Precision Engineering 59 (2019) 1–17

Contents lists available at ScienceDirect

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

A lightweight interpolation algorithm for short-segmented machining tool T

paths to realize vibration avoidance, high accuracy, and short machining
time
Takehiro Hayasakaa,∗, Kazuaki Minouraa, Kousuke Ishizakia, Eiji Shamotoa, Burak Sencerb
a
Graduate School of Engineering, Nagoya University, Furo-cho, Chikusa-ku, Nagoya, 464-8603, Japan
b
College of Engineering, Oregon State University, 219A Dearborn Hall, Corvallis, Oregon, 97331, USA

ARTICLE INFO ABSTRACT

Keywords: A computationally efficient FIR-filtering based path-smoothing algorithm which simultaneously realizes vibra-
Trajectory generation tion avoidance, high accuracy, and short machining time is proposed in this paper. Unlike the case of long G-line
Short-segmented G-line blocks blocks where only the adjacent blocks affect the cornering error of a specific corner, more than two blocks affect
FIR filter the error in the case of short-segmented blocks. To satisfy the tolerance error, point-to-point (P2P) technique can
Cornering error
be applied, but its machining time will be excessively elongated due to full stops at each corner. Alternatively,
Block splitting
motions with the delay times of FIR filters fully-overlapped, which are available on commercially-installed NC
systems, can realize short machining time, but they cannot satisfy the tolerance error due to the filtered tra-
jectories accompanies by high speed. Other methods such as spline fitting may satisfy the tolerance error and
realize short machining time, but they will allow vibration of the machine tool structure since these motions are
not allowed to be filtered for satisfaction of the tolerance. Therefore, no method exists which realizes vibration
avoidance, high accuracy, and short machining time all at the same time. For the first time in the literature, a
method is proposed which realizes all of the above requirements. The proposed algorithm bases on a kinematic
smoothing scheme where no spline-fitting based geometric smoothing is required, and the blended path geo-
metry is only controlled by optimizing the feedrate (speed) profiles along a span of short G01 and G02/G03
moves. FIR filtering is applied to avoid the inertial excitation of the machine tool structure, and a novel “block
splitting” method is proposed to keep elongation time of the G-line blocks the minimum. The effectiveness of the
proposed method is validated through a series of experiments by comparison with conventional methods.

1. Introduction by the NC unit of the machine.

G01 commands are interpolated to generate point-to-point (P2P)
Numerical control (NC) units have become a key component in motion in which the tool motion is linearly interpolated along the linear
determining accuracy and productivity of modern CNC machine tools. path, and it undergoes a full stop at the end. P2P linear interpolation is
Amongst other secondary operational functions, the most critical task of mainly used during drilling and tapping processes, or for secondary
a NC unit is to interpret the reference tool-path from the user supplied operations such as tool-change or probing. During typical high-speed
G-code (CL-file), and interpolate coordinated, rapid and accurate multi- machining operation, G01 and G02/G03 commands are interpolated in
axis motion in real-time. Although some machine-specific G-codes are a non-stop fashion where the tool motion is never stopped at the end of
capable of describing complex machining tool-paths (geometries) in the segment. Instead, feed motion is blended along consecutive seg-
parametric spline [1,2], or NURBS [3] curve format, standard G-codes ments and non-stop interpolation is realized. Since original linear path
are programmed by using basic G01 (linear) or G02/G03 (circular) segments are only position (G 0 ) continuous, velocity (C1) and accel-
interpolation commands. Most low-cost CAM systems discretize circular eration (C 2 ) motion kinematic continuity can be achieved by blending
and curved geometries into smaller linear segments and generate G- the geometry and increasing its continuity in the first place. Typically,
codes only composed of G01 linear moves. Hence, generation of smooth this type of blending is called the “local corner smoothing” where two
and accurate motion along tool-paths programmed by short linear G01 consecutive linear segments are blended using tangent arcs [4,5],
and circular G02/G03 moves is a critical task, which must be performed parametric curves [6,7], or splines [8,9] locally. Since the original path

∗
Corresponding author.
E-mail address: [email protected] (T. Hayasaka).

https://doi.org/10.1016/j.precisioneng.2019.05.006
Received 7 May 2019; Accepted 21 May 2019
Available online 23 May 2019
0141-6359/ © 2019 Published by Elsevier Inc.
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

filter taps [10,11] to interpolate tool motion with jerk-limited kine-

matics [12]. FIR filters provide distinct advantages over geometry-
based smoothing. Firstly, they do not need spline fitting. They directly
smoothen the kinematic profiles of the motion in time domain, which
makes them computationally efficient. Furthermore, since they can be
applied as moving average filters in real-time [10], they can be im-
plemented easily even in low-cost micro-processors. Note that filtering-
based approaches still generate blending errors due to the low-pass
characteristics of the filter, which must be controlled to realize an ac-
curate motion [11,12].
Fig. 1. G-line command. Past research has already considered FIR filtering [10] with blend
error control [11,12] for tool-paths that are composed of long linear
G01 segments. For these types of tool-paths, local corner blend geo-
is altered, the resulting blending errors must be constrained to ensure metry can be controlled accurately and easily [12]. However, when the
final part tolerances [4–9]. Notice that controlling the blend geometry original tool-path is programmed with very short linear segments, the
and associated path errors is a computationally expensive task, which control of blending errors becomes complicated. This is due to the fact
must be completed in real-time for every G-line to avoid motion in- that more than a single corner must be blended at the same time, and
terruptions. The curvature of the blend geometry should be optimized the blend sections overlap with each other. Control of the geometric
so that rapid feed motion can be realized for better productivity [8]. blending error is computationally stringent. So far, machine tool lit-
Moreover, structural vibrations cannot be suppressed since these paths erature addressed this problem with “global path smoothing” [13]
cannot be filtered in order to meet the tolerance. Thus, generating the where a long spline is fitted while iteratively checking for the blending
fastest and the most accurate motion with minimum vibration and errors [14–16]. This approach suffers from several drawbacks. Firstly,
computational cost is the overachieving goal of modern NC system the inertial excitation of the machine tool structure occurs since FIR
manufacturers where NC units become a critical component in de- filtering is not used. Secondly, least-squares (LS)-based spline fitting [2]
termining productivity of modern high-speed machine tools. is inherently a non-linear operation. Control of the fitting errors re-
Recent machine tool literature focuses on alternative techniques quires several iterations, and error convergence may not be strictly in a
that base on dynamic filtering to blend linear tool-paths and generate linear fashion limiting its implementation in practice. Thirdly, after the
non-stop feed motion [10]. These filtering techniques mainly use FIR spline fitting is completed, a tangential feed (velocity) profile must be

Fig. 2. Difference between original and filtered trajectory profiles.

2
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

Fig. 3. Fourier transform results of acceleration before and after filtering.

planned in real-time to ensure that axis kinematic limits are respected machining time. The paper is organized as follows. Section 2 presents
[1,3,15]. Exceeding acceleration/torque limits of the drives results in fundamentals of FIR-filtering based interpolation and control of local
actuator saturation and jeopardizes final part quality and even the blending errors through feedrate optimization accompanied by “block
machine health. splitting”. Section 3 introduces the novel technique to control blending
For the first time, this paper proposes a computationally efficient errors along short-segmented tool-paths. Section 4 shows experimental
FIR-filtering based path-smoothing algorithm, which accurately blends validation on a commercial machine tool and validates the effectiveness
short-segmented tool-paths within user specified error tolerances. The of the proposed technique.
proposed algorithm bases on kinematic smoothing scheme [11,12,17]
where no spline-fitting based geometric smoothing is required. The 2. FIR-filtering based path interpolation and blending error
blended path geometry is only controlled by optimizing the feedrate control
(speed) profiles along a span of short G01 and G02/G03 moves. Thus,
the feed profiling and blending are merged and controlled by the tan- 2.1. FIR-filtering based linear point-to-point (P2P) interpolation
gential speed. FIR filtering is used to interpolate the position commands
for vibration avoidance. Therefore, the proposed algorithm simulta- FIR filters in the form of the following transfer function [10,12] in
neously satisfies vibration avoidance, high-accuracy, and short Laplace(s) domain

Fig. 5. Image of error function for the case of two long blocks.
Fig. 4. Cornering error during non-stop interpolation of consecutive G-lines.

3
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

Fig. 6. Block splitting approach to reduce machining time.

Fig. 8. Corner smoothing error trend in short-segmented tool-path.

Fig. 7. Non-stop motion generated along short-segmented tool-path.

4
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

Fig. 9. Example of short-segmented G-line blocks and affecting range of a

corner.

1 e sTd
F (s ) =
s (1)

are used to interpolate G01 and G02/03 path commands with smooth
kinematic profiles [12]. As shown in Eq. (1), a 1st order FIR filter
contains a delay element e sTd controlled by the delay time Td and a free
integrator (1/ s ). As a result, when a velocity pulse is convolved with a
FIR filter, it simply integrates the input pulse and generates a trape-
zoidal velocity profile.
Fig. 1 provides an illustrative example of this process. A L = 10 mm
long linear motion is commanded as a velocity pulse with a feedrate of
F = 10000 mm/min and for a duration of Tv = 60 msec. The velocity
pulse is convolved with the FIR filter (Eq. (1)) with a time constant of
Td = 25 msec, and the time resolution (unit time) is set to 1 msec As
shown in Fig. 2, the resultant feedrate (speed) profile is a trapezoid with
a pulse acceleration transient. Notice that the duration of the ramp is
equal to the FIR filter delay. Hence, the maximum acceleration Amax
demanded is simply controlled by the commanded velocity pulse am-
plitude and the time constant of the FIR filter, that is Amax = F / Td .
As mentioned above, the maximum acceleration is either controlled
by the filter time constant or the feedrate command. In practice, the
maximum feedrate (F ) is described by the machine tool component
specifications, and a pre-determined Td is necessary to limit the max-
imum acceleration. The filter time constant Td is selected to mitigate
unwanted vibrations [10,11]. Although statically very rigid, typical
machine tool structures suffer from limited dynamic stiffness due to
their weak structural components and exhibit various resonances.
Major contributors to those vibrations are the rocking and column
modes of the machine tool, which introduce lightly damped low fre-
quency resonances [18,19]. As the feed drives accelerate and decelerate Fig. 10. Flow chart of the proposed algorithm.
the workpiece-table during high acceleration, frequency spectrum of
inertial forces can excite those lightly damped resonances and induce
unwanted forced vibrations [11,18]. The filter delay is selected to avoid frequency spectrum of the FIR filter can be computed from Eq. (1) as
those resonances as follows. Notice that filter time constants for vi-
bration avoidance are selected first, and an additional filter is added if 1 e j Td A sin( Td/2)
F (s ) = F (s ) =
necessary for acceleration limitation. j
s=j (2)
Fig. 3 shows the frequency ( ) spectrum of the original and filtered
acceleration profiles. Since the original velocity profile is a pulse, its As shown in Fig. 3, the FIR-filtered trajectory generates notches at
acceleration becomes an impulse (see Fig. 2(c)), and hence it induces Td = 2k = 2k /Td where k = 1,2, …, . Hence, by simply
equal amount of excitation along infinite frequency range. The matching the time constant of the filter to the period of the vibration

5
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

Fig. 11. (a) Trajectory profile, (b) Step 0, (c) Step 1a, (d) Step 2a, and (d) Step 3a.

Td = 1/ fn , it becomes possible to avoid exciting resonances with given by 2 consecutive linear moves which form a corner with an ex-
minimum filter delay. Note that multiple filters need to be used to avoid ternal angle of . When the tool-path is interpolated at constant speed
excitation of multiple resonances, and the total filter time delay equals using FIR filtering, a corner smoothing error is generated. The largest
to the sum of the individual time delays, e.g. filters with time delays of deviation from the corner, extremum of the smoothing error, occurs in
T1 and T2 are used to avoid resonances at f1 and f2 , respectively, and the the middle of the cornering transition and denoted by 1 (see Fig. 4).
total time delay Td = T1 + T2 . As introduced by Sencer et al. [11], the cornering error 1 by a single
FIR filter can be computed analytically based on the feedrate F , cor-
2.2. Corner blending (smoothing) error control nering angle , and FIR filter time delay Td as

1 cos
Previous section introduced that FIR filters can be utilized to in- 1 = FTd
32 (3)
terpolate P2P motion and generate sharp corners. This section presents
how sharp corners can be rounded and a non-stop contouring motion Equation (3) shows that the cornering error has a linear relationship
could be interpolated. An example of a cornering error during non-stop with the feedrate F and time constant Td . Also notice that the filter time
interpolation of linear moves is illustrated in Fig. 4. The tool-path is constant Td is predetermined for vibration mitigation. Therefore, the

Fig. 12. (a) Trajectory profile, (b) Step 3a, (c) Step 1b, (d) Step 2b, and (d) Step 3b

6
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

Fig. 13. (a) Trajectory profile, (b) Step 3a, and (c) Step 1c.

cornering error can simply be constrained by altering the feedrate. If

the feedrate of the two blocks are all reduced by the same rate , the
reduced cornering error ( ) is linearly scaled as:
( )= 1 (4)
An image of a linear error function against the reduction rate is
shown in Fig. 5. Optimum opt to meet the tolerance error tol can simply
be found from Eq. (4) as
opt = tol/ 1 (5)
As presented above, the feedrate in the vicinity of the corner is
critical. By increasing the cornering speed, the corner blending error
increases linearly obeying Eq. (3). The feedrate within the range of
± Td/2 affect the targeted cornering error. A novel method called the
“block splitting” is proposed to reduce the total machining time. This
method is explained in Fig. 6. For convenience, the NC blocks of x and y
directions are expressed in the same figure.
Since only the feedrate within the range of ± Td/2 affect the tar-
geted cornering error, blocks that affect the cornering error are sepa-
rated at that point. In the example in Fig. 6, Block A is split to A1 and
A2, and Block B is split to B1 and B2. Then, the feedrates of A2 and B1 are
reduced by the optimal reduction rate opt to satisfy the cornering error Fig. 14. Image of additional control point at middle of arc and reduction of
tolerance, tol . Note that the feedrates of A1 and B2 are kept the same as feedrate.
the original profiles; thus, the feedrate reduction is kept within the
minimum range. The final result is shown in Fig. 6(c), and it can be
observed that the cornering error is the same as that of Fig. 6(b) while cases where more than two blocks have an effect on the targeted corner,
the total machining time is shorter than that of Fig. 6(b). namely Tv < Td/2 where Tv is the pulse length of a block.
Secondly, a more complex case with multiple short-segmented G- In the case of short blocks, the derivation of the cornering error is
line blocks is described, and the trajectory is shown in Fig. 7. The time more complicated than as presented in Eq. (3) since more than two
constant of the FIR filter is the same as the first case, but the G-line blocks affect the targeted corner. As a result, the cornering error be-
blocks are extremely short. Note that the term “short” is used to express comes a nonlinear function of the feed reduction rate, , and needs

7
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

reducing feedrate on a single G-line alters deviation of the interpolated

trajectory formed by more than two blocks. The general idea that the
feedrate determines the cornering error is still effective. Although there
is a need of repetitive calculation, where the optimum reduction rate
opt can be determined, ( opt ) = tol can be achieved. Note that the idea
of block splitting has been explained earlier for two long blocks, but it
can be expanded to multiple short blocks. In other words, only the
feedrates within the range of ± Td/2 affect the targeted cornering error,
and only the feedrates in that range need to be reduced. An image of a
nonlinear error function against the reduction rate is shown in Fig. 8,
which is the actual example of the error function of the targeted corner
in Fig. 7. For example, Newton-Raphson iterations can be used to de-
termine opt in a computationally efficient manner.
As described above, for one targeted corner affects more than two
blocks. This means that reducing the feedrate on a single G-line alters
cornering errors of not only both corners on its ends but also some other
preceding/succeeding corners. Therefore, the feedrate reduction and its
corresponding block splitting to control each targeted cornering error is
just necessary but not sufficient for global cornering error control. To
control all cornering errors, the feedrate is reduced by the minimum
reduction rate min( i ) of each split region in the proposed method.
The concepts of optimal feedrate reduction and block splitting are
explained using only a single filter, but they are also true for multiple
filters. For example, if there are three filters, which is a typical example
when suppressing two structural modes and an additional filter for
limitation of the maximum acceleration, with time constants of T1, T2 ,
and T3 , Td is set to T1 + T2 + T3 , and the same characteristics can be
observed as explained above.

Fig. 15. Experimental setup. 3. Accurate blending of short-segmented G-line blocks

A trajectory generation method is proposed to simultaneously rea-

lize the acceleration limitation, minimum vibration, high cornering
numerical iteration for the optimum value to be found. More precisely, accuracy, and short machining time. An example of a case for multiple
if the corners are well separated, only affects the adjacent two blocks short-segmented G-line blocks is shown in Fig. 9. As described in the
of the targeted corner (see Fig. 4). However, on a short-segmented tool- earlier discussion, each corner affects other corners within the range of
path, a half of an affecting position range Td/2 becomes longer than a G- ± Td/2 . The cornering errors are predicted and these ranges, e.g. Corner
line block Tv , and hence affects more than two blocks. Therefore, P3 in Fig. 9 is affected by the blocks from P0 to P7, are identified along

Fig. 16. Measured free vibration versus (a) time and (b) frequency.

8
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

with the reduction rates , and the optimum feedrates are planned. At Step 1, the cornering errors i are predicted for each corner. If all
Thereafter, blocks are split if required. The flow chart of the proposed cornering errors meet the tolerance tol , then all Steps end here. If not,
algorithm is shown in Fig. 10, and key components are explained as the algorithm moves to Step 2.
follows. At Step 2, if i > tol the optimum feed reduction rate i is calculated
At Step 0, the conditions (original G-line codes, tolerance error, and for each corner where i = tol , and it is memorized in the buffer along
time constants to suppress resonances of the machine tool) are input with the affecting position range psi to pei of that corner. After all is
into the system. calculated, the algorithm moves to Step 3.

Fig. 17. (a) Original tool-path overlaid with the interpolated trajectory, (b) feedrate profile, and (c) measured vibrations.

9
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

At Step 3, regions are made at the memorized psi ’s and pei ’s, and the
minimum feed reduction rates min( i ) are calculated at each region.
Then, the G-line blocks are split at the positions psi ’s and pei ’s where
min( i ) changes, and the feedrates are reduced by min( i ) . After ne-
cessary splitting is executed, the algorithm moves to Step 1.
Note that it is occasionally required to go through the loop of Step 1
to Step 3 again even after each cornering error meets the tolerance error
in the first loop. This problem occurs when the cornering direction
changes as shown in the example inFigs. 11–13. At the initial path after
the filtering in Step 1a shown in Fig. 11, the cornering errors at Corner
2 and Corner 3 are greater than the tolerance error. Therefore, Corner 2
and Corner 3 require the surrounding blocks to reduce their feedrates
by the reduction rates 2 and 3, respectively, which are calculated in
Step 2a. Due to these requirements, Block 1 and Block 3 are split to 1-1,
1-2, 1-3 and 3-1, 3-2, respectively, in Step 3a. Notice that the region
defined by Corner 3 and pe2 is not adopted since the min(ai ) for that
region is 3, i.e. 2 corresponding to pe2 is not minimum for that region.
However, after the feedrates are reduced, the cornering errors at Corner
3-2 and Corner 4 again exceeds the tolerance error as shown in Step 1b
of Fig. 12(c), which is due to the change in the cornering direction from
Corner 3 to Corner 4. Thus, these original corners require the feedrates
within the range of ± Td/2 to be reduced, and the surrounding feedrates Fig. 18. Original tool-path.
are reduced by 3 1 and 4 , and the corresponding Block 3-1 and Block
4 are split to 3-1-1, 3-1-2, 3-1-3 and 4-1, 4-2, 4-3, respectively, in Step
3b. After this step, all cornering errors meet the tolerance error of
400 μm as shown in Fig. 13, and all G-lines are finalized.
Note that this method is explained based on straight paths (G01
commands), but it is nearly the same with arc paths with the following
additional steps. In order to calculate the cornering error for the arc
path, an additional control point is made at the middle of the arc, which
is only used for calculation of the cornering error in Step 1. In addition,
if there is a requirement to reduce the filtered feed speeds of the G-line
block of the arc in Step 3, the feedrates of the whole arc are reduced
regardless of the relation between the arc length and the time constants
of the filters. This is simply because the circular path has a constant
curvature over the whole arc. An image of these additional steps is
shown in Fig. 14.
By using the proposed method, all of the requirements can be sa-
tisfied at once. In addition, when machining with back-and-forth paths
with a pick feed, they need to be identically equal to each other, or
otherwise their difference will cause unwanted micro-bumps in the
machined area. The proposed method can also be adopted for these
cases, since it does not have a feed-direction dependency, i.e. the time
constants of the filters do not change with the feed direction and the
resulting motion paths of the machine tool does not change.
The proposed algorithm is presented in this section for off-line
calculation. However, it can also be executed in real-time where each
step in the flow chart of Fig. 9 can run in parallel with a delay of more
than the affecting position range (corresponding to Td/2 ). When all the
errors i meet the tolerance in the beginning range of 0 to Td in Step 1,
the G-lines in the beginning range of 0 to Td/2 can be finalized because Fig. 19. Interpolated trajectory for P2P technique.
they do not affect the errors in the following range. Then, the position
along the G-line that corresponds to Td/2 is updated to 0, and the par-
allel execution of the flows continues until the last G-line is finalized. A commercial 3-axis vertical machining center shown in Fig. 15 is
utilized. The time resolution (unit time) of its numerical control unit is
4. Experimental validation 2 msec In order to demonstrate the vibration mitigation functionality of
FIR-filtering based trajectory generation, a pair of flexible steel beams
Experiments are carried out to validate effectiveness of the proposed are placed on the x - y table, and accelerometers ( x : 339A30, y : 356A01,
technique in terms of accuracy, vibration, and speed (machining time). PCB Electronics Inc.) are used to measure the vibration at the tip of the

10
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

Fig. 20. (a) Feedrate profile and (b) measured vibration for P2P technique.

beams. Fig. 16 shows examples of the free response of the flexible interpolation technique which fully overlaps acceleration/deceleration
beams. As shown, the dominant resonances in the x and y directions are transients to blend sharp corners, (C) a spline interpolation technique
measured to be fx = 5.81 Hz and fy = 9.69 Hz, respectively. In order to [8], which blends sharp corners with micro Bezier splines, (D) the
avoid exciting those resonances, the time constants of the FIR filters are proposed technique without block splitting, which is introduced in
selected as Tx = 172 msec and Ty = 104 msec, respectively. The total Section 2.2, and finally (E) the final proposed technique which includes
filter delay becomes Td = Tx + Ty = 276 msec. Please note that in order the block splitting. The corner blending error tolerance is set to
to synchronize the motion, identical filters are applied to interpolate x tol=500 μm. Experimental results are given in Fig. 17, which shows the
and y axes motion. The use of 2 FIR filter ensures generation of jerk- original tool-path (G-lines) overlaid with the interpolated trajectory,
limited trajectory. the feedrate profile, and measured vibrations of the beams.
At first, two long G-line blocks with a single corner is interpolated to As observed from Fig. 17, the P2P technique allows machining of
demonstrate the functionality of the proposed interpolation technique sharp corners by ensuring a full stop at the block junction (corner). Due
and to compare it against the state of the art. Five different strategies to the full stop at the corner, the motion time is longer compared to (B)
are evaluated on this simple linear tool-path. The motion starts at and (D). The commercial non-stop high-speed interpolation technique
xs = 0 mm, ys = 0 mm, moves to the corner at x c = 150 mm, yc = 0 mm, follows an opposite strategy. It blends the single corner and generates a
and completes at x e = 150 mm, ye = 150 mm. The commanded feedrate non-stop motion. However, corner tolerance is not controlled. As a
for both segments (G-lines) is F = 10000 mm/min. The five strategies result, parts machined by this method can easily exceed desired pro-
are evaluated: (A) P2P interpolation technique directly on the com- duction tolerance and be scrapped. Please notice that since both tech-
mercial machine tool, (B) the commercial non-stop motion niques are FIR based, frequency spectrum of the acceleration

11
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

Fig. 21. (a) Original tool-path (G-lines) overlaid with the interpolated trajectory and (b) measured cornering errors for commercial non-stop high-speed interpolation
technique.

commands are controlled and hence the adverse effects of residual vi- feedrate reduction and block splitting is applied, the cornering error is
brations are avoided. The spline interpolation technique [8] can gen- confined within the tolerance, and the feedrate is only reduced around
erate non-stop and accurate motion around the corner. However, it the corner, which yields significantly reduced motion duration. This
cannot control frequency spectrum of the reference trajectory, and thus simple example showcases the potential of the proposed real-time in-
residual vibrations cannot be avoided. In fact, these severe vibrations terpolation technique to generate rapid and accurate feed motion.
can easily destroy the surface quality and cause scrap parts in actual Finally, the proposed technique is applied on a complex short-seg-
production environment. The proposed technique without the block mented tool-path shown in Fig. 18 and benchmarked against other
splitting functionality can generate non-stop and accurate motion while techniques. Please notice that the spline-based corner smoothing tech-
avoiding unwanted vibrations. However, total motion duration is nique [8] is excluded from this comparison since it induces large vi-
longer since feedrate of the entire block is reduced to meet the cor- brations making it not suitable for production environment. Further-
nering error tolerance. Lastly, when the proposed technique including more, its computational expense limits its real-time implementation on

12
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

Fig. 22. (a) Feedrate profile and (b) measured vibration for commercial non-stop high-speed interpolation technique.

commercial numerical control (NC) systems. The motion starts at the each single G-line. As a result, the total motion time becomes T = 61.928
origin x = y = 0 , and passes through the points A – I and returns back to sec.
the origin. The feedrate is set to F = 10000 mm/min. The corner tol- Next, the non-stop motion interpolation functionality, which fully
erance is kept unchanged from above. The resonant frequencies of the overlaps acceleration/deceleration motions by the adopted time con-
beams are remeasured before this experiment, and they are fx = 5.93 stants, is tested. As a result, the machine executes its own corner
Hz and fy = 9.63. The corresponding time constants of the FIR filters are smoothing algorithm and generates non-stop motion, which is suitable
Tx = 168 msec and Ty = 104 . Firstly, the P2P interpolation of the com- for high-speed machining. Fig. 21(a) shows the commercial non-stop
mercial machine tool is evaluated, and experimental results are shown motion generated along the tool-path. As shown in Fig. 22(a), sharp
in Figs. 19 and 20. As described above, time constants of the filters used corners are interpolated with non-stop motion and overall cycle time is
in the commercial NC system are tuned to eliminate unwanted forced reduced by 92% to T = 4.736 sec compared to the above result of P2P.
vibrations. In the P2P strategy, tool motion undergoes a full stop at the The resultant interpolation (cornering) errors are measured and de-
end of each segment, as shown in Fig. 20(a). Therefore, sharp corners picted in Fig. 21(b). As observed, in the commercial non-stop motion
can be machined, and no errors are observed in Fig. 19. Also shown in strategy, cornering errors could not be controlled. The NC system
Fig. 20(b), owing to the FIR filtering frequency spectrum, the flexible simply interpolates corner regions at user set speed in the G-line. Based
beams are not excited by the inertial forces generated during accel- on the sharpness of the corners (external angles) and commanded fee-
eration and deceleration sections. Hence, the beams and the table are drates, resultant cornering errors vary and may easily exceed user de-
moving as an almost rigid body together. However, P2P motion de- sired part tolerances prompting scrap parts, which is the key short-
mands large acceleration/deceleration to realize a full stop at the end of coming of commercial or state-of-the-art systems. Notice that, as it was

13
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

Fig. 23. (a) Original tool-path (G-lines) overlaid with the interpolated trajectory and (b) measured cornering errors for proposed technique with only feedrate
reduction.

for the case of the P2P technique, FIR filtering helps avoid excitement of the machine tries to accelerate to the desired feedrate with maximum
beam dynamics even during high-speed motion as shown in Fig. 22(b), possible acceleration within 2 msec The time resolution is set to 2 msec
but unwanted vibrations are large compared to those observed in the for all the following experiments. The G-line moves obtained by the
P2P technique because of the larger feedrates, i.e. larger accelerations proposed algorithm are interpolated and FIR-filtered with the time re-
and larger inertial forces. solution of 2 msec to generate the reference trajectories, and the tra-
Lastly, the proposed algorithm presented in Section 3 is tested on jectory positions and corresponding feedrates are then inputted to the
the same commercial machine. In order to test the proposed trajectory machine as short segmented G-lines, and air cutting experiments are
generation technique, the reference trajectory generator of the machine conducted. The commanded feedrates for all blocks are F =10000 mm/
is switched off. This is achieved by setting the reference trajectory pre- min, and the tolerance error is set to tol=500 μm. For comparison, the
filter time constant parameter in the NC unit to its lowest possible value proposed method without block splitting, or it could also be described
of 2 msec This ensures that when a reference tool-path is commanded, as the commercial non-stop motion technique accompanied with only

14
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

Fig. 24. (a) Feedrate profile and (b) measured vibration for proposed technique with only feedrate reduction.

feedrate reduction to meet the tolerance error, is also tested so that the T =25.866 s. Even though the proposed method is slower than the non-
effect of the block splitting can be easily shown. stop motion of the commercial machine tool, the cornering errors are
The results of the non-stop motion with only feedrate reduction and confined within the set contouring tolerance delivering both high-speed
with both feedrate reduction and block splitting are shown in Figs. 23 and high-accuracy multi-axis motion.
and 24 and Figs. 25 and 26, respectively. Focusing on the cornering The above off-line generation of the reference trajectory by the
errors shown in Figs. 23(b) and 25(b), it can be observed that the proposed algorithm is completed in 1.30 s, utilizing the developed
cornering errors by both methods are confined below the user specified Matlab code and a notebook computer with Intel Core i5-8250U run-
tolerance value tol , and the Corners A – I are controlled to be equal to ning at 1.60 GHz. Since the time of 1.30 s spent for the proposed al-
the tolerance since they are sharp corners compared to the others. gorithm is much less than the machining (motion) time of 10.738 s, it is
Considering about the vibration of the beams as shown in Figs. 24(b) considered that the developed algorithm can also be executed in real-
and 26(b), they cannot be observed in both methods due to the FIR time.
filtering. As with the total motion time shown in Figs. 24(a) and 26(a),
they are both much faster than the P2P motion but slower than the non- 5. Conclusions
stop motion generated by the commercial NC system of the machine
tool. Moreover, the final motion time of the proposed technique This research has presented a novel non-stop real-time interpolation
T =10.738 s is much faster than that without block splitting, i.e. scheme for short-segmented discrete tool-paths. It has shown that FIR-

15
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

Fig. 25. (a) Original tool-path (G-lines) overlaid with the interpolated trajectory and (b) measured cornering errors for proposed technique with both feedrate
reduction and block splitting.

filtering based interpolation can be employed to interpolate short-seg- advantage becomes even more pronounced along longer tool-paths. As
mented tool-paths within user defined error tolerances and generate compared to the commercial/state-of-the-art high-speed interpolation
rapid feed motion. The proposed method completely eliminates the technique, the proposed technique composed of the optimum feedrate
need of geometry-based smoothing strategies to generate a smooth reduction and the block splitting can constrain the blending errors for
motion. Instead, FIR filter characteristics and tangential feedrate is used short-segmented tool-paths with minimum elongation of machining
to control the blending errors. Experimental validation has been con- time. It can be concluded that the proposed algorithm for short-seg-
ducted on a commercial machine tool and shows that the proposed mented tool-paths realizes all of the practical requirements, i.e. vibra-
technique can generate accurate and rapid feed motion while avoiding tion avoidance, high accuracy, and short machining time, which could
unwanted vibrations. As compared to the P2P-based technique, the not be achieved by any conventional algorithms.
overall machining (motion) time is reduced by 83%. Notice that this

16
T. Hayasaka, et al. Precision Engineering 59 (2019) 1–17

Fig. 26. (a) Feedrate profile and (b) measured vibration for proposed technique with both feedrate reduction and block splitting.

References segments. Int J Mach Tool Manuf 2013;65:88–98.

[10] Biagiotti L, Melchiorri C. FIR filters for online trajectory planning with time-and
frequency-domain specifications. Contr Eng Pract 2012;20(12):1385–99.
[1] Erkorkmaz K, Altintas Y. Quintic spline interpolation with minimal feed fluctuation. [11] Sencer B, Ishizaki K, Shamoto E. High speed cornering strategy with confined
ASME 2003 international mechanical engineering congress and exposition. 2003. p. contour error and vibration suppression for CNC machine tools. CIRP Ann - Manuf
523–31. Technol 2015;64(1):369–72.
[2] Piegl L, Tiller W. The NURBS book. Springer Science & Business Media; 2012. [12] Tajima S, Sencer B, Shamoto E. Accurate interpolation of machining tool-paths
[3] Brecher C, Lange S, Merz M, Niehaus F, Wenzel C, Winterschladen M, Weck M. based on FIR filtering. Precis Eng 2018;52:332–44.
NURBS based ultra-precision free-form machining. CIRP Ann - Manuf Technol [13] Tajima S, Sencer B. Global tool-path smoothing for CNC machine tools with unin-
2006;55(1):547–50. terrupted acceleration. Int J Mach Tool Manuf 2017;121:81–95.
[4] Jouaneh MK, Wang Z, Dornfeld DA. Trajectory planning for coordinated motion of a [14] Zhao H, Zhu L, Ding H. A real-time look-ahead interpolation methodology with
robot and a positioning table. Part 1. Path specification. IEEE Trans Robot Autom curvature-continuous B-spline transition scheme for CNC machining of short line
1990;6:735–45. segments. Int J Mach Tool Manuf 2013;65:88–98.
[5] Jouaneh MK, Dornfeld DA, Tomizuka M. Trajectory planning for coordinated mo- [15] Zhang Q, Gao X, Li H, Zhao M. Minimum time corner transition algorithm with
tion of a robot and a positioning table. Part 2. Optimal trajectory specification. IEEE confined feedrate and axial acceleration for nc machining along linear tool path. Int
Trans Robot Autom 1990;6:746–59. J Adv Manuf Technol 2017;89(1–4):941–56.
[6] Erkorkmaz K, Yeung CH, Altintas Y. Virtual CNC system. Part II. High speed con- [16] Lin MT, Tsai MS, Yau HT. Development of a dynamics-based NURBS interpolator
touring application. Int J Mach Tool Manuf 2006;46(10):1124–38. with real-time look-ahead algorithm. Int J Mach Tool Manuf 2007;47(15):2246–62.
[7] Zhang L, You Y, He J, Yang X. The transition algorithm based on parametric spline [17] Tajima S, Sencer B. Kinematic corner smoothing for high speed machine tools. Int J
curve for high-speed machining of continuous short line segments. Int J Adv Manuf Mach Tool Manuf 2016;108:27–43.
Technol 2011;52:245–54. [18] Sencer B, Dumanli A, Yamada Y. Spline interpolation with optimal frequency
[8] Sencer B, Ishizaki K, Shamoto E. A curvature optimal sharp corner smoothing al- spectrum for vibration avoidance. CIRP Ann - Manuf Technol 2018;67(1):377–80.
gorithm for high-speed feed motion generation of NC systems along linear tool [19] Uriarte L, Zatarain M, Axinte D, Yagüe-Fabra J, Ihlenfeldt S, Eguia J, Olarra A.
paths. Int J Adv Manuf Technol 2015;76(9–12):1977–92. Machine tools for large parts. CIRP Ann - Manuf Technol 2013;62(2):731–50.
[9] Zhao H, Zhu L, Ding H. A real-time look-ahead interpolation methodology with
curvature-continuous B-spline transition scheme for CNC machining of short line

17