PED-Vol.

64, Manufacturing Sciem:e and Engineering

ASME 1993
CNC INTERPOLATORS:
ALGORITHMS AND ANALYSIS
Y. Koren and C. C. Lo
Department of Mechanical Engineering
University of Michigan
Ann Arbor, Michigan
M. Shpitalni
Laboratory for Computer Graphics and CAD
Faculty of Mechanical Engineering
Technion
Haifa, Israel
ABSTRACT
CAD systems today interpolate general curves by dividing
each curve into many straight-line segments which are
downloaded to the CNC. Determining the number of lines to
be transferred from the CAD to the CNC poses a conflict
between the desired precision of the part and the feedrate
fidelity. The current method results in severe variations in the
feedrate, leading, in turn, to variations in the surface
smoothness and a substantial increase in machining time.
These problems are caused by the acceleration/deceleration at
the ends of each segment. Moreover, the problems are
inherent in the CNC interpolator, as is thoroughly discussed in
this paper. These problems can be solved by the development
of curve interpolation algorithms for CNC. In this paper, a
real-time interpolation algorithm for curves presented in their
parametric forms is proposed and compared with the existing
CAD interpolators. Analysis shows that with this new
interpolator, a constant feed is maintained along the cut and
the machining time is as expected. In addition, the amount of
geometric information transferred from the CAD system to the
CNC is reduced by orders of magnitude. Moreover, the contour
errors caused by the new interpolator are much smaller than
those caused by conventional CAD interpolators.
Keywords: CNC, CAD, Interpolators, Curves.
INTRODUCTION
Most computer aided design (CAD) systems provide the
designer with tools for defining two- and three-dimensional
curves and surfaces. By contrast, conventional computerized
numerically controlled (CNC) machines generally support
only the functions of straight line and circular interpolations
[e.g., Koren, 1983]. This gap between the well developed
theory of representing curves and surfaces in CAD systems and
the limited capabilities offered by the numerical control
interpolators imposes severe difficulties on rapid and accurate
83
machining of surfaces and curves, indicating an obvious need
for a general curve interpolator for CNC machines.
In this paper, various possibilities for implementing curve
interpolators on CNC machines are analyzed, and a new
scheme is suggested. First, the current method used for
machining of curves is discussed, and the difficulties
associated with it are explored. Then, the need for a real-time
interpolator for general curves is presented. Finally, a new
scheme is proposed for real-time interpolators for general
curves presented in parametric form.
Figure 1 shows the current method for machining a part.
CAD systems are used to define the geometry of the part, and
the CNC must then drive the machine tool to machine this
geometry. The part geometry is transferred to the CNC by
means of a part program consisting mainly of motion
commands. These motion commands must be translated in real
time by the CNC interpolator into a special form before being
sent to the control loops for execution (i.e., driving the
machine tool).
The motion commands must fit the interpolator capabilities
so that the interpolator can translate them in real time. That
is, a typical CNC can process only straight line and circular arc
motion commands. Thus, in order to drive the machine tool
along a curve, the curve must be broken into a set of line
segments which approximates the curve to a desired accuracy
(i.e., tolerance). The segmentation of the curve into line
segments is performed in the CAD system by off-line
calculations.
In summary, the current procedure for machining curves is
as follows:
(a) The designer defines a shape by a set of geometrical
entities. An entity may be a line, an arc, or a curve.
(b) If the entity is a line or a circular arc, it is transferred
directly as such to the CNC by the part program.
(c) If the entity to be machined is a curve, it is first
divided into line segments in the CAD system. The
resulting line segments are then transferred to the part
CAD CNC
FIG. 1: CURRENT METHOD FOR MACHINING OF
CURVES.
program. Their original curve is not even known to
the CNC.
OPTIMAL SEGMENTATION OF CURVES
Inherent in the current method for machining of curves is a
conflict regarding the number of line segments into which a
curve should be divided by the CAD system. On the one hand,
the number of segments should be maximized for two reasons:
( l) To better approximate the curve and reduce the contour
error, and
(2) To minimize the effect of segmentation which causes
discontinuities in the first derivatives along the path.
These discontinuities, in turn, lead to deterioration of
the smoothness of curves and surfaces and necessitate
additional treatment (e.g., polishing).
The difficulty in this maximization, however, is the huge
number (millions) of segments required for machining of
surfaces. Therefore, on the other hand, the number of
segments should be kept to a minimum for the following
reasons:
( 1) Each segment is treated by the CNC as an individual
line. If many short segments are transferred, the tool
may never reach the desired feedrate due to the
automatic acceleration and deceleration applied at the
beginning and end of each segment (feature of the
control). The result is that the feedrate along the curve
is not constant which, in turn, deteriorates the surface
finish (in milling) or part dimension (in laser cutting).
In addition, the machining time is increased because
the mean feedrate is less than the desired feedrate.
(2) A similar effect is caused by interpolators of the
Reference-Word type (see below). With these
interpolators, a large number of segments increases
the feedrate variations, reduces the average feedrate,
and increases the machining time.
(3) The CNC's memory is very small (and very expensive)
compared to the number of segments to be stored for a
part with complex surfaces.
84
(4) The communication load between the CAD and the
CNC, that may cause consequent errors, should be
reduced.
In practice, the approach of minimizing the number of
segments has been adopted by industry due to memory and
communication constraints. The minimum number of
segments is dictated by the allowed tolerance and depends on
the curvature and length of the curve. However, the resulting
discontinuities along the curve necessitate additional
treatment. Note that even with the minimization approach a
huge number of segments (may reach millions) is still needed
in order to maintain reasonable tolerance (e.g., 10 ~ m ) . Thus,
the current method is not adequate for machining of curves and
surfaces.
Consequently, a new CAD/CNC procedure should be
developed which will enable high and constant feedrate along
the curve. This concept is shown in Figure 2. In the proposed
approach the CAD segmentation algorithm has been removed
from the CAD system. Instead, a new interpolator which can
interpolate general curves in real time has been added to the
CNC. This enables the CAD system to transfer to the CNC
only information about the curve; the CNC then interpolates it
using the real-time curve interpolator.
CAD CNC
Machine Tools
FIG. 2: SUGGESTED METHOD FOR MACHINING OF
CURVES.
OFF-LINE CURVE INTERPOLATORS
One method for describing curves which treats the x and y
axes symmetrically is the parametric form
x = x(u)
y=y(u)
(1)
where u is an arbitrary parameter usually 0 ~ u ~ 1 . Most
CAD/CAM systems use parametric forms to represent curves.
The parametric form is very convenient for controlling multi-
axis machine tools, where each axis is individually driven.
Furthermore, the parametric form enables direct calculation of
x and y as functions of u, and the extension from 2-D to 3-D is
straightforward.
The selection of u is critical to the proper operation of an
interpolator that calculates x(u) and y(u) in real time. Three
conditions must be satisfied in the selection of u:
1. The parameter u must be independent of the
geometrical parameters of the curve.
2. The representation of x(u) and y(u) must be explicit for
u to allow direct, fast calculation of x(u) and y(u).
3. Successive u's must divide the curve uniformly, i.e.,
into equal segments .6.s. Since each segment is
machined during one sampling period T, this uniform
subdivision of the curve guarantees cutting at a
constant feedrate (i.e., velocity) V, where V=.6.srf.
While the first two requirements are usually satisfied, the
third condition is satisfied only if u is a linear function of s,
where s is the curve arc length, and then the position is
represented as x(s) and y(s). This condition is easily satisfied
for straight lines and circles, as shown below:
For a straight line given by its two endpoints (xb YJ) and
(xz, yz):
For a circle with center at (XJ, YJ) and radius of a:
x(s) = XJ +a cos (sf a)
y(s) = Yl +a sin (sf a)
n;a
for 0 :>; s :>;-. However, satisfying the third condition in the
2
general case is more complex.
In order to determine x(s) and y(s) in the general case, the
intrinsic equations of the curve must be solved [Faux & Pratt,
1979]. The intrinsic equation of a curve defines the
relationship between its arc length s and the angle <p between
the tangent at point P and the X-axis. The curvature at point P
is obtained from the intrinsic equation by the formula:
d<p
K=-
ds
where JC= lip and p is the radius of curvature at point P.
(2)
Alternatively, the curve may be described parametrically in
terms of the arc length by the equations x=x(s) and y=y(s).
The functions x(s) and y(s) are then related to <p by the
equations
dx
-= cos<p
ds
dy .
-=sm<p
ds
(3)
85
If we differentiate these equations with respect to s, and
substitute K for d<p , dx for cos <p, and dy for sin <p, we
ds ds ds
obtain the simultaneous differential equations:
(4)
These two second-order equations represent an interpolation
algorithm and can in principle be solved to determine x(s) and
y(s) for any given curvature function K(s). Appropriate
numerical procedures have been described by Adams [1975)].
For a general curve, however, these are time consuming
calculations that fit only off-line applications. Other
solutions should be sought.
The common approach in off-line interpolation is to divide
the curve into equal increments .6.u, to determine the
corresponding [x(uk), y(uk)l and to drive the tool along a
straight line between [x(uk-1), y(uk_I)] and [x(uk), y(uk)].
Obviously, a larger number of segments results in a better
tracking of the curve. However, in the previous section, we
claimed that with reference-word interpolators, dividing a
curve into a large number of segments reduces the average
feedrate, increases the feedrate variations, and increases the
machining time.
Although this observation is extremely important, it was
never discussed in the literature. Consequently, CAD/CNC
programmers are not aware of it, thus causing many problems
in production of parts with complex surfaces. To facilitate the
explanation of this point, let us use an example of a CNC
system with a resolution of BLU=0.01 mm and interpolation
sampling period of T=0.01 sec. A straight line (which
simulates here a general curve) has to be machined in one axis
at a feedrate of V=4 mm/sec (i.e., 4 BLU/sampling-period).
When the CNC linear interpolator receives the input, in
terms of the length of the line l and the feedrate V =4, it starts
to issue control commands of "move 4 BLUs" at every
sampling period T. The accumulated velocity commands is the
position reference. The last command in the line might be
shorter, to complete the production of exactly l units. For a
line of 22 BLUs, for example, the interpolator produces the
position reference shown in Fig. 3a. The last incremental
position is smaller. Figure 3b shows the corresponding
velocity command.
During the first five sampling periods, the velocity
command is V=4BLUs/0.01 sec=4 mm/sec, but it is only 2
mm/sec in the last period. The smaller velocity in the last
period is typical.
Now let us examine the effect of curve segmentation for a
straight line of l = 55 BLUs. The selection of a short line
enables us to draw the results. We analyze four cases:
(a) The line is produced as one segment of 55 BLUs.
(b) The line is divided into three segments, two of 18
BLUs and one of 19 BLUs (2 x 18 +19 =55).
(c) The line is divided into six segments, five of 9 BLUs
and one of 10 BLUs (5 x 9 + 10 =55).
(d) The line is divided into 11 segments, each of 5 BLUs.
;r--
24
20
16
12
8
4,P..-..1
0
0
2
(a)
I I
2 (b) 3
4
I
4
~
5 6
FIG. 3: (A) POSITION COMMAND ISSUED BY LINEAR
INTERPOLATOR, AND (B) THE CONSEQUENT VELOCITY
COMMAND.
The corresponding velocity commands are depicted in Fig.
4. If the length of a segment is not integer multiples of 4, its
last incremental position is smaller than 4, which means that
the velocity command is reduced during the last sampling
period. Figure 4 shows that a division of a line into a larger
number of segments increases the feedrate variations and
reduces the average feedrate, which, in turn, increases the
machining time. The machining time of the 55 BLU, as shown
in Fig. 4, is 14T if the whole line is treated as a single segment
(Case a). The machining time increases to 1ST if the line is
divided into 3 segments (Case b); to 18T if the line is divided
into 6 segments (c); and, to 22T if 11 segments are used. The
situation is even more complex when several axes are involved
in the motion and especially for general curves.
To further clarify the point, assume that the curve
{
x=ll.9u
3
-29.8u
2
+32.9u+5.0 OSuSl
y = 47.6u
3
-41.7u
2
+ 16.55u+2.5
is stored in a CAD system and has to be machined on a CNC at
a velocity of V = 1.2 m/min = 20 mm/sec. Also assume
that the sampling time of the CNC is 0.01 sec. In the CAD
system, the parameter u is incremented uniformly in equal
increments ~ u -- this is the segmentation of the curve. The
more segments in the curve, the smaller the contour error, but
larger deviations from the average feedrate are expected. The
results are shown in Fig. 5.
As a consequence, we see that very fine segmentation causes
severe problems in maintaining a uniform and desired V. This
and the other reasons discussed in Sec. 2 above show the
importance of a reliable real-time CNC interpolator.
86
V[mn/sec]
,.:1
'
~
.. L
14
~ ) : 1
'
'
l
u u
~
I5
.. ,
~
(c)
0 I ~ .._ t
4
(d)
~ 8J214161820iz.,.
FIG. 4: VELOCITY COMMANDS ISSUED BY THE CNC
INTERPOLATOR VERSUS
~
QJ
]
_.
~
= ..
QJ
....
~
>
NUMBER OF SAMPLING PERIODS.
20
19
18
17
16
15
0 20 40 60 80
Segments
1000
100
e
~
10 ..
0
..
..
QJ
100
FIG. 5: THE EFFECT OF SEGMENTATION ON THE
CONTOUR ERROR AND AVERAGE FEEDRATE.
TYPES OF CNC INTERPOLATORS
Two main kinds of interpolators exist in CNC systems.
( 1) Reference-Pulse Interpolators which continuously
calculate the increment to the next reference point at
the resolution level (pulse mode), e.g., the DDA
(Digital Differential Analyzer) method [Koren, 1976;
Koren & Masory, 1981, Papaioannou, 1979].
(2) Reference-Word Interpolators which calculate the next
reference point based on constant sampling time T
[e.g., Masory & Koren, 1982].
The selected interpolation method depends on the control
method. The proposed interpolator in this paper is of the more
common method-- the reference-word interpolator.
Several researchers have developed real-time interpolators
for curve generation. Koren [1976] proposed an interpolator
for standard parabolic curves; Sata et. al. [1981] developed a
real-time interpolator for cubic Bezier curves which can, in
fact, interpolate any cubic function. This interpolator can of
course interpolate general parabolic curves by using only a
2nd order Bezier curve. Makino [1988, 1991] developed
special interpolators for high-speed machines. Stadelmann
[1989] proposed a high-order interpolator for complex spatial
geometry that requires a given velocity profile. In this
interpolator the transition between two successive segments is
continuous. However, this interpolator does not guarantee
that all interpolated points lie on the curve. Recently Chou
and Yang [1991, 1992] proposed a new interpolator for curves
represented in their parametric forms. Their interpolator
indeed interpolates the curves correctly but it does not fit real-
time applications. They solved the equation that relates the
time t to parameter u, namely t=f(u), while for real time
realization the parameter u should be related to time, namely
u=g(t). Solving u=g(t) requires a mathematical development
that is given in this paper.
Proposed interpolation schemes may be differentiated by
the method used to represent the curve. Two representation
methods exist with corresponding interpolation schemes. The
first scheme deals with the implicitly defined curve f(x,y)=O
[Lo, 1992]; and, the second scheme deals with the parametric
curves r(u)=x(u)l+y(u)]. (The extension to 3-D is also
discussed in this paper.)
Note that in the particular case where y can be expressed
explicitly as a function of x, namely, y=f(x), the situation is
equivalent to the case of parametric representation. In these
cases, the function can be expressed as:
x(u)=u
y(u)=f(u)
In this paper only interpolators for curves presented in their
parametric form are discussed. The straightforward approach
for real-time interpolators might be the one utilized in CAD
systems: Increment the parameter u uniformly, namely in
equal small increments L1u, and calculate the corresponding
Xk+l andyk+l at each sampling periodT.
The approach, however, has two main drawbacks:
(1) The length of the steps where

is not equal, but

the tool traverses them in equal time intervals T.
Consequently, the feedrate along the curve is not
constant.
(2) The optimal size of the increment is not known. If
is too small, the resultant are too small as
well, and the system slows down. If is too big, the
resultant are too big and the position accuracy is
not maintained.
As a consequence, a real-time interpolator cannot be based
on a uniform segmentation of the curve according to u. CAD
systems, however, are based on equal segmentation of u to
represent a curve by lines. But, in CAD systems, the lines are
relatively longer and a velocity command is issued for each
87
line, whereas in real-time interpolators the velocity is
obtained automatically by
This situation motivated the development of a new real-time
interpolation method for general curves expressed in their
parametric form. This method requires only small computation
time and therefore fits real-time applications.
The task of the real-time interpolator for a general curve can
now pe specified.
Given: (1) a curve i'(u),
(2) the desiredfeedrate V[mm/sec] along the
curve,
( 3) the sampling time interval T[ sec], and
( 4) the current reference point on the curve
Rk(XkoYk),
find the next reference point to which the tool should be
driven in T seconds while maintaining a constant feedrate
v.
This implies that the next reference point Rk+ 1 (Xk+ loYk+ 1)
must lie on the desired curve and must obey 11Rk+1-Rkii=VT.
In the following section, a new approach for real-time
reference-word interpolators is proposed. Such an interpolator
fits the case of parametric representation and is capable of
dealing with general curves. We assume that the given curve is
the curve along which the center of the tool must be driven.
The new real-time interpolator is compared with the
conventional interpolator.
INTERPOLATOR BASED ON UNIFORM-LENGTH
SEGMENTS
The key idea for real-time interpolators of curves
represented in parametric form is that the segmentation should
be based on curve segments of equal length rather than on equal
Accordingly, the proposed method determines
successive values of u such that the curve segments
(machined at each sampling period T) are constants, which, in
turn, guarantees a constant feedrate (i.e., velocity).
The feedrate V(u) along the curve is defined by
V(u)= ds
dt du dt
or
du V
-=--
dt dsfdu
where
and
, dx
X=-
du
' dy
y =-
du
, dz
z =-
du
(5)
(6)
(7)
\"
Substituting Eq. (7) into (6) yields
(8)
A solution of Eq. (8) that is performed at short computation
time is the heart of a real-time interpolator for curves given in
parametric forms [Lo, 1992; Huang and Yang, 1992; Huang,
1992]. The solution of Eq. (8) for a constant V gives the
required u(t). However, because the solution of Eq. (8) is
difficult in the general case, we may use a recursive solution
based on Taylor's expansion around t=kT:
(9)
. du .. d
2
u
where u denotes - and u denotes -
2
-.
dt dt
If T is very small and the curve does not have small radii of
curvature, even a first-order approximation is adequate:
(10)
This equation usually holds when
(11)
Substituting Eq. (8) in Eq. (10) yields the equation of the
proposed interpolator:
(12)
In cases where Eq. (11) is not satisfied, we may use two
terms of Taylor's expansion
(13)
Equation (13) holds for small T, defined by
(14)
The expression for ii may be derived from Eq. (8)
88
(15)
where ds/du is given in Eq. (7), and differentiating Eq. (7)
yields
(16)
where
" d2y
y =--
2
and
du
The first derivative of u is given in Eq. (8)
(17)
and the second derivative is given by substituting Eqs. (7) and
(16) into (15).
-v
2
(x' x"+y'y"+z' z")
.. kk kk kk
uk= 2
{xf+yk2+zf)
(18)
Substituting Eqs. (17) and (18) into (13) yields [Lo, 1992;
Huang, 1992]
VT
uk+l = uk+---,.=====
x'2 + y'2 + z'2
k k k
(VT)
2
(xkxk+ YkYk+ zkzk)
2( xk2 + yk2 + zln2
(19)
The explicit expressions of the first and second derivatives
of x and y with respect to u are calculated and substituted into
Eq. (19). The new proposed interpolator is based on
calculating the new value uk+l from Eq. (19), and using it in
the next iteration to calculate x(uk+I), y(uk+I) and z(uk+I).
In order to check the condition ( 11) at which the
approximation (10) is valid, Eqs. (17) and (18) are substituted
into (11), yielding
(20)
If condition (20) is not satisfied, and Eq. (12) is used in the
interpolation process, a feedrate error exists. Note, however,
that the interpolated points are always on the curve. An
interpolator based on Eq. (19) yields negligible feedrate errors
for all practical cases.
It is important to show how this interpolator behaves if
applied to a circular arc in the X-Y plane.
x(u) =a cos u
y(u) =a sin u
foro::; u : : ; ~ .
2
Differentiating these equations to obtain
dx .
-=-asmu
du
dy =acosu
du
and substituting them in Eq. (11) results in:
namely, the changes in Uk are independent of k, they are
constant and equal to VT/a (where a is the radius of the arc).
This means that a perfect circle is produced.
To evaluate the proposed interpolator, 2-D and 3-D cubic
polynomial curves were simulated and compared with the
conventional method, which approximates the curve by line
segments for which .6.u=constant. The evaluation is given
below.
Example 1
The 2-D cubic curve
{
x = -140u
3
+90u
2
+ 90u
y = -90u
2
+ 90u
was simulated with V = 1.5 m/min, T = 0.01 sec.
The contour shape of this 2-D curve is shown in Fig. 6. The
proposed interpolator was simulated and results in very small
contour errors compared to conventional CAD interpolators,
as shown in Fig. 7 and in Table 1. The variations in the
velocity commands that sent to the control loops for the
proposed interpolator are depicted in Fig. 8a for Eq. 12 and
Fig. 8b for Eq. 19. Both are negligible for most applications.
Adding the second derivative of u to the approximation results
in slightly smaller contour errors and smaller velocity
variations. The calculation time, however, is increased, which
increases the lower bound on T. Therefore, for this example,
the approximation based on Eq. 12 is adequate.
89
Y [mm]
:3.0l
&7
18.4
\
I
I
X [mrn]
ll.84 23.68 35.52 47.36 59.21
FIG. 6: A 2-D CUBIC POLYNOMIAL CURVE.
The segmentation in the conventional CAD interpolator,
based on small equal .6.u (e.g., .6.u=0.02 for 50 segments),
results in correspondingly short .6.s segments (e.g., average of
1.9 mm). Because of the method by which linear interpolators
operate, the short segments cause large feedrate errors, as
shown in Fig. 9. As expected and as shown in Table 1,
increasing the number of segments results in smaller contour
errors, but increases the average deviation from the
programmed feedrate (V=25 mm/sec in this example). This
simulation does not include the acceleration/deceleration effect
that even deteriorates the situation in off-line CAD
segmentation.
In other CAD systems the approximation of the parametric
curves by a set of line segments is based on assigning
constant tolerance, rather than constant u [Loney & Ozsoy,
1987]. The curve in these cases is approximated by a smaller
number of lines, but the fundamental problem inherent to off-
line interpolation will exist. However, we made the
comparison to the method utilizing constant .6.u since it can be
a candidate for real-time interpolators, whereas the constant
tolerance method has high computation load and cannot be
implemented in real time.
-
TABLE 1: INTERPOLATION OF A 2D CUBIC SPLINE.
Proposed Interpolator:
Equation used:
EmaxJJlffi 1
OVavN
oVmaxN
Conventional
# of line segments
Emax[Jllll 1
OVavN
oVmaxN
23.00, Position
I Error [!J.m)




tl.5o-j


Eq. 12
2.6
0.2%
1.7%
Interpolator:
30
61
3.8%
50
22.3
6.6%
Eq. 19
2.5
0.01%
0.07%
70
11.2
9.2%
Arbitrary, and may reach up to 98%
/\
/ \
I '
I \CAD
1 \ incerpolation
/ '\
Proposed
/ ............ Incerpolation
....... :\. __ ...... -' ..... ___ u
0.00 L __
0.00
7 I I I
O.lO 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 LOO
FIG. 7: CONTOUR ERRORS CAUSED BY THE
PROPOSED AND CONVENTIONAL INTERPOLATORS.
90
Verror (%]
1\
Proposed
Interpolator
1\

I
I
I
I
!
\ '

t(sec]
2.00
0.00 0.38 0.76 l.l3 l.Sl l.89 2.27 2.65 3.02 3.40 3.7:
Verror [%]
0.07
0.06
0.05
0.04
0.03
i\
i
I
II
II
I.
Proposed i
J
0.02
O.Dl Interpolator /
, I
:I
'!\ I !
\I

0.0
-0.01
0.02
t[sec]
0.04
0.00 0.38 0.76 1.13 LSI 1.89 2.27 2.65 3.02 3.40
FIG. 8: VARIATIONS IN THE VELOCITY COMMANDS
CAUSED BY THE PROPOSED INTERPOLATOR.
{A) EO. 12; {B) EO. 19.
Verror [%]
t[sec]
-100.00 L....__...,.--,---,---,-__...,.--.-.:...._-,---,-__...,.----,
0.00 0.81 1.61 2.42 3.22 4.03
FIG. 9: VARIATIONS IN THE VELOCITY COMMANDS TO
THE CONTROL LOOPS CAUSED BY CAD
CONVENTIONAL INTERPOLATOR.
Example 2
The 3D cubic curve
l
x=ll.9u
3
-29.8u
2
+32.9u+5.0
3 2
y= 47.6u -41.7u +16.55u+2.5
z = 11.9u
3
-5.95u
2
+ 9.05u+5.0
was simulated with V= 1.2 m/min, T=O.Ol sec.
The simulations results for a 3D cubic spline are similar to
those for a 2D cubic spline. The results are summarized in
Table 2.
Based on the above examples, as well as on many others
that we tried, we may conclude the following:
( 1) The conventional approach needs a very large number
of line segments to guarantee a small contour error.
Note that in the first (2-D) example the length of the
tool path is about 94.5 mm, but 70 segments are
needed to approximate it with a tolerance of 11 11m.
This subdivision caused a decrease of about 9% in the
average feedrate.
(2) The feedrate error caused by the proposed interpolator
depends on the number of terms used to approximate
the next spatial parameter Uk +I If the requirement is
not very strict (e.g., loVmaxiN < 3% ), only the first
term in Taylor's expansion is needed, and, in tum, the
91
computation is very fast. In Example 1, the numbers
were loVmaxiN = 1.7% when only the first term was
used, and 0.07% when the second term was used as
well.
(3) In Example 2 (3-dimensional curve) even better results
were achieved. The results are better both with the
conventional and the proposed interpolator. The
reason for this is that the 3-D curve has larger minimal
radius of curvature. The proposed interpolator still
offers higher performance.
TABLE 2: INTERPOLATION OF A 3D CUBIC SPLINE.
Proposed
interpolator:
Equation used:
EmaxJilm]
oVmaxN
Conventional
Interpolator:
# of line segments
Emax(llm)
CONCLUSIONS
Eq. 12
0.80
1.3%
10
77
Eq.l9
0.78
0.035%
30
8.8
60
2.2
The amount of geometric information transferred from the
CAD to the CNC must be minimized but must still enable the
machining of any general curve. These conflicting
requirements can be satisfied only by the development of
real-time interpolators for general curves. The most common
method to present curves is by their parametric forms.
Accordingly, a new type of CNC interpolators which can
handle general curves was introduced. The only assumption
made is that the first two derivatives exist, as is the case for
smooth curves. The performance of the interpolator, evaluated
in terms of precision and feedrate deviations, is superior to all
existing interpolation methods.
ACKNOWLEDGEMENTS
This research was partially sponsored by NSF grant #DDM-
9114131 and by the Japan TS Research Fund through the
Technion's Vice President for Research grant #033-0861. The
paper was written while Yoram Koren was a Lady Davis
Visiting Professor at the Technion. The authors acknowledge
the support of the Lady Davis Foundation.
REFERENCES
Adams, J.A., 1975, "The Intrinsic Method for Curve
Definition," Computer Aided Design, Vol. 7, No. 4, pp. 243-
249.
Chou, J.J. and Yang, D.C.H., 1991, "Command Generation
for Three-Axis CNC Machining," Trans. of ASME, Journal of
Engineering for Industry, Vol. 113, August, pp. 305-310.
Chou, J.J. and Yang, D.C.H., 1992, "On the Generation of
Coordinated Motion of Fine-Axis CNC/CMM Machines,"
Trans. of ASME, Journal of Engineering for Industry, Vol.
114, February, pp. 15-22.
Faux, I.D. and Pratt, M.J., Computational Geometry for
Design and Manufacture, Ellis Horwood Publishers, 1979.
Huang, J.T. and Yang, D.C.H., 1992, "Precision Command
Generation for Computer Controlled Machines," PED-Vol. 58,
Nova, pp. 85-104.
Huang, J.T., 1992, "Design and Application of a New CNC
Command Generator for CAD/CAM Integration," Ph.D.
Thesis, UCLA.
Koren, Y., 1976, "Interpolator for a Computer Numerical
Control System," IEEE Transactions on Computers, Vol. C-
25, No. 1, Jan., pp. 32-37.
Koren, Y. and Masory, 0., 1981, "Reference-Pulse Circular
Interpolators for CNC systems," Trans. of ASME, Journal of
Engineering for Industry, Vol. 103, Feb., pp. 131-136.
Koren, Y. and Masory, 0., 1982, "Reference-Word Circular
Interpolators for CNC systems," Trans. of ASME, Journal of
Engineering for Industry, Vol. 104, Nov., pp. 400-405.
Koren, Y., 1983, Computer Control of Manufacturing
Systems, McGraw-Hill, New York.
Lien, T.K., 1980, "Coordinate Transformation in CNC
Systems for Automatic Handling Machines," CIRP Manuf
Syst., Vol. 9, No. 1, pp. 49-60.
Lo, C.C ., 1992, "Cross-coupling Control of Multi-Axis
Manufacturing," Ph.D. Thesis, The University of Michigan,
Ann Arbor.
Loney, G.C. and Ozsoy, T.M., 1987, "NC Machining of
Free Form Surfaces," CAD, Vol. 19, No.2, pp. 85-90.
Makino, H., 1988, "Clothoidal Interpolation - A New Tool
for High-Speed Continuous Path Control," Annals of the
CIRP, Vol. 37, No. 1.
Makino, H. and Ohde, T., 1991, "Motion Control of the
Direct Drive Actuator," Annals of the CIRP, Vol. 40, No. 1,
pp. 375-378.
Papaioannou, S. G., 1979, "Interpolation Algorithms for
Numerical Control," Computers in Industry, Vol. 1, pp. 27-40.
Sata, T., Kimura, F., Okada, N. and Hosaka, M., 1981, "A
New Method of NC Interpolation for Machining the Sculptured
Surface," Annals of the CIRP, Vol. 30, No. 1, pp. 369-372.
Stadelmann, R., 1989, "Computation of Nominal Path
Values to Generate Various Special Curves for Machine,"
Annals of the CIRP, Vol. 38, No. 1, pp. 373-376.
----------- -------
92