ORIGINAL ARTICLE

Chatter Identification of Face Milling

Operation via Time-Frequency and
Fourier Analysis

Ching-Chih Wei1, Meng-Kun Liu1,*, and Guo-Hua Huang1

1Department of Mechanical Engineering, National Taiwan University of Science and Technology, 106 Taipei city, TAIWAN
(Received 30 September 2015; Accepted 8 January 2016; Published on line 1 March 2016)
*Corresponding author: [email protected]
DOI: 10.5875/ausmt.v6i1.1018
Abstract: Chatter is generally defined as self-generated vibrations from the interaction between tool and workpiece.
It is the most critical vibration in machining operations and can decrease the surface quality and cause the premature
tool wear. Chatter creates erroneous vibrations on the workpiece surface. However, even though the spatial response
of chatter is still bounded, its frequency response would become unstably broadband, emitting high-pitched sounds.
The deterioration in both time and frequency domains implies that chatter is a route-to-chaos process. To capture the
time-frequency characteristics of chatter, instantaneous frequency is applied to extract the relevant force vibration of
machining chatter, and the energy ratio approach is used to develop a discrete stability lobe diagram of a face-milling
operation. The proposed method is validated by measuring the surface roughness of the workpiece.
Keywords: chatter, milling, Hilbert-Huang transform, Fourier transform, stability lobe diagram

testing and is represented in the frequency response

function (FRF). With this information, the stability lobe
Introduction
diagram could predict cutting conditions through a
Machining operations result in many kinds of mechanistic force model of the milling process. As shown
vibrations from the machine, tool or workpiece. The most in Fig. 1, the stable cutting region is defined by the spindle
critical such vibration is vibration generated from speed and DOC. Chatter occurs when the cutting
interaction between the tool and the workpiece. This so- parameters are chosen above the curves, otherwise the
called “chatter” results in tool wear and reduces cutting process is stable. The establishment of a stability
workpiece surface quality. Operators usually seek to lobe diagram assumes that the tool vibration can be
alleviate chatter by reducing cutting parameters such as approximated by a second order linear differential
feed rate, spindle speed and depth of cut (DOC), but such equation, where the FRF is used to determine the
measures also decrease the material removal rate (MRR) marginal DOC. The stable cutting region is seen as a “good”
and thus overall throughput. To increase both cutting alignment (i.e., in phase) between the tool and the
efficiency and fabrication quality, it is imperative to workpiece vibration, while the unstable cutting region
establish the relation between various cutting parameters means the vibrations are out of phase [2]. By drawing the
and the incidence of chatter. Currently, the stability lobe stability lobe diagram in advance, the operator could
diagram is widely used to identify a stable cutting region easily choose and design the optimal cutting parameters
in turning and milling operations. In the late 1950s, Tobias to maximize cutting efficiency. To validate the cutting
and Fishwick [1] explained the regeneration mechanism stability margin in the machining operation, Fast Fourier
and developed the two-dimensional stability lobe theory, Transform (FFT) is usually used to analyze the vibration
which relates the cutting stability to the function of signal [3][4]. Once chatter occurs, excessively high
spindle speed and DOC. The stability lobe diagram has frequencies would appear on the frequency spectrum.
been widely applied in numerical methods. Generally, However, previous studies have indicated that the
machine tool dynamics are identified through impact regenerative cutting process contains rich nonlinear

www.ausmt.org 25 auSMT Vol. 6 No. 1 (2016)

Copyright © 2016 International Journal of Automation and Smart Technology
 ORIGINAL ARTICLE Chatter Identification of Face Milling Operation via Time-Frequency and Fourier Analysis

characteristics, and the linear approximation used in the instantaneous frequency, giving both time and frequency
stability lobe diagram would inevitably misinterpret the resolution concurrently. EEMD has been used to
genuine features of nonlinear vibration [5]. Machining decompose vibration signals into multiple IMFs; FFT was
chatter is a route-to-chaos process with deterioration in then applied to compare the frequency of stable and
both the time and frequency domains. Despite the time unstable cutting to identify the chatter frequency [8].
response of the vibration being bounded, when chatter HHT was applied directly to obtain a time-frequency
occurs the frequency response would become unstably diagram of the vibration signal, and the marginal
broadband. Hence, to capture the bona fide features of spectrum was used to observe the frequency distribution
the machining status, it is necessary to monitor time and [9]. In general, previous studies measured the cutting
frequency response simultaneously. force along the X and Y direction of an end-milling process,
20
which is the main cause of tool vibration. A two-direction
force sensor is required to measure the cutting force.
This research seeks to measure the signal of the
15
cutting force in the Z (vertical) direction in a face-milling
operation to analyze the signal and identify chatter. To
DOC (mm)

10 capture the genuine characteristics of the machining

status, EEMD is applied to decompose the force signal into
5 multiple IMFs. Those related to the machining process will
be extracted and combined, while the irrelevant IMFs will
0
be ignored. Hence the EEMD process serves as an
0 0.5 1 1.5 2 2.5 3
Spindle Speed (rpm) 4
empirical filter to evaluate machining vibration and
x 10
Figure 1. Stability lobe diagram. discard environmental noise. FFT will be used to analyze
the combined IMFs and a method to calculate energy ratio
Recently, instantaneous frequency (IF) using the will be proposed to determine the occurrence of chatter.
Hilbert-Huang transform (HHT) has been used to Finally, a discrete stability lobe diagram is drawn and
investigate machining vibration. Ensemble empirical validated according to the surface roughness of the
mode decomposition (EEMD) applies the shifting process workpiece.
to decompose the nonlinear broadband response into
orthogonal sets of mono-component called intrinsic mode
function (IMF). EEMD can separate a complex time series Experimental Design and Setup
into simple oscillatory modes [6][7]. The Hilbert transform Figure 2(a) shows a 3-axis milling machine used for
is then applied to each IMF and resolves the the cutting experiment, with its specification summarized
Ching-Chih Wei received his B.Eng. and M.Eng. Degrees in Mechanical in Table 1. This cutting machine differs from a CNC
engineering from National Taiwan University of Science and Technology, in machine center in two key respects. First, because the
2012 and 2014 respectively. He has worked in Precision 5-axis CNC Machining
Center in National Taiwan University of Science and Technology for 2 years. torque of its spindle is limited, it only removes a small
His work includes conducting CNC training programs and managing CNC amount of chip each time so that a high spindle speed is
machines. He enrolled the Ph.D. program in National Taiwan University of
Science and Technology in 2014. His current research is mainly about 5-axis required to increase cutting capability. Second, this
machine and machining technique, including Kinematics, Cutting machine is designed to provide a low DOC and a high feed
optimization, Signal analysis, Computer programming, Mechanical design
and Mechatronics.
rate during high speed cutting (HSC), so it has low stiffness
and small inertia. The cutting experiment uses a 6 mm
Meng-Kun Liu received his bachelor degree from National Chiao Tung
University and master degree from National Taiwan University, both in
end-mill tool as shown in Fig. 2(b), and the cutting force
mechanical engineering. He obtained his PhD in Mechanical Engineering during face milling is measured using a Kistler 6-axis
from Texas A&M University in 2012. Since 2015, He joined the department dynamometer as shown in Fig. 2(c), with the specification
of Mechanical Engineering in National Taiwan University of Science and
Technology as an assistant professor. Dr. Liu’s research interests include time- summarized in Table 2. The sampling rate in the
frequency analysis of manufacturing process, sensorless fault diagnosis of experiment is 10000 Hz, which is 50 times of the
induction motors, robot force control and chaos control. He has co-
authored numerous journal articles, conference proceedings and a book maximum frequency at 12000 rpm. The signal is then
published by Wiley. analyzed using HHT and FFT. After the signal of the cutting
Guo-Hua Huang was graduated from National Taiwan University of Science
force in the Z direction (vertical) is analyzed, the results
and Technology and enrolled the M.Eng. Program in the same college in 2014. are compared against the surface roughness, which is
His electives are mainly regarding the signal processing and the structural
design. In addition, he has done the mechanical test of materials in Lab for
measured by surface metrology equipment as shown in
one years. He is currently researching into the application of the multi-axis Fig. 2(d), and the relation among cutting parameters,
rapid prototyping machine about using multiple materials for combine. cutting force and the surface quality can be established.

www.ausmt.org 26 auSMT Vol. 6 No. 1 (2016)

Copyright © 2016 International Journal of Automation and Smart Technology
 Ching-Chih Wei, Meng-Kun Liu, and Guo-Hua Huang

the DOC of 0.24 mm is not efficient for rough milling, thus

to determine optimal cutting efficiency and quality, the
DOC is designed from 2 mm to 12 mm with an interval of
2 mm in each step. Using the recommended cutting
parameters and cutting capability, the spindle speed is set
from 6000 rpm to 12000 rpm with an interval of 1000 rpm.
Because the tool is a 2-flute end-mill and the feed per
(a) (b) tooth is set as 0.01 mm/tooth, the feed rate can be
calculated by Eq. (2). The second experiment uses a flat
workpiece and the DOC is fixed at 2 mm. The feed rate is
changed from 400 mm/min to 1000 mm/min with an
interval of 200 mm/min, and the adjustment of spindle
speed is the same as in the first experiment.
MRR = ap•ae•F
MRR: Material Removal Rate
(c) (d)
ap: Axial depth of cut (1)
Figure 2. Experimental equipment: (a) CNC milling machine; (b) 6 mm
end-mill tool; (c) 6-axis dynamometer; (d) Surface metrology equipment. ae: Radial depth of cut
F: Feed rate
CNC milling machine
F = f •z •N
Travels X / Y / Z 300 / 350 / 150 mm F: Feed rate
Max. feed rate 10 m / min f: Feed per tooth (2)
z: Number of flutes
Max. spindle speed 12000 rpm N: Spindle speed
Repeatability 0.01 mm
Table 1. Specification of CNC milling machine. 6 mm end-mill tool (aluminum workpiece)

Spindle speed 7900 rpm

6-axis dynamometer
Feed rate 424 mm/min
Fx, Fy, Fz ±10 kN
Measuring range Material Removal Rate 305.28 mm3/min
Mx, My, Mz ±500 Nm Table 3. Cutting parameters recommendation.
Maximum sampling rate 125kS / s
Table 2. Specification of 6-axis dynamometer.

As previously mentioned, the three important

cutting parameters are feed rate, spindle speed and DOC.
The experiment is divided into two categories, one sets
the feed rate as a constant, and investigates the relation
Figure 3. Designed ladder workpiece and fixture.
between spindle speed and DOC; the other sets the DOC
as a constant, and investigates the relation between
spindle speed and feed rate. Meanwhile, the DOC can be Methods of Signal Analysis
classified as the axial depth of cut (ap) and radial depth of
cut (ae). In both experiments, the radial depth of cut is This research investigates chatter in a face milling
set as 50% of the tool radius, which is 3 mm, and the DOC operation from a time-frequency perspective. The force in
in this research always refers to the axial depth of cut. the vertical direction is measured and analyzed as shown
The first experiment uses a ladder-shaped in Fig. 4, and Fig. 5 presents the analysis flow. Both
workpiece (shown in Fig. 3) and a base fixture is designed experiments begin by measuring the force signal in the
to be fixed on the 6-axis dynamometer. The vertical direction. EEMD is used to decompose the original
recommended cutting parameters are summarized in signal into multiple independent IMFs. Then
Table 3. With the recommended MRR, the critical DOC is instantaneous frequency (IF) is calculated from each IMF
about 0.24 mm, which is calculated from Eq. (1). However, using the Hilbert transform. With the IF result, the

www.ausmt.org 27 auSMT Vol. 6 No. 1 (2016)

Copyright © 2016 International Journal of Automation and Smart Technology
 ORIGINAL ARTICLE Chatter Identification of Face Milling Operation via Time-Frequency and Fourier Analysis

dominant IMFs which best represent the occurrence of

chatter can be selected and combined to form a delegated
signal. Next, the dominant frequencies of this delegated
signal can be found by using Fast Fourier Transform (FFT),
and chatter can be identified by observing the energy
change of the dominant frequencies. Finally, the surface
roughness of the workpiece is measured to validate the
result and construct the discrete stability lobe diagram.
Figure 6. Process of Hilbert-Huang Transform (HHT).

A decomposition of the original time series signal

into n IMFs can be represented as:
n
x(t )   C (t )  R(t )
j 1
j (3)

where C_j (t) is the jth intrinsic mode function and the
residue R(t) can be either the mean trend or a constant.
Equation (3) also indicates that the original signal can be
synthesized by IMFs and residue, so the completeness of
this decomposition can be proofed numerically. The ratio
of each IMF energy compared to the total energy can be
calculated as:
T
 C (t ) dt
2
Figure 4. Measured force signal in z-axis. j
rj  0
T
(4)
 x(t ) dt
2
0
The energy ratio of each IMF can be used to determine the
dominate dominant frequency component within the
input signal x(t). To resolve the instantaneous frequency,
the Hilbert transform is applied to each IMF and
constructs an analytic signal Z_j (t).
Z j (t )  C j (t )  iH C j (t )  
 C j (t )  iy j (t )
i ( j (t ))
 a j (t )e (5)
Figure 5. Flow chart of the methods.
where the Hilbert transform is:
P  C j ( )
Methods of Signal Analysis 
y j (t )  H C j (t ) 
  t  
 d  (6)

Instantaneous frequency has been previously used P is the Cauchy principle value. Then
to investigate the nonlinear features of a time-delayed
a j (t )  C 2j (t )  y 2j (t ) and the instantaneous frequency
feedback oscillator, along with phase plots, bifurcation
diagrams, and Fourier spectrum [10]. Ensemble empirical of the jth IMF can be defined as:
mode decomposition (EEMD) applies the shifting process 1 d j (t ) (7)
fj 
to decompose the nonlinear broadband response into 2 dt
orthogonal set of mono-components called intrinsic mod Thus the frequency modulation of a complex signal can be
function (IMF). EEMD can separate a complex time series retrieved. The integral of the Hilbert spectrum throughout
into simple oscillatory modes. Each IMF has the same the entire time span can be defined as a marginal
number of extrema and zero crossings so that it is spectrum as follows [6]:
T
 H (, t)dt
inherently symmetrical with respect to the local mean
h( )  (8)
defined by the average of the upper and lower cubic spline 0
envelope without resorting to any pre-defined basis. The marginal spectrum offers a measure of the total
EEMD is needed to deal with data from nonstationary and energy contribution from each frequency value. It
nonlinear processes. The EEMD process is shown in Fig. 6 represents the accumulated energy distribution over the
and is elucidated in [7]. entire data span in a probabilistic sense.

www.ausmt.org 28 auSMT Vol. 6 No. 1 (2016)

Copyright © 2016 International Journal of Automation and Smart Technology
 Ching-Chih Wei, Meng-Kun Liu, and Guo-Hua Huang

Results and Discussion

Experiment I: Constant feed rate

First, the cutting force signal is decomposed into

multiple IMFs using EEMD, as shown in Fig. 7. Then each
IMF is converted to an instantaneous frequency using the
Hilbert transform, as shown in Fig. 8. It is shown that, even
in stable cutting conditions, the measured vertical force
contains rich nonlinear dynamics. The instantaneous
frequency of each IMF in Fig. 8 is not constant, but rather
shows intermittent variations, a characteristic of the
regenerative effect. To observe each instantaneous
frequency from a probability perspective, they are
transformed to a marginal spectrum, as shown in Fig. 9,
which shows the bandwidth of each IMF under different
DOCs. The spindle speed in this experiment ranges from
6000 rpm to 12000 rpm, with corresponding spindle
frequencies of 100 Hz to 200 Hz. Given that harmonics
may be as much as four times the spindle frequency, we
observed frequencies between 100Hz and 1000Hz. Figure
8 shows temporal intermittent variations and represents
the real frequency in each IMF, thus we can retain IMF4,
IMF5, and IMF6 and omit the irrelevant IMFs.
When the spindle speed is set to 6000 rpm, the
instantaneous frequencies of IMF4, IMF5 and IMF6 are
Figure 7. IMF signal (spindle speed is 6000 rpm and DOC is 2 mm).
transformed into a marginal spectrum as shown in Fig. 9.
With the DOC increment, the bandwidth of each marginal
spectrum becomes broader and the spectrum peak, i.e.,
the frequency with the highest probability of occurring,
shifts to the right, meaning a higher overall frequency.
Take IMF 6 in Fig. 9 for example: When the DOC increases,
the peak frequency is still obvious but it shifts from 100Hz
to 150Hz and the bandwidth widens. When the spindle
speed is set to 12000 rpm, the change of marginal
spectrum has the same tendency as the spindle speed at
6000 rpm. As shown in Fig. 10, the dominant frequency at
200 Hz is also obvious in IMF6 when the DOC is 2mm, but
the bandwidth is wider with the larger DOC. As a result,
the marginal spectrum of the specific IMFs can reflect
subtle changes to the machining status.
To design a quantifiable index to identify machining
chatter, the signals of IMF4, IMF5, and IMF6 are combined
and FFT is applied to extract the dominate frequencies, as
shown in Fig. 11. The frequency of 200 Hz is found to be
the dominant frequency caused by spindle speed, 400 Hz
is the tool natural frequency in two-flute cutting, and 600
Hz and 800 Hz are respectively the three-time and four-
time frequencies caused by the spindle vibration. The
excessive frequencies on the Fourier spectrum increase
with the DOC. Moreover, based on experimental
observations, the force signal which has a broadband
distribution in frequency can be identified as the
occurrence of chatter, as shown in Fig. 11(c). Figure 8. Instantaneous frequency of each IMF (spindle speed is 6000
rpm and DOC is 2 mm).

www.ausmt.org 29 auSMT Vol. 6 No. 1 (2016)

Copyright © 2016 International Journal of Automation and Smart Technology
ORIGINAL ARTICLE Chatter Identification of Face Milling Operation via Time-Frequency and Fourier Analysis

Figure 9. Marginal spectrum (spindle speed is 6000 rpm).

Figure 10. Marginal spectrum (spindle speed is 12000 rpm).

www.ausmt.org 30 auSMT Vol. 6 No. 1 (2016)

Copyright © 2016 International Journal of Automation and Smart Technology
 Ching-Chih Wei, Meng-Kun Liu, and Guo-Hua Huang

A, B, and P(f) are measured based on the Fourier

spectrum of the combined signal of IMF4, IMF5 and IMF6
in Fig. 11, which refers to frequencies below 1000 Hz, so
the frequency in the Fourier spectrum is chosen in the
same range. The design of the energy ratio in Eq. (9) is
based on the assumption that chatter will produce
excessive frequencies, and the energy generated will
dilute the ratio of the dominant frequencies’ energy with
respect to the overall energy. Therefore a threshold can be
set to determine the drop in the energy ratio, indicating
the occurrence of chatter. To validate this assumption, the
(a)
surface roughness of the workpiece is measured to
compare the energy ratio to the surface quality. As a result,
the discrete stability lobe diagram is constructed based on
the proposed criteria. Because there the FFT results may
include noise, a threshold value is set to filter low-effect
frequencies whose energy is less than 1/20 of the energy
from the dominant frequency generated by the spindle
speed. After filtering the signal, the energy ratio of the
dominant frequencies under different cutting parameters
is developed, as shown in Table 4. Holding spindle speed
constant, the ratio decreases as the DOC increases. The
(b) surface roughness of the workpiece is then measured and
the results are shown in Table 5. From Tables 4 and 5, the
energy ratio and surface roughness can be presented
together in Fig. 12. It can be seen that the energy ratio
above 0.25 has a better surface roughness within 1 μm. By
setting the energy ratio 0.25 as the margin of chatter
occurrence, the discrete stability lobe diagram can be
constructed, as shown in Fig. 13. Figure 14 shows the
workpiece surface after cutting, which can be classified as
stable and unstable regions according to surface
roughness, and those regions are completely consistent
with the results of the discrete stability lobe diagram.
(c)
Figure 11. FFT of merged IMF (spindle speed is 12000 rpm) (a) DOC is Experiment II: Constant DOC
2mm; (b) DOC is 8 mm; (c) DOC is 12 mm.
The signal analysis process in experiment II is the
Although chatter can be observed from Fig. 11(c), same that in experiment I. First, the cutting force signal
chatter must still be determined quantitatively. The FFT is decomposed into multiple IMFs using EEMD, and then
results of the combined IMF show the magnitude of the each IMF is converted into a marginal spectrum. In Fig. 15,
main cutting frequencies is much higher than the others, the dominant frequency of IMF6 is about 133 Hz, which is
which means the frequencies of the main cutting signals created by a spindle speed of 8000 rpm. The bandwidth in
dominate the rest of the vibration signal. Figure 11 shows each IMF exhibits no obvious difference, and the peak
the energy ratio between two dominant frequencies frequency remains the same, meaning that the change in
(spindle frequency and tool dominant frequency) and the machining status is opaque. Likewise, the dominant
combined signal can be designed as: frequency of IMF6 in Fig. 16 is about 200Hz, due to the
( A  B) spindle speed of 12000 rpm. Among four different feed
Energy ratio  (9) rate conditions, the dominant frequency is much clearer

f 1000
P( f ) when the feed rate in 400 mm/min, indicating that this
f 0
where rate produces a small nonlinear vibration effect and
A: Magnitude of spindle frequency possibly the best surface quality. The marginal spectrum
B: Magnitude of tool natural frequency in other three conditions shows no obvious difference,
P(f): Magnitude of frequency f which implies similar cutting conditions.
www.ausmt.org 31 auSMT Vol. 6 No. 1 (2016)
Copyright © 2016 International Journal of Automation and Smart Technology
 ORIGINAL ARTICLE Chatter Identification of Face Milling Operation via Time-Frequency and Fourier Analysis

For further discussion, the signals of IMF4, IMF5 and

IMF6 are combined and processed using FFT analysis. The
FFT result for a spindle speed of 12000 rpm is shown in Fig.
17. As in the FFT analysis in experiment I, the dominant
frequency and tool natural frequency are easily identified.
As the feed rate increases, the abnormal vibration
becomes more prominent, especially at a feed rate of 800
mm/min.
Figure 18 shows the workpiece surface after cutting.
The region is divided into stable and unstable cutting
according to the feed rate and spindle speed. The energy
ratio is then constructed in Table 6 and the surface
roughness is recorded in Table 7. The relation between the
energy ratio and the surface roughness is drawn in Fig. 19.
However, the machine in this research has low stiffness
and may vibrate easily. Even though the energy ratio is less
than 0.1, the surface roughness is around 1 μm which is Gray area: energy ration greater than 0.25 (stable cutting
still acceptable. Hence the stability margin for experiment region)
II is opaque. But for most results, the tendency is that, as Table 4. Energy ratio of dominant frequency to other frequencies.

the energy ratio increases, the surface roughness

converges to the better region that is mostly less than 1
μm.

Figure 12. Relation of energy ratio and surface roughness.

Gray area: surface roughness smaller than 1μm (stable
cutting region)
Table 5. Surface roughness results.

Figure 13 Discrete stability lobe diagram.

Figure 14. Milling result of experiment I, (1) stable cutting region, (2)
unstable cutting region.

www.ausmt.org 32 auSMT Vol. 6 No. 1 (2016)

Copyright © 2016 International Journal of Automation and Smart Technology
 Ching-Chih Wei, Meng-Kun Liu, and Guo-Hua Huang

Figure 15. Marginal spectrum with spindle speed at 8000 rpm.

Figure 15. (cont.) Marginal spectrum with spindle speed at 8000 rpm.

www.ausmt.org 33 auSMT Vol. 6 No. 1 (2016)

Copyright © 2016 International Journal of Automation and Smart Technology
ORIGINAL ARTICLE Chatter Identification of Face Milling Operation via Time-Frequency and Fourier Analysis

Figure 16. Marginal spectrum with spindle speed at 12000 rpm.

Figure 16. (cont.) Marginal spectrum with spindle speed at 12000 rpm.

www.ausmt.org 34 auSMT Vol. 6 No. 1 (2016)

Copyright © 2016 International Journal of Automation and Smart Technology
 Ching-Chih Wei, Meng-Kun Liu, and Guo-Hua Huang

(a)
Figure 18. Milling result of experiment II.

Figure 19. Relation of energy ratio and surface roughness.

(b)

Table 6. Energy ratio of dominant frequency to other frequencies.

(c)

Table 7. Surface roughness results.

(d)
Figure 17. FFT of combined IMF with spindle speed is at 12000 rpm, (a)
Feed rate is 400 mm/min; (b) Feed rate is 600 mm/min; (c) Feed rate is Conclusion
800 mm/min; (d) Feed rate is 1000 mm/min.
A generic chatter identification approach is
proposed in this research. To capture the genuine time-
www.ausmt.org 35 auSMT Vol. 6 No. 1 (2016)
Copyright © 2016 International Journal of Automation and Smart Technology
 ORIGINAL ARTICLE Chatter Identification of Face Milling Operation via Time-Frequency and Fourier Analysis

frequency characteristics of chatter, the Hilbert-Huang Engineering Transactions of the ASME, vol. 126, pp.
transform is used as an empirical filter to extract the 459-466, 2004.
relevant machining force signal from a face-milling doi: 10.1115/1.1765139
machine. Different from previous studies, only the vertical [5] C. Steve Suh and M.-K. Liu, Control of Cutting
force is measured for analysis. The occurrence of chatter Vibration and Machining Instability: A Time-
is identified by the energy ratio of the dominant Frequency Approach for Precision, Micro and Nano
frequencies on the FFT spectrum. As a result, a discrete Machining. John Wiley & Sons, 2013.
stability lobe diagram is developed by setting a threshold [6] N.-E. Huang, Z. Shen, S.-R. Long, M.-C. Wu, H.-H. Shih,
for the energy ratio, and the surface roughness of the Q. Zheng, N.-C. Yen, C.-C. Tung, H.-H. Liu, “The
workpiece is measured to validate this approach. The empirical mode decomposition and the Hilbert
results suggest the surface roughness converges to a spectrum for nonlinear and non-stationary time
better quality within 1 μm as the energy ratio is increased. series analysis,” Proceedings of the Royal Society of
Therefore, the energy ratio proposed in this research can London A: Mathematical, Physical and Engineering
be used as a reference to identify superior cutting Sciences, vol. 454, no. 1971, pp. 903–995, 1998.
conditions. doi: 10.1098/rspa.1998.0193
[7] N.-E. Huang and S.-S. Shen. Hilbert-Huang transform
and its applications. World Scientific, 2005.
References
[8] H. Cao, K. Zhou, and X. Chen, "Chatter identification
[1] S. A. Tobias and W. Fishwick, "Theory of regenerative in end milling process based on EEMD and nonlinear
machine tool chatter," The Engineer, vol. 205, pp. dimensionless indicators," International Journal of
199-203, 1958. Machine Tools and Manufacture, vol. 92, pp. 52-59,
[2] T. L. Schmitz and K. Scott Smith, Machining 2015.
Dynamics: Frequency Response to Improved doi: 10.1016/j.ijmachtools.2015.03.002
Productivity. Springer Science & Business Media, [9] W. Peng, Z. Hu, L. Yuan, and P. Zhu, "Chatter
2008. identification using HHT for boring process," in
[3] Y. Altintas, G. Stepan, D. Merdol, and Z. Dombovari, Proceedings of SPIE - The International Society for
"Chatter stability of milling in frequency and Optical Engineering, Beijing, China, 2013, p.904316.
discrete time domain," CIRP Journal of doi: 10.1117/12.2037988
Manufacturing Science and Technology, vol. 1, pp. [10] C.-S. Suh and B. Yang, “On the nonlinear features of
35-44, 2008. time-delayed feedback oscillators,”
doi: 10.1016/j.cirpj.2008.06.003 Communications in Nonlinear Science and
[4] S. D. Merdol and Y. Altintas, "Multi frequency Numerical Simulation, Vol. 9, No. 5, pp. 515-529,
solution of chatter stability for low immersion 2004.
milling," Journal of Manufacturing Science and doi: 10.1016/S1007-5704(03)00008-X

www.ausmt.org 36 auSMT Vol. 6 No. 1 (2016)

Copyright © 2016 International Journal of Automation and Smart Technology