Tool Monitoring Using Image Processing

Published on January 2017 | Categories: Documents | Downloads: 71 | Comments: 0 | Views: 431
of 21
Download PDF   Embed   Report

Comments

Content


Review
Application of digital image processing in tool condition monitoring:
A review
S. Dutta
a
, S.K. Pal
b,
*, S. Mukhopadhyay
c
, R. Sen
a
a
CSIR-Central Mechanical Engineering Research Institute, Durgapur, India
b
Mechanical Engineering Department, Indian Institute of Technology, Kharagpur, India
c
Electronics and Electrical Communication Engineering Department, Indian Institute of Technology, Kharagpur, India
Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
1.1. Advantages and disadvantages of DIP for tool condition monitoring. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
2. Digital image processing techniques. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
3. Lighting systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
4. Direct TCM techniques using image processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
4.1. Two dimensional techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
4.2. Three dimensional techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
5. Indirect TCM techniques using image processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
5.1. Online techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
5.2. Offline techniques. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
1. Introduction
In any machining process, high quality of the final product is the
ultimate aim. The trend towards automation in machining has
been driven by the need to maintain high product quality with
improving production rate and the potential economic benefits of
automation in machining are significant as well. These process
improvements can be possible by monitoring and control of
machining process. Tool condition monitoring (TCM) is very much
inevitable for reducing machine tool downtime. Reduction of
machine tool downtime improves production rate, significantly.
Excessive wear and breakage of the cutting tool is one severe cause
of downtime. Dull or damaged cutting tool can put extra strain on
the machine tool as well as surface finish of the machined part.
Cutting speed can increase 10–50% with appropriate TCM
techniques [105]. In a TCM system, acquisition of machining
process data viz. cutting force, sound energy, power, current,
surface finish, vibration, temperature, etc., which are influenced by
cutting tool geometry and machining process conditions, has been
performed through high level intelligent sensors viz. dynamome-
ter, acoustic emission sensor, power and current sensor, surface
profiler or vision based system, accelerometer, pyrometer [121].
The acquired sensory information are then filtered and processed
through signal processing and some relevant features are extracted
CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232
A R T I C L E I N F O
Article history:
Available online 4 April 2013
Keywords:
Tool condition monitoring
Digital image processing
Surface texture
Tool wear
A B S T R A C T
Tool condition monitoring is gaining a parallel development with the advancement of automatic
manufacturing processes in the last thirty years due to the increasing need for improvement of product
quality. The advances of digital image processing techniques used in tool condition monitoring are an
important research interest due to the improvement of machine vision system, computing hardware and
non-tactile application. In this paper, a reviewof development of digital image processing techniques in
tool condition monitoring is discussed and finally a conclusion is drawn about required systematic
research in this field.
ß 2013 CIRP.
* Corresponding author. Tel.: +91 3222 282996; fax: +91 3222 255303.
E-mail address: [email protected] (S.K. Pal).
Contents lists available at SciVerse ScienceDirect
CIRP Journal of Manufacturing Science and Technology
j ou r nal h o mepage: w ww. el sevi er . co m/ l oc at e/ c i r p j
1755-5817/$ – see front matter ß 2013 CIRP.
http://dx.doi.org/10.1016/j.cirpj.2013.02.005
from the results of signal processing techniques. Then prediction of
process data and process optimization can be possible using design
of experiment (DoE) and artificial intelligence (AI) techniques from
the extracted and selected features. Comparison of actual and
predicted values of selected features are also required to find out
the precision of that technique. Then optimized data are fed to the
machine controller and servo mechanism which can control the
machining process. Elbestawi et al. [34] comprehensively classified
different sensor systems for monitoring different output process
parameters viz. dimensions, cutting force, feed force, spindle
motor and acoustic emissions used in turning, milling and drilling
operations. Two excellent case studies have been conducted by
them using proposed multiple principal component fuzzy neural
network for classification of sharp tool, slightly worn tool, medium
worn tool, severe worn tool and breakage in turning and drilling
experiment using force, vibration and power signal. An online
monitoring of chipping in drilling process has also been conducted
by them using vibration signal with 97% success rate. Roth et al.
[106] emphasized wireless, integrated and embedded low cost
sensors; wavelet, time-frequency and time-scale analysis as a
signal processing approach; artificial neural network (ANN) and
support vector machine approach for assessment of tool condition;
hidden Markov model and recurrent neural network for the
prediction purpose in their comprehensive review of TCM for
turning, milling, drilling and grinding processes. Nebot and
Subiro´ n [92] reviewed the TCM systems of machining and
proposed a generic methodology combining DoE and ANN for
improved process modelling and prediction. Teti et al. [121] made
a comprehensive review on intelligent sensors for monitoring and
control of advanced machining operation. They also mentioned the
real industrial implementation of the intelligent sensor systems for
TCM of advanced machining of complex-shaped parts made of
super alloy. Chandrasekaran et al. [19] made an comprehensive
literature review on the application of soft computing techniques
viz. neural network, fuzzy logic, genetic algorithm, simulated
annealing, ant colony optimization and particle swarm optimiza-
tion on turning, milling, grinding and drilling operations for
optimization of cutting conditions with minimum cost machining
with maximum production rate based on prediction of process
outputs viz. surface finish, cutting force and tool wear.
The product quality is principally dependent on the machined
surface. The surface quality is mainly dependent on the cutting tool
wear. Cutting tool wear is dependent upon cutting conditions,
work and tool material, tool geometry. There are four modes of
cutting tool wears, such as, adhesive wear due to shear plane
deformation, abrasive wear due to hard particles cutting, diffusion
wear due to high temperature and fracture wear due to fatigue.
Four principal types of wear occur in cutting tool and they are nose
wear, flank wear, crater wear and notch wear. Flank wear (as
shown in Fig. 1) occurs due to rubbing between tool flank surface
and work piece. Flank wear is specified by maximum flank wear
width (VB
max
) or mean flank wear width (VB
mean
). Tool life
criterion is mainly dependent on the VB
mean
. Cutting tools are
experiencing three stages of wear [29] viz. initial wear (during first
few minutes), steady-state (cutting tool quality slowly deterio-
rates) and severe wear (rapid deterioration as the tool reaches the
end of its life). Crater wear are produced at the due the high
temperature for chip-tool interaction. This wear is characterized
by the crater depth and crater area.
Principally, tool condition monitoring systems can be classified
into two groups. They are, (a) direct techniques and (b) indirect
techniques. In direct techniques, flank wear width, crater depth
and crater area are measured directly either with tool maker’s
microscope, 3D surface profiler, optical microscope or scanning
electron microscope (off-line method) or with CCD camera (in-
process method). In indirect techniques, the measured parameters
or signals (viz. force, acoustic emission, current, power, surface
finish, etc.) of the cutting process allow for drawing conclusions
upon the degree of tool wear. Normally, these tool wear
monitoring systems are based upon the comparison of a reference
signal of an optimized cutting process with the actual process
signal [127]. These techniques have predominantly been imple-
mented, employing such varied technologies as acoustic emission,
cutting force, spindle current, and vibration sensors [99]. However,
there are some limitations of these methods. To overcome those
limitations, research is going on to identify the degree of tool wear
by analyzing surface texture of machined surfaces with digital
image processing technique from the images of machined surfaces.
There is a wide range of application of digital image processing
(DIP) using machine vision in machining processes like control of
surface quality, tool wear measurements, work piece surface
texture measurements, etc.
1.1. Advantages and disadvantages of DIP for tool condition
monitoring
There are some advantages of using digital image processing
techniques over other techniques to monitor any manufacturing
process. Such as, (1) it applies no force or load to the surface texture
under examination; (2) it is a non-contact, in-process application
[63]; (3) this monitoring system is more flexible and inexpensive
than other systems; (4) this system can be operated and controlled
from a remote location, so it is very much helpful for unmanned
production system; (5) this technique is not dependent on the
frequency of the chatter, directionality as acoustic emission (AE)
sensors are dependent on those factors; also, the AE sensors are
mainly detecting tool breakage in machining [17,102,29]. Thus, the
monitoring of progressive wears of cutting tool is very difficult
using AE sensors; (6) vibration sensors (accelerometer) can
monitor tool breakage, out of tolerance parts and machine
collisions [52]; the progressive wear monitoring has not been
possible using vibration sensors; (7) DIP technique is not affected
Fig. 1. Flank wear and notch wear from the microscopic image of a tool insert.
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 213
by the high frequency forces as this high frequency forces cannot
be taken by dynamometer; also the force sensors are sensitive to
machine vibrations [53]; (8) to monitor and control a machining
process, the fusion of several sensors (AE sensor, dynamometer,
vibration signatures, etc.) is required, which is not at all cost
effective [52]; (9) however, the machined surface image carries the
information of tool imprint as well as the change of tool geometry
[9]; thus, a roughness, waviness and form information can be
obtained by analyzing a machined surface image [15]; (10) a 2D
information can be obtained from a machined surface image which
is not possible to get by a 1D surface profiler [122]; (11) also, the
information of machining parameters can be obtained from
machined surface images [31]; (12) the development of CCD
cameras has also contributed to the acceptance of industrial image
processing, since CCD cameras are less sensitive to the adverse
industrial environment; (13) optical image processing has brought
about the possibility of adding, subtracting, multiplying, storing
and even performing different image transforms using optical
devices; (14) three dimensional surface roughness of machined
surface can be measured, accurately, using scanning type 3D
surface profiler [1,23,88,95]; however, these 3D measurements are
not effective for in-process or online tool condition monitoring due
to uneconomic time, cost ineffectiveness and inaccessibility to the
machine tools; to overcome this situation, a machine vision based
system can be useful for monitoring purpose. However, there are
some limitations for using machine vision system in tool condition
monitoring techniques also [141]. (1) An appropriate illumination
system, robust image processing algorithm, protection from
machining noises (chips, dirts, etc.) are very much essential for
the successful implementation of this technique [9]. (2) Monitor-
ing of drill parts using DIP are very difficult due to its inaccessibility
[51]. However, a method to monitor deep hole parts has been
developed in recent years [84].
This paper is composed of five major components. The first
component presents an overview of digital image processing
techniques used for tool condition monitoring. The second
explains lighting systems which are used in TCM. The third
presents direct TCM techniques using digital image processing. The
fourth component presents different in-direct TCM techniques
using image processing. And the final and last component draws
overall conclusions and suggests future directions for TCM
research through digital image processing technique.
2. Digital image processing techniques
Image acquisition is the first step of any machine vision system.
In case of TCM, images of cutting tool (rake face or flank surface) or
work piece surface are captured with a CCD (Charged Coupled
Device) camera or CMOS (Complementary Metal-Oxide Semicon-
ductor) digital camera. CCD camera is comprised of CCD sensor
which is an array of photosensitive elements to collect electrical
charges generated by absorbed photons. Those electrical charges
are then converted to an electrical signal which is converted to a
digital image via frame grabber. Finally, the image is transferred to
a PC for processing purpose [50]. CMOS is different from CCD
sensor by its faster capturing rate. CMOS sensor can acquire frames
faster than CCD camera. But the sensitivity of CMOS sensor is much
less than CCD sensor. To create a digital image, a conversion is
needed from the continuous sensed data into digital form. This
involves two processes: sampling and quantization. Digitization of
coordinate values and amplitude values are called sampling and
quantization. Image magnification is also possible by linear
interpolation, cubic interpolation, cubic convolution interpolation
etc. Different types of neighbourhood operations are also needed
for further processing [41].
From the illumination point of view, an Image f(x, y) may be
characterized by two components: (1) the amount of source
illumination incident on the scene, and (2) the amount of
illumination reflected by the objects. Appropriately, these are
called the illumination and reflectance components and are
denoted by i(x, y) and r(x, y), respectively. The two functions
combine as a product to form f(x, y),
f ðx; yÞ ¼ iðx; yÞrðx; yÞ (1)
Image pre-processing is required for the improvement of
images by contrast stretching, histogram equalization, noise
reduction by filtering, inhomogeneous illumination compensa-
tion etc. To increase contrast in an image, contrast stretching and
histogram equalization are two mostly used techniques. To
reduce noise, low pass filtering is very important technique. It
includes image smoothing by using low pass filtering in both
spatial and frequency domains. In spatial low pass filtering, a
filter mask is convolved with the image matrix to reduce
unwanted noise present in the image (image smoothing). Order
statistics or median filter is used to remove impulse noise in an
image (image smoothing). Butterworth and Gaussian low pass
filters are some common low pass filters in frequency domain.
High pass filters are used to enhance the sharpness of an image
(image sharpening). Unsharp masking (to emphasize high
frequency components with retaining low frequency compo-
nents), Laplacian filter (second order filter) are some spatial high
pass filters used for image sharpening purpose [41]. Image
filtering and enhancement operations are very much essential to
reduce the noise of the images specially for cutting tool images,
because there are a chance of noise due to the dirt, oils, dust of
machining on the object surface. The low-pass filtering (e.g.
median filter, Gaussian filter, etc.) is useful to reduce the noises
present in the cutting tool wear images and machined surface
images. Also the high pass filtering technique can be useful to
enhance tool wear profile and for clear identifications of feed
marks in machined surface images.
After pre-processing, image segmentation and edge detection
are generally done to segment the worn region of cutting tool
from the unworn region and also to detect the edges of the feed
lines of the machined surface images. Image segmentation is the
method of partitioning an image into multiple regions according
to a given criterion. Feature-state based techniques collect pixel/
region properties into feature vectors and then use such vectors
for assigning them to classes, by choosing some threshold
values. While feature-state based techniques do not take into
account spatial relationships among pixels, image-domain based
techniques do take them into account; for example, split and
merge techniques divide and merge adjacent regions according
to similarity measurements; region growing techniques aggre-
gate adjacent pixels starting fromrandomseeds (region centres),
again by comparing pixel values. Watershed-based segmenta-
tion technique can be useful for micro and nano surface
topography. Watershed analysis, which consists in reasoning
over a surface topography in terms of hills and dales, actually
originates from the work by Maxwell on geographical analysis.
Watershed-based surface segmentation consists in partitioning
the surface topography into regions classified as hills (areas from
which maximum uphill paths lead to one particular peak) or
dales (areas from which maximum downhill paths lead to one
particular pit), the boundaries between hills being watercourse
lines, and the boundaries between dales being watershed lines
[2].
The edge detection operation is used to detect significant edges
of an image by calculating image gradient and direction. Gradient
and direction of an image f(x, y) are defined in Eqs. (2) and (3),
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 214
respectively.
G
x
G
y
_ _
¼
d f
dx
d f
dy
_
¸
¸
_
_
¸
¸
_
(2)
uðx; yÞ ¼ tan
À1
G
y
G
x
_ _
(3)
where u is measured with respect to the x-axis.
Robert operator (sensitive to noise), Sobel operator, Prewitt
edge operator are some first order edge detectors which are very
useful for automatic detection of tool wear profile. Canny edge
detector is widely used in the field of machine vision because of its
noise immunity and capability to detect true edge points with
minimum error. In Canny edge detection method, the image is first
convolved with Gaussian smoothing filter with standard deviation
s. This operation is followed by gradient computation on the
resultant smoothed image. Non-maxima suppression, double
thresholding and edge threshold selection with Bayes decision
theory are the steps to implement Canny edge detection. Gradient
images of tool flank wear (experimentally obtained from milling
operation) and machined surface (experimentally obtained from
turning operation) using Canny edge detector are shown in Fig. 2. A
wear profile or edges of surface texture can be obtained by this
method. The edge detector based on double derivative is used to
detect only those points as edge points which possess local
maxima in the gradient values. Laplacian and Laplacian of Gaussian
are the most commonly used double derivative-based edge
detectors.
For partitioning a digital image into multiple regions, grey level
thresholding techniques are computationally inexpensive. Based
on some optimal threshold, an image can be partitioned into
multiple regions. For example, to partition the flank wear profile
from its background, thresholding techniques are generally used. A
very common thresholding technique used in tool wear measure-
ment is Otsu’s optimal thresholding technique. In this technique, a
class, C
0
is formed with all the grey value V(k) for a grey level
intensity, k and all the other form another class, C
1
. Optimal k value
is selected for maximum between-class variance. In bi-level
thresholding technique images are partitioned into foreground and
background segments and in multilevel or dynamic thresholding,
images are divided into more than two segments. In entropy-based
thresholding, the threshold value is selected in such a way, so that
the total entropy value of foreground and background is maximum
[2]. Thresholding techniques are important for binarization of flank
wear profile.
After edge detection and thresholding, morphological opera-
tions viz. erosion, dilation, closing, opening are important tools for
completing the wear profile, accurately. In this operation, a
noiseless morphology is obtained by introducing or removing
some points or grey values in a profile [41].
Tool condition monitoring via surface texture of machined
parts are mainly dependent on the texture analysis method. This
method can be applied after pre-processing. Texture is a repeated
pattern, whichis a set of local statistics or attributes vary slowly or
remain approximately periodic. Primitive in texture is a con-
nected set of pixels, characterized by a set of attributes
(coarseness and directionality). For example, in case of turned
surface, a repetitive feed marks can be obtained as texture
primitives. Texture analysis can be done using statistical,
geometrical, model-based and signal processing based methods.
In statistical method a texture is modelled as a randomfield and a
statistical probability density function model is fitted to the
spatial distribution of intensities in the texture. Higher-order
statistics like run-length statistics, second order statistics like
grey level co-occurrence matrix (GLCM) can be used as statistical
texture classifiers. In geometric texture analysis method, the
analysis depends upon the geometric properties of texture
primitives. Voronoi tessellation, Zucker’s model are some of the
geometric texture analysis methods. In model based methods,
texture analysis is done with some signal model like, Markov
random field, Gibbs random field, Derin-Elliot, auto-binomial,
fractal (self-similarity) models are some mathematical model-
based texture analysis methods. In signal-processing based
texture analysis, spatial domain filtering, Fourier-domain filter-
ing, Gabor and wavelet analysis are some common texture
analysis methods [125].
3. Lighting systems
Lighting system is the most important and critical aspect to
receive a proper image for image processing. Due to inhomoge-
neous illumination for improper lighting set-up, the information
from images will not be sufficient for any machine vision
application. Several researches give strong importance on lighting
set-up for tool condition monitoring using image processing.
Lighting systems required are varying depending on applications
viz. for capturing tool wear image and machined surface image.
Weis [132] tried to capture the tool wear image using a diode flash
light incorporated with a infrared band filter, which helped to
enhance the tool wear region with respect to the background.
Kurada andBradley [73] usedtwofibre-optic guides to capture the
tool wear regions. They used it to obtain adequate contrast
between the worn and unworn tool regions. Pfeifer and Weigers
[99] used ring of LEDs attached with camera to capture the proper
illuminated images of tool inserts fromdifferent angle. Kim et al.
[70] used a fibre optic light surrounding the lens to illuminate the
flank face portion of a 4-fluted end mill. They also examined that
the best measurement of flank wear can be possible with a high
power lighting (60 W). Jurkovic et al. [58] utilized a halogen light
to illuminate the rake and flank face of the cutting tool and a laser
diode and accessories to obtain a structured light pattern on the
face of the tool to detect the tool wear by the deformation of
structured light on the rake face. Wang et al. [131] used a fibre
optic guided light to illuminate the flank portion of each insert
attached to a 4-fluted milling tool holder and capture the
successive images in a slow rotating condition by using a laser
trigger with very less blurring. A white light from a fluorescent
ring as well as light from a fibre bundle was used to minimize
specular reflections on capture the tool images by Kerr et al. [68].
So, highly illuminated and directional lighting is required to
capture the tool wear region as to get a very accurately
illuminated image. Wong et al. [134] used a 5 mW He–Ne laser
0.8 mm-diameter beamfor focusing onto the machined surface by
a lens at an incident angle of 308 for capturing the centre of the
pattern. Then the reflected light pattern was formed on a screen
made of white coated glass from where the scattered pattern was
grabbed using a CCD camera. The setup was covered in order to
minimize interference from ambient light and a consistent
lighting condition for all the tests has been provided. But the
actual image of the machined surface is required instead of
reflectedpattern. Tsai et al. [123], triedto obtaina homogeneously
illuminated machined surface image by a regular fluorescent light
source which was situated at an angle of approximately 108
incidence with respect to the normal of the specimen surface. The
camera was also set up at an angle of approximately 108 with
respect to the normal of the specimen surface to obtain image at
the direction of light. But this set-up may only be useful for flat
specimens not for curved surfaces. Bradley and Wong [16] used a
fibre optic guided illumination source and a lighting fixture. A
uniform illumination of the machined surface was ensured by
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 215
changing the position of lighting fixture. During surface assess-
ment, the specimen was positioned on the platformso that the lay
marks were perpendicular to the longer dimension of the CCD
sensor. The light source was positioned at a distance of 8 cm from
the surface, as this provided the best image contrast. In this
technique, the images of flat specimens (end milled) were
captured but the images of turned surface (i.e. curved surfaces)
were not obtained. Lee et al. [78] used a diffused, blue light source
situated at an angle of approximately 458 incidence with respect
to the machined (turned) surface specimen to accomplish the
illumination of the specimens. Alegre et al. [4], explained about a
diffusedlighting system(a DC regulated light source with infrared
interference filter for cool illumination) for capturing images of
turned parts. They also used a square continuous diffused
illuminator for getting diffused illumination in the camera axis.
The last lighting system is most appropriate for obtaining a
homogeneously illuminated image of turned or curved parts. A
cover can be used to reduce the interference of ambient lighting in
industrial environment.
4. Direct TCM techniques using image processing
There are two predominant wear mechanisms for a cutting
tool’s useful life: flank wear and crater wear. Flank wear occurs
on the relief face of the tool and is mainly attributed to the
rubbing action of the tool on the machined surface. Crater wear
occurs on the rake face of the tool and changes the chip-tool
interface, thus affecting the cutting process. Tool wears increases
progressively during machining. It depends on the type of tool
material, cutting conditions and lubricant selected. Online
measurement of tool wear by image processing after taking
images of cutting tool through machine vision system is under
research. This technique is coming under the area of direct tool
condition monitoring. Flank wear can directly be determined by
capturing images of cutting tool but a more complex technique is
required to determine the crater depth [59]. Cutting tool wears
have been measured by two dimensional and three dimensional
techniques in various researches which are described in the
following sections.
Fig. 2. (a) Milling tool wear image and (b) corresponding gradient image using Canny edge detector (c) turned surface image and (d) corresponding gradient image using
Canny edge detector.
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 216
4.1. Two dimensional techniques
Flank wears are determined by two dimensional techniques.
Kurada and Bradley [74] made a review on advances of machine
vision sensors which are used to obtain information about the
cutting tool and machined part. They made the comparison of
advancement of machine vision techniques up to previous decade.
They emphasized the laboratory level development.
Kurada and Bradley [73] did a pioneering work for direct tool
condition monitoring by capturing images of tool flank wear by
using two fibre optic guided lights and CCD camera. Both lights
were adjusted for illuminating the tool flank wear region. They first
calibrated the image in terms of two factors in horizontal and
vertical direction to convert pixel unit to length unit (micron). In
their work, they used texture-based image segmentation tech-
nique step by step using image enhancement (using cascaded
median filter) to reduce noise, image segmentation to extract the
flank wear region from background (using variance operator),
global thresholding, feature extraction by morphological operation
(blob analysis) and flank wear calculation (by boundary and
regional descriptors). However, they tried it out in offline using
video zoom microscope.
In case of offline techniques, all the time cutting inserts or
cutting tools has to be disengaged from the machine tool. Thus, this
is very much time consuming and may be erroneous for proper
alignment of the cutting tool. For this reason, Weis [132] did a
pioneering work to capture the tool wear region of a milling insert
without disengaging the insert from the tool holder. Also the tool
wear region has been enhanced and the background has been
faded out with the help of an infra red band filter at the time of
image acquisition. Diode flash light has also been synchronized
with the CCD camera to capture a perfect tool wear region. A
dilation and binary operation has been applied on flank face image
to measure the flank wear width. They mainly have given the
importance to the lighting system for online capturing of tool wear
images, accurately. However, image processing methods in their
technique has been given a second priority. Tauno and Lembit
[120] developed a software for detection of flank wear using non-
linear median filter to remove noise and a Roberts filter operator
for edge detection. This system provided the automatic measure-
ment of surface area, average wear land length and perimeters of
flank wear profile. However, their method could not be utilized for
fully automated measurement. Pfeifer and Wiegers [99] captured
images of tool inserts with a ring light in different angles of
incidence. Then they compared those captured images and
reduced inhomogeneous illumination problem for even complex
cutting edges. However, they did not check their technique for
different wear conditions. Sortino [116] developed a flank wear
measuring software by using a new edge detection method from a
colour image. In this statistical filtering method, the neighbour-
hood pixels of a pixel were considered as a set and the mean and
standard deviation of each set have been calculated for each
fundamental colour (red, green, blue). Then, a comparison
parameter, D
edge
, has been evolved from the set parameters (i.e.
mean and standard deviation). Finally, the edge was detected
accurately for higher D
edge
values and cutting edge, borderline
between worn zone and oxidized zone, borderline between
oxidized zone and tool surface has been detected. However, the
accuracy of this wear measuring system is limited for low flank
wear width as the resolution is 10 mm. [58] used a specular
reflection by structured lighting for the appearance and charac-
terization of insert surface using projection of a line stripe to
determine the deepness and furrowness of rake or flank face of a
tool. But this method requires very much complex and costly set up
for image acquisition. Also the method did not use any three
dimensional model to show the depth profile of crater as well as
this method could not be helpful to measure the crater depth in
grooved inserts. Wang et al. [131] developed an automated system
to capture and process successive images of moving inserts to
measure flank wear in milling using cross correlation technique
between successive image pairs. The method developed by them is
a robust technique to remove noises using a novel parallel
scanning technique. However, the method is a threshold depen-
dent method where the accuracy of the measurement is dependent
on the selected threshold value and the method has not been very
much useful to measure the coated carbide insert due to the mal
interaction between the lighting and the coating material. To
recover the limitation of threshold on the accuracy of measure-
ment, they deployed another technique based on the moment
invariance to select the exact bottom portion of a flank wear profile
with maximum 15 mm error and minimum 3 mm error compared
to the measurement obtained from microscope [128–130]. They
also measured the flank wear of coated carbide insert, successfully.
Though this system was independent of the thresholding, but it
was dependent on the accuracy of a reference line with respect to
whom the flank wear width was determined. The computation
time for this method was 2 s which was not at all practical for real
time measurement. Fadare and Oni [35], evaluated flank and notch
wear using the insert images. Tool insert images were first filtered
by Weiner filtering. Length, width, area, equivalent diameter,
centroid, major axis length, minor axis length, solidity, eccentricity
and orientation were the extracted descriptors of wear. They have
taken tool insert images in a dark room with the help of two
incandescent light sources. Maximum absolute difference of
measurements between microscope and vision system was
3.13%. An overall tool wear indicator, namely, Tool Wear Index
(TWI) was also derived from the extracted wear descriptors, which
was a highly reliable tool wear indicator. A very good systematic
variation was obtained in Fadare’s work. However, C/C++
programming language can give faster result than MATLAB and
may be used for real time application. A better lighting conditions
such as fibre optic guided light or diffused ring light is required to
implement their method in industrial environment. Liang et al.
[83] utilized an image registration and mutual information based
method to recognize the change of nose radius of TiN-coated, TiCN-
coated and TiAlN-coated carbide milling inserts for progressive
milling operation. They also used the similarity metrics to describe
the nose radius. However, their method is quite difficult for the
measurement of crater and flank wear. Sahabi and Ratnam [109]
measured nose radius of the turning insert online from the
silhouette image of the cutting tool tip. They utilized median and
Weiner filtering to reduce the image noise; applied morphological
operations to reduce the noise due to micro-dust particle; used a
conformity method for reducing the misalignment error; applied
thresholding and subtraction of worn and unworn tool to measure
the nose wear area. They used this technique for turning with
various cutting speeds. However, they did not quantify the flank
wear width. To improve their method for measuring the flank wear
width in the zone nearer to the nose of the cutting tool, they
utilized the information of nose radius and machined surface
roughness profile using machine vision system in online for
turning operation [108]. They have obtained an mean deviation of
7.7% and 5.5% between the flank wear determined by their method
and flank wear measured by using tool maker’s microscope from
nose radius and surface roughness profile, respectively. However,
their method is very difficult to implement in ultra-precision
machining with low feed rate.
Kim et al. [70] has been developed a magnetic jig for fixing the
camera and lighting system to accomplish the objective of on
machine tool measurement of flank wear for a 4-fluted end mill.
They compared the signal to noise ratio of measurements using
microscope and CCD camera incorporating with a novel jig. They
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 217
inserted the fibre optic guided lighting system into the lens for
further improvement. However, this work is more prone to the
measuring system instead of image processing technique.
Kerr et al. [68] utilized four different texture analysis
techniques namely histogram based processing, grey level co-
occurrence technique, frequency domain based technique and
fractal method to analyze the texture of the worn region of turning
and milling insert. They obtained the best result by frequency
domain or Fourier spectrum analysis techniques because this
technique is position and illumination invariant. However, they
have captured the tool tip portion of turning insert instead of the
flank face portion which is not a standard practice. Jackson et al.
[49] has been proposed a novel technique for accurate edge
detection algorithm utilizing neural network technique for tool
wear detection. They have utilized the scanning electron
microscopic images of the flank wear images of a 4-fluted high
speed steel milling cutter. However, they did not do any online
monitoring using CCD or CMOS camera.
Lanzetta [76] proposed an automated and flexible vision based
sensor system incorporating the measurement and classifications
of tool wears. The resolution of their sensor was 40 mm/pixel.
However, several tests at different cutting conditions might be
required to establish this technique. Measurement of cutting
inserts with chip breaker, the effect of noises from dirt, oil chips on
the insert surfaces has not been addressed in his research. Schmitt
et al. [112] developed a flexible and automated tool flank wear
measurement system incorporating ring illuminators and CCD
camera where full illuminated and side illuminated images of flank
wear portion of cutting inserts have been captured and processed.
The full illuminated image was required for main cutting edge
detection and cutting edge corner detection whereas side-
illumination was used for flank wear profile segmentation and
wear measurement. They applied region of interest selection, Sobel
filter technique for cutting tool edge enhancement, morphological
opening and closing for reduction of enhanced noise and line
interpolation for getting the accurate cutting edge on full
illuminated image. They applied linear transformation of histo-
gram to brightened the wear area; thresholding and morphological
opening and closing for noise elimination; blob analysis for
detection of blobs situated outside the wear profile; creation of a
dynamic region of interest (ROI) to detect the best initial point for
contour detection and snake algorithm for contour detection on
side illuminated image. Then they measured the average flank
wear width and maximum flank wear width with 4.4 mm
resolution after calibrating the vision system using a chequerboard
pattern. Though their system is highly accurate but the computa-
tional time was not emphasized in their work. Stemmer et al. [117]
applied a neural network classification technique to classify the
flank wear and breakage of the cutting tool with 4% error using
image processing. They have observed by their machine vision
system that the flank wear featured a sharper and brighter textures
whereas the breakage of the tool consisted of smooth and rough
textures. Based on this phenomenon, they have classified the types
of wear and also measured flank wear area, maximum and average
flank wear width automatically, using Canny edge detection and
line interpolation, pre-filtering and blob analysis, active contour
detection by snakes algorithm, wear classification and measure-
ment with a resolution of 4.4 mm/pixel. However, the chipping,
crater wear, notch wear and nose wear were not classified in their
technique for different variety of cutting tools. A faster classifica-
tion approach was proposed by Castejo´ n et al. [18,13]. They
estimated different wear levels (low, medium, high) of the tool
insert by means of the discriminant analysis of nine geometrical
descriptors and assessed by means of the Fowlkes–Mallows index
and also Zernike, Legendre, Hu, Taubin and Flusser invariant
moments were used to characterize the shape of the worn region of
flank wear zone in the binary images to classify the three wear
levels viz. low, medium and high wear. Hu descriptor was found to
be the best one in their work. However, this technique may be
more simple and useful for indirect monitoring technique by
processing the machined surface images. Alegre et al. [6] computed
the average and maximum flank wear width based on contour
signature of the binary image of flank wear profile. Contour
signature is a vector whose elements are the distance between the
centroid of the contour and the boundary pixel points. The number
of elements can be chosen by the user. They have chosen 40 and
100 number of elements and on this basis they classified the low
wear and high wear inserts used in turning experiment using k-
nearest neighbourhood (k-NN) classification and multi layer
perceptron neural network (MLPNN) classification. Finally, they
concluded that the wear classification result was best by using the
MLPNN, average flank wear width and 40 element signature with
5.1% minimum error. However, the classification has been based on
two classes of wear only and the system calibration has not been
done in their work.
Atli et al. [11] developed a new measure namely DEFROL
(deviation from linearity) to classify between sharp and dull
drilling tool from their images. However, the emphasis has been
done on the change of point angle and linearity deviation of the
cutting edges due to the wearing effect, but no study has been
taken care regarding the flank wear in this technique. So this
technique was not suitable to measure the flank wear which was
used to define the tool life in standard practice (ISO 3685) [48].
Makki et al. [86] did a real time capturing of drill bit image at the
time of 100 rpm rotation. Then they processed those captured
images by edge detection and accurate segmentation technique to
find out the tool wear (only the deviation of the lip portion) and
tool run-out in the image plane. However, the measurement of
flank wear and tool run out perpendicular to the image plane
cannot be possible by their technique.
Liang and Chiou [82] has been introduced a flank wear
measurement technique of multilayer coated twist drill by using
image processing. They have detected the edges of the wear profile
on the cutting plane using spatial moment edge detector with sub-
pixel accuracy and also they smoothed the edges using B-spline
technique. After that they have applied the Gaussian low-pass
filtering technique for smoothing the curvature curve and finally,
applied a statistical process control measure to select the accurate
threshold value for extracting the accurate wear profile for precise
measurement of maximum flank wear width. For improving the
wear measurement technique of twist drill used for micromachin-
ing, Su et al. [118] studied the feasibility of measuring the flank
wear in a micro drill of diameter 0.2 mm for drilling 10-layered PCB
(Printed Circuit Board) with digital image processing technique.
They measured wear area, average and maximum wear height by
the help of an automated edge detection algorithm for cutting
plane segmentation with 0.996 mm/pixel resolution in 1 s. The
advantage of this technique is automatic detection of reference line
and wear profile of microdrill irrespective of the position of the
object. However, this technique is only useful when the cutting
plane image is clearer or when no smearing occurs. There was a
problem to differentiate between the smeared part of cutting plane
and the clearance part of the micro drill. To overcome this problem,
Duan et al. [30] has been applied a level set based technique for
accurate segmentation of cutting plane of micro drill bit. They have
fused the segmented image and thresholded image to get an
accurate result. They also observed that a significant change of
area, width and length has been occurred due to the wear of the
micro drill bit used for PCB manufacturing. However, they have
proposed a future scope to reduce the computational time. Xiong
et al. [135] had also used the variational level set based method by
eliminating the need for re-initialization of the zero level set
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 218
function for accurate segmentation of wear contour of cutting
inserts used for milling operation. They measured the tool wear
area by this method. However, the measurement of flank wear
width had been missing in their research.
Otieno et al. [96], studied flank wears of two fluted micro end
mills of diameter 1 mm, 0.625 mm and 0.25 mm with digital image
processing techniques using filtering and thresholding by XOR
operator. But any edge detection, tool wear quantification and
wear classification was not performed.
Inoue et al. [47] made a generalized approach by detecting
defects in rod-shaped cutting tool via edge detection (by Prewitt
operator) and extracted image parameters after performing
discrete Fourier transform (DFT) on the edge detected image.
However, many unstudied defects cannot be possible to recog-
nized by this system.
Jackson et al. [49] utilized a neural image processing method for
accurate detection of very small wear developed in very small
diameter milling cutter on the environmental scanning electron
microscopic (ESEM) images of tool. They have even measured the
small average wear of 5 mm developed in a 9.5 mm diameter
milling cutter. Though this technique is very much useful for
micro-machining, but the method is very much difficult to use
online.
Grain fracture, bond fracture and attritous wear are three types
of pre-dominant wears in grinding wheel. Wear flats are developed
on the grinding wheel surface due to attritous wear. Consequently,
the increasing rate of wear flats area develops heat and burn the
workpiece. But the automatic and precise segmentation of true
wear flats are quite challenging task from the wheel surface
images. An edge detection approach after thresholding were
utilized to distinguish true wear flats from its background [138].
However, the accurate selection of intensity threshold and edge
threshold was a difficult task. To overcome this problem, Lachance
et al. [75] utilized a region growing method for segmenting the true
wear flats from its background. However, some morphological
techniques can be utilized for more accurate computation of wear
flat area. Heger and Pandit [43] captured the images of grinding
wheel surface by multidirectional illumination and image fusion
for obtaining more detailed information. Then they have utilized
multi-scale wavelet transform and classification technique for
distinguishing the grains and cavities on the surface. A new
approach to discriminate the fresh and worn out grinding wheels,
progressively, has been established by Arunachalam and Rama-
moorthy [10]. They extracted some texture descriptors for
describing the condition of grinding wheel surface utilizing
histogram based, GLCM based and fractal based texture analysis
methods on the wheel surface images taken at different progres-
sive time. However, no explanation regarding the variations of
selected features with the progressive wear has been encountered.
In the area of integrated circuit (IC) manufacturing, the surface
of stamped tool or cutting dust has been monitored real time by
Kashiwagi et al. [62]. They captured the surface image of cutting
dust and determined the width of stamped line by using image
histogram and cross-correlation technique. They observed that the
width was decreasing with the increase of cutting time or decrease
of tool sharpness.
4.2. Three dimensional techniques
Three dimensional measurement techniques are used to
measure the crater depth accurately. Yang and Kwon [137,136]
first used a microscope equipped with a CCD sensors to capture
noisy images of rake face of an worn out tool insert and measured
the depth of crater in different levels of wear by automatic focusing
technique. They have used image consolidation and median
filtering to remove high frequency noises without blurring from
rake face image. Then they thresholded optimally for segmenting
the worn region from the background and detected the crater
contour by using Laplacian method. Edge linking and dilation
methods incorporating eight neighbourhood chain coding have
been applied on that contour to get an accurate shape of crater
region. A Laplacian criterion function incorporating an infinite
impulse response (IIR) filter has been used for getting the focused
position along z-direction. A hybrid search algorithm with
polynomial interpolation and golden search technique has been
utilized to improve the accuracy of the automated focusing
technique, in this method. This way they measured the crater
depth. They used seven features (four were related to flank wear
and three were related to crater wear) to classify flank wear, crater
wear, chipping and fracture. A mathematical model was intro-
duced in their work to obtain flank wear profile from crater wear
contour. Then they selected 12 input nodes (each node contains
seven feature parameters) and 4 output nodes (flank wear, crater
wear, chipping and fracture) in a multi-layer perceptron (MLP)
neural network to classify four types of wear. All the tests were
done on a P20 cemented carbide tool insert without chip breaker.
Though the work is pioneering the crater depth measurement very
accurately, but the 3D map of crater region has not been evaluated
by this offline technique. Also, it may be difficult to use their
technique for insert with chip breaker due to the major undulation
of rake surface. Ramamoorthy and co-workers [61,100] used image
processing with stereo vision technique with only a single CCD
camera to determine the depth of each point in the crater. Trends of
tool wear pattern were then analyzed with a MLPNN algorithm,
where inputs were speed, feed, depth of cut and cutting time and
output parameters were flank wear width and crater wear depth.
However, the crater depth estimation less than 125 mm could not
be obtained accurately by this technique. Also some pre-
processing algorithm were required to eliminated the noises from
dirt, chip, oil etc. on the rake face to make the method possible in
online.
Ng and Moon [93] proposed a technique for 3D measurement of
tool wear for micro milling tool (50 mm diameter) by capturing
images with varying the tool and camera plane distance with
15 mm resolution. Then they have re-constructed 3D image from
the captured images using digital focus measurement. Finally, they
proposed that the tool wear measurement could be possible by
combining the actual 3D image and the 3D CAD model of the tool.
However, no depth measurement had been performed in their
work.
Devillez et al. [24] utilized white light interferometry technique
to measure the depth of crater wear and determined the optimal
cutting conditions (cutting speed and feed rate) to get the best
surface finish in orthogonal dry turning of 42CrMo4 steel with a
uncoated carbide insert. In white light interferometry technique, a
vertical scanning has been performed to get the best focus
positions for each and every point presented in the object to be
measured. White light is used to get the high resolution (sub-
nanometer) and high precision measurements over a wider area.
However, this technique is an offline technique and the measure-
ment of crater depth of grooved inserts or inserts with chip breaker
is quite challenging for this technique. Dawson and Kurfess [22]
used a computational metrology technique to determine the flank
wear and crater wear rate of a coated and uncoated cubic boron
nitride (CBN) tool for progressive wear monitoring in offline. They
have acquired the data of the worn out cutting insert by using
white light interferometry and compute the volume reduction in
the insert by comparing those data with the CAD model of fresh
insert developed by using computational metrology. However, no
grooved insert has been used in their technique. Wang et al. [128–
131] measured various parameters viz. crater depth, crater width,
crater centre and crater front distance of crater wear by
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 219
Table 1
Direct TCM techniques based on image processing.
Researcher Illumination sys-
tem
Image processing Type of tool wear
measurement
Machining Remarks
Galante et al. [40] Diffused lighting Thresholding Flank wear Turning Offline, 2D technique
Weis [132] Diode flash light
with infra-red band
filter
Dilation and thresholding Flank wear
measurement
Milling No evaluation of accuracy
Kurada and
Bradley [73]
Fibre optic guided
light
Image enhancement, Image
segmentation, thresholding,
morphological operation
Flank wear
measurement
Turning Offline
Tauno and
Lembit [120]
Blue light source Median filter, Robert’s edge
detector, thresholding
Flank wear
measurement
Turning, milling Offline, 8% error
Pfeifer and
Weigers [99]
Ring of LED Method to set optimum
incidence angle of lighting for
controlled illumination
Flank wear
measurement
Turning, milling Online
Sortino [116] Median filtering
Statistical filter for edge
detection
Flank wear
measurement
Generalized
for insert
Offline
Flank wear
Jurkovic et al. [58] Halogen light along
with a laser diode
Manual measurement using a
image processing software
Flank wear and
deformation of
laser light pattern
on rake face
Tool inserts Manual measurement, crater
depth measurement has not
been done
Wang et al.
[128–131]
Laser trigger
synchronized with
camera, fibre optic
guided light
Find critical area, find reference
line, pixel to pixel scan for
measuring VB
max
from reference
line
Flank wear
(captured when
tool is moving)
Milling inserts Online, max error 15 mm,
difficult to measure coated
carbide inserts
Liang et al. [83] Backlighting Image registration, spatial
transformation, image
subtraction, similarity analysis
Nose wear Inserts Difficult to implement for
flank wear width
measurement
Sahabi and
Ratnam [108]
Backlighting Weiner filter, thresholding,
detection and subtraction of
worn and unworn profile in polar
co-ordinate
Flank wear from
nose radius and
surface roughness
profile
Inserts 7.7% (from nose) and 5.5%
error (from surface
roughness), difficult to
implement in very low feed
application
Fadare and Oni [35] 2 incandescent
light sources
inclined at 458
Weiner filter, shadow removing,
canny edge detection, pixel
counting
Flank wear Inserts Sensitive to the fluctuation of
ambient light
Kerr et al. [68] White ring light,
fibre optic guided
light
Unsharp mask, manual
measurement, histogram
analysis, GLCM analysis, Fourier
spectrum analysis, fractal
analysis
Flank wear
measurement via
texture descriptors
Turning inserts,
end mill cutter
Texture analysis of wear
region, no automatic
measurement of wear
Lanzetta [76] Structured lighting
with Laser
Resolution enhancement,
averaging, segmentation
Flank and crater
wear
Generalized
for insert
The effect of dirt, oils on
inserts did not address
Schmitt et al. [112] and
Stemmer et al. [117]
Ring light (for full
and side
illumination)
Sobel filter, line interpolation,
histogram transformation,
morphological opening &
closing, blob analysis, contour
detection for measurement; NN
for flank wear and breakage
classification
Flank wear
measurement,
wear and breakage
classification
Milling Resolution 4.4 mm,
classification error 4%; the
method has not been applied
for different variety of cutting
inserts
Castejo´ n et al. [18] and
Barriero et al. [13]
DC regulated light
with square
continuous
diffused
illuminator
Low pass filter, cropping,
histogram stretching, manual
segmentation, moment invariant
methods (Zernike, Legendre, Hu,
Taubin, Flusser), and linear
discriminant analysis for
classification
Classification of
low, medium and
high wear
Inserts 99.88% discrimination for
Hu’s descriptor, no wear
prediction has been
performed
Alegre et al. [6] DC regulated light
with square
continuous
diffused
illuminator
Contour signature based on
Canny edge detected image, k-
NN and MLPNN for classification
Classification of
low and high wear
Inserts 5.1% classification error;
three levels of wear
classification is needed
Atli et al. [11] Silhouette image of
tool
Canny edge detection,
measurement of deviation from
linearity of tool tip
Drill-bit Drilling Only useful for drilling; Flank
wear width cannot be
measured
Makki et al. [86] Silhouette image of
tool captured at
100–1500 r.p.m
Canny edge detection, best
fitting algorithm
Tool run out
detection
Drilling Tool run-out perpendicular to
the image plane had not been
measured
Liang and Chiou [82] Circular back
lighting
Spatial moment edge detection,
edge sorting, B-spline
smoothing, gaussian LPF,
thresholding, morphological
operation
Flank wear
detection for
progressive
machining
Multi-layer
twist drill
Results were not compared
with the microscopic wear
measurement; applicable for
no smear image
Su et al. [118] Circular lighting Accurate edge detection
proposed, rotation, automated
measurement
Flank wear
detection for
progressive
machining
Micro drill-bit
(for PCB drilling)
Resolution 0.996 mm/pixel;
only applicable when no
smearing in cutting plane
image
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 220
reconstructing a 3D crater profile by capturing four fringe patterns
with four phase shifting angles. No scanning is required in this
method unlike white interferometry technique. However, the
accuracy of the measurement is dependent on the fringe width or
fringe pattern.
Table 1 summarizes the application of digital image processing
in direct tool wear monitoring.
So, in direct technique, condition monitoring is done by
analyzing the change in geometry of the cutting tool. Chatter,
vibration, cutting force change etc. are not taken into account with
cutting tool observation whereas surface finish can emphasize
those changes as well as change in tool geometry. So, researchers
are going to take the measurement of surface finish through
indirect TCM techniques using image processing of machined
surface images.
5. Indirect TCM techniques using image processing
Diverse properties play an important role in the surface finish of
metallic parts, e.g. mechanical strength, wear resistance of the
surfaces or geometrical and dimensional quality of the parts. These
properties are directly related to the surface finish level, which is
dependent on the manufacturing process parameters and the
materials used. Thus, the measurement of the surface finish has
been a research matter of special interest during last sixty years in
machining sector. There are tactile and non-tactile techniques to
assess the surface quality of the machined parts. In tactile
techniques, surface roughness parameters are measured using a
stylus instrument; whereas in non-tactile method, surface
roughness parameters are obtained from the images of machined
surface textures. But there is a chance of scratches on soft materials
in tactile techniques due to the tracking of stylus on measurable
surface; whereas non-tactile techniques are becoming more
advantageous due to the advancement of computer vision
technology. While tactile techniques characterize a linear track
over the surface of the part, the computer vision techniques allow
characterizing whole areas of the surface of the part, providing
more information [8,111,113]. Besides, computer vision techni-
ques take measures faster, as images are captured in almost no
time and so they can be implemented in the machine. According to
this, it is possible to apply these techniques for controlling the
processes in real time on an autonomous manner. An exhaustive
validity check can also be made to every single part produced.
Continuous advances have been made in sensing technologies and,
particularly, in the vision sensors that have been specially
enhanced in capabilities with lower cost. The advances made in
the image processing technology also provide more reliable
solutions than before. In all, computer vision is a very useful
non-invasive technique for the industrial environment. The use of
these systems in other monitoring operations in machining
processes has proved [5,18] an important reduction in the cycle
time and the resources. In this field, two guidelines should
be remarked: the study in spatial domain and in frequency
domain [56,133]. Indirect tool condition monitoring using image
Table 1 (Continued )
Researcher Illumination sys-
tem
Image processing Type of tool wear
measurement
Machining Remarks
Duan et al. [30] Front lighting with
LED
Histogram generation, level set
based contour segmentation,
histogram based contour
segmentation, fusion of both
segmentation, wear
measurement
Flank wear
detection for
progressive
machining
Micro drill-bit
(for PCB drilling)
Capable to remove the noise
due to smearing; More
computation time
Xiong et al. [135] Fluorescent high
frequency linear
light
Variational level set based
segmentation, no need for re-
initialization of zero level set
Tool wear area Milling inserts No measurement of flank
wear width
Otieno et al. [96] Dome light with
low intensity back
lighting
Histogram equalization,
Gaussian filtering, XOR
operation for edge detection
Micro-Milling tool No measurement of wear
Yasui et al. [138] Microscope Thresholding, edge detection to
segment the wear flats from its
background
Grinding wheel
wear
Grinding Accuracy is low, possibility
for detection of false wear
flats
Lachance et al. [75] Fibre optic guided
light with beam
splitter
Thresholding, region growing Progressive wear of
grinding wheel
Grinding Morphological operations
will be lead to more accurate
segmentation
Prasad and
Ramamoorthy [100]
White light Histogram, GLCM and fractal-
based texture analysis
Progressive wear of
grinding wheel
Grinding Simple, faster but less
accurate
Karthik et al. [61] Automatic focusing
at various height
(interpolation and
search technique
for improving
accuracy)
Image consolidation, median
filtering, thresholding, laplacian
contour detection, edge linking,
dilation, chain coding, MLPNN
for classification
Flank and crater
wear (depth)
measurement and
classification
Turning inserts Leads to 3D measurement;
flank wear, crater wear,
chipping and breakage were
classified; 3D map for crater
wear has not been evaluated;
difficult for grooved inserts
Prasad and
Ramamoorthy [100]
Stereo vision
technique using
law of triangulation
Stereo image processing for
getting the 3D map of crater,
MLPNN
Flank wear and
crater wear
prediction and
progressive wear
measurement
Turning inserts Less accurate technique for
crater depth less than
125 mm; no technique to
reduce the noises from dirt,
dust, oil etc. difficult for
grooved inserts
Devillez et al. [24] White light
interferometer
White light interferometry by
automatic and varying focusing
Crater depth
measurement
Inserts Difficult to measure grooved
inserts
Dawson and
Kurfess [22]
White light
interferometer
Volume reduction measurement
of tool from fusion of CAD model
and surface profile
Crater depth
measurement
Inserts Difficult to measure grooved
inserts
Wang et al. [130] LCD projector for
fringe creation on
rake surface
3D reconstruction using phase
shifting method from 4 fringe
patterns with 4 phase shifting
angle
4 parameters of
crater wear
measurement
Inserts Difficult to measure grooved
inserts
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 221
processing can extract surface finish descriptors from images of
machined surface textures. There are mainly two techniques for
tool condition monitoring from the images of the machined
surface: online and offline. In online method, images of just
machined surfaces are captured using CCD or CMOS camera.
Online techniques are mainly useful for long and heavy parts. In
offline method, surface images are taken after finishing a number
of components. Generally, small and lightweight parts are
measured using offline techniques. Some researches on online
and offline methods are discussed in the following sections.
5.1. Online techniques
Gupta and Raman [42] measured the surface roughness of pre-
turned cylindrical bar utilizing the images of the laser scatter
pattern developed on the turned surface image, when the bar was
rotating with speeds ranging from 140 to 285 r.p.m. They extracted
first order statistical texture descriptors based on the grey level
histogram of images. They have also concluded from their study
that the ambient lighting and the speed of rotation were not
affecting the extracted surface roughness. However, there was no
correlation study of vision-based surface finish descriptors with
the stylus-based surface roughness and progressive flank wear
width. Ho et al. [44] did turning experiments in various feed,
cutting speed and depth of cut combinations and simultaneously
captured the machined surface images. Then they found out the
grey level average (G
a
) values of those images. After that adaptive
neuro-fuzzy inference system (ANFIS) was applied with inputs as
feed, cutting speed, depth of cut, G
a
and output was average surface
roughness (R
a
). At the time of machining the R
a
values were also
measured from the machined surface with the help of a stylus
instrument. The maximum prediction error of this process was
4.55%. However, only the grey level value, which is a first order
statistical texture descriptor, of the surface image has been
determined in this regard and no other higher order statistical
descriptors has been extracted in their research. [119] took the
images of turned surfaces at 57 different cutting speed, feed and
depth of cut combination by a camera and then they calculated a
parameter (G
a
, arithmetic average of grey level) from those images.
At the time of experiment, surface roughness (R
a
) were also
measured. Afterwards, cutting speed, feed, depth of cut and G
a
were used as inputs in a polynomial network with self organized
adaptive learning ability to predict the surface roughness. They
found a good correlation between predicted and measured surface
roughness with maximum 14% error. As an improvement of the
previous work, Lee et al. [77] utilized both spatial and frequency
domain properties of machined surface images without consider-
ing the machining conditions as the inputs of an abductive network
to predict the surface roughness. They have considered the
frequency co-ordinates, maximum eigen value of the covariance
matrix of normalized power spectrum and the standard deviation
of grey level as the inputs of abductive network. They have
considered both the spatial and frequency domain properties of
image texture for their analysis. Lee et al. [79] had further
improved their model to predict the surface roughness from the
image texture descriptors namely, spatial frequency, arithmetic
mean grey level and standard deviation of grey level by using
ANFIS. They achieved lesser deviation between predicted and
measured surface roughness (maximum 8%) compared to the
polynomial network technique. It can also be observed from their
results that error is less for high surface roughness values.
However, these methods have been applied on turning operation
with only one combination of cutting tool and workpiece material.
Also it has not been carried out for progressive wear monitoring.
Akbari et al. [3] predicted the surface roughness of milled surfaces
by using four texture descriptors, namely, arithmetic mean,
standard deviation, average surface roughness and root mean
square surface roughness based on grey level histogram of
machined surface images as inputs to a multi layer perceptron
neural network (MLPNN). Though an entire surface area has been
evaluated to get the more accurate estimation but no quantitative
error estimation with respect to the stylus based surface roughness
or tool wear has been reported by them. Narayanan et al. [91]
presented a genetic algorithm based EHW (evolvable hardware)
chip for noise removal from the milled surface images captured by
CCD camera. The surface image has been enhanced by 62.5% using
their system and then the G
a
value has been adopted as vision-
based surface roughness. This technique can be used to evaluate
the surface roughness of machined surface. However, an ANFIS
programme may be incorporated to their hardware for more
accurate prediction of surface roughness. Sarma et al. [110] turned
a glass fibre reinforced plastic (GFRP) composite hollow bar and
measured R
a
and G
a
values, simultaneously for each experiment
during machining. After that a correlation has been achieved from
those two values by linear regression analysis. Also a normalized
power spectrum was obtained from the experimental images and
the power spectral density was reducing with the improvement of
surface finish for increasing cutting speeds. However, no explana-
tion about the blurring effect due to the capturing of images during
machining was present in their work. Jian and Jin [55] introduced a
fast online surface texture analysis method to characterized the
machined surface images with straight feed marks. They first
binarize the machined surface images and then converted all the
pixels along a vertical line into 1 if the number of 1-valued pixels in
that vertical line is more than 50% of total number of pixels along
that line. Then they calculated the width between two consecutive
white lines and taken the average of all texture width in an image.
This texture width was characterized as a roughness descriptors
according to them. But their method is very much crude and less
accurate method. Palani and Natarajan [97] did an online
prediction of surface roughness values using cutting speed, feed,
depth of cut, major peak frequency, principal component magni-
tude square, G
a
as the input to a BPNN in end milling application.
The prediction error between the predicted and stylus based
surface roughness was 2.47%. However, their technique can also be
implemented for progressive wear monitoring.
Kassim et al. [67] turned AISI 1045 and AISI 4340 workpiece
materials by coated and uncoated carbide inserts until the inserts
reached to a catastrophic failure. Time to time, the images of the
surface textures of machined surfaces were captured by a CCD
camera with high magnification lens and the wear value of the
inserts were taken after each pass. Then the surface texture images
were processed using Sobel operation, thresholding operation and
column projection technique. The column projection technique
was used to normalize the image and to reduce the effect of non-
uniform illumination. As a result they got a uniform pattern for
machined surface machined by a sharp tool and irregular pattern
for machined surface machined with a dull tool. On the other hand,
gradient images (after Sobel operation) were analyzed by run
length statistics approach with six parameters. Then two sets of
machined surfaces (in set A, AISI 1045 workpieces were turned at
the feed rate of 0.4 mm/rev and at a cutting speed of 220 m/min
and in set B, the cutting speed was selected at 120 m/min to turn
AISI 4340 workpieces with a feed rate of 0.3 mm/rev) were
analyzed by the above mentioned methods and those sets were
clearly classified. However, no correlation study with surface
roughness has been performed by them. Mannan et al. [87] did
image and sound analyses and combined the features extracted
from both analyses to train a radial basis function neural network
(RBFNN) for predicting different states of the tool flank wear
corresponding to the applied features. Also flank wear was
measured using an optical microscope for validity check. They
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 222
tried to monitor the condition of a sharp, a semi-dull and a dull tool
by this technique. However, they did not analyze the error of
prediction. Kassim et al. [64] introduced a procedure to define
edges of surface texture obtained from turning, end milling and
face milling operation by connectivity oriented fast Hough
transform parameters like spread of orientation, average line
length, main texture orientation and total fitting error. This
connectivity oriented fast Hough transform process was faster and
less computationally complex than standard Hough transform
technique which was used to analyze the uniformity of surface
textures obtained from sharp and dull tools. Then the tool wear
was then predicted by using a MLPNN where inputs were taken
from the parameters of processed images. However, they did not
get any correlation for image number 3–5. Kassim et al. [63] also
showed that run length statistics technique for the detection of
surface textures machined by sharp tool and dull tool was faster
and better than column projection technique and connectivity
oriented fast Hough transform technique. Column projection
analysis technique was working well for highly regular surfaces
whereas Hough transform technique was extracting line segments
for variety of length. With the features extracted from run length
matrix, they classified the sharp tool and dull tool by applying
Mahalanobis distance classifier. Also they compensated inhomo-
geneous illumination of the texture images through an excellent
way. However, they did not get any systematic trend of variation
between image texture parameters and machining time. The image
descriptors were not normalized and no correlation study of image
texture descriptors with progressive tool wear or surface
roughness has been indicated in their work.
In a very recent study, Datta et al. [21] captured the turned
surface images for progressive wear of a uncoated carbide tool and
analyzed those images using a grey level co-occurrence matrix
(GLCM) technique based texture analysis. They also find a linear
correlation between the extracted features, namely, contrast and
homogeneity with the tool wear in terms of slope of the linear fit
and a fitting parameter, coefficient of determination. It has also
been observed from their study that the selection of GLCM
parameters viz. pixel pair spacing and direction is very much
important to get the accurate results as the distribution of feed
marks are varying with the variation of machining conditions (feed
rate and depth of cut). However, they did not mention about any
method to find the optimum pixel pair distance. As an improve-
ment of the previous technique, Dutta et al. [31] has been proposed
a novel technique to find the optimum pixel pair spacing
parameter to get an accurate result by texture analysis of machined
surfaces with the progressive tool wear. They got a periodic
relation of extracted texture descriptors viz. contrast and
homogeneity with the different pixel pair spacing. Utilizing this
periodic property, they found out the periodicity using Fourier
power spectral density technique and later on they found the
optimum pixel pair spacing parameter of GLCM. However,
the optimum pixel pair spacing is also varying dependent on
the change of feed rate. They got a very good correlation of
extracted descriptors with tool wear and surface roughness.
However, they did not do any experiment to detect the progressive
tool wear of coated carbide tools.
Fractal analysis of surface texture for tool wear monitoring was
proposed by Kassim et al. [66] to deal with high directionality and
self-affinity of end-milled surfaces and a hidden Markov model
(HMM) was used to differentiate the states of tool wear.
Anisotropic nature of end-milled and turned surface textures
was analyzed by fractal analysis along different directions to the
entire image by Kassim et al. [65]. They used a 13-element feature
vector to train the HMM model for classifying four distinct states of
tool condition. However, no estimation of classification error has
been encountered in their study. Kang et al. [60] used a fractal
analysis technique to study the variation of fractal dimension with
measured surface roughness, wear values with machining time for
different feed combination for high-speed end milling of high-
hardened material by a coated carbide tool. However, no
quantitative analysis of correlation of fractal dimension with
flank wear or surface roughness was done.
Persson [98] established a non-contact method to measure the
surface roughness by incorporating angular speckle correlation
technique. A speckle pattern created on the machined surface with
the help of a coherent He–Ne laser and captured at different angle
of illumination. Then a correlation between those captured speckle
pattern at different angle of illumination has been calculated. The
lower correlation value has been observed for rougher surfaces.
Though this technique can be used for the in-process measurement
of surface roughness but the accuracy of this method is limited by
the proper angular positioning of the set-up. However, this
limitation can be overcome by using a laser interferometric
technique for tilt measurement of the set up.
With a different approach, Li et al. [79] has been introduced an
wavelet packet analysis of machined surface images obtained from
turning operation. They got a good correlation between the
extracted feature, namely, high frequency energy distribution ratio
with progressive cutting tool wear. However, a systematic
quantitative correlation analyses was missing in their study.
5.2. Offline techniques
Luk and Huynh [85] analyzed the grey level histogram of the
machined surface image to characterize surface roughness. They
found the ratio of the spread and the mean value of the distribution
to be a nonlinear, increasing function of R
a
. Since their method was
based solely on the grey level histogram, it was sensitive to the
uniformity and degree of illumination present. In addition, no
information regarding the spatial distribution of periodic features
could be obtained from the grey level histogram. Hoy and Yu [45]
adopted the algorithm of Luk and Huynh [85] to characterize the
surface quality of turned and milled specimens. They found one
exception where the ratio of the spread and the mean of the grey-
level distribution was not a strictly increasing function of surface
roughness and, therefore, the value of the ratio might lead to
incorrect measurement. They also addressed the possibility of
using the Fourier transform (FT) to characterize surface roughness
in the frequency domain. However, only simple visual judgement
of surface images in the frequency plane was discussed. No
quantitative description of FT features for the measurement of
surface roughness was proposed. Al-kindi et al. [7] examined the
use of a digital image system in the assessment of surface quality.
The measure of surface roughness was based on spacing between
grey level peaks and the number of grey level peaks per unit length
of a scanned line in the grey level image. This 1D based technique
did not fully utilize the 2D information of the surface image, and is
sensitive to choice of lay direction, lighting and noise. Cuthbert and
Huynh [20] increased the sophistication of the analysis by applying
a statistical texture analysis on the optical Fourier transform
pattern created on the ground surface images. Then they calculated
the mean, standard deviation, skewness, kurtosis, and root mean
square height of the grey level histogram of the image. There were
two limitations of this technique. Only surfaces upto an average
surface roughness of 0.4 mm could be inspected, as rougher
surfaces tend to create a diffused pattern in the camera. Precise and
complex alignment of the imaging optics was required, thereby
making it difficult to the use in online inspection. Jetley and Selven
[54] used the projection of a reflection pattern of a beam of low
power (1 mW) He–Ne laser light from ground surface. Then the
pattern was analyzed and characterized using blob area, thresh-
olding and hence correlated to the surface roughness. But the
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 223
result in this technique is sensitive to grazing angle of lighting
system. That might be eliminated by using any filtering technique
to remove the inhomogeneous illumination. As noted by Elbestawi
et al. [33,107], the conventional roughness measure (i.e. R
a
), when
plotted as a function of the distance, the tool had travelled on the
part (or cutting time), undergone a complex evolution. They
showed that R
a
increased steadily with cutting time but then
dropped markedly as the tool showed marked wear. The drop in R
a
at long distances was due to heating and ductile surface
deformation by the worn tool. Therefore, R
a
alone was not a
reliable tool wear indicator, even though a stylus profiler
measurement indicated an acceptable R
a
value. These two
phenomena could generate a surface of incorrect form and texture.
Ramamoorthy and Radhakrishnan [103] and Kiran et al. [71] had
utilized the grey level intensity histogram for establishing some
features for roughness evaluation of ground, milled and shaped
surface images but they did not correlate those parameters with
tool wear or profiler-based surface roughness. Wong et al. [134]
used a 0.8 mm-diameter 5 mW He–Ne laser beam and focused it
on the turned surface to get a laser scatter pattern. Then the mean
and standard deviation of the captured laser scatter pattern image
were calculated. But they did not show any definite correlation
between the measured surface roughness of the machined surface
and the intensity distribution of the scattered light pattern.
Although, they found quite good correlation between tool wear
and intensity distribution of the scattered light pattern for most of
the work pieces, they did not show any systematic variations of
image features with machining time. Younis [140] analyzed the
scatter pattern created by white light on the ground surface and
derived a vision-based surface roughness parameter. They
computed the vision-based surface roughness parameter based
on the squared difference of a pixel value with its 8-neighbour-
hood. They studied the correlation of stylus-based surface
roughness and vision-based surface roughness for tool steel,
copper and brass material and found the linear correlation co-
efficient ranging from 0.79 to 0.92. They have concluded that the
correlation were varying for different material depending on
different modes of tearing and fracture for grinding of different
materials. However, more experiments were needed with different
cutting conditions for the establishment of their technique.
Whitehead et al. [133] compared contact and laser stylus methods
for roughness measurement. Kumar et al. [72] magnified original
images of shaped, milled and ground surfaces using cubic
convolution interpolation technique and enhanced the edges
using a linear edge crispening algorithm. Based on the surface
image features, a parameter called G
a
was estimated using
regression analysis, for the original images and for the magnified
and enhanced images. Finally, a comparison has been carried out to
establish a correlation between magnification index, G
a
and
surface roughness. However, more number of image features
might be evaluated to get more accurate results. Dhanasekar and
Ramamoorthy [27] applied a geometric search technique for edge
detection following a pre-processing the machined surface images
of shaped, milled and ground parts. A good correlation has been
encountered between the surface roughness (G
a
) obtained by the
vision-based system and by the stylus-based surface profiler.
However, the correlation of the vision-based surface roughness
with progressive tool wear has not been addressed in their work.
Khalifa et al. [69] used magnification, edge enhancement,
statistical and texture analysis of turned surface images to detect
chatter. Those images were enhanced using composite Laplacian
filter. After that G
a
, histogram mean, standard deviation, variance
were evaluated from filtered images. Subsequently, the GLCM
analysis of enhanced images were also performed. Energy, entropy
and inertia features were extracted from GLCM to discriminate
chatter-rich and chatter-free images. Al-kindi and Shirinzadeh [8]
proposed a method named intensity–topography compatibility
(ITC) for characterizing the image data by three components,
namely, lightning, reflectance and surface characteristics. They
extracted the value of the all surface roughness parameters viz.
average roughness, root mean square roughness, maximum peak
to valley, maximum valley depth, maximum peak height,
skewness, kurtosis etc. from grey level histogram. However, no
wear or surface roughness correlation study has been performed in
their technique. Elango and Karunamoorthy [32] studied the
variation of grazing angle of a diffused light on face turning
specimen turned at different striation angle with design of
experiment using Taguchi’s L9 orthogonal array and analysis of
variance technique (ANOVA). They considered the G
a
value as a
texture descriptor and find a 758 optimum grazing angle and 908
striation inclination to achieve accurate G
a
. However, this method
can suitably be applied for progressive wear monitoring purpose.
[28] enhanced the resolution of the ground and milled surface
image by using projection on convex sets (POCSs) algorithm. Then
they extracted three texture descriptors using frequency domain
based and histogram based texture analyses from the recon-
structed image. Finally, they predicted the surface roughness using
group method of data handling (GMDH) and compared those
predicted values with stylus based surface roughness. However, no
prediction error had been reported in their work. Zhongxiang et al.
[143] captured images of plain ground, plained, plain milled and
end milled specimens using a digital camera and then those images
were pre-processed by median filtering, greyscale equalization and
histogram conversion amplification methods. The image data were
analyzed by normalized cross-correlation and surface fitting
techniques by using CAS software. They extracted four features,
namely, mean, standard deviation, root mean square value (3D)
and kurtosis (3D) from those pre-processed images and found a
correlation between the surface roughness (obtained from stylus
measurement) and those extracted features. Texture analysis of
higher order statistics may produce better and robust results in
their technique. Dhanasekar and Ramamoorthy [26] pioneered to
capture the moving machined surface images of milled and ground
specimens and then deblurred those images using Richardson–
Lucy restoration algorithm. Those deblurred images were pre-
processed to compensate the inhomogeneous illumination. After-
wards, the spatial frequency, arithmetic mean value and standard
deviations were extracted as texture features. An artificial neural
network (ANN) was used with these three features as inputs to
predict the vision based surface roughness. Then they plotted
predicted result with experimental result and got coefficient of
determination (R
2
) values of 0.923 and 0.841 for milling and
grinding, respectively, for deblurred and restored images. Also they
showed that the R
2
values of restored images were much better
than R
2
values for non-restored images. An online tool condition
monitoring using laser vibrometer and CCD camera has been
performed by Prasad et al. [101]. They extracted the acousto-optic
signal of machining and the 3D surface roughness of machined
surface images using a surface metrology software, namely,
TRUEMAP for progressive wearing of cutting tool in face turning
operation. However, the computation time required for surface
roughness measurement has not been mentioned in their work.
Gadelmawla [36] did an automatic surface characterization by
calculating the grey level co-occurrence matrix of 10 types of
machined surface images with varying GLCM parameters, distance
and orientation. They have also extracted four features from the
GLCMs and observed that most of the features, except standard
deviation of the GLCM, were behaving differently with respect to
sensitivity for rough (turned) and smooth (lapped) surfaces.
Gadelmawla et al. [37,38] developed a reverse engineering
software for detecting and predicting the machining conditions,
cutting speed, feed rate and depth of cut, from the machined
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 224
surface images using the GLCM texture descriptors. However, they
have not optimized any of the GLCM parameters. Also they have
only tested this method for milled surface images only. Gadel-
mawla [39] predicted average surface roughness (R
a
) values from
the texture descriptors extracted from the GLCM of turned surface
images with only a single combination of GLCM parameters for
different machining conditions. The error between the measured
R
a
value by stylus method and the predicted R
a
value is Æ7%.
However, the distance parameter of GLCM could be optimized for
getting more accurate and precise result.
Myshkin et al. [89] introduced a special type of co-occurrence
technique with the concept of multi-level roughness analysis
to determine the surface roughness for nanometer scale
deviations obtained from the atomic force microscope (AFM)
images. However, no quantitative analysis has been done in their
study.
Tsai et al. [123] investigated Fourier power spectrum of shaped
and milled surface images with various maximum surface
roughness. The maximum surface roughness values were mea-
sured using a stylus-based surface profiler. They found image of
the surface patterns of the shaped specimens were more regular
and present less noise than those of the milled specimens. They
further found a monotonically decreasing trends for feature major
peak frequency, principal component magnitude squared, central
power spectrum percentage and monotonically increasing trends
for average power spectrum with increasing values of measured
surface roughness for both the shaped and milled parts. Further-
more they used two artificial neural network (ANN) techniques for
classification of roughness features in fixed and arbitrary orienta-
tions of surfaces. Then they selected major peak frequency as the
best feature for both shaped and milled specimen in fixed
orientation, because, it was the distance between the major peak
and the origin, so it was a robust measure to overcome the effect of
lighting of the environment. However, they only did the surface
finish measurement for flat surfaces not for curved parts. Tsai and
Wu [124] used a Gabor filter-based technique for an automated
classification of defective and non-defective surfaces from the
surface images. They convolved the image with a 2D Gabor
function, which is an oriented complex sinusoidal grating
modulated by a 2D Gaussian function. Then they have selected
the best parameter of the Gabor function, such that the energy of
the convolved image was zero, using exhaustive search method.
Then a threshold value has been chosen using statistical control
method for distinguishing the homogeneous and non-homoge-
neous surface texture. However, a very accurate controlled set-up
for capturing the surface images are required for practical
accomplishment of their method. Dhanasekar et al. [25] captured
speckle patterns of machined surfaces (ground and milled) using a
collimated laser beam (He–Ne laser, 10 mW, l = 532 nm) and a
CCD camera. Then, pre-processing of speckle images was carried
out to remove unwanted intensity variations due to ambient
lighting change, etc. The speckle images were filtered by Butter-
worth filter and then the centralized fast Fourier transform (FFT)
was determined. After that average and integrated peak spectral
intensity coefficient and autocorrelation coefficient in X, Y and
diagonal directions were determined. The width of autocorrelation
functions for smooth and rough images were varied. The spectral
speckle correlation (auto-correlation) technique for surface
roughness assessment had been used before and after pre-
processing of speckle images. They were then compared to stylus
values (R
a
). It was found that autocorrelation parameters after pre-
processing had a better correlation (i.e. higher correlation
coefficient) with the average surface roughness (R
a
) measured
for the milled and ground components. To get more accurate result,
image model for compensating inhomogeneous illumination [14]
could be used in their work.
Josso et al. [57] analyzed and classified eight surface images
obtained from eight types of engineering processes viz. casting,
grinding, gritblasting, hand filing, horizontal milling, linishing,
shotblasting, vertical milling. They have developed a space-
frequency representation of surface texture using frequency
normalized wavelet transform (FNWT) and extracted some surface
finish descriptors. Then they classified those eight types of surfaces
using discriminant and cluster analysis approach. However, there
is a high chance of misclassification between similar types of
texture viz. milling and grinding. So, they compared continuous
wavelet transform (CWT), standard and scaled discrete wavelet
transform (DWT) methods and concluded that the standard
discrete wavelet transform associated with cluster analysis was
the best method for classification purpose. In their another work
[55], they tried to measure the form, waviness and roughness of
machined surfaces images by using FNWT. Niola et al. [94] tried to
reduce the problem of brightness variation on surface images at
different lighting condition by enhancing images of machined,
ground and polished surfaces using Haar wavelet transform.
However, no surface finish descriptors were extracted from the
surface images, in their study.
Ramana and Ramamoorthy [104] classified ground, milled and
shaped images based on GLCM, amplitude varying rate approach
and run length statistical technique. However, they did not decide
about the best feature for vision-based surface roughness
measurement. Also they did not do any quantitative correlation
study between vision based and stylus based surface roughness.
Bradley and Wong [16] presented the performance of three image-
processing algorithms, namely, analysis of the intensity histogram,
image frequency domain analysis and spatial domain surface
texture analysis for evaluating the tool condition from face milled
surface images. Though, the histogram based technique revealed a
proper trend for the progressive wear of face milling tool but it was
very much influenced by the lighting condition. Frequency domain
technique was much less sensitive to inhomogeneous illumination
than the histogram based approach. The major advantage of a
texture-based method was the dependence on localized similari-
ties in the image structure. The absolute value of illumination
intensity was not critical; the illumination must be sufficient to
highlight image features. Similarly, the method was not sensitive
to the angle of illumination, except for extreme cases where the
axis of illumination approached 908. They showed a systematic
variation of texture parameters with machining time. However, no
quantitative correlation has been reported by them. Zhang et al.
[142] developed an accurate defect detection and classification
system by extracting the best features from discrete cosine
transform (DCT), Laws filter bank, Gabor filter bank, GLCM. They
used support vector machine (SVM) and RBFNN for classification
purpose. They have got a 82% success using the combination of
Gabor filter, GLCM and SVM. Singh and Mishra [115] classified
different types of spangles obtained due to the galvanization of
steel sheets using GLCM and Laws texture descriptors with RBFNN.
They achieved 80% accuracy of classification. Their approach can
also be used for progressive wear monitoring. Alegre et al. [4] used
first order statistical texture analysis, GLCM method and Laws
method to evaluate turned surface images and classified two
roughness classes using k-NN technique. Best result was obtained
by using Laws method, in their study. In a different approach,
Bamberger et al. [12] compared three methods for examining the
chatter marks produced at the time of machining in valve seat of
automotive parts from the images of the valve seats. They
compared three image processing based techniques, namely, circle
fitting, circularity and GLCM method to classify accepted and
rejected parts. Though, they selected the appropriate distance
parameter of GLCM, manually, but it is needed to develop an
automatic method for detection of optimized distance parameter.
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 225
Table 2
Indirect TCM techniques based on image processing.
Researcher Illumination system Image processing algorithm Applied in Remarks
Wong et al. [134] He–Ne laser Mean and standard deviation of
laser pattern created on machined
surface
Turning Offline; no study on correlation
and progressive wear
Gupta and Raman [42] HeNe laser, circular
variable attenuator
Histogram based 1st order
statistical texture analysis
Turning (moving and
static condition); surface
roughness measurement
Online; no correlation study
between vision-based surface
roughness and stylus-based
surface roughness and
progressive tool wear; no
discussion about blurring due to
movement
Tarng and Lee [119] 2 Light sources
situated at an acute
angle with the axis of
workpiece
Determination of G
a
, polynomial
network with self organized
adaptive learning (feed, speed,
depth of cut and G
a
as input, R
a
as
output)
Turning; R
a
prediction Online; prediction error
(max) = 14%; extraction of 1
descriptor only; no prediction of
tool wear
Ho et al. [44] 2 Light sources
situated at an acute
angle with the axis of
workpiece
Determination of G
a
, ANFIS (feed,
speed, depth of cut and G
a
as input,
R
a
as output)
Turning, R
a
prediction Online; prediction of R
a
using
ANFIS prediction error
(max) = 4.55%; extraction of 1
descriptor only; no prediction of
tool wear
Lee et al. [78] A diffused blue light in
458 inclination
Standard deviation of grey level, two
frequency domain parameters and
abductive network (input as 3
texture descriptors, output as R
a
)
Turning, R
a
prediction Online; max prediction
error = 14.96%; no prediction of
tool wear
Lee et al. [79] A diffused blue light in
458 inclination
Standard deviation of grey level, two
frequency domain parameters and
ANFIS (input as 3 texture
descriptors, output as R
a
)
Turning, R
a
prediction Online; max prediction
error = 8%; no prediction of tool
wear
Akbari et al. [3] Scattered pattern of
light
Histogram based 1st order
statistical texture analysis (four
descriptors) & MLPNN
Milling, R
a
prediction Online; No quantification of
prediction error; No prediction
of tool wear
Narayanan et al. [91] An evolvable hardware Image enhancement, determination
of G
a
, genetic algorithm
Milling; Surface
roughness measure
Online; no quantification of
prediction error; no prediction of
tool wear
Sarma et al. [110] Determination of G
a
, frequency
domain analysis
Turning GFRP composite
with PCD tool
No study for progressive wear
monitoring
Palani and Natarajan [97] Frequency and spatial domain based
texture analysis, BPNN
End milling, R
a
prediction No study for progressive wear
monitoring
Kassim et al. [67] Sobel operation, thresholding,
column projection (CP) (applied on
thresholded images), run-length
statistics (RLS) (applied on grey
level images)
Turning; progressive wear
monitoring
Online; Progressive wear
monitoring; classification
between sharp tool and dull tool
in various machining; no
correlation study with R
a
Mannan et al. [87] Sobel operation, thresholding, CP,
RLS, extraction of AE parameters
using wavelet analysis, RBFNN for
flank wear prediction
Turning; progressive wear
monitoring
Online; monitor sharp, semi-dull
and dull tool; no quantification
of prediction error
Kassim et al. [64] Canny edge detection, connectivity
oriented fast Hough transform,
MLPNN for FW prediction
Turning, end milling, face
milling; progressive wear
monitoring
Online; no quantification and
comparison of prediction error
Kassim et al. [63] Compensating inhomogeneous
illumination compensation,
comparison of CP, connectivity
oriented fast Hough transform and
RLS, Mahalanobis distance classifier
for classification of sharp and dull
tool
Turning; progressive wear
monitoring and
classification
Online; RLS was selected as the
best technique depending only
on a single cutting condition;
classification between two wear
state only; more
experimentation needed
Datta et al. [21] Diffused light GLCM technique Turning; progressive wear
monitoring
Online; extraction of best feature
depending only on a single
cutting condition; No
optimization of GLCM
parameters
Datta et al. [31] Diffused light GLCM technique with optimized
pixel pair spacing (pps) parameter
Turning; Progressive wear
monitoring
Online; Optimization of pps
developed; applicable for any
periodic textures; no study to
monitor coated carbide tool
Kassim et al. [66] Fractal with HMM End milling; Classification Online; No estimation of
classification error
Kassim et al. [65] 3D fractal with HMM End milling; classification
of 4 states of wear
Online; no estimation of
classification error
Kang et al. [60] Fractal; progressive variation study
with surface roughness and tool
wear
High speed end milling
(with coated carbide)
Online; no estimation of
correlation parameter
Li et al. [81] Diffused light Wavelet packet decomposition Turning; progressive wear
monitoring
Online; no correlation analysis
with tool wear
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 226
Table 2 (Continued )
Researcher Illumination system Image processing algorithm Applied in Remarks
Hoy and Yu [45] Diffused white light Histogram analysis, 2D FFT analysis Turning, milling Offline; no progressive wear
monitoring
Cuthbert and Hynh [20] He–Ne laser, spatial
filter, beam splitter
and mirror
Histogram based 1st order
statistical texture analysis
Grinding Offline; complex attenuator;
difficult to implement for high
roughness values; no
progressive wear monitoring
Jetley and Selven [54] He–Ne laser Blob analysis, thresholding Grinding Offline; no progressive wear
monitoring
Ramamoorthy and
Radhakrisnan [103]
GLCM analysis Grinding, shaping, milling Offline; no correlation parameter
study
Kiran et al. [71] Diffused light; light
sectioning; phase
shifting with grating
projection
Frame averaging; low pass filtering;
2nd order co-occurrence statistics;
three lighting methods were
compared for rough, medium rough
and smooth images
Grinding, milling, shaping Offline; mainly the comparison
of three types of lighting; no
roughness evaluation
Younis [140] White light Neighbourhood processing Grinding (different
material)
Offline; coefficient of variation
8.6%
Coefficient of determination (R
2
)
0.79–0.92; no progressive wear
study
Kumar et al. [72] Cubic convolution interpolation,
linear edge crispening,
Determination of G
a
Shaping, milling, grinding Offline; no progressive wear
monitoring
Khalifa et al. [69] Edge enhancement, magnification,
statistical texture analysis (1st and
2nd order), calculation of G
a
value
Chatter detection in
turning
Discrimination between chatter-
rich and chatter-free process
from surface images
Al-kindi and
Shirinzadeh [8]
Ambient light Comparison between two lighting
models viz. intensity topography
compatibility and light diffused
model, extraction of optical surface
roughness parameters from 1st
order statistics
Face milling No correlation study with
progressive wear
Elango and
Karunamoorthy [32]
Diffused light at
different grazing angle
Determination of G
a
, Taguchi’s
orthogonal array and ANOVA
Face turning No correlation study with
progressive wear
Dhanasekar and
Ramamoorthy [28]
White light POCS for reconstruction of high
resolution image, frequency domain
and histogram based texture
analysis, GMDH
Milling, grinding (R
a
prediction)
No prediction error analysis, no
correlation study with
progressive wear
Zhongxiang et al. [143] Stereo zoom
microscope, halogen
lamp
Median filtering, histogram
conversion, histogram
homogenization, calculation of 3D
roughness
Paning, plain milling, end
milling, grinding
No correlation study with
progressive wear
Dhanasekar and
Ramamoorthy [26]
Richardson–Lucy algorithm for
deblurring, frequency and spatial
domain based texture analysis, ANN
Milling, grinding, R
a
prediction
Correlation coefficient 0.923 and
0.841 for milling and grinding,
No correlation study with
progressive wear
Gadelmawla[36] Microscope GLCM, study the effect of pps Face turning No optimization of pps value, No
correlation study with
progressive wear
Gadelmawla
et al. [37,38]
Microscope GLCM Milling, Reverse
engineering for cutting
conditions
No optimization of pps value, No
correlation study with
progressive wear
Gadelmawla [39] Microscope GLCM Face turning, Correlation
with R
a
No optimization of pps value, No
correlation study with
progressive wear
Tsai et al. [123] Fluorescent light
source
Fourier analysis, ANN Shaping, Milling No correlation study with
progressive wear
Tsai and Wu [124] Gabor filtering, classification of
defective and non-defective parts,
Milling No mention of success rate; no
progressive wear or surface
roughness study
Dhanasekar et al. [25] He–Ne laser Speckle pattern, Butterworth
filtering, Fourier analysis,
Autocorrelation
Grinding, milling No correlation study with
progressive wear
Josso et al. [57] Frequency normalized wavelet
transform, discriminant and cluster
analysis
Classification of ground,
milled, cast surfaces etc.
No correlation study with
progressive wear
Josso et al. [56] Frequency normalized wavelet
transform,
Form, waviness,
roughness measurement
No correlation study with
progressive wear
Niola et al. [94] Haar wavelet for reduction of
inhomogeneous illumination
Milling, grinding,
polishing
No extraction of surface finish
parameters
Raman
andRamamoorthy [104]
GLCM, amplitude varying rate
method, RLS
Classification of ground,
milled, shaped surfaces
No correlation study with stylus
based surface roughness; no
progressive wear study
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 227
Table 2 (Continued )
Researcher Illumination system Image processing algorithm Applied in Remarks
Bradley and
Wong [16]
Fibre optic guided light
(regulated)
Frame averaging, Gaussian filtering,
median filtering, after filtering:
image histogram analysis,
frequency domain analysis, texture
segmentation
Face milling, progressive
wear study
Comparison between histogram
analysis, frequency domain
analysis and texture
segmentation; no correlation
analysis of vision-based surface
finish with tool wear
Zhang et al. [142] DCT, Laws filter, Gabor filter, GLCM,
Shape features, SVM with RBFNN
kernel
Defect detection and
classification in ground
and polished surfaces
82% success rate using the
combination of Gabor filter and
GLCM with SVM
Alegre et al. [4] DC regulated light with
SCDI
First order statistical texture
analysis, GLCM, Laws method, k-NN
classification
Turning No progressive wear study
Nakao [90] Fibre optic light Thesholding, component labelling Drilling burr
measurement
3% and 2% error in measuring
burr thickness and height
Yoon and
Chung [139]
Halogen (front light)
LED (back light)
Edge detection (burr width
measurement), Shape from focus
(burr height measurement)
Micro-drilling 0.1 mm resolution; less than
0.5 mm accuracy
Sharan and
Onwubolu [114]
High intensity spot
lighting
Burr profile measurement Milling 2.2 mm resolution
Fig. 3. Flow diagram of proposed tool condition monitoring technique using digital image processing.
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 228
Ikonen and Toivanen [46] proposed an algorithm that gave
priority to a pixel in the tail so as to calculate the minimum
distance in a curved space so that it helped in calculating the
roughness in a faster and more efficient manner.
Vesselenyi et al. [126] utilized 2D box counting method and
found nine parameters as roughness descriptor by linear, second
order and third order polynomial fitting on shaped, ground and
polished surface images of different surface roughness. Then
they classified them using C-means clustering. However, more
number of samples were required to proof the suitability of their
method.
Quality of honed surfaces was also determined by Leo´ n et al.
[80] using image processing technique. He quantified the groove
textures and defects of honed cylinder bore in frequency domain.
In frequency domain, the groove texture of interest was
separated from the other defects such as groove interrupts,
holes, cracks, flakes, material defects, graphite lamellae, material
smearings, smudgy groove edges and foreign bodies. The images
were taken from fax film replicas of honed surfaces. The images
were enhanced by contrast stretching. Digital image processing
was also used in chatter identification and burr detection in
machining.
Nakao [90] captured images of drilling burrs and then
processed to monitor drilling process. Here the conventional
image processing techniques such as the binary image proces-
sing, the noise reduction and the labelling were applied to
measure image data. Here burr height and thickness were
measured fromthe processed image using co-ordinate data. Yoon
et al. [139] used edge detection algorithm to measure hole quality
and burr width in micro-drilled holes. They also measured burr
height with ‘Shape From Focus’ (SFF) method. Here a halogen
light was used as a front light and LED was used as a backlight for
getting uniform illumination. Sharan and Onwubolu [114]
measured the burr profile of milled parts with 2.2 mm system
resolution.
In most of the research, the variation of vision based surface
texture descriptors with machining time were not studied for
progressive wear monitoring. Also there is a requirement to
normalize the texture or wear descriptors for reducing the effects
of lighting variations. Research in this area is requiring a detailed
study with various work tool material combination with various
cutting parameters for different machining application to establish
a robust monitoring system.
The indirect tool condition monitoring techniques, using image
processing are summarized in Table 2.
6. Conclusions
In this paper, the application of image processing technology
applied for tool condition monitoring is discussed. For real time
tool condition monitoring with noncontact techniques, the image
processing algorithms can be used for enhancing the automation
capability in unmanned machining centres.
The digital image processing techniques are very useful for fast
and easier automatic detection of various types of tool wear (such
as crater wear, tool chipping and tool fracture) which are very
difficult to recognize by other modes. Textural analysis techniques
are playing a predominant role for tool condition monitoring via
assessment of machined surface quality. Future research should be
aimed to develop a robust system (including lighting, camera and
faster algorithm) for real time tool condition monitoring tech-
nique. However, a noncontact and less costly tool condition
monitoring technique can be established with the help of digital
image processing techniques through robust machine vision
system. Some established observations from the review are
discussed below:
1. Diffused lighting system (such as LED light, fibre optic guided
light, white light) and a high speed CCD camera should be
utilized to enhance the image capturing capability in real time
monitoring.
2. In direct TCM technique, image pre-processing, image thresh-
olding, edge detection and morphological operation along with
texture analysis technique can be used for getting the faster
outputs.
3. Gaussian filtering along with illumination compensation
technique has a good impact in image pre-processing
operation.
4. Canny edge detection method can be a fast and strong edge
detection technique for direct TCM.
5. Textural analysis (in both spatial and frequency domain) can be
a strong and fast technique for having a good correlation with
the surface roughness and tool flank wear data in case of
indirect monitoring technique.
6. Pattern classification should be used to classify between sharp,
semi-dull and dull tools in indirect TCM.
7. ANFIS (Adaptive Neuro Fuzzy Inference System) is a very
robust tool for accurate prediction of tool wear in indirect TCM.
8. Crater wear can be measured using stereo imaging algorithm
with a single camera.
9. Faster detection of the effect of vibrations, machine noise,
cutting tool condition, etc. can be possible from indirect TCM in
comparison with direct TCM.
10. Systematic variation of image parameters with machining time
should be studied and established to get the full benefit of
machine vision based tool condition monitoring approach.
Based on the above conclusions, a tool condition monitoring
technique using digital image processing can be used in future
research as shown in Fig. 3. There both direct and indirect
methods can be combined, where direct technique will be used to
validate the results of indirect techniques in a single experimental
set-up.
References
[1] Abouelatta, O.B., 2010, 3D Surface Roughness Measurement Using a Light
Sectioning Vision System, in: Proceedings of the World Congress on
Engineering (30th June–2nd July 2010, London, U.K., vol. I, WCE),
pp.698–703.
[2] Acharya, T., Roy, A.K., 2005, Image Processing: Principles and Applications,
Wiley-Interscience, New Jersey.
[3] Akbari, A.A., Fard, A.M., Chegini, A.G., 2006, An Effective Image Based Surface
Roughness Estimation Approach Using Neural Network, IEEE World Automa-
tion Congress (WAC) (24–26th July 2006, Budapest, Hungary, IEEE), pp.1–6.
[4] Alegre, E., Barreiro, J., Castejo´ n, M., Suarez, S., 2008, Computer Vision and
Classification Techniques on the Surface Finish Control in Machining Pro-
cesses, in Campilho A, Kamel M, (Eds.) Image Analysis and Recognition, LNCS
5112. Springer-Verlag, Berlin, pp. pp.1101–1110.
[5] Alegre, E., Barreiro, J., Fernandez, R.A., Castejn, M., 2006, Design of a Computer
Vision System to Estimate Tool Wearing, Materials Science Forum, 526:61–
66.
[6] Alegre, E., Rodrı´guez, R.A., Barreiro, J., Ruiz, J., 2009, Use of Contour Signatures
and Classification Methods to Optimize the Tool Life in Metal Machining,
Estonian Journal of Engineering, 15:3–12.
[7] Al-Kindi, G.A., Banl, R.M., Gilt, K.F., 1992, An Application of Machine Vision in
the Automated Inspection of Engineering Surface, International Journal of
Production Research, 30:241–253.
[8] Al-Kindi, G.A., Shirinzadeh, B., 2007, An Evaluation of Surface Roughness
Parameters Measurement Using Vision-based Data, International Journal of
Machine Tools and Manufacture, 47:697–708.
[9] Al-Kindi, G., Zughaer, H., 2012, An Approach to Improved CNC Machining
Using Vision-based System, Materials and Manufacturing Processes, 27:765–
774.
[10] Arunachalam, N., Ramamoorthy, B., 2007, Texture Analysis for Grinding
Wheel Wear Assessment Using Machine Vision, Proceedings of the Institu-
tion of Mechanical Engineers Part B Journal of Engineering, 221:419–430.
[11] Atli, A.V., Urhan, O., Ertu¨ rk, S., So¨ nmez, M., 2006, A Computer Vision-based
Fast Approach to Drilling Tool Condition Monitoring, P I Mech Eng B-J Eng,
220:1409–1415.
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 229
[12] Bamberger, H., Ramachandran, S., Hong, E., Katz, R., 2011, Identification of
Machining Chatter Marks on Surfaces of Automotive Valve Seats, Journal of
Manufacturing Science and Engineering-Transactions of the ASME,
133:0410031–0410037.
[13] Barreiro, J., Castejon, M., Alegre, E., Hernandez, L.K., 2008, Use of Descriptors
Based on Moments From Digital Images for Tool Wear Monitoring, Interna-
tional Journal of Machine Tools and Manufacture, 48:1005–1013.
[14] Beyerer, J., Leo´ n, F.P., 1997, Suppression of Inhomogeneities in Images of
Surfaces, Optical Engineering, 36:85–93.
[15] Boryczko, A., 2013, Effect of Waviness and Roughness Components on
Transverse Profiles of Turned Surfaces, Measurement, 46:688–696.
[16] Bradley, C., Wong, Y.S., 2001, Surface Texture Indicators of Tool Wear – A
Machine Vision Approach, International Journal of Advanced Manufacturing
Technology, 17:435–443.
[17] Byrne, G., Dornfeld, D., lnasaki, I., Ketteler, G., Konig, W., Teti, R., 1995, Tool
Condition Monitoring (TCM) – The Status of Research and Industrial Appli-
cation, CIRP Annals-Manufacturing Technology, 44:541–567.
[18] Castejon, M., Alegre, E., Barreiro, J., Hernandez, L.K., 2007, On-line Tool Wear
Monitoring Using Geometric Descriptors From Digital Images, International
Journal of Machine Tools and Manufacture, 47:1847–1853.
[19] Chandrasekaran, M., Muralidhar, M., Krishna, M.C., Dixit, U.S., 2010, Applica-
tion of Soft Computing Techniques in Machining Performance Prediction and
Optimization: A Literature Review, International Journal of Advanced
Manufacturing Technology, 46:445–464.
[20] Cuthbert, L., Huynh, V., 1992, Statistical Analysis of Optical Fourier Transform
Patterns for Surface Texture Measurement, Measurement Science and Tech-
nology, 3:740–745.
[21] Datta, A., Dutta, S., Pal, S.K., Sen, R., Mukhopadhyay, S., 2012, Texture Analysis
of Turned Surface Images Using Grey Level Co-occurrence Technique,
Advances in Materials Research, 365:38–43.
[22] Dawson, T.G., Kurfess, T.R., 2005, Quantification of Tool Wear Using White
Light Interferometry and Three-dimensional Computational Metrology, In-
ternational Journal of Machine Tools and Manufacture, 45:591–596.
[23] Demircioglu, P., Durakbasa, M.N., 2011, Investigations on Machined Metal
Surfaces Through the Stylus Type and Optical 3D Instruments and Their
Mathematical Modelling With the Help of Statistical Techniques, Measure-
ment, 44:611–619.
[24] Devillez, A., Lesko, S., Mozerc, W., 2004, Cutting Tool Crater Wear Measure-
ment With White Light Interferometry, Wear, 256:56–65.
[25] Dhanasekar, B., Krishna Mohan, N., Bhaduri, B., Ramamoorthy, B., 2008,
Evaluation of Surface Roughness Based on Monochromatic Speckle Correla-
tion Using Image Processing, Precision Engineering, 32:196–206.
[26] Dhanasekar, B., Ramamoorthy, B., 2010, Restoration of Blurred Images for
Surface Roughness Evaluation Using Machine Vision, Tribology International,
43:268–276.
[27] Dhanasekar, B., Ramamoorthy, B., 2006, Evaluation of Surface Roughness
Using a Image Processing and Machine Vision System, MAPAN-Journal of
Metrology Society of India, 21:9–15.
[28] Dhanasekar, B., Ramamoorthy, B., 2008, Assessment of Surface Roughness
Based on Super Resolution Reconstruction Algorithm, International Journal of
Advanced Manufacturing Technology, 35:1191–1205.
[29] Dimla, E., Dimla, S., 2000, Sensor signals for Tool-wear Monitoring in Metal
Cutting Operations—A Review of Methods, International Journal of Machine
Tools and Manufacture, 40:1073–1098.
[30] Duan, G., Chen, Y.W., Sukegawa, T., 2010, Automatic Optical Flank Wear
Measurement of Microdrills Using Level Set for Cutting Plane Segmentation,
Machine Vision and Applications, 21:667–676.
[31] Dutta, S., Datta, A., Chakladar, N.D., Pal, S.K., Mukhopadhyay, S., Sen, R., 2012,
Detection of Tool Condition From the Turned Surface Images Using an
Accurate Grey Level Co-occurrence Technique, Precision Engineering,
36:458–466.
[32] Elango, V., Karunamoorthy, L., 2008, Effect of Lighting Conditions in the
Study of Surface Roughness by Machine Vision – An Experimental Design
Approach, International Journal of Advanced Manufacturing Technology,
37:92–103.
[33] Elbestawi, M.A., Ismail, F., Yuen, K.M., 1994, Surface Topography Characteri-
zation in Finish Milling, International Journal of Machine Tools and Manu-
facture, 34:245–255.
[34] Elbestawi, M.A., Dumitrescu, M., Ng, E.G., 2006, Tool Condition Monitoring in
Machining Condition Monitoring and Control for Intelligent Manufacturing,
Springer Series in Advanced Manufacturing, 55–82.
[35] Fadare, D.A., Oni, A.O., 2009, Development and Application of a Machine
Vision System for Measurement of Tool Wear, ARPN Journal of Engineering
and Applied Sciences, 4:42–49.
[36] Gadelmawla, E.S., 2004, A Vision System for Surface Roughness Characteri-
zation Using the Gray Level Co-occurrence Matrix, NDT&E international,
37:577–582.
[37] Gadelmawla, E.S., Eladawi, A.E., Abouelatta, O.B., Elewa, I.M., 2008, Investi-
gation of the Cutting Conditions in Milling Operations Using Image Texture
Features, Proceedings of the Institution of Mechanical Engineers Part B
Journal of Engineering, 222:1395–1404.
[38] Gadelmawla, E.S., Eladawi, A.E., Abouelatta, O.B., Elewa, I.M., 2009, Applica-
tion of Computer Vision for the Prediction of Cutting Conditions in Milling
Operations, Proceedings of the Institution of Mechanical Engineers Part B
Journal of Engineering, 223:791–800.
[39] Gadelmawla, E.S., 2011, Estimation of Surface Roughness for Turning Opera-
tions Using Image Texture Features, Proceedings of the Institution of Me-
chanical Engineers Part B Journal of Engineering, 225:1281–1292.
[40] Galante, G., Piacentini, M., Ruisi, V.F., 1991, Surface Roughness Detection by
Tool Image Processing, Wear, 148:211–220.
[41] Gonzalez, R.C., Woods, R.E., 2002, Digital Image Processing, 2nd ed. Prentice-
Hall Inc., New Jersey.
[42] Gupta, M., Raman, S., 2001, Machine Vision Assisted Characterization of
Machined Surfaces, International Journal of Production Research, 39:759–
784.
[43] Heger, T., Pandit, M., 2004, Optical Wear Assessment for Grinding Tools,
Journal of Electronic Imaging, 13:450–461.
[44] Ho, S.Y., Lee, K.C., Chen, S.S., Ho, S.J., 2002, Accurate Modeling and Prediction
of Surface Roughness by Computer Vision in Turning Operations Using an
Adaptive Neuro-fuzzy Inference System, International Journal of Machine
Tools and Manufacture, 42:1441–1446.
[45] Hoy, D.E.P., Yu, F., 1991, Surface Quality Assessment Using Computer Vision
Methods, Journal of Materials Processing Technology, 28:265–274.
[46] Ikonen, L., Toivanen, P.J., 2007, Distance and Nearest Neighbor Transforms on
Gray-level Surfaces, Pattern Recognition Letters, 28:604–612.
[47] Inoue, S., Konishi, M., Imai, J., 2009, Surface Defect Inspection of a Cutting Tool
by Image Processing with Neural Networks, Memoirs of the Faculty of
Engineering Okayama University, 43:55–60.
[48] ISO3685: Tool-life Testing with Single-point Turning Tools, 1993.
[49] Jackson, M.J., Robinson, G.M., Hyde, L.J., Rhodes, R., 2006, Neural Image
Processing of the Wear of Cutting Tools Coated with Thin Films, Journal of
Materials Engineering and Performance, 15:223–229.
[50] Ja¨ hne, B., 2002, Digital Image Processing, Springer-Verlag Heidelberg, Berlin.
[51] Jantunen, E., 2002, A Summary of Methods Applied to Tool Condition Moni-
toring in Drilling, International Journal of Machine Tools and Manufacture,
42:997–1010.
[52] Jemielniak, K., 1999, Commercial Tool Condition Monitoring Systems, Inter-
national Journal of Advanced Manufacturing Technology, 15:711–721.
[53] Jemielniak, K., Arrazola, P.J., 2008, Application of AE and Cutting Force Signals
in Tool Condition Monitoring in Micro-milling, CIRP Journal of Manufacturing
Science and Technology, 1:97–102.
[54] Jetley, S., Selven, D., 1993, Applying Machining Vision to Surface Texture
Analysis, in: Proceedings of the 36th Midwest Symposium on Circuits and
Systems (16–18th August, 1993, Detroit, USA: IEEE), pp.1456–1459.
[55] Jian, Z., Jin, Z., 2010, Surface Roughness Measure Based on Average Texture
Cycle, Second International Conference on Intelligent Human-Machine Sys-
tems and Cybernetics (Nanjing, China), .
[56] Josso, B., Burton, D.R., Lalor, M.J., 2002, Frequency Normalised Wavelet
Transform for Surface Roughness Analysis and Characterisation, Wear,
252:491–500.
[57] Josso, B., Burton, D.R., Lalor, M.J., 2001, Wavelet Strategy for Surface Rough-
ness Analysis and Characterisation, Computer Methods in Applied Mechanics
and Engineering, 191:829–842.
[58] Jurkovic, J., Korosec, M., Kopac, J., 2005, New Approach in Tool Wear Mea-
suring Technique Using CCD Vision System, International Journal of Machine
Tools and Manufacture, 45:1023–1030.
[59] Kai C, (Ed.) (2006), Machining Dynamics: Fundamentals, Applications and
Practices. Springer Series in Advanced Manufacturing Springer-Verlag.
[60] Kang, M.C., Kim, J.S., Kim, K.H., 2005, Fractal Dimension Analysis of Machined
Surface Depending on Coated Tool Wear, Surface and Coatings Technology,
193:259–265.
[61] Karthik, A., Chandra, S., Ramamoorthy, B., Das, S., 1997, 3D Tool Wear
Measurement and Visualisation Using Stereo Imaging, International Journal
of Machine Tools and Manufacture, 37:1573–1581.
[62] Kashiwagi, H., Nagayamal, Y., Shibuta, K., Watanabe, H., Oka, H., 2003,
Estimation of Tool Wear by Use of Image Processing of Cutting Dust, SICE
Annual Conference (Fukui, Japan), .
[63] Kassim, A.A., Mannan, M.A., Mian, Z., 2007, Texture Analysis Methods for Tool
Condition Monitoring, Image and Vision Computing, 25:1080–1090.
[64] Kassim, A.A., Mian, Z., Mannan, M.A., 2004, Connectivity Oriented Fast
Hough Transform for Tool Wear Monitoring, Pattern Recognition,
37:1925–1933.
[65] Kassim, A.A., Mian, Z., Mannan, M.A., 2006, Tool Condition Classification
Using Hidden Markov Model Based on Fractal Analysis of Machined Surface
Textures, Machine Vision and Applications, 17:327–336.
[66] Kassim, A.A., Mian, Z., Mannan, M.A., 2002, Texture Analysis Using Fractals for
Tool Wear Monitoring, in: Proceedings of IEEE International Conference on
Image processing (24–28th June), pp.105–108.
[67] Kassim, A.A., Mannan, M.A., Jing, M., 2000, Machine Tool Condition Monitor-
ing Using Workpiece Surface Texture Analysis, Machine Vision and Applica-
tions, 11:257–263.
[68] Kerr, D., Pengilley, J., Garwood, R., 2006, Assessment and Visualisation of
Machine Tool Wear Using Computer Vision, International Journal of Ad-
vanced Manufacturing Technology, 28:781–791.
[69] Khalifa, O.O., Densibali, A., Faris, W., 2006, Image Processing for Chatter
Identification in Machining Processes, International Journal of Advanced
Manufacturing Technology, 31:443–449.
[70] Kim, J.H., Moon, D.K., Lee, D.W., Kim, J.S., Kang, M.C., Kim, K.H., 2002, Tool
Wear Measuring Technique on the Machine Using CCD and Exclusive jig,
Journal of Materials Processing Technology, 130–131:668–674.
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 230
[71] Kiran, M., Ramamoorthy, B., Radhakrishnan, V., 1998, Evaluation of Surface
Roughness by Vision System, International Journal of Machine Tools and
Manufacture, 38:685–690.
[72] Kumar, R., Kulashekar, P., Dhanasekar, B., Ramamoorthy, B., 2005, Application
of Digital Image Magnification for Surface Roughness Evaluation Using
Machine Vision, International Journal of Machine Tools and Manufacture,
45:228–234.
[73] Kurada, S., Bradley, C., 1997, A Machine Vision System for Tool Wear Assess-
ment, Tribology International, 30:295–304.
[74] Kurada, S., Bradley, C., 1997, A Review of Machine Vision Sensors for Tool
Condition Monitoring, Computers in Industry, 34:55–72.
[75] Lachance, S., Bauer, R., Warkentin, A., 2004, Application of Region Growing
Method to Evaluate the Surface Condition of Grinding Wheels, International
Journal of Machine Tools and Manufacture, 44:823–829.
[76] Lanzetta, M., 2001, A New Flexible High-resolution Vision Sensor for Tool
Condition Monitoring, Journal of Materials Processing Technology, 119:73–
82.
[77] Lee, B.Y., Juan, H., Yu, S.F., 2002, A Study of Computer Vision for Measuring
Surface Roughness in the Turning Process, International Journal of Advanced
Manufacturing Technology, 19:295–301.
[78] Lee, B.Y., Yu, S.F., Juan, H., 2004, The Model of Surface Roughness Inspection
by Vision System in Turning, Mechatronics, 14:129–141.
[79] Lee, K.C., Ho, S.J., Ho, S.Y., 2005, Accurate Estimation of Surface Roughness
From Texture Features of the Surface Image Using an Adaptive Neuro-fuzzy
Inference System, Precision Engineering, 29:95–100.
[80] Leo´ n, F.P., 2005, An Objective Measure of the Quality of Honed Surfaces,
Proceedings of SPIE, 5856:287–295.
[81] Li, P.Y., Hao, C.Y., Zhu, S.W., 2007, Machining Tools Wear Condition Detection
Based on Wavelet Packet, in: Proceedings of the Sixth International Confer-
ence on Machine Learning and Cybernetics (Hong Kong), .
[82] Liang, Y.T., Chiou, Y.C., 2006, An effective Drilling Wear Measurement Based
on Visual Inspection Technique, in: Proceedings of the 2006 Joint Conference
on Information Sciences (JCIS 2006) (Kaohsiung, Taiwan), .
[83] Liang, Y.T., Chiou, Y.C., Louh, C.J., 2005, Automatic Wear Measurement of Ti-
based Coatings Milling via Image Registration, IAPR Conference on Machine
Vision Applications (Tsukuba Science City, Japan), .
[84] Liu, W., Zheng, X., Liu, S., Jia, Z., 2012, A Roughness Measurement Method
Based on Genetic Algorithm and Neural Network for Microheterogeneous
Surface in Deep-hole Parts, Journal of Circuits Systems and Computers,
21:1250005–1250019.
[85] Luk, F., Huynh, V., 1987, A Vision System for In-process Surface Quality
Assessment, in: Proceedings of the Vision ‘87 SME Conference (Detroit,
Michigan), pp.1243–1258.
[86] Makki, H., Heinemann, R.K., Hinduja, S., Owodunni, O.O., 2009, Online De-
termination of Tool Run-out and Wear Using Machine Vision and Image
Processing Techniques, Innovative Production Machines and Systems (Cardiff
University, Wales, UK), .
[87] Mannan, M.A., Kassim, A.A., Jing, M., 2000, Application of Image and Sound
Analysis Techniques to Monitor the Condition of Cutting Tools, Pattern
Recognition Letters, 21:969–979.
[88] McBride, J., Maul, C., 2004, The 3D Measurement and Analysis of High
Precision Surfaces Using Confocal Optical Methods, IEICE – Transactions
on Electronics, E87-C:1261–1267.
[89] Myshkin, N.K., Grigoriev, A.Y., Chizhik, S.A., Choi, K.Y., Petrokovets, M.I., 2003,
Surface Roughness and Texture Analysis in Microscale, Wear, 254:1001–
1009.
[90] Nakao, Y., 2001. Measurement of Drilling Burr by Image Processing Tech-
nique. Available from: http://www.aspe.net/publications/Annual_2001/PDF/
POSTERS/.../1142.PDF (accessed 06.07.10).
[91] Narayanan, M.R., Gowri, S., Krishna, M.M., 2007, On line Surface Roughness
Measurement Using Image Processing and Machine Vision, in: Proceedings of
the World Congress on Engineering 2007 (WCE, London, UK), .
[92] Nebot, J.V.A., Subiro´ n, F.R., 2010, A Review of Machining Monitoring Systems
Based on Artificial Intelligence Process Models, International Journal of
Advanced Manufacturing Technology, 47:237–257.
[93] Ng, K.W., Moon, K.S., 2001, Measurement of 3-D Tool Wear Based on Focus
Error and Micro-coordinate Measuring System, in: Proceedings of the 16th
Annual Meeting of American Society for Precision Engineering (Virginia,
USA), .
[94] Niola, V.N., Nasti, G., Quaremba, G., 2005, A Problem of Emphasizing Features
of a Surface Roughness by Means the Discrete Wavelet Transform, Journal of
Materials Processing Technology, 164–165:1410–1415.
[95] Okada, S., Imade, M., Miyauchi, H., Isomoto, K., Miyoshi, T., Sumimoto, T.,
Yamamoto, H., 1998, 3-D Shape Measurement of Free-form Machined Sur-
faces by Optical Ring Imaging System, in: Proceedings of the 24th Annual
Conference of the IEEE Industrial Electronics Society (31st August–4th Sep-
tember, 1998, Aachen, Germany), pp.1284–1289.
[96] Otieno, A., Pedapati, C., Wan, X., Zhang, H., 2006. Imaging and wear analysis of
micro-tools using machine vision. Indian Institute of Technology Kharagpur.
Available from: http://www.ijme.us/cd_06/PDF/IT%20301-071.pdf (accessed
18.07.10).
[97] Palani, S., Natarajan, U., 2011, Prediction of Surface Roughness in CNC End
Milling by Machine Vision System Using Artificial Neural Network Based on
2D Fourier Transform, International Journal of Advanced Manufacturing
Technology, 54:1033–1042.
[98] Persson, U., 2006, Surface Roughness Measurement on Machined Surfaces
Using Angular Speckle Correlation, Journal of Materials Processing Technol-
ogy, 180:233–238.
[99] Pfeifer, T., Wiegers, L., 2000, Reliable Tool Wear Monitoring by Optimized
Image and Illumination Control in Machine Vision, Measurement, 28:209–
218.
[100] Prasad, K.N., Ramamoorthy, B., 2001, Tool Wear Evaluation by Stereo Vision
and Prediction by Artificial Neural Network, Journal of Materials Processing
Technology, 112:43–52.
[101] Prasad, B.S., Sarcar, M.M.M., Ben, B.S., 2011, Surface Textural Analysis Using
Acousto Optic Emission and Vision-based 3D Surface Topography—A Base for
Online Tool Condition Monitoring in Face Turning, International Journal of
Advanced Manufacturing Technology, 55:1025–1035.
[102] Prickett, P.W., Johns, C., 1999, An Overview of Approaches to End Milling Tool
Monitoring, International Journal of Machine Tools and Manufacture,
39:105–122.
[103] Ramamoorthy, B., Radhakrishnan, V., 1993, Statistical Approaches to Surface
Texture Classification, Wear, 167:155–161.
[104] Ramana, K.V., Ramamoorthy, B., 1996, Statistical Methods to Compare the
Texture Features of Machined Surfaces, Pattern Recognition, 29:1447–1459.
[105] Rehorn, A.G., Jiang, J., Orban, P.E., 2005, State-of-the-Art Methods and Results
in Tool Condition Monitoring: A Review, International Journal of Advanced
Manufacturing Technology, 26:693–710.
[106] Roth, J.T., Djurdjanovic, D., Yang, X., Mears, L., Kurfess, T., 2010, Quality and
Inspection of Machining Operations – Tool Condition Monitoring, Journal of
Manufacturing Science and Engineering-Transactions of the ASME, 132.
041015-1–041015-16.
[107] Russ, J.C., 1998, Fractal Dimension Measurement of Engineering Surfaces,
International Journal of Machine Tools and Manufacture, 38:567–571.
[108] Sahabi, H.H., Ratnam, M.M., 2009, Assessment of Flank Wear and Nose Radius
Wear From Workpiece Roughness Profile in Turning Operation Using Ma-
chine Vision, International Journal of Advanced Manufacturing Technology,
43:11–21.
[109] Sahabi, H.H., Ratnam, M.M., 2008, On-line Monitoring of Tool Wear in
Turning Operation in the Presence of Tool Misalignment, International
Journal of Advanced Manufacturing Technology, 38:718–727.
[110] Sarma, P.M.M.S., Karunamoorthy, L., Palanikumar, K., 2009, Surface Rough-
ness Parameters Evaluation in Machining GFRP Composites by PCD Tool
Using Digital Image Processing, Journal of Reinforced Plastics and Compo-
sites, 28:1567–1585.
[111] Schmahling, J., Hamprecht, F.A., Hoffmann, D.M.P., 2006, A Three-dimension-
al Measure of Surface Roughness Based on Mathematical Morphology, In-
ternational Journal of Machine Tools and Manufacture, 46:1764–1769.
[112] Schmitt, R., Hermes, R., Stemmer, M., Pavim, A., Deschamps, F., Moraes, D.,
2005, Machine Vision Prototype for Flank Wear Measurement on Milling
Tools, 38th CIRP Manufacturing Systems Seminar (Florianopolis, Brazil), .
[113] Senin, N., Ziliotti, M., Groppetti, R., 2007, Three-dimensional Surface Topog-
raphy Segmentation Through Clustering, Wear, 262:395–410.
[114] Sharan, R.V., Onwubolu, G.C., 2011, Measurement of End-milling Burr Using
Image Processing Techniques, Proceedings of the Institution of Mechanical
Engineers Part B Journal of Engineering, 225:448–452.
[115] Singh, V., Mishra, R., 2006, Developing a Machine Vision System for Spangle
Classification Using Image Processing and Artificial Neural Network, Surface
and Coatings Technology, 201:2813–2817.
[116] Sortino, M., 2003, Application of Statistical Filtering for Optical Detection of
Tool Wear, International Journal of Machine Tools and Manufacture, 43:493–
497.
[117] Stemmer, M., Pavim, A., Adur, M., Deschamps, F., Schmitt, R., Hermes, R.,
2005, Machine Vision and Neural Networks Applied to Wear Classification on
Cutting Tools, Poceedings of the EOS Conference on Industrial Imaging and
Machine Vision (Munich, Germany), .
[118] Su, J.C., Huang, C.K., Tarng, Y.S., 2006, An automated Flank Wear Measure-
ment of Microdrills Using Machine Vision, Journal of Materials Processing
Technology, 180:328–335.
[119] Tarng, Y.S., Lee, B.Y., 2001, Surface Roughness Inspection by Computer Vision
in Turning Operations, International Journal of Machine Tools and Manufac-
ture, 41:1251–1263.
[120] O. Tauno, K. Lembit, 2000. Digital tool wear measuring video system, Estonia.
Available from: http://mess.me.ttu.ee/Otto/DAAAM.pdf (Accessed 08.06.10).
[121] Teti, R., Jemielniak, K., O’Donnell, G., Dornfeld, D., 2010, Advanced Monitoring
of Machining Operations, CIRP Annals-Manufacturing Technology, 59:717–
739.
[122] Tomkiewicz, A.Z., 2010, Estimation of Surface Roughness Parameter Based on
Machined Surface Image, Metrology and Measurement Systems XVII, 493–
504.
[123] Tsai, D.M., Chen, J.J., Chert, J.F., 1998, A Vision System for Surface Roughness
Assessment Using Neural Networks, International Journal of Advanced
Manufacturing Technology, 14:412–422.
[124] Tsai, D.M., Wu, S.K., 2000, Automated Surface Inspection Using Gabor Filters,
International Journal of Advanced Manufacturing Technology, 16:474–482.
[125] Tuceryan, M., Jain, A.K., 1998, Texture Analysis, in Chen CH, Pau LF, Wang PSP,
(Eds.) The Handbook of Pattern Recognition and Computer Vision. World
Scientific, Singapore, pp. pp.207–248.
[126] Vesselenyi, T., Dzitac, I., Dzitac, S., Vaida, V., 2008, Surface Roughness Image
Analysis Using Quasi-fractal Characteristics and Fuzzy Clustering Methods,
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 231
International Journal of Computers Communications and Control, 3:304–
316.
[127] Wang L, Gao RX, (Eds.) (2006), Condition Monitoring and Control for Intelli-
gent Manufacturing. Springer Series in Advanced Manufacturing. Springer-
Verlag Limited, London.
[128] Wang, W.H., Hong, G.S., Wong, Y.S., 2006, Flank Wear Measurement by a
Threshold Independent Method with Sub-pixel Accuracy, International Jour-
nal of Machine Tools and Manufacture, 46:199–207.
[129] Wang, W., Hong, G.S., Wong, Y.S., 2006, Flank Wear Measurement by a
Threshold Independent Method with Sub-pixel Accuracy, International Jour-
nal of Machine Tools and Manufacture, 46:199–207.
[130] Wang, W.H., Wong, Y.S., Hong, G.S., 2006, 3D Measurement of Crater Wear by
Phase Shifting Method, Wear, 261:164–171.
[131] Wang, W., Wong, Y.S., Hong, G.S., 2005, Flank Wear Measurement by Suc-
cessive Image Analysis, Computers and Industrial, 56:816–830.
[132] W. Weis, 1993. Tool Wear Measurement on basis of optical sensors, vision
systems and neuronal networks (application milling). WESCON’93. San
Fransisco, USA.
[133] Whitehead, S.A., Shearer, A.C., Watts, D.C., Wilson, N.H.F., 1999, Comparison
of Two Stylus Methods for Measuring Surface Texture, Dental Materials,
15:79–86.
[134] Wong, F.S., Nee, A.F.C., Li, X.Q., Reisdorj, C., 1997, Tool Condition Monitoring
Using Laser Scatter Pattern, Journal of Materials Processing Technology,
63:205–210.
[135] Xiong, G., Liu, J., Avila, A., 2011. Cutting Tool Wear Measurement by Using
Active Contour Model Based Image Processing, in: Proceedings of the 2011
IEEE International Conference on Mechatronics and Automation (Beijing,
China).
[136] Yang, M., Kwon, O., 1998, A Tool Condition Recognition System Using Image
Processing, Control Engineering Practice, 6:1389–1395.
[137] Yang, M., Kwon, O., 1996, Crater Wear Measurement Using Computer Vision
and Automatic Focusing, Journal of Materials Processing Technology,
58:362–367.
[138] Yasui, H., Haraki, Y., Sakata, M., 2001, Development of Automatic Image
Processing System for Evaluation of Wheel Surface Condition in Ultra-
smoothness Grinding, in: Proceedings of the 16th Annual Meeting of Ameri-
can Society for Precision Engineering (Virginia, USA), .
[139] Yoon, H., Chung, S.C., 2004, Vision Inspection of Micro-drilling Processes on
the Machine Tool, Transactions of NAMRI/SME, 32:391–394.
[140] Younis, M.A., 1998, On line Surface Roughness Measurements Using Image
Processing Towards an Adaptive Control, Computers and Industrial Engi-
neering, 35:49–52.
[141] Yu, J., 2012, Machine Tool Condition Monitoring Based on an Adaptive
Gaussian Mixture Model, Journal of Manufacturing Science and Engineer-
ing-Transactions of the ASME, 134. 031004-1–031004-13.
[142] Zhang, X., Krewet, C., Kuhlenko¨ tter, B., 2006, Automatic Classification of
Defects on the Product Surface in Grinding and Polishing, International
Journal of Machine Tools and Manufacture, 46:59–69.
[143] Zhongxiang, H., Lei, Z., Jiaxu, T., Xuehong, M., Xiaojun, S., 2009, Evaluation of
Three-dimensional Surface Roughness Parameters Based on Digital Image
Processing, International Journal of Advanced Manufacturing Technology,
40:342–348.
S. Dutta et al. / CIRP Journal of Manufacturing Science and Technology 6 (2013) 212–232 232

Sponsor Documents

Or use your account on DocShare.tips

Hide

Forgot your password?

Or register your new account on DocShare.tips

Hide

Lost your password? Please enter your email address. You will receive a link to create a new password.

Back to log-in

Close