84 / APRIL 2008

INTERNATIONAL
JOURNAL
OF PRECISION
ENGINEERING
AND MANUFACTURING
Vol. 9, No.2,
INTERNATIONAL
JOURNAL
OF PRECISION
ENGINEERING
AND MANUFACTURING
Vol.pp.84-92
9 No.2

Parametric NURBS Curve Interpolators : A

Review
Sekar Mohan1, Sung-Hwan Kweon1, Dong-Mok Lee1 and Seung-Han Yang1,#
1 School of Mechanical Engineering, Kyungpook National University, 1370 Sankyuk-dong, Buk-gu, Deagu, South Korea, 702-701
# Corresponding Author / E-mail: [email protected], TEL: +82-53-950-6569 FAX: +082-53-950-6550
KEYWORDS : NURBS, CNC, Free-form machining, Parametric curve Interpolator

Free-form shapes which were once considered as an aesthetic feature are now an important functional
requirement. CNC industries are looking for a compact solution for reproducing free-form shapes as conventional
interpolation models are inadequate. The parametric curve interpolator developed in the last decade has clearly
emerged as favorite among its contemporaries in recent years. At present intense research has been done on
parametric curve interpolators and interesting developments are reported. Out of the various parametric
representations for curves and surfaces, NURBS has been standardized and widely used in free-form shape design.
This paper presents a review of various methods of parametric interpolation for NURBS and discusses the salient
features, problems and solutions. Recent approaches on variable feedrate interpolation, parameter compensation
are also reviewed and research trends are addressed finally.
Manuscript received: February 11, 2008 / Accepted: March 14, 2008

1. Introduction
The introduction of computers to control the machine tool has
significantly improved automation of the machining process. The
availability of low cost microprocessors with increased computing
power and capabilities has led to major advancements in the design
and implementation of Computer Numerical Control (CNC) systems.
Nowadays major portion of processing the part data, interpolation and
servo control are handled through software components. CNC
interpolators have also witnessed significant advances. In contrast to
the Numerical Control (NC) hardwired interpolators which were built
on multiple Digital Differential Analyzers, CNC interpolators are
implemented as software. The main function of the interpolator is to
generate position commands for reproduction of a prescribed shape of
the part with minimum errors. Modern CNC interpolators also have
subroutines, macros, special cycles and look-ahead features. In
addition to linear and circular interpolators, parabolic and elliptical
interpolators are also developed for specialized processes like laser
beam cutting, etc.1,2
Developments in CAD industry witnessed a very rapid pace of
growth due to sophisticated and better mathematical models and
computation facilities that can exactly describe the free-form shapes.
Free-form shapes once used for aesthetics have become an essential
functional requirement in modern product design. Although
machining of complex free-form shapes became a reality with
conventional interpolation methods, accurate reproduction of these
could not achieved, since they are based only on standard motions:
linear and circular.
This necessity motivated researchers to explore new methods for
interpolation. One such interpolator which has received considerable
attention over the last decade is the parametric interpolator. Long
usage of parametric curves to represent the geometry in CAD systems

is an important reason behind this development. It was at this stage,

that, most of the CAD free-form shape design widely followed NonUniform Rational B-Spline (NURBS) representation. In recent years
research on NURBS based parametric interpolators has intensified
and many new algorithms and methods are proposed. This paper
presents a discussion on problems with existing interpolation methods
to appreciate the need of parametric interpolators. This review later
focuses on the recent developments from the perspective of
parametric NURBS curve interpolators. Accomplishments relevant to
the above are also mentioned.

2. An Introduction to NURBS
In CAD/CAM, parametric curves are preferred over implicit
representation due to their extension to three dimensional spaces and
easier implementation in computer. Parametric representation is
suitable to readily obtain cutter offsets for CNC and also lends itself
to piecewise description of curves and surfaces.3 Although these
features enabled description of intricate shapes in CAD, a unified
representation was not made until NURBS was proposed. NURBS
provided this possibility because of its excellent features and
adherence to existing CAD standards. NURBS permits manipulation
of weight and control points thereby offering ease of design of both
standard analytical shapes (e.g. conics) and free-form curves.4 It is
also independent of coordinate system which makes it suitable for
easier and faster transformations. Closed shapes, large slopes, local
control and geometric intuitiveness are some of the other advantages.
Hence CAD/CAM had no hesitation in choosing NURBS as the
common representation of parametric curves. However the NC
technology originated well ahead of it and could not implement it
straightforward. NURBS remained long as a fancy till the

Copyright (c) 2008 by KSPE

INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING

Vol. 9, No.2

development of curve interpolators.

A NURBS curve C( u ) is a vector valued piecewise rational
polynomial function of the form

APRIL 2008 / 85

reliable communication facilities between the CAM and CNC I/O

systems to avoid data starvation, data noise and data loss.

C( u ) =

Ri ,p ( u )Pi

(1)

i =0

where Ri , p ( u ) =

wi N i , p ( u )
n

(2)

N j , p ( u )w j
j =0

The B-spline basis function of degree p is defined recursively as

1 if u i u u i +1
N i ,0 ( u ) =
0 otherwise

(3)

ui+ p+1 u
u ui
Ni,p ( u ) =
Ni,p1( u ) +
Ni+1,p1( u )
ui+ p ui
ui+ p+1 ui+1

(4)

Fig. 1 Conventional approach to machine NURBS curve

where Pi are the control points forming a control polygon, wi are

the weights and Ri,p(u) are the p th degree rational basis functions
defined over a non-uniform knot vector u [0 1] with u as nondimensional curve parameter. Degree p , number of knots m + 1 and
number of control points n can be related by

m = n+ p +1

3.1.2 Non-utilization of Maximum Feedrate

In a typical CNC machine, the G codes are executed in the
controller in a sequential manner. Each block of code has its own
inherent acceleration and deceleration phases consuming much of the
machining time. In addition to overall increase of machining time, the
prescribed feedrate Vc may never be reached and the feedrate
fluctuates between the blocks as shown in Fig. 2. These effects are
more pronounced with the decrease of segment length.

(5)
Vc=42mm/s is not reached

With another parameter v , Eq. 1 can easily be extended to define a

NURBS surface. Piegl5 and Rogers6 give excellent introduction on
the subject of NURBS and may be referred for more details.

3. Interpolation of Free Form Shapes Designed with

NURBS

Fig. 2 Fluctuating feedrate due to segmentation8

3.1.3 Fluctuation of Feedrate

In this paper we presume that free-form curves and surfaces in
CAD are designed with parametric NURBS though other forms such
as explicit and implicit non-parametric forms of representation are
purely admissible. In addition we assume that the NURBS tool path is
available after performing necessary offsetting calculations. The
strategies to generate position commands for NURBS can be
classified into 1) approximation methods, 2) incremental methods and
3) direct curve interpolation methods.

The actual command generation process of conventional linear

interpolator repeats at a sampling interval of T s and the commands
are transmitted to the CNC system as a sequence of reference pulses
or binary word. During each cycle the tool position is incremented by
TV BLUs where BLU (Basic Length Unit) is the length the
machine can move when a control pulse is given. Hence the resulting
number of intermediate commands (N) must be an integer value,
given by,9,10

3.1 Approximating with Linear and Circular Segments

l
N = int
+ 0.5

VT

(6)

In the approximation method, the complex CAD geometry is

discretized into simpler entities offline and the corresponding Gcodes are used for real-time interpolation. A more detailed description
can be found in the report by Jensen et al.7 A CAM program is
usually engaged to automatically generate part programs after
performing necessary offsets. The resulting part program consists of
several lines of G codes or APT statements that correspond to
consecutive linear segments or a combination of both.8 Fig. 1 shows
this approach. Understandably the number of segments must be
maximized for better approximation and surface finish. On the other
hand increase in number of segments will lead to the problems
mentioned below.

This cannot be achieved in all the cases and hence the last
increment pertaining to a shorter length is also carried out in the same
sampling time thus reducing the overall feedrate. Otherwise the
interpolator adjusts the feedrate such that the length of the line
segment is an integer multiple of modified feedrate Vc. This effect is
practically negligible as long as the length of the line segment is big,
which is the case of normal interpolation. For shorter segments, the
above approach results in fluctuation of feedrate and causes poor
surface finish. In most cases additional hand work such as polishing is
required.

3.1.1 Constraints on Resources

3.2 Approximating with Polynomials

The stringent tolerance for precision machining results in a

voluminous part program which usually is not possible to execute
from memory of the CNC controller and hence is split into smaller
part programs. This results in interruption of machining process and
increase in machining time. Nowadays, a powerful CAM program
executed from a remote computer is employed to continuously
transfer the commands to the CNC controller. But this requires

The next choice is to fit the discrete data of cutter paths with
polynomials. Parametric form of polynomials is preferred due to the
advantages stated above. Although it is possible to fit an n th order
polynomial with a dimensionless parameter through data points, a
piecewise approach to approximate the entire curve satisfying the
continuity conditions at the connecting points is widely followed.11
Important ones are the cubic spline and quintic spline interpolation

86 / APRIL 2008

INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING

methods where the parameterization is done based on the chord

length or arc length. Cubic polynomials could represent the shape
well but they suffer from feedrate fluctuations and hence quintic
polynomials parameterized with near arc length are used.12 The
continuity constraints of the composite curve consisting of n quintic
segments are obtained from pre-fitting the cubic spline. The arc
length is approximated using a nonlinear integral equation solved on a
least-squares basis. In another work Chou et al.11 succeeded in
relating the time domain with curve parameter to produce a feedrate
profile derived from kinematic constraints. These approaches have
limited application because some parts of the segment may not
participate in machining at certain feedrates. This can be overcome by
adding more data points that improve arc-length parameterization.13
In the method proposed by Erkorkmaz et al.14 a series of reference
points are connected by quintic splines while the step size is based on
the total travel length. The acceleration and deceleration planned
reference trajectory can be reconstructed at the servo sampling
frequency in real time. Though the feedrate fluctuations are
comparatively less, the above methods require considerable offline
processing and involve higher order polynomials to describe intricate
curves.

3.3 Incremental Methods

intervals.9,10,18 Unlike the above interpolators direct curve

interpolators generate real-time position commands that lie on the
original curve. Since segmentation is not required additional
resources of memory and communication bottlenecks can be avoided.
A study done by Yang et al.9 comparing the linear and parameter
approaches shows that parametric interpolator behaves far better than
its counterpart in terms of memory size requirement, feedrate
fluctuation and position accuracy.
CNC
NURBS
CAD
Model

Part program

Real time
NURBS
Interpolator

Servo
loops

4.1 Derivation of Equations for Parametric Interpolator

In direct curve interpolation the successive parameter
corresponding to the sampling period is evaluated using numerical
integration. On substitution of the resulting parameter in curve
equations such as Eq. 1, the coordinate points can be generated in
real-time. The numerical approximation using Taylors series is
explained below.
Considering a parametric curve C( u ) , the change of u in time
domain is derived from the feedrate V ( u ) as follows:

dC( u )
dt

V(u ) =

C(u)

NURBS data:
Control points
and knot,weight
vectors

Fig. 4 Parametric approach does not require segmentation10

Parametric Curve C(u)

Vol. 9 No.2

(7)

(xi+1, yi+1, zi+1)

n

where ||.|| denotes the Euclidean norm in 3D space. Since u( t ) is a

strictly monotonically increasing function, the feedrate upon
application of chain rule becomes

dP

(xi,yi,zi)
Z
Current interpolated
point

dP

Possible candidate
points
Vector representing
next position

Fig. 3 Non-Orthogonal method searching for next best position15

In incremental methods, the next interpolation point is estimated
from a set of nearby points such that it lies closest to the curve.15 In
this approach incremental moves of the machine tool are obtained
from two conditions namely, direction and step size. The exact
direction of the successive moves is chosen based on the tangent
vectors. The movement can be done one axis at a time (orthogonal) or
simultaneously (non-orthogonal). The non-orthogonal method (see
Fig. 3) is preferred for machining NURBS. The closest point near the
curve can be computed by imposing the other condition: the normal
distance from the curve is a minimum (nt = 0). Integer programming
method is used to solve the unknown steps in terms of BLU and the
corresponding unknown parameter values are computed with
Newton-Raphson method.16,17
In all these methods it is not guaranteed that the interpolated
points lie on the originally designed curve. In case of 3D
curve/surface interpolation, information of surface is required to
avoid deviation of path and probable undercutting. Compared to the
direct curve interpolator which follows this section, the algorithms are
complex and require computation of arc-length for which no closed
form solution is available except for lines.

4. Direct Parametric Curve Interpolation

Parametric curve interpolators (Fig. 4) proposed to overcome the
disadvantages of conventional interpolators use Euler or Taylors
expansion to compute increments of parameter at subsequent time

du
V(u )
=
dt
dC( u )
du

(8)

dC( u )
=
dt

where

dx 2 dy 2 dz 2
+ +
du
du
du

(9)

Since the derivative cannot be expressed in closed form, it is

expanded using Taylors series for time t=kTs where k denotes the
sampling number.
u k +1 = u k + T

V ( kt )
dC ( u )
du u =u

(10)
k

T 2 A( kt )

2 dC( u )
du

dC( u ) d 2 C( u )
V 2 ( kt )

du 2
du
dC( u )
du

+ HOT

4
u =u k

For simplicity first order Taylors expression with a constant feedrate

is widely used.

4.2 Problems with Parametric Approach

The parametric approach is not free of problems as it does not use
the information about the intrinsic geometry of the curve. The
approximation involved in Taylors series truncation induces error
during the evaluation of the successive parameter values. Although
these errors are not cumulative, they cause feedrate error which is
more pronounced when a small curve tolerance and sharp corners are
negotiated. The feedrate error is defined as follows:

feed

s
Vc
= t
Vc

(11)

INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING

Vol. 9, No.2

where Vc is the commanded feedrate and s is the actual distance

traveled. s can be computed from the parametric values obtained
from Taylors series (Eq. 10).
Although feedrate errors are of important concern and researched
to a great extent as explained later, the feedrate errors are less
problematic compared to conventional linear interpolators.9 Inclusion
of higher order terms of Taylors equation can significantly reduce the
feedrate errors albeit at the cost of additional computational load.
The parametric approach fails when the curve derivatives are zero.
This is a very rare case and arises in case of cusps which will not
occur if the curve is a polynomial in terms of non-dimensional
parameter.10 When multiple curves are interpolated exact end may not
be reached. This will cause similar problems posed by the linear
interpolator, though the impact is very little.

5. Types of Parametric NURBS Interpolators

NURBS representation is fundamentally different from the
parametric polynomials mentioned in section 3.2 though they can
approximately represent a NURBS curve. NURBS uses the concept
of weighed sum of the control points determined by the combination
of basis functions. Free-form shapes designed earlier were composed
of Bezier and B-Splines. In one of the pioneering work, Sata et al. 19
proposed an incremental procedure for interpolating Bezier curves
using an approximating step function that can be performed with
addition and bitwise operations. Another approach proposed by
Hermann20 used recursive subdivision algorithm to approximate the
B-Spline curve piece-wise. However better approaches are available
now.
NURBS parametric interpolators can be broadly classified into
1) Uniform increment
2) Constant feedrate
3) Variable feedrate interpolators
The uniform approach is the simplest of the above interpolation
methods but not practicable. Though constant feedrate interpolator is
widely preferred, new designs of variable feedrate interpolators are
also favored.

5.1 Uniform Interpolators

The straightforward approach to implement a parametric
interpolator for NURBS is to uniformly increase u . (uk+1 = uk + u
where u is a constant increment to curve parameter u ). Despite the
simplicity of implementation of this method, it is difficult to obtain
the optimal increment size. To overcome this problem a small value
of u is preferred. Obviously this small value will increase the
machining time. In addition, feedrate fluctuations are bound to
happen as the uniform step size does not correspond to uniform
length of travel in Cartesian space. The fluctuations ultimately
decrease the surface finish of the machined part. Bedi et al.21,22 used
uniform increment to parameter for interpolating B-splines and
suggested to position the control points of the curve in uniform
distances to reduce fluctuations. But this method imposes severe
restriction on the part of CAD designer and requires reparameterization of the original curve.

5.2 Constant Feedrate Interpolators

Constant feedrate interpolators employ Taylors series to compute
successive parametric values according to Eq. 10 or its simplified
form: first order Taylors series. The constant feedrate must be
selected such that it does not result in chord error deviation along the
entire path. Cheng et al. 23 compared real time implementation of
NURBS interpolators employing various methods such as first order,
second order, R-K method etc., and suggested first order (execution
time:1ms) for faster real-time applications and second order
(execution time:2ms) for reduced feedrate error and accurate
positioning. Zhang et al.24 implemented a real time NURBS
interpolator and used a look-up table for the calculation of derivatives

APRIL 2008 / 87

by way of reduction in computation. It is also possible to reduce the

computational effort by storing adjacent basis functions for
calculating the higher order terms.25 Extending these ideas real time
surface interpolators for parametric and NURBS curves were
developed.26,27 The authors used iso-parametric curves where one of
the parameter directions u or v is kept constant while parametric
curve interpolation is carried on the other.

5.2.1 Trajectory Planning for Constant Feedrate Interpolators

Trajectory planning involves ACC/DEC planning for start and
stop motions. G-Codes used in conventional interpolation have these
features built-in but these must be planned explicitly for NURBS
interpolation. For constant feedrate interpolators which start with the
commanded feedrate, the machine stages can be directed to reach the
starting point of the tool path from another point with the commanded
feedrate. However it requires additional codes and careful planning.
Hence it is ideal to plan the acceleration and deceleration phases
during the interpolation process itself.
Different types of profiles for start and stop motions can be
planned. A trapezoidal feedrate velocity profile is used by Zhang24
where the time for acceleration and deceleration are computed by
approximating the distance of the NURBS curve. A known profile of
feedrate allows easier calculation of feedrate at every sampling travel
and the same can be used to find successive parametric increments. If
the curve length is insufficient to accommodate trapezoidal profile, a
triangular profile can be followed. Alternative profiles such as bell
and exponential feedrate profiles are also demonstrated.25.

5.3 Curvature Dependent Interpolators

Curvature dependent interpolators also use the Taylors series
expansion. Instead of constant feedrate, a variable feedrate depending
on the instantaneous curvature of the NURBS curve is used. In one
such method28 a linear relationship between the ratio of curvature
corresponding to half the maximum commanded speed and the
instantaneous current curvature is assumed. The commanded feedrate
is followed at the flatter curve regions. Substituting the variable
feedrate in the Taylors expansion one can find the parameter
increments at successive time intervals. But this method has no
explicit control over chord error since the assumption does not take
this into account. Chord error must be considered as it determines the
size of the tolerance band and surface finish of the machined surface.
To achieve this objective, the feedrate is adaptively adjusted so that
the chord error does not exceed the allowable maximum value. In the
adaptive interpolation method29 the feedrate is expressed as a function
of chord error and instantaneous curvature of the NURBS curve. The
relationship between feedrate V(uk) and chord error is given by the
following expression

V ( uk ) =

2
Ts

k 2 ( k max )2

(12)

where max is the user specified maximum chord error. The

instantaneous radius of curvature can be calculated from

k =

, =

C( 1 )( uk ) C( 2 )( uk )
C

(1)

( uk )

(13)

where is the curvature and C ( 1 ) ( u ) and C ( 2 ) ( u ) are the first

and second derivatives respectively. Substituting the expression of Eq.
12 for V(uk) in Eq. 10, uk+1 can be found. Thus the adaptive
interpolator computes feedrate V(uk) which is kept constant at most
times and changed adaptively when the chord error deviates from the
given tolerance max.
Variable feedrate interpolator is also preferred for constant
material removal and surface finish though the criteria to compute the
feedrate differ from the above. Tikhon et al.30 succeeded in
expressing the feedrate at every sampling interval as function of

88 / APRIL 2008

curvature, tool radius and depth of cut so as to obtain constant

removal of material. In another work which focused on surface finish,
feedrate has been expressed in terms of roughness, radius of the cutter
and pick feed.31
As the feedrate depends on radius of curvature, a large change in
curvature will usually lead to abrupt change in velocity which may
exceed the ACC/DEC and jerk limits of the machine tool. Unless they
are controlled, the drives may saturate and excite the drive train.
Feedrate planning for these limits is therefore essential to implement
the variable feedrate interpolator. They are discussed in section 8.

5.4 Five-axis Interpolators

The above NURBS curve interpolators are primarily meant for 3
axis machining but can also be extended to 5-axis CNC machines
with an additional inverse kinematics module.32 The module
computes the orientation of tool and work table. In contrast to the
conventional approach, the orientation of the tool with the surfacenormal enables production of smoother NURBS surface. These
concepts can be extended to other complex applications also.33

5.5 Other Methods

Arc length parameterization gives exact distance of travel along
the curve and is best suited for constant feedrate applications in CNC
machining. Unfortunately for NURBS, there is no closed form
analytical solution for arc length except for straight lines.34 Though
methods based on numerical quadrature exist they are timeconsuming and unpredictable and hence cannot be applied for realtime CNC applications. However it is possible to use arc length
parameterization when the parametric interval is sufficiently small.
Utilizing this concept, the interpolated points generated at consecutive
sampling time intervals with conventional linear interpolator can be
mapped to the curve parameter by a quintic polynomial. Parametric
curve interpolation with a near constant feedrate is realized upon
substituting this value in Eq.1.35 A preprocessing stage for
conventional interpolation and a search for the segment for a
corresponding arc length is unavoidable.
Generally, for complex curve/surface machining researchers
develop specialized interpolation schemes for improved
performance.36 It is also possible to reconstruct NURBS curve from
the linearly interpolated data though at the cost of discretization error.
In one such method37 the least-squares curve fitting technique is used
to fit a series of discrete points, with predetermined number of control
points. The tolerance is also constantly checked and in case of
violation the limits of the control points are increased. In a recent
development, Li et al.38 developed 5-axis NURBS interpolator to
interpolate trajectories generated with traditional interpolator in the
form of NURBS. It is worth noting that existing G-codes which are
available in conventional format can be interpolated in real-time in
NURBS parametric form with this approach.

6. Computing Chord Error

In all the above mentioned types of interpolators, chord error
plays an important role in the surface quality of the machining part. It
is an important criterion to select feedrate (in case of curve
interpolators) or step length (in case of conventional interpolators).
The chord error deviation can be defined as the maximum orthogonal
distance between the chord formed in the interval of two successive
interpolated points and the desired NURBS curve. As the
computation of chord error for NURBS is not straightforward, several
methods to estimate chord are reported in literature.

6.1 Bisection Method

The classic method proposed by Faux et al.3 considers that, when

the chord error is at maximum, the curve tangent must be
perpendicular to the perpendicular of chord segment. The segment
under consideration (formed with successive interpolated points) is

INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING

Vol. 9 No.2

re-parameterized and the condition of orthogonality between the

chord vector and the chordal deviation vector is used to find the chord
error. The chord error vector with magnitude (chordal length) can be
obtained from the commanded feedrate at a given sampling time. The
resulting chord error corresponding to this length may however, be
more than the desired value and a solution can be found iteratively.
The number of iterations can be reduced by bisection method which
employs an upper and lower bound. Though this may result in a
quicker convergence, the number of iterations is unknown and hence
is suitable for offline applications only.39 Bisection results in error but
can be controlled within limits by adjusting the bounds. Due to the
complexity and increased computational loads this method is not
preferred for NURBS.

6.2 Circular Approximation Method

The most widely used approximation method is circular
approximation method since it is reasonable to approximate a small
NURBS segment with its osculating circle. Employing Pythagoras
theorem 3, 29 chord error k can be approximated as (Fig. 5)
L

k = k k 2 k
2

(14)

where k is radius of the curvature and Lk is the distance between

points C( u k ) and C( u k +1 ) . Upon further simplification of the
above equation the feedrate can be expressed as

v=

2 2
Ts

(15)

C(uk+1)

Circle
P(uk+1)
L(uk)
C(uk)= P(uk)
k
NURBS
Fig. 5 Circular approximation to NURBS

6.3 Arc Substitution Method

The circular approximation may not yield appropriate results at
sharp corners of NURBS tool path. In such cases, the radius can be
approximated. A sharp corner is identified based on tangents40 defined
by two lines formed between the successive interpolation points. An
arc whose radius is defined based on the angle between them and its
geometric derivatives are used to compute the instantaneous radius as
follows.

k =

( )

sin 2
cos

(16)

where is the angle between the two lines formed between two
successive interpolation points and is the permitted chord error.41

7. Parameter Compensation Methods

One of the major drawbacks of curve interpolators is feedrate
error caused by the approximation of the truncated Taylors series.
Although higher order methods exist42 the recursive nature of
NURBS rational functions make this unsuitable for real-time
implementation and hence most of the works prefer to employ the
first order Taylors series. Feedrate errors are especially more at high

Vol. 9, No.2

7.1 Predictor Method

The predictor method is based on the concept of feedback
proposed by Lo40 which uses finite differences to compute the
derivatives of the parametric curve. Although the resulting algorithm
is simple and fast, a large feedrate error is inevitable because of
multiple approximations (finite differences and Taylors expansion).
To reduce this, a revised feedrate is computed corresponding to
coordinate points obtained by updating the current parameter.
However calculating intermediate coordinate points of NURBS
requires computation of recursive rational functions which is as costly
as computing derivatives.

7.2 Predictor Corrector Method

Although feedrate error is reduced using the above approach,
control over the feedrate error is not ensured. In the Predictor
Corrector approach the feedrate error can be brought within limits by
repeatedly iterating the predicted parameter with an updated
parameter. Since the number of such iterations is unknown, the
corrector approach may result in unstable computation time.23 In a
later work Tsai et al.43 derived convergence conditions for real time
implementation.

7.3 Speed Controlled Interpolation

In this method, the interpolated points at (k+1)th time are also

approximated in addition to the Taylors expansion. This enables the
feedrate at kth interval to be expressed in terms of coordinates at uk+1.
Based on this concept, Yeh et al.44 expressed a quadratic equation
with the compensatory parametric value. Even though the feedrate is
maintained constant at most parts of the curve, the feedrate deviates at
certain parameter values. The presence of acceleration term prevents
using second order Taylors expansion.42 To use acceleration further
approximation is needed as the acceleration of NURBS curve cannot
be computed explicitly unlike feedrate. However it can be used with
Taylors first order method as the acceleration term is not involved.45

Various methods have been proposed to modify the profile to

satisfy the ACC/DEC and jerk limits. Most of these methods rely on
the concept of detecting the regions where the feedrate deviates from
the commanded feedrate. Such regions are called as Feed Rate
Sensitive (FRS) corners or key regions. A FRS corner consists of
segments of deceleration and acceleration with the former following
the latter. Depending upon the various chord error methods described
in section 6, a constant feedrate segment may also occur in between.
In the offline look-ahead methods proposed by Yong et al.39, an
ACC/DEC limited feedrate profile simultaneously satisfying chord
error is obtained by analyzing the change in feedrate between
consecutive interpolated points. The feedrate profile is adjusted if the
change violates the limiting values of ACC/DEC. This method is
further improved by Liu et al.47 by incorporating jerk control. They
identified sharp points from the information of geometric derivatives
and gradually reduced the feedrate to zero. Further, the corresponding
jerk along the path is distributed based on a jerk reshaping method.
With these methods it is not possible to reach the exact parameter
values corresponding to start and stop point of FRS corner, since,
Taylors expansion cannot compute subsequent parameter increments
so as to reach the predefined curve parameter ue. Any deviation from
the exact value of ue will result in chord error as well as ACC/DEC
and jerk values exceeding the limits. Hence it is imperative that, to
reach the exact ue value the constant feedrate segment feedrate
segments must be recomputed. But such modification of profile is
possible only if the arc length between the start and stop points is
sufficient so that adequate number of iterations of u could be carried
out. This can be overcome if the new parameter value corresponds to
one of the previously interpolated points. A method using this idea is
implemented by Du et al.41 for real-time look-ahead implementation
using a fixed number of look-ahead points due to restricted buffer size.

[ u1 ,V1 ]n

[ u4 ,V4 ]n
profiles violating
chord error

Decelerating
segment

8. Incorporating Machining Dynamics in the Variable

Feedrate NURBS Interpolator
One of the latest improvements in CNC machining is high speed
machining. Latest CNC machine tools with modern tooling enable
machining at very high speeds and feeds resulting in enhanced
process economics. While machining an intricate shape, non-optimal
feedrates might be required in order to satisfy the chord error. This is
usually overcome by means of look-ahead methods incorporated in
the CNC controller. For a typical CNC machine using conventional
linear interpolator, look-ahead method can foresee a certain number
of blocks and plan the feedrate according to ACC/DEC limits of the
machine tool. But in common parametric curve interpolation, the
whole curve is executed as a single instruction. The parametric nature
of NURBS prevents the usage of conventional ACC/DEC and jerk
control methods as they distort the entire feedrate profile.

(n+1)

Corner (n)

Accelerating
segment
Corner (n+2)

[ u int ,Vint ] n ,n+ 2

Feed rate

7.4 Other Methods

A more practical approach is to compute a new feedrate instead of
adjusting the parameter value with a compensatory value, thereby
retaining the original parameter. The new commanded feedrate can be
computed from distance of travel between the interpolated points. But
it incurs additional computational load to find the distance and
feedrate. A feed forward controller design is also required.25 In
another effort to compensate the parameter, Park et al.46 related BLU
and chord error as the ratio of angles between the actual and
estimated interpolated points assuming near arc approximation.

curvature points and a feedrate error of 5% is obtained with first order

methods and 0.2% with the second order methods.9,40 The parameter
compensation method is an alternate to higher order Taylors series
and is used to reduce the truncation error.

APRIL 2008 / 89

Feed rate mm/s

INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING

[ u2 ,V2 ]n

[ uint ,Vint ] n +1,n + 2 switching points

[ u3 ,V3 ]n

Constant feedrate
Curve parameter- u

Fig. 6 Interaction between adjacent feedrate profiles

When a complex NURBS profile is adaptively interpolated,
several key regions (FRS corners) will form depending on the
commanded feedrate. When a NURBS profile is planned for
ACC/DEC limits the resulting new ACC/DEC corrected segments
may overlap with neighboring corners (shown as dotted lines in Fig.
6). The complex interaction of adjacent feedrate corners thus makes
the computed look-ahead profile with the above procedures invalid. A
new profile passing through the intersection points [uint,Vint]n,n+2 and
[uint,Vint]n+1,n+2 can be used.
Sun et al.48 proposed a method to compute the feedrate profile
without using look-ahead method explicitly. The feedrate sensitive
regions are found by relating acceleration, chord error and sampling
time together to a permissible slope of a curve, defined by parameter
and curvature. However, jerk control was not implemented in their
method. It also requires the conversion of feedrate profile curve to a
NURBS curve. Considering real time implementation, an additional
NURBS curve for feedrate profile will increase the computational
load significantly.
A cubic curve fitting procedure is proposed to overcome this
problem.49 In this method coefficients of the cubic feedrate curve are
computed considering ACC/DEC and jerk limits. Expressing the

90 / APRIL 2008

feedrate profile as cubic curve and relating it to knot vectors of the

NURBS curve the information can be easily embedded with the
modified control points and passed to the controller so that a jerk
limited feedrate profile can be realized in real-time. Since velocity
can be computed based on the cubic feedrate profile, the CNC
controller is relieved from computing the second derivative (Eq. 12).
As every segment of a FRS corner must be checked with all the other
FRS corners for possible interaction, the intersection algorithm
however becomes computationally intense and new intersecting
patterns may emerge. When the length of the cubic feedrate segment
becomes short it becomes difficult to reach the feedrate at the
intersection points. Therefore the feedrate profile continuity may be
lost resulting in violation of ACC/DEC and jerk limits.

Fig. 7 Chord error free jerk limited cubic feedrate profile

In a modified approach the feedrate profile can be fitted first,
using cubic curve with ACC/DEC and jerk constraints also being
satisfied and later adjusted for chord error.50 As the cubic curve passes
through the region of chord error, the starting point [u1,V1] is moved
backwards to a previously interpolated value. In case of accelerating
profile it is shifted forward. Fig. 7 shows this method. In the similar
manner, cubic feedrate profiles are constructed with subsequent
feedrate switching points and a complete feedrate profile satisfying
the chord error and ACC/DEC and jerk limits is obtained.
In general, real-time systems for NURBS interpolation require
multiple modules/stages to satisfy ACC/DEC and jerk limits.46
Modules to obtain information on geometry, plan for look-ahead and
interpolate are often executed as multiple threads of varying priority.
Recently Lin et al.51 proposed a look-ahead method integrating servo
dynamics. In this approach the segments with sharp corners are
identified considering instantaneous curvature and acceleration. The
dynamic feedrates at sharp corners are computed from the system
frequency that satisfies the steady state contour error tolerance.
Further a jerk limited module is employed to optimize different types
of feedrate profiles which arise due to path constraints. Nam and
Yang52 proposed a generalized variable feedrate interpolator which
can handle NURBS and other curves also. The interpolator computes
subsequent parameters of composite curves confining to path and
kinematic constraints by estimating the curve length in real-time. The
feedrate profile also satisfies the minimum time criterion by following
a trapezoidal acceleration profile. With improved controller
designs53,54 real-time systems for NURBS can result in precise
machining of complex shapes at high speeds.

9. Research Directives
At present the aerospace, automobile, die and mold and jewelry
industries are the prominent users of freeform shapes designed with
NURBS and in the future, this will extend to other applications also.
Parametric NURBS curve interpolators are very useful and the
implementation of G-Codes for NURBS in commercial controllers
such as FANUC shows their success.55
Currently research is focused on removing an important drawback
of parametric interpolators the feedrate error with various models.
Though preprocessing the NURBS curve offers a realistic estimation
of parameter, it needs further refinement. Variable feedrate

INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING

Vol. 9 No.2

interpolation is given much importance as high speed machining

became a reality. It has attracted much attention as it is unnecessary to
command the machine at a lower constant feedrate than optimal.
However the command generation process must consider the machine
tool limits so that trajectory generated is smooth and does not saturate
feed drives. A varying feedrate interpolator will also result in chatter
vibration which causes rapid tool wear and poor surface finish.
Enhanced variable feedrate interpolators with ACC/DEC and jerk
control are also designed and evaluated and further improvements are
expected.
The possibility of approximating NURBS with Pythagorean
hodographs34 and Quintic splines is interesting. PH curves have
closed form solution to arc length and can be expressed in Bernstein
form. On contrary to NURBS, Quintic splines12,13,14,56 can satisfy
dynamics of both the axes individually and show desirable properties.
But they are not much popular like NURBS.
Another area requiring more attention is machining of curves and
surfaces with 5-axis CNCs. For machining of such free-form shapes
the CNC industry still depends largely on conventional linear
segmentation as the procedures for generating tool paths are well
defined. In 5-axis CNC the cutting tool must be treated as rigid body
and must be driven with angular feedrate in addition to translation.
Though the theory of linear parametric interpolator is well developed,
parametric angular feedrate interpolation must be given more
atttention.57,58

10. Conclusions
This paper has summarized the research results of parametric
curve interpolators for NURBS from the aspect of command
generation algorithms. A comparison with other types of interpolation
approaches is also presented. Though the arc-length parameterized
approach is less prone to feedrate error, the parametric curve
approach permits real-time implementation of free-form curves and
surface with reduced data transfer and contour error and is preferred.
Parameter compensation methods, chord error computation methods
and recent designs of variable feedrate interpolators are also discussed.
These methods show promising results at the research level. This
encourages to think that in the imminent future precise and highspeed machining of NURBS curves and surfaces will be realized.

REFERENCES
1. Koren, Y., Computer Control of Manufacturing Systems,
McGraw-Hill, pp. 115-142, 1998.
2. Koren, Y., Control of Machine Tools, ASME Journal of
Manufacturing Science, Vol. 119, Issue 4B, pp. 749-755, 1997.
3. Faux, L. D. and Pratt, M. J., Computational Geometry for
Design and Manufacture, Ellis Horwood, pp. 261-277, 1979.
4. Piegl, L., On NURBS: A Survey, IEEE Computer Graphics and
Applications, Vol. 11, No. 1, pp. 55-71, 1991.
5. Piegl, L. and Tiller. W., The NURBS Book, Springer, pp. 117138, 1997.
6. Rogers, D. F., An Introduction to NURBS: with Historical
Perspective, Morgan Kaufmann Publishers Inc., pp. 129-150,
2001.
7. Jensen, C. G. and Anderson, D. D., A Review of Numerically
Controlled Methods for Finish Sculptured-Surface Machining,
IIE Transactions, Vol. 28, No. 1, pp. 30-39, 1996.
8. Vickers, G. W. and Bradley, C., Curved Surface Machining
Through Circular Arc Interpolation, Computers in Industry, Vol.
19, No. 3, pp. 29-37, 1992.

INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING

Vol. 9, No.2

9. Yang, D. C. H. and Kong, T., Parametric Interpolator versus

Linear Interpolator for Precision CNC Machining, ComputerAided Design, Vol. 26, No. 3, pp. 225-234, 1994.
10. Shiptalni, M., Koren. Y. and Lo, C. C., Real-Time Curve
Interpolators, Computer-Aided Design, Vol. 26, No. 11, pp. 832838, 1994.
11. Chou, J. J. and Yang, D. C. H., Command Generation for Three
Axis CNC Machining, ASME Journal of Engineering for
Industry, Vol. 113, No. 3, pp. 306-310, 1991.
12. Wang, F. C. and Yang, D. C. H., Nearly Arc Length
Parameterized Quintic Spline Interpolation for Precision
Machining, Computer-Aided Design, Vol. 5, No. 5, pp. 281-288,
1993.
13. Wang, F. C. and Wright, P. K., Open Architecture Controllers for
Machine Tools, Part 2: A Real Time Quintic Spline Interpolators,
ASME Journal of Manufacturing Science, Vol. 120, No. 2, pp.
425-432, 1998.
14. Erkorkmaz, K. and Altintas, Y., High Speed CNC System
Design Part 1: Jerk Limited Trajectory Generation and Quintic
Spline Interpolation, International Journal of Machine Tool and
Manufacture, Vol. 41, No. 9, pp. 1323-1345, 2002.
15. Papaioannou, S. G., Interpolation Algorithms for Numerical
Control, Computers in Industry, Vol. 1, No. 1, pp. 27-40, 1979.
16. Kritsis, D., High Precision Interpolation Algorithm for 3D
Parametric Curve Generation, Computer-Aided Design, Vol. 26,
No. 11, pp. 850-856, 1994.
17. Omirou, S. L., Space Curve Interpolation for CNC Machines,
Journal of Materials Processing Technology, Vol. 141, No. 3, pp.
343-350, 2003.
18. Huang, J. T. and Yang, D. C. H., A Generalized Interpolator for
Command Generation of Parametric Curves in Computer
Controller Machines, ASME Japan-USA symposium on Flexible
Automation, pp. 393-399, 1992.
19. Sata, T., Kimura, F., Okada, N. and Hosaka, M., A New Method
of NC Interpolation for Machining Sculptured Surfaces, Annals
of CIRP, Vol. 30, No. 1, pp. 369-372, 1980.
20. Hermann, G.., Patch Programming: The Integration of Motion
Planning into Numerical Control, Computers in Industry, Vol. 5,
No. 4, pp. 351-359, 1984.
21. Bedi, S., Ali, I. and Quan, N., Advanced Interpolation
Techniques for NC Machines, ASME Journal of Engineering for
Industry, Vol. 115, No. 3, pp. 329-336, 1993.

APRIL 2008 / 91

26. Lin, R. S., Real-time Surface Interpolator for 3D Parametric

Surface Machining on 3-Axis Machine Tools, International
Journal of Machine Tool and Manufacture, Vol. 40, No. 10, pp.
1513-1526, 2000.
27. Tsai, M. C., Cheng, C. W. and Cheng, M. Y., A Real-time
Surface Interpolator for Precision Three Axis CNC Machining,
International Journal of Machine Tool and Manufacture, Vol. 43,
No. 12, pp. 1217-1227, 2003.
28. Zhiming, X., Jincheng. C. and Zhengjin, F., Performance
Evaluation of a Real-Time Interpolation Algorithm for NURBS
Curve, International Journal of Advanced Manufacturing
Technology, Vol. 20, No. 4, pp. 270-276, 2002.
29. Yeh, S. S. and Hsu, P. L., Adaptive Feedrate Interpolation for
Parametric Curves with Confined Chord Error, Computer-Aided
Design, Vol. 34, No. 3, pp. 229-237, 2002.
30. Tikhon, M., Ko, T. J., Lee, S. H. and Kim, H. S., NURBS
Interpolator for Constant Material Removal Rate in Open NC
Machine Tools, International Journal of Machine Tool and
Manufacture, Vol. 44, No. 2, pp. 237-245, 2004.
31. Choi, I. H., Yang, M. Y., Hong, W. P. and Jung, T. S., Curve
Interpolation with Variable Feedrate for Surface Requirement,
International Journal of Advanced Manufacturing Technology,
Vol. 25, No. 3/4, pp. 325-333, 2005.
32. Koren, Y. and Lin, R. S., Five-Axis Surface Interpolators,
Annals of CIRP, Vol. 44, No.1, pp. 379-382, 1995.
33. Kang, C. G., Online Trajectory Planning for a PUMA Robot,
International Journal of Precision Engineering and Manufacturing,
Vol. 8, No. 4, pp. 51-56, 2007.
34. Farouki, R. T. and Shah, S., Real-time CNC interpolators for
Pythagorean Hodograph curves, Computer-Aided Geometric
Design, Vol. 13, No. 7, pp. 583-600, 1996.
35. Jeong, S. Y., Choi, Y. J. and Park, P. G.., Parametric Interpolation
using Sampled Data, Computer-Aided Design, Vol. 38, No. 1, pp.
39-47, 2006.
36. Kim, H. T., Yang, H. J. and Kim, S. C., Control Method for the
Tool Path in Aspherical Surface Grinding and Polishing,
International Journal of Precision Engineering and Manufacturing,
Vol. 7, No. 4, pp. 51-56, 2006.
37. Yau, H. T. and Kuo, M. J., NURBS Machining and Feed Rate
Adjustment for High-Speed Cutting of Complex Sculptured
Surfaces, International Journal of Prod. Research, Vol. 39, No. 1,
pp. 21- 41, 2001.

22. Bedi, S. and Quan, N., Spline Interpolation Technique for NC

Machines, Computers in Industry, Vol. 18, No. 3, pp. 307-313,
1992.

38. Li, W., Liu, Y. Yamazaki, K. Fujisima, M. and Mori, M., The
Design of a NURBS Pre-Interpolator for Five-Axis Machining,
International Journal of Advanced Manufacturing Technology, in
print (DOI 10.1007/s00170-006-0905-8).

23. Cheng, M. Y., Tsai, M. C. and Kuo, J. C., Real time NURBS
Command Generators for CNC Servo Controllers, International
Journal of Machine Tool and Manufacture, Vol. 42, No. 7, pp.
801-813, 2001.

39. Yong, T. and Narayanasami, R., A Parametric Interpolator with

confined Chord Errors, Acceleration and Deceleration for NC
Machining, Computer-Aided Design, Vol. 35, No. 13, pp. 12491259, 2003.

24. Zhang, Q. G. and Greenway, R. B., Development and

Interpolation of a NURBS Curve Motion Interpolator, Robotics
and Computer-Integrated Manufacturing, Vol. 14, No. 1, pp. 2736, 1998.

40. Lo, C., Feedback Interpolators, ASME Journal

Manufacturing Science, Vol. 119, No. 4, pp. 587-592, 1997.

25. Cheng, C. W. and Tsai, M. C., Real-time Variable Feedrate

NURBS Curve Interpolators for CNC Machining, International
Journal of Advanced Manufacturing Technology, Vol. 23, pp.
865-873, 2004.

of

41. Du, D., Liu, Y., Yan, C. and Li, C., An Accurate Adaptive
Parametric Curve Interpolator for NURBS Curve Interpolator,
International Journal of Advanced Manufacturing Technology,
Vol. 32, No. 9/10, pp. 999-1008, 2007.
42. Farouki, R. T. and Tsai, Y. F., Exact Taylor Series Coefficients
for Variable Feedrate CNC Curve Interpolators, Computer-

92 / APRIL 2008

Aided Design, Vol. 35, No. 2, pp. 155-165, 2001.

43. Tsai, M. C. and Cheng, C. W., Real-time Predictor Corrector
Interpolator For CNC Machining, ASME Journal of
Manufacturing Engineering and Science, Vol. 125, No. 3, pp.
449-460, 2003.
44. Yeh, S. S and Hsu, P. L., The Speed Controlled Interpolator for
Machining Parametric Curves, Computer-Aided Design, Vol. 31,
No. 5, pp. 349-357, 1999.
45. Yau, C., Du, D. and Li, C., Design of a Real-time Adaptive
Interpolator with Parameter Compensation, International Journal
of Advanced Manufacturing Technology, Vol. 35, No. 1/2, pp.
169-178, 2007.
46. Park, J. H., Nam, S. H. and Yang, M. Y., Development of a Realtime Trajectory Generator for NURBS Interpolator based on a 2
Stage Interpolation Method, International Journal of Advanced
Manufacturing Technology, Vol. 26, No. 4, pp. 359-365, 2005.
47. Liu, X., Ahamd, F., Yamazaki, K. and Mori, M., Adaptive
Interpolation Scheme for NURBS Curves with the Integration of
Machining Dynamics, International Journal of Machine Tool
and Manufacture, Vol. 45, No. 4/5, pp. 433-444, 2005.
48. Sun, Y., Wang, J. and Guo, D., Guide Curve based Interpolation
Scheme of Parametric Curves for Precision Machining,
International Journal of Machine Tool and Manufacture, Vol. 46,
No. 3/4, pp. 235-242, 2006.
49. Sekar, M., Narayanan, V. N. and Yang, S. H., Design of Jerk
Bounded Feedrate with Ripple Effect for Adaptive NURBS
Interpolators, International Journal of Advanced Manufacturing
Technology, in print (DOI: 10.1007/s00170-007-992-1).
50. Lee, D. M., Sekar, M. and Yang, S. H., NURBS Interpolator
using Minimum Time Method for High Speed Machining,
Proceeding of Leading Edge Manufacturing-21, pp. 187-190,
2007.
51. Lin, M. T., Tsai, M. S and Yau, H. T., Development of a
Dynamics-Based NURBS Interpolator with Real-Time LookAhead Algorithm, International Journal of Machine Tool and
Manufacture, Vol. 47, No. 15, pp. 2246-2262, 2007.
52. Nam, S. H. and Yang, M. Y., A Study on a Generalized
Parametric Interpolator with Real-Time Jerk Limited
Acceleration, Computer-Aided Design, Vol. 36, No. 1, pp. 27-36,
2004.
53. Lee, M. H., Yang, S. H., Kim, Y. S., Roh, Y. R. and Kim, C., A
Multi-Axis Contour Error Controller for Free Form Curves,
JSME International Journal Series C, Mechanical Systems,
Machine Elements and Manufacturing, Vol. 47, No. 1, pp. 144149, 2004.
54. Jee, S. C. and Lee, H. C., An Integrated Approach to the
Analysis and Design of a Three-Axis Cross-Coupling Control
System, International Journal of Precision Engineering and
Manufacturing, Vol. 8, No. 2, pp. 59-63, 2007.
55. Shima, A., Sasaki, T., Ohtsuki, T. and Wakinotani, Y., 64-bit
RISC-based Series 15 NURBS Interpolation, FANUC Technical
Review, Vol. 9, No. 1, pp. 23-28, 1996.
56. Erkorkmaz, K. and Altintas. Y., Quintic Spline Interpolation with
Minimal Feed Fluctuation, ASME Journal of Manufacturing
Science and Engineering, Vol. 127, No. 2, pp. 339-349, 2005.
57. Sarma, R. and Rao, A., Discretizors and Interpolators for FiveAxis CNC Machines, ASME Journal of Manufacturing Science
and Engineering, Vol. 122, No. 1, pp. 191-197, 2000.

INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING

Vol. 9 No.2

58. Xu, H. Y., Dai, J. S. and Tam, H. Y., Angular Interpolations for
Planar Implicit Curves, International Journal of Prod. Research,
Vol. 40, No. 9, pp. 2125-2140, 2002.