the CAD environment can not
geometry is imported from other CAD systems than currently be used due to its implicit nature.
used or is created as a different type of curve. Therefore, A single curve segment in the CAD environment is most often
some curves should be replaced with Bézier curves, which are a geometric representation of the third-degree polynomials.
used for further modeling. The indication of the accuracy of Depending on the limitations, it assumes the mathematical form
the approximation is the distance between the curves, which of the Hermite curve or the Bézier curve (Figure 1) [1–3]. In both
should be as small as possible. It should be mentioned that cases, the four conditions are required to describe the curve
certain curves can be approximated by Bézier curves with segment. For the Hermite curve, there are two points (begin and
some error (e.g., circle arc) [4–7]. The methods of automatic end point) and two vectors tangent at the starting and ending
curve conversion implemented in the CAD environment do points. For the Bézier curve, there are four control points, where
not always give a satisfactory result. This is the result of the P P define the beginning and the end of the segment and the
1 and 2

assumption of maintaining the directions of tangent vectors sections P P and P P determine the directions of tangents at the
1 2 3 4

at the ends of the approximated curve [2]. The user does not beginning and the end of the segment.
have access to the mathematical model of the curve in the In the CATIA environment, in the generative shape design (GSD)
CAD environment, so another solution should be used [8]. module, there is available a spline in form similar to Hermite,
The article presents the methodology for approximating where a starting and ending point, senses and magnitudes of
the curve by the Bézier curve using optimization methods the tangential vectors at these points and additionally sides and
implemented in the CATIA environment, such as the radii of curvature can be specified (Figure 2a).

*
Corresponding author: Mariusz Sobolak
E-mail: msobolak@prz.edu.pl

120
 Sobolak et al.: Method of approximating curves

tangency of the circle arc as a defining element (type: From

curve), and the vector magnitudes (tangent tension) were
determined by parameters respectively at the beginning of the
spline t1 and at the end of t2. The parameter relations t1and
t2 to spline tangents appeared in the model tree (Figure 5).
The task with the curve approximating consists of selecting
such magnitudes of the tangent vectors of the spline, at
the beginning and the end, that the spline approximates
Figure 1. The Hermite curve (a) and the Bézier curve (b). the curve as accurate as possible. The maximum distance
of approximating spline to the curve can be a measure of
Additionally, there is possibility to select the surface or the approximation accuracy. In the CATIA environment, there is
plane, where the spline has to be modeled. In the FreeStyle a tool to measure the distance of curves (distance analysis),
module, the spline is available in the form of the Bézier curve but the measurement result is not an evident parameter
(3D curve) with an active option of creating with the control
points (Figure 2b).

2. Curve Approximation Method

The method of curve approximating by the spline is presented

on the example of the circle arc approximation. In the CATIA
environment, a circular arc with a radius of 100 mm, an angle
of 90°, a center at point of 0.0.0 lying in the x,y plane was
created (Figure 3).
Two parameters of a real-number type named t1 and t2 were
created and initial values t1 =1 and t2 = 1 were assigned.
These are the initially assumed values of the tangent vectors
at the beginning and the end of the spline, respectively.
In the GSD module the spline was created, which the beginning
is one of the circle arc end, and the second one is the other
end of the circle arc (Figure 4). At the start and the end point,
the tangential vector of the spline was determined by selecting Figure 3. Circular arc that will be approximated with the spline.

Figure 2. Single curve segments: (a) spline in the GSD module and (b) Bézier curve in the free-style module.

121
Advances in Manufacturing Science and Technology

(that can be used by the user as a parameter). Therefore, (point on the approximated curve, point on the approximating
it is necessary to prepare a new, specific measurement curve). The relevant pairs of points lying in the same planes
procedure where the maximum curves distance can be are indicated in brackets. The task is to minimize the distance;
parameterized. The procedure requires repeating operations, hence, it is needed to choose the largest value from among
therefore a programming language can be used, but it is not specified parameters l1÷l9. For this purpose, a distance
shown here. Distances measurement should be discretized. parameter lmax is created with assigned formula (Figure 9):
The points, which the distances are measured, are created max(l1/1mm, l 2 /1mm, l 3 /1mm, l 4 /1mm, l 5 /1mm, l 6 /1mm, l 7 /1mm, l 8 /1mm, l 9 / 1mm) *1mm

with the Points and Planes Repetition command (Figure 6). As the max function is in operation only for real-type
In the considered example, the division into 10 segments was parameters, the values of parameters l1 ÷l9 are divided by 1
adopted, which requires to use 9 points. If the programming mm, and finally the maximum value is multiplied by 1 mm to
language is used, the number of points can be much greater. give back the length unit.
At each point dividing the approximated curve, a normal Now, by changing the values of parameters t1 and t2
plane to this curve must be additionally created (the active (magnitudes of tangent vectors), it is possible to change the
option: Create normal planes also). The measurement of the maximum distance between curves. This distance should
distance from the approximating curve is realized in these be minimized. This can be realized using the optimization
planes. command in the Product Engineering Optimizer module
Afterwards, the points that are created as a result of (Figure 10).
intersection of the spline with the created planes should be The minimization has to be selected as the type of optimization,
determined. If the option Create in a new body was selected the lmax variable as the optimized parameter, t1 and t2 as
in the previous command, points and planes were created parameters influencing the value lmax. For parameters t1
in the new geometrical set (Figure 7). That allows selecting and t2, their range can be specified. The simulated annealing
all planes in the Intersection command by choosing the
geometrical set where they are located (Figure 8).
The distances between points in the individual planes should
be described by variables. For this purpose, the distance
parameters (length) are created from l1 to l9, which the
distance measurement formulas are given as: distance

Figure 6. The points and planes creation.

Figure 4. Defining the spline.

Figure 5. A fragment of the model tree in the CATIA environment

with the assumed parameters t1 and t2 and their relations with the Figure 7. A fragment of the model tree with points and planes in a
tangent vectors. new geometrical set.

122
 Sobolak et al.: Method of approximating curves

Figure 8. Determination of points as intersection result of the spline

with a set of planes.

Figure 10. The distance lmax minimization.

similar (the circle arc is symmetrical). In this case, it was

Figure 9. Created parameters and formulas. possible to use one parameter t1 = t2 , however, in the case
of asymmetrical curves, they should be independent.
algorithm was chosen. Because this algorithm may be unstable, A comparison of both curves is shown in Figure 11. The
the criteria for interrupting the calculation are assumed: the distance analysis tool was used. The maximum distance
maximum number of updates 200, the number of consecutive between the curves is 0.081 mm. To visualize the distance,
updates without improvements 50, the maximum calculation comb box analysis with displayed envelope was activated.
time 5 minutes. After determining the optimization parameters, The spline oscillation is seen. This is due to the fact that
it can be run. After performing calculations in the computations the spline is a parametric representation of the curve with
results tab, the optimization results in the form of a list are third-degree polynomials. This shows that the circle cannot
available. The solutions that are closest to the expected one be approximated very precisely with a spline. Unfortunately,
should be chosen from the list. Sometimes, it is not the last this note also applies to other cyclic curves such as involute,
solution on the list; hence, it is convenient to use the Excel extended involute, cycloid, and so on. Therefore, some
spreadsheet. If the save optimization, data option was selected curves in CAD environments are normally represented in a
before optimization, the program asked for a file name where different way than with parametric cubic curves.
the results are saved in the Excel format. In this file, it is possible In a similar way, the curve segment can be approximated by the
to sort by lmax and select the smallest value lmax, read Bézier curve. In this case, two parameters d and d as the length
1 2

parameter values for t1 and t2. This is the first approximation type should be created. Then at the beginning and the end of the
but not the most appropriate that can be obtained. At this stage, approximated curve, the tangent segment with the appropriate
it is possible to limit the change range of t1 and t2 and then lengths d and d should be modeled (Figure 12). The control
1 2

perform the optimization again. points P and P are the endpoints of these segments.
2 3

The following results were obtained for the approximated Having the points P P P P it is possible to create a Bézier
1, 2, 3, 4

arc: t1= 1.168367213, t2= 1.168367209. The values are very curve segment (3D Curve command in the FreeStyle module).

123
Advances in Manufacturing Science and Technology

The type of creation by the control points should be chosen the end of the curve. It is assumed that the coordinates of
and appropriate points should be selected (Figure 13). points P and P are the variables.
2 3

The curve approximation is realized in a similar way to that In the CATIA environment, the variables x2, y2, x3, y3 are
shown earlier for the B-Spline curve. The variable parameters created and they are corresponding to the coordinates of
in this case are the lengths of the tangent segments d and d .
1 2
the points P and P (Figure 15). The values of variables x2,
2 3

Figure 14 shows the comparison of a circle fragment with an y2, x3, y3 are initially assumed. Control points P and P are
2 3

approximating Bézier curve. In this case, the oscillation of the created on the plane where the section of the approximated
approximation curve also occurred. circle was located (Figure 16).
The Design of Experiment (DOE) method can be used to Having the points P P P P the Bézier curve segment is
1, 2, 3, 4

determine the curve approximation by a spline. This method created similarly as in the previous example (3D Curve
is beneficial if there are more variables affecting the result. command in the FreeStyle module). The type of creation
The method will be shown on the example of approximation by the control points should be chosen and the appropriate
of a circle segment by the Bézier curve. This time the points points should be selected.
are not fixed with tangent straight lines at the beginning and After the Bézier curve was created, the matching procedure

Figure 13. The Bézier curve creating; 1—approximated circular arc,

2—Bézier curve, and 3—Bézier polygon.

Figure 11. Analysis of the curves distance.

Figure 14. Comparison of the circle arc with the approximating

Bézier curve.

Figure 12. Tangent segment to the curve.

124
 Sobolak et al.: Method of approximating curves

is identical to the B-Spline curve up to the optimization stage. Product Engineering Optimizer module) should be used.
A set of planes which intersect curves is created. Variables The input parameters are the values of coordinates x2, y2,
corresponding to the distance of points in individual planes x3, y3 (Figure 17). For each coordinate, the minimum value
and a formula determining the maximum distance of lmax (Inf. Range), the maximum (Sup. Range), and the number
from among are defined. of levels (Nb of Levels) are declared. The number of levels
To determine the parameters values x2, y2, x3, y3, the design defines the discretization step. The lmax distance is the output
method using the DOE experiment (Design of Experiment, parameter. After starting the experiment (Run DOE), a table
of all possible combinations of dimensions is created. The
obtained table in Excel format can be sorted according to the
lmax column. Then the values of the parameters x2, y2, x3,
y3 for the smallest lmax can be read. Because many solutions
are created (in the shown example for 21 levels for each
variable x2, y2, x3, y3 194481 combinations are generated), it
Figure 15. Variables corresponding to coordinates of control points is strongly advised to assume not to many levels in this step.
P2 and P3, After determining the first approximation, the initial data can
be modified by narrowing the search.
The results x2 = 55.29, y2 = 99.92, x3 = 99.92, y3 = 55.29
were obtained. This means that the points P and P do not lie
2 3

on the lines that are tangent to the circle arc at the beginning
and the end. The Bézier curve in this case approximates the
circle arc with a smaller error, but the number of oscillations
increased (Figure 18).

3. Examples of Curve Approximation

Figure 16. Creating a control point. As examples of approximation the involute and fillet curve
were presented. These curves can be found in the flank
profile of gears (Figure 19).
Involute and fillet curve were created by parametric equations
directly in the CAD environment [9, 10]. Modeling with
parametric equations in CATIA is not the subject of this article,
so it is not presented.

3.1. An example of involute approximation (B-Spline)

The involute was created for the data: module m = 5 mm,
α n

number of teeth z = 20, pressure angle n = 20°, spur gear.

Figure 17. Design of Experiment (DOE). Figure 18. Comparison of the circle arc with the approximating
Bézier curve.

125
Advances in Manufacturing Science and Technology

The involute was approximated by a single spline segment. The comparison shows that the maximum distance is ca.
The values of tangent vectors were determined: at the 0.005 mm, so in the industrial practice it is not satisfactory.
beginning of involute at the basic diameter t1 = 0.386, on Oscillations of the spline are visible. In order to increase
the addendum diameter: t2 = 1.568 (Figure 20). The distance accuracy, the involute should be approximated by several
analysis of the involute and approximated curve was executed segments of the spline.
(Figure 21).
3.2. An example of fillet curve approximation (Bézier
curve)
The fillet curve was modeled in the CATIA environment using
the parametric equation of the extended involute in the plane x,y.
To approximate fillet curve with the Bézier curve, the variables
x2, y2, x3, y3 corresponding to the coordinates of the control
points P and P were created and the initial values were
2 3

assigned (Figure 22). The starting and ending point of the

previously created curve is the control point P and P of the
1 4

Bézier curve, respectively.

Then the points with these coordinates were created (Figure
23).
Figure 19. Gear flank profile; 1—involute and 2—fillet curve. Having the points P P P P the Bézier curve segment (3D
1, 2, 3, 4

Curve command in the FreeStyle module) was created. The

type of creation by the control points was chosen and the
corresponding points were selected (Figure 24).
Figure 20. Values of tangent vectors in the model tree. On the approximated fillet curve, a discrete set of nine points
was created with planes perpendicular to the curve. The
Bézier curve was intersected with these planes and a set of
points was obtained (Figure 25).

Figure 21. The distance analysis of the involute and approximated

curve.

Figure 22. Coordinates of points P2 and P3 in the model tree.

Figure 23. Creating points: (a) P2 and (b) P3.

126
 Sobolak et al.: Method of approximating curves

Variables l1÷ l9 were created and the formulas for the

distance points measuring were assigned. Next the formula
determining the maximum distance was defined (lmax).
Pre-values of the parameters x2, y2, x3, y3 were determined
using the simulated annealing algorithm (Figure 26).
Figure 27 shows preliminary results of the optimization.
Next step of the approximation was made using the DOE
method, assuming appropriate values ranges of variables x2,
y2, x3, y3 near the values determined in the previous stage
(Figure 28).
Figure 29 shows a fragment of an Excel spreadsheet with
sorted results. It is seen that for the assumed number of
levels the same maximum distance of curves was obtained
for the first five sets of parameters. Values of parameters were
changed every 0.01 mm. Figure 30 shows the comparison of
Figure 24. Modeling of the Bézier curve; 1—approximated fillet
curve and 2—Bézier curve.

Figure 27. Preliminary results of the optimization: (a) parameters in

the model tree and (b) curves: 1—fillet and 2—Bézier.

Figure 25. The points preparation for the distance measuring; 1—

approximated fillet curve and 2—Bézier curve.

Figure 26. A fragment of the optimization window. Figure 28. The preparation of the experiment.

127
Advances in Manufacturing Science and Technology

the profile modeling in a CAD environment should be greater

than obtained in machining process. The accuracy in CAD
system should be at least one order of magnitude greater.
Thus, this approximation can be used in CAD demonstrative
models. The presented method of involute curves modeling
by B-Spline or Bézier curves in the CAD environment causes
the curve’s oscillation. This is due to the fact that the spline
function is described by the third-degree polynomial. To
increase accuracy, the approximated curve should be divided
Figure 29. A fragment of an Excel spreadsheet with sorted results.
into segments. It should be mentioned that oscillation will
always occur, but the amplitude will be adequately lower.
In workshop practice, the fillet curve as well as the involute
should be divided into several segments. In the example
under consideration, accuracy of 0.001 mm was achieved by
dividing the curve for involutes into four segments and for the
fillet curve into nine segments, respectively. The continuity of
curve contact must be ensured between these segments.

References

[1] G. FARIN: Curves and surfaces for computer aided geometric

design: a practical guide. 3rd ed. Academic Press, Boston, MA
1993.
[2] H. PRAUTZSCH, W. BOEHM, M. PALUSZNY: Bézier and B-spline
techniques. Springer-Verlag, Heidelberg 2002.
[3] H.N. FITTERA, A.B. PANDEY, D.D. PATEL, J.M. MISTRY: A re-
view on approaches for handling Bézier curves in CAD for man-
Figure 30. Comparison of the fillet curve and the approximating
Bézier curve. ufacturing. Procedia Eng., 97(2014), 1155–1166.
[4] F. HIGUCHI, S. GOFUKU, T. MAEKAWA, H. MUKUNDAN, N.M.
PATRIKALAKIS: Approximation of involute curves for CAD-sys-
tem processing. Eng. Comput., 23(2007), 207–214.
[5] Y.J. AHN, C. HOFFMANN, Y.S. KIM: Curvature continuous offset
the fillet curve and approximating Bézier curve. approximation based on circle approximation using quadratic
Bézier biarcs. Comput.-Aided Des., 43(2011), 1011–1017.
[6] S.H. KIM, Y.J. AHN: An approximation of circular arcs by quartic
4. Conclusions
Bézier curves. Comput.-Aided Des., 39(2007), 490–493.
[7] J. SÁNCHEZ-REYES: Bézier representation of epitrochoids and
The presented examples show that to approximate the curve, hypotrochoids. Comput.-Aided Des., 31(1999), 747–750.
the approximating method with Bézier curve using control [8] M. BATSCH, T. MARKOWSKI, S. LEGUTKO, G.M. KROLCZYK:
points P and P as “free” points can be applied. That means
2 3
Measurement and mathematical model of convexo-concave
that it should be assumed that the points are not located on Novikov gear mesh. Measurement, 125(2018), 516–525.
the tangents at the beginning and the end of the approximate [9] S. BABU, K.M. ABUBACKER: Development of involute profiled
curve. spur gear model with excel spreadsheet, solidworks and CAD
The approximation of involute or fillet curve with the single technique. Int. J. Mech. Eng., 5(2018), 5–11.
segment curve should not be used if very high accuracy is [10] G.G. REY, G.B. GORDILLO: Parametric geometric modeling of a
needed. Industrial practice indicates that the accuracy of spur gear using solidworks. Gear Solutions., 15(2017), 30–36.

128