1 s2.0 S0890695509000418 Main

Wavelet analysis of sensor signals for tool condition monitoring:
A review and some new results

Zhu Kunpeng
, Wong Yoke San, Hong Geok Soon

Department of Mechanical Engineering, National University of Singapore, Singapore 119260, Singapore
a r t i c l e i n f o
Article history:
Received 9 May 2008
Received in revised form
10 February 2009
Accepted 12 February 2009
Available online 27 February 2009
Keywords:
Wavelet
Tool condition monitoring
a b s t r a c t
This paper reviews the state-of-the-art of wavelet analysis for tool condition monitoring (TCM).
Wavelet analysis has been the most important non-stationary signal processing tool today, and
popular in machining sensor signal analysis. Based on the nature of monitored signals, wavelet
approaches are introduced and the superiorities of wavelet analysis to Fourier methods
are discussed for TCM. According to the multiresolution, sparsity and localization properties of wavelet
transform, literatures are reviewed in ve categories in TCM: timefrequency analysis of machining
signal, signal denoising, feature extraction, singularity analysis for tool state estimation, and
density estimation for tool wear classication. This review provides a comprehensive survey of the
current work on wavelet approaches to TCM and also proposes two new prospects for future studies in
this area.
& 2009 Elsevier Ltd. All rights reserved.
Contents
1. Overview of tool condition monitoring. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537
1.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537
1.2. TCM as a pattern recognition problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538
2. Wavelet, wavelet transform, and properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
2.1. Limitation of timefrequency resolutions of Fourier methods. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
2.2. Wavelet and continuous wavelet analysis (CWT). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 540
2.3. Discrete wavelet transform (DWT) and wavelet packet decomposition (WPD) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541
2.4. Useful properties of wavelet transform for TCM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541
3. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541
3.1. Timefrequency analysis of TCM sensor signals with wavelet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 542
3.2. Wavelet denoising in TCM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544
3.3. Feature extraction and dimension reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545
3.4. Singularity analysis for tool wear detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 547
3.5. Wavelet probability density estimation for tool
state classication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549
4. Conclusion and future studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 550
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 550
1. Overview of tool condition monitoring
1.1. Introduction
During machining, the contact between the cutting tool,
workpiece, and the chips imposes pressure on the tool and
causes the shape of the tool to change, either gradually as
tool wear or abruptly as tool fracture or breakage [1]. In
tool condition monitoring, the aim is to apply appropriate
sensor signal processing and pattern recognition techniques to
identify and predict the cutting tool state, so as to reduce
loss brought about by tool wear or tool failure. An effective
tool condition monitoring (TCM) system can improve productivity
and ensure workpiece quality, and hence, has a major inuence
on machining efciency [2]. Tool condition monitoring has
been extensively studied by many researchers since the late
1980s. Many of the reported research works are reviewed in
[35].
ARTICLE IN PRESS
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/ijmactool
International Journal of Machine Tools & Manufacture
0890-6955/$ - see front matter & 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijmachtools.2009.02.003
Corresponding author.
E-mail address: mpezhuk@nus.edu.sg (K.P. Zhu).
International Journal of Machine Tools & Manufacture 49 (2009) 537553
Since tool condition is typically dened according to the
geometrical changes in the tool, direct monitoring methods such
as vision and optical approaches, which measure the geometric
parameters of the cutting tool, have been developed [68]. The
direct methods have advantages of capturing actual geometric
changes arising from wear of tool. However, direct measurements
are very difcult to implement because of the continuous contact
between the tool and the workpiece, and almost impossible due to
the presence of coolant uids. The difculties severely limit the
application of direct approach. The indirect approaches are
achieved by correlating or deducing suitable sensor signals to
tool wear states. They have the advantages of less complicated
setup and suitability for practical application. This paper focuses
on indirect approaches. For indirect approaches, tool condition is
not captured directly, but estimated from the measurable signal
feature. This signal feature is extracted through signal processing
steps (Fig. 1) for sensitive and robust representation of its
corresponding state.
Indirect methods such as those based on sensing of the cutting
forces [915], vibrations [1620], acoustic emission (AE) [2126],
and motor/feed current [2732] have been the most employed
and reported for TCM. Detailed works on the design and
implementation of these indirect approaches for TCM have been
reported in [3335].
1.2. TCM as a pattern recognition problem
The problem of TCM can be considered as a typical
pattern recognition problem. The objectives of TCM
can be formally specied to be a search for the most probable
state C
i
given the extracted measurable signal feature y(t)
at time t. This is a dynamic inference problem since the
tool state is not estimated only with prior knowledge, but also
adapt to the current features. This is somewhat of Bayesian
inference [36].
Hence, as the pattern recognition problem, the aim of TCM is to
nd,
TCM : arg max
i
pC
i
jy (1)
ARTICLE IN PRESS
Nomenclature
AE acoustic emission
AR autoregressive
ART adaptive resonance theory
CQF conjugate quadratic lters
CWD ChoiWilliams distribution
CWT continuous wavelet transform
DFT discrete Fourier transform
DWT discrete wavelet transform
EMD empirical mode decomposition
FDR Fishers discriminant ratio
FFT fast Fourier transform
FWT fast wavelet transform
GMM Gaussian mixture models
HMM hidden Markov models
ICA independent component analysis
KLT KarhunenLoeve transform
LE Lipschitz exponent
LDA linear discriminant analysis
MLP multilayer perceptron
MRA multiresolution
NN neural networks
PCA principal component analysis
pdf probability density function
PSD power spectrum density
SNR signal-to-noise ratio
SOM self-organizing map
STFT short-time Fourier transform
TCM tool condition monitoring
WPD wavelet packet decomposition
WT wavelet transform
WTMM wavelet transform modulus maxima
WVD WignerVille distribution
Symbols
f(t) any mathematical signal f(t)AL
R
2
x(t) any sensory signal from machining
y(t) extracted features
^
f o Fourier transform of f(t)
T
s
sampling interval
f
s
sampling frequency
C
i
class i
/x(t),y(t)S inner product of signal x(t) and y(t)
u position parameter of wavelet function
s scale parameter of wavelet function
s
t
the resolution of time
s
o
the resolution of frequency
c(t) wavelet function
j(t) scaling function
h(n) low-pass lter
g(n) high-pass lter
c
j,k
scaling coefcient
d
j,k
wavelet coefcient
W
j,n,k
wavelet packet
d
n
j;k
wavelet packet coefcient
Q quality factor
a Lipschitz exponent
S
i
the covariance matrix of class i
S
W
within-class covariance matrix
S
B
between-class covariance matrix
Machining

Setup
Signals
Force
AE
Vibration
Current
Image
Signal Processing

Amplitude analysis
Fourier analysis
Wavelet analysis
Statistical moments
Time series
Tool State
Wear
Chipping
Breakage
Failure
Chatter
Sensors
Classifier
Time series
LDA
NN
Clustering
HMM
features estimation
Fig. 1. The framework of TCM.
K.P. Zhu et al. / International Journal of Machine Tools & Manufacture 49 (2009) 537553 538
or in the physical form
TCM : arg max
tool state i
ptool statejsignal features (2)
The TCM system can be achieved in a three-step procedure as
pattern recognition (Fig. 2). The sensor signal x(t) is rstly pre-
processed to remove the noise and prepare the data for feature
extraction. Then the information relevant to pattern classication
is extracted from x(t) to a feature vector y(t). The task of feature
extraction is to enhance the characteristics of the various tool
wear classes and suppress or lter off the normal background. The
nal stage is state classication. Feature vector y(t) is assigned to
one of the K tool wear state, C
1
,C
2
,y,C
K
, by the classier based on
a certain type of classication criteria.
Earlier study on TCM classier is mainly carried out with time-
series analysis [9,37,38]. With these methods, a threshold value
needs to be set between the normal and abnormal tool states.
However, the threshold value varies with cutting conditions and is
difcult to determine. To improve the performance of TCM, more
advanced methods have been developed. Neural networks (NNs)
are most studied and gained most success in practical applications
[3944], due to its capability in learning and non-linear mapping
of features and tool state. Besides NNs, other pattern recognition
methods, such as fuzzy clustering approaches [45,46], linear
discriminant analysis (LDA) [37], Gaussian mixture models
(GMMs) [47], combination of regression and neuro-fuzzy techni-
ques [48], self-organizing feature maps (SOM) [8,49], hidden
Markov models (HMMs) [50,51], support vector machine (SVM)
[52], and rough set [53] have also been studied and applied to
TCM by many researchers.
Table 1 lists them according to their classication approaches.
Note that the list is not totally exhaustive but serves to be
representative of known TCM approaches, where most of them
also involve wavelet applications to one of the three stages of TCM
(Fig. 2).
2. Wavelet, wavelet transform, and properties
2.1. Limitation of timefrequency resolutions of Fourier methods
We call any square integrable real function f(t)AL
2
(R) a signal.
For the signal f(t), the Fourier transform
^
f o is obtained by the
inner product of f(t) with a sinusoidal wave e
jot
,
^
f o f t; e
jot
_ _

_
1
1
f te
jot
dt (3)
It transforms the signal f(t) from the time domain to the
frequency domain o and is viewed as the basis of modern signal
processing. The fast Fourier transform (FFT) [54] is the standard
method for observing signals in the frequency domain and has
been widely studied in TCM, such as those [11,55,56]. In spite of its
earlier popularity, Fourier transform has certain serious theore-
tical drawbacks in processing machining signals. This is because
the
^
f o is the integration of f(t) for all times tA(N,+N)
(function 3), and this globally inclusive of information makes it
difcult to analyze any local property of f(t) from
^
f o. To
overcome this limitation, Gabor [57] introduced a sliding window
function g(t) to the Fourier transform and obtains a localized
ARTICLE IN PRESS
Table 1
Tool condition estimation scopes.
Approaches References Comments
Time series AR, ARMA [9,37,38,127] Linear, physical meaning, good for stationary machining like
turning, not good for non-stationary machining like milling, need
set threshold for classication
Neural networks MLP [1214,19,4144,120,126,130] Iterative MSE optimization, sensitive to network structure, non-
linear classication, slow training
SOM [8,40,49,128] Non-linear and iterative clustering, suitable for low-dimension
feature space
ART [24,40,67,68,83,131] Based on competitive learning: fast incremental learning ability,
good self-adaptive ability
SVM [52] Maximizing the margin between classes with minimum number
of support vectors, metric dependent, non-linear, good
generalization, slow training
Others [122,123] Sensitive to training parameters, non-linear classication, robust
to outliers
Pattern recognition k-means [12,13] K-clusters, the nearest mean decides the cluster, good for
Gaussian signal with equal covariance
Fuzzy methods [12,13,45,46,76,132] Need initializing clusters and class membership
Gaussian mixture models [47,148] Each state is assumed to be drawn number of underlying
Gaussian distributions, soft membership, better than k-means
clustering, need estimate components
PCA/KLT [91,92,96,126] Linear, second-order statistics based on eigenvector
decomposition, good for Gaussian signal
LDA [12,13,23,37,38,61,96,124] Supervised linear classier, using MSE for optimization, better
than PCA for classication, need Gaussian assumption of signal
Stochastic models Hidden Markov models [50,51,129] Simple structure, good in generalization, good in non-linear and
non-stationary machining signals, need to train many small
models
Fig. 2. TCM as a pattern recognition system.
timefrequency atom f:
f
u;o
t e
jot
gt u (4)
The resultant transform is named short-time Fourier transform
(STFT):
STFTf t f t; f
u;o
_

_
1
1
f tgt ue
jot
dt (5)
The corresponding energy density |STFT(f(t))|
2
is called a
spectrogram, which is widely used for timefrequency analysis
before wavelet, and applied to TCM [49,58].
However, the frequency resolution s
2
o
and time resolution s
2
t
are constant for both time and frequency (Fig. 3(c)), and according
to Heisenberg uncertainty principle [59],
s
2
t
s
2
o
X
1
4
(6)
equality holds if f(t) is a Gaussian. It states that high resolution
both in frequency and time cannot be attained at the same time.
As shown in Fig. 3, while the original time signal f(t) has innite
time resolution (Fig. 3(a)) but no frequency information, the
Fourier transform
^
f o has innite frequency resolution (Fig. 3(b))
but provides no time information. The timefrequency resolution
of STFT is constant (Fig. 3(c)).
On the other hand, the above-mentioned approaches assume
that the sensor signals are stationary. However, due to the nature
of manufacturing processes, the signals are usually non-stationary
[12,60]. Constant time and frequency resolutions of STFT are not
suitable for the analysis of non-stationary signal. For example,
Mori et al. [61] took the FFT of the thrust force signal and found
that the spectra cannot capture the localized aspect of the saw-
tooth signal, but instead it spread the information across the
transformed signal. Wavelet analysis overcomes the drawbacks of
Fourier methods and permits adaptive timefrequency represen-
tation. Gong et al. [62] have shown that the wavelet analysis is
more sensitive and reliable than the Fourier analysis for
recognizing the tool wear states in turning. Yoon and Chin [63]
also veried the reliability of the wavelet transform method
compared the spectra method of FFT. The signal processing
approaches that deal with non-stationary signals are more
appropriate for process monitoring.
2.2. Wavelet and continuous wavelet analysis (CWT)
To meet the needs for adaptive timefrequency analysis in
applied mathematics, physics, and engineering, the wavelet
theory was developed in the late 1980s by Mallat [64,59], and
Daubechies [65,66]. It has been widely used to analyze machining
signals for tool wear monitoring since Tansel et al. [67,68].
Let c
s,u
(t), s,uAR,s40, be a family of functions dened as
translations and re-scales of a single function c
s;u
t 2 L
2
R [59],
c
s;u
t
1
s
p c
t u
s
_ _
(7)
where s is the scaling parameter and u the position parameter.
The wavelet c
s,u
(t) has the following basic properties:
_
1
1
ct dt 0;
_
1
1
c
2
t dt 1 (8)
These properties indicate that the wavelet is a small wave:
oscillate around zero (zero mean) and has limited support area
(nite energy), as shown in Fig. 4(a). The wavelet c
s,u
(t) has to
meet the admissibility condition for the transformation to be
invertible [65].
Continuous wavelet transform of signal f(t) is dened as
W
c
f s; u f t; c
s;u
t
_

1
s
p
_
1
1
f tc
t u
s
_ _
dt (9)
The timefrequency resolution of wavelet transform is illu-
strated in Fig. 4(b). The signal is localized in the area with time
width D
t
: u
0
1=2a
0
s
t
; u
0
1=2a
0
s
t
and frequency width
D
o
: Z=a
0
s
o
=2a
0
; Z=Z s
o
=2a
0
. Compared with the
STFT, whose timefrequency resolution is constant Fig. 3(c), the
timefrequency resolution of the wavelet transform (WT) de-
pends on the frequency of the signal. At high frequencies, the
wavelet reaches at a high time resolution but at a low-frequency
resolution; whereas at low frequencies, high-frequency resolution
and low time resolution can be obtained.
On the other side, the quality factor of WT is kept constant
in the timefrequency plane. For mother c(t/a), the quality
ARTICLE IN PRESS
t
t
1
u
2
u t 0
Fig. 3. Timefrequency resolutions of Fourier transform and wavelet transform.

0 2 4
-2
-1
0
1
2
Daubechies Wavelet
u

(
u
,
s
)
-50 0 50
-1
-0.5
0
0.5
1
u
Morlet Wavelet

(
u
,
s
)
t
a
a
0 t
a
0
a
u
0
u t 0
0
a
Fig. 4. Typical wavelets and adaptive timefrequency resolution of WT: u is position of the wavelet and s the scale of the wavelet.
factor dened,
Q
a
Central frequency=bandwidth Z=a=s
o
=a Z=s
o
and for c(t/a
0
),
Q
a
0
Central frequency=bandwidth Z=a
0
=s
o
=a
0
Q
a
As a result, no matter what u is (u40), c(t/a), and c(t/a
0
)
are constant with quality factor. Fig. 4(d) illustrates the
bandwidth and central frequency varies with the different scale
u. Such adaptability of timefrequency analysis is called math-
ematical micro-scope and benets the signal analysis crossing
scales. This multiscale analysis is preferred in TCM to get both
coarse global information and ne localized details when
necessary.
In physical interpretation, the coefcients of the wavelet
transform indicate the variation of energy of the signal with time
and frequency. In engineering applications, the square of the
coefcients of the CWT is often called as scalogram, dened
as Eq. (10), which has been used for machinery fault diagnostics,
and TCM.
SC
t
s; u; c W
c
f s; u
2
(10)
2.3. Discrete wavelet transform (DWT) and wavelet packet
decomposition (WPD)
Continuous wavelet transforms are recognized as effective
tools for both stationary and non-stationary signals. However, it
involves much redundant information and is computationally
very slow. Discrete wavelet transform was developed by Mallat
with fast algorithm based on the conjugate quadratic lters (CQF)
[64]. Wavelet and scaling functions at different scales are
generated from a single scaling function f(t) with two-scale
difference equations [59]:
ft
2
p
k
hkf2t k (11)
ct
2
p
k
gkf2t k (12)
where g(k) (1)
k
h(1k), and the h(k) and g(k) are viewed as
lter coefcients of low-pass and high-pass lters, and l is the
lter length. f(t) And c(t) are scaling and wavelet functions at
scale j 1, respectively. In an orthogonal wavelet expansion, a set
of recursive relationships governs scaling and wavelet coefcients
at different scales and translations as follows:
c
j;k

l
h
l2k
c
j1;l
(13)
d
j;k

l
g
l2k
c
j1;l
(14)
where c
j,k
, d
j,k
are scaling and wavelet coefcients derived from
the projection of the signal onto the space of scaling f
j,k
(t) and
wavelet functions c
j,k
(t), respectively.
Fig. 5(a) illustrates the 5-level MRA analysis of a cutting force
signal sampled at 6000Hz. With a 5-level MRA analysis, the
corresponding frequency bands are separated as illustrated in
Fig. 5(b). By DWT, the signal f(t) is decomposes into two parts:
low-pass approximation coefcients and high-pass detail coef-
cients. The next step then decomposes the new approximation
coefcients.
The DWT lead to a loss of useful information at high frequency
because successive details are no longer analyzed. We need to
double the sampling rate for higher frequency analysis, which
however involves more data and computation. Wavelet packet
decomposition [69] is a generalization of wavelet decomposition
at higher frequencies. In the wavelet packet decomposition, each
approximate and detail coefcients are recursively decomposed.
The library of wavelet packet basis functions fW
n
g
1
n0
can be
obtained from a given W
0
as follows:
W
j
2n

2
p
k
hkW
n
2t k (15)
W
j
2n1

2
p
k
gkW
n
2t k (16)
where the functions W
0
and W
1
are set to the scaling function f(x)
and the mother wavelet function c(x), respectively. The imple-
mentation of the wavelet packets leads to a tree-structured
decomposition, thereby implying that both the outputs of the
low-pass and high-pass lters are recursively decomposed.
The wavelet packet coefcients are then produced from the
integral:
d
j
n;k

_
W
j;n;k
tf t dt (17)
It should be emphasized that Eq. (18) allows many possible
combinations of wavelet packet functions to be selected in order
to optimally characterize the signal. Several criteria have been
developed to select the best basis for these purposes [59,70].
Due to the benets of wavelet decomposition, wavelet
methods have been studied in all aspects in TCM, such as
timefrequency analysis, signal denoising, feature extraction and
compression, or directly used as classier for TCM. Basic theories
of these approaches are introduced and literatures are reviewed
in Section 3.
2.4. Useful properties of wavelet transform for TCM
The most important property of wavelet useful in tool
condition monitoring is its sparse representation of signal. The
wavelet expansion coefcients c
j,k
and d
j,k
decay rapidly with
increase in j and k, and only a few large coefcients exist
while the others are small. By setting a suitable threshold, the
undesired noise is ltered. This is the essence of wavelet
denoising, and compression [71,72]. Another important property
of wavelet transform that determines its applications is its
localization property, as discussed in Section 2.1. Unlike Fourier
transform that spans the entire time period, wavelet transform
localize the time and frequency description of the signal, and
reveals the signal behavior in certain time and its corresponding
frequency property, which is generally useful for uncovering
different localized features associated with various different tool
states.
3. Applications
We rst presents a review of the timefrequency analysis of
machining signals, signal denoising, and feature extraction of
wavelet applications for TCM. These three applications are
studied in most of the literature reviewed. Two new approaches,
called singular detection and density estimation with wavelet
for TCM, are also introduced with real experiments and
discussed for TCM. Though few papers are found on these two
approaches on TCM, papers related to machinery condition
monitoring and fault diagnosis are reviewed to show the
prospect of these applications in TCM. Table 2 lists the references
reviewed according to different wavelet features used for TCM in
this paper.
ARTICLE IN PRESS
3.1. Timefrequency analysis of TCM sensor signals with wavelet
In the timefrequency analysis of TCM sensor signals, the
scalogram SC
t
(a,b;c), wavelet coefcient d
j,k
, scaling coefcient
c
j,k
, wavelet packet coefcients d
j
k
, and their wavelet domain
statistics (i.e.: mean, variance, etc.) are used as criterion in
condition discrimination in TCM after some manipulation. The
corresponding literatures reviewed in this section are summar-
ized in Table 3.
Yesilyurt [73] uses the scalogram and mean frequency of
scalogram of vibration signals in milling breakage detection.
The mean frequency of scalogram characterized the energy
density of the signal in a certain period. It was found that the
feed rate was highly correlated to the mean frequency of
scalogram, and the mean frequency variation is quite sensitive
to the presence of fault. Khraisheh et al. [74] found that CWT is
suitable for analyzing the transient in primary chatter. The
transient boundary and the built-up edge were successfully
identied by wavelet transform. With CWT, as discussed earlier,
it is a redundant transform. We generally encounter the problem
of overlapping, as a large amount of redundant information exists
after CWT; the overlapping may have the effect of smearing the
localized features for TCM. Minimizing the effect of overlapping
and improving the localization are still problems in CWT. These
are partially studied in [75] with an exact WT for gear fault
detection.
DWT is preferable in the timefrequency analysis because of
no redundancy and fast computation. Gong et al. [62] applied
DWT to monitor the ank wear states in turning. It was found that
the 5-level coefcients were sensitive to the ank wear and
cutting conditions. The normalized 5-level mean wavelet coef-
cients were used as parameters of the ank wear state recogni-
tion. The experimental results showed that it was more reliable
than FFT analysis in turning TCM. Li et al. [76] discussed a tool
breakage monitoring system based on DWT of acoustic emission
and feed current signals. The experimental results show overall
98.5% reliability and the good real-time monitoring capability of
the DWT for detecting tool breakage during drilling. Fu et al. [77]
identied saw tooth and screeching behavior in the thrust force
signal by using convolution masks to extract various features from
the DWT coefcients. The drawback of this method is that the
proles of the convolution masks are derived from idealized
simulated signals, and the shapes of the simulated signals
signicantly inuence the output results. Suh et al. [78] developed
ARTICLE IN PRESS
0 200 400 600 800 1000 1200 1400 1600 1800
-5
0
5
10
Cutting force and it's multiresolution decomposition
500 1000 1500
-5
0
5
10
Approximation level 5
500 1000 1500
-5
0
5
Detail level 5
500 1000 1500
-5
0
5
10
500 1000 1500
-5
0
5
Detail level 4
500 1000 1500
-5
0
5
10
500 1000 1500
-5
0
5
Detail level 3
500 1000 1500
-5
0
5
10
500 1000 1500
-5
0
5
Detail level 2
500 1000 1500
-5
0
5
10
500 1000 1500
-5
0
5
Detail level 1
V
0
V
1
V
2
V
3
V
4
W
4
W
3
W
2
W
1
0 47Hz 94Hz 188Hz 750Hz 3000Hz
W
5
V
5
1500Hz 375Hz
Fig. 5. (a) Five-level MRA analysis of cutting force: abscissa is the location of sampled cutting force, and ordinate is the force amplitude in Newton. (b) Frequency band
separation of 5-level MRA analysis.
a DWT-based algorithm to capture the variations of periodicity,
scaling coefcients, and wavelet coefcients. The three para-
meters were used to characterize stable, conditionally stable, and
unstable milling conditions. It was claimed that this approach can
accurately detect the transition of the system from one state to
the other.
Yoon and Chin [63] applied with the standard deviation ratio of
DWT coefcients of cutting force to detect chatter. It was found
that the detail coefcient parameters of the third or fourth level
were desirable for detection of chatter with spindle speed of
5001300rpm in end milling. The identication of chatter
through wavelet was also investigated by Berger et al. [79] who
analyzed the ratios of the mean absolute deviations of detail
coefcients of cutting forces. Similar study was also reported in
[80], which claimed that the variance of wavelet coefcients are
sensitive to tool wear and little inuenced by the variation of
working conditions, and thus provide a robust description of tool
wear in milling. It is reasonable to use the deviation as standard to
detect chatter because the signal amplitude varies largely before
and after the chatter period, but how to generate a robust decision
measure (i.e. threshold) is still challenging. Luis Alfonso et al. [81]
described a method using current signals to estimate the tool
conditions by using the DWT of cutting force, and through an
autocorrelation algorithm to evaluate the tool wear in the form of
an asymmetry weighting function. The authors reported that the
asymmetry value increases according to the states of the tool
wear. This approach is promising since it uses the current signals,
which is easy to implement and do not interrupt the machining.
ARTICLE IN PRESS
Table 2
Overview of the selected references with wavelet applications.
Wavelet Wavelet features Tool condition Monitored signal Reference
CWT Scalegram: SC
t
(a,b,c) mean, variance of Scalegram Drill breakage Current [29]
Drill wear Current [29]
Mill breakage Vibration [73]
Milling chatter Vibration [74]
DWT c
j,k
, d
j,k
, Mean, variance moments, and cumulants of DWT Drill breakage Current [29]
Force [61,67,68,77]
AE [76,132]
Drill wear Force [47,81]
Vibration [130]
AE [132]
Curremt [29]
Milling chatter Force [63,78,79]
Milling tool wear Force [80,125]
Milling fracture Current [111]
Turning tool wear AE [122,123]
Force [62,124,126]
Vibration [120,129]
Turning failure Force [110]
Grinding too wear Force [109]
WPD d
n
j;k
, Mean, variance moments, and cumulants of WPD Drilling wear Vibration [60]
Drilling breakage Current [46]
Turning chatter Force [60]
Turning chipping/breakage AE [24,82]
Turning wear AE [24,82]
Current [121]
Vibration [128]
Force [128,131]
MP
f
m1
n0
R
n
; f
g
n
_ _
f
g
n
R
m Drilling failure Force [85]
Table 3
Timefrequency properties of the transforms.
Wavelet Reference Transform Suitability Comments
CWT [7375] Linear Stationary, non-stationary Redundant transform, computationally slow
DWT [62,7681] Linear Stationary, non-stationary No redundancy, computationally fast, only low-
frequency components iteratively decomposed
WPD [24,60,82,83,121,127,131] Linear Stationary, non-stationary Redundant transform, both low- and high-frequency
components iteratively decomposed
MP [85] Linear Stationary, non-stationary No need to be orthogonal basis, acted also as a classier
STFT [49,58,62,63] Linear Stationary Can not adaptively change time and frequency resolution
WVD/CWD [8689] Quadratic Stationary, non-stationary Accurate timefrequency distribution, but involves
cross-interference terms
KLT [91,92] Linear Stationary A PCA approach, can not capture signals non-Gaussian
property
The problem is that the current signal may not be sensitive in all
conditions.
Wavelet transform can be very helpful if it is used as a signal
separation tool in TCM. Niu et al. [24] and Niu [82] separate the
acoustic emission signal into burst and continuous components
by WPD. The burst signal is suitable for the detection of transient
tool conditions, such as chipping and tool fracture, etc., and the
continuous signal is suitable for determination of tool wear. This
approach is much alike the one presented in [83]. This is
promising as only one signal is used to carry different monitoring
tasks.
The above WT approaches generally need feature extractors
followed by classiers for TCM. A modied WPD analysis, the
matching pursuit (MP) [84], handles both of the problems in an
integrated manner. It decomposes the signal into a linear
expansion of waveforms that are selected from a dictionary of
basis functions. It chooses waveforms that best match the signal
structure iteratively. As a pattern recognition method, MP adapts
to specic machining monitoring tasks and do not need feature
extraction or classication. Fu et al. [85] applied the MP to predict
small drill bit failure, with different types of drilling behavior from
the thrust force. It was revealed that the MP approach performed
satisfactorily with Gaussian, Haar, and Gabor wavelet dictionaries
in detecting small drill bit pre-failure. This approach simplies the
structure of TCM. The problem is that the threshold selection
scheme in the wear detector is hard to generalize.
In some cases, the energy variation of non-stationary signal is
examined for TCM. Quadratic timefrequency methods such as
the WignerVille distribution (WVD) and the ChoiWilliams
distribution (CWD) [86] are more precisely in representation of
the energy distribution. Gillespie and Atlas [87] stated that WVD
provides good stationary and non-stationary representation, and
provide alternatives for timefrequency analysis in TCM. These
methods are widely applied to mechanical signal analysis for
machinery condition monitoring [88,89], and shows good time
frequency energy distribution for fault detection. But in practice
these methods are limited by the existence of interference terms,
even if they are attenuated by some approaches [59]. Another
transform similar to WT is the KarhunenLoeve transform (KLT)
[90]. Tumer et al. [91,92] proposed the KLT to decompose
vibration signals from milling into fundamental eigenvectors,
and use the eigenvectors to indicate the changes in the fault
patterns. This is an improvement over methods which project the
data onto pre-determined wave functions, such as FT with sines
and cosines. But this approach is in essence the principal
component analysis (PCA), and it only uses the second-order
statistics and de-correlates the signals with eigenvector decom-
position. It cannot capture non-Gaussian properties (high-order
statistics), which is very important in characterizing signals used
in machining monitoring.
3.2. Wavelet denoising in TCM
Noise always exists in machining, especially encountered in
high precision machining [9395]. Denoising is a practical
problem in TCM. Houshmand and Elijah Kannatey-Asibu [96]
found that the spectra of the AE signal are highly contaminated by
noise. This noise is so high that the rst principal component of
spectra has no discriminatory power at all. But denoising is
generally not concerned in TCM area. This may due to the high
signal to noise ratio (SNR) in conventional machining, and the
noise imposes little effect on the nal decision. On the other hand,
in statistical analysis such as mean or moving average [97], the
noise is diminished by the averaging. But in high precision and
micro machining, the machining signal is typically very small, and
as a result the SNR is relatively low [93,95,98]. The noise has to be
removed before further analysis for TCM.
The model of signal with noise is [99]
yt
i
f t
i
t
i
(18)
where the function f(t) represents the desired signal, while the
remaining part e
i
is the noise. For the wavelet denoising, applying
DWT to the noisy data, we obtain the wavelet coefcients,
d Wyt
i
Wf t
i
Wt
i
(19)
Because smaller coefcients are usually contributed by data
noise, thresholding out these coefcients has the effect of
removing the data noise. In wavelet thresholding, after setting
some coefcients to zeros, the reconstructed (denoised) signal is
obtained by inverse transformation:
^
f t W
1 ^
d (20)
Fig. 6 illustrates the thresholding and reconstruction process.
Donoho [71] Donoho and Johnstone [72] developed several
wavelet-based thresholding techniques such as hard thresholding
and soft thresholding to nd an optimal estimate
^
f t from the
noisy data. The choice of threshold l is crucial: small/large
threshold values will produce estimates that tend to overt/
undert the data. They proposed a universal threshold
l ^ s

2logn
_
, where n is the number of observations and ^ s is
an estimate of the noise variance, which is unknown and needs to
be estimated from noisy samples. Despite the triviality of such a
threshold, they showed that the resulting wavelet estimator is
asymptotically near-minimax among all estimators within the
whole range of the Besov space [100]. Fig. 7 illustrated the
thresholding approaches. The hard thresholding employs a keep-
or-kill rule (Fig. 7(a)), while the soft thresholding is a shrink-or-
kill rule (Fig. 7(b)).
Mathematically speaking, both thresholding approaches have
drawbacks: hard thresholding is unstable, and sensitive to small
changes, while soft thresholding is a bias estimate rule. To
overcome these drawbacks, Gao and Bruce [101] developed the
rm threshold thresholding. The resulting wavelet thresholding
ARTICLE IN PRESS

Fig. 6. Wavelet denoising scheme.
Fig. 7. (a) Hard thresholding. (b) Soft thresholding. Wavelet denoising.
offers a balance between the two approaches. Other ways to
overcome this problem is to estimate level-dependent thresholds
with generalized cross-validation [102], or utilize a prior informa-
tion with Bayes shrinkage [103,104] for signal denoising, which
are exclusively studied in statistics literature and not applied to
mechanical domain however. (See Jansen [105] for a comprehen-
sive study on wavelet denoising.)
Menon et al. [106] used the wavelet-based method to eliminate
the background noise, which was a problem when using the AE to
detect small fatigue cracks in rotor head components. Bukkapat-
nam et al. [107] modied Donohos thresholding method for
chaotic signal with multiplicative noise. It was found that the
method can separate chaotic signal from worn tool in machining,
and suitable for on-line implementation. The DWT was formerly
applied by the authors as signal separation approach for the
denoising [108], which was not thresholding, but as a matched
lter with MRA decomposition. Kwak and Ha [109] described the
use of the grinding force signal with noise reduction to detect the
dressing time based on DWT. As a result of denoising, the grinding
force signal was successfully used to detect the need for dressing.
The wavelet denoising method was found to be more effective
than the FFT ltering technique. Kwak [110] furthered this
approach in turning. DWT is used in both denoising and detecting
tool failure. The DWT coefcients of the cutting force signal
showed that the onset time of tool failure and chatter vibration
was successfully detected. Li and Guan [111] proposed a wavelet-
based denoising to extract marked features from the feed-motor
current signals to indicate the minor cutting edge fracture. It was
found that the best denoising approach was to utilize a third
Symmlet mother wavelet function in combination with cross-
validation threshold determination and soft thresholding.
Wavelet denoising approaches are also studied in mechanical
system condition monitoring, which is quite similar to the TCM
approach. Qiu et al. [112] compared the performance of wavelet
decomposition-based denoising and wavelet lter-based denois-
ing methods on signals from mechanical defects. The comparison
result demonstrates that wavelet lter is more suitable and
reliable to detect a weak signature of mechanical impulse-like
defect signal, whereas the wavelet decomposition denoising
method can achieve satisfactory results on smooth signal
detection. Lin and Qu [113], and Lin et al. [114] improved the
soft-thresholding method by utilizing a prior information on the
probability density of the impulse, which matched with Morlet
wavelet. The timefrequency resolution can be adapted to
different signals of interest. It was claimed that the method
performed excellently when applied to denoise gear and bearing
vibration signals with a low SNR.
However, the existing wavelet denoising methods reported in
the literature rely heavily on white Gaussian noise and relative
energy levels of wavelet coefcients [115]. In practical machining,
the noise is generally not Gaussian. This is because the noise is not
purely random but correlated with working conditions. Fig. 8
shows the residue of the denoised force after soft thresholding.
The PSD is nearly evenly distributed in all frequencies, and a
Gaussian signal ts it quite well. The wavelet denoising methods
meet problems in denoising non-Gaussian noise, which is
discussed in details in [98]. It was found that the thresholds
dened by Donoho [71] Donoho and Johnstone [72] are too small
when we met super-Gaussian (normalized kurtosis40) noise
[116,117]. In this condition, the noise can be regarded as a certain
signal and the wavelet decomposition coefcients are not evenly
distributed among all scales as Gaussian noise, and as a result
some of the noise coefcients are not small. Under this condition,
independent component analysis (ICA) [118,119] for noise separa-
tion is more effective than wavelet denoising as shown in [98].
Fig. 9 is the recorded real noise during the interval of machining.
Apparently, it is quite different from the estimated noise in Fig. 8
both in time and frequency domain, and is super-Gaussian.
3.3. Feature extraction and dimension reduction
Apart from the original intention of the WT for timefrequency
analysis, the important and successful application of wavelet in
ARTICLE IN PRESS
0 0.2 0.4 0.6 0.8 1
-0.5
0
0.5
time (s)
wavelet residue
a
m
p
l
i
t
u
d
e

(
N
)
-0.3 -0.2 -0.1 0 0.1 0.2 0.3
0
2
4
6
probability density

p
d
f
sample value
residue
Gaussian fit
0 500 1000 1500 2000 2500 3000
0.02
0.04
0.06
0.08
frequency (Hz)
PSD
P
S
D

v
a
l
u
e
-200 -100 0 100 200
-0.2
0
0.2
0.4
0.6
0.8
Autocorrelations
step
a
u
t
o
c
o
r
r
e
l
a
t
i
o
n

v
a
l
u
e
Fig. 8. Illustrative of the residue of wavelet denoised force. (a) Estimated noise, (b) noise distribution compared with Gaussian distribution, (c) the corresponding power
spectrum density and (d) the autocorrelation coefcients.
TCM is feature extraction and dimension reduction. The wavelet
coefcients usually need to be extracted for robust and effective
representation of different tool states in TCM. The extraction
approaches are generally based on some statistical measurements
(i.e. mean, variance, Euclidian distance, etc.) to maximize the
discriminability or classication among different tool states.
Related literatures are summarized in Table 4 according to their
approaches.
Moria et al. [61] developed a method for extracting pre-failure
features from the cutting force to predict breakage of a small drill
bit to discriminate different tool states. The DWT coefcients are
reduced to three indices: energy, waviness, and irregularity. The
problem is it is difcult to determine the index function for
classication. Tansel et al. [67] used the DWT coefcients of the
thrust force as an input to (ART-2) to predict micro drill bit
breakage in peck drilling. The authors later enhanced this
technique by encoding the raw data before inputting it to the
ART-2 [68]. However, the encoding technique relies excessively on
statistical abstractions that cannot be explained in terms of the
machining system behavior during failure. Thus, the drilling
monitoring method lacks a basis for generalization to other types
of drilling processes. Hong et al. [120] decomposed dynamic
cutting force signal into different frequency bands by DWT, and
two features, mean values and variances of the local maxima,
were extracted from the decomposed signal for each frequency
band. These features are input to MLP for the tool state
classication. It was shown that the features extracted by wavelet
transform are less sensitive to changing cutting conditions and
the MLP has high classication rate. Similar study has also been
reported in [121]. Kamarthi and Pittner [122], and Pittner and
Kamarthi [123] investigate a ank wear estimation technique in
turning through wavelet representation of acoustic emission
signals. The energies of DWT coefcients are used for ank wear
estimation. The ank wear estimation from the recurrent neural
network is fairly accurate and indicates that the FWT representa-
tion of AE signals is more effective and sensitive than Fourier
transform representation. Kamarti et al. [124] extended this study
with a new feature extraction method based on the FWT. The
features of AE signal are calculated by the Euclidean norms of the
frequency clusters. They showed that the proposed method can
efciently extract important features related to progressive ank
wear. Choi et al. [125] studied cutting force trends and tool wear
effects in ramp cut machining. This study is challenging because
the depth of cut is continuously changing in ramp cuts. Wavelet
analysis is applied to cutting forces from a progressively worn
tool. The root mean square value of the approximation coefcients
extracted as features and linear regression are used for tool wear
estimation. It is reported that for small depth of cut, the linear
regression can estimate the tool wear with error below 6%.
A typical example of wavelet feature extraction was presented
by Wu and Du [60]. They introduced an automatic feature
extraction and assessment procedure using a wavelet packet
transform in TCM. To identify the effectiveness of the selected
features in both time and frequency domains, four criteria such as
cross-correlation and cross-coherence of signal and reconstructed
signal, correlation of the residue, and power spectrum of the
residue, are proposed. The proposed method is tested for chatter
monitoring in turning and tool wear monitoring in drilling.
It was found that WPD can capture important features of signal
that are sensitive to the machining conditions, e.g. chatter and
different states of tool wear, but is insensitive to the variation of
working conditions and noises. Accordingly, accurate and reliable
on-line monitoring decisions can be made. Wu et al. [126]
furthered the study with a real-time implementation of WPD
ARTICLE IN PRESS
0 0.2 0.4 0.6 0.8 1
-0.4
-0.2
0
time (s)
noise
a
m
p
l
i
t
u
d
e

(
N
)
-0.5 -0.4 -0.3 -0.2 -0.1
0
5
10
probability density

sample value
p
d
f
real noise
Gaussian fit
0 1000 2000 3000
0
0.2
0.4
0.6
0.8
frquency (Hz)
PSD
P
S
D

v
a
l
u
e
-200 -100 0 100 200
0.7
0.8
0.9
1
step
a
u
t
o
c
o
r
r
e
l
a
t
i
o
n

v
a
l
u
e
autocorrelation
Fig. 9. Recorded real noise. (a) Noise signal, (b) noise distribution compared to Gaussian distribution, (c) the corresponding power spectrum density and (d) the
autocorrelation coefcients.
Table 4
Feature extraction with wavelet analysis.
Features Reference Comments
DWT [61,67,68,120,122125,129,130,132] Statistics of DWT coefcients, need
succeeding classier or set a
threshold for classication
WPD [24,60,83,121,126128,131,133136] Statistics of WPD coefcients, need
succeeding classier or set a
threshold for classication
MP [85] Acted as both feature extraction and
classication
to motor current signal. The wavelet packet features were
obtained under normal tool condition. Based on the principal
components analysis the alarm thresholds are then constructed
for decision making. WPD approach for the feature extraction is
also studied in [127], while the classication is based on a MLP
neural network.
Scheffer and Heyns [128] adapted the approach of [60] and
used the correlation coefcient with an ideal trend function to
select features during machining. The monitoring system auto-
matically selected features displaying the most consistent trend
towards tool wear. The time and frequency domains, time-series
coefcients and wavelet packet features were extracted. The WPD
features consist 5 of 13 features nally selected. Wang et al. [129]
extracted DWT coefcients and designed a codebook vector to
convert the feature vectors into a symbolic sequence. The
effectiveness of the approach was evaluated under different
conditions. The problem of this approach is that the vector
quantization is targeted for HMM modeling, and may largely
degrade the sensitivity of the wavelet features, since it is averaged
over a long period. Abu-Mahfouz [130] found that the average
harmonic wavelet coefcients and the maximum entropy spec-
trum peaks are more efcient in training the MLP than the time
domain features in monitoring of twist drill wear. Other features,
such as normalized mean wavelet packet coefcients of the feed
force, which is affected not only by the ank wear but also by the
severe crater wear observed in high-speed machining, are tried by
Obikawa and Shinozuka [131]. Li et al. [132] applied DWT to
decompose turning AE signals into uncorrelated components.
These components were treated as features and thresholded with
Shewhart control charts for TCM. The case study demonstrated
that the new method is able to detect serious tool wear or
breakage in the machining process. However, the threshold
selection is not mentioned and it is difcult to decide the range
limit.
The following case study shows a feature extraction approach
for milling TCM [133]. Milling force is highly non-stationary and
WT is more appropriate than FT approaches. The features are
extracted with wavelet packet coefcients. It is decomposed into 5
levels for the signal representation. There are overall 62 wavelet
packets in this decomposition. These high-dimension features
need to be reduced for less computation and robust representa-
tion. A feature selection algorithm is adapted as Fishers
discriminant ratio (FDR) [134] to discriminate features from
different classes:
FDR
c
i
c
jai
m
i
m
j
2
s
2
i
s
2
j
(21)
where the subscripts i, j refer to the mean (m
i
,m
j
) and scatter
matrix (s
i
,s
j
) corresponding to the features under investigation for
the classes o
i
, o
j
, respectively. Features are then ranked with their
class discriminating ability according to their FRD. Their FDR
score is demonstrated in Fig. 10 under different tool wear level.
The features are normalized between [0,1] and classied by a
succeeding classier.
Besides TCM, interesting and efcient wavelet feature extrac-
tion approaches are also studied widely in machinery condition
monitoring, which are valuable references to TCM. Wang and Gao
[135] selected wavelet packet feature with principal component
analysis in machine defect detection. Liu and Ling [136]
introduced mutual information to search redundant wavelet
packets. This algorithm extracts those informative wavelets which
carry important but less redundant information about machinery
faults. It was found that the proposed feature extraction method
performed much better than the PCA in detecting diesel engine
malfunctions.
3.4. Singularity analysis for tool wear detection
In signal analysis, singularity is thought of as either an abrupt
change or a sudden shift of the signals value to a different level.
This corresponds to the phenomenon happened in machining
such as chipping and tool breakage. The original goal of
singularity analysis is to estimate the localization and degrees of
a signals abrupt changes or image edges [137,138]. It is estimated
from wavelet coefcient regularity at certain point and the rate of
decay of the coefcients near this point. It has been studied in
machinery condition monitoring lately [139147], but little is
studied in TCM. Because it is robust and stable in capturing signal
changes, it can be expected that these singularity-based features
will provide valuable study in TCM. We introduce the idea of
wavelet singular analysis rst and then reviewed some papers
related to machinery condition monitoring, and then illustrate the
application to milling TCM.
The singularity of the signal is measured by the Lipschitz
exponent (LE) [137]. Assume the signal f(t) has the following
property at the vicinity of t
0
:
f t
0
h P
n
h
pA h
a
; noaon 1 (22)
where P
n
(t) is an n-order polynomial and pass the point f(t
0
), and h
is small enough, the LE of f(t) at t
0
is the superemum of the a for
which function (23) holds. It follows directly that a determines
f(t)s continuously differentiable times (regular) at t
0
: the higher
the a, the more regular the function f(t). Mallat and Hwang [137]
have shown that the local regularity of a function (signal) is
related to the propagation across scale of its wavelet maxima and
the decay of the wavelet transform modulus in the time-scale
plane. A modulus maxima line connects all these local maxima
points.
With the WT of f(t), and in the vicinity of t,
Wf
s
t
pAs
a
or log Wf
s
t
plogA alogs (23)

where A is a constant related to the wavelet c(t). This function
connects the wavelet scale j and the Holder exponent a. The A and
a can be computed by setting the equality in function (24). The
function also shows the relationship between log wavelet trans-
form modulus maxima (WTMM) and the wavelet scale j (or a in
CWT). The Holder exponent a indicates the degree of singularity
and is often used as a feature in machine condition monitoring.
The singularity analysis with wavelet for TCM was attempted
by Chen and Li in [139]. The authors discussed the WTMM
approach but it was not implemented as they claimed that the
WTMM was easily inuenced by the environment noise and
varying cutting conditions; it could not represent the cutting tool
conditions accurately. Therefore, the WTMM approach was not
preferred but chose the wavelet coefcient norm and their
ARTICLE IN PRESS
0 5 10 15 20 25 30 35
0
5
10
15
20
25
the number of wavelet packet features
d
i
s
c
r
i
m
i
n
a
t
e

v
a
l
u
e
slight wear
medium wear
severe wear
Fig. 10. Discriminant properties of different wavelet packets.
statistics as features to indicate the tool condition instead. This
approach is more robust in case of noise but it is also subject to
the limitation in choosing threshold, which also varies with
working conditions. Besides, the AE signal can be denoised rst to
apply the WTMM scheme.
Most applications of wavelet for singularity detection were
applied such as change point analysis [140], and non-linear
system identication [141], which are similar to the TCM in
machining. Hambaba and Hu [142] used wavelet transform to
determine the Holder exponent value of a gear response at
different scale levels. By tting an autoregressive moving average
(ARMA) model to the wavelet-transformed data, analysis of the
residual error was used to indicate the presence of damage in the
gear. Sun et al. [143] proposed singularity analysis for bearing
defect diagnosis. Through modifying the intensity of the wavelet
transform modulus maxima, defect-related vibration signature
was highlighted and could be easily associated with the bearing
defect characteristic frequencies. Loutridis and Trochidis [144]
employed the Lipschitz exponent to investigate gear faults. It was
shown that the Hoelder exponent for each type of fault exhibits a
constant value, not affected by changing working conditions, and
can be used as a stable indication of different gear faults. It was
also showed that the Holder exponent was effective in capturing
non-stationary nature of signals and sensitive for identifying
structure damage [145]. Peng et al. [146] examined shaft orbits
using the wavelet modulus maxima. Four LE-based features are
extracted to classify the shaft orbit. Peng et al. [147] furthered
this study and presented a novel singularity-based fault feature
extraction from vibration signals. The data are collected
under different machine health conditions that show different
patterns of singularities measured quantitatively by the
Lipschitz exponents. The number of Lipschitz exponents per
rotation, the mean value and the relative standard deviation
of the Lipschitz exponents that are obtained from the extracted
features for singularity representation. The results show that the
three parameters are excellent fault features for fault pattern
recognition.
Fig. 11 shows an example study with modulus maxima for
detection of milling tool at different wear state. Fig. 11a shows a
cutting force signal from a fresh tool and severe worn tool.
Lipschitz exponent extracted from the WTMM (Fig. 11(b)). Though
they do not change much, Lipschitz exponents offer a stable tool
ARTICLE IN PRESS
0 500 1000 1500 2000 2500 3000 3500 4000 4500
-2
-1
0
1
2
force sample
f
o
r
c
e

a
m
p
l
i
t
u
d
e

(
N
)
Slight wear
0 500 1000 1500 2000 2500 3000 3500 4000 4500
-2
0
2
4
force sample
f
o
r
c
e

a
m
p
l
i
t
u
d
e

(
N
)
Severe wear
0 500 1000 1500 2000 2500 3000 3500 4000 4500
-1.5
-1
-0.5
0
0.5
Holder estimates at time instants T
sample
H
E

v
a
l
u
e
0 500 1000 1500 2000 2500 3000 3500 4000 4500
-1
-0.5
0
0.5
1
Holder estimates at time instants T
force sample
H
E

v
a
l
u
e
-0.85 -0.8 -0.75 -0.7 -0.65
0.8
0.85
0.9
0.95
1
Singularity Spectrum
Singularity Singularity

s
p
e
c
t
r
u
m

v
a
l
u
e
-0.3 -0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
Singularity Spectrum
s
p
e
c
t
r
u
m

v
a
l
u
e
Fig. 11. (a) Cutting forces of fresh tool and severe ank wear tool. (b) Lipschitz exponent of slight wear tool and severe wear tool. (c) The singular spectrum of the slight
wear tool and severe wear tool. Modulus maxima for slight wear tool and severe wear tool.
to detect the degrees of tool wear level as shown. The spectrum of
LEs from two tool states is illustrated in Fig. 11(c). It is observed
that with the fault severity increasing, the milling force signals
singularities and singularity ranges will increase as well, and
therefore we can evaluate the fault severity through measuring
the force signals singularities and singularity ranges.
Carefully attention should also be made upon choosing wavelet
in the WTMM approach. Not all wavelets WTMM location
corresponds to the abrupt change points, and it happens only
when the applied wavelet is the rst-order differentiation of a
certain smooth function (i.e. Gaussian). If it is the second-order
differentiation of a smooth function, the change point will be the
WTs zero crossing position, not the WTMM position. As in this
study it applies the rst-order differentiation of Gaussian function
in this study of Fig. 11. The regularity is another important
criterion in the selection wavelet. Usually, the selected wavelet
must be sufciently regular, which means a larger vanish moment
(see [65] for details about vanish moment); otherwise some
singularities would be overlooked. Additionally, the noise will
inuence the performance of the wavelet greatly, therefore before
the singularity is detected, the signal preprocessing must be
carried out.
3.5. Wavelet probability density estimation for tool
state classication
Probability density estimation is a classic problem in statistics.
The density estimation problem in statistics is in fact the pattern
recognition problem in engineering as discussed in Section 1.2:
assigning classes to higher probability density groups. There is
little study of probability density application to TCM except
[47,148]. Heck and Chou [47], and Chou and Heck [148] reported
studies of wavelet density estimation for TCM. They use a
Gaussian mixture model to approximate the wavelet coefcients
from machining signals. Different tool states are represented with
different GMM, and then choose the maximum class conditional
probability for state decision when feature extracted for classi-
cation. In other area as chemical process monitoring, Safavi et al.
[149] present an application of wavelets analysis to density
estimation and process monitoring. The resulting density function
was used to dene a normal operating region for the process so
that any future abnormal changes in the process can be
monitored. Results of applying these techniques to chemical
process are also presented to be effective in monitoring the states.
In the structural health monitoring, Krishnan and Kiremidjian
[150] applied a time-series-based detection algorithm with
GMMs. The structural vibration signals are modeled with ARMA
rst, and then the rst three AR coefcients are chosen as feature
vectors. A Gaussian mixture model is used to model the feature
vector. Damage is detected by comparing the Mahalanobis
distance of the extracted AR coefcients and those of the
undamaged. This approach provides a useful framework for data
fusion, where different measurements such as strains, tempera-
ture, and humidity could be used for a more robust damage
decision.
The most important and popular density estimation approach
before wavelet density estimation is kernel density estimation
[151]. Wavelet is brought attention in density estimation lately
[152]. The differences between kernel and wavelet estimates are
mainly explained by the ability of the wavelet method to take into
account local gaps in the data distribution. This new approach is
very promising, since smaller structures superimposed onto a
larger one are detected by this technique, especially when small
samples are investigated. Thus, wavelet solutions appear to be
better suited for classication studies.
Wavelet density estimation is developed from non-parametric
regression problem in statistics example [153,154]. Wavelet
density estimation opens the problem of density estimation for
dependent observations, while almost all other density estimators
fail in dealing. Let X
1
,X
1
,y,X
n
be a sequence of independently and
identically distributed random variables, with density: h h(x).
This density is unknown and to be estimate. Suppose
_
h(x)
2
dx is
nite, and approximated in wavelet series:
hx
k2Z
c
j;k
f
j;k
t
J
j1
k2Z
d
J;k
c
J;k
t (24)
ARTICLE IN PRESS
-1 -0.5 0 0.5 1 1.5
-2
0
2
4
6
8
10
x
p
(
x
)
-1 -0.5 0 0.5 1 1.5 2
-2
0
2
4
6
x
p
(
x
)
-2 -1 0 1 2 3 4
-1
0
1
2
x
p
(
x
)
0 5 10 15 20 25
0
0.1
0.2
0.3
0.4
0.5
Pmax
Pmin
Fig. 12. Density estimation with wavelet and state estimation. (a) slight wear pdf, (b) medium wear pdf, (c) severe wear pdf and (d) Bayesian classier (MAP).
To estimate h, it is sufcient to estimate the coordinates c
j,k
and
d
J,k
by the sample mean:
^ c
j;k

1
n
n
i1
f
j;k
X
i
;
^
d
J;k

1
n
n
i1
c
J;k
X
i
(25)
Inserting this estimate to the wavelet series expansion of h, we
get an estimate
^
hx
k2Z
^ c
j;k
f
j;k
t
J
j1
k2Z
^
d
J;k
c
J;k
t (26)
The estimate
^
hx is then used as real probability density for
application.
The idea can be easily adapt to tool wear classication with the
following steps:
(1) Estimate the respective density distributions of tool wear
levels provided by the machining data and measured tool
wear, like the supervised neural network training.
(2) Match the present estimated signal density to the known
density.
(3) Select the maximum posterior probability density as tool
condition with Bayes classier.
Fig. 12(a)(c) is the learned density from the measurement,
representing three possible tool state. The estimated density is
matched to these tree conditions, and the likelihood of these
matching is found in Fig. 12(d). We conclude that the state highest
probability is what we estimate.
The wavelet density estimation generally applied together with
GMM because this approach is more exible since it can
approximate wider densities. This idea is similar to those of NN,
HMM, and PR approaches in nature, where the approaches nd
the most probable state by the pdfs they belong to. Wavelet
density estimation is more powerful as it can approximate wider
densities while PR approaches such as LDA assume Gaussian
density as a prior, the later is generally not true of machining
sensory signals however. More studies can be expected in TCM
with this approach.
4. Conclusion and future studies
It is shown that WT is competitive to other signal analysis
approaches because of its MRA, sparsity, and localization proper-
ties. It is effective in analyzing non-stationary machining sensor
signals. Based on the benets of WT discussed, applications of
wavelet in TCM are reviewed in timefrequency analysis, denois-
ing, feature extraction, singularity analysis, and density estima-
tion. It achieves a lot of success in TCM overall but many more
need to be further studied.
As discussed, the WT presents a natural MRA analysis and
keeps the quality factor constant under different scales. These are
to the benet of characterizing signals crossing scales properties
with WT. These advantages of WT should be taken over other
methods such as FT and PCA in the applications but not only use
WT as lter banks and decompose the signal into various
frequency bands. For the wavelet denoising, the general thresh-
olding approaches are not preferred in TCM because of its
Gaussian noise assumption, because in TCM the noise is generally
correlated to machining conditions and not Gaussian. As reference
noise signal can always be collected during TCM, apply Bayesian
approaches that incorporate this a prior information will be quite
helpful in signal denoising. WT has great potential in detecting
abrupt changes of tool conditions in TCM. It is very robust as the
LE component is insensitive to changing working conditions and is
be very helpful in TCM. For future studies, the wavelet density
estimation can approximate wider densities than other density
estimation approaches. The density estimation actually incorpo-
rates both feature extraction and classication steps in the TCM
and is concise and easy to implement. More studies can be
expected in these areas.
References
[1] Milton C. Shaw, Metal Cutting Principles, second ed., Oxford University Press,
New York, 2005.
[2] Geoffrey Boothroyd, A. Winston, Knight, Fundamentals of Machining and
Machine Tools, Addison-Wesley, Redwood City, CA, 1989.
[3] G. Byrne, D. Dornfeld, I. Inasaki, G. Ketteler, W. Konig, R. Teti, Tool condition
monitoring (TCM)the status of research and industrial application, Annals
of the CIRP 44 (1995) 541.
[4] D.E. Dimla, Sensor signal for tool-wear monitoring in metal cutting
operationsa review of methods, International Journal of Machine Tools
and Manufacture 40 (2000) 10731098.
[5] S.Y. Liang, L.H. Rogelio, G.L. Robert, Machining process monitoring and
control: the state-of-the-art, Journal of Manufacturing Science and En-
gineering 126 (2004) 297310.
[6] D.C.D. Oguamanam, H. Raafat, S.M. Taboun, A machine vision system for
wear monitoring and breakage detection of single-point cutting tools,
Computers in Industrial Engineering 26 (3) (1994) 575598.
[7] S. Kurada, C. Bradley, A review of machine vision sensors for tool condition
monitoring, Computers in Industry 34 (1) (1997) 5572.
[8] W.H. Wang, G.S. Hong, Y.S. Wong, K.P. Zhu, Sensor fusion for on-line tool
condition monitoring in milling, International Journal of Production
Research 45 (21) (2007) 50595116.
[9] Y. Altintas, In-process detection of tool breakages using time series
monitoring of cutting forces, International Journal of Machine Tools and
Manufacture 28 (2) (1988) 157172.
[10] Y. Altintas, I. Yellowley, In process detection of tool failure in milling using
cutting force models, Journal of Engineering for IndustryTransactions of
the ASME 111 (1989) 149157.
[11] M.A. Elbestawi, T.A. Papazariou, R.X. Du, In-process monitoring of tool wear
in milling using cutting force signature, International Journal of Machine
Tools and Manufacture 31 (1) (1991) 5573.
[12] R. Du, M.A. Elbestawi, S.M. Wu, Computer automated monitoring of
manufacturing processes: Part 1. Monitoring decision-making methods,
Journal of Engineering for IndustryTransactions of the ASME May (1995).
[13] R. Du, M.A. Elbestawi, S.M. Wu, Computer automated monitoring of
manufacturing processes: Part 2. Applications, Journal of Engineering for
IndustryTransactions of the ASME May (1995).
[14] H. Saglam, A. Unuvar, Tool condition monitoring in milling on cutting forces
by a neural network, International Journal of Production Research 41 (7)
(2003) 15191532.
[15] G. Sutter, A. Molinari, Analysis of the cutting force components and friction
in high speed machining, Journal of Manufacturing Science and Engineer-
ingTransactions of the ASME 127 (May) (2005) 245250.
[16] J.C. Chen, W.L. Chen, A tool breakage detection system using an accel-
erometer sensor, Journal of Intelligent Manufacturing 10 (2) (1999) 187197.
[17] T.I. El-Wardany, D. Gaoand, M.A. Elbestawi, Tool condition monitoring in
drilling using vibration signature analysis, International Journal of Machine
Tools and Manufacture 36 (6) (1996) 687711.
[18] D.E. Dimla Sr., P.M. Lister, On-line metal cutting tool condition monitoring I:
force and vibration analyses, International Journal of Machine Tools and
Manufacture 40 (2000) 739768.
[19] D.E. Dimla Sr., P.M. Lister, On-line metal cutting tool condition monitoring II:
tool-state classication using multi-layer perceptron neural networks,
International Journal of Machine Tools and Manufacture 40 (2000) 769781.
[20] B.S. Berger, I. Minis, J. Harley, M. Rokni, M. Papadopoulos, Wavelet based
cutting state, identication, Journal of Sound and Vibration 213 (5) (1998)
813827.
[21] E.N. Diei, D.A. Dornfeld, A model of tool fracture generated acoustic emission
during machining, Journal of Engineering for IndustryTransactions of the
ASME 109 (1987) 227233.
[22] M. Liu, S.Y. Liang, Analytical modeling of acoustic emission for monitoring of
peripheral milling process, International Journal of Machine Tools and
Manufacture 31 (4) (1991) 589606.
[23] AliA Houshmand, Elijah Kannatey-Asibu Jr., Gray D. Herrin, A dynamic model
for tool detection using acoustic emission, Mechanical Systems and Signal
Processing 9 (4) (1995) 415428.
[24] Y.M. Niu, Y.S. Wong, G.S. Hong, Multi-category classication of tool
conditions using wavelet packets and ART2 network, Journal of Manufactur-
ing Science and TechnologyTransactions of the ASME 120 (1998) 807815.
[25] P. Srinivasa Pai, P.K. Ramakrishna Rao, Acoustic emission analysis for tool
wear monitoring in face milling, International Journal of Production
Research 40 (5) (2002) 10811093.
[26] J. Sun, G.S. Hong, M. Rahman, Y.S. Wong, Identication of feature set for
effective tool condition monitoring by acoustic emission sensing, Interna-
tional Journal of Production Research 42 (5) (2004) 901918.
ARTICLE IN PRESS
[27] Y. Altintas, Prediction of cutting forces and tool breakage in milling from
feed drive current measurement, Journal of Engineering for Industry
Transactions of the ASME 114 (1992) 386392.
[28] X. Li, On-line detection of the breakage of small diameter drills using current
signature wavelet transform, International Journal of Machine Tools and
Manufacture 39 (1999) 157164.
[29] X. Li, S.K. Tso, Jun Wang, Real time tool condition monitoring using wavelet
transforms and fuzzy techniques, IEEE Transactions on Systems, Man, and
Cybernetics Part C: Applications and Reviews 31 (3) (2000) 352357.
[30] X. Li, Detection of tool ute fracture in end milling using feed-motor current
signatures, IEEE/ASME Transactions on Mechatronics 6 (2001) 491498.
[31] B.Y. Lee, Y.S. Tarng, Application of the discrete wavelet transform to the
monitoring of tool failure in end milling using the spindle motor current,
International Journal of Advanced Manufacturing Technology 15 (1999)
238243.
[32] H.K. Tonshoff, X. Li, C. Lapp, Application of fast Haar transform and
concurrent learning to tool-breakage detection in milling, IEEE/ASME
Transactions on Mechatronics 8 (2003) 414417.
[33] C.S. Leem, D.A. Dornfeld, Design and implementation of sensor based tool
wear monitoring systems, Mechanical Systems and Signal Processing 10 (4)
(1996) 439458.
[34] A. Ghasempoor, J. Jeswiet, T.N. Moore, Real time implementation of on-line
tool condition monitoring in turning, International Journal of Machine Tools
and Manufacture 39 (12) (1999) 18831902.
[35] T.H.C. Childs, K. Maekawa, T. Obikawa, Y. Yamane, Metal Machining: Theory
and Applications, Elsevier, Amsterdam, 2000.
[36] Richard O. Duda, Peter E. Hart, David G. Stork, Pattern Classication, second
ed., Wiley Interscience, New York, 2001.
[37] S.Y. Liang, D.A. Dornfeld, Detection of cutting tool-wear using adaptive time
series modeling of acoustic emission signal, Sensors for Manufacturing
Transactions of the ASME (1987) 2738.
[38] S.A. Kumar, H.V. Ravindra, Y.G. Srinivasa, In-process tool wear monitoring
through time series modeling and pattern recognition, International Journal
of Production Research 35 (3) (1997) 739751.
[39] Bernhard Sick, Review: on-line and indirect tool wear monitoring in turning
with articial neural networks: a review of more than a decade of research,
Mechanical Systems and Signal Processing 16 (4) (2002) 487546.
[40] R.G. Silva, R.L. Reuben, K.J. Baker, S.J. Wilcox, Tool wear monitoring of turning
operations by neural network and expert system classication of a feature
set generated from multiple sensors, Mechanical Systems and Signal
Processing 12 (2) (1998) 319332.
[41] J.H. Lee, D.E. Kim, S.J. Lee, Application of neural networks to ank
wear prediction, Mechanical Systems and Signal Processing 10 (3) (1996)
265276.
[42] H. Saglam, A. Unuvar, Tool condition monitoring in milling based on cutting
forces by a neural network, International Journal of Production Research 41
(7) (2003) 15191532.
[43] A. Ghasempoor, J. Jeswiet, T.N. Moore, Real time implementation of on-line
tool condition monitoring in turning, International Journal of Machine Tools
and Manufacture 39 (1999) 18831902.
[44] S.C. Lin a, R.J. Lin, Tool wear monitoring in face milling using force signals,
Wear 198 (1996) 136142.
[45] S. Li, M.A. Elbestawi, Fuzzy clustering for automated tool condition
monitoring in machining, Mechanical Systems and Signal Processing 10
(5) (1996) 533550.
[46] Xiaoli Li, Yuan Zhejun, Tool wear monitoring with wavelet packet transform-
fuzzy clustering method, Wear 219 (2) (1998) 145154.
[47] Larry P. Heck, Kenneth C. Chou, Gaussian mixture model classiers for
machine monitoring, ICASSP-94, IEEE International Conference on Acoustics
Speech and Signal Processing 6 (6) (1994) VI/133VI/136.
[48] R. Kothamasu, S.H. Huang, W.H. Verduin, Comparison of computational
intelligence and statistical methods in condition monitoring for hard
turning, International Journal of Production Research 43 (3) (2005) 597610.
[49] Lane M.D. Owsley, Les E. Atlas, Gary D. Bernard, Self-organizing feature maps
and hidden Markov models for machine-tool monitoring, IEEE Transactions
on Signal Processing 45 (11) (1997) 27872796.
[50] Huseyin Metin Ertunc, Cuneyt Oysu, Drill wear monitoring using cutting
force signals, Mechatronics 14 (2004) 533548.
[51] P. Baruah, R.B. Chinnam, HMMs for diagnostics and prognostics in machining
processes, International Journal of Production Research 43 (6) (2005)
12751293.
[52] J. Sun, G.S. Hong, Y.S. Wong, M. Rahman, Z.G. Wang, Effective training data
selection in tool condition monitoring system, International Journal of
Machine Tools and Manufacture 46 (2006) 218224.
[53] Z.J. Pasek, Exploration of rough sets theory use for manufacturing process
monitoring, IMechE 2006 Proceedings ImechE, vol. 220, Part B: Journal of
Engineering Manufacture, pp. 335374.
[54] Alan V. Oppenheim, Alan S. Willsky, S. Nawab Nawab, Syed ham
Nawab, Signals and Systems, second ed., Prentice-Hall, Englewood Cliffs,
NJ, 1996.
[55] Y.S. Tarng, B.Y. Lee, A sensor for the detection of tool breakage in NC milling,
Journal of Materials Processing Technology 36 (1993) 259272.
[56] Rixin Zhu, Shiv G. Kapoor, Richard E. DeVor, A model-based monitoring and
fault diagnosis methodology for free-form surface machining process,
Journal of Manufacturing Science and EngineeringTransactions of the
ASME 125 (August) (2003).
[57] D. Gabor, Theory of communication, Journal of the Institutional Electrical
Engineering 93 (III) (1946) 429457.
[58] M.J. Dowling, Application of non-stationary analysis to machinery monitor-
ing, ICASSP-93, IEEE International Conference on Acoustics, Speech, and
Signal Processing 1 (1993) 5962.
[59] S.G. Mallat, A Wavelet Tour of Signal Processing, second ed., Academic Press,
New York, 1999.
[60] Ya Wu, R. Du, Feature extraction and assessment using wavelet packets for
monitoring of machining process, Mechanical Systems and Signal Processing
10 (1) (1996) 2953.
[61] K. Mori, N. Kasashima, J.C. Fu, K. Muto, Prediction of small drill bit breakage
by wavelet transforms and linear discriminant functions, International
Journal of Machine Tools and Manufacture 39 (1999) 14711484.
[62] W. Gong, T. Obikawa, T. Shirakashi, Monitoring of tool wear states in turning
based on wavelet analysis, JSME International Journal (Series C) 40 (1997)
447453.
[63] M.C. Yoon, D.H. Chin, Cutting force monitoring in the end milling operation
for chatter detection, IMechE 2005 Proceedings IMechE, vol. 219, Part B:
Journal of Engineering Manufacture, pp. 455465.
[64] S.G. Mallat, A theory of multiresolution signal decomposition: the wavelet
representation, IEEE Transactions on Pattern and Machine Intelligence 11 (7)
(1989) 674693.
[65] I. Daubechies, The wavelet transformation, timefrequency localization
and signal analysis, IEEE Transactions on Information Theory 36 (1990)
9611005.
[66] I. Daubeches, Ten Lectures on Wavelets, SIAM, Philadelphia, PA, 1992.
[67] I.N. Tansel, C. Mekdeci, O. Rodriguez, B. Uragun, Monitoring drill conditions
with wavelet based encoding and neural networks, International Journal of
Machine Tools and Manufacture 33 (1993) 559575.
[68] I.N. Tansel, C. Mekdeci, C. McLaughlin, Detection of tool failure in end
milling with wavelet transformations and neural networks (WT-NN),
International Journal of Machine Tools and Manufacture 35 (August)
(1995) 11371147.
[69] M.V. Wickerhauser, R.R. Coifman, Entropy based methods for best
basis selection, IEEE Transactions on Information Theory 38 (2) (1992)
719746.
[70] Z.K. Peng, F.L. Chu, Review: application of the wavelet transform in machine
condition monitoring and fault diagnostics: a review with bibliography,
Mechanical Systems and Signal Processing 18 (2004) 199221.
[71] D.L. Donoho, De-noising by soft-thresholding, IEEE Transactions on Informa-
tion Theory 41 (3) (1995) 613627.
[72] D.L. Donoho, I.M. Johnstone, Adapting to unknown smoothness via wavelet
shrinkage, Journal of the American Statistical Association 90 (432) (1995)
12001224.
[73] Isa Yesilyurt, End mill breakage detection using mean frequency analysis of
scalogram, International Journal of Machine Tools and Manufacture 46 (34)
(2006) 450458.
[74] M.K. Khraisheh, C. Pezeshki, A.E. Bayoumi, Time series based analysis for
primary chatter in metal cutting, Journal of Sound and Vibration 180 (1)
(1995) 6787.
[75] Peter W. Tse, Wen-Xian Yang, H.Y. Tam, Machine fault diagnosis through an
effective exact wavelet analysis, Journal of Sound and Vibration 277 (2004)
10051024.
[76] X. Li, S. Dong, Z. Yuan, Discrete wavelet transform for tool breakage
monitoring, International Journal of Machine Tools and Manufacture 39
(1999) 19351944.
[77] J.C. Fu, C.A. Troy, K. Mori, The prediction of small drill bit breakage by
wavelet based joint time-scale analysis, CIRP ProceedingsManufacturing
Systems 25 (1996) 385389.
[78] C.S. Suh, P.P. Khurjekar, B. Yang, Characterisation and identication of
dynamic instability in milling operation, Mechanical Systems and Signal
Processing 16 (5) (2002) 853872.
[79] B.S. Berger, I. Minis, J. Harley, M. Rokni, M. Papadopoulos, Wavelet based
cutting state identication, Journal of Sound and Vibration 213 (5) (1998)
813827.
[80] F. Klocke, M. Reuber, H. Kratz, Application of a wavelet-based signal analysis
for evaluating the tool state in cutting operations, Industrial Electronics
Society, IECON 2000, in: Twenty-sixth Annual Conference of the IEEE, vol. 3,
2228 October 2000, pp. 19671972.
[81] F.G. Luis Alfonso, H.R. Gilberto, P.V. Rocio, R.T. Rene de Jesus, L.T. Wbaldo,
Sensorless tool failure monitoring system for drilling machines, Interna-
tional Journal of Machine Tools and Manufacture 46 (2006) 381386.
[82] Y.M. Niu, Information-driven tool condition monitoring techniques, Ph.D.
Thesis, Department of Mechanical Engineering, Faculty of Engineering,
National University of Singapore, 2001.
[83] B. Liu, Selection of wavelet packet basis for rotating machinery fault
diagnosis, Journal of Sound and Vibration 284 (2005) 567582.
[84] S.G. Mallat, Z. Zhang, Matching pursuits with timefrequency dictionaries,
IEEE Transactions on Signal Processing 41 (1993) 33973415.
[85] J.C. Fu, C.A. Troy, P.J. Phillips, A matching pursuit approach to small drill bit
breakage prediction, International Journal of Production Research 37 (14)
(1999) 32473261.
[86] L. Cohen, Timefrequency distributiona review, Proceedings of the IEEE 77
(1989) 941981.
[87] B.W. Gillespie, L.E. Atlas, Optimizing TimeFrequency Kernels for Classica-
tion, IEEE Transactions on Signal Processing 49 (3) (2001) 485496.
ARTICLE IN PRESS
[88] S.K. Lee, P.R. White, Higher-order timefrequency analysis and its applica-
tion to fault detection in rotating machinery, Mechanical Systems and Signal
Processing 11 (4) (1997) 637650.
[89] W.J. Staszewski, K. Worden, G.R. Tomlinson, Timefrequency analysis in
gearbox fault detection using the WignerVille distribution and pattern
recognition, Mechanical Systems and Signal Processing 11 (5) (1997)
673692.
[90] Michael Kirby, Geometric Data Analysis: An Empirical Approach to
Dimensionality Reduction and the Study of Patterns, Wiley-Interscience,
New York, 2000.
[91] I. Tumer, K. Wood, I. Busch-Vishniac, 1997, Monitoring fault condition during
manufacturing using the KarhunenLoeve transform, in: ASME Mechanical
Vibration and Noise Conference, System Health Monitoring Symposium,
vol. DETC97-VIB4234.
[92] I.Y. Tumer, K.L. Wood, I.J. Busch-Vishniac, Monitoring of signals from
manufacturing processes using the KarhunenLoee Ve transform, Mechan-
ical Systems and Signal Processing 14 (6) (2000) 10111026.
[93] I.N. Tansel, O. Rodriguez, M. Trujillo, E. Paz, W. Li, Micro-end-millingI.
Wear and breakage, International Journal of Machine Tools and Manufacture
38 (1998) 14191436.
[94] I.N. Tansel, T.T. Arkan, W.Y. Bao, N. Mahendrakar, B. Shisler, D. Smith, M.
McCool, Tool wear estimation in micro-machining. Part II: neural-network-
based periodic inspector for nonmetals, International Journal of Machine
Tools and Manufacture 40 (2000) 609620.
[95] K.P. Zhu, Y.S. Wong, G.S. Hong, Noise-robust tool condition monitoring in
micro-milling with hidden Markov models, in: P. Bhanu (Ed.), Soft
Computing Applications in Industry, Studies in Fuzziness and Soft Comput-
ing 226, pp. 2346, Springer, 2008.
[96] Ali A. Houshmand, J.R. Elijah Kannatey-Asibu, Statistical process control for
cutting tool of acoustic emission monitoring, Mechanical Systems and Signal
Processing 3 (4) (1989) 405424.
[97] P.T. Huang, J.C. Chen, C.Y. Chou, A statistical approach in detecting tool
breakage in end milling operations, Journal of Industrial Technology 15 (3)
(1999).
[98] K.P. Zhu, G.S. Hong, Y.S. Wong, W.H. Wang, Cutting force denoising in micro-
milling tool condition monitoring, International Journal of Production
Research 46 (16) (2008) 43914408.
[99] M. Jansen, Wavelet denoising, Ph.D. Thesis, Department of Computer
Science, Katholieke Universiteit Leuven, Belgium, October 2001.
[100] D.L. Donoho, I.M. Johnstone, Minimax estimation via wavelet shrinkage,
Annals of Statistics 26 (1998) 879921.
[101] H.Y. Gao, A.G. Bruce, WaveShrink with rm shrinkage, Statistica Sinica 7
(1997) 855874.
[102] G.P. Nason, Wavelet shrinkage using cross-validation, Journal of the Royal
Statistical Society B 58 (1996) 463479.
[103] H.A. Chipman, E.D. Kolaczyk, R.E. McCulloch, Adaptive Bayesian wavelet
shrinkage, Journal of the American Statistical Association 92 (1997)
14131421.
[104] F. Abramovich, T. Sapatinas, Bayesian approach to wavelet decomposition
and shrinkage, in: P. Mu ller, B. Vidakovic (Eds.), Bayesian Inference in
Wavelet Based Models, Lecture Notes in Statistics, vol. 141, Springer,
New York, 1999, pp. 3350.
[105] M. Jansen, Noise Reduction by Wavelet Thresholding, Lecture Notes in
Statistics, vol. 161, Springer, Berlin, 2001.
[106] S. Menon, J.N. Schoess, R. Hamza, D. Busch, Wavelet-based acoustic emission
detection method with adaptive thresholding, Proceedings of the SPIE 3986
(2000) 7177.
[107] S.T.S. Bukkapatnam, S.R.T. Kumara, A. Lakhtakia, P. Srinivasan, The
neighborhood method and its coupling with the wavelet method for
signal separation of chaotic signals, Signal Processing 82 (2002) 13511374.
[108] S.T.S. Bukkapatnam, Soundar R.T. Kumara, A. Lakhtakia, Fractal estimation of
ank wear in turning, Journal of Dynamic Systems, Measurement, and
ControlTransactions of the ASME 122 (March) (2000) 8994.
[109] Jae-Seob Kwak, Man-Kyung Ha, Detection of dressing time using the
grinding force signal based on the discrete wavelet decomposition, The
International Journal of Advanced Manufacturing Technology 23 (12)
(2004) 8792.
[110] Jae-Seob Kwak, Application of wavelet transform technique to detect tool
failure in turning operations, The International Journal of Advanced
Manufacturing Technology 28 (1112) (2006) 10781083.
[111] Xiaoli Li, X.P. Guan, Timefrequency-analysis-based minor cutting edge
fracture detection during end milling, Mechanical Systems and Signal
Processing 18 (2004) 14851496.
[112] Hai Qiu, Jay Lee, Jing Lin, Gang Yu, Wavelet lter-based weak signature
detection method and its application on rolling element bearing prognostics,
Journal of Sound and Vibration 289 (2006) 10661090.
[113] Jing Lin, Liangsheng Qu, Journal of sound and vibration, feature extraction
based on Morlet wavelet and its application for mechanical fault diagnosis,
Journal of Sound and Vibration 234 (1) (2000) 135148.
[114] J. Lin, M.J. Zuo, K.R. Fyfe, Mechanical fault detection based on the wavelet
de-noising technique, Journal of Vibration and Acoustics 126 (2004)
916.
[115] D.B. Percival, A.T. Walden, Wavelet Methods for Time Series Analysis,
Cambridge University Press, Cambridge, 2000.
[116] A. Antoniadis, D. Leporini, J.C. Pesquet, Wavelet thresholding for some
classes of non-Gaussian noise, Statistica Neerlandica 56 (4) (2002) 434453.
[117] R. Averkamp, C. Houdr, Wavelet thresholding for non-necessarily Gaussian
noise: idealism, Annals of Statistics 31 (2003) 110151.
[118] J.F. Cardoso, Blind signal separation: statistical principles, Proceedings of the
IEEE 9 (10) (1998) 20092025.
[119] A. Hyvarinen, J. Karhunen, E. Oja, Independent Component Analysis, Wiley,
New York, 2001.
[120] G.S. Hong, M. Rahman, Q. Zhou, Using neural networks for tool condition
monitoring based on wavelet decomposition, International Journal of
Machine Tools and Manufacturing (1996) 3655.
[121] Q. Zhou, G.S. Hong, M. Rahman, A new tool life criterion for tool condition
monitoring using a neural network, Engineering Applications of Articial
Intelligence 8 (5) (1995) 579588.
[122] S.V. Kamarthi, S. Pittner, Fast Fourier and wavelet transform for ank wear
estimationa comparison, Mechanical Systems and Signal Processing 11
(November) (1997) 791809.
[123] S. Pittner, V. Kamarthi, Feature extraction from wavelet coefcients for
pattern recognition tasks, IEEE Transactions on Pattern Analysis and
Machine Intelligence 21 (1) (1999) 8388.
[124] S.V. Kamarti, S.R.T. Kumara, P.H. Cohen, Flank wear estimation in turning
through wavelet representation of acoustic emission signals, Journal of
Manufacturing Science and EngineeringTransactions of the ASME 122
(2000) 1219.
[125] Y. Choi, R. Narayanaswami, A. Chandra, Tool wear monitoring in ramp cuts in
end milling using the wavelet transform, International Journal of Advanced
Manufacturing Technology 23 (2004) 419428.
[126] Ya Wu, Philippe Escande, R. Du, A new method for real-time tool condition
monitoring in transfer machining stations, Journal of Manufacturing Science
and Engineering 123 (May) (2001) 339347.
[127] S.T.S. Bukkapatnam, S.R.T. Kumara, A. Lakhtakia, Analysis of acoustic
emission in machining, Journal of Manufacturing Science and Engineer-
ingTransactions of the ASME 121 (1999) 568576.
[128] C. Scheffer, P.S. Heyns, Wear monitoring in turning operations using
vibration and strain measurements, Mechanical Systems and Signal
Processing 15 (6) (2001) 11851202.
[129] Litao Wang, Mostafa G. Mehrabi, Elijah Kannatey-Asibu, Hidden Markov
model-based tool wear monitoring in turning, Journal of Manufacturing
Science and EngineeringTransactions of the ASME 124 (August) (2002)
652658.
[130] I. Abu-Mahfouz, Drilling wear detection and classication using vibration
signals and articial neural network, International Journal of Machine Tools
and Manufacture 43 (7) (2003) 707720.
[131] Toshiyuki Obikawa, Jun Shinozuka, Monitoring of ank wear of coated tools
in high speed machining with a neural network ART2, International Journal
of Machine Tools and Manufacture 44 (2004) 13111318.
[132] X. Li, Y. Yao, Z. Yuan, On-line tool condition monitoring system with wavelet
fuzzy neural network, Journal of Intelligent Manufacturing 8 (1997)
271276.
[133] K.P. Zhu, Y.S. Wong, G.S. Hong, Multi-category micro-milling tool wear
classication with continuous hidden Markov models, Mechanical Systems
and Signal Processing 23 (2009) 547560.
[134] S. Theodoridis, K. Koutroumbas, Pattern Recognition, Academic Press, New
York, 2003.
[135] C. Wang, R.X. Gao, Wavelet transform with spectral post-processing for
enhanced feature extraction, IEEE Transactions on Instrumentation and
Measurement 52 (2003) 12961301.
[136] B. Liu, S.F. Ling, On the selection of informative wavelets for machinery
diagnosis, Mechanical Systems and Signal Processing 13 (1) (1999) 145162.
[137] S.G. Mallat, W.L. Hwang, Singularity detection and processing with wavelets,
IEEE Transactions on Information Theory 38 (1992) 617643.
[138] S. Mallat, S. Zhong, Characterization of signals from multiscale edges, IEEE
Transactions on Pattern Analysis and Machine Intelligence 14 (7) (1992).
[139] Xiaozhi Chen, Beizhi Li, Acoustic emission method for tool condition
monitoring based on wavelet analysis, International Journal of Advanced
Manufacturing Technology 33 (2007) 968976.
[140] R.T. Ogden, J.D. Lynch, Bayesian analysis of change-point models, Lecture
Notes in Statistics, vol. 141, 1999, pp. 6782.
[141] W.J. Staszewski, Identication of non-linear systems using multi-scale ridges
and skeletons of the wavelet transform, Journal of Sound and Vibration 214
(4) (1998) 639658.
[142] A. Hambaba, A.E. Huy, in: IEEE Aerospace Conference Proceedings, Multi-
resolution error detection on early fatigue cracks in gears, vol. 6, 2000, pp.
367372.
[143] Qiao Sun, Ying Tang, Wei Yang Lu, Yuan Ji, Feature extraction with discrete
wavelet transform for drill wear monitoring, Journal of Vibration and
Control 11 (11) (2005) 13751396.
[144] S. Loutridis, A. Trochidis, Classication of gear faults using Hoelder
exponents, Mechanical Systems and Signal Processing 18 (2004) 10091030.
[145] A.N. Robertson, C.R. Farrar, H. Sohn, Singularity detection for structural
health monitoring using holder exponents, Mechanical Systems and Signal
Processing 17 (6) (2003) 11631184.
[146] Z. Peng, Y. He, Z. Chen, F. Chu, Identication of the shaft orbit for rotating
machines using wavelet modulus maxima, Mechanical Systems and Signal
Processing 16 (2002) 623635.
[147] Z.K. Peng, F.L. Chu, Peter, W. Tse, Singularity analysis of the vibration signals
by means of wavelet modulus maximal method, Mechanical Systems and
Signal Processing 21 (2) (2007) 780794.
ARTICLE IN PRESS
[148] Kenneth C. Chou, Larry P. Heck, A Multiscale Stochastic Modeling Approach
to the Monitoring of Mechanical Systems, IEEE Press, New York, 1994.
[149] A.A. Safavi, J. Chen, J.A. Romagnoli, Wavelet-based density estimation and
application to process monitoring, Articial Intelligence in Chemical
Engineering 43 (5) (1997) 12271241.
[150] K. Krishnan Nair, Anne S. Kiremidjian, Time series based structural damage
detection algorithm using Gaussian mixtures modeling, Journal of Dynamic
Systems, Measurement, and ControlTransactions of the ASME 129 (May)
(2007) 285293.
[151] W. Hardle, G. Kerkyacharian, D. Picard, A. Tsybakov, Wavelets, Approxima-
tion and Statistical Applications, Lecture Notes in Statistics, vol. 129,
Springer, New York, 1998.
[152] Brani Vidakovic, Statistical Modeling by Wavelet, Wiley, New York,
1999.
[153] M. Vannucci, Nonparametric density estimation using wavelets: a review.
Discussion Paper 95-26, ISDS, Duke University, USA, 1995.
[154] F. Abramovich, T.C. Bailey, T. Sapatinas, Wavelet analysis and its statistical
applications, The Statistician 49 (2000) 129.
ARTICLE IN PRESS

1 s2.0 S0890695509000418 Main

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 s2.0 S0890695509000418 Main

Uploaded by

Copyright:

Available Formats

Wavelet analysis of sensor signals for tool condition monitoring:

A review and some new results

, Wong Yoke San, Hong Geok Soon

Fig. 3. Timefrequency resolutions of Fourier transform and wavelet transform.

plogA alogs (23)

You might also like