Content uploaded by P.Y. Sevilla-Camacho
Author content
All content in this area was uploaded by P.Y. Sevilla-Camacho on Aug 29, 2022
Content may be subject to copyright.
1 23
The International Journal of
Advanced Manufacturing Technology
ISSN 0268-3768
Volume 81
Combined 5-8
Int J Adv Manuf Technol (2015)
81:1187-1194
DOI 10.1007/s00170-015-7302-0
Tool failure detection method for high-
speed milling using vibration signal and
reconfigurable bandpass digital filtering
P.Y.Sevilla-Camacho, J.B.Robles-
Ocampo, J.Muñiz-Soria & F.Lee-
Orantes
1 23
Your article is protected by copyright and
all rights are held exclusively by Springer-
Verlag London. This e-offprint is for personal
use only and shall not be self-archived
in electronic repositories. If you wish to
self-archive your article, please use the
accepted manuscript version for posting on
your own website. You may further deposit
the accepted manuscript version in any
repository, provided it is only made publicly
available 12 months after official publication
or later and provided acknowledgement is
given to the original source of publication
and a link is inserted to the published article
on Springer's website. The link must be
accompanied by the following text: "The final
publication is available at link.springer.com”.
ORIGINAL ARTICLE
Tool failure detection method for high-speed milling using
vibration signal and reconfigurable bandpass digital filtering
P. Y. Sevilla-Camacho
1
&J. B. Robles-Ocampo
2
&J. Muñiz-Soria
2
&F. Lee-Orantes
1
Received: 7 November 2014 /Accepted: 11 May 2015 /Published online: 19 May 2015
#Springer-Verlag London 2015
Abstract This paper presents a monitoring method for on-
line detection and indication of the occurrence of a cutting
tool failure during high-speed face milling. The method con-
sists of processing of the vibration signal using a
reconfigurable infinite impulse response (IIR) bandpass digi-
tal filter and statistical techniques. The healthy tool threshold
and the filter passband are adjusted and configured based on
the cutting parameters that were set up during the machining
process. For this process, sets of filter coefficients are pre-
calculated for a number of defined insert passing frequencies
ranges. The method is verified on-line during machining tests
that are carried out at different tool failure levels and using
various cutting parameters. In all experimental tests, the meth-
od allows the tool condition to be detected and indicated cor-
rectly. The proposed method is therefore shown to be simple,
fast, computationally efficient, and reliable for the detection
and indication of the presence of several types of tool failures
for various cutting parameters, and the use of this method does
not require any modification of the machine tool structure.
Keywords Tool failure .High-speed machining .Vibrat ion
signal .Reconfigurable bandpass digital filter .Tool condition
monitoring
1 Introduction
In manufacturing industries, process automation is one of the
main issues of interest. High-speed machining (HSM) is one
example of a highly automated process. However, despite this
high level of automation, the quality of the machining of the
workpiece cannot be always guaranteed because of the cutting
tool conditions. When a tool is worn or broken and the prob-
lem is not detected during the machining process, it affects
both the production performance and the production costs.
Therefore, cutting tool condition monitoring (TCM) is a major
issue in HSM. However, there are very few dedicated TCM
research projects for HSM. Some of the monitoring algo-
rithms that have been developed and reported are based on
measurement of physical variables related to the cutting tool
conditions, such as the feed and spindle motor currents [1,2],
the cutting force [3,4], and the tool vibration [5–10]. This last
signal has been widely used because it offers better character-
istics than other tool monitoring variables, including (1) a
periodic shape that resembles the cutting force, (2) an ability
to provide sufficient information about the tool’scondition,
(3) robustness, (4) reliability, (5) applicability, (6) low cost,
and (7) easy measurement implementation without any need
for machine tool modifications.
However, these algorithms are intended for specific ma-
chining operations with fixed cutting conditions, and this
limits their industrial application because machining processes
are carried out using variable cutting parameters. Other limi-
tations of some of these systems include low processing
speeds and inappropriate analysis techniques. This is because
nonlinear and nonstationary high frequency signals are gener-
ated during HSM processes, and the proposed techniques re-
quire high computing power to provide a reliable indication of
the cutting tool condition. Also, the TCM analysis techniques
in some studies are based on fast Fourier transforms (FFTs);
*P. Y. Sevilla-Camacho
perlitas13@yahoo.com
1
Ingeniería Mecatrónica, Universidad Politécnica de Chiapas, Tuxtla
Gutiérrez, Chiapas 29082, México
2
Cuerpo Académico de Energía y Sustentabilidad, Universidad
Politécnica de Chiapas, Tuxtla Gutiérrez, Chiapas 29082, México
Int J Adv Manuf Technol (2015) 81:1187–1194
DOI 10.1007/s00170-015-7302-0
Author's personal copy
however, the effectiveness of this signal spectral analysis
method is limited, because FFTs cannot identify transient
and nonlinear signals. Finally, one other limitation of these
systems is the type of sensors used, which are sometimes
expensive or can be difficult to place on the machine tool.
This paper presents a reliable and reconfigurable monitoring
method for on-line detection and indication of the presence of
a tool failure during HSM. The vibration signal is obtained
using an accelerometer and is processed digitally using a
reconfigurable infinite impulse response (IIR) bandpass digi-
tal filter (4th order).
The use of digital filter for TCM also has been proposed in
[11]. However, this method presents some differences in com-
parison with our proposed method. In [11], only tool breakage
is detected. This detection is based on that, for a broken cutter,
the frequency strength of the same frequency components
increases. For this reason, a bandpass filter between the tool
passing frequency (TPF) and insert passing frequency (IPF)
minus tool passing frequency is applied. The overall peak to
peak difference of the filtered samples in spindle load for each
tool rotation is used to calculate the relative energy index
(REI) and indicated the tool condition. Unlike of the afore-
mentioned method, our proposed method detects differenttool
failure levels. The detection is based on any change in the tool
condition that generates changes in the frequency components
of the vibration signal. The major changes are located between
the IPF and the first harmonic. For that, a bandpass filter is
applied for these cutoff frequencies. Then, the filtered signal is
processed using statistical techniques such as sum of absolute
values and arithmetic mean value methods. The processed
vibration signal is compared with a preset healthy tool thresh-
old to detect and indicate through an alarm signal the presence
of a tool failure. The digital filter passband and the healthy
tool threshold are configured and set up as functions of the
cutting parameters.
2 Tool failure detection algorithm
The proposed detection algorithm consists of the detection of
vibration signal that is generated during the cutting process
(Fig. 1). This signal is detected by an accelerometer and ac-
quired using a data acquisition device (DAQ). Prior to the
acquisition of the data, the vibration signal is conditioned
using an antialiasing filter, which is implemented with a
−40 dB/decade low-pass Butterworth analog filter. This filter
limits the acceleration signal to a bandwidth of 2.4 kHz,
allowing a sampling frequency (f
s
) of 25 kHz. The sampled
signal is then processed using the reconfigurable IIR bandpass
digital filter (4th order). This filtering process provides an
output signal that only contains representative features of cut-
ting tool failures and thus enables failure detection. The digital
filter passband is reconfigured as a function of IPF, which
depends on the cutting parameters that are set up during the
machining process. For this process, sets of filter coefficients
are pre-calculated for a number of IPF ranges.
The healthy tool threshold is obtained by applying statisti-
cal techniques to the digitally filtered vibration signal. This
signal is acquired from a machining test that was carried out
using a healthy tool and using the same cutting parameters as
the monitored machining test.
The statistical techniques used to obtain the healthy tool
threshold value are the arithmetic mean value and sum of
absolute values techniques.
2.1 Reconfigurable IIR digital f iltering
The proposed algorithm is based on the 4th-order IIR
bandpass digital filtering of a vibration signal, which is ac-
quired using an accelerometer. The upper and lower cutoff
frequencies of the filter band are the IPF and its first harmonic,
respectively. The IPF is given by the following expression:
IPF ¼szð1Þ
where sis the spindle speed (rps) and zis the number of inserts
on the cutting tool.
Bandpass digital filtering is proposed because any change
in the tool condition generates changes in the frequency com-
ponents of the vibration signal, i.e., a milling process carried
out using a healthy cutting tool generates different frequencies
to those generated using a damaged tool. Frequencies between
the IPF and its harmonics are generated when the cutting tool
is defective, but this does not occur when the tool is healthy.
The major changes are located between the IPF and the first
harmonic [10]. Based on these vibration patterns, which de-
pend on the tool condition, bandpass filtering can be used to
extract important and representative features of cutting tool
failures and thus ease their detection. The use of digital filter-
ing as a signal processing technique is also considered because
it offers computational efficiency, high accuracy, high perfor-
mance, and simple, fast reconfiguration of the filter
parameters.
In this paper, an IIR digital filter is used because it has a
much better frequency response than a finite impulse response
(FIR) filter of the same order. Another advantage of IIR filters
over FIR filters is that IIR filters usually require fewer coeffi-
cients to execute similar filtering operations, and IIR filters are
also faster and require less memory space and computing
power. These factors enable these filters to detect the tool
conditions and indicate them on-line.
An IIR digital filter has the following equation:
ynðÞ¼
X
N
i¼0
aiyn−iðÞþ
X
N
j¼0
bjxn−jðÞ ð2Þ
1188 Int J Adv Manuf Technol (2015) 81:1187–1194
Author's personal copy
where a
i
and b
j
are the constant coefficients and Ndenotes the
filter order.
The constant coefficients are a set of constants that are
calculated and stored using any standard digital filter design
tool. This tool also determines the minimum order filter.
The constant coefficient calculus is designed for a specific
set of bandpass filter frequency specifications, such as the
sampling frequency, passband frequencies, and stopband fre-
quencies, while the minimum order filter calculus is determine
for a specific set of bandpass filter magnitude specifications,
such as passband ripple and stopband attenuation.
Therefore, when the signal is filtered for a specific pass-
band and magnitude specifications, Eq. (2) uses the required
set of stored coefficients and the calculated minimum order
filter.
In industrial applications, the cutting process is carried out
at various spindle speeds with numerous inserts on the tool.
These changes in the cutting parameters generate different
values for the IPF and its harmonics. Therefore, the digital
filter passband must be reconfigured as a function of the cut-
ting parameters that were set up during the machining process.
The digital filter reconfiguration is performed by changing the
set of filter coefficients. In this research, FDATool of MATL
AB was used to calculate and store the sets of filter coeffi-
cients for a fixed set of several passbands and stopband fre-
quencies. The values of passband and stopband frequencies
were defined in function of a number of predefined IPF ranges
and its first harmonic (see Table 1). For example, if the IPF
frequency is a value between 300 and 325 Hz, then its first
harmonic is a value between 600 and 650, respectively. There-
fore, the passband needs to be located between 300 and
650 Hz, which are the minimum and maximum frequency
values. In this case, the proposed passband was located from
400 to 500 Hz, whereas the stopband frequencies were 350
and 550 Hz.
Fig. 1 Detection algorithm
Table 1 Digital filter passbands and stopbands for predefined IPF
ranges
IPF range (Hz) Passband frequencies (Hz) Stopband frequencies (Hz)
Fpass1 Fpass2 Fstop1 Fstop2
300≤IPF ≤325 400 500 350 550
325< IPF≤350 425 550 375 600
350< IPF≤375 450 600 400 650
375< IPF≤400 475 650 425 700
400< IPF≤425 500 700 450 750
425< IPF≤450 525 750 475 800
450< IPF≤475 550 800 500 850
475< IPF≤500 575 850 525 900
500< IPF≤525 600 900 550 950
525< IPF≤550 625 950 575 1000
550< IPF≤575 650 1000 600 1050
575< IPF≤600 675 1050 625 1100
600< IPF≤625 700 1100 650 1150
625< IPF≤650 725 1150 675 1200
650< IPF≤675 750 1200 700 1250
675< IPF≤700 775 1250 725 1300
700< IPF≤725 800 1300 750 1350
725< IPF≤750 825 1350 775 1400
750< IPF≤775 850 1400 800 1450
775< IPF≤800 875 1450 825 1500
Int J Adv Manuf Technol (2015) 81:1187–1194 1189
Author's personal copy
The calculated filter coefficients are returned as row vectors
band aof length N+1.
For this example, the set of calculated filter coefficients is
showninTable2. This set of filter coefficients for the
reconfigurable IIR bandpass digital filter is the same for all
IPFs in that range.
The FDATool also calculated the minimum order for filters
with passband ripple of 1 dB and stopband attenuation of
3 dB. For this case, the calculated order was 4th.
2.2 Healthy tool threshold
In this algorithm, arithmetic mean value and sum of absolute
values of the digitally filtered vibration signal is taken as a
reference value to set up a healthy tool threshold.
The arithmetic mean value is applied to obtain a mathemat-
ical representation of the typical value of the sum of absolute
and to reduce the noise in the data.
u¼1
8 SPRðÞ
X
8 SPRðÞ
i¼1
yi
jj ð3Þ
The healthy tool threshold is determined by monitoring
eight complete spindle rotations (8 [SPR]) with the tool in
healthy cutting condition. The number of samples for a single
SPR is not a constant due to the sampling frequency (f
s
)isset
to 25 kHz and the spindle speed (s) is variable. The number of
SPR is obtained according to Eq. (4).
SPR ¼fs
s
¼25 kHz
s
ð4Þ
Finally, the healthy tool threshold is calculated by adding a
tolerance value to the arithmetic mean value obtained. This
tolerance was obtained experimentally.
T¼uþu¼2uð5Þ
2.3 Sum of absolute values of the filtered signal
The detection algorithm is based on the sum of absolute values
of the digitally filtered vibration signals [see Eq. (6)]. This
statistical technique is used to obtain a numerical parameter
to describe the tool condition.
The number of processed samples is for one complete SPR
[see Eq. (4)]. The vibration signal is acquired using the same
cutting parameters that were used for the preset of the healthy
tool threshold.
u¼X
SPR
i¼1
yi
jj ð6Þ
2.4 Decision-making
The decision-making is based on a comparison between a set
healthy tool threshold (see Eq. [5]) and the sum of absolute
values of the digitally filtered vibration signal [see Eq. (6)].
The comparison is made to indicate the tool condition. When
the sum of absolute values (u) is greater than the set healthy
tool threshold (T), an alarm signal is activated to indicate
cutting tool failure. Otherwise, the alarm signal remains inac-
tive. This indicator thus permits operator interventions during
the machining process.
3Experimentalsetup
To test the proposed method, a DM 4326 numerically con-
trolled, three-axis, high-speed vertical machining center is
used. The workpiece material used throughout the experi-
ments is aluminum alloy 6061-T6 in rectangular blocks. The
cutting tool is an indexable face mill with a diameter of 60 mm
and three triangular carbide inserts. The workpiece feed direc-
tion is along the positive X-axis.
The experimental setup is shown in Fig. 2. The vibration
signal, generated during the cutting process in the Xmachin-
ing directions, is detected using a biaxial ADXL321 acceler-
ometer with an acceleration range of ±18 g, which is fitted to
the vise jaws.
The signal is acquired using an NI USB-6215M series
multifunction DAQ device. The acquired signal is processed
and analyzed on-line using LABVIEW™. Prior to the acqui-
sition of the data, the vibration signal is conditioned using an
antialiasing filter.
A series of machining tests is carried out during the down
milling process with an axial depth of cut (a
p
) of 3 mm, a
radial depth of cut (a
e
) of 30 mm, and variable cutting speeds
(V
c
), feed rates (V
f
), and cutting tool conditions (see Table 2).
All the experiments were performed with dry inserts.
Figure 3shows the various condition levels of the cutting
inserts that were used for the tests. The insert in a healthy
cutting condition is shown in the left column, and the middle
and right columns show inserts with partial and severe wear,
respectively. The widths of the wear land were 0.5 mm for a
partial wear and 1 mm for a severe wear. In addition,
Tabl e 2 Set of calculated filter coefficients for an IPF between 300 and
325 Hz
i0 1234
a1−3.9089 5.7545 −3.7813 0.9358
b5.3229e−04 0 −0.0011 0 5.3229e−04
1190 Int J Adv Manuf Technol (2015) 81:1187–1194
Author's personal copy
machining experiments are performed with one insert missing.
These experiments with the missing insert are intended to
simulate very severe failure or breakage of the cutting tool.
4 Analysis of experimental data and results
Figure 4a shows the waveforms of the original vibration
signals, which were obtained from experimental tests car-
ried out under various tool conditions, using a cutting tool
with three inserts, a spindle speed of 117 rps, and a feed rate
of 73 mm/s. As shown in the figure, the waveforms are
composed of several frequency components that belong to
the various parameters that are part of the cutting process.
To verify the frequency components of the signals, the sig-
nals were processed by the continuous wavelet transform
(CWT). Figure 5shows the 2D contour plot of the CWT
time-frequency maps. As can be seen, only frequencies of
351, 702, and 1053 Hz are generated when the cutting tool
is healthy. These frequencies correspond to the IPF and its
harmonics, respectively. However, when the cutting tool
presents a failure, additional nonlinear and transient fre-
quencies are generated. These dominant peak frequencies
appear between the IPF and its first harmonic, which are
351 and 702 Hz, respectively.
It is important to mention that the CWT time-frequency
maps of all the tests (see Table 2) present the same behaviors
aforementioned for both healthy and failure tool. But it is
obvious for their respective IPF.
The presence of others frequency components, which do
not belong to the tool condition, makes it difficult to use a
conventional signal processing technique for tool failure
detection. To overcome this problem, a reconfigurable IIR
bandpass digital filter (4th order) was applied to the original
vibration signals to extract only the frequency components
that are representative of the cutting tool condition.
Figure 4b shows the waveforms of the vibration signals
when processed using the reconfigurable IIR bandpass digital
filter. This figure clearly shows that the waveform level ob-
tained when using a healthy cutting tool is smaller than the
Fig. 2 Experimental setup
Fig. 3 Condition levels of the
cutting inserts used for the
experimental tests
Int J Adv Manuf Technol (2015) 81:1187–1194 1191
Author's personal copy
levels obtained when using the damaged cutting tools. This
difference in the levels occurs because the digital filter only
passes signal frequencies within a specific frequency range
and rejects all other frequencies outside this range.
Fig. 4 Waveforms of original
and filtered vibration signals
acquired from the machining tests
performed at s=117 rps, V
f
=
73 mm/s, cutting tool with three
inserts, and different tool
conditions
Fig. 5 CWT time-frequency
maps of vibration signals from the
machining tests performed at s=
117 rps, V
f
=73 mm/s, and
different tool conditions: athree
healthy inserts, b2 healthy inserts
and 1 partially worn insert, c2
healthy inserts and 1 severely
worn insert, and d2healthy
inserts and 1 missing insert
1192 Int J Adv Manuf Technol (2015) 81:1187–1194
Author's personal copy
For this experimental test, the IPF is 351 Hz. Therefore, the
digital filter passband is set to 450–600Hz(seeTable1). If the
filter is applied to the vibration signal generated with healthy
cutting tool, the generated frequencies of 351, 702, and
1053 Hz are rejected. As a result, the waveform level of the
filtered signal is close to 0. However, if the filter is applied to
the vibration signal generated with a damaged cutting tool, the
frequencies between the range of 450 and 600 Hz are passed.
And as a result, a waveform level for the filtered signals is
generated.
Analysis of the digital filtered signals allows on-line detection
of the presence of tool failure in graphical form. To obtain a
numerical parameter that allows the cutting tool condition to be
indicated to the machining operators, the statistical technique of
the sum of absolute values was then applied to the digitally
filtered signals. Figure 6shows the final results obtained when
applying this technique to vibration signals acquired from the
machining tests performed at s=117 rps, V
f
=73mm/s,anddif-
ferent tool conditions. The first 14 results correspond to the vi-
bration signal from experimental tests carried out using a healthy
tool.Theresultsfromtests15to35correspondtotestscarried
out using a tool with one partially worn insert, while those from
tests36to56correspondtotestscarriedoutusingatoolwithone
severely worn insert, and finally those from tests 57 to 70 corre-
spond to tests carried out using a tool with one insert missing.
From Fig. 6, it is evident that all values obtained from the
tests using the damaged tool are higher than the set healthy
tool threshold. In contrast, all values obtained from the tests
using a healthy tool are smaller than the set healthy tool
threshold. As a result, an alarm could be activated to indicate
the occurrence of cutting tool failure.
Several experimental tests (see Tables 2and 3) using a
variety of cutting parameters were performed to validate the
algorithm. In all tests, similar results were obtained. This
clearly demonstrates that the proposed method indicates and
detects correctly the cutting tool condition. Also, this demon-
strates that the different spindle speeds and feed rates do not
interfere with the detection.
5Conclusions
This paper demonstrates that the proposed vibration signal
processing method, which is based on a reconfigurable IIR
Tabl e 3 Experimental test arrangement
Spindle speed
(rps)
Feed rate
(mm/s)
Tool condition
117 73 3 healthy inserts
117 73 2 healthy inserts and 1 partially worn
insert
117 73 2 healthy inserts and 1 severely worn
insert
117 73 2 healthy inserts and 1 missing insert
125 75 3 healthy inserts
125 75 2 healthy inserts and 1 partially worn
insert
125 75 2 healthy inserts and 1 severely worn
insert
125 75 2 healthy inserts and 1 missing insert
133 67 3 healthy inserts
133 67 2 healthy inserts and 1 partially worn
insert
133 67 2 healthy inserts and 1 severely worn
insert
133 67 2 healthy inserts and 1 missing insert
142 83 3 healthy inserts
142 83 2 healthy inserts and 1 partially worn
insert
142 83 2 healthy inserts and 1 severely worn
insert
142 83 2 healthy inserts and 1 missing insert
Fig. 6 Sum of absolute values of
the digitally filtered vibration
signals acquired from the
machining tests performed at s=
117 rps, V
f
=73 mm/s, and
different tool conditions
Int J Adv Manuf Technol (2015) 81:1187–1194 1193
Author's personal copy
bandpass digital filter (4th order) and statistical techniques,
allows the on-line detection and indication of the occurrence
of cutting tool failure. The bandpass digital filter is used to
extract an important representative feature related to cutting
tool failure. This feature is the presence of frequencies located
between the IPF and its harmonics.
The use of IIR bandpass digital filtering for failure detec-
tion offers several advantages, including easy implementation,
simple detection of the tool conditions, fast and simple recon-
figuration of the filter parameters, and rapid processing with
very little memory space and computing power required.
The sum of absolute values technique, when applied to the
digitally filtered vibration signal, generates a numerical pa-
rameter that is used to indicate the tool condition. When a
cutting tool fails, this parameter is higher than the set healthy
tool threshold.
An important element of the proposed method is that the
healthy tool threshold used for decision-making and the pass-
band of the digital filter are both adjusted and configured
based on the cutting parameters that were set up during the
machining process. Therefore, this method is suitable for in-
dustrial applications. In addition, the reconfiguration feature
of the proposed method means that it can be applied to differ-
ent machining processes where the occurrence of tool failure
generates frequencies in the range between the IPF and its
harmonics.
Another important conclusion is that the installation of the
accelerometer does not require any modification of the ma-
chine tool structure and does not interfere with the cutting
process. This sensor also reduces the cost of implementation
of the TCM method when compared with other methods.
All experimental results showed 100 % detection and indi-
cation of cutting tool failure. The results presented here dem-
onstrate that that the proposed method is a simple, efficient,
fast, reliable, and economical method for on-line detection and
indication of the presence of several types of tool failure with
various cutting parameters in high-speed face milling.
The algorithm has been tested using a personal computer as
the processor unit, but it can also be implemented with a
system-on-chip (SoC) approach to obtain an on-line, real-
time, and compact monitoring system.
References
1. Sevilla-Camacho PY, Herrera-Ruiz G, Robles-Ocampo JB,
Jáuregui-Correa JC (2011) Tool breakage detection in CNC high-
speed millingbased in feed-motor current signals. Int J Adv Manuf
Technol 53:1141–1148. doi:10.1007/s00170-010-2907-9
2. Rivero A, Lopez de Lacalle LN, Penalva ML (2008) Tool wear
detection in dry high-speed milling based upon the analysis of ma-
chine internal signals. Mechatronics 18:627–633. doi:10.1016/j.
mechatronics.2008.06.008
3. Zhang S, Li JF, Liu FS, Jiang F (2008) Tool wear in high-speed
milling of Ti-6Al-4V alloy. Key Eng Mater 375–376:435–439. doi:
10.4028/www.scientific.net/KEM.375-376.435
4. Zhang S, Li JF, Sun J, Jiang F (2010) Tool wear and cutting forces
variation in high-speed end-milling Ti-6 Al-4V alloy. Int J Adv
Manuf Technol 46:69–78. doi:10.1007/s00170-009-2077-9
5. Haber RE, Jiménez JE, Peres CR, Alique JR (2004) An investiga-
tion of tool-wear monitoring in a high-speed machining process.
Sensors Actuators 116:539–545. doi:10.1016/j.sna.2004.05.017
6. Zeng H, Thoe TB, Li X, Zhou J (2006) Multi-modal sensing for
machine health monitoring in high speed machining. 4th IEEE Int
Conf on Ind Informatics 1217–1222. doi: 10.1109/INDIN.2006.
275812
7. Dimla DE Sr (2002) The correlation of vibration signal features to
cutting tool wear in a metal turning operation. Int J Adv Manuf
Technol 19:705–713. doi:10.1007/s001700200080
8. Orhan S, Er OA, Camuscu N, Aslan E (2007) Tool wear evaluation
by vibration analysis during end milling of AISI D3 cold work tool
steel with 35 HRC hardness. NDT&E Int 40:121–126. doi:10.1016/
j.ndteint.2006.09.006
9. Zhang JZ, Chen JC (2008) Tool condition monitoring in an end-
milling operation based on the vibration signal collected through a
microcontroller-based data acquisition system. Int J Adv Manuf
Technol 39:118–128. doi:10.1007/s00170-007-1186-6
10. Sevilla PY, Jauregui JC, Herrera G, Robles JB (2013) Efficient
method for detecting tool failures in high-speed machining process.
Proc Inst Mech Eng B J Eng Manuf 227:473–482. doi:10.1177/
0954405412473906
11. Siddiqui RA, Amer W, Ahsan Q, Grosvenor RI, Prickett PW (2007)
Multi-band infinite impulse response filtering using
microcontrollers for e-Monitoring applications. Microprocess
Microsyst 31:370–380. doi:10.1016/j.micpro.2007.02.007
1194 Int J Adv Manuf Technol (2015) 81:1187–1194
Author's personal copy
