Research Article

Advances in Mechanical Engineering

2021, Vol. 13(8) 1–18
Ó The Author(s) 2021
A corner smoothing algorithm using DOI: 10.1177/16878140211040671
journals.sagepub.com/home/ade
Trajectory Pattern Method (TPM) for
high-speed and high-quality machining

Hao Li1,2 , Jahangir Rastegar3, Baosheng Wang1,2,

Wenjiang Wu4,5 and Zhuwen Yan1,2

Abstract
In micro-line segments machining, transition curves with high harmonic components are more prone to causing vibration
issues in the feed drive system, which affects machining efficiency and quality severely. To construct low harmonic trajec-
tories, this paper proposes a corner smoothing algorithm that uses the Trajectory Pattern Method (TPM). The transition
curve construction and axial motion scheduling are performed with a specified fundamental frequency in one step, which
reduces the smoothing process time and avoids excitation of natural modes of vibration of the system. The synthesized
trajectories and axial kinematic profiles are all smooth and only contain the selected fundamental frequency and its first
two odd harmonics, which minimizes the number of high harmonic components in the required actuation forces/torques
and avoids excitation of the system modes of vibration. Linear programming is used to synthesize the trajectories. The
proposed algorithm is shown to achieve near time-optimal trajectories. The provided experimental analysis and compar-
isons demonstrate that the proposed algorithm achieves smooth axial kinematic profiles with low harmonic contents,
which would improve machining efficiency and quality.

Keywords
Trajectory pattern method, low harmonic, vibration, corner smoothing, axial motion planning

Date received: 2 April 2021; accepted: 30 July 2021

Handling Editor: James Baldwin

Introduction
The most common practice in CNC machining is to 1
express tool paths by small line-segments. Research Industrial Technology Research Institute of Intelligent Equipment,
Nanjing Institute of Technology, Nanjing, China
work into high-speed and high-quality machining has 2
Jiangsu Provincial Engineering Laboratory of Intelligent Manufacturing
concentrated on methods to improve smoothness of Equipment, Nanjing Institute of Technology, Nanjing, China
3
trajectories. To this end, several researchers have inves- Department of Mechanical Engineering, Stony Brook University, Stony
tigated methods for smooth tool path generation.1,2 4
Brook, NY, USA
Shenyang Institute of Computing Technology, Chinese Academy of
Some researchers have focused on the Local Smooth
Sciences, Shenyang, China
Method (LSM), in which transition curves are con- 5
Shenyang CASNC Technology Co., Ltd, Shenyang, China
structed at the junction of small line-segments to
improve corner smoothness and velocity since it is sim- Corresponding author:
ple to implement and is easily controlled. For example, Hao Li, Industrial Technology Research Institute of Intelligent Equipment,
Jiangsu Provincial Engineering Laboratory of Intelligent Manufacturing
Lv et al.3 used arcs to blend small line-segments. Zhang Equipment, Nanjing Institute of Technology, 1 Hongjing Road, Nanjing
et al.4 constructed transition curves with quintic poly- 211167, China.
nomials and Ernesto and Farouki5 used rational Email: [email protected]

Creative Commons CC BY: This article is distributed under the terms of the Creative Commons Attribution 4.0 License
(https://creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work
without further permission provided the original work is attributed as specified on the SAGE and Open Access pages
(https://us.sagepub.com/en-us/nam/open-access-at-sage).
2 Advances in Mechanical Engineering

quadratic Bézier curves. These three algorithms, how- as those based on polynomial, Akima, or B spline
ever, only provide G1 continuity, that is, continuous curves contain significant high harmonic components.
velocity, at the corner, which would lead to vibration In addition, the nonlinearity of the machine system
caused due to the present normal acceleration. To dynamics would require actuating forces/torques with
obtain G2 continuity, that is, continuous curvature, Du even higher harmonic content to follow the prescribed
et al.6 smoothed small line-segments with cubic Bézier trajectories. The lack of control over the harmonic con-
curves. The calculation of a proper transition length, tent of the required actuating forces/torques and their
however, requires a significant amount of time. To frequencies in currently available and published meth-
reduce vibration during the machining process, ods for trajectory synthesis leads to excitation of natu-
Pythagorean-Hodograph (PH) spline has drawn a lot ral modes of vibration of machining systems. The
of attention for corner smoothing studies because it commonly used actuator input filtering does not allevi-
provides analytical solution of arc-length related to ate the problem since the filtered components are
spline parameters. Shi et al.7,8 developed corner needed to achieve the prescribed motion and results in
smoothing algorithms for multi-axis tool paths based path and trajectory errors.
on quintic PH splines, which avoided the approxima- Rastegar15 introduced the concept of trajectory pat-
tion errors of arc-length calculation. In order to simul- terns and their application to synthesis of trajectories
taneously achieve real-time interpolation and higher with low harmonic content such that the harmonic con-
order continuous motion, Hu et al.9 proposed a C3 tent of the required actuation forces/torques is con-
continuous analytical corner smoothing algorithm by trolled and does not contain significant high harmonic
inserting the specially designed micro PH splines content.15–18 The Trajectory Pattern Method (TPM)
among linear tool path segments. Then, Hu et al.10 was then introduced for trajectory synthesis and model-
extended the algorithm to construct C3 continuous tra- based control of high-speed and precision machin-
jectories for five-axis machine tools. Fan et al.11 ery.15,16 Using TPM, the fundamental frequency of the
enhanced the continuity further by utilizing two quartic synthesized trajectories can be selected such that
Bézier curves to achieve G3 continuity, that is, smooth together with all its harmonics that appear in the trajec-
curvature. Smooth transition curves and tangential tory and the required actuating forces/torques would
motion profiles can be achieved by above algorithms. not excite the natural modes of vibration of the system.
However, the methods require two computational In TPM, trajectory patterns are classes of parametric
steps. In the first step, parametric curves are used at the trajectories that are used to synthesize trajectories with
corners to construct the transition schemes. In the sec- the desired characteristics and derive the system inverse
ond step, tangent motion profiles are scheduled based dynamics in an algebraic form.15,16 The actuating
on the synthesized trajectories. These algorithms are forces/torques can then be readily optimized and used
therefore inefficient. In addition, sudden changes in the in a model-based control schema to achieve high-speed
axis velocity, acceleration, and/or jerk profiles are and high-precision operation of the system.
encountered since only tangential kinematic limits are It has been shown that the trajectory patterns that
considered. are constructed with sinusoidal functions with an
In recent years, one LSM has drawn researchers’ appropriate fundamental frequency can be used to
attention, in which in one step, the transition schema synthesize trajectories with low harmonic content that
construction and motion scheduling are performed. would satisfy the dynamic response limitations of the
Biagiotti and Melchiorri12 smoothed tool paths using system actuators while avoiding excitation of the natu-
finite impulse filters. The method, however, leads to ral modes of vibration of the system.16,18
large contour errors and unavoidable time delay. In this paper, a corner smoothing algorithm is pre-
Tajima and Sencer13 developed a direct axis motion sented that uses harmonic based trajectory patterns for
profile blending approach to obtain transition curves. high-speed and high-quality machining. In Section 2, it
Sudden changes in axis jerk may, however, occur at is shown that transition curve construction and axis
corners due to the use of a method based on jerk- velocity planning with a specified fundamental fre-
limited acceleration and deceleration process. Zhang quency can be performed in one step using the TPM.
and Du14 used a jounce-limited velocity planning The synthesized trajectories and axis kinematic profiles
method to achieve transition curves and smooth axis are smooth and only contain the fundamental fre-
acceleration profiles, but axis jerk profiles are still not quency and its first two harmonics. Linear program-
smooth. ming is used to synthesize the trajectories and the
It is important to note that when expressed in proposed algorithm is shown to achieve near time-
Fourier Series, all above synthesized trajectories, such optimal trajectories. In Section 3, the effectiveness of
Li et al. 3

the proposed algorithm is illustrated through the results Transition schema construction
of computer simulations studies. Conclusions of the Figure 1 shows the flow diagram of the developed tran-
present study are presented in Section 4. sition schema. The transition schema construction
includes five modules: an initialization module; a
boundary conditions definition module; a motion pro-
Low harmonic transition schema file constraint definition module; the contour error con-
In this section, a Trajectory Pattern Method (TPM) straint definition module; and a trajectory parameters
based transition schema, which achieves smooth axis calculation module. The above modules are described
motion profiles with chord error and axis kinematic in more detail in the following sections.
constraints is presented. In this scheme, trajectory pat-
terns that are synthesized with a fundamental frequency Initialization module.. Figure 2 presents the proposed
and its first two harmonics are used to achieve mini- transition schema. In Figure 2, let the point Pi,s be
mum trajectory harmonic content.18 the origin to establish the x–y coordinate system and
lengths of line-segments Pi21Pi and PiPi + 1 be li
and li + 1, respectively. The two line-segments Pi21Pi
Trajectory pattern and PiPi + 1 form a corner, with u1 indicating the angle
The class of trajectory patterns used in the present between the vector Pi1 Pi and the positive x-axis and
study is harmonic based and is formed by a fundamen- u2 indicating the angle between the vectors Pi1 Pi and
tal sinusoidal function and (n 2 1) number of its Pi Pi + 1 . The dotted curve is the transition schema that
harmonics: is to be synthesized by the proposed algorithm, for
which the corner length for the line-segments Pi, sPi
n
X and PiPi,e is hereinafter indicated by Li,c. To simplify
f (t) = ½ai cos (2pift) + bi sin (2pift), n 2 N ð1Þ the transition curve construction process and obtain
i=0
smooth axial motion profiles as shown in Figure 3, the
where ai and bi are trajectory parameters; t is the time, line-segments Pi,sPi and PiPi,e are set to have equal
lengths and equal initial and final tangential velocities
t 2 ½0, T , and T is the motion period; f is the selected
of vi,c, accelerations of ai,c = 0 and jerks of ji,c = 0. The
fundamental frequency of the trajectory pattern.
point Pi,m as shown in Figure 2 is the mid-point of the
In this paper, the trajectory pattern (1) that consists
transition schema.
of a fundamental frequency sinusoidal function and its
The user is to provide the required chord error e and
first two odd harmonics is used to construct the kine-
the fundamental frequency f of the trajectory pattern,
matic functions of the transition schema. The velocity
which must be selected such that together with its first
function is then expressed as,
two odd harmonics, the natural modes of vibration of
the system are not excited.
v(t) = a0 + a1 cos (2pft) + a3 cos (6pft)
ð2Þ
+ a5 cos (10pft)
Boundary conditions. From the indicated end conditions
From equation (2), the acceleration and jerk functions and equations (2)–(5), the following boundary condi-
are then derived as, tions are obtained,

a(t) =  2pfa1 sin (2pft)  6pfa3 sin (6pft) sx (0) = ac = 0 ð6Þ

ð3Þ
 10pfa5 sin (10pft) vx (0) = a0 + a1 + a3 + a5 = vi, c cosu1 ð7Þ
ax (0) = 0 ð8Þ
j(t) =  (2pf )2 a1 cos (2pft)  (6pf )2 a3 cos (6pft)
ð4Þ
 (10pf )2 a5 cos (10pft) jx (0) =  (2pf )2 a1  (6pf )2 a3  (10pf )2 a5 = 0 ð9Þ
vx (T ) = a0 + a1 cos (2pfT) + a3 cos (6pfT)
By integrating equation (2) with respect to the time t, ð10Þ
the displacement function is derived as, + a5 cos (10pfT) = vi, c cos(u1 + u2 )
ax (T ) =  2pfa1 sin (2pfT )  6pfa3 sin (6pfT)
a1 sin (2pft) a3 sin (6pft) ð11Þ
s(t) = ac + a0 t + +  10pfa5 sin (10pfT) = 0
2pf 6pf
ð5Þ
a5 sin (10pft) jx (T ) =  (2pf )2 a1 cos (2pfT)  (6pf )2 a3 cos (6pfT)
+
10pf  (10pf )2 a5 cos (10pfT) = 0
ð12Þ
4 Advances in Mechanical Engineering

Figure 1. Flow diagram of the developed transition schema.

From equations (6)–(12), it is clear that if the cycle time vx (0) = a0 + a1 + a3 + a5 = vi, c cosu1 ð14Þ
T is set equal to half the fundamental frequency period, vx (T ) = a0  a1  a3  a5 = vi, c cos(u1 + u2 ) ð15Þ
p
that is, if T = 2pf = 2f1 , equation (11) is then always
satisfied and the required equality condition that have jx (0) =  jx (T ) =  a1  9a3  25a5 = 0 ð16Þ
to be satisfied become the following:
Motion profile constrains. To achieve the desired motion
sx (0) = ac = 0 ð13Þ
profiles shown in Figure 3, the condition a(t) ø 0,
Li et al. 5

The functions g1(t), g2(t), and g3(t) with arbitrary

amplitudes and the resulting acceleration a(t) are
plotted as shown in Figure 4. As expected, and can be
seen in Figure 4, the extreme values of a(t) can only
occur at one of the extreme points of the fundamen-
tal frequency function g1(t), or the functions g2(t)
and g3(t). This is easily seen to be the case since the
functions g2(t) and g3(t) are the first two odd harmo-
nics of the fundamental sinusoidal function g1(t).
Therefore, the condition, a(t) ø 0, t 2 ½0, T, can be
satisfied only if a(t) ø 0 at all extreme points of
g1(t), g2(t), and g3(t).
The extreme points of the functions g1(t), g2(t), and
g3(t) may be calculated as follows:
As an example, for the function g1(t), the extreme of
this sin(j) function occurs at j = (2k + 1)p/2, k 2 Z,
Figure 2. Transition schema. where Z is the set of all integers. Then from equation
(18), the extreme point equation of g1(t) are at,

2k + 1
2pft = p, k2Z ð21Þ
t 2 ½0, T  must be guaranteed, which indicates that v(t) 2
is a monotonically increasing function from vs to ve. Therefore, from equation (18) for the time t 2 ½0, T, k
From equation (3), it can be seen that the acceleration is limited to,
a(t) is a sum of three sine functions and can be
expressed as, 1 1
 łkł , k2Z ð22Þ
2 2
a(t) = g1 (t) + g2 (t) + g3 (t) ð17Þ
Then in general, for all integers k satisfying equation
where (22), the extreme point times are determined for the
function from equation (21). For the case of the g1(t),
g1 (t) =  2pfa1 sin (2pft) ð18Þ the extreme point is readily seen to be at,
g2 (t) =  6pfa3 sin (6pft) ð19Þ
1
t= ð23Þ
g3 (t) =  10pfa5 sin (10pft) ð20Þ 4f

(a) (b)

(c) (d)

Figure 3. Smooth axial motion profiles: (a) displacement, (b) velocity, (c) acceleration, and (d) jerk.
6 Advances in Mechanical Engineering

Figure 4. Acceleration component profiles.

For the functions g2(t) and g3(t), the extreme point The corner length Li,c is then determined from equation
times are similarly determined to be at (see Figure 4), (26) as,
 
1 1 3 1 7 a0
t1 = , t2 = , t3 = , t4 = , t5 = , Li, c = ac + =½cos u1 + cos (u1 + u2 ) ð27Þ
20f 12f 20f 4f 20f 2f
5 9
t6 = , t7 = The projection of corner error ei on the x-axis is seen in
12f 20f
Figure 2 to be,
ð24Þ
 
p u2
Now due to symmetry of the functions about t = T/2, ei, x = ei cos + u1 + ð28Þ
2 2
to achieve the desired condition of a(t) ø 0, t 2 ½0, T 
for the synthesized motion profiles, the trajectory pat- The x coordinate of the point Pi is also seen to be,
tern parameters must be calculated such that the follow-
ing inequality conditions are satisfies, xi = Li, c cos u1 ð29Þ

a(ti ) ø 0, i = 1, 2, 3, 4 ð25Þ From equation (5) at the time t = T/2, the x coordinate
of the point Pi,m is,
 
Contour error constraints. To control machining accuracy, T a0 T a1 sin (pfT)
xi, m = sx = ac + +
the chord error should be bounded. In Figure 2, the 2 2 2pf
transition schema is seen to be symmetric about the a3 sin (3pfT ) a5 sin (5pfT)
+ +
bisector of the angle \Pi1 Pi Pi + 1 since the initial and 6pf 10pf
final tangential velocities, accelerations, and jerks are a0 a1 a3 a5
the same. The distance between the points Pi and Pi,m = ac + +  + ð30Þ
4f 2pf 6pf 10pf
is therefore the contour error maximum ei. The com-
puting process of the maximum is as followed: As can be seen in Figure 2, the x-axis contour error ei,x
For the corner motion displacement during the is,
period T, from Figure 2, the corner length Li,c and the
displacement sx(T), equation (5), are expressed as, ei, x = xi, m  xi ð31Þ
Substitute equations (27)–(30) into equation (31), the
Li, c ½cos u1 + cos (u1 + u2 ) = sx (T)
maximum chord error ei is obtained as,
a1 sin (2pfT) a3 sin (6pfT)
= ac + a0 T + + 
a0 a1 a3 a5

2pf 6pf ei = ac r1 + r2 +  + =r3 ð32Þ
ð26Þ 4f 2pf 6pf 10pf
a5 sin (10pfT) a0
+ = ac +
10pf 2f where
Li et al. 7

cos (u1 + u2 ) with sinu1, sin(u1 + u2) and sin(p/2 + u1 + u2/2),

r1 = ð33Þ
cos (u1 + u2 ) + cos u1 respectively, in equations (6)–(41).
cos (u1 + u2 )  cos u1 To synchronize the corner motion, the x-axis and y-
r2 = ð34Þ axis motion trajectories should be constructed with the
cos (u1 + u2 ) + cos u1
  same corner velocity vi,c. The process can be as follows:
p u2
r3 = cos + u1 + ð35Þ
2 2 Step 1. Assume that vi,c,x, vi,c,y are corner velocities
computed by the described x-axis and y-axis trajec-
To control the machining accuracy, we should guaran- tory synthesis process, respectively. Let vi,c be the
tee 0 ł ei ł e. smaller one of vi,c,x and vi,c,y as,

Trajectory parameters calculation. To calculate the trajec-  

tory parameters for the x-axis motion profiles, the ini- vi, c = min vi, c, x , vi, c, y ð42Þ
tial corner velocity vi,c is set at its desired value. From
equations (14) and (15), the trajectory parameter a0 is
Step 2. If vi,c,x is the minimum, use the process
then derived as,
described in the Section 2.2 and recalculate the
a0 = vi, c ½cos u1 + cos (u1 + u2 )=2 ð36Þ y-axis trajectory parameters with vi,c. If vi,c,y is the
minimum, similarly recalculate the x-axis trajectory
Substitute equations (13) and (36) into equation (27), parameters with vi,c.
the corner velocity vi,c in terms of the corner length Li,c Step 3. By substituting the calculated trajectory
is given as, parameters in equations (2)–(5), the synthesized tra-
jectories are obtained. As can be seen, the trajectory
vi, c = 4fLi, c ð37Þ velocity, acceleration, and jerk only contain sinusoi-
dal functions with the selected fundamental fre-
Now let Li,c be defined as the shorter of half the lengths
quency and its first two odd functions. As a result,
of the line-segments Pi21Pi and PiPi + 1, that is, li and
by the proper selection of the fundamental fre-
li + 1, respectively, as,
quency, excitation of the natural modes of vibration

li li + 1
 of the system is avoided.
Li, c = min , ð38Þ
2 2
Simulation validations
Substituting equation (38) into equation (37), the initial
velocity vi,c is then calculated. In this section, to demonstrate the effectiveness and
It is noted that the boundary conditions, equations performance of the proposed corner smoothing algo-
(13)–(16); motion profile constraints, equation (25); rithm, the results of performed simulations on a typical
and the contour error constrain, equation (32), are all selected system is presented and compared with results
linear functions of the trajectory parameters ac, a0, a1, that would be obtained with one of the commonly used
a3, and a5. Thus, their optimal values can be calculated trajectory synthesis algorithms. Both algorithms were
using linear programming. The linear programming implemented in Visual Studio 2013 and MATLAB
problem for determining optimal trajectory parameters 2015 was used for plotting the results. The following
to achieve minimum chord error ei becomes, typical settings were used: a fundamental frequency of
f = 16 Hz, a sampling period of T = 2 ms, a chord
min (ei  e) ð39Þ error limit of e = 0.1 mm, a programmed feedrate of
8 F = 100 mm/s, acceleration limits of the drive axes
ac = 0
Ax_max = Ay_max = 2000 mm/s2, and jerk limits of
>
>
a0 + a1 + a3 + a5 = vi, c cosu1
<
subject to : the drive axes Jx_max = Jy_max = 100,000 mm/s3.
a  a1  a3  a5 = vi, c cos(u1 + u2 )
: 0
>
>
a1  9a3  25a5 = 0
ð40Þ Case one

a(ti ) ø 0, i = 1, 2, 3, 4 To evaluate the proposed algorithm, a jerk-limited
ð41Þ algorithm,13 which constructs the transition schema
e ø ei ø 0
and jerk-limited axial kinematic profiles in one step, is
If the motion is in a deceleration phase, then the accel- used as the comparison algorithm. As shown in Figure
eration constraint becomes a(t) ł 0, t 2 ½0, T . For the 5, a simple pattern with one acute angle is indicated as
y-axis, its trajectory parameters are computed by repla- the desired tool path. The synthesized transition curves
cing cosu1, cos(u1 + u2), and cos(p/2 + u1 + u2/2) by applying the two algorithms are plotted. The cycle
8 Advances in Mechanical Engineering

speed motions would in general be beyond their

dynamic response and could excite natural modes of
vibration of the system. As expected, Figure 7, the velo-
city, acceleration, and jerk profiles of the synthesized
trajectory with the proposed algorithm only include the
fundamental frequency and its first two odd harmonics.
To explain how trajectories with high harmonic
components affect vibration of the machine tool, an
undamped, 1-DOF, linear spring-mass system modeling
the behavior of a vibration isolation system shown in
Figure 8(a) is selected. Let the spring have a spring rate
and free length of k and l, respectively. The mass of the
block is m. The displacements of the mass and the sys-
tem from their initial position are u and d, respectively.
From the Newton’s second law, the system equation of
motion is seen to be,

Figure 5. Tool path and synthesized trajectories. mu00 =  k(u  d  l) ð43Þ

Let z = u  d  l, z00 = u00  d 00 , and set the natural fre-

time with the proposed algorithm was calculated to be
quency of the spring-mass system as
0.0313 s, which is 9.5% faster than the jerk-limited algo-
rithm of 0.0346 s. Figure 6(a) shows that neither algo- rffiffiffiffi
1 k
rithm exceeds the specified chord error limit e. From fn = = 144Hz ð44Þ
Figures 5 and 6(b) to (f), we can see that the tangential 2p m
and axial displacement and velocity profiles are all Then, equation (43) can be rewritten as,
smooth. From Figure 6(g) to (i), it is observed that the
tangential acceleration profiles of both algorithms are z00 =  ð2pfn Þ2 z  d 00 ð45Þ
smooth, while the axial acceleration profiles of the com-
parison algorithm only have G0 continuity. In addition, Now as an example, take the y-axis motion profiles cal-
since the comparison algorithm constructs axial kine- culated using the aforementioned two algorithms as
matic profiles under axial acceleration and jerk limits, input to the dynamic system described by equation (45).
its jerk profiles exhibit sudden changes as shown in The input motion d(t) is then y-axis acceleration profile,
Figure 6(j) to (l). Noting that since such unsmooth that is, d 00 = ay (t).
acceleration and jerk profiles cannot be realized in the Substituting d 00 = ay (t) into equation (45) and sol-
feed drive systems, the synthesized trajectories would ving the resulting second order ordinary differential
most likely excite natural modes of vibration of the sys- equation, the vibrational motion of the mass m of the
tem. The proposed algorithm, however, has smooth spring-mass system of Figure 8(a) is determined.
tangential and axial velocity, acceleration, and jerk pro- As can be seen in Figure 8(b), the displacement z,
files, as shown in Figure 6, which the feed drive system which indicated the tool displacement, of the proposed
should be capable of realizing without exciting natural algorithm is smoother than that of the jerk-limited
modes of vibration of the system. The proposed algo- algorithm. In addition, when the y-axis motion ends,
rithm should, therefore, achieve significantly better the mass m comes to a stop with the proposed algo-
machining quality. rithm, while the mass of the jerk-limited algorithm con-
To evaluate the proposed algorithm further, fre- tinues to vibrate with its natural mode. It is noted that
quency spectrum distribution of the synthesized axial the trajectories synthesized using the jerk-limited algo-
kinematic profiles is studied using Fourier integration. rithm contain a significant number of high harmonic
In the plots of Figure 7, it can be seen that the x-axis content, which would in general cause excitation of the
and y-axis displacements, velocities, accelerations, and system modes of vibration irrespective of their frequen-
jerks of the trajectories synthesized using the aforemen- cies. The synthesized trajectories of the proposed algo-
tioned jerk-limited algorithm contain a considerable rithm, however, only contain the selected fundamental
number of high harmonic components, that is, harmo- frequency and two of its odd harmonics in its velocity,
nics with frequencies above the second odd harmonic acceleration, and jerk terms (and even in their higher
of the fundamental frequency of the trajectory pattern. order derivatives), which would minimize the number
As a result, the related actuating forces/torques require of high harmonic components in the required actuation
producing high frequency components, which for high forces/torques. The fundamental frequency of the
Li et al. 9

(a) (b)

(c)

(d) (e)

(f)

Figure 6. Continued
10 Advances in Mechanical Engineering

(g) (h)

(i)

(j) (k)

(l)

Figure 6. Comparison of the kinematic profiles for the Case one: (a) contour error, (b) x-axis displacement, (c) y-axis
displacement, (d) tangential velocity, (e) x-axis velocity, (f) y-axis velocity, (g) tangential acceleration, (h) x-axis acceleration, (i) y-axis
acceleration, (j) tangential jerk, (k) x-axis jerk, and (l) y-axis jerk.
Li et al. 11

(a) (b) (c) (d)

(e) (f) (g) (h)

(i) (j) (k) (l)

(m) (n) (o) (p)

Figure 7. Comparison of the Fourier analysis for the Case one: (a) x-axis displacement of the jerk-limited algorithm, (b) x-axis
displacement of the proposed algorithm, (c) y-axis displacement of the jerk-limited algorithm, (d) y-axis displacement of the
proposed algorithm, (e) x-axis velocity of the jerk-limited algorithm, (f) x-axis velocity of the proposed algorithm, (g) y-axis velocity
of the jerk-limited algorithm, (h) y-axis velocity of the proposed algorithm, (i) x-axis acceleration of the jerk-limited algorithm, (j) x-
axis acceleration of the proposed algorithm, (k) y-axis acceleration of the jerk-limited algorithm, (l) y-axis acceleration of the
proposed algorithm, (m) x-axis jerk of the jerk-limited algorithm, (n) x-axis jerk of the proposed algorithm, (o) y-axis jerk of the jerk-
limited algorithm, and (p) y-axis jerk of the proposed algorithm.

trajectory pattern can also be selected to avoid natural Case two

modes of vibration of the system. As a result, the sys- In this Case, a polynomial algorithm,4 a jerk-limited
tem should be capable of operating with minimal vibra- algorithm,13 and a jounce-limited algorithm14 are
tion and achieve higher quality machining. implemented as the basis for comparison with the
12 Advances in Mechanical Engineering

(a) (b)

Figure 8. (a) Mass-spring system and (b) its vibrations.

(a)

(b)

Figure 9. (a) Trident-shaped tool path and (b) enlarged corners.

proposed Trajectory Pattern Method (TPM) based uses a jerk-limited velocity scheduling method to obtain
algorithm. The polynomial algorithm constructs the the transition schema and jerk-limited axial kinematic
transition schema with the quintic polynomial in arc profiles in one step in time domain. The jounce-limited
length domain, then achieves tangential kinematic pro- algorithm utilized a jounce-limited velocity scheduling
files by using a traditional tangential trigonometric method to construct smoother acceleration transition
velocity scheduling method. The jerk-limited algorithm profiles, which can reach G3 continuity. The proposed
Li et al. 13

(a)

(b) (c)

(d) (e)

(f)

Figure 10. Continued

14 Advances in Mechanical Engineering

(g)

Figure 10. Comparison of kinematic profiles for Case two: (a) tangential velocity, (b) x-axis velocity, (c) y-axis velocity, (d) x-axis
acceleration, (e) y-axis acceleration, (f) x-axis jerk, and (g) y-axis jerk.

algorithm adopts the TPM to get the transition schema contents are shown in Figure 11. As can be seen in the
and jerk-smooth axial kinematic profiles in one step in plots of Figure 11(a) to (c) and (e) to (g) for the polyno-
time domain as described in Section 2. mial, jerk-limited and jounce-limited algorithms,
The simulation conditions of this Case are the same respectively, the axial acceleration profiles of these
as those of the Case one. Figure 9(a) shows the trident- three algorithms contain a large number of high fre-
shaped pattern that is selected as the tool path. The quency components since their trajectories are synthe-
synthesized corner tool paths with the above three algo- sized by polynomial basis curves. The trajectories of
rithms are shown in Figure 9(b). the proposed algorithm, however, are constructed with
The comparison of the simulation results is demon- only three harmonic functions, a fundamental fre-
strated in the plots of Figure 10. Figure 10(a) shows quency sinusoidal function and its first two odd harmo-
that tangential velocities of the four algorithms, which nics, in the present simulations with a fundamental
are all smooth and do not exceed the programmed fee- frequency of 16 Hz and the first two odd harmonic fre-
drate F. As shown in Figure 10(b) and (c), axial velocity quencies of 48 and 80 Hz. The plots of (c) and (f) of
profiles of the jerk-limited, jounce-limited, and pro- Figure 11 also shows that the maximum axial accelera-
posed algorithms are smooth and similar, while those tion frequency for synthesized trajectory with the pro-
of the polynomial algorithm is not smooth at the cor- posed algorithm is 80 Hz, which should minimize the
ners. In addition, Figure 10(d) to (g) show that axial number of high harmonic components in the required
acceleration and jerk profiles of the polynomial algo- actuation forces/torques and minimize the excitation of
rithm exceed specified limitations and fluctuate fre- the system modes of vibrations. The proposed algo-
quently around corners since this algorithm does not rithm should, therefore, yield a significantly better
use an axial velocity planning method, but a traditional machining quality and flexibility.
tangential velocity scheduling method. In the case of The calculated computing and machining times and
jerk-limited algorithm, its axial acceleration profiles are contour errors for the polynomial, jerk-limited, jounce-
also not smooth and its axial jerk profiles exhibits sud- limited, and proposed algorithms are presented in
den changes at the beginning and end of acceleration or Table 1. The results show that the computing time cost
deceleration periods around the corners. These fluctua- for the polynomial algorithm is almost twice that of the
tions and sudden changes cannot be realized in the feed jerk-limited, jounce-limited, and proposed algorithms.
drive system, which makes the machining process prone It can also be seen that the machining time spent by the
to vibration. proposed algorithm is the shortest (1.7203 s), which is
As can be seen in the plots of Figure 10, in contrast to 4.7%, 1.8%, and 5.3% shorter than the jounce-limited
polynomial, jerk-limited, and jounce-limited algorithms, (1.8058 s), jerk-limited (1.7515 s), and polynomial
the proposed algorithm has smooth axial velocity, accel- (1.8165 s) algorithms, respectively. The reason for the
eration, and jerk profiles, which should make it achievable proposed algorithm providing the shortest machining
by the drive systems of the machinery system. time is due to the use of an axial velocity planning
Fourier analysis results of the axial acceleration pro- method and since no limit on the axial jerk maximum.
files of the four algorithms showing their frequency It can, therefore, be concluded that the proposed
Li et al. 15

(a) (b)

(c) (d)

(e) (f)

(g) (h)

Figure 11. Comparison of frequency distribution for Case two: (a) x-axis acceleration profile of polynomial algorithm, (b) x-axis
acceleration profile of jerk-limited algorithm, (c) x-axis acceleration profile of jounce-limited algorithm, (d) x-axis acceleration profile
of proposed algorithm, (e) y-axis acceleration profile of polynomial algorithm, (f) y-axis acceleration profile of jerk-limited algorithm,
(g) y-axis acceleration profile of jounce-limited algorithm, and (h) y-axis acceleration profile of proposed algorithm.
16 Advances in Mechanical Engineering

Figure 12. Contour errors.

Table 1. Comparison of computing and machining times and contour error.

Algorithm Computing time (s) Machining time (s) Contour error (mm)
Max Mean

Polynomial 0.7836 1.8165 0.0870 0.0023

Jerk-limited 0.4253 1.7515 0.0749 0.0019
Jounce-limited 0.4505 1.8058 0.0747 0.0020
Proposed 0.4742 1.7203 0.0740 0.0017

algorithm is near time-optimal and satisfy the require- frequency sinusoidal function and its first two odd har-
ment of real-time machining, while yielding a signifi- monics in time domain.
cantly better machining quality. In comparison with previously published algorithms,
Figure 12 shows that none of the four algorithms the proposed algorithm has the following advantages:
exceeds the prescribed chord error e, and that the maxi- (I) the transition schema and axial kinematic profiles
mum always happens around the corners. As can be are constructed with a sinusoidal function with a
seen in Table 1, the contour error maximum for the selected fundamental frequency in one step, which
proposed algorithm is 0.0740 mm, which is 0.9%, 1.2%, reduces the smoothing process time, while avoiding
and 14.9% lower than those of the jounce-limited, jerk- excitation of the natural modes of vibration of the
limited and polynomial algorithms, respectively. In machinery system; (II) all velocity, acceleration, and
addition, compared with the other three algorithms, the jerk profiles are expressed by a sinusoidal function con-
average contour error for the proposed algorithm is sisting of a selected frequency harmonic and its first
also smaller, that is, it is 0.0017 mm as compared to two odd harmonics, which minimizes the number of
0.0023 mm for the polynomial, 0.0019 mm for the jerk- high frequency harmonic components in the required
limited, and 0.0020 mm for the jounce-limited algo- actuation forces/torques, reducing their dynamic
rithms. The proposed algorithm can, therefore, better response requirements, and by proper selection of the
satisfy machining accuracy requirements, while yielding trajectory fundamental frequency, the excitation of the
a significantly better machining quality. system modes of vibration is minimized; (III) the pro-
posed algorithm can realize smooth axial kinematic
control action under chord error and dynamic
Conclusions
response limitations of the axial drives, noting that for
In this paper, a corner smoothing algorithm using the high-speed and high-quality machining, smooth axial
Trajectory Pattern Method (TPM) for high-speed and kinematic profiles are more important than smooth
high-quality machining is proposed. In this algorithm, tangential kinematic profiles; and (IV) the proposed
the transition schema and axial kinematic profiles are algorithm is near time-optimal and can significantly
constructed simultaneously with a fundamental improve machining efficiency and quality.
Li et al. 17

Since the proposed algorithm is a general corner 2. Sun S, Yu D, Wang C, et al. A smooth tool path gen-
smoothing method, it could be extended to five-axis eration and real-time interpolation algorithm based on
machines and its machining efficiency and quality B-spline curves. Adv Mech Eng 2018; 10: 1–14.
would be improved more significantly. However, com- 3. Lv Q, Zhang H, Yang KM, et al. Study on the method
pared to the three-axis machine, there are mainly two of increasing turning velocity during CNC continuous
machining. Manuf Technol Mach Tool 2008; 7: 79–83.
challenges for a five-axis corner smooth algorithm.
4. Zhang XH, Yu D, Yang DS, et al. Corner curve transi-
First, the tool tip position and tool orientation need to tion interpolation algorithm for high speed machining of
be smoothed and synchronized. Second, due to the micro-line segment. J Mech Eng 2010; 46: 183–191.
non-linear relationship between tool orientation vectors 5. Ernesto CA and Farouki RT. High-speed cornering by
and rotary axis coordinates, the smooth motion of the CNC machines under prescribed bounds on axis accel-
rotary axis is not capable of being guaranteed by the erations and toolpath contour error. Int J Adv Manuf
smooth process of tool orientation vectors. Especially Technol 2012; 58: 327–338.
when tool orientation vectors approach singular points, 6. Du X, Huang J and Zhu LM. A locally optimal transi-
fierce vibrations of the rotary axis would cause the tion method with analytical calculation of transition
servo alarm, even damage machine parts. Efforts length for computer numerical control machining of
should be, therefore, made to obtain smooth motion of short line segments. Proc IMechE, Part B: J Engineering
Manufacture 2018; 232: 2409–2419.
rotary axes under the tool orientation deviation limita-
7. Shi J, Bi QZ, Wang YH, et al. Development of real-time
tion in the five-axis tool center point interpolation look-ahead methodology based on Quintic PH curve
processing. with G2 continuity for high-speed machining. Appl Mech
Future work will focus on two aspects. First, we are Mater 2013; 464: 258–264.
in the process of implementing the proposed algorithm 8. Shi J, Bi Q, Zhu L, et al. Corner rounding of linear five-
and more comparison algorithms on an open-source axis tool path by dual PH curves blending. Int J Mach
CNC system to analyze the performance further. Tools Manuf 2015; 88: 223–236.
Second, more limitation conditions will be considered 9. Hu Q, Chen Y, Jin X, et al. A real-time C3 continuous
in the proposed algorithm. We plan to publish the local corner smoothing and interpolation algorithm for
results shortly after. CNC machine tools. J Manuf Sci Eng 2019; 141:
041004.
10. Hu Q, Chen Y, Jin X, et al. A real-time C3 continuous
Declaration of conflicting interests tool path smoothing and interpolation algorithm for five-
The author(s) declared no potential conflicts of interest with axis machine tools. J Manuf Sci Eng 2020; 142: 1–43.
respect to the research, authorship, and/or publication of this 11. Fan W, Lee CH and Chen JH. A realtime curvature-
article. smooth interpolation scheme and motion planning for
CNC machining of short line segments. Int J Mach Tools
Manuf 2015; 96: 27–46.
Funding 12. Biagiotti L and Melchiorri C. FIR filters for online trajec-
The author(s) disclosed receipt of the following financial sup- tory planning with time- and frequency-domain specifica-
port for the research, authorship, and/or publication of this tions. Control Eng Pract 2012; 20: 1385–1399.
article: This work was supported by the Scientific Research 13. Tajima S and Sencer B. Kinematic corner smoothing for
Startup Foundation for High-level Talents of Nanjing high speed machine tools. Int J Mach Tools Manuf 2016;
Institute of Technology (Grant No. YKJ2019112), the 108: 27–43.
Natural Science Foundation of Jiangsu Provincial of China 14. Zhang L and Du J. Acceleration smoothing algorithm
(Grant No. BK20181024), and the Shuangchuang Doctoral based on jounce limited for corner motion in high-speed
Project of Jiangsu Province. machining. Int J Adv Manuf Technol 2018; 95: 1487–1504.
15. Rastegar J. (IEEE). A basis trajectory approach to the
inverse dynamics formulation of robot manipulators. In:
ORCID iDs IEEE international conference on systems engineering,
Hao Li https://orcid.org/0000-0003-3928-6870 Pittsburgh, PA, 9–11 August 1990, paper no.
Zhuwen Yan https://orcid.org/0000-0001-9794-315X ICSYSE1990-203158, pp.308–311. New York: IEEE.
16. Fardanesh B and Rastegar J. A new model based track-
ing controller for robot manipulators using the trajectory
References pattern inverse dynamics. In: IEEE international confer-
1. Hendrawan YM, Simba KR and Uchiyama N. Iterative ence on robotics and automation, Nice, France, 12–14
learning contouring controller based on trajectory gener- May 1992, paper no. ROBOT1992-219938, pp.2152–
ation with linearly interpolated contour error estimation 2157. New York: IEEE.
and Bézier reposition trajectory for computerized numer- 17. Rastegar J, Tu Q, Mattice M, et al. Trajectory pattern
ical control machine tool feed drive systems. Adv Mech method applied to a nonlinear flexible pointing system.
Eng 2019; 11: 1–15. In: American control conference, San Francisco, CA, 2–4
18 Advances in Mechanical Engineering

June 1993, paper no. ACC1993-4793148, pp.1622–1626. international design engineering technical conferences and
New York: IEEE. computers and information in engineering conference,
18. Rastegar J and Feng D. On the minimum harmonic tra- Washington, DC, 28–31 August 2011, paper no.
jectory pattern for point-to-point motion. In: ASME DETC2011-47724, pp.359–363. New York: ASME.