Precision Engineering 65 (2020) 23–31

Contents lists available at ScienceDirect

Precision Engineering
journal homepage: http://www.elsevier.com/locate/precision

A novel curvature circle iterative algorithm for contour error control of

multi-axis CNC machine tools
Ming Yang a, Jixiang Yang b, Limin Zhu a, *, Xi Yu b
a
State Key Laboratory of Mechanical System and Vibration, School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai, 200240, China
b
State Key Laboratory of Digital Manufacturing Equipment and Technology, School of Mechanical Science and Engineering, Huazhong University of Science and
Technology, Wuhan, Hubei, 430074, China

A R T I C L E I N F O A B S T R A C T

Keywords: High-precision real-time estimation of contouring errors is a prerequisite for contouring errors control of multi-
Contouring error estimation axis CNC machine tools. This paper focuses on developing a nearest point projection curvature circle iterative
Contouring control (NPP–CCI) algorithm to achieve real-time estimation of multi-axis contouring errors. It is found that the tradi­
CNC
tional curvature circle iterative (CCI) method has two major shortcomings. The first is that the iterative process
Curvature circle iterative
may terminate incorrectly at the local contour position, and the other is that the actual tool position and local
curvature circle are not necessarily coplanar in three-dimensional space, which would lead to inaccurate
calculation of the delay time parameter and eventually affect the estimation accuracy. In order to address the
problem of false termination, an index method is used to find the closest reference position with respect to the
actual position. At the same time, the projection technology is proposed to overcome the problem met in
extending the planar curvature circle iterative method to the spatial applications. The proposed NPP-CCI algo­
rithm is more suitable for spatial contouring errors estimation in tracking complex trajectories and has higher
estimation accuracy than the traditional CCI algorithm. Various experiments with different tool paths are con­
ducted on an in-house developed multi-axis experimental platform to verify the effectiveness of the proposed
algorithm. The experimental results show that the NPP-CCI algorithm can estimate the contouring errors with
higher accuracy than the traditional CCI algorithm, and with the help of real-time computation and compen­
sation, the contouring errors are reduced by more than 44% in terms of the MAX and RMS values.

1. Introduction conducted various motion control studies in the recent decades. There
are two main strategies: One is to improve the accuracy of contour by
Multi-axis CNC machine tools are widely used in machining various reducing the tracking error, and the other is to directly model and
industrial parts because of their high machining accuracy and reliability. control the contouring errors to realize contouring control.
With the development of aviation, aerospace and energy industries, the There are two main methods to improve the tracking accuracy of
surface quality requirements for complex components have increased CNC machine tools. The first method is to generate the well-designed
significantly. Accordingly, the machining accuracy requirements of the interpolation trajectories, and the second method is to employ the
multi-axis CNC machine tools are also enhanced. However, due to the advanced controllers. For the first strategy, in order to avoid the influ­
bandwidth limitations of the motor drive and the inevitable external ence of the singular regions on the tracking accuracy, Yang et al. [3]
disturbances, such as nonlinear friction, the tracking errors always presented a singularity avoidance method by modifying the control
appear for each axis [1,2]. Because of the coupling among multiple axes, points of the tool orientation spline. Wan et al. [4] proposed a two-step
multi-axis tracking error could cause the actual position of the tool to method. The first step is to project the three-dimensional curve repre­
deviate from the desired trajectory, which is described as the contouring senting the tool orientation path onto a two-dimensional plane, and the
errors. Obviously, contouring errors directly affect the quality of the second step is to use an optimization strategy to locally deform the tool
surface of the machined workpiece. In order to improve the contour path to avoid the singular regions. For the second strategy, Tomizuka [5]
accuracy of multi-axis CNC machine tools, many researchers have developed a feedforward controller named the Zero Phase Error

* Corresponding author.
E-mail address: [email protected] (L. Zhu).

https://doi.org/10.1016/j.precisioneng.2020.05.005
Received 17 March 2020; Received in revised form 26 April 2020; Accepted 11 May 2020
Available online 14 May 2020
0141-6359/© 2020 Elsevier Inc. All rights reserved.
M. Yang et al. Precision Engineering 65 (2020) 23–31

Tracking Control (ZPETC) which improves tracking accuracy through contouring error estimation methods in regions of large curvature, but
pole/zero cancellation and phase cancellation. However, due to the this method was currently only applicable to a biaxial system.
open loop characteristics, the feedforward controller is sensitive to the In addition to the above works of estimating contour errors by linear
changes in system parameters and external disturbances. To address this or curved approximation methods, there are also researchers who
problem, Weck and Ye [6] removed the high frequency content of the employed the iterative methods to estimate contour errors. El Khalick
control commands through a low pass filter with zero phase character­ et al. [22] proposed a contouring errors estimation procedure for
istics. Altintas et al. [7] proposed an adaptive sliding mode controller for three-dimensional machining tasks, which was based on the iterative
feed drives. It is robust against uncertainties in the drive’s parameters, estimation of the local curvature of the reference trajectory and co­
maximizes the bandwidth within physical limitations, and compensates ordinates transformation. Later, El Khalick et al. [23] developed a novel
for external disturbances such as friction and cutting force. Erkorkmaz sliding mode contouring controller with a nonlinear sliding surface to
et al. [8] proposed a systematic approach for designing a control law improve the machining accuracy for three-dimensional workpieces.
which provide a high tracking bandwidth as well as adequate distur­ However, the three-dimensional contouring errors estimation algorithm
bance rejection and parameter variation robustness, in order to mini­ in literature [22] has two main flaws. The first is that a false termination
mize the tracking errors in each axis. Yang et al. [9] proposed a tracking may occur at the local contour position during the iterative process. The
error pre-compensation method based on a novel friction model for other is that it is not accurate enough to calculate the delay time
feedforward compensation to reduce the tracking error and improve the parameter due to the non-coplanarity of the actual tool position and
multi-axis contour accuracy. local curvature circle.
Although improving the tracking accuracy can improve the contour In order to alleviate the disadvantages of the curvature circle itera­
accuracy to some extent, due to the coupling among multiple axes, high tive (CCI) algorithm [22], a nearest point projection curvature circle
tracking accuracy of each axis does not necessarily ensure satisfactory iterative (NPP–CCI) algorithm is proposed in this paper. The two prob­
contour accuracy [10,11]. Therefore, another way to improve the ac­ lems are solved by the nearest point index method and the projection
curacy of the contour is to directly model and control the contouring method, respectively. The remainder of this paper is organized as fol­
errors. Since the information of contouring errors is a prerequisite for lows. In Section 2, after briefly introducing the characteristics of the CCI
contouring errors control of multi-axis CNC machine tools, it is vital to algorithm, the solutions to the two main flaws of the original algorithm
develop real-time contouring calculation algorithms with high accuracy. are proposed, and the NPP-CCI contouring errors estimation algorithm is
However, the calculation of the exact contouring errors of free-from then developed. Experimental results are given in Section 3, and con­
trajectories is very computationally intensive, it is difficult to meet the clusions are drawn in Section 4.
real-time requirements. Hence, many researchers have attempted to
calculate the approximate contouring errors rather than the exact ones 2. Multi-axis contouring errors estimation algorithm
for contouring errors control. The pioneering research on contouring
errors modeling and control was carried out by Koren [12], who 2.1. Traditional CCI algorithm
modeled the contouring errors by exploiting the linear relationship be­
tween the contouring errors and the tracking errors, and then developed As is shown in Fig. 1, in the fixed frame Σ W , the curve o is the desired
a cross-coupled controller (CCC) to improve the contour accuracy. P. K. contour curve of a machined part followed by the feed drive system; the
Kulkarni et al. [13] derived the mathematical formulation of a class of variable r ¼ ½rX ; rY ; rZ �T is the desired position of the machined part at
cross-coupled controller from the point of view of optimal control. time t and the corresponding actual position of the feed drive system is
Subsequently, Koren et al. [14] proposed a variable-gain cross-coupled q ¼ ½qX ; qY ; qZ �T . The collection of the tracking errors in three feed drive
controller based on an improved contouring errors estimation algorithm
axis, eW ¼ ½eX ; eY ; eZ �T , is defined as the difference between the reference
to track the nonlinear trajectory. Yeh et al. [15] proposed a contouring
position r and the actual position q , which can be expressed as
errors estimation method based on the tangential approximation to the
free-form path. Although this algorithm needs a small computational eW ¼ ½eX ; eY ; eZ �T ¼ r q (1)
load, the estimation accuracy is greatly affected by the amount of the
The above errors are defined in the coordinate frame Σ W .
tracking error. To relieve this problem, Yang et al. [11] proposed an
Contouring error is defined as the shortest distance between the
algorithm based on the coordinate transformation and circular approx­
actual position and the desired reference trajectory. As is shown in
imation, in which the contouring error vector was represented by the
Fig. 1, the contouring error corresponding to the actual position q is ec .
coordinate of the actual tool tip position with respect to a specific virtual
Since it is difficult to obtain the analytical expression of the contouring
coordinate frame. Xi et al. [16] proposed a method for estimating the
error, the method of approximate estimation is generally accepted in
contouring errors of linear and circular paths based on the Taylor series
consideration of the calculation load. The CCI algorithm [22,23] is such
expansion. However, this method is not applicable to the free form
a method, and its procedure is summarized as follows:
contours. To address this problem, Huo et al. [17] developed the
generalized Taylor series expansion method for estimating the con­
(1) As shown in Fig. 1(a), a local coordinate frame Σ L is defined with
touring errors of arbitrary free form contour. Later, Huo et al. [18]
origin at r and axes t, n and b, where the t axis is along the
proposed a generalized cross-coupling control (GCCC) method for con­
tangential direction of o at r, the n axis is along the normal di­
touring errors control. This method can be applied to arbitrary free form
rection of o at r, and the b axis is along the bi-normal direction
contour without prior knowledge of the contour function or velocity
that is normal to t and n. For the parametric trajectory, the
estimator. Zhu et al. [19] proposed a novel approach for calculating
tangential, normal, and bi-normal vectors, denoted as t, n and b,
contouring errors of arbitrary contours in real-time based on the
respectively, are calculated at a time t as follows:
second-order Taylor approximant of the point-to-curve distance func­
tion. In order to improve the accuracy of Taylor expansion for con­ 8 r_
t¼
touring errors estimation, Song et al. [20] used the Taylor expansion to >
>
>
> krk
_
>
approximate the ideal contour to a third-order arc length parameterized >
< r€
curve, and then obtained the contouring errors by calculating the dis­ n¼ (2)
>
> rk
k€
tance between the actual position and the approximate contour. Yang >
>
>
> r_ r€
et al. [21] proposed a method for efficiently estimating the contour error :
b¼t�n¼ �
of a two-dimensional plane, which can eliminate sudden change of krk
_ rk
k€
contouring error vector directions that are encountered by traditional

24
M. Yang et al. Precision Engineering 65 (2020) 23–31

Fig. 1. Contour error estimation based on curvature circle iteration: (a) first iteration and (b) second iteration [22].

contour error, ec , than the distance between point r and point q.

(2) The curvature at the desired position r is calculated as follows: (6) The second iteration is illustrated in Fig. 1(b), where the esti­
mated point s2 is closer to the actual position point q than the
1 kr_ � r€k
¼ ; r_ 6¼ 0 (3) formerly estimated point s1 Accordingly, it is possible to obtain
R _ 3
krk an estimated point sm which is closest to the actual position q on
the desired trajectory. There are two criteria to terminate the
where R denotes the curvature radius which is the reciprocal of the iterative process. The first one is that the error magnitude kem k ¼
curvature. ksm qk begins to increase during the iterative process, and the
second one is to set the maximum number of iteration times
(3) The circle of curvature is located in the t n plane and perpen­ directly with a trade-off between the calculation load and esti­
dicular to b. The center of the circle c can be determined using a mation accuracy.
coordinate transformation between the fixed and the local coor­
dinate frames: The above is a brief overview of the algorithm for estimation con­
T touring errors based on traditional CCI method. However, this algorithm
c ¼ ½cX ; cY ; cZ � ¼ r þ L½0; R; 0�
(4) has two obvious flaws in estimating the three-dimensional contouring
L ¼ ½t n b�
errors. The first is that the algorithm may erroneously terminate at the
local contour position during the iteration process. The other is that due
(4) The angle θ can be calculated as follows: to the non-coplanarity of the actual tool position and the local curvature
a⋅n circle, the delay time parameter calculated by Eq. (6) is not accurate
1
θ ¼ cos
kak⋅knk enough, which will eventually damage the estimation accuracy. This
(5) paper will use the index method and projection technology to solve these
c q
a¼
kc qk two problems, and the specific content is detailed in the next section.

(5) A new local coordinate frame Σ L1 is then defined at point s1 ¼ 2.2. NPP-CCI algorithm
rðt1 Þ, where time t1 is the time elapsed for moving throughout the
segment r s with a constant velocity equaling to the desired As mentioned in the previous section, the iterative algorithm has two
velocity at r. Here s denotes the point on the curvature circle that termination conditions, one of which is that the error magnitude kem k ¼
is closest to the actual position q. It is assumed that the time ksm qk begins to increase during the iterative process. However, this
required to pass the segment r s is the same as that required to termination condition does not take into account the problem of existing
pass the segment r s1 on the desired trajectory. The delayed local nearest estimated points. As shown in Fig. 2, sm 1 , sm and smþ1 are
time t1 is calculated as follows: contouring points that are estimated sequentially by the CCI algorithm.
And the distances of these three estimated contouring points to the
t1 ¼ t td actual position point q are L1 , L2 and L3 , respectively, and there is the
R⋅θ (6) following relationship:
td ¼ ; r_ 6¼ 0
krk
_ �
L3 < L1 < L2
(7)
The estimated point s1 is closer to the actual position point q than with L1 ¼ kq sm 1 k; L2 ¼ kq sm k; L3 ¼ kq smþ1 k
point r. The distance between point s1 and point q is closer to the actual
According to the first termination condition of the traditional CCI

25
M. Yang et al. Precision Engineering 65 (2020) 23–31

curvature is located on the plane t0 n0 and the center of the circle is c0 .

The actual position q is on the straight line along the direction of vector
b0 , and it do not coincide with point s0 . It should be noted that here it is
just to simplify the analysis of the flaw in the original algorithm, so it is
assumed that the actual position q is on the line along the direction of
vector b0 . Naturally, s0 is the point on the desired trajectory that is
closest to the actual position q, so s0 should be the final estimated
contour position. However, if the first termination condition of the
original iterative algorithm is temporarily ignored, the iterative process
Fig. 2. The diagram of local nearest contour position met in the estima­ continues until the maximum number of iterations is reached. Moreover,
tion process. according to Eq. (5) and Eq. (6), the delay time calculated by the angle
between the vectors ðs0 qÞ and n0 with the value of π 2 will eventually
=

algorithm, the iterative algorithm would terminate when it arrives at impair the estimation accuracy of the contouring error. Thanks to the
point sm , and point sm 1 will be determined as the contouring position first termination condition of the original algorithm, inappropriate
and L1 is the contouring error. However, point sm 1 is not the actual estimation in this case may not occur. Although this unreasonable esti­
contouring position closest to the actual position q, and it is only a local mation does not occur in this situation, there is a situation that may
nearest contouring position on the segment r sm . Obviously, the first cause an unreasonable estimation that affects the final estimation ac­
termination condition of the original iterative algorithm would curacy. As shown in Fig. 3(b), the currently estimated contour position
adversely affect the estimation accuracy of the contouring errors. s0 is located near the ideal contour position sc , and the angle θ0 used to
In order to address this problem, an index-based method is proposed iteratively estimate the next contour position is the angle between the
to find the reference position closest to the actual position at first, and vectors ðs0 qÞ and n0 , which will be used to calculate the delay time by
then start the iterative calculation at this nearest reference point. Since Eq. (6). The calculated delay time is over-estimated because the angle θ0
is too large, and the contour error is obtained at the estimated point s0 .
0
the interpolated reference positions generated by the CNC system can be �0 �
stored in a buffer, the nearest reference position relative to the actual Since the Euclidean norm �s0 q� > ks0 qk, the iterative estimation
tool tip position q can be found by calculating and comparing the dis­ process would terminate according to the first iteration termination
tances between actual position q and each of the reference positions in criterion. s0 will be the final estimated contour position corresponding
the vicinity of the actual position. Following this idea, the following to the current actual position. The above case are likely to occur in
calculation is performed for every reference position command in a practical applications. The reason why the angle used to calculate the
certain scope around the desired reference position rd , and the reference delay time is uncontrollable is that the plane of the local curvature circle
point, named the nearest position rn , corresponding to the smallest d is changes as the iterative process progresses. As a result, the corre­
picked out: sponding angle used to calculate the delay time in three-dimensional
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi space would also vary greatly.
d ¼ ðq rd i ÞT ðq rd i Þ; with i 2 ½ N; N� (8) To address this difficulty, a projection-based method is proposed to
improve the traditional CCI algorithm. As shown in Fig. 3(c), point q is
0

where N determines the searching scope around the desired reference the projection of the actual position q on the plane t0 n0 , which con­
position command rd . The principle of choosing a suitable N is to tains the curvature circle. The angle used to calculate the delay time is
consider the magnitude of tracking errors along the trajectory, but it is no longer the angle θ0 between the vectors ðs0 qÞ and n0 , but the angle
often constrained by the computational power of the CNC’s CPU [24]. θ0 between the vectors ðs0 q Þ and n0 . Assuming that the currently
0 0

When the nearest reference position is found, the value of the corre­ estimated contour position s0 is the ideal contour position, the vector
sponding subscript i can be also obtained, and finally the time parameter ðs0 qÞ is definitely perpendicular to the tangent vector t0 at s0 , and the
tn corresponding to the nearest reference position rn can be obtained. projection ðs0 q Þ of ðs0 qÞ on the plane t0 n0 coincides with the
0

Another flaw of the original iterative algorithm is that the calculation normal vector n0 at s0 . At this time, the angle between the vectors
of delay time parameter is not accurate enough because it does not take
ðs0 q Þ and n0 is zero. It means that the angle obtained by using the
0

into account the fact that the actual position is not coplanar with the
projection technique to calculate the delay time is more appropriate, and
local curvature circle. As shown in Fig. 3(a), a local coordinate frame Σ L0
the angle would gradually approach to zero. As a result, the estimated
is defined and its origin is at s0 and axes are t0 , n0 and b0 . The circle of
contour position would gradually approach to the ideal contour position

Fig. 3. The diagram of the case that the actual position and the local curvature circle are not coplanar.

26
M. Yang et al. Precision Engineering 65 (2020) 23–31

as much as possible. follows

In each iteration, the projection of the actual position on the plane 8 �π � 2 3
ti ni in the local coordinate system Σ Li must be re-calculated. Taking >
>
x ¼ 2 cos
10
t mm
>
> 6 7
Fig. 3(c) as an example, the projection point q of q on the plane t0 n0 > pffiffi � π �
0
< 6 7
can be obtained by the following formula: y ¼ 2sin t mm t 2 6
60 : 0:001 : 20
7
7 (10)
> 10 6 7
>
> � � 4 5
> p ffi ffi
qT b0 sT0 b0 : z ¼ 2sin 2π t mm
>
(9)
0
q ¼q b0 10
bT0 b0
8 � � 2 3
After the projection operation, the delay time is calculated by >
� t� 2π
> x¼ 1þ cos t mm
changing q with q in Eq. (5). >
0
>
> 10 10 6 7
>
> 6 7
The first and second iterations of the proposed NPP-CCI algorithm < � t � � �
2π
6
6
7
7
t 2 0 : 0:001 : 30 (11)
are illustrated in Fig. 4, where rn is the reference interpolation position >
>
y ¼ 1 þ
10
sin
10
t mm 6
6
7
7
> 6 7
closest to the actual position q found by the index method, and q1 and q2 >
0 0
>
> 4 5
>
: t
are the projection of the actual position on the corresponding planes in z¼
10
mm
the first and second iterations, respectively. Other notations are with the
same as those employed in the CCI method. The basic flow chart of the The tracking times for the two path are 20 and 30 s, respectively. The
proposed NPP-CCI contouring errors estimation algorithm is shown in tool path profiles and curvatures are illustrated in Fig. 8.
Fig. 5. Firstly, the path of saddle-shaped curve is used to test the accuracies
of four contouring errors estimation algorithms under different iteration
3. Experiments times. The four contour error estimation algorithms are the original CCI
algorithm, the CCI algorithm combined with the nearest point index
In order to evaluate the effectiveness of the proposed NPP-CCI con­ technology, the CCI algorithm combined with the projection technique,
touring error estimation algorithm, various experiments are carried out and the NPP-CCI algorithm. The computational results are summarized
on an in-house developed table-tilting five-axis CNC platform as shown in Fig. 9. As shown in Fig. 9 (a), when the number of iterations of the
in Fig. 6, where each axis is driven by a Yaskawa AC servomotor with four algorithms exceeds five times, the maximum (MAX) and the root
built-in current control loop. The CNC platform is equipped with an mean square (RMS) values of the discrepancy between the actual and the
open real-time controller board, dSPACE 1103, the key processor board estimated contouring errors tend to be substantially stable. Therefore, in
of which is a 604 PowerPC with 400 MHz processing frequency. The order to balance the estimation accuracy and the computational load,
drives are controlled by using PID controllers at a sampling frequency of the maximum number of iterations selected in this paper is five. The
1 kHz as shown in Fig. 7, where r, q and e represent the reference po­ actual contouring errors represented in Fig. 9 are numerically calculated
sition, actual position and tracking error, respectively. Ka is the current off-line by the Newton-Raphson iteration method as introduced in
amplifier gain and Kt is the torque constant of the motor. J and B are the Ref. [25]. From Fig. 9 (a), it can be seen that neither the nearest point
equivalent inertia and viscous damping of the motor shaft, respectively. index technology nor the projection technology cannot significantly
rg is the transmission gain from angular to linear motion of the ball- improve the estimation accuracy of the original CCI algorithm, and even
screw. It should be noted that only the X, Y and Z axes are used in the lead to the decrease of the estimation accuracy. However, when the two
experiments. technologies are simultaneously used, the estimation accuracy of the
To compare the effectiveness of the proposed NNP-CCI algorithm CCI algorithm exhibits a great improvement.
with that of the traditional CCI algorithm, two test trajectories are As shown in Fig. 9 (b) (c) and Table 1, the proposed NPP-CCI algo­
employed in the experiments. The first path is a saddle-shaped curve, rithm has better estimation accuracy than the original CCI algorithm in
which has been used in Ref. [22], and the second path is a spatial spiral general, and only at very few locations, the accuracy of the proposed
curve. The mathematical expressions of the two trajectories are as NPP-CCI algorithm is not as high as that of the original method. The

Fig. 4. Contour error estimation using the proposed NPP-CCI algorithm: (a) first iteration and (b) second iteration.

27
M. Yang et al. Precision Engineering 65 (2020) 23–31

Fig. 5. The flow chart of the proposed NPP-CCI algorithm for contouring errors estimation.

Fig. 6. The five-axis experimental platform.

Fig. 7. The block-diagram of the axis drive using PID controller.

28
M. Yang et al. Precision Engineering 65 (2020) 23–31

Eq. (6), and however, the local curvature arc does not fully reflect the
true path and the tools do not necessarily move along the path with a
constant speed. Although the algorithm proposed in this paper cannot
completely solve this problem, compared with the original method, it
has a significant improvement in the estimation accuracy of the con­
touring errors. In order to further evaluate the accuracy of the proposed
algorithm, the contouring error components on individual axes are
depicted in Fig. 10. It can be seen that the proposed algorithm has better
estimation accuracy than the original CCI algorithm.
In order to verify the real-time performance of the proposed esti­
mation algorithm in the contouring errors control, the experimental
results with and without the contour error compensation are also
compared. Here, the contouring errors control algorithm introduced in
Ref. [26] is adopted, which directly compensates the contouring errors

Table 1
Comparison of contouring error discrepancies between the estimated and actual
Fig. 8. The tool path profiles and curvatures: (a) saddle-shaped curve [22], (b) values in tracking the saddle-shaped and spiral paths with different algorithms.
spiral curve. Saddle-shaped curve Spiral curve

Max(mm) RMS(mm) Max(mm) RMS(mm)

reason for this phenomenon is that the delay time parameter is calcu­
CCI algorithm 0.0066 1.5524e-3 7.8663e-3 3.4668e-3
lated by assuming that the delay time is approximately equal to the time
NPP-CCI algorithm 0.0048 7.5017e-4 4.8308e-4 1.0250e-4
elapsed for traveling the local arc R⋅θ with a constant speed, as shown in

Fig. 9. Experimental results of different estimation algorithms for different paths: (a) the MAX and RMS values of the estimation discrepancy for the four different
estimation algorithms with different iterative times in tracking the saddle-shaped path, (b) comparison of estimation accuracies of two algorithms in tracking the
saddle-shaped curve, (c) comparison of estimation accuracies of two algorithms in tracking the spiral curve.

29
M. Yang et al. Precision Engineering 65 (2020) 23–31

component of each axis through a control gain m to the position com­ tool/workpiece deformation caused by cutting forces when focusing on
mand of the corresponding driver in real-time. More detailed de­ the effect of the contouring error estimation and control algorithms. This
scriptions of the controller design can refer to Ref. [26]. The controller kind of experiments were commonly adopted to evaluate the effective­
gain selected in the experiment is m ¼ 1, and the experimental results ness of the tool path generation or smoothing algorithm, such as [10,
show that the contouring errors after compensation is significantly 11].
reduced as shown in Fig. 11. Specifically, as shown in Table 2, for the
saddle-shaped testing path, the MAX and RMS values of the contouring 4. Conclusion
errors are reduced by 44.72% and 48.86%, respectively; for the spiral
testing path, the MAX and RMS values of the contouring errors are In this paper, the nearest point projection curvature circle iterative
reduced by 50.04% and 49.69%, respectively. Hence, the proposed (NPP–CCI) algorithm is proposed to improve the estimation accuracy of
contouring error estimation algorithm can be readily incorporated into contouring errors in tracking complicated spatial trajectories. Compared
the real-time contouring error control algorithm to improve the con­ with the traditional CCI algorithm, the proposed NPP-CCI algorithm has
touring accuracy of multi-axis CNC machine tools. the following two advantages. The first one is that it alleviates the
It should be pointed out that the current study did not take into ac­ problem of false termination by using the index method. The second one
count the impact of cutting force and cutting temperature on the quality is that it considers the problem that the actual tool position and local
of machining parts. In fact, as shown in literature [27–29], the cutting curvature circle are not necessarily coplanar in three-dimensional space,
force and cutting temperature have a significant influence on the quality and thus introduces the projection technique to calculate the delay time
of machining parts. The reason why not to choose metal cutting in ex­ parameter more accurately. Experiment results on an in-house devel­
periments is to exclude the error sources caused by thermal errors, oped multi-axis experimental platform confirm the effectiveness of the

Fig. 10. Comparison of the estimated and actual contouring error components on individual axes. (a) saddle-shaped curve; (b) spiral curve.

30
M. Yang et al. Precision Engineering 65 (2020) 23–31

[2] Tang L, Landers RG. Multiaxis contour control-the state of the art. IEEE Trans Contr
Syst Technol 2013;21:1997–2010. https://doi.org/10.1109/TCST.2012.2235179.
[3] Yang J, Yuen A. An analytical local corner smoothing algorithm for five-axis CNC
machining. Int J Mach Tool Manufact 2017;123:22–35. https://doi.org/10.1016/j.
ijmachtools.2017.07.007.
[4] Wan M, Liu Y, Xing WJ, Zhang WH. Singularity avoidance for five-axis machine
tools through introducing geometrical constraints. Int J Mach Tool Manufact 2018;
127:1–13. https://doi.org/10.1016/j.ijmachtools.2017.12.006.
[5] Tomizuka M. Zero phase error tracking algorithm for digital control. J Dyn Syst
Meas Contr 1987;109:65–8. https://doi.org/10.1115/1.3143822.
[6] Weck M, Ye G. Sharp corner tracking using the IKF control strategy. CIRP Ann -
Manuf Technol 1990;39:437–41. https://doi.org/10.1016/S0007-8506(07)61091-
9.
[7] Altintas Y, Erkorkmaz K, Zhu WH. Sliding mode controller design for high speed
feed drives. CIRP Ann Manuf Technol 2000;49:265–70. https://doi.org/10.1016/
S0007-8506(07)62943-6.
[8] Erkorkmaz K, Altintas Y. High speed CNC system design. Part III: high speed
tracking and contouring control of feed drives. Int J Mach Tool Manufact 2001;41:
1637–58. https://doi.org/10.1016/S0890-6955(01)00004-9.
[9] Yang M, Yang J, Ding H. A two-stage friction model and its application in tracking
error pre-compensation of CNC machine tools. Precis Eng 2017;51:426–36.
https://doi.org/10.1016/j.precisioneng.2017.09.014.
[10] Zhang K, Yuen A, Altintas Y. Pre-compensation of contour errors in five-axis CNC
machine tools. Int J Mach Tool Manufact 2013;74:1–11. https://doi.org/10.1016/
j.ijmachtools.2013.07.003.
[11] Yang J, Li Z. A novel contour error estimation for position loop-based cross-coupled
control. IEEE/ASME Trans. Mechatronics. 2011;16:643–55. https://doi.org/
Fig. 11. Contouring error compensation results in tracking two different paths. 10.1109/TMECH.2010.2048718.
(a) saddle-shaped curve; (b) spiral curve. [12] Koren Y. Cross-coupled biaxial computer control for manufacturing systems. J Dyn
Syst Meas Contr 1980;102:265–72.
[13] Kulkarni PK, Srinivasan K. Optimal contouring control of multi-axial feed drive
serwomechanisms. J. Manuf. Sci. Eng. Trans. ASME. 1989;111:140–8. https://doi.
Table 2 org/10.1115/1.3188743.
Comparison of contouring error discrepancies between the estimated and the [14] Koren Y, Lo C-C. Variable-gain cross-coupling controller for contouring. CIRP Ann.
actual values in tracking two different paths with and without error Technol. 1991;40:371–4.
[15] Yeh SS, Hsu PL. Estimation of the contouring error vector for the cross-coupled
compensation.
control design. IEEE/ASME Trans. Mechatronics. 2002;7:44–51. https://doi.org/
Saddle-shaped curve Spiral curve 10.1109/3516.990886.
[16] Xi X-C, Poo A-N, Hong G-S, Huo F. Experimental implementation of Taylor series
Max(mm) RMS(mm) Max(mm) RMS(mm) expansion error compensation on a bi-axial CNC machine. Int J Adv Manuf Technol
2010;53:285–99. https://doi.org/10.1007/s00170-010-2843-8.
Uncompensated 0.0881 0.0526 0.1407 0.0978 [17] Huo F, Xi XC, Poo AN. Generalized Taylor series expansion for free-form two-
Compensated 0.0487 0.0269 0.0703 0.0492 dimensional contour error compensation. Int J Mach Tool Manufact 2012;53:91–9.
� �
Compensated 44.72% 48.86% 50.04% 49.69% https://doi.org/10.1016/j.ijmachtools.2011.10.001.
1 *
Unompensated [18] Huo F, Poo AN. Improving contouring accuracy by using generalized cross-coupled
100% control. Int J Mach Tool Manufact 2012;63:49–57. https://doi.org/10.1016/j.
ijmachtools.2012.07.012.
[19] Zhu L, Zhao H, Ding H. Real-time contouring error estimation for multi-axis motion
systems using the second-order approximation. Int J Mach Tool Manufact 2013;68:
NPP-CCI algorithm. When tracking a saddle-shaped trajectory, the NPP- 75–80. https://doi.org/10.1016/j.ijmachtools.2013.01.008.
CCI algorithm can reduce the MAX and RMS values of the estimation [20] Song D-N, Zhong Y-G, Ma J-W. Third-order contour-error estimation for arbitrary
deviation by 27.27% and 51.68%, respectively. Also, the NPP-CCI al­ free-form paths in contour-following tasks. Precis Eng 2019;60:85–92. https://doi.
org/10.1016/j.precisioneng.2019.07.009.
gorithm is applied to real-time contouring control, and the contouring [21] Yang Xiao, Seethaler Rudolf, Zhan Chengpeng. Dun Lu, and wanhua zhao, A novel
errors are reduced by more than 44% in terms of the MAX and RMS contouring error estimation method for contouring control. IEEE/ASME Trans.
values. Furthermore, the proposed algorithm is currently aimed to three- Mechatronics. 2019;24:1902–7. https://doi.org/10.1109/LRA.2016.2579746.
[22] El Khalick A, Uchiyama MN. Contouring controller design based on iterative
axis CNC machining, and the future work will extend it to five-axis CNC contour error estimation for three-dimensional machining. Robot Comput
machining. Integrated Manuf 2011;27:802–7. https://doi.org/10.1016/j.rcim.2011.01.001.
[23] El Khalick AM, Uchiyama N, Sano S. Sliding mode contouring control design using
nonlinear sliding surface for three-dimensional machining. Int J Mach Tool
Declaration of competing interest Manufact 2013;65:8–14. https://doi.org/10.1016/j.ijmachtools.2012.07.004.
[24] Yang M, Yang J, Ding H. A high accuracy on-line estimation algorithm of five-axis
contouring errors based on three-point arc approximation. Int J Mach Tool
The authors declare that they have no known competing financial Manufact 2018;130–131:73–84. https://doi.org/10.1016/j.
interests or personal relationships that could have appeared to influence ijmachtools.2018.04.001.
the work reported in this paper. [25] Yang J, Ding H, Zhao H, Yan S. A generalized online estimation algorithm of multi-
axis contouring errors for CNC machine tools with rotary axes. Int J Adv Manuf
Technol 2015;84:1239–51. https://doi.org/10.1007/s00170-015-7783-x.
Acknowledgement [26] Yang J, Altintas Y. A generalized on-line estimation and control of five-axis
contouring errors of CNC machine tools. Int J Mach Tool Manufact 2015;88:9–23.
https://doi.org/10.1016/j.ijmachtools.2014.08.004.
This research is supported by the National Natural Science Founda­ [27] Ning J, Nguyen V, Liang SY. Analytical modeling of machining forces of ultra-fine-
tion of China under Grant Nos. 91648202 and 91948301. grained titanium. Int J Adv Manuf Technol 2019;101:627–36. https://doi.org/
10.1007/s00170-018-2889-6.
[28] Ning J, Liang SY. Predictive modeling of machining temperatures with force-
References temperature correlation using cutting mechanics and constitutive relation.
Materials (Basel) 2019;12. https://doi.org/10.3390/ma12020284.
[1] Jia Z, Ma J, Song D, Wang F, Liu W. A review of contouring-error reduction method [29] Ning J, Nguyen V, Huang Y, Hartwig KT, Liang SY. Constitutive modeling of ultra-
in multi-axis CNC machining. Int J Mach Tool Manufact 2018;125:34–54. https:// fine-grained titanium flow stress for machining temperature prediction. Bio-Design
doi.org/10.1016/j.ijmachtools.2017.10.008. Manuf. 2019;2:153–60. https://doi.org/10.1007/s42242-019-00044-9.

31