ON-LINE MEASUREMENT AND DETECTION OF BORING TOOL WEAR
Robert Jay Jolley
B.S., University of California, Berkeley, 1992
THESIS
Submitted in partial satisfaction of
the requirements for the degree of
MASTER OF SCIENCE
in
MECHANICAL ENGINEERING
at
CALIFORNIA STATE UNIVERSITY, SACRAMENTO
FALL
2011
ONLINE MEASUREMENT AND DETECTION OF BORING TOOL WEAR
A Thesis
by
Robert Jay Jolley
Approved by:
__________________________________, Committee Chair
Tien-I Liu
__________________________________, Second Reader
Kenneth Sprott
____________________________
Date
ii
Student: Robert Jay Jolley
I certify that this student has met the requirements for format contained in the University
format manual, and that this thesis is suitable for shelving in the Library and credit is to
be awarded for the thesis.
__________________________, Graduate Coordinator
Akihiko Kumagai
Department of Mechanical Engineering
iii
___________________
Date
Abstract
of
ONLINE MEASUREMENT AND DETECTION OF BORING TOOL WEAR
by
Robert Jay Jolley
In order to assure precision and quality of boring operations, on-line tool condition
monitoring is essential. In this research, Counterpropagation Neural Networks (CPN’s)
and Adaptive Neuro-Fuzzy Inference Systems (ANFIS) are used in conjunction with
features extracted from 3-axis cutting force data for the on-line measurement and
detection of tool wear for precision boring.
Force data was collected for carbide inserts during the boring of titanium parts. At the
end of each boring operation, the average flank wear width was measured to determine
the cutting tool conditions. Measurements were accomplished with the aid of a
toolmaker’s microscope.
Fourteen features were extracted from the cutting force data. In order to determine
which features showed the best indication of cutting tool conditions, a Sequential
Forward Search (SFS) algorithm was utilized to reduce the number of features for on-line
measurements and detection of precision boring of titanium parts. The selected two most
iv
prominent features were kurtosis of longitudinal force and average of the ratio between
tangential force and radial force.
On-line classification showed excellent results, using both a 2x30x1 CPN and a 2x2
ANFIS, of being able to predict tool conditions on-line with 100% accuracy. On-line
measurements also produced exceedingly successful results with a minimum error for a
1x10 ANFIS of 0.87% and a minimum error for a 3x69x1 CPN of 8.46%.
_______________________, Committee Chair
Tien-I Liu
_______________________
Date
v
ACKNOWLEDGEMENTS
This thesis would not be possible without the support of my loving wife Janelle and
children Jared and Jacob. As they have provided inspiration in my life, I hope my
accomplishments will also inspire them.
I would like to acknowledge and thank my thesis advisor Dr. Tien-I Liu for providing
this research opportunity as well as his guidance and assistance in the creation of this
document.
I would also like to acknowledge the professors, and staff, within the Mechanical
Engineering Department at California State University, Sacramento, for providing a great
environment in which to study and learn.
vi
TABLE OF CONTENTS
Page
Acknowledgements ............................................................................................................ vi
List of Tables ..................................................................................................................... ix
List of Figures ......................................................................................................................x
Chapter
1. INTRODUCTION ..........................................................................................................1
2. EXPERIMENTATION AND DATA ACQUISITION ..................................................3
3. FEATURE EXTRACTION AND SELECTION............................................................6
3.1 Feature Extraction ............................................................................................ 6
3.2 Feature Selection .............................................................................................. 8
4. ARTIFICIAL INTELLIGENCE TECHNIQUES.........................................................12
4.1 Counterpropagation Neural Networks (CPN’s) ............................................. 12
4.2 Adaptive Neuro-Fuzzy Inference System (ANFIS) ....................................... 15
5. ON-LINE MONITORING AND MEASUREMENTS OF TOOL WEAR FOR
TITANIUM BORING ...................................................................................................21
5.1 Training .......................................................................................................... 21
5.2 On-Line Classification ................................................................................... 24
5.3 On-Line Measurement ................................................................................... 29
6. CONCLUSIONS...........................................................................................................35
Appendix A Extracted Features and Average Flank Wear Measurements ........................37
Appendix B CPN Network Training Data .........................................................................48
Appendix C ANFIS Network Training Data .....................................................................58
vii
Appendix D Data for On-line Testing ...............................................................................68
Appendix E CPN On-line Classification Outputs ..............................................................70
Appendix F ANFIS On-line Classification Outputs ..........................................................76
Appendix G CPN On-line Measurement Outputs .............................................................96
Appendix H ANFIS On-line Measurement Outputs ........................................................102
References ........................................................................................................................122
viii
LIST OF TABLES
Page
Table 1 List of Fourteen Extracted Features .................................................................. 6
Table 2 Outcome of Feature Selection Process............................................................ 11
Table 3 Experimental Data Used in the Training Process ........................................... 21
Table 4 Data for On-line Classification and Measurements ........................................ 24
Table 5 CPN On-line Classification Success Rate ....................................................... 25
Table 6 ANFIS On-line Classification Success Rate ................................................... 26
Table 7 ANFIS On-line Classification Success Rate ................................................... 27
Table 8 Actual Outputs for the Best On-line Classification Structures ....................... 28
Table 9 CPN On-line Measurement Average Output Error ......................................... 30
Table 10 ANFIS On-line Measurement Average Output Error ................................... 31
Table 11 ANFIS On-line Measurement Average Output Error ................................... 32
Table 12 Best Results for On-line Measurements ...................................................... 33
ix
LIST OF FIGURES
Page
Figure 1 SEM Photograph of a worn out tool ................................................................ 4
Figure 2 Schematic Diagram of Boring Experiment...................................................... 5
Figure 3 2x30x1 CPN Architecture.............................................................................. 13
Figure 4 2x3 ANFIS Architecture ................................................................................ 16
Figure 5 Comparison of Actual Tool Wear and On-line Measurements .................... 33
x
1
Chapter 1
INTRODUCTION
To remain viable in today’s global economy, companies need to continually enhance
the accuracy and quality of manufacturing processes and eliminate machine down time.
This is achievable by monitoring the health of manufacturing machinery, through sensors
attached to the equipment, in a process known as Condition Based Maintenance (CBM).
Facilitated by advances in sensor technology and condition monitoring techniques, CBM
enables continuous improvements in any modern manufacturing facility [1, 2].
An important part of CBM for machining operations is tool condition monitoring
(TCM). The condition of a cutting tool has a direct relationship on the precision and
quality of a machined part. With the ability to monitor the tool condition, the machining
precision and quality can be assured and the machine down time can be greatly reduced.
However, owing to its nonlinear and stochastic nature, predicting or monitoring tool wear
is a difficult task [3].
Fortunately, the use of artificial neural networks in conjunction with indirect
monitoring methods has produced promising results [4]. To this point, TCM research has
been conducted utilizing cutting forces, acoustic emissions, vibration, and thermal
measurements for the monitoring of turning, drilling, milling and similar machining
operations [5-22].
This research explores the viability of using Counterpropagation Neural Networks
(CPN’s) and Adaptive Neuro-Fuzzy Inference Systems (ANFIS) along with features
extracted from the cutting forces of titanium boring operations for on-line monitoring and
2
measurement of tool wear. Section 2 illustrates experimentation and data acquisition.
Feature extraction and selection are discussed in section 3. Artificial Intelligence
Techniques are discussed in section 4. On-line monitoring and measurements of tool
wear for titanium boring are explained in section 5. Conclusions are given in section 6.
3
Chapter 2
EXPERIMENTATION AND DATA ACQUISITION
The material machined, during data acquisition, was Titanium 6Al4V with a hardness
of Rc 28. The raw stock had an inside diameter of 38.1 mm (1.5 inches), an outside
diameter of 116.84 mm (4.6 inches), and a length of 57.15 mm (2.25 inches). The
material was bored from an inner diameter of 38.1 mm (1.5 inches) to 88.9 mm (3.5
inches). Kennametal TNMG-332 triangular C-2 carbide inserts, mounted on a
Kennametal A16-DTFNR3 boring bar, were used on a Lodge and Shipley automatic lathe
to perform the boring operation. The cutting speed of the boring operation was 1193.8
mm/sec (235 fpm), the feedrate was 0.3048 mm/rev (0.012 ipr), and the depth of cut was
2.54 mm (0.100 inch). After each boring operation, the tool wear was measured using a
DoAll toolmaker’s microscope. Average flank wear width was used to indicate the extent
of the tool degradation, with an average flank wear width in excess of .300 mm indicating
a worn out tool. A Scanning Electron Microscope (SEM) photograph of a worn out tool is
shown in Figure1.
4
Figure 1 SEM Photograph of a worn out tool
A Montronix FS13CXPX 3-axis dynamometer, installed within the lathe tool mount,
was used to acquire the three cutting force components: the radial force, the longitudinal
force, and the tangential force. A Montronix TSA203-1 miniature 3-axis charge amplifier
was used for initial amplification of the force signals and a customized module was added
to provide additional amplification and filtering of the signals. The filter was an antialiasing low pass filter with a cutoff frequency of 415 Hz. After being amplified and
filtered, the conditioned signals were digitized and stored on a designated computer for
further computation. The digitization was performed, at a sampling frequency of 1 kHz,
by a data acquisition (DAQ) board within the computer. LabView software was utilized
to facilitate data collection and storage. The schematic diagram of the experimental setup
is shown in Figure 2.
5
Titanium
Part
Dynamometer
Charge
Amplifier
Carbide
Insert
Amplifier
& Filter
Computer
Boring
Bar
Figure 2 Schematic Diagram of Boring Experiment
Data was collected, over the usable range, for ten triangular carbide inserts. The
average flank wear measurements for these inserts were recorded for values up to the
.300 mm usability threshold [23]. Fifty four data sets were acquired in boring
experiments.
6
Chapter 3
FEATURE EXTRACTION AND SELECTION
3.1 Feature Extraction
Since the wear is measured at the end of each boring operation the force data collected
during each sequence is representative of the respective wear measurement. Feature
extraction is the process of extracting usable parameters from this acquired force data to
provide useful inputs to the artificial neural networks [24, 25].
Based on the previous research for the monitoring and diagnosis of cutting tools [7, 8],
fourteen features were extracted from the raw force data. These fourteen features are
listed in Table 1 and are described in the following paragraphs.
Table 1 List of Fourteen Extracted Features
INDEX
NUMBER
1
FEATURE DESCRIPTION
Average of radial force (fx)
2
Average of longitudinal force (fy)
3
Average of tangential force (fz)
4
Average of fz/fx
5
Average of fz/fy
6
Root Mean Square (RMS) of radial force (fx)
7
Root Mean Square (RMS) of longitudinal force (fy)
8
Root Mean Square (RMS) of tangential force (fz)
9
Skewness of radial force (fx)
10
Skewness of longitudinal force (fy)
11
Skewness of tangential force (fz)
12
Kurtosis of radial force (fx)
13
Kurtosis of longitudinal force (fy)
14
Kurtosis of tangential force (fz)
7
The average values of radial, longitudinal, and tangential forces are calculated from
the beginning point to the end point of the boring operation. The values are defined as
follows:
𝑛
1
𝐴𝑣𝑒𝑟𝑎𝑔𝑒 = ∑ 𝑓𝑖
𝑛
(1)
𝑖=1
where n is number of data points and fi is the value of cutting force for the ith data point.
The pair of ratios between two cutting forces are defined as follows:
𝑛
𝑓𝑧 1
𝑓𝑧𝑖
= ∑
𝑓𝑥 𝑛
𝑓𝑥𝑖
(2)
𝑖=1
𝑛
𝑓𝑧 1
𝑓𝑧𝑖
= ∑
𝑓𝑦 𝑛
𝑓𝑦𝑖
(3)
𝑖=1
where (𝑓𝑥𝑖 ), (𝑓𝑦𝑖 ), and (𝑓𝑧𝑖 ) are the forces in the x, y, and z directions, respectively, and n
is the number of data points.
RMS of the radial, longitudinal, and tangential forces are expressed by:
𝑛
1
𝑅𝑀𝑆 = √ ∑ 𝑓𝑖2
𝑛
(4)
𝑖=1
where fi is the value of cutting force on the ith point and n is total number of data points.
Skewness characterizes the degree of asymmetry of a distribution around its mean.
Positive skewness indicates a distribution with an asymmetric tail extending toward more
positive values and negative skewness indicates a distribution with an asymmetric tail
extending toward more negative values. The equation for skewness is defined as:
8
𝑛
3
𝑛
𝑓𝑖 − 𝑓 ̅
𝑠𝑘𝑒𝑤𝑛𝑒𝑠𝑠 =
∑(
)
(𝑛 − 1)(𝑛 − 2)
𝑠
(5)
𝑖=1
where s is the sample standard deviation, n is the total number of data points, f i is the ith
cutting force, and f is the average of the cutting forces.
Kurtosis characterizes the relative flatness of a distribution as compared with the
normal distribution. Positive kurtosis indicates a relatively peaked distribution and
negative kurtosis indicates a relatively flat distribution. Kurtosis is defined as:
𝑛
4
𝑛(𝑛 + 1)
𝑓𝑖 − 𝑓 ̅
3(𝑛 − 1)2
𝑘𝑢𝑟𝑡𝑜𝑠𝑖𝑠 = {
∑(
) }−
(𝑛 − 1)(𝑛 − 2)(𝑛 − 3)
(𝑛 − 2)(𝑛 − 3)
𝑠
(6)
𝑖=1
where s is the sample standard deviation and n is the total number of data, f i is the ith
cutting force, and f is the average of the cutting forces.
3.2 Feature Selection
To determine which of the fourteen features produce high sensitivity to tool wear and
low sensitivity to noise, the following feature selection technique is utilized based on the
Euclidean Distance Measurement and Sequential Forward Search (SFS) Algorithm.
The minimum number of training samples required, for the following feature selection
technique, is defined as:
𝑁 ≥ 2(𝑑 + 1)
where N is the number of training samples and d is the number of features. This is
extremely important in order to conduct the training process adequately [26]. With 14
features being evaluated and 54 sets of training data, this requirement is satisfied.
(7)
9
This technique takes a D dimensional measurement vector and determines the d
features which maximize a criterion representing the signal to noise ratio of these
features. For this process, the concept of the Euclidean Distance Measurement is used.
The formula used in the work for the feature selection is:
𝐽 = 𝑡𝑟𝑎𝑐𝑒(𝑆𝑤 −1 𝑆𝑏 )
(8)
where Sw is the within class scatter matrix, Sb is the between class scatter matrix, J
represents the signal to noise ratio of the feature vectors.
The scatter matrices Sw and Sb are defined as follows:
𝑐
𝑆𝑤 = ∑ 𝑆𝑖
(9)
𝑖=1
𝑛𝑖
𝑆𝑖 = ∑(𝑥𝑖𝑘 − 𝑚𝑖 )(𝑥𝑖𝑘 − 𝑚𝑖 )𝑡
(10)
𝑘=1
𝑛𝑖
1
𝑚𝑖 = ∑ 𝑥𝑖𝑘
𝑛𝑖
(11)
𝑘=1
𝑐
𝑆𝑏 = ∑ 𝑛𝑖 (𝑚𝑖 − 𝑚)(𝑚𝑖 − 𝑚)𝑡
(12)
𝑖=1
c
1
𝑚 = ∑ 𝑛𝑖 𝑚𝑖
𝑐
(13)
i=1
where Si is the within class scatter of the ith class, mi is the mean vector of the ith class
training patterns, ni is the number of training vectors of the ith class, xik is the training
vector of the ith class, c is the number of classes, and m is the total mean vector.
10
The SFS Algorithm is also used as part of the process [27]. This algorithm works for
the feature selection as described in the following:
1. Select the feature from D features in the measurement vector which maximizes
J. Let’s call this feature f1.
2. Pair each of the remaining D-1 features with f1 and calculate J based upon
Equation (8) for all the pairs. The pair which maximizes J is selected.
3. Repeat the previous step until all d features have been selected.
The SFS algorithm cannot guarantee to find the best feature set. However, this
algorithm is efficient and is capable of obtaining the feature sets whose signal to noise
ratio is quite close to the optimum.
The results of the feature selection process for the fourteen selected features are listed
in Table 2. The five feature combinations, which were utilized as inputs to the neural
networks, are those in feature space 1 through 5 from Table 2.
11
Table 2 Outcome of Feature Selection Process
FEATURE
SPACE
1
2
3
4
5
6
7
8
9
10
11
12
13
14
FEATURE(S) SELECTED BY INDEX
NUMBER
#13
#13 #4
#13 #4 #2
#13 #4 #2 #10
#13 #4 #2 #10 #11
#13 #4 #2 #10 #11 #6
#13 #4 #2 #10 #11 #6 #1
#13 #4 #2 #10 #11 #6 #1 #3
#13 #4 #2 #10 #11 #6 #1 #3 #5
#13 #4 #2 #10 #11 #6 #1 #3 #5 #8
#13 #4 #2 #10 #11 #6 #1 #3 #5 #8 #7
#13 #4 #2 #10 #11 #6 #1 #3 #5 #8 #7 #9
#13 #4 #2 #10 #11 #6 #1 #3 #5 #8 #7 #9 #12
#13 #4 #2 #10 #11 #6 #1 #3 #5 #8 #7 #9 #12 #14
12
Chapter 4
ARTIFICIAL INTELLIGENCE TECHNIQUES
4.1 Counterpropagation Neural Networks (CPN’s)
The Counterpropagation Neural Network (CPN) combines features from both the
Kohonen Self Organizing Map network and the Grossberg Outstar network. As such, the
CPN is a member of the mapping network family and is derived from the work of Dr.
Robert Hecht-Nielsen [28]. Features of CPN include the use of the Euclidean distance as
a measure of closeness and a conscience mechanism.
The Euclidean distance is used as a measure of closeness of the weight vector to the
input vector. Furthermore, the limitation of weight vector initialization is addressed by
the conscience mechanism, which ensures that each Kohonen weight is mapped into a
viable vector space.
The basic CPN architecture, shown in Figure 3, is composed of three layers; the input
layer, the Kohonen layer, and the Grossberg layer.
13
Figure 3 2x30x1 CPN Architecture
The input vector is presented to the input layer and then propagated to the Kohonen
layer. The Euclidean distance between the input vector and the Kohonen weight vector is
calculated according to equation (14):
𝑚
𝑑𝑖 = ‖𝐾𝑖 − 𝑋‖ = √∑(𝑘𝑖𝑗 − 𝑥𝑗 )
2
(14)
𝑗=1
where 𝑘𝑖𝑗 is the 𝑗th component of the 𝑖th Kohonen weight vector, 𝑥𝑗 is the 𝑗th component
of the input vector and 𝑚 is the number of neurons in the input layer.
During normal mode, the Kohonen neuron with the least Euclidean distance is
declared the ‘winner’. This neuron then outputs a value of one to the corresponding
neuron of the Grossberg layer. All other Kohonen neurons output a value of zero.
14
1
𝑧𝑖 = {
0
𝑖𝑓 𝑑𝑖 ≤ 𝑑𝑗 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑗 = 1 … 𝑚
𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
(15)
The Grossberg layer then multiples the “winning” Kohonen output by a weighted value.
The Grossberg output is represented by:
𝑦 ′ = 𝐺𝑖 𝑧𝑖
(16)
where 𝑦′ is the output, 𝐺𝑖 is the weight vector of the 𝑖th Grossberg neuron, and 𝑧𝑖 is the
output of the 𝑖th Kohonen neuron.
Training of the CPN is accomplished in a two step process. The first step involves the
manipulation of the Kohonen weight vectors to achieve an equiprobable distribution
relative to the applied training data set. Starting with a random distribution, Kohonen
vectors are compared to test data to determine the neuron with the least distance. Once
the minimum distance is selected, the Kohonen weight vectors are adjusted according to:
𝑘𝑖𝑛𝑒𝑤 = 𝑘𝑖𝑜𝑙𝑑 + 𝛼(𝑥 − 𝑘𝑖𝑜𝑙𝑑 )𝑧𝑖 + 𝛽(𝑥 − 𝑘𝑖𝑜𝑙𝑑 )(1 − 𝑧𝑖 )
(17)
where 𝑘𝑖 is the ith Kohonen weight vector, 𝑧𝑖 is the output of the ith Kohonen neuron, 𝑥
is the input vector, and 𝛼 and 𝛽 are parameters set by the user. The 𝛼 term applies only to
the biased winner and the 𝛽 term applies to all others. The 𝛽 parameter is normally set to
zero, so that only the winning neurons weight is adjusted and all other neuron weights
remain unchanged.
To assure that all Kohonen vectors are viably mapped, the use of a conscience
mechanism is employed. The conscience mechanism adds a bias to the Euclidean
distance calculation.
𝑑′𝑖 = {
𝑑
𝑖𝑓 𝑊𝑖𝑛 𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦𝑖 ≤ 𝑇
𝑑𝑖 − 𝑏𝑖 𝑖𝑓 𝑊𝑖𝑛 𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦𝑖 ≥ 𝑇
(18)
15
where 𝑑′𝑖 is the new distance calculation, while 𝑑 and 𝑇 are parameters set by the user.
These parameters allow neurons that lose frequently to be drawn into regions where
training occurs. The bias value is calculated by:
1
𝑏𝑖 = 𝑐 ( − 𝑝𝑖 )
𝑛
(19)
where 𝑏𝑖 is the bias value, 𝑝𝑖 is the 𝑊𝑖𝑛 𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦, 𝑛 is the number of Kohonen
neurons, and 𝑐 is a parameter set by the user.
After the Kohonen mapping has been completed, the Grossberg training is
accomplished through an iterative process according to:
𝐺𝑖𝑛𝑒𝑤
𝐺𝑖𝑜𝑙𝑑
𝑖𝑓 𝑧𝑖 = 0
= { 𝑜𝑙𝑑
𝑜𝑙𝑑
𝐺𝑖 + 𝑎(𝑦 − 𝐺𝑖 ) 𝑖𝑓 𝑧𝑖 = 1
(20)
where 𝐺𝑖 is the 𝑖th Grossberg weight vector, 𝑎 is a network parameter, 𝑦 is the output
value of the training data, and zi is the output of the respective Kohonen neuron.
4.2 Adaptive Neuro-Fuzzy Inference System (ANFIS)
The Adaptive Nuero-Fuzzy Inference System combines the features of the Fuzzy
Inference System (FIS) with a learning capability similar to that of a neural network. In
this configuration, a Sugeno type FIS network is required with adjustable variables
defining the input and output membership functions.
ANFIS is made up of five layers [29]. They are the input, the Input Membership
Functions (IMF), the rules, the Output Membership Functions (OMF), and the
defuzzification output. An ANFIS structure of 2 inputs and 3 membership functions (2x3)
is shown in Figure 4.
16
Figure 4 2x3 ANFIS Architecture
The functions selected for the input and output membership functions, as well as the
quantity of IMF’s, are user specified parameters. In this research, the IMF’s are defined
as one of seven possible functions. These seven input membership functions are
described in the following.
The generalized bell function (gbell) is represented by
𝑓(𝑥; 𝑎, 𝑏, 𝑐) =
1
(𝑥 − 𝑐) 2𝑏
1+| 𝑎 |
where 𝑎 and 𝑏 vary the width of the curve and the parameter 𝑐 locates the center.
(21)
17
The Gaussian distribution curve (gauss) is represented by
𝑓(𝑥; 𝜎, 𝑐) =
−(𝑥−𝑐)2
𝑒 2𝜎2
(22)
where 𝑐 is the curve mean and 𝜎 is the variance.
The difference between two sigmoidal functions (dsig) is represented by
𝑓(𝑥; 𝑎1 , 𝑎2 , 𝑐1 , 𝑐2 ) =
1
1+
𝑒 −𝑎1 (𝑥−𝑐1 )
−
1
1+
𝑒 −𝑎2 (𝑥−𝑐2 )
(23)
where 𝑎1 and 𝑎2 control the slopes and 𝑐1 and 𝑐2 control the points of inflection of curves
1 and 2.
The product of two sigmoidal functions (psig) is represented by
1
1
𝑓(𝑥; 𝑎1 , 𝑎2 , 𝑐1 , 𝑐2 ) = (
)
(
)
1 + 𝑒 −𝑎1 (𝑥−𝑐1 ) 1 + 𝑒 −𝑎2 (𝑥−𝑐2 )
(24)
where 𝑎1 and 𝑎2 control the slopes and 𝑐1 and 𝑐2 control the points of inflection of curves
1 and 2.
The trapezoidal function (trap) is represented by
𝑥−𝑎 𝑑−𝑥
𝑓(𝑥; 𝑎, 𝑏, 𝑐, 𝑑) = max (𝑚𝑖𝑛 (
, 1,
) , 0)
𝑏−𝑎
𝑑−𝑐
where 𝑎 and 𝑑 control the base points of the trapezoid and 𝑏 and 𝑐 control the top
corners.
(25)
18
The pi curve (pi) is represented by
0,
𝑥−𝑎 2
2(
) ,
𝑏−𝑎
𝑥−𝑏 2
1 − 2(
) ,
𝑏−𝑎
1,
𝑓(𝑥; 𝑎, 𝑏, 𝑐, 𝑑) =
𝑥−𝑐 2
1 − 2(
) ,
𝑑−𝑐
𝑥−𝑑 2
2(
) ,
𝑑−𝑐
{
0,
𝑥≤𝑎
𝑎≤𝑥≤
𝑎+𝑏
2
𝑎+𝑏
≤𝑥≤𝑏
2
𝑏≤𝑥≤𝑐
𝑐+𝑑
𝑐≤𝑥≤
2
𝑐+𝑑
≤𝑥≤𝑑
2
}
𝑥≥𝑑
(26)
where 𝑎 and 𝑑 control the base points of the pi curve and 𝑏 and 𝑐 control the top corners.
The triangular function (tri) is represented by
𝑓(𝑥; 𝑎, 𝑏, 𝑐) = max (𝑚𝑖𝑛 (
𝑥−𝑎 𝑐−𝑥
,
) , 0)
𝑏−𝑎 𝑐−𝑏
(27)
where 𝑎 and 𝑐 set the left and right base points of the triangle and 𝑏 sets the location of
the triangle peak.
During data evaluation, the input layer accepts the input vector and routes the
individual data components to the respective IMF. Using this input data, the Input
Membership Functions calculate an output value 𝑓𝑖𝑗 (𝑥𝑖 ). Where 𝑥𝑖 is the 𝑖th component
of the input vector and 𝑓𝑖𝑗 is the 𝑗th IMF associated with the 𝑖th component of the input
vector.
Within the rules layer, the 𝑓𝑖𝑗 (𝑥𝑖 ) values are combined into fuzzy logic statements.
The format of these rules is expressed by:
𝑅𝑘 = 𝑓1𝑗 (𝑥1 ) AND 𝑓2𝑗 (𝑥2 ) AND … AND 𝑓𝑖𝑗 (𝑥𝑖 )
(28)
where 𝑅𝑘 is the 𝑘th rule output, 𝑖 is the input neuron, and 𝑗 is an IMF associated with the
19
respective input, such that all unique rule combinations are created. The ‘AND’ operator
in this equation performs either a multiplicative function if a probabilistic ‘AND’ is
selected, as shown in equation (29), or a minimum function if a fuzzy logic ‘AND’ is
selected as shown in equation (30).
𝑅𝑘 = 𝑓1𝑗 (𝑥1 ) ∗ 𝑓2𝑗 (𝑥2 ) ∗ … ∗ 𝑓𝑖𝑗 (𝑥𝑖 )
(29)
𝑅𝑘 = min{𝑓1𝑗 (𝑥1 ), 𝑓2𝑗 (𝑥2 ), … , 𝑓𝑖𝑗 (𝑥𝑖 )}
(30)
For example, the third rule, in Figure 4, would be defined as:
𝑅3 = 𝑓11 (𝑥1 ) 𝑎𝑛𝑑 𝑓23 (𝑥2 )
(31)
The number of these rules is defined by the input neurons 𝑖 and the IMF 𝑗 according to
the relationship:
Number of rules = 𝑗 𝑖
(32)
There are 9 (32) rules in the case of a 2 dimensional input vector with 3 IMF’s per input.
A constant (zeroth order) equation is used for the Output Measurement Function so
that the output of the OMF layer is a weighted value of 𝑅𝑘 .
𝐺𝑘 = 𝑑𝑘 ∗ 𝑅𝑘
(33)
where 𝑘 specifies the corresponding rule number and 𝑑𝑘 is the adjustable parameter.
The Defuzzification Output layer calculates a weighted average for Rule outputs as
follows:
𝑗𝑖
𝑗𝑖
𝑜𝑢𝑡𝑝𝑢𝑡 = ∑ 𝐺𝑘 ÷ ∑ 𝑅(𝑘)
𝑘=1
(34)
𝑘=1
where the letter 𝑘 specifies the corresponding rule number and 𝑗 𝑖 is the number of rules.
20
The learning algorithm used for the training process is a hybrid type and is based on
the output error. This algorithm is a combination of both the backpropagation and least
square methods. The IMF variables and the OMF constants are the parameters
manipulated during training. The total number of IMF variables for the network is equal
to:
𝐼𝑀𝐹𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 = 𝑎 ∗ 𝑏 ∗ 𝑐
(35)
where 𝑎 is the number of input neurons, 𝑏 is the number of IMF’s per input neuron, and 𝑐
is the number of variables required to define the IMF function. For a 2x3 (gbell) ANFIS,
there are 18 (2*3*3) IMF parameters. The number of OMF adjustable parameters for the
network is equal to the number of rules as defined in equation (32). So, for the same 2x3
(gbell) ANFIS, there are 9 (32) OMF adjustable parameters. Therefore, this ANFIS
structure has 27 parameters being manipulated during the training process. Upon
completion of training, the ANFIS represents a nonlinear function to correlate the input
and output relationship of the training data.
21
Chapter 5
ON-LINE MONITORING AND MEASUREMENTS OF TOOL WEAR FOR
TITANIUM BORING
5.1 Training
For the purpose of network training, the features extracted from the experimental data
were used as listed in Table 3.
Table 3 Experimental Data Used in the Training Process
Kurtosis
fy
-0.5527
-0.1554
1.3176
0.4589
-0.7969
-0.2318
-0.1866
-0.1652
-0.4091
-0.5129
0.0227
0.3484
-0.2635
0.3225
0.0014
-0.1244
0.1144
-0.1382
0.1271
-0.2721
-0.2326
0.2729
Avg.
fz/fx
1.1667
1.1696
1.1458
1.1290
1.0780
1.1670
1.1722
1.1952
1.1692
1.1789
1.1677
1.1878
1.1600
1.1686
1.1405
1.1691
1.1404
1.1511
1.1312
1.1153
1.0854
1.1456
Skew
Avg. fy
fy
1.3411 -0.2674
1.3916 -0.3626
1.3109 -0.3186
1.4746 -0.2095
1.2420 -0.0254
1.0706 0.2510
1.0823 0.0418
1.2402 -0.0192
1.2538 -0.0015
1.2332 -0.0062
1.3849 0.0764
1.5163 -0.5647
0.6323 0.1549
1.1228 -0.0724
1.0235 -0.1959
1.2914 0.2098
1.2065 0.3639
0.9970 0.1179
1.0027 0.1047
1.0370 -0.2267
1.0848 -0.1334
1.0837 0.1999
Skew
fz
-0.1056
-0.0382
-0.0515
-0.0468
-0.1853
0.2398
0.1234
0.0582
0.0817
0.0459
-0.3217
-1.0211
0.1528
0.0036
-0.1428
-0.3085
-0.2596
-0.0508
0.0608
-0.5323
0.0069
-0.1265
Ave.
flank
Wear
(mm)
0.140
0.184
0.239
0.260
*
0.160
0.158
0.214
0.222
0.277
*
0.177
0.221
0.198
0.230
0.263
*
0.140
0.143
0.167
0.218
0.215
Classification
Value (mm)
0.3
0.3
0.3
0.3
0.7
0.3
0.3
0.3
0.3
0.3
0.7
0.3
0.3
0.3
0.3
0.3
0.7
0.3
0.3
0.3
0.3
0.3
22
0.6714 1.1172 0.9179 0.0092 -0.2476
*
0.7
0.8106 1.1354 1.6056 0.0467 -0.3753 0.257
0.3
-0.6812 1.1207 0.6991 0.1423 -0.2232 0.281
0.3
-0.4564 1.1840 1.1497 0.5766 0.0003 0.267
0.3
0.5706 1.0396 0.9444 0.9037 0.5452 0.297
0.3
1.0470 1.1000 0.9889 0.8387 0.3409
*
0.7
0.2391 1.1202 1.2456 0.5397 0.3005 0.139
0.3
-0.5313 1.1330 1.2451 0.0560 0.2668 0.179
0.3
-0.3152 1.1606 1.4728 0.0231 0.2941 0.241
0.3
-0.4642 1.1538 1.2302 -0.1282 -0.3870 0.234
0.3
-0.5572 1.1684 1.8335 -0.1105 0.1981
*
0.7
0.2879 1.1028 1.1140 0.2684 -0.2958 0.120
0.3
-0.7193 1.1010 0.8913 -0.2069 0.0097 0.139
0.3
-0.2432 1.0870 1.1223 -0.3812 -0.3799 0.210
0.3
-0.2455 1.1638 1.3895 -0.4210 -0.2757 0.270
0.3
0.8384 1.1531 1.3154 0.2214 0.1490
*
0.7
0.4673 1.1498 1.1899 0.0044 0.1037 0.162
0.3
-0.1446 1.1866 1.6724 -0.2450 -0.0264 0.205
0.3
1.5720 1.1865 1.7225 -0.0893 0.1383 0.269
0.3
1.7118 1.2077 1.9929 -0.3451 -0.1431 0.296
0.3
-0.0504 1.1914 1.9230 0.3805 0.0582
*
0.7
0.2050 1.1164 1.3695 0.8558 -0.5229 0.140
0.3
-0.2534 1.1543 1.5688 0.2270 -0.5781 0.232
0.3
0.6315 1.1785 1.8812 0.7212 -0.0927 0.266
0.3
-0.4312 1.1694 1.8134 -0.0492 -0.4826 0.298
0.3
1.2758 1.0252 1.1680 2.5997 -1.1414
*
0.7
* Average flank wear width could not be measured practically due to
severe tool failure. Therefore, these worn-out tools were not used for the
training for on-line measurements.
There were 48 training data sets for on-line classification and 39 training data sets for
on-line measurement.
Linear Interpolation was implemented by adding three linearly interpolated sets of
data between each of the original data sets, with the exception that no interpolation data
23
was added between the last and next to last data sets so as to increase the number of data
in the training process. This resulted in a total of 138 training sets for on-line
classifications and 129 training sets for on-line measurements.
For the training of on-line classification, the tool conditions were classified into two
categories: useable and worn out tools. Outputs for usable tools were given a value of 0.3
while outputs for worn out tools were given a value of 0.7.
For the CPN, the input data was normalized to a value between 0.1 and 0.9 using:
𝑎𝑖 =
0.8
(𝑟 − 𝑟𝑚𝑖𝑛 ) + 0.1
𝑟𝑚𝑎𝑥 − 𝑟𝑚𝑖𝑛 𝑖
(36)
where 𝑎𝑖 is the normalized data, rmax and rmin are the maximum and minimum value of the
raw data respectively, and 𝑟𝑖 is the 𝑖th raw input data.
Seventy CPN network structures were used for both on-line classification and on-line
measurement. The inputs ranged from 1 to 5 while the values for Kohonen neurons were
between 30 and 69 with a step size of three. The training process composed of 1000
iterations for the training of the Kohonen section and an additional 1000 iterations for the
training of the Grossberg section.
For ANFIS, two hundred and sixty-six different network structures were used for both
on-line classification and on-line measurement. These were composed of seven different
IMF’s, nineteen different structures, and two different Rule ‘AND’ functions. ANFIS
utilized the hybrid training method, a constant OMF, and 50 training epochs.
24
5.2 On-Line Classification
On-line classification of the boring insert is a very important part of the TCM process.
Improper tool classification can result in either a tool being replaced before necessary,
resulting in additional tool cost, or the continual use of a worn out tool with part quality
degradation and the added risk of tool breakage. Therefore, to be viable, a network must
be able to predict tool classification with 100% accuracy.
Immediately following network training, on-line classification was performed with
independent data sets to determine which networks demonstrated the on-line ability to
classify tools as either usable or worn out with 100% accuracy. Table 4 lists the
independent data sets.
Table 4 Data for On-line Classification and Measurements
Ave.
flank
Kurtosis Avg.
Avg.
Skew
Skew
Wear Classification
fy
fz/fx
fy
fy
fz
(mm)
Value (mm)
-0.7955 1.1691 1.2869 0.3805 0.0582 0.159
≤ 0.5
-0.4954 1.2185 1.8413 0.8558 -0.5229 0.244
≤ 0.5
1.0545 1.1697 1.4311 0.2270 -0.5781 0.242
≤ 0.5
-0.5058 1.0870 1.0247 0.7212 -0.0927 0.219
≤ 0.5
-0.2675 1.0906 1.1096 -0.0492 -0.4826 0.231
≤ 0.5
7.5529 1.0782 1.1618 2.5997 -1.1414
*
> 0.5
* Average flank wear width could not be measured practically due to
severe tool failure. Therefore, these worn-out tools were not used for the
training for on-line measurements.
For the evaluation of on-line classification results, network output values less than or
equal to 0.5 designate a usable tool while values exceeding 0.5 are classified as unusable.
Table 5 contains the on-line classification performance for the CPN network structures
25
while Tables 6 and 7 display the on-line classification performance for the ANFIS
network structures.
Number of Kohonen Neurons
Table 5 CPN On-line Classification Success Rate
30
33
36
39
42
45
48
51
54
57
60
63
66
69
1
66.67%
83.33%
66.67%
66.67%
83.33%
66.67%
66.67%
66.67%
66.67%
66.67%
66.67%
66.67%
66.67%
66.67%
Number of Inputs
2
3
4
100.00% 100.00% 83.33%
100.00% 100.00% 100.00%
100.00% 100.00% 100.00%
100.00% 100.00% 100.00%
100.00% 83.33% 100.00%
100.00% 100.00% 100.00%
100.00% 100.00% 83.33%
100.00% 50.00% 100.00%
100.00% 66.67% 100.00%
100.00% 83.33% 100.00%
100.00% 83.33% 100.00%
83.33% 100.00% 100.00%
100.00% 100.00% 83.33%
100.00% 100.00% 100.00%
5
83.33%
83.33%
100.00%
83.33%
83.33%
83.33%
83.33%
100.00%
100.00%
83.33%
83.33%
100.00%
83.33%
100.00%
26
Table 6 ANFIS On-line Classification Success Rate
(ANFIS with probabilistic ‘AND’ used in all the fuzzy rules)
(Number of Inputs) x (Number of IMF's)
Type of Input Measurement Function
tri
trap
gbell
gauss
dsig
psig
pi
1x4 83.33% 83.33% 66.67% 66.67% 66.67% 66.67% 83.33%
1x5 66.67% 83.33% 66.67% 66.67% 66.67% 66.67% 83.33%
1x6 66.67% 66.67% 66.67% 66.67% 83.33% 83.33% 83.33%
1x7 66.67% 66.67% 66.67% 66.67% 83.33% 83.33% 66.67%
1x8 66.67%
*
83.33% 83.33% 66.67% 66.67% 66.67%
1x9 66.67% 66.67% 66.67% 66.67% 66.67% 66.67% 66.67%
1x10 66.67% 66.67% 66.67% 66.67% 66.67% 66.67% 66.67%
1x11 66.67% 66.67% 66.67% 66.67% 66.67% 66.67% 66.67%
1x12 66.67% 66.67% 66.67% 50.00% 66.67% 66.67% 66.67%
2x2 83.33% 83.33% 100.00% 100.00% 100.00% 100.00% 83.33%
2x3
*
*
100.00% 100.00% 83.33% 100.00% 83.33%
2x4 83.33% 83.33% 100.00% 83.33% 83.33% 83.33% 83.33%
2x5
*
*
83.33% 66.67% 66.67% 66.67% 83.33%
2x6
*
*
66.67% 83.33% 66.67% 66.67% 83.33%
3x2 66.67% 83.33% 100.00% 83.33% 100.00% 100.00% 83.33%
3x3 66.67%
*
83.33% 100.00% 66.67% 66.67% 66.67%
3x4 66.67%
*
100.00% 66.67% 83.33% 100.00% 83.33%
4x2
*
83.33% 83.33% 66.67% 100.00% 100.00% 83.33%
5x2 33.33% 83.33% 66.67% 66.67% 100.00% 100.00% 83.33%
* Invalid function configurations occur which made the training process infeasible.
27
(Number of Inputs) x (Number of IMF's)
Table 7 ANFIS On-line Classification Success Rate
(ANFIS with fuzzy ‘AND’ used in all the fuzzy rules)
tri
1x4 83.33%
1x5 83.33%
1x6 83.33%
1x7 83.33%
1x8 83.33%
1x9 83.33%
1x10 66.67%
1x11 66.67%
1x12 66.67%
2x2 83.33%
2x3 83.33%
2x4 83.33%
2x5 83.33%
2x6 66.67%
3x2 66.67%
3x3 66.67%
3x4 83.33%
4x2 66.67%
5x2 83.33%
Type of Input Measurement Function
trap
gbell
gauss
dsig
83.33% 83.33% 83.33% 83.33%
83.33% 83.33% 83.33% 83.33%
83.33% 83.33% 83.33% 83.33%
83.33% 83.33% 83.33% 83.33%
83.33% 83.33% 83.33% 83.33%
83.33% 83.33% 83.33% 83.33%
83.33% 83.33% 83.33% 83.33%
83.33% 83.33% 83.33% 83.33%
83.33% 66.67% 66.67% 83.33%
83.33% 83.33% 83.33% 83.33%
83.33% 100.00% 83.33% 83.33%
83.33% 83.33% 83.33% 83.33%
83.33% 66.67% 83.33% 83.33%
66.67% 66.67% 50.00% 66.67%
83.33% 83.33% 83.33% 83.33%
66.67% 100.00% 66.67% 66.67%
83.33% 100.00% 83.33% 66.67%
83.33% 83.33% 83.33% 100.00%
83.33% 83.33% 83.33% 83.33%
psig
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
100.00%
83.33%
83.33%
66.67%
83.33%
83.33%
83.33%
100.00%
83.33%
pi
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
66.67%
66.67%
83.33%
83.33%
83.33%
83.33%
83.33%
Several network structures, for both CPN and ANFIS, were able to classify tools with
100% accuracy thus demonstrating the viability of using both CPN and ANFIS networks
for the purpose of on-line tool classification.
Choosing the least complex network structure simplifies the on-line process through
the minimization of computational requirements. Both the CPN and ANFIS structures
were able to predict a worn out tool with 100% accuracy using a minimum of two inputs
28
(kurtosis of longitudinal force, average of the ratio between tangential force and radial
force). The least complex structure, within each network, was the 2x30x1 CPN and the
2x2 (gauss) ANFIS. The complete set of data for these two structures is shown in Table
8.
Table 8 Actual Outputs for the Best On-line Classification Structures
Classification
Criteria
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
> 0.5
2x30x1 CPN
2x2 (gauss) ANFIS
On-line
Classification
0.4511
0.4275
0.4770
0.3386
0.3000
0.5314
On-line
Classification
0.3162
0.3164
0.3205
0.2885
0.3125
0.6721
The best results for the ANFIS networks were produced by the ‘gbell’, ‘gauss’, ‘dsig’,
and ‘psig’ IMFs. These structures were able to handle the complex nature of the training
data and deliver satisfactory results. However, the ‘trap’, ‘pi’, and ‘tri’ IMFs were unable
to predict a worn out tool with a 100% success rate. Some of these also encountered
invalid function configurations during the training process and hence the training process
became infeasible as shown in Table 6.
The CPN networks did not encounter any training difficulties. However, with
equivalent results, the ANFIS would be the preferred approach due to its superb
performance in regard to on-line measurements. Other desirable features of the ANFIS
network are that the input data does not require normalization and since there are fewer
29
training iterations (50 epochs as opposed to 2000 iterations required for CPN training), a
significantly shorter training time.
5.3 On-Line Measurement
On-line measurements provide useful information about tool status. The better the online estimation of the average flank wear width the more precisely the tool degradation
can be predicted. In this work, the average of the absolute output errors from the on-line
measurement results are used to gage the capability of a network to predict the average
flank wear width so as to forecast the cutting tool degradation.
Immediately following network training, on-line measurements were performed with
other independent data sets. Table 9 contains the on-line measurement performance for
the CPN network structures while Tables 10 and 11 display the on-line measurement
results for the ANFIS approach using probabilistic and fuzzy logic ‘AND’ operations.
30
Number of Kohonen Neurons
Table 9 CPN On-line Measurement Average Output Error
30
33
36
39
42
45
48
51
54
57
60
63
66
69
1
13.82%
12.96%
13.82%
13.39%
19.23%
12.37%
12.95%
17.37%
12.29%
12.29%
13.24%
17.12%
14.42%
13.40%
Number of Inputs
2
3
4
16.38% 27.30% 24.29%
17.01% 20.96% 26.74%
17.01% 18.04% 28.66%
17.44% 12.33% 26.72%
11.78% 14.85% 27.96%
14.72% 18.23% 27.70%
14.41% 12.82% 28.39%
13.28% 11.00% 28.39%
14.73% 18.63% 31.39%
12.58% 12.28% 28.39%
14.36% 11.14% 27.76%
12.92% 12.28% 32.13%
14.36% 19.19% 31.89%
12.94% 9.93% 29.81%
5
31.58%
27.82%
30.02%
25.31%
29.28%
23.60%
31.33%
23.44%
29.28%
28.65%
28.65%
23.78%
28.96%
28.55%
31
Table 10 ANFIS On-line Measurement Average Output Error
(ANFIS with probabilistic ‘AND’ used in all the fuzzy rules)
(Number of Inputs) x (Number of IMF's)
Type of Input Measurement Function
tri
trap
gbell
gauss
dsig
psig
pi
1x4 13.49% 12.46% 12.47% 11.80% 11.97% 11.96% 11.60%
1x5 13.29% 12.62% 12.03% 11.84% 11.52% 11.44% 11.74%
1x6 12.81% 12.74% 11.67% 11.55% 11.39% 11.39% 12.05%
1x7 12.42% 12.19% 10.48% 12.07% 12.40% 12.40% 11.71%
1x8
9.73% 10.57% 11.03% 11.04% 10.44% 10.44% 9.08%
1x9 10.14% 10.09% 10.84% 10.62% 10.20% 10.20% 9.27%
1x10 10.17% 11.37% 10.52% 10.62% 10.79% 10.81% 11.11%
1x11 9.98% 10.72% 5.73% 11.43% 15.99% 15.99% 12.70%
1x12 9.98% 11.09% 9.01% 11.44% 8.65%
8.70% 14.70%
2x2 12.20% 18.57% 21.43% 17.32% 18.15% 18.15% 18.28%
2x3 19.28%
*
15.69% 20.26% 24.18% 24.18% 13.13%
2x4 20.86% 22.10% 23.15% 20.93% 25.83% 25.83% 25.02%
2x5
*
*
26.60% 15.58% 20.85% 20.89% 31.43%
2x6 51.17%
*
51.67% 64.26% 156.27% 209.00% 14.70%
3x2 26.35% 16.98% 19.81% 22.32% 16.35% 16.35% 16.41%
3x3 19.36% 20.10% 55.48% 31.35% 24.52% 24.53% 23.30%
3x4 51.31%
*
35.02% 35.18% 98.07% 98.06% 25.95%
4x2 59.60% 29.51% 29.87% 27.23% 37.99% 37.99% 37.72%
5x2 137.23% 70.91% 130.68% 87.68% 73.20% 73.20% 74.72%
* Invalid function configurations occur which made the training process infeasible.
32
(Number of Inputs) x (Number of IMF's)
Table 11 ANFIS On-line Measurement Average Output Error
(ANFIS with fuzzy ‘AND’ used in all the fuzzy rules)
Type of Input Measurement Function
tri
trap
gbell
gauss
dsig
psig
1x4 13.14% 12.73% 13.09% 12.79% 12.60% 12.60%
1x5 12.52% 12.90% 12.76% 13.35% 11.71% 11.91%
1x6 11.93% 12.64% 11.41% 11.02% 11.40% 11.41%
1x7 12.24% 12.76% 11.37% 12.64% 11.35% 11.45%
1x8 12.48% 11.70% 12.50% 12.29% 11.20% 11.10%
1x9 10.49% 11.08% 10.93% 10.87% 10.95% 10.95%
1x10 10.34% 12.11% 11.06% 4.07%
11.28% 11.28%
1x11 10.71% 12.66% 5.62%
7.21%
11.31% 11.22%
1x12 10.58% 11.93% 11.69% 5.53%
7.82%
6.03%
2x2 16.16% 19.09% 26.64% 24.73% 18.87% 18.87%
2x3 20.84% 14.70% 16.32% 15.84% 15.09% 15.33%
2x4 19.23% 18.14% 18.47% 22.33% 18.14% 18.14%
2x5 26.45% 36.07% 27.16% 23.17% 19.81% 19.46%
2x6 36.05% 54.55% 43.62% 21.76% 35.15% 35.15%
3x2 18.75% 18.85% 15.26% 17.17% 16.80% 16.80%
3x3 22.23% 54.54% 28.40% 20.44% 28.86% 28.86%
3x4 136.09% 31.57% 67.35% 244.06% 51.33% 51.33%
4x2 34.89% 25.85% 35.86% 36.76% 35.25% 35.25%
5x2 34.57% 41.13% 66.21% 44.90% 236.35% 236.35%
pi
12.68%
12.77%
12.55%
12.78%
11.48%
11.11%
12.41%
12.64%
12.36%
18.86%
13.48%
21.04%
39.05%
82.93%
16.37%
23.69%
92.37%
35.94%
36.72%
The best performing structures, within each network type, were the 3x69x1 CPN and
the 1x10 (gauss) ANFIS using a fuzzy logic ‘AND’ operator. The average output errors
for these networks were 9.93% and 4.07% respectively. The maximum and minimum
errors for the 3x69x1 CPN network were 13.43% and 8.46% while the maximum and
minimum errors for the 1x10 (gauss) ANFIS network were 11.15% and 0.87%. The
complete set of output data for these two structures is shown in Table 12 and Figure 5.
33
Table 12 Best Results for On-line Measurements
CPN 3x69x1
Measured
Flank
Wear
(mm)
0.159
0.244
0.242
0.219
0.231
On-line
Flank
Wear
(mm)
0.146
0.221
0.218
0.200
0.200
ANFIS 1x10 (gauss)
Percent
Error
-8.46%
-9.35%
-9.73%
-8.69%
-13.43%
On-line
Flank
Wear
(mm)
0.1571
0.2168
0.2441
0.2168
0.2168
Percent
Error
-1.19%
-11.15%
0.87%
-1.00%
-6.15%
Average Flank Wear Width (mm)
0.3
0.25
0.2
Measured Wear
0.15
3x69x1 CPN
0.1
1x10 (gauss) ANFIS
0.05
0
0
1
2
3
4
5
6
Numberof Boring Operations
Figure 5 Comparison of Actual Tool Wear and On-line Measurements
These results clearly demonstrate the viability of using both CPN and ANFIS for the
purpose of on-line tool measurements.
34
Just like in the on-line classification section, not all ANFIS structures were able to
handle the complex nature of the training data. Some of the ‘trap’ and ‘tri’ IMF structures
encountered training difficulties due to invalid function configurations in the training
process as shown in Table 10. Still the preferred network would be the 1x10 (gauss)
ANFIS with fuzzy logic ‘AND’ operation as this network requires only a single input,
has low average error, and a minimum error of only 0.87%.
The 1x10 (gauss) ANFIS on-line measurements can be used for the prediction of tool
degradation up to the point where the average flank wear width approaches the 0.3 mm
threshold with the 2x2 (gauss) ANFIS being invoked at this point as a very precise
indication of the need for tool replacement.
35
Chapter 6
CONCLUSIONS
Based upon the above descriptions the following conclusions can be drawn:
1. Fourteen features can be extracted from the three components of cutting forces.
These fourteen features are: averages of radial force, tangential force and
longitudinal force, average of the ratio between tangential and radial forces,
average of the ratio between longitudinal and radial forces, root mean square of
radial, longitudinal, and tangential forces, skewness and kurtosis of tangential,
longitudinal, and radial forces.
2. Use of a Sequential Forward Search algorithm, to both select the best
combination and minimize the number of features used, is beneficial in
performing on-line monitoring of boring tools effectively and economically.
3. Both CPN and ANFIS networks can be trained, using features extracted from
cutting force measurements, to successfully perform on-line classification of
boring tools. Multiple structures were able to classify tool conditions with
100% accuracy, with minimal network complexities being the 2x30x1 CPN
and the 2x2 (gauss) ANFIS.
4. Both CPN and ANFIS networks can be trained, using features extracted from
cutting force measurements, to accurately measure tool wear. Excellent results
were achieved using a 1x10 (gauss) ANFIS structure with a minimum error of
0.87% while reasonably good results were obtained using a 3x69x1 CPN
structure with a minimum error of 8.46%.
36
5. Using ANFIS networks and only two features, a system may be created that
both improves part quality and avoids excessive tool wear or breakage. A 1x10
ANFIS can provide continuous monitoring of cutting tool degradation while a
2x2 ANFIS for on-line classification facilitates worn out or broken tool
replacement and enhances quality and safety of the boring process.
37
APPENDIX A
Extracted Features and Average Flank Wear Measurements
38
Table A1: Extracted Features and Average Flank Wear Measurements
Insert 1
Avg. fx
Avg. fy
Avg. fz
RMS fx
RMS fy
RMS fz
Avg. fz/fx
Avg. fz/fy
Skew fx
Skew fy
Skew fz
Kurtosis fx
Kurtosis fy
Kurtosis fz
Std Dev fx
Std Dev fy
Std Dev fz
Wear (mm)
1
3.4719
1.3411
4.0506
3.4725
1.3426
4.0513
1.1667
3.0255
-0.2845
-0.2674
-0.1056
0.5675
-0.5527
0.1248
0.0631
0.0625
0.0796
0.140
Run Number
2
3
4
3.2790 3.1748 3.6788
1.3916 1.3109 1.4746
3.8348 3.6375 4.1520
3.2794 3.1753 3.6794
1.3926 1.3116 1.4753
3.8354 3.6381 4.1542
1.1696 1.1458 1.1290
2.7589 2.7774 2.8181
-0.1271 -0.2734 -0.7391
-0.3626 -0.3186 -0.2095
-0.0382 -0.0515 -0.0468
0.4717 0.9468 3.4717
-0.1554 1.3176 0.4589
0.1393 0.4903 -0.6979
0.0524 0.0545 0.0675
0.0515 0.0420 0.0480
0.0670 0.0690 0.1371
0.184
0.239
0.260
5
3.6382
1.2420
3.9199
3.6389
1.2440
3.9252
1.0780
3.1584
-0.5875
-0.0254
-0.1853
2.5618
-0.7969
-1.2718
0.0718
0.0703
0.2034
> 0.300
39
Table A2: Extracted Features and Average Flank Wear Measurements
Insert 2
Avg. fx
Avg. fy
Avg. fz
RMS fx
RMS fy
RMS fz
Avg. fz/fx
Avg. fz/fy
Skew fx
Skew fy
Skew fz
Kurtosis fx
Kurtosis fy
Kurtosis fz
Std Dev fx
Std Dev fy
Std Dev fz
Wear (mm)
1
3.1458
1.0706
3.6700
3.1472
1.0718
3.6719
1.1670
3.4315
0.0583
0.2510
0.2398
-0.0815
-0.2318
-0.4211
0.0938
0.0500
0.1201
0.160
2
3.0402
1.0823
3.5633
3.0409
1.0836
3.5640
1.1722
3.2988
-0.1031
0.0418
0.1234
0.2312
-0.1866
0.1686
0.0627
0.0531
0.0703
0.158
Run Number
3
4
3.0603 3.1068
1.2402 1.2538
3.6570 3.6319
3.0608 3.1074
1.2414 1.2557
3.6576 3.6325
1.1952 1.1692
2.9534 2.9045
-0.0170 0.0102
-0.0192 -0.0015
0.0582 0.0817
0.3564 -0.0458
-0.1652 -0.4091
0.3206 0.0324
0.0594 0.0621
0.0548 0.0682
0.0682 0.0693
0.214
0.222
5
3.1308
1.2332
3.6899
3.1318
1.2352
3.6910
1.1789
3.0001
-0.0727
-0.0062
0.0459
-0.3027
-0.5129
-0.1646
0.0787
0.0712
0.0874
0.277
6
3.2914
1.3849
3.8423
3.2921
1.3863
3.8433
1.1677
2.7791
-0.1369
0.0764
-0.3217
0.1622
0.0227
0.3227
0.0676
0.0623
0.0891
> 0.300
40
Table A3: Extracted Features and Average Flank Wear Measurements
Insert 3
Avg. fx
Avg. fy
Avg. fz
RMS fx
RMS fy
RMS fz
Avg. fz/fx
Avg. fz/fy
Skew fx
Skew fy
Skew fz
Kurtosis fx
Kurtosis fy
Kurtosis fz
Std Dev fx
Std Dev fy
Std Dev fz
Wear (mm)
1
3.7181
1.5163
4.4165
3.7206
1.5202
4.4203
1.1878
2.9206
-1.2874
-0.5647
-1.0211
3.2790
0.3484
2.2039
0.1354
0.1086
0.1837
0.177
2
2.5518
0.6323
2.9595
2.5525
0.6348
2.9603
1.1600
4.7162
-0.0301
0.1549
0.1528
0.0700
-0.2635
0.2442
0.0608
0.0564
0.0685
0.221
Run Number
3
4
2.6123 2.7419
1.1228 1.0235
3.0251 3.1266
2.6130 2.7426
1.2141 1.0253
3.0530 3.1275
1.1686 1.1405
2.7237 3.0644
-0.0561 -0.2848
-0.0724 -0.1959
0.0036 -0.1428
0.2538 1.3127
0.3225 0.0014
0.3030 0.4777
0.0613 0.0607
0.0536 0.0598
0.0714 0.0742
0.198
0.230
5
3.1573
1.2914
3.6895
3.1586
1.2949
3.6907
1.1691
2.8710
-0.2725
0.2098
-0.3085
0.7464
-0.1244
0.6115
0.0920
0.0949
0.0940
0.263
6
3.2489
1.2065
3.7035
3.2499
1.2102
3.7058
1.1404
3.0832
0.0622
0.3639
-0.2596
0.3476
0.1144
-0.5860
0.0810
0.0958
0.1295
> 0.300
41
Table A4: Extracted Features and Average Flank Wear Measurements
Insert 4
Avg. fx
Avg. fy
Avg. fz
RMS fx
RMS fy
RMS fz
Avg. fz/fx
Avg. fz/fy
Skew fx
Skew fy
Skew fz
Kurtosis fx
Kurtosis fy
Kurtosis fz
Std Dev fx
Std Dev fy
Std Dev fz
Wear (mm)
1
3.2511
0.9970
3.7419
3.2521
0.9980
3.7429
1.1511
3.7578
-0.0618
0.1179
-0.0508
0.3365
-0.1382
0.4987
0.0775
0.0431
0.0886
0.140
2
3.1658
1.0027
3.5809
3.1665
1.0035
3.5818
1.1312
3.5758
-0.0428
0.1047
0.0608
0.1128
0.1271
0.1733
0.0656
0.0402
0.0790
0.143
Run Number
3
4
3.2416 3.3441
1.0370 1.0848
3.6160 3.6278
3.2424 3.3450
1.0392 1.0863
3.6217 3.6321
1.1153 1.0854
3.4919 3.3490
-0.4103 0.0952
-0.2267 -0.1334
-0.5323 0.0069
0.9043 0.5301
-0.2721 -0.2326
-0.7196 -0.9697
0.0715 0.0739
0.0680 0.0583
0.2023 0.1763
0.167
0.218
5
3.4903
1.0837
3.9984
3.4921
1.0867
4.0010
1.1456
3.7052
-0.4096
0.1999
-0.1265
0.8502
0.2729
0.2342
0.1106
0.0808
0.1436
0.215
6
3.5584
0.9179
3.9754
3.5595
0.9197
3.9771
1.1172
4.3451
-0.3363
0.0092
-0.2476
1.1006
0.6714
0.9408
0.0860
0.0577
0.1155
> 0.300
42
Table A5: Extracted Features and Average Flank Wear Measurements
Insert 5
Avg. fx
Avg. fy
Avg. fz
RMS fx
RMS fy
RMS fz
Avg. fz/fx
Avg. fz/fy
Skew fx
Skew fy
Skew fz
Kurtosis fx
Kurtosis fy
Kurtosis fz
Std Dev fx
Std Dev fy
Std Dev fz
Wear (mm)
1
3.0740
1.6056
3.4936
3.0784
1.6077
3.4963
1.1354
2.1827
0.1183
0.0467
-0.3753
0.9093
0.8106
0.0383
0.0761
0.0822
0.1369
0.257
Run Number
2
3
4
2.3511 3.2283 3.3801
0.6991 1.1497 0.9444
2.6344 3.6082 3.5095
2.3536 3.2294 3.3819
0.7188 1.1586 0.9542
2.6376 3.6130 3.5192
1.1207 1.1840 1.0396
3.9826 3.1691 3.7582
-0.2138 0.0144 0.3135
0.1423 0.5766 0.9037
-0.2232 0.0003 0.5452
-0.1387 0.2336 -0.0636
-0.6812 -0.4564 0.5706
-0.4334 -0.8393 -0.7859
0.1070 0.0826 0.1120
0.1671 0.1432 0.1362
0.1297 0.1874 0.2623
0.281
0.267
0.297
5
3.4617
0.9889
3.8079
3.4630
0.9918
3.8115
1.1000
3.8625
-0.1178
0.8387
0.3409
-0.0378
1.0470
-0.1868
0.0939
0.0758
0.1675
> 0.300
43
Table A6: Extracted Features and Average Flank Wear Measurements
Insert 6
Avg. fx
Avg. fy
Avg. fz
RMS fx
RMS fy
RMS fz
Avg. fz/fx
Avg. fz/fy
Skew fx
Skew fy
Skew fz
Kurtosis fx
Kurtosis fy
Kurtosis fz
Std Dev fx
Std Dev fy
Std Dev fz
Wear (mm)
1
3.3970
1.2456
3.8051
3.3986
1.2469
3.8073
1.1202
3.0580
0.2125
0.5397
0.3005
0.0680
0.2391
0.1236
0.1045
0.0577
0.1311
0.139
Run Number
2
3
4
3.1742 3.1172 3.0459
1.2451 1.4728 1.2302
3.5960 3.6172 3.5141
3.1750 3.1179 3.0466
1.2467 1.4739 1.2322
3.5970 3.6181 3.5162
1.1330 1.1606 1.1538
2.8947 2.4590 2.8642
0.0939 -0.0657 0.0916
0.0560 0.0231 -0.1282
0.2668 0.2941 -0.3870
0.1190 0.8509 0.3955
-0.5313 -0.3152 -0.4642
0.4378 0.6202 0.0703
0.0696 0.0686 0.0666
0.0638 0.0576 0.0693
0.0841 0.0815 0.1209
0.179
0.241
0.234
5
2.9456
1.8335
3.4409
2.9463
1.8364
3.4447
1.1684
1.8799
0.1813
-0.1105
0.1981
0.2421
-0.5572
-0.9336
0.0640
0.1031
0.1607
> 0.300
44
Table A7: Extracted Features and Average Flank Wear Measurements
Insert 7
Avg. fx
Avg. fy
Avg. fz
RMS fx
RMS fy
RMS fz
Avg. fz/fx
Avg. fz/fy
Skew fx
Skew fy
Skew fz
Kurtosis fx
Kurtosis fy
Kurtosis fz
Std Dev fx
Std Dev fy
Std Dev fz
Wear (mm)
1
3.0150
1.2869
3.5243
3.0161
1.2919
3.5257
1.1691
2.7596
-0.2195
0.3805
0.0582
0.9014
-0.7955
0.7979
0.0819
0.1142
0.1013
0.159
2
2.8910
1.8413
3.5192
2.8945
1.8644
3.5215
1.2185
1.9614
-0.8122
0.8558
-0.5229
0.0284
-0.4954
0.6978
0.1435
0.2927
0.1276
0.244
Run Number
3
4
3.2338 3.5707
1.4311 1.0247
3.7824 3.8749
3.2347 3.5731
1.4337 1.0420
3.7846 3.8801
1.1697 1.0870
2.6530 3.8787
-0.3142 0.2735
0.2270 0.7212
-0.5781 -0.0927
1.3750 0.2865
1.0545 -0.5058
0.7163 -0.7660
0.0800 0.1324
0.0865 0.1891
0.1286 0.2018
0.242
0.219
5
3.7549
1.1096
4.0927
3.7565
1.1116
4.0958
1.0906
3.6957
0.2137
-0.0492
-0.4826
1.2419
-0.2675
0.3036
0.1065
0.0655
0.1590
0.231
6
3.7973
1.1618
4.0958
3.7999
1.1719
4.1030
1.0782
3.5786
-1.7299
2.5997
-1.1414
4.2819
7.5529
1.4683
0.1414
0.1540
0.2434
> 0.300
45
Table A8: Extracted Features and Average Flank Wear Measurements
Insert 8
Avg. fx
Avg. fy
Avg. fz
RMS fx
RMS fy
RMS fz
Avg. fz/fx
Avg. fz/fy
Skew fx
Skew fy
Skew fz
Kurtosis fx
Kurtosis fy
Kurtosis fz
Std Dev fx
Std Dev fy
Std Dev fz
Wear (mm)
1
3.3289
1.1140
3.6702
3.3302
1.1156
3.6724
1.1028
3.3034
-0.6733
-0.4857
0.0863
1.8570
0.2879
-0.3356
0.0914
0.0591
0.1267
0.120
Run Number
2
3
4
3.2399 3.2791 3.1737
0.8913 1.1223 1.3895
3.5665 3.5628 3.6922
3.2407 3.2799 3.1757
0.8944 1.1240 1.3972
3.5713 3.5686 3.6940
1.1010 1.0870 1.1638
4.0150 3.1773 2.6895
0.0525 0.1186 -0.0672
0.1186 -0.0289 0.7248
0.1436 0.0511 0.2435
0.9929 0.4936 -0.6926
-0.7193 -0.2432 -0.2455
-1.0030 -1.1124 -0.1349
0.0710 0.0700 0.1122
0.0741 0.0622 0.1457
0.1849 0.2024 0.1157
0.139
0.210
0.270
5
3.2973
1.3154
3.8013
3.2981
1.3199
3.8024
1.1531
2.9077
-0.0160
1.2105
0.1831
0.2864
0.8384
0.3384
0.0723
0.1087
0.0913
> 0.300
46
Table A9: Extracted Features and Average Flank Wear Measurements
Insert 9
Avg. fx
Avg. fy
Avg. fz
RMS fx
RMS fy
RMS fz
Avg. fz/fx
Avg. fz/fy
Skew fx
Skew fy
Skew fz
Kurtosis fx
Kurtosis fy
Kurtosis fz
Std Dev fx
Std Dev fy
Std Dev fz
Wear (mm)
1
3.0100
1.1899
3.4603
3.0111
1.1917
3.4619
1.1498
2.9158
-0.1338
0.3343
-0.1721
1.4439
0.4673
1.3768
0.0831
0.0659
0.1047
0.162
Run Number
2
3
4
3.3747 3.1301 3.0257
1.6724 1.7225 1.9929
4.0041 3.7134 3.6538
3.3758 3.1311 3.0270
1.6752 1.7246 1.9955
4.0059 3.7149 3.6561
1.1866 1.1865 1.2077
2.4029 2.1613 1.8391
-0.3301 -0.0163 -0.2811
0.2684 -0.2069 -0.3812
-0.2958 0.0097 -0.3799
1.3932 2.0935 1.2896
-0.1446 1.5720 1.7118
0.8367 1.9858 0.3940
0.0863 0.0804 0.0860
0.0968 0.0859 0.1024
0.1216 0.1030 0.1305
0.205
0.269
0.296
5
3.1134
1.9230
3.7065
3.1154
1.9280
3.7081
1.1914
1.9371
-0.2533
-0.4210
-0.2757
1.9963
-0.0504
1.1945
0.1113
0.1384
0.1104
> 0.300
47
Table A10: Extracted Features and Average Flank Wear Measurements
Insert 10
Avg. fx
Avg. fy
Avg. fz
RMS fx
RMS fy
RMS fz
Avg. fz/fx
Avg. fz/fy
Skew fx
Skew fy
Skew fz
Kurtosis fx
Kurtosis fy
Kurtosis fz
Std Dev fx
Std Dev fy
Std Dev fz
Wear (mm)
1
3.3452
1.3695
3.7344
3.3466
1.3714
3.7368
1.1164
2.7328
-0.1009
0.2214
0.1490
0.5216
0.2050
0.4566
0.0986
0.0719
0.1327
0.140
Run Number
2
3
4
3.4213 3.3535 3.1948
1.5688 1.8812 1.8134
3.9487 3.9513 3.7353
3.4228 3.3552 3.1967
1.5732 1.8842 1.8207
3.9507 3.9536 3.7378
1.1543 1.1785 1.1694
2.5309 2.1072 2.0791
-0.0526 -0.6732 -0.0565
0.0044 -0.2450 -0.0893
0.1037 -0.0264 0.1383
0.5851 3.6335 0.3298
-0.2534 0.6315 -0.4312
0.5998 0.6810 0.2682
0.1010 0.1076 0.1087
0.1176 0.1065 0.1632
0.1265 0.1330 0.1381
0.232
0.266
0.298
5
2.3939
1.1680
2.4538
2.3970
1.1750
2.4580
1.0252
2.1306
-0.4689
-0.3451
-0.1431
1.1324
1.2758
0.5925
0.1231
0.1282
0.1436
> 0.300
48
APPENDIX B
CPN Network Training Data
49
Table B1: CPN Network Training Data
Insert #1
Tool Run Kurtosis
#
#
fy
1
0.1234
0.1329
0.1424
0.1519
2
0.1615
0.1967
0.2320
0.2673
3
0.3026
0.2820
0.2615
0.2409
4
0.2203
5
0.1000
Avg.
fz/fx
0.6856
0.6886
0.6916
0.6946
0.6976
0.6730
0.6484
0.6237
0.5991
0.5817
0.5644
0.5470
0.5296
0.3185
Avg. fy
0.7014
0.7052
0.7091
0.7129
0.7168
0.7106
0.7045
0.6983
0.6922
0.7046
0.7171
0.7296
0.7421
0.6712
Skew
fy
0.1752
0.1691
0.1631
0.1571
0.1511
0.1539
0.1567
0.1594
0.1622
0.1691
0.1760
0.1829
0.1898
0.2363
Skew
fz
0.5913
0.5993
0.6073
0.6153
0.6233
0.6217
0.6201
0.6185
0.6170
0.6175
0.6181
0.6186
0.6192
0.5535
Actual
Wear
(mm)
0.140
0.151
0.162
0.173
0.184
0.198
0.212
0.225
0.239
0.244
0.250
0.255
0.260
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
50
Table B2: CPN Network Training Data (cont.)
Insert #2
Tool Run Kurtosis
#
#
fy
1
0.1541
0.1552
0.1563
0.1574
2
0.1585
0.1590
0.1595
0.1600
3
0.1605
0.1547
0.1488
0.1430
4
0.1372
0.1347
0.1322
0.1297
5
0.1272
6
0.1785
Avg.
fz/fx
0.6869
0.6922
0.6976
0.7030
0.7084
0.7322
0.7560
0.7798
0.8036
0.7767
0.7498
0.7229
0.6960
0.7060
0.7160
0.7261
0.7361
0.6898
Avg. fy
0.6189
0.6198
0.6207
0.6216
0.6225
0.6345
0.6466
0.6586
0.6706
0.6717
0.6727
0.6737
0.6748
0.6732
0.6716
0.6701
0.6685
0.7147
Skew
fy
0.3062
0.2930
0.2798
0.2666
0.2533
0.2495
0.2456
0.2418
0.2379
0.2390
0.2401
0.2413
0.2424
0.2421
0.2418
0.2415
0.2412
0.2621
Skew
fz
0.7551
0.7413
0.7275
0.7137
0.6999
0.6922
0.6845
0.6767
0.6690
0.6718
0.6746
0.6774
0.6801
0.6759
0.6717
0.6674
0.6632
0.4888
Actual
Wear
(mm)
0.160
0.160
0.159
0.159
0.158
0.172
0.186
0.200
0.214
0.216
0.218
0.220
0.222
0.236
0.250
0.263
0.277
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
51
Table B3: CPN Network Training Data (cont.)
Insert #3
Tool Run Kurtosis
#
#
fy
1
0.2097
0.1951
0.1804
0.1658
2
0.1511
0.1651
0.1792
0.1932
3
0.2073
0.1996
0.1919
0.1842
4
0.1765
0.1735
0.1705
0.1674
5
0.1644
6
0.1873
Avg.
fz/fx
0.7729
0.7442
0.7154
0.6867
0.6579
0.6668
0.6757
0.6846
0.6935
0.6644
0.6353
0.6063
0.5772
0.6068
0.6364
0.6660
0.6956
0.5768
Avg. fy
0.7548
0.6874
0.6201
0.5527
0.4854
0.5227
0.5601
0.5975
0.6348
0.6273
0.6197
0.6122
0.6046
0.6250
0.6454
0.6658
0.6862
0.6604
Skew
fy
0.1000
0.1455
0.1910
0.2364
0.2819
0.2676
0.2532
0.2388
0.2245
0.2167
0.2088
0.2010
0.1932
0.2189
0.2445
0.2702
0.2958
0.3348
Skew
fz
0.1571
0.2963
0.4355
0.5747
0.7139
0.6962
0.6785
0.6608
0.6431
0.6257
0.6084
0.5910
0.5737
0.5540
0.5344
0.5147
0.4951
0.5183
Actual
Wear
(mm)
0.177
0.188
0.199
0.210
0.221
0.215
0.210
0.204
0.198
0.206
0.214
0.222
0.230
0.238
0.247
0.255
0.263
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
52
Table B4: CPN Network Training Data (cont.)
Insert #4
Tool Run Kurtosis
#
#
fy
1
0.1631
0.1695
0.1758
0.1822
2
0.1885
0.1790
0.1694
0.1598
3
0.1503
0.1512
0.1522
0.1531
4
0.1541
0.1662
0.1783
0.1904
5
0.2025
6
0.2407
Avg.
fz/fx
0.6211
0.6005
0.5799
0.5593
0.5387
0.5222
0.5058
0.4893
0.4729
0.4420
0.4110
0.3801
0.3491
0.4114
0.4737
0.5360
0.5983
0.4808
Avg. fy
0.5965
0.5969
0.5974
0.5978
0.5982
0.6009
0.6035
0.6061
0.6087
0.6123
0.6160
0.6196
0.6233
0.6232
0.6231
0.6230
0.6229
0.5724
Skew
fy
0.2726
0.2717
0.2709
0.2701
0.2692
0.2483
0.2273
0.2064
0.1855
0.1913
0.1972
0.2031
0.2090
0.2301
0.2512
0.2722
0.2933
0.2451
Skew
fz
0.6173
0.6305
0.6438
0.6570
0.6702
0.5999
0.5296
0.4592
0.3889
0.4529
0.5168
0.5807
0.6447
0.6289
0.6130
0.5972
0.5814
0.5240
Actual
Wear
(mm)
0.140
0.141
0.142
0.142
0.143
0.149
0.155
0.161
0.167
0.180
0.193
0.205
0.218
0.217
0.217
0.216
0.215
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
53
Table B5: CPN Network Training Data (cont.)
Insert #5
Tool Run Kurtosis
#
#
fy
1
0.2540
0.2183
0.1826
0.1468
2
0.1111
0.1165
0.1219
0.1272
3
0.1326
0.1572
0.1818
0.2064
4
0.2310
5
0.2767
Avg.
fz/fx
0.5561
0.5409
0.5257
0.5105
0.4952
0.5607
0.6262
0.6917
0.7572
0.6078
0.4584
0.3090
0.1596
0.4096
Avg. fy
0.7820
0.7129
0.6439
0.5748
0.5057
0.5401
0.5744
0.6087
0.6430
0.6274
0.6118
0.5961
0.5805
0.5940
Skew
fy
0.2546
0.2606
0.2667
0.2727
0.2787
0.3062
0.3336
0.3611
0.3885
0.4092
0.4299
0.4506
0.4712
0.4548
Skew
fz
0.4634
0.4814
0.4995
0.5175
0.5355
0.5620
0.5885
0.6150
0.6415
0.7062
0.7708
0.8354
0.9000
0.8031
Actual
Wear
(mm)
0.257
0.263
0.269
0.275
0.281
0.278
0.274
0.271
0.267
0.275
0.282
0.290
0.297
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
54
Table B6: CPN Network Training Data (cont.)
Insert #6
Tool Run Kurtosis
#
#
fy
1
0.1993
0.1808
0.1624
0.1439
2
0.1254
0.1306
0.1358
0.1410
3
0.1462
0.1426
0.1390
0.1354
4
0.1319
5
0.1230
Avg.
fz/fx
0.4932
0.5064
0.5197
0.5329
0.5461
0.5747
0.6033
0.6318
0.6604
0.6533
0.6463
0.6393
0.6322
0.6927
Avg. fy
0.6723
0.6722
0.6722
0.6722
0.6721
0.6895
0.7068
0.7242
0.7415
0.7230
0.7045
0.6861
0.6676
0.8514
Skew
fy
0.3792
0.3486
0.3181
0.2875
0.2569
0.2548
0.2528
0.2507
0.2486
0.2390
0.2295
0.2199
0.2104
0.2148
Skew
fz
0.7839
0.7799
0.7759
0.7719
0.7679
0.7712
0.7744
0.7777
0.7809
0.7001
0.6194
0.5386
0.4578
0.7354
Actual
Wear
(mm)
0.139
0.149
0.159
0.169
0.179
0.195
0.210
0.226
0.241
0.239
0.238
0.236
0.234
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
55
Table B7: CPN Network Training Data (cont.)
Insert #8
Tool Run Kurtosis
#
#
fy
1
0.2039
0.1798
0.1557
0.1316
2
0.1074
0.1188
0.1302
0.1416
3
0.1531
0.1530
0.1529
0.1529
4
0.1528
5
0.2567
Avg.
fz/fx
0.4212
0.4193
0.4174
0.4156
0.4137
0.3992
0.3847
0.3703
0.3558
0.4352
0.5147
0.5942
0.6736
0.6293
Avg. fy
0.6322
0.6152
0.5982
0.5813
0.5643
0.5819
0.5995
0.6171
0.6347
0.6551
0.6754
0.6958
0.7161
0.6935
Skew
fy
0.1200
0.1582
0.1964
0.2346
0.2727
0.2634
0.2541
0.2448
0.2355
0.2831
0.3307
0.3784
0.4260
0.5488
Skew
fz
0.6823
0.6891
0.6959
0.7027
0.7095
0.6985
0.6876
0.6766
0.6656
0.6885
0.7113
0.7341
0.7569
0.7282
Actual
Wear
(mm)
0.120
0.125
0.130
0.134
0.139
0.157
0.175
0.192
0.210
0.225
0.240
0.255
0.270
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
56
Table B8: CPN Network Training Data (cont.)
Insert #9
Tool Run Kurtosis
#
#
fy
1
0.2211
0.2065
0.1918
0.1772
2
0.1625
0.2036
0.2447
0.2858
3
0.3270
0.3303
0.3337
0.3370
4
0.3404
5
0.1715
Avg.
fz/fx
0.6157
0.6538
0.6918
0.7299
0.7680
0.7679
0.7678
0.7677
0.7676
0.7895
0.8114
0.8334
0.8553
0.7878
Avg. fy
0.6553
0.6921
0.7288
0.7656
0.8023
0.8061
0.8100
0.8138
0.8176
0.8382
0.8588
0.8794
0.9000
0.8787
Skew
fy
0.3273
0.3231
0.3189
0.3148
0.3106
0.2806
0.2505
0.2205
0.1905
0.1794
0.1684
0.1574
0.1464
0.1363
Skew
fz
0.5598
0.5451
0.5304
0.5158
0.5011
0.5373
0.5735
0.6098
0.6460
0.5998
0.5536
0.5074
0.4612
0.5106
Actual
Wear
(mm)
0.162
0.173
0.184
0.194
0.205
0.221
0.237
0.253
0.269
0.276
0.283
0.289
0.296
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
57
Table B9: CPN Network Training Data (cont.)
Insert #10
Tool Run Kurtosis
#
#
fy
1
0.1960
0.1850
0.1740
0.1631
2
0.1521
0.1733
0.1945
0.2157
3
0.2369
0.2114
0.1859
0.1605
4
0.1350
5
0.2986
Avg.
fz/fx
0.4774
0.5167
0.5559
0.5951
0.6343
0.6593
0.6844
0.7094
0.7345
0.7250
0.7156
0.7062
0.6968
0.1000
Avg. fy
0.7100
0.7252
0.7404
0.7556
0.7708
0.7946
0.8184
0.8422
0.8660
0.8608
0.8556
0.8505
0.8453
0.6486
Skew
fy
0.2987
0.2850
0.2713
0.2576
0.2439
0.2281
0.2123
0.1966
0.1808
0.1907
0.2005
0.2103
0.2202
0.1555
Skew
fz
0.7121
0.7067
0.7013
0.6960
0.6906
0.6752
0.6597
0.6443
0.6289
0.6484
0.6679
0.6875
0.7070
0.5735
Actual
Wear
(mm)
0.140
0.163
0.186
0.209
0.232
0.241
0.249
0.258
0.266
0.274
0.282
0.290
0.298
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
58
APPENDIX C
ANFIS Network Training Data
59
Table C1: ANFIS Network Training Data
Insert #1
Tool Run Kurtosis
#
#
fy
1
-0.5527
-0.4534
-0.3541
-0.2547
2
-0.1554
0.2129
0.5811
0.9494
3
1.3176
1.1029
0.8883
0.6736
4
0.4589
5
-0.7969
Avg.
fz/fx
1.1667
1.1674
1.1682
1.1689
1.1696
1.1637
1.1577
1.1518
1.1458
1.1416
1.1374
1.1332
1.1290
1.0780
Skew
Avg. fy
fy
1.3411 -0.2674
1.3537 -0.2912
1.3664 -0.3150
1.3790 -0.3388
1.3916 -0.3626
1.3714 -0.3516
1.3513 -0.3406
1.3311 -0.3296
1.3109 -0.3186
1.3518 -0.2913
1.3928 -0.2641
1.4337 -0.2368
1.4746 -0.2095
1.2420 -0.0254
Skew
fz
-0.1056
-0.0888
-0.0719
-0.0551
-0.0382
-0.0415
-0.0449
-0.0482
-0.0515
-0.0503
-0.0492
-0.0480
-0.0468
-0.1853
Actual
Wear
(mm)
0.140
0.151
0.162
0.173
0.184
0.198
0.212
0.225
0.239
0.244
0.250
0.255
0.260
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
60
Table C2: ANFIS Network Training Data (cont.)
Insert #2
Tool Run Kurtosis
#
#
fy
1
-0.2318
-0.2205
-0.2092
-0.1979
2
-0.1866
-0.1813
-0.1759
-0.1706
3
-0.1652
-0.2262
-0.2872
-0.3481
4
-0.4091
-0.4351
-0.4610
-0.4870
5
-0.5129
6
0.0227
Avg.
fz/fx
1.1670
1.1683
1.1696
1.1709
1.1722
1.1780
1.1837
1.1895
1.1952
1.1887
1.1822
1.1757
1.1692
1.1716
1.1741
1.1765
1.1789
1.1677
Skew
Skew
Avg. fy
fy
fz
1.0706 0.2510 0.2398
1.0735 0.1987 0.2107
1.0765 0.1464 0.1816
1.0794 0.0941 0.1525
1.0823 0.0418 0.1234
1.1218 0.0266 0.1071
1.1613 0.0113 0.0908
1.2007 -0.0040 0.0745
1.2402 -0.0192 0.0582
1.2436 -0.0148 0.0641
1.2470 -0.0104 0.0700
1.2504 -0.0059 0.0758
1.2538 -0.0015 0.0817
1.2487 -0.0027 0.0728
1.2435 -0.0039 0.0638
1.2384 -0.0050 0.0549
1.2332 -0.0062 0.0459
1.3849 0.0764 -0.3217
Actual
Wear
(mm)
0.160
0.160
0.159
0.159
0.158
0.172
0.186
0.200
0.214
0.216
0.218
0.220
0.222
0.236
0.250
0.263
0.277
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
61
Table C3: ANFIS Network Training Data (cont.)
Insert #3
Tool Run Kurtosis
#
#
fy
1
0.3484
0.1954
0.0425
-0.1105
2
-0.2635
-0.1170
0.0295
0.1760
3
0.3225
0.2422
0.1620
0.0817
4
0.0014
-0.0301
-0.0615
-0.0930
5
-0.1244
6
0.1144
Avg.
fz/fx
1.1878
1.1809
1.1739
1.1670
1.1600
1.1622
1.1643
1.1665
1.1686
1.1616
1.1546
1.1475
1.1405
1.1477
1.1548
1.1620
1.1691
1.1404
Skew
Avg. fy
fy
1.5163 -0.5647
1.2953 -0.3848
1.0743 -0.2049
0.8533 -0.0250
0.6323 0.1549
0.7549 0.0981
0.8776 0.0413
1.0002 -0.0156
1.1228 -0.0724
1.0980 -0.1033
1.0732 -0.1342
1.0483 -0.1650
1.0235 -0.1959
1.0905 -0.0945
1.1575 0.0069
1.2244 0.1084
1.2914 0.2098
1.2065 0.3639
Skew
fz
-1.0211
-0.7276
-0.4342
-0.1407
0.1528
0.1155
0.0782
0.0409
0.0036
-0.0330
-0.0696
-0.1062
-0.1428
-0.1842
-0.2257
-0.2671
-0.3085
-0.2596
Actual
Wear
(mm)
0.177
0.188
0.199
0.210
0.221
0.215
0.210
0.204
0.198
0.206
0.214
0.222
0.230
0.238
0.247
0.255
0.263
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
62
Table C4: ANFIS Network Training Data (cont.)
Insert #4
Tool Run Kurtosis
#
#
fy
1
-0.1382
-0.0719
-0.0056
0.0608
2
0.1271
0.0273
-0.0725
-0.1723
3
-0.2721
-0.2622
-0.2524
-0.2425
4
-0.2326
-0.1062
0.0202
0.1465
5
0.2729
6
0.6714
Avg.
fz/fx
1.1511
1.1461
1.1412
1.1362
1.1312
1.1272
1.1233
1.1193
1.1153
1.1078
1.1004
1.0929
1.0854
1.1005
1.1155
1.1306
1.1456
1.1172
Skew
Avg. fy
fy
0.9970 0.1179
0.9984 0.1146
0.9999 0.1113
1.0013 0.1080
1.0027 0.1047
1.0113 0.0219
1.0199 -0.0610
1.0284 -0.1439
1.0370 -0.2267
1.0490 -0.2034
1.0609 -0.1801
1.0729 -0.1567
1.0848 -0.1334
1.0845 -0.0501
1.0843 0.0333
1.0840 0.1166
1.0837 0.1999
0.9179 0.0092
Skew
fz
-0.0508
-0.0229
0.0050
0.0329
0.0608
-0.0875
-0.2358
-0.3840
-0.5323
-0.3975
-0.2627
-0.1279
0.0069
-0.0265
-0.0598
-0.0932
-0.1265
-0.2476
Actual
Wear
(mm)
0.140
0.141
0.142
0.142
0.143
0.149
0.155
0.161
0.167
0.180
0.193
0.205
0.218
0.217
0.217
0.216
0.215
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
63
Table C5: ANFIS Network Training Data (cont.)
Insert #5
Tool Run Kurtosis
#
#
fy
1
0.8106
0.4377
0.0647
-0.3083
2
-0.6812
-0.6250
-0.5688
-0.5126
3
-0.4564
-0.1997
0.0571
0.3139
4
0.5706
5
1.0470
Avg.
fz/fx
1.1354
1.1317
1.1281
1.1244
1.1207
1.1365
1.1524
1.1682
1.1840
1.1479
1.1118
1.0757
1.0396
1.1000
Avg. fy
1.6056
1.3790
1.1524
0.9257
0.6991
0.8118
0.9244
1.0371
1.1497
1.0984
1.0471
0.9957
0.9444
0.9889
Skew
fy
0.0467
0.0706
0.0945
0.1184
0.1423
0.2509
0.3595
0.4680
0.5766
0.6584
0.7402
0.8219
0.9037
0.8387
Skew
fz
-0.3753
-0.3373
-0.2993
-0.2612
-0.2232
-0.1673
-0.1115
-0.0556
0.0003
0.1365
0.2728
0.4090
0.5452
0.3409
Actual
Wear
(mm)
0.257
0.263
0.269
0.275
0.281
0.278
0.274
0.271
0.267
0.275
0.282
0.290
0.297
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
64
Table C6: ANFIS Network Training Data (cont.)
Insert #6
Tool Run Kurtosis
#
#
fy
1
0.2391
0.0465
-0.1461
-0.3387
2
-0.5313
-0.4773
-0.4233
-0.3692
3
-0.3152
-0.3525
-0.3897
-0.4270
4
-0.4642
5
-0.5572
Avg.
fz/fx
1.1202
1.1234
1.1266
1.1298
1.1330
1.1399
1.1468
1.1537
1.1606
1.1589
1.1572
1.1555
1.1538
1.1684
Skew
Skew
Avg. fy
fy
fz
1.2456 0.5397 0.3005
1.2455 0.4188 0.2921
1.2454 0.2979 0.2837
1.2452 0.1769 0.2752
1.2451 0.0560 0.2668
1.3020 0.0478 0.2736
1.3590 0.0396 0.2805
1.4159 0.0313 0.2873
1.4728 0.0231 0.2941
1.4122 -0.0147 0.1238
1.3515 -0.0526 -0.0465
1.2909 -0.0904 -0.2167
1.2302 -0.1282 -0.3870
1.8335 -0.1105 0.1981
Actual
Wear
(mm)
0.139
0.149
0.159
0.169
0.179
0.195
0.210
0.226
0.241
0.239
0.238
0.236
0.234
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
65
Table C7: ANFIS Network Training Data (cont.)
Insert #8
Tool Run Kurtosis
#
#
fy
1
0.2879
0.0361
-0.2157
-0.4675
2
-0.7193
-0.6003
-0.4813
-0.3622
3
-0.2432
-0.2438
-0.2444
-0.2449
4
-0.2455
5
0.8384
Avg.
fz/fx
1.1028
1.1024
1.1019
1.1015
1.1010
1.0975
1.0940
1.0905
1.0870
1.1062
1.1254
1.1446
1.1638
1.1531
Skew
Avg. fy
fy
1.1140 -0.4857
1.0583 -0.3346
1.0027 -0.1836
0.9470 -0.0325
0.8913 0.1186
0.9491 0.0817
1.0068 0.0449
1.0646 0.0080
1.1223 -0.0289
1.1891 0.1595
1.2559 0.3480
1.3227 0.5364
1.3895 0.7248
1.3154 1.2105
Skew
fz
0.0863
0.1006
0.1150
0.1293
0.1436
0.1205
0.0974
0.0742
0.0511
0.0992
0.1473
0.1954
0.2435
0.1831
Actual
Wear
(mm)
0.120
0.125
0.130
0.134
0.139
0.157
0.175
0.192
0.210
0.225
0.240
0.255
0.270
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
66
Table C8: ANFIS Network Training Data (cont.)
Insert #9
Tool Run Kurtosis
#
#
fy
1
0.4673
0.3143
0.1614
0.0084
2
-0.1446
0.2846
0.7137
1.1429
3
1.5720
1.6070
1.6419
1.6769
4
1.7118
5
-0.0504
Avg.
fz/fx
1.1498
1.1590
1.1682
1.1774
1.1866
1.1866
1.1866
1.1865
1.1865
1.1918
1.1971
1.2024
1.2077
1.1914
Skew
Avg. fy
fy
1.1899 0.3343
1.3105 0.3178
1.4312 0.3014
1.5518 0.2849
1.6724 0.2684
1.6849 0.1496
1.6975 0.0308
1.7100 -0.0881
1.7225 -0.2069
1.7901 -0.2505
1.8577 -0.2941
1.9253 -0.3376
1.9929 -0.3812
1.9230 -0.4210
Skew
fz
-0.1721
-0.2030
-0.2340
-0.2649
-0.2958
-0.2194
-0.1431
-0.0667
0.0097
-0.0877
-0.1851
-0.2825
-0.3799
-0.2757
Actual
Wear
(mm)
0.162
0.173
0.184
0.194
0.205
0.221
0.237
0.253
0.269
0.276
0.283
0.289
0.296
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
67
Table C9: ANFIS Network Training Data (cont.)
Insert #10
Tool Run Kurtosis
#
#
fy
1
0.2050
0.0904
-0.0242
-0.1388
2
-0.2534
-0.0322
0.1891
0.4103
3
0.6315
0.3658
0.1002
-0.1655
4
-0.4312
5
1.2758
Avg.
fz/fx
1.1164
1.1259
1.1354
1.1448
1.1543
1.1604
1.1664
1.1725
1.1785
1.1762
1.1740
1.1717
1.1694
1.0252
Skew
Skew
Avg. fy
fy
fz
1.3695 0.2214 0.1490
1.4193 0.1672 0.1377
1.4692 0.1129 0.1264
1.5190 0.0587 0.1150
1.5688 0.0044 0.1037
1.6469 -0.0580 0.0712
1.7250 -0.1203 0.0387
1.8031 -0.1827 0.0061
1.8812 -0.2450 -0.0264
1.8643 -0.2061 0.0148
1.8473 -0.1672 0.0560
1.8304 -0.1282 0.0971
1.8134 -0.0893 0.1383
1.1680 -0.3451 -0.1431
Actual
Wear
(mm)
0.140
0.163
0.186
0.209
0.232
0.241
0.249
0.258
0.266
0.274
0.282
0.290
0.298
*
Classification
(mm)
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.3
0.7
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
68
APPENDIX D
Data for On-Line Testing
69
Table D1: CPN On-line Data
Insert #7
Tool
#
Run
#
1
2
3
4
5
6
Kurtosis
fy
0.1001
0.1289
0.2774
0.1279
0.1507
0.9000
Avg.
fz/fx
0.6956
0.9000
0.6980
0.3558
0.3707
0.3193
Avg.
fy
0.6849
0.8538
0.7288
0.6050
0.6308
0.6467
Skew
fy
0.3390
0.4591
0.3002
0.4251
0.2303
0.9000
Skew
fz
0.6690
0.3934
0.3672
0.5974
0.4125
0.1000
Ave.
flank
Wear
(mm)
0.159
0.244
0.242
0.219
0.231
*
Classification
Value (mm)
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
> 0.5
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
Table D2: ANFIS On-line Data
Insert #7
Tool
#
Run
#
1
2
3
4
5
6
Kurtosis
fy
-0.7955
-0.4954
1.0545
-0.5058
-0.2675
7.5529
Avg.
fz/fx
1.1691
1.2185
1.1697
1.0870
1.0906
1.0782
Avg.
fy
1.2869
1.8413
1.4311
1.0247
1.1096
1.1618
Skew
fy
0.3805
0.8558
0.2270
0.7212
-0.0492
2.5997
Skew
fz
0.0582
-0.5229
-0.5781
-0.0927
-0.4826
-1.1414
Ave.
flank
Wear
(mm)
0.159
0.244
0.242
0.219
0.231
*
Classification
Value (mm)
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
> 0.5
* Average flank wear width could not be measured practically due to severe tool failure.
Therefore, these worn-out tools were not used for the training for on-line measurements.
70
APPENDIX E
CPN On-line Classification Outputs
71
Table E1: CPN On-line Classification Outputs
Inputs x 30 x 1 (CPN)
Number of
Inputs
#1
1
0.3585
2
0.4511
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.3000
0.4275
0.3678
0.3000
0.3000
≤ 0.5
Run Number
#3
#4
0.5314 0.3000
0.4770 0.3386
0.3725 0.3263
0.3386 0.5314
0.3000 0.3263
≤ 0.5
≤ 0.5
#5
0.3000
0.3000
0.3263
0.3092
0.3000
#6
0.3000
0.5314
0.5314
0.7000
0.3386
≤ 0.5
> 0.5
Success
Rate
66.67%
100.00%
100.00%
83.33%
83.33%
Table E2: CPN On-line Classification Outputs
Inputs x 33 x 1 (CPN)
Number of
Inputs
#1
1
0.3940
2
0.4770
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.3000
0.4365
0.3678
0.3000
0.4511
≤ 0.5
Run Number
#3
#4
0.4290 0.3000
0.3000 0.3386
0.3000 0.3263
0.3386 0.3000
0.3386 0.3000
≤ 0.5
≤ 0.5
#5
0.3000
0.3386
0.3263
0.3000
0.3000
#6
0.3000
0.5314
0.7000
0.7000
0.3585
≤ 0.5
> 0.5
Success
Rate
83.33%
100.00%
100.00%
100.00%
83.33%
Table E3: CPN On-line Classification Outputs
Inputs x 36 x 1 (CPN)
Number of
Inputs
#1
1
0.3940
2
0.4770
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.3000
0.4365
0.3995
0.3000
0.3000
≤ 0.5
Run Number
#3
#4
0.5314 0.3000
0.3000 0.3386
0.3000 0.3263
0.3386 0.3000
0.3585 0.3263
≤ 0.5
≤ 0.5
#5
0.3000
0.3386
0.3263
0.3386
0.3000
#6
0.3000
0.5314
0.7000
0.7000
0.7000
≤ 0.5
> 0.5
Success
Rate
66.67%
100.00%
100.00%
100.00%
100.00%
72
Table E4: CPN On-line Classification Outputs
Inputs x 39 x 1 (CPN)
Number of
Inputs
#1
1
0.3940
2
0.4511
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.3000
0.4365
0.3000
0.3000
0.3000
≤ 0.5
Run Number
#3
#4
0.5314 0.3000
0.3000 0.3386
0.3000 0.3386
0.3940 0.3000
0.3585 0.3000
≤ 0.5
≤ 0.5
#5
0.3000
0.3386
0.3386
0.3386
0.3000
#6
0.3000
0.7000
0.7000
0.7000
0.3000
≤ 0.5
> 0.5
Success
Rate
66.67%
100.00%
100.00%
100.00%
83.33%
Table E5: CPN On-line Classification Outputs
Inputs x 42 x 1 (CPN)
Number of
Inputs
#1
1
0.3940
2
0.4511
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.3000
0.4365
0.3000
0.3000
0.3000
≤ 0.5
Run Number
#3
#4
0.4290 0.3000
0.3000 0.3386
0.5314 0.3263
0.3585 0.3000
0.3940 0.3000
≤ 0.5
≤ 0.5
#5
0.3000
0.3386
0.3263
0.3263
0.3000
#6
0.3000
0.7000
0.7000
0.7000
0.3000
≤ 0.5
> 0.5
Success
Rate
83.33%
100.00%
83.33%
100.00%
83.33%
Table E6: CPN On-line Classification Outputs
Inputs x 45 x 1 (CPN)
Number of
Inputs
#1
1
0.3940
2
0.4770
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.3000
0.4511
0.3000
0.3000
0.4770
≤ 0.5
Run Number
#3
#4
0.7000 0.3000
0.3000 0.3585
0.3000 0.3386
0.3000 0.4770
0.3585 0.3000
≤ 0.5
≤ 0.5
#5
0.3000
0.3000
0.3386
0.3386
0.3000
#6
0.3000
0.7000
0.7000
0.7000
0.3000
≤ 0.5
> 0.5
Success
Rate
66.67%
100.00%
100.00%
100.00%
83.33%
73
Table E7: CPN On-line Classification Outputs
Inputs x 48 x 1 (CPN)
Number of
Inputs
#1
1
0.3940
2
0.4770
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.3000
0.4770
0.3000
0.3000
0.3000
≤ 0.5
Run Number
#3
#4
0.5314 0.3000
0.3000 0.3386
0.3000 0.3263
0.3940 0.5314
0.3585 0.3000
≤ 0.5
≤ 0.5
#5
0.3000
0.3386
0.3263
0.3386
0.3000
#6
0.3000
0.7000
0.7000
0.7000
0.3000
≤ 0.5
> 0.5
Success
Rate
66.67%
100.00%
100.00%
83.33%
83.33%
Table E8: CPN On-line Classification Outputs
Inputs x 51 x 1 (CPN)
Number of
Inputs
#1
1
0.3940
2
0.4770
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.4365
0.4365
0.5314
0.3000
0.3000
≤ 0.5
Run Number
#3
#4
0.7000 0.4365
0.3000 0.3585
0.5314 0.3000
0.3000 0.3000
0.3940 0.3000
≤ 0.5
≤ 0.5
#5
0.3000
0.3000
0.3000
0.3386
0.3585
#6
0.3000
0.7000
0.3000
0.7000
0.7000
≤ 0.5
> 0.5
Success
Rate
66.67%
100.00%
50.00%
100.00%
100.00%
Table E9: CPN On-line Classification Outputs
Inputs x 54 x 1 (CPN)
Number of
Inputs
#1
1
0.3940
2
0.4770
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.3000
0.4770
0.3000
0.3000
0.3000
≤ 0.5
Run Number
#3
#4
0.7000 0.3000
0.3000 0.3386
0.7000 0.3585
0.3000 0.3000
0.3940 0.3386
≤ 0.5
≤ 0.5
#5
0.3000
0.3386
0.3585
0.3263
0.3000
#6
0.3000
0.5314
0.3000
0.7000
0.7000
≤ 0.5
> 0.5
Success
Rate
66.67%
100.00%
66.67%
100.00%
100.00%
74
Table E10: CPN On-line Classification Outputs
Inputs x 57 x 1 (CPN)
Number of
Inputs
#1
1
0.3940
2
0.4770
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.3000
0.4511
0.3000
0.3000
0.3000
≤ 0.5
Run Number
#3
#4
0.7000 0.3000
0.3000 0.3940
0.3000 0.3000
0.3000 0.3000
0.3940 0.3386
≤ 0.5
≤ 0.5
#5
0.3000
0.3000
0.3000
0.3386
0.3000
#6
0.3000
0.5314
0.3000
0.7000
0.3000
≤ 0.5
> 0.5
Success
Rate
66.67%
100.00%
83.33%
100.00%
83.33%
Table E11: CPN On-line Classification Outputs
Inputs x 60 x 1 (CPN)
Number of
Inputs
#1
1
0.3940
2
0.4770
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.3000
0.4365
0.3000
0.3000
0.3000
≤ 0.5
Run Number
#3
#4
0.5314 0.3000
0.3000 0.3585
0.3000 0.3585
0.3000 0.3000
0.3940 0.3000
≤ 0.5
≤ 0.5
#5
0.3000
0.3000
0.3585
0.3386
0.3000
#6
0.3000
0.7000
0.3000
0.7000
0.3000
≤ 0.5
> 0.5
Success
Rate
66.67%
100.00%
83.33%
100.00%
83.33%
Table E12: CPN On-line Classification Outputs
Inputs x 63 x 1 (CPN)
Number of
Inputs
#1
1
0.3940
2
0.5314
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.3000
0.4770
0.3000
0.3000
0.3000
≤ 0.5
Run Number
#3
#4
0.7000 0.3000
0.3000 0.3000
0.3000 0.3585
0.3000 0.3000
0.3000 0.3000
≤ 0.5
≤ 0.5
#5
0.3000
0.3000
0.3585
0.3585
0.3000
#6
0.3000
0.7000
0.7000
0.7000
0.7000
≤ 0.5
> 0.5
Success
Rate
66.67%
83.33%
100.00%
100.00%
100.00%
75
Table E13: CPN On-line Classification Outputs
Inputs x 66 x 1 (CPN)
Number of
Inputs
#1
1
0.3940
2
0.4770
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.3000
0.4770
0.3000
0.3000
0.3000
≤ 0.5
Run Number
#3
#4
0.7000 0.3000
0.3000 0.3000
0.3000 0.3000
0.3000 0.5314
0.3940 0.3000
≤ 0.5
≤ 0.5
#5
0.3000
0.3000
0.3000
0.3585
0.3000
#6
0.3000
0.7000
0.7000
0.7000
0.3000
≤ 0.5
> 0.5
Success
Rate
66.67%
100.00%
100.00%
83.33%
83.33%
Table E14: CPN On-line Classification Outputs
Inputs x 69 x 1 (CPN)
Number of
Inputs
#1
1
0.4686
2
0.4770
3
0.3000
4
0.3000
5
0.3000
Success
≤ 0.5
Criteria
#2
0.3000
0.4365
0.3000
0.3000
0.3000
≤ 0.5
Run Number
#3
#4
0.7000 0.3000
0.3000 0.3940
0.3000 0.3000
0.3000 0.3000
0.3940 0.3000
≤ 0.5
≤ 0.5
#5
0.3000
0.3000
0.3000
0.3000
0.3000
#6
0.3000
0.5314
0.7000
0.5314
0.7000
≤ 0.5
> 0.5
Success
Rate
66.67%
100.00%
100.00%
100.00%
100.00%
76
APPENDIX F
ANFIS On-line Classification Outputs
77
Table F1: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x4 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.5861
0.6633
0.6736
0.6735
0.4035
0.4281
0.3522
#2
0.3129
0.3134
0.3132
0.3132
0.3103
0.3105
0.3321
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4397 0.3129
0.4442 0.3134
0.4360 0.3134
0.4360 0.3134
0.4335 0.3103
0.4478 0.3105
0.4190 0.3328
≤ 0.5
≤ 0.5
#5
0.3129
0.3134
0.3122
0.3122
0.3103
0.3105
0.3176
#6
0.3215
0.3134
0.5000
0.2988
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
66.67%
66.67%
66.67%
66.67%
83.33%
83.33%
83.33%
Table F2: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x4 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.3411
0.3459
0.3372
0.3372
0.3387
0.3356
0.3443
#2
0.3330
0.3321
0.3334
0.3334
0.3342
0.3345
0.3286
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4079 0.3337
0.4112 0.3329
0.4036 0.3340
0.4036 0.3340
0.4054 0.3351
0.4033 0.3349
0.4141 0.3291
≤ 0.5
≤ 0.5
#5
0.3126
0.3132
0.3132
0.3132
0.3142
0.3122
0.3165
#6
0.3701
0.4130
0.5000
0.2960
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
78
Table F3: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x5 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.6364
0.6684
0.6797
0.6799
0.4219
0.4000
0.5241
#2
0.3132
0.3120
0.3158
0.3158
0.3116
0.3133
0.3062
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4383 0.3132
0.4458 0.3120
0.4362 0.3158
0.4362 0.3158
0.4516 0.3116
0.4725 0.3133
0.4666 0.3115
≤ 0.5
≤ 0.5
#5
0.3136
0.3121
0.3157
0.3157
0.3116
0.3133
0.3052
#6
0.4057
0.3120
0.5000
0.2991
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
66.67%
66.67%
66.67%
66.67%
83.33%
83.33%
66.67%
Table F4: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x5 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.3668
0.3643
0.3688
0.3688
0.3669
0.3689
0.3683
#2
0.3397
0.3373
0.3427
0.3427
0.3413
0.3452
0.3363
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4492 0.3418
0.4480 0.3388
0.4470 0.3457
0.4470 0.3457
0.4464 0.3437
0.4398 0.3494
0.4510 0.3375
≤ 0.5
≤ 0.5
#5
0.3109
0.3115
0.3101
0.3101
0.3103
0.3109
0.3119
#6
0.3573
0.4573
0.5000
0.2978
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
79
Table F5: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x6 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.6767
0.6702
0.4864
0.4864
0.5600
0.4000
0.5553
#2
0.3060
0.3063
0.3168
0.3168
0.3082
0.3001
0.3068
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4559 0.3061
0.4421 0.3063
0.4627 0.3216
0.4627 0.3216
0.4600 0.3183
0.4600 0.3001
0.4364 0.3121
≤ 0.5
≤ 0.5
#5
0.3068
0.3066
0.2952
0.2952
0.2966
0.3001
0.3014
#6
0.3109
0.2968
0.5000
0.2982
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
66.67%
66.67%
83.33%
83.33%
66.67%
83.33%
66.67%
Table F6: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x6 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.4081
0.4050
0.4071
0.4071
0.4048
0.4032
0.4078
#2
0.3307
0.3354
0.3249
0.3249
0.3273
0.3151
0.3392
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4435 0.3348
0.4364 0.3387
0.4433 0.3299
0.4433 0.3299
0.4437 0.3327
0.4504 0.3208
0.4340 0.3416
≤ 0.5
≤ 0.5
#5
0.2998
0.2991
0.3006
0.3006
0.3018
0.3028
0.2949
#6
0.3565
0.4403
0.5000
0.2962
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
80
Table F7: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x7 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.6766
0.6724
0.4834
0.4834
0.5883
0.5205
0.5411
#2
0.3044
0.3055
0.3195
0.3195
0.3054
0.2964
0.3163
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4331 0.3045
0.4327 0.3055
0.4421 0.3256
0.4421 0.3256
0.4385 0.3146
0.4488 0.2964
0.4448 0.3192
≤ 0.5
≤ 0.5
#5
0.3042
0.3055
0.2945
0.2945
0.2952
0.2964
0.2966
#6
0.3390
0.2940
0.5000
0.2960
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
66.67%
66.67%
83.33%
83.33%
66.67%
66.67%
66.67%
Table F8: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x7 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.4396
0.4482
0.4285
0.4285
0.4252
0.4117
0.4603
#2
0.3123
0.3186
0.3062
0.3062
0.2990
0.3023
0.3277
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4415 0.3158
0.4559 0.3231
0.4385 0.3090
0.4385 0.3090
0.4374 0.3002
0.4396 0.3024
0.4429 0.3323
≤ 0.5
≤ 0.5
#5
0.2955
0.2931
0.2990
0.2990
0.2990
0.3023
0.2935
#6
0.3682
0.4565
0.5000
0.2945
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
81
Table F9: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x8 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.4891
0.5058
0.5153
0.5157
0.5633
0.5516
#2
0.3274
0.3141
0.3104
0.3095
Training
0.3217
0.3112
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4321 0.3318
0.4164 0.3181
0.4144 0.3186
0.4145 0.3176
error
IMF
0.4333 0.3285
0.4564 0.3124
≤ 0.5
≤ 0.5
#5
0.2845
0.2939
0.2937
0.2940
b>c
0.2983
0.2840
#6
0.3598
0.6567
0.5000
0.2976
Success
Rate
83.33%
83.33%
66.67%
66.67%
0.5000
0.5000
66.67%
66.67%
≤ 0.5
> 0.5
Table F10: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x8 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.4537
0.4690
0.4387
0.4387
0.4266
0.4168
0.4855
#2
0.3066
0.3074
0.3052
0.3052
0.3057
0.3086
0.3121
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4470 0.3087
0.4562 0.3112
0.4226 0.3062
0.4226 0.3062
0.4154 0.3057
0.4195 0.3086
0.4553 0.3182
≤ 0.5
≤ 0.5
#5
0.3068
0.3039
0.3103
0.3103
0.3121
0.3123
0.3040
#6
0.4232
0.4575
0.5000
0.2884
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
82
Table F11: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x9 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.6661
0.6686
0.6779
0.6782
0.6958
0.6977
0.6453
#2
0.3200
0.3171
0.3205
0.3207
0.3156
0.3157
0.3080
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4274 0.3206
0.3813 0.3172
0.3865 0.3213
0.3867 0.3216
0.3800 0.3156
0.3800 0.3157
0.4072 0.3160
≤ 0.5
≤ 0.5
#5
0.2960
0.2963
0.2973
0.2972
0.2998
0.2989
0.3002
#6
0.2981
0.5000
0.5000
0.2979
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
66.67%
66.67%
66.67%
66.67%
66.67%
66.67%
66.67%
Table F12: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x9 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.4641
0.4764
0.4496
0.4496
0.4374
0.4331
0.4941
#2
0.3107
0.3093
0.3110
0.3110
0.3128
0.3135
0.3036
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4082 0.3121
0.4114 0.3118
0.4039 0.3113
0.4039 0.3113
0.4017 0.3128
0.4028 0.3135
0.4092 0.3102
≤ 0.5
≤ 0.5
#5
0.3057
0.3036
0.3069
0.3069
0.3071
0.3076
0.3030
#6
0.3719
0.5000
0.5000
0.2817
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83
Table F13: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x10
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
Run Number
#1
0.6577
0.6705
0.6739
0.6747
0.6955
0.6983
0.6741
#2
0.3161
0.3211
0.3066
0.3049
0.3198
0.3185
0.3186
#3
0.4814
0.4672
0.4975
0.4976
0.5000
0.5000
0.4090
#4
0.3211
0.3220
0.3129
0.3102
0.3198
0.3185
0.3230
#5
0.2975
0.2977
0.2989
0.2991
0.2982
0.2990
0.3030
#6
0.2952
0.5000
0.5000
0.3022
0.5000
0.5000
0.5000
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
> 0.5
Success
Rate
66.67%
66.67%
66.67%
66.67%
66.67%
66.67%
66.67%
Table F14: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x10
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
Run Number
#1
0.4760
0.4855
0.4634
0.4634
0.4544
0.4465
0.5015
#2
0.3142
0.3124
0.3156
0.3156
0.3167
0.3174
0.3061
#3
0.4564
0.4497
0.4670
0.4670
0.4762
0.4920
0.4017
#4
0.3154
0.3143
0.3158
0.3158
0.3167
0.3174
0.3066
#5
0.2961
0.2954
0.2969
0.2969
0.2974
0.2981
0.2957
#6
0.3652
0.5000
0.5000
0.3077
0.5000
0.5000
0.5000
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
> 0.5
Success
Rate
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
66.67%
84
Table F15: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x11
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
Run Number
#1
0.6639
0.6775
0.6772
0.6769
0.6959
0.6992
0.6250
#2
0.3225
0.3211
0.3047
0.3041
0.3232
0.3226
0.3057
#3
0.4131
0.3637
0.4704
0.4676
0.4970
0.4986
0.4360
#4
0.3254
0.3224
0.3103
0.3095
0.3257
0.3261
0.3154
#5
0.2961
0.2930
0.2986
0.2987
0.2955
0.2969
0.2963
#6
0.4967
0.5000
0.5000
0.3031
0.5000
0.5000
0.5000
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
> 0.5
Success
Rate
66.67%
66.67%
66.67%
66.67%
66.67%
66.67%
66.67%
Table F16: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x11
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
Run Number
#1
0.4896
0.4962
0.4795
0.4795
0.4703
0.4611
0.5020
#2
0.3167
0.3148
0.3198
0.3198
0.3214
0.3208
0.3138
#3
0.4397
0.4600
0.4460
0.4460
0.4460
0.4646
0.4336
#4
0.3179
0.3166
0.3201
0.3201
0.3214
0.3208
0.3154
#5
0.2938
0.2925
0.2945
0.2945
0.2958
0.2972
0.2890
#6
0.3574
0.5000
0.5000
0.3035
0.5000
0.5000
0.5000
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
> 0.5
Success
Rate
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
66.67%
85
Table F17: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x12
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
Run Number
#1
0.6605
0.6918
0.6798
0.6800
0.7000
0.7000
0.6942
#2
0.3109
0.3188
0.3038
0.3037
0.3184
0.2990
0.3205
#3
0.3583
0.5726
0.3830
0.3831
0.3800
0.3800
0.4095
#4
0.3164
0.3238
0.3085
0.3084
0.3232
0.3039
0.3237
#5
0.2983
0.2983
0.2994
0.2994
0.2956
0.2934
0.2957
#6
0.2883
0.2969
0.5000
0.3027
0.5000
0.5000
0.5000
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
> 0.5
Success
Rate
66.67%
50.00%
66.67%
66.67%
66.67%
66.67%
66.67%
Table F18: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x12
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
Run Number
#1
0.5040
0.5090
0.4973
0.4973
0.4911
0.4804
0.5127
#2
0.3190
0.3153
0.3238
0.3238
0.3272
0.3278
0.3150
#3
0.4095
0.4096
0.4094
0.4094
0.4060
0.4086
0.4203
#4
0.3208
0.3175
0.3254
0.3254
0.3291
0.3280
0.3167
#5
0.2962
0.2970
0.2962
0.2962
0.2955
0.2967
0.2973
#6
0.3571
0.5000
0.5000
0.3023
0.5000
0.5000
0.5000
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
≤ 0.5
> 0.5
Success
Rate
66.67%
66.67%
83.33%
83.33%
83.33%
83.33%
66.67%
86
Table F19: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
2x2 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.3181
0.3162
0.3179
0.3179
0.3189
0.3169
0.3148
#2
0.3184
0.3164
0.3179
0.3179
0.3189
0.3169
0.3176
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.3110 0.2915
0.3205 0.2885
0.3187 0.3045
0.3187 0.3045
0.3210 0.2979
0.3269 0.3148
0.3199 0.3023
≤ 0.5
≤ 0.5
#5
0.3047
0.3125
0.3051
0.3051
0.2979
0.3148
0.3218
#6
0.5829
0.6721
0.6583
0.6583
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
100.00%
100.00%
100.00%
100.00%
83.33%
83.33%
83.33%
Table F20: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
2x2 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.2972
0.2912
0.3158
0.3158
0.3150
0.3159
0.2817
#2
0.2950
0.2870
0.3159
0.3159
0.3150
0.3159
0.2808
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.3146 0.2971
0.3170 0.2985
0.3127 0.3016
0.3127 0.3016
0.3135 0.3023
0.3208 0.3078
0.3259 0.3058
≤ 0.5
≤ 0.5
#5
0.3106
0.3151
0.3026
0.3026
0.3030
0.3080
0.3215
#6
0.4420
0.4734
0.4694
0.4694
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
87
Table F21: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
2x3 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.3086
0.3039
0.3088
0.3089
0.3073
≤ 0.5
#2
0.3106
0.3075
0.3090
0.3091
Training
0.3073
Training
≤ 0.5
Run Number
#3
#4
0.3157 0.3454
0.3197 0.3502
0.3163 0.3159
0.3164 0.3161
error
IMF
0.3207 0.2955
error
IMF
≤ 0.5
≤ 0.5
#5
0.3102
0.3243
0.3026
0.3018
b>c
0.2994
a>b
#6
0.7367
0.5050
0.5000
1.0516
Success
Rate
100.00%
100.00%
83.33%
100.00%
0.5000
83.33%
≤ 0.5
> 0.5
Table F22: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
2x3 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.3277
0.3212
0.3150
0.3149
0.3152
0.3135
0.3407
#2
0.3722
0.3508
0.3271
0.3272
0.3282
0.3187
0.3678
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.3215 0.3560
0.3222 0.3481
0.3256 0.3826
0.3257 0.3826
0.3348 0.3560
0.3287 0.3055
0.3516 0.3613
≤ 0.5
≤ 0.5
#5
0.3519
0.3503
0.3477
0.3477
0.3018
0.2999
0.3403
#6
0.6436
0.2074
0.5000
3.0007
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
100.00%
83.33%
83.33%
100.00%
83.33%
83.33%
83.33%
88
Table F23: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
2x4 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.3300
0.3305
0.3282
0.3280
0.3551
0.5000
0.5000
#2
-1.5112
-12.2362
-45.5911
-45.5804
0.2019
0.5000
0.5000
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.3414 0.3933
0.3439 0.3712
0.3440 0.3480
0.3440 0.3480
0.3385 0.3596
0.5000 0.5000
0.5000 0.5000
≤ 0.5
≤ 0.5
#5
0.2955
0.2793
0.2999
0.2999
0.2924
0.5000
0.5000
#6
0.5605
-0.8822
0.5000
-3.0528
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
100.00%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
Table F24: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
2x4 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.3170
0.3096
0.3195
0.3195
0.3209
0.3246
0.2715
#2
0.2903
0.2604
0.2822
0.2822
0.3052
0.2825
0.2728
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.3418 0.3845
0.2960 0.3787
0.3464 0.3427
0.3464 0.3427
0.3447 0.3755
0.3564 0.3791
0.2813 0.3006
≤ 0.5
≤ 0.5
#5
0.3116
0.3104
0.3008
0.3008
0.3030
0.2993
0.2859
#6
0.1957
0.1654
0.5000
-4.6235
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
89
Table F25: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
2x5 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.4185
0.3772
0.4279
0.4279
0.4338
≤ 0.5
#2
-33.1151
-22.7308
-0.9883
-0.9791
Training
-0.1506
Training
≤ 0.5
Run Number
#3
#4
0.3164 0.3365
0.5690 0.3525
1.4923 0.4168
1.4961 0.4168
error
IMF
0.0045 0.4507
error
IMF
≤ 0.5
≤ 0.5
#5
0.2969
0.2935
0.3006
0.3006
a>b
0.3001
a>b
#6
-19.7372
-33.8544
0.5000
0.0357
Success
Rate
83.33%
66.67%
66.67%
66.67%
0.5000
83.33%
≤ 0.5
> 0.5
Table F26: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
2x5 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.3986
0.3774
0.3956
0.3956
0.3877
0.3855
0.3923
#2
0.0892
0.1532
0.1370
0.1370
0.1880
0.1647
0.2067
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.5689 0.3900
0.4815 0.3734
0.4668 0.4663
0.4668 0.4663
0.4610 0.4606
0.4635 0.5280
0.3058 0.3840
≤ 0.5
≤ 0.5
#5
0.2954
0.2920
0.2901
0.2901
0.2932
0.2947
0.2931
#6
-4.0652
-10.6947
0.5000
-0.6320
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
66.67%
83.33%
83.33%
83.33%
83.33%
66.67%
83.33%
90
Table F27: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
2x6 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.5683
0.4267
0.5037
0.5041
0.5000
≤ 0.5
#2
-4.4725
-11.9043
-0.9918
-0.9781
Training
0.5000
Training
≤ 0.5
Run Number
#3
#4
0.4255 0.2673
0.3384 0.3371
0.2994 0.3114
0.2994 0.3113
error
IMF
0.5000 0.5000
error
IMF
≤ 0.5
≤ 0.5
#5
0.3043
0.2990
0.3010
0.3009
c>d
0.5000
b>c
#6
-7.1768
-0.4858
0.5000
0.0104
Success
Rate
66.67%
83.33%
66.67%
66.67%
0.5000
83.33%
≤ 0.5
> 0.5
Table F28: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
2x6 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.5619
0.5290
0.5637
0.5639
0.5421
0.5865
0.4996
#2
0.1690
0.1499
0.2113
0.2131
0.2631
0.3472
0.1818
≤ 0.5
≤ 0.5
Run Number
#3
#4
1.2967 0.3473
0.6702 0.3118
0.4891 0.3458
0.5529 0.3507
0.1688 0.3108
0.1483 0.3005
0.9022 0.3593
≤ 0.5
≤ 0.5
#5
0.2989
0.3049
0.2997
0.2731
0.3002
0.3001
0.2759
#6
3.5399
-2.2040
0.5000
0.6991
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
66.67%
50.00%
66.67%
66.67%
66.67%
66.67%
66.67%
91
Table F29: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
3x2 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.3147
0.3060
0.3202
0.3202
0.3158
0.3145
0.3106
#2
0.4673
0.6835
0.3360
0.3360
0.3546
0.3355
0.5947
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.2952 0.2908
0.3075 0.2803
0.3128 0.2987
0.3128 0.2987
0.3532 0.2801
0.2887 0.3003
0.2922 0.2866
≤ 0.5
≤ 0.5
#5
0.3035
0.3028
0.3046
0.3046
0.2924
0.3045
0.3179
#6
0.5374
0.5613
0.6689
0.6689
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
100.00%
83.33%
100.00%
100.00%
83.33%
83.33%
66.67%
Table F30 ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
3x2 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.3209
0.3257
0.3106
0.3106
0.3066
0.3123
0.5005
#2
0.3796
0.4016
0.3451
0.3451
0.3593
0.3449
0.4795
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.3471 0.2968
0.3422 0.3018
0.3246 0.2982
0.3246 0.2982
0.3563 0.2987
0.3223 0.2972
0.3156 0.2929
≤ 0.5
≤ 0.5
#5
0.3140
0.3188
0.3091
0.3091
0.2978
0.3070
0.3223
#6
0.2881
0.2626
0.3113
0.3113
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
66.67%
92
Table F31: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
3x3 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.2873
0.2883
0.2896
0.2896
0.2909
0.2793
#2
13.0920
0.2025
16.5799
16.5800
Training
3.7361
0.6809
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.3543 0.3032
0.4184 0.3587
0.3359 0.3207
0.3359 0.3207
error
IMF
0.3210 0.3022
0.1350 0.3415
≤ 0.5
≤ 0.5
#5
0.3003
0.3218
0.2945
0.2945
c>d
0.3005
0.3085
#6
0.6312
0.9888
0.5000
-0.7561
Success
Rate
83.33%
100.00%
66.67%
66.67%
0.5000
0.5000
66.67%
66.67%
≤ 0.5
> 0.5
Table F32: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
3x3 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.2981
0.2934
0.2916
0.2910
0.2914
0.2933
0.2660
#2
-4.3122
1.1249
0.6980
0.8780
0.8759
0.4190
1.2177
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4328 0.3049
0.3846 0.3643
0.3200 0.3478
0.3304 0.3470
0.3102 0.2893
0.2895 0.2877
0.3529 0.2978
≤ 0.5
≤ 0.5
#5
0.2944
0.3103
0.3224
0.3225
0.2969
0.2940
0.3003
#6
0.7486
0.2534
0.5000
12.9870
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
100.00%
66.67%
66.67%
83.33%
66.67%
83.33%
66.67%
93
Table F33: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
3x4 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.2906
0.2846
0.3065
0.3065
0.2976
0.2560
#2
-5.1489
-0.8301
-7.4646
-7.4647
Training
-43.8786
2.3175
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.3922 0.2928
0.5176 0.2940
-0.7348 0.3700
-0.7348 0.3700
error
IMF
0.2986 0.3019
-0.4917 0.3086
≤ 0.5
≤ 0.5
#5
0.3271
0.3080
0.3133
0.3133
c>d
0.2917
0.3025
#6
0.7226
0.1394
0.5000
39.1938
Success
Rate
100.00%
66.67%
83.33%
100.00%
0.5000
0.5000
83.33%
66.67%
≤ 0.5
> 0.5
Table F34: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
3x4 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.2550
0.2707
0.2907
0.2907
0.2896
0.2890
0.2154
#2
-7.0764
-1.0894
-3.3032
-3.3031
-3.4432
-8.9100
-0.9554
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4537 0.3056
0.3698 0.2968
1.4800 0.3134
1.4798 0.3134
0.1362 0.3237
0.3043 0.3363
0.0126 0.2961
≤ 0.5
≤ 0.5
#5
0.2925
0.2937
0.3033
0.3033
0.2989
0.2993
0.2970
#6
4.0247
-1.1663
0.5000
3.4187
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
100.00%
83.33%
66.67%
83.33%
83.33%
83.33%
83.33%
94
Table F35: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
4x2 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.2753
0.2587
0.3164
0.3164
0.3099
0.3133
#2
0.1534
0.1282
0.3017
0.3017
0.2472
0.2650
Training
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4392 0.3690
0.4077 0.6339
0.4037 0.2215
0.4037 0.2215
0.4174 0.1902
0.3784 0.2380
error
IMF
≤ 0.5
≤ 0.5
#5
0.3273
0.3242
0.3189
0.3189
0.3593
0.3260
a>b
#6
-2.4124
-6.7138
2.0829
2.0829
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
83.33%
66.67%
100.00%
100.00%
83.33%
83.33%
Table F36: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
4x2 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.2669
0.2598
0.3077
0.3077
0.3079
0.3119
0.5210
#2
0.0401
-0.1672
0.2622
0.2622
0.2099
0.2686
-0.2712
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.4453 0.1830
0.4498 0.2176
0.4148 0.1419
0.4148 0.1419
0.4667 0.1231
0.4161 0.1586
0.2563 0.1942
≤ 0.5
≤ 0.5
#5
0.3452
0.3485
0.3477
0.3477
0.3753
0.3565
0.3466
#6
0.3499
0.2966
0.5590
0.5590
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
83.33%
83.33%
100.00%
100.00%
83.33%
83.33%
66.67%
95
Table F37: ANFIS On-line Classification Outputs
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
5x2 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.2962
0.2840
0.2955
0.2955
0.2933
0.3025
0.2631
#2
-0.1371
-3.0408
-0.5366
-0.5366
-0.1886
-0.9891
-1.5519
≤ 0.5
≤ 0.5
Run Number
#3
#4
-0.0257 1.5480
0.1723 3.6901
0.1688 0.4894
0.1688 0.4894
-0.1153 0.0552
-0.1663 0.3904
0.5310 0.6805
≤ 0.5
≤ 0.5
#5
0.6475
0.5820
0.4398
0.4398
0.4819
0.4920
0.5190
#6
14.7470
39.0584
53.0248
53.0248
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
66.67%
66.67%
100.00%
100.00%
83.33%
83.33%
33.33%
Table F38: ANFIS On-line Classification Outputs
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
5x2 (ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Success
Criteria
#1
0.3612
0.3914
0.2909
0.2909
0.3068
0.2974
0.4640
#2
-0.5883
-0.7800
-0.3566
-0.3566
-0.0980
-0.2713
-0.9188
≤ 0.5
≤ 0.5
Run Number
#3
#4
0.1721 0.3356
0.1122 0.3104
0.0979 0.1024
0.0979 0.1024
0.0867 0.1959
0.0448 0.1324
0.2318 0.3318
≤ 0.5
≤ 0.5
#5
0.4513
0.4392
0.3532
0.3532
0.3631
0.3649
0.4448
#6
0.1677
0.0087
0.4929
0.4929
0.5000
0.5000
0.5000
≤ 0.5
> 0.5
Success
Rate
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
83.33%
96
APPENDIX G
CPN On-line Measurement Outputs
97
Table G1: CPN On-line Measurement Outputs (Percent Error)
Inputs x 30 x 1 (CPN)
Number of
Inputs
#1
1
33.72%
2
36.82%
3
54.14%
4
69.29%
5
70.10%
Measured
0.159
Value (mm)
Run Number
#2
#3
#4
-13.82% -2.94%
4.24%
-16.25% -6.70% -8.69%
-18.20% -23.23% -18.35%
-15.20% -9.94% -13.57%
-17.55% -16.87% -29.26%
0.244
0.242
0.219
#5
-14.38%
-13.43%
-22.59%
-13.43%
-24.12%
Average
Error
13.82%
16.38%
27.30%
24.29%
31.58%
0.231
Table G2: CPN On-line Measurement Outputs (Percent Error)
Inputs x 33 x 1 (CPN)
Number of
Inputs
#1
1
33.72%
2
36.82%
3
54.14%
4
69.29%
5
70.10%
Measured
0.159
Value (mm)
Run Number
#2
#3
#4
-11.58% -2.94% -1.48%
-16.25% 9.89%
-8.69%
-1.18% -20.97% -11.97%
-20.33% -19.68% -13.57%
-12.05% -3.56% -29.26%
0.244
0.242
0.219
#5
-15.09%
-13.43%
-16.55%
-10.84%
-24.12%
Average
Error
12.96%
17.02%
20.96%
26.74%
27.82%
0.231
Table G3: CPN On-line Measurement Outputs (Percent Error)
Inputs x 36 x 1 (CPN)
Number of
Inputs
#1
1
33.72%
2
36.82%
3
45.92%
4
69.29%
5
69.01%
Measured
0.159
Value (mm)
Run Number
#2
#3
#4
-14.74% -2.94%
4.24%
-16.25% 9.89%
-8.69%
-1.18% -20.97% -8.69%
-17.55% -16.87% 28.77%
-15.20% -9.94% -29.26%
0.244
0.242
0.219
#5
-13.49%
-13.43%
-13.43%
-10.84%
-26.68%
0.231
Average
Error
13.83%
17.02%
18.04%
28.66%
30.02%
98
Table G4: CPN On-line Measurement Outputs (Percent Error)
Inputs x 39 x 1 (CPN)
Number of
Inputs
#1
1
33.72%
2
36.82%
3
-1.44%
4
69.01%
5
70.14%
Measured
0.159
Value (mm)
Run Number
#2
#3
#4
-13.82% 2.76% -3.98%
-18.36% 9.89% -8.69%
-18.20% -19.88% -8.69%
-15.20% -9.94% 28.77%
-12.05% -14.65% -5.61%
0.244
0.242
0.219
#5
-12.68%
-13.43%
-13.43%
-10.66%
-24.12%
Average
Error
13.39%
17.44%
12.33%
26.72%
25.31%
0.231
Table G5: CPN On-line Measurement Outputs (Percent Error)
Inputs x 42 x 1 (CPN)
Run Number
Number of
Inputs
#1
#2
#3
#4
1
45.74% 10.78%
2.76%
23.43%
2
18.16% -14.43% 4.20%
-8.69%
3
14.08% -18.20% -19.88% -8.69%
4
70.10% -15.20% -14.88% 28.77%
5
67.59% -9.35% -11.89% -30.86%
Measured
0.159
0.244
0.242
0.219
Value (mm)
#5
-13.46%
-13.43%
-13.43%
-10.84%
-26.68%
Average
Error
19.23%
11.78%
14.86%
27.96%
29.27%
0.231
Table G6: CPN On-line Measurement Outputs (Percent Error)
Inputs x 45 x 1 (CPN)
Run Number
Number of
Inputs
#1
#2
#3
#4
1
31.66% 1.21%
2.76% 12.77%
2
29.47% -16.25% -5.77% -8.69%
3
49.97% -9.35%
-9.73% -8.69%
4
69.01% -15.20% -14.88% 28.77%
5
70.10% -12.05% -3.56% -5.61%
Measured
0.159
0.244
0.242
0.219
Value (mm)
#5
-13.46%
-13.43%
-13.43%
-10.66%
-26.68%
0.231
Average
Error
12.37%
14.72%
18.23%
27.70%
23.60%
99
Table G7: CPN On-line Measurement Outputs (Percent Error)
Inputs x 48 x 1 (CPN)
Number of
Inputs
#1
1
31.66%
2
29.47%
3
14.08%
4
69.01%
5
69.01%
Measured
0.159
Value (mm)
Run Number
#2
#3
#4
1.21%
0.95%
12.77%
-16.25% 9.89%
-5.76%
-18.20% -9.73% -8.69%
-15.20% -14.88% 32.19%
-15.20% -14.88% -30.86%
0.244
0.242
0.219
#5
-18.16%
-10.66%
-13.43%
-10.66%
-26.68%
Average
Error
12.95%
14.41%
12.83%
28.39%
31.33%
0.231
Table G8: CPN On-line Measurement Outputs (Percent Error)
Inputs x 51 x 1 (CPN)
Number of
Inputs
#1
1
45.74%
2
18.16%
3
-4.96%
4
69.01%
5
69.01%
Measured
0.159
Value (mm)
Run Number
#2
#3
#4
-14.74% 0.95% 10.33%
-16.25% 9.89% -8.69%
-18.20% -9.73% -8.69%
-15.20% -14.88% 32.19%
-15.20% -3.56% -5.61%
0.244
0.242
0.219
#5
-15.09%
-13.43%
-13.43%
-10.66%
-23.82%
Average
Error
17.37%
13.28%
11.00%
28.39%
23.44%
0.231
Table G9: CPN On-line Measurement Outputs (Percent Error)
Inputs x 54 x 1 (CPN)
Number of
Inputs
#1
1
31.66%
2
29.47%
3
45.92%
4
69.01%
5
70.10%
Measured
0.159
Value (mm)
Run Number
#2
#3
#4
-10.46% 0.95%
-0.24%
-13.90% 8.18%
-8.69%
-18.20% -6.90% -8.69%
-15.20% -30.81% 32.19%
-15.20% -3.56% -30.86%
0.244
0.242
0.219
#5
-18.16%
-13.43%
-13.43%
-9.73%
-26.68%
0.231
Average
Error
12.29%
14.73%
18.63%
31.39%
29.28%
100
Table G10: CPN On-line Measurement Outputs (Percent Error)
Inputs x 57 x 1 (CPN)
Number of
Inputs
#1
1
31.66%
2
18.16%
3
-8.46%
4
69.01%
5
69.01%
Measured
0.159
Value (mm)
Run Number
#2
#3
#4
-10.46% 0.95% -0.24%
-14.43% 8.18% -8.69%
-18.20% -12.60% -8.69%
-15.20% -14.88% 32.19%
-15.20% -3.56% 28.77%
0.244
0.242
0.219
#5
-18.16%
-13.43%
-13.43%
-10.66%
-26.68%
Average
Error
12.29%
12.58%
12.28%
28.39%
28.64%
0.231
Table G11: CPN On-line Measurement Outputs (Percent Error)
Inputs x 60 x 1 (CPN)
Number of
Inputs
#1
1
31.66%
2
29.47%
3
-8.46%
4
69.01%
5
69.01%
Measured
0.159
Value (mm)
Run Number
#2
#3
#4
-13.82% 0.95% -3.98%
-14.43% -5.77% -8.69%
-18.20% -6.90% -8.69%
-9.35% -21.92% 28.77%
-15.20% -3.56% 28.77%
0.244
0.242
0.219
#5
-15.81%
-13.43%
-13.43%
-9.73%
-26.68%
Average
Error
13.24%
14.36%
11.14%
27.76%
28.64%
0.231
Table G12: CPN On-line Measurement Outputs (Percent Error)
Inputs x 63 x 1 (CPN)
Number of
Inputs
#1
1
31.66%
2
18.16%
3
-8.46%
4
69.01%
5
69.01%
Measured
0.159
Value (mm)
Run Number
#2
#3
#4
12.18%
0.95% 24.99%
-14.43% 9.89% -8.69%
-18.20% -12.60% -8.69%
-15.20% -30.81% 32.19%
-15.20% -3.56% -4.43%
0.244
0.242
0.219
#5
-15.81%
-13.43%
-13.43%
-13.43%
-26.68%
0.231
Average
Error
17.12%
12.92%
12.28%
32.13%
23.78%
101
Table G13: CPN On-line Measurement Outputs (Percent Error)
Inputs x 66 x 1 (CPN)
Number of
Inputs
#1
1
31.66%
2
29.47%
3
45.92%
4
68.88%
5
68.88%
Measured
0.159
Value (mm)
Run Number
#2
#3
#4
-13.82% 0.95%
10.33%
-14.43% -5.77% -8.69%
-18.20% -9.73% -8.69%
-20.33% -30.81% 28.77%
-15.20% 7.00% -27.03%
0.244
0.242
0.219
#5
-15.35%
-13.43%
-13.43%
-10.66%
-26.68%
Average
Error
14.42%
14.36%
19.19%
31.89%
28.96%
0.231
Table G14: CPN On-line Measurement Outputs (Percent Error)
Inputs x 69 x 1 (CPN)
Run Number
Number of
Inputs
#1
#2
#3
#4
1
31.66% 0.62%
2.76%
12.11%
2
18.16% -16.25% 8.18%
-8.69%
3
-8.46% -9.35% -9.73% -8.69%
4
69.01% -15.20% -26.36% 28.77%
5
69.01% -15.20% -3.56% -30.86%
Measured
0.159
0.244
0.242
0.219
Value (mm)
#5
-19.85%
-13.43%
-13.43%
-9.73%
-24.12%
0.231
Average
Error
13.40%
12.94%
9.93%
29.81%
28.55%
102
APPENDIX H
ANFIS On-line Measurement Outputs
103
Table H1: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x4
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
39.20%
38.29%
38.45%
38.45%
39.39%
38.53%
43.91%
#2
-9.51%
-9.89%
-9.81%
-9.81%
-9.17%
-9.73%
-10.88%
#3
2.83%
1.15%
1.84%
1.84%
2.88%
2.34%
2.36%
#4
0.85%
0.40%
0.50%
0.49%
1.20%
0.57%
-0.51%
#5
-9.96%
-9.28%
-9.23%
-9.22%
-9.67%
-6.81%
-9.81%
0.159
0.244
0.242
0.219
0.231
Average
Error
12.47%
11.80%
11.97%
11.96%
12.46%
11.60%
13.49%
Table H2: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x4
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
39.37%
37.23%
38.62%
38.62%
38.99%
38.93%
37.30%
#2
-10.00%
-10.82%
-9.80%
-9.80%
-9.43%
-9.47%
-10.53%
#3
6.24%
6.40%
4.50%
4.50%
4.34%
3.47%
8.14%
#4
0.32%
-0.59%
0.55%
0.55%
0.91%
0.87%
-0.32%
#5
-9.52%
-8.92%
-9.52%
-9.52%
-9.96%
-10.65%
-9.39%
0.159
0.244
0.242
0.219
0.231
Average
Error
13.09%
12.79%
12.60%
12.60%
12.73%
12.68%
13.14%
104
Table H3: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x5
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
39.27%
38.74%
38.34%
38.23%
41.01%
39.22%
45.16%
#2
-9.60%
-9.72%
-9.86%
-9.93%
-8.60%
-9.28%
-9.76%
#3
0.73%
0.44%
0.13%
0.06%
0.45%
0.59%
-0.63%
#4
0.76%
0.61%
0.43%
0.35%
2.18%
1.08%
0.77%
#5
-9.81%
-9.69%
-8.85%
-8.64%
-10.84%
-8.54%
-10.11%
0.159
0.244
0.242
0.219
0.231
Average
Error
12.03%
11.84%
11.52%
11.44%
12.62%
11.74%
13.29%
Table H4: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x5
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
40.13%
41.76%
38.24%
32.14%
39.62%
39.31%
39.69%
#2
-9.18%
-8.77%
-9.92%
-13.89%
-9.02%
-9.22%
-8.98%
#3
2.40%
2.98%
1.45%
0.41%
1.07%
1.16%
1.78%
#4
1.23%
1.78%
0.37%
-4.06%
1.37%
1.14%
1.42%
#5
-10.87%
-11.47%
-8.57%
-9.05%
-13.42%
-13.03%
-10.74%
0.159
0.244
0.242
0.219
0.231
Average
Error
12.76%
13.35%
11.71%
11.91%
12.90%
12.77%
12.52%
105
Table H5: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x6
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
38.53%
38.14%
38.14%
38.14%
41.21%
39.56%
43.44%
#2
-9.79%
-9.98%
-9.98%
-9.98%
-9.67%
-9.06%
-9.52%
#3
-0.35%
-0.27%
-0.28%
-0.28%
0.34%
-0.61%
-0.22%
#4
0.52%
0.30%
0.29%
0.29%
0.86%
1.32%
0.99%
#5
-9.17%
-9.07%
-8.28%
-8.28%
-11.60%
-9.69%
-9.89%
0.159
0.244
0.242
0.219
0.231
Average
Error
11.67%
11.55%
11.39%
11.39%
12.74%
12.05%
12.81%
Table H6: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x6
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
38.18%
36.60%
38.11%
38.11%
39.50%
39.25%
38.81%
#2
-10.00%
-10.98%
-10.00%
-10.00%
-9.10%
-9.26%
-9.55%
#3
0.17%
0.00%
0.62%
0.62%
0.83%
0.79%
0.37%
#4
0.27%
-0.82%
0.27%
0.27%
1.28%
1.10%
0.78%
#5
-8.44%
-6.71%
-8.01%
-8.05%
-12.51%
-12.34%
-10.13%
0.159
0.244
0.242
0.219
0.231
Average
Error
11.41%
11.02%
11.40%
11.41%
12.64%
12.55%
11.93%
106
Table H7: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x7
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
30.99%
38.99%
41.04%
41.03%
39.47%
39.64%
41.69%
#2
-9.27%
-10.20%
-10.39%
-10.39%
-10.78%
-9.85%
-10.13%
#3
1.23%
1.61%
0.88%
0.89%
1.08%
0.41%
-0.40%
#4
0.79%
0.13%
0.05%
0.05%
-0.40%
0.62%
0.26%
#5
-10.11%
-9.42%
-9.63%
-9.62%
-9.21%
-8.04%
-9.64%
0.159
0.244
0.242
0.219
0.231
Average
Error
10.48%
12.07%
12.40%
12.40%
12.19%
11.71%
12.42%
Table H8: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x7
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
37.86%
38.81%
37.80%
37.99%
37.92%
37.30%
38.87%
#2
-10.16%
-9.55%
-10.20%
-10.08%
-12.66%
-13.24%
-9.51%
#3
2.73%
2.19%
1.61%
1.61%
1.49%
1.07%
1.94%
#4
0.09%
0.78%
0.05%
0.18%
-2.42%
-2.88%
0.82%
#5
-6.02%
-11.86%
-7.10%
-7.40%
-9.31%
-9.39%
-10.04%
0.159
0.244
0.242
0.219
0.231
Average
Error
11.37%
12.64%
11.35%
11.45%
12.76%
12.78%
12.24%
107
Table H9: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x8
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
35.46%
36.17%
34.64%
34.64%
35.89%
30.03%
27.54%
#2
-9.91%
-10.18%
-9.82%
-9.82%
-10.01%
-8.80%
-9.07%
#3
0.89%
0.90%
-0.35%
-0.35%
-0.15%
-0.10%
-0.80%
#4
0.09%
-0.15%
0.27%
0.27%
0.12%
1.61%
1.11%
#5
-8.80%
-7.82%
-7.10%
-7.10%
-6.69%
-4.85%
-10.13%
0.159
0.244
0.242
0.219
0.231
Average
Error
11.03%
11.04%
10.44%
10.44%
10.57%
9.08%
9.73%
Table H10: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x8
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
39.75%
38.74%
38.55%
38.24%
37.55%
35.60%
39.25%
#2
-9.18%
-9.59%
-9.71%
-9.92%
-11.89%
-12.13%
-9.26%
#3
-0.58%
-1.28%
0.00%
0.00%
0.12%
0.33%
-2.02%
#4
1.23%
0.73%
0.59%
0.37%
-1.69%
-2.10%
1.10%
#5
-11.77%
-11.08%
-7.14%
-6.97%
-7.23%
-7.23%
-10.78%
0.159
0.244
0.242
0.219
0.231
Average
Error
12.50%
12.29%
11.20%
11.10%
11.70%
11.48%
12.48%
108
Table H11: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x9
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
35.15%
36.07%
34.46%
34.46%
33.05%
30.03%
30.04%
#2
-9.80%
-10.36%
-10.07%
-10.06%
-9.30%
-8.81%
-9.11%
#3
-0.44%
-0.09%
-0.40%
-0.40%
0.55%
-0.52%
0.95%
#4
0.26%
-0.28%
0.01%
0.02%
1.06%
1.60%
0.94%
#5
-8.53%
-6.28%
-6.07%
-6.08%
-6.47%
-5.38%
-9.65%
0.159
0.244
0.242
0.219
0.231
Average
Error
10.84%
10.62%
10.20%
10.20%
10.09%
9.27%
10.14%
Table H12: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x9
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
34.91%
34.34%
36.16%
36.16%
35.35%
35.72%
32.26%
#2
-9.71%
-9.80%
-10.00%
-10.00%
-9.51%
-9.51%
-10.12%
#3
0.83%
0.83%
0.74%
0.74%
0.70%
0.79%
0.87%
#4
0.50%
0.37%
0.27%
0.27%
0.82%
0.82%
-0.09%
#5
-8.70%
-9.05%
-7.58%
-7.58%
-9.00%
-8.70%
-9.09%
0.159
0.244
0.242
0.219
0.231
Average
Error
10.93%
10.87%
10.95%
10.95%
11.08%
11.11%
10.49%
109
Table H13: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x10
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
35.47%
34.11%
33.33%
33.31%
32.16%
30.36%
30.30%
#2
-9.76%
-9.57%
-7.69%
-7.68%
-8.13%
-7.25%
-9.11%
#3
1.27%
1.78%
-0.87%
-0.92%
1.55%
2.01%
0.60%
#4
0.43%
0.64%
2.47%
2.53%
2.36%
3.33%
0.93%
#5
-5.69%
-7.02%
-9.57%
-9.59%
-12.63%
-12.59%
-9.89%
0.159
0.244
0.242
0.219
0.231
Average
Error
10.52%
10.62%
10.79%
10.81%
11.37%
11.11%
10.17%
Table H14: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x10
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
35.91%
-1.19%
37.92%
37.92%
36.73%
37.55%
30.63%
#2
-9.84%
-11.15%
-10.37%
-10.37%
-9.59%
-10.00%
-8.85%
#3
1.36%
0.87%
1.69%
1.69%
1.74%
2.02%
0.74%
#4
0.46%
-1.00%
-0.14%
-0.14%
0.73%
0.27%
1.23%
#5
-7.75%
-6.15%
-6.28%
-6.28%
-11.77%
-12.21%
-10.26%
0.159
0.244
0.242
0.219
0.231
Average
Error
11.06%
4.07%
11.28%
11.28%
12.11%
12.41%
10.34%
110
Table H15: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x11
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
4.49%
32.53%
-55.74%
-55.73%
26.50%
36.89%
28.67%
#2
-9.04%
-9.88%
-9.00%
-9.01%
-7.58%
-8.70%
-8.24%
#3
-5.59%
-5.80%
-5.48%
-5.48%
-4.41%
-4.66%
-0.71%
#4
1.14%
0.25%
1.16%
1.16%
2.97%
1.72%
2.23%
#5
-8.39%
-8.68%
-8.59%
-8.58%
-12.16%
-11.55%
-10.04%
0.159
0.244
0.242
0.219
0.231
Average
Error
5.73%
11.43%
15.99%
15.99%
10.72%
12.70%
9.98%
Table H16: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x11
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
8.18%
17.55%
37.55%
37.67%
38.11%
38.55%
31.26%
#2
-10.90%
-9.67%
-10.57%
-10.29%
-9.39%
-9.47%
-8.44%
#3
-2.40%
-2.02%
-2.15%
-2.15%
-2.69%
-2.11%
-0.99%
#4
-0.73%
0.64%
-0.37%
-0.05%
0.96%
0.87%
1.69%
#5
-5.89%
-6.19%
-5.93%
-5.93%
-12.16%
-12.21%
-11.17%
0.159
0.244
0.242
0.219
0.231
Average
Error
5.62%
7.21%
11.31%
11.22%
12.66%
12.64%
10.71%
111
Table H17: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
1x12
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
25.10%
30.59%
23.32%
23.29%
36.10%
31.45%
27.21%
#2
-8.72%
-9.71%
-8.00%
-8.00%
-8.75%
-14.34%
-7.23%
#3
-1.40%
-5.99%
-2.34%
-2.59%
-0.88%
-13.64%
-0.68%
#4
1.26%
0.40%
1.84%
1.84%
1.67%
-4.57%
3.75%
#5
-8.59%
-10.53%
-7.77%
-7.76%
-8.06%
-9.52%
-11.05%
0.159
0.244
0.242
0.219
0.231
Average
Error
9.01%
11.44%
8.65%
8.70%
11.09%
14.70%
9.98%
Table H18: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
1x12
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
35.91%
4.65%
22.96%
-14.47%
37.17%
39.50%
30.44%
#2
-9.71%
-9.51%
-9.55%
-9.75%
-8.85%
-9.06%
-8.07%
#3
-0.25%
-0.17%
-0.41%
-0.41%
-0.41%
-0.37%
-0.58%
#4
0.59%
0.82%
0.78%
0.55%
1.55%
1.32%
2.10%
#5
-11.99%
-12.51%
-5.41%
-4.98%
-11.65%
-11.56%
-11.69%
0.159
0.244
0.242
0.219
0.231
Average
Error
11.69%
5.53%
7.82%
6.03%
11.93%
12.36%
10.58%
112
Table H19: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
2x2
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
36.22%
32.19%
34.45%
34.45%
34.60%
34.42%
29.89%
#2
-23.20%
-13.00%
-12.38%
-12.38%
-12.29%
-12.41%
-2.02%
#3
-6.82%
2.87%
5.68%
5.68%
5.59%
6.26%
-0.38%
#4
-19.46%
-18.39%
-17.00%
-17.00%
-18.07%
-17.01%
-12.67%
#5
-21.44%
-20.15%
-21.23%
-21.23%
-22.32%
-21.32%
-16.04%
0.159
0.244
0.242
0.219
0.231
Average
Error
21.43%
17.32%
18.15%
18.15%
18.57%
18.28%
12.20%
Table H20: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
2x2
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
36.92%
37.92%
33.77%
33.77%
34.40%
34.21%
-14.59%
#2
-46.39%
-44.92%
-12.75%
-12.75%
-12.42%
-12.54%
-28.16%
#3
-10.04%
-9.38%
5.37%
5.37%
4.83%
5.12%
-9.50%
#4
-18.17%
-13.47%
-19.27%
-19.27%
-19.91%
-19.13%
-13.52%
#5
-21.69%
-17.97%
-23.20%
-23.20%
-23.90%
-23.29%
-15.02%
0.159
0.244
0.242
0.219
0.231
Average
Error
26.64%
24.73%
18.87%
18.87%
19.09%
18.86%
16.16%
113
Table H21: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
2x3
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
40.25%
43.08%
39.88%
39.88%
Training
33.15%
43.14%
#2
-9.19%
-8.07%
53.90%
53.90%
error
-2.19%
-6.92%
#3
-1.61%
-0.31%
-4.23%
-4.23%
IMF
-9.94%
2.60%
#4
-15.02%
-35.44%
-8.85%
-8.85%
b>c
-4.61%
-23.80%
#5
-12.40%
-14.39%
-14.02%
-14.02%
0.159
0.244
0.242
0.219
0.231
-15.76%
-19.94%
Average
Error
15.69%
20.26%
24.18%
24.18%
13.13%
19.28%
Table H22: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
2x3
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
39.25%
36.86%
35.47%
37.61%
38.55%
36.42%
43.14%
#2
-8.81%
-10.82%
-11.72%
-10.37%
-7.91%
-11.11%
-4.67%
#3
0.25%
4.50%
-3.47%
-3.31%
-0.91%
-1.40%
3.31%
#4
-17.99%
-12.37%
-4.02%
-10.55%
-15.75%
-8.63%
-34.61%
#5
-15.32%
-14.63%
-20.78%
-14.81%
-10.39%
-9.83%
-18.48%
0.159
0.244
0.242
0.219
0.231
Average
Error
16.32%
15.84%
15.09%
15.33%
14.70%
13.48%
20.84%
114
Table H23: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
2x4
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
40.54%
52.74%
45.44%
45.44%
45.44%
47.32%
52.03%
#2
13.64%
13.50%
9.89%
9.89%
22.24%
18.62%
-12.83%
#3
-2.09%
3.48%
-11.98%
-11.98%
-1.12%
-12.63%
3.53%
#4
-41.10%
-18.61%
-45.35%
-45.35%
-19.13%
-22.11%
-24.50%
#5
-18.40%
-16.33%
-16.47%
-16.47%
-22.56%
-24.44%
-11.40%
0.159
0.244
0.242
0.219
0.231
Average
Error
23.15%
20.93%
25.83%
25.83%
22.10%
25.02%
20.86%
Table H24: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
2x4
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
56.10%
47.67%
48.55%
48.55%
48.87%
44.03%
56.79%
#2
2.17%
10.94%
14.75%
14.75%
1.19%
19.96%
-18.28%
#3
-0.54%
1.49%
-3.31%
-3.31%
0.21%
-0.37%
4.09%
#4
-21.10%
-37.85%
-4.79%
-4.79%
-18.58%
-18.58%
-1.05%
#5
-12.47%
-13.68%
-19.31%
-19.31%
-21.86%
-22.25%
-15.93%
0.159
0.244
0.242
0.219
0.231
Average
Error
18.47%
22.33%
18.14%
18.14%
18.14%
21.04%
19.23%
115
Table H25: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
2x5
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
59.64%
53.12%
57.82%
57.77%
Training
62.53%
Training
#2
-2.60%
18.58%
19.03%
19.34%
error
43.39%
error
#3
2.67%
4.35%
0.54%
0.47%
IMF
2.34%
IMF
#4
50.53%
-7.44%
-15.73%
-15.72%
a>b
-31.33%
a>b
#5
-13.51%
-12.66%
-11.15%
-11.16%
0.159
0.244
0.242
0.219
0.231
-17.55%
Average
Error
25.79%
19.23%
20.85%
20.89%
31.43%
Table H26: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
2x5
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
60.69%
52.83%
36.16%
36.16%
51.38%
52.45%
56.10%
#2
15.70%
22.91%
-14.71%
-14.96%
65.94%
83.32%
-10.12%
#3
-6.61%
-3.72%
3.55%
3.55%
-9.55%
-7.69%
-10.33%
#4
-39.59%
-22.28%
-19.41%
-18.08%
-38.72%
-36.26%
-42.56%
#5
-13.20%
-14.11%
-25.19%
-24.55%
-14.76%
-15.54%
-13.16%
0.159
0.244
0.242
0.219
0.231
Average
Error
27.16%
23.17%
19.81%
19.46%
36.07%
39.05%
26.45%
116
Table H27: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
2x6
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
#2
#3
#4
41.45% 172.67% 10.19% 53.90%
44.46%
22.24% 216.09% 24.28%
61.37% 685.93% -1.53% -20.00%
60.30% -411.75% 10.99% 92.66%
Training
error
IMF
a>b
113.96% -36.19%
2.34% -21.86%
51.45% 133.22% 22.95% -36.68%
0.159
0.244
0.242
0.219
Average
Error
#5
-15.91%
-14.15%
-11.76%
-17.66%
58.82%
64.24%
156.12%
118.67%
-12.84%
-11.55%
37.44%
51.17%
0.231
Table H28: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
2x6
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
#2
#3
26.98% 135.41% -38.14%
40.06% 19.22%
26.61%
38.62% 70.04%
-1.86%
38.62% 70.04%
-1.86%
49.56% -100.00% -96.57%
49.37% -100.00% -236.61%
63.71% 52.25%
10.54%
0.159
0.244
0.242
#4
2.01%
-8.13%
-52.37%
-52.37%
-14.89%
-15.34%
-37.17%
#5
-15.54%
-14.76%
-12.86%
-12.86%
-11.73%
-13.33%
-16.58%
0.219
0.231
Average
Error
43.62%
21.76%
35.15%
35.15%
54.55%
82.93%
36.05%
117
Table H29: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
3x2
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
41.00%
44.82%
36.32%
36.32%
37.05%
35.94%
49.24%
#2
-1.13%
3.25%
-5.08%
-5.08%
-2.73%
-5.23%
9.80%
#3
-9.79%
-15.47%
1.73%
1.73%
-5.63%
1.99%
-17.94%
#4
-20.81%
-22.84%
-15.86%
-15.86%
-16.80%
-16.71%
-29.18%
#5
-26.32%
-25.20%
-22.74%
-22.74%
-22.68%
-22.20%
-25.57%
0.159
0.244
0.242
0.219
0.231
Average
Error
19.81%
22.32%
16.35%
16.35%
16.98%
16.41%
26.35%
Table H30: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
3x2
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
39.06%
41.70%
35.28%
35.28%
38.05%
36.92%
44.97%
#2
-2.91%
-2.83%
-7.30%
-7.30%
-3.24%
-5.33%
0.82%
#3
-6.82%
-9.88%
7.02%
7.02%
-11.28%
2.02%
-14.26%
#4
3.88%
-7.63%
-16.30%
-16.30%
-17.81%
-15.75%
-14.66%
#5
-23.64%
-23.81%
-18.10%
-18.10%
-23.85%
-21.82%
-19.05%
0.159
0.244
0.242
0.219
0.231
Average
Error
15.26%
17.17%
16.80%
16.80%
18.85%
16.37%
18.75%
118
Table H31: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
3x3
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
#2
41.32% -184.52%
45.99% 69.10%
37.07% -45.17%
37.07% -45.18%
37.64% -20.71%
34.73% -36.28%
48.31%
5.80%
0.159
0.244
#3
-17.20%
-18.27%
-10.52%
-10.52%
-15.67%
-17.93%
-11.62%
#4
-29.53%
-11.00%
-17.42%
-17.42%
-15.36%
-15.39%
-18.59%
#5
-4.81%
-12.37%
-12.44%
-12.44%
-11.10%
-11.31%
-12.47%
0.242
0.219
0.231
Average
Error
55.48%
31.35%
24.52%
24.53%
20.10%
23.13%
19.36%
Table H32: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
3x3
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
#2
41.95% 15.04%
46.16%
6.93%
38.05% -26.84%
38.05% -26.84%
39.31% -206.56%
34.40% 18.16%
40.75% 12.34%
0.159
0.244
#3
-13.97%
-10.29%
-15.45%
-15.45%
1.57%
-21.61%
-10.21%
0.242
#4
#5
-64.02% -7.01%
-29.09% -9.74%
-55.39% -8.57%
-55.39% -8.57%
-20.91% -4.33%
-39.00% -5.28%
-35.07% -12.77%
0.219
0.231
Average
Error
28.40%
20.44%
28.86%
28.86%
54.54%
23.69%
22.23%
119
Table H33: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
3x4
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
#2
#3
#4
#5
38.58% 91.65% 10.00% -27.66% -7.23%
42.76% 177.15% -8.36%
-7.85% -11.75%
25.79% 446.70% -2.33%
1.97% -13.54%
25.79% 446.69% -2.33%
1.97% -13.54%
Training
error
IMF
a>b
39.62% 47.03%
-6.52% -30.00% -23.15%
50.37% -76.74% 115.48% -6.81% -7.16%
0.159
0.244
0.242
0.219
Average
Error
35.02%
49.57%
98.07%
98.06%
29.26%
51.31%
0.231
Table H34: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
3x4
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
15.28%
37.67%
35.22%
35.22%
37.99%
39.81%
82.01%
#2
247.05%
-1138.57%
170.98%
170.98%
-61.68%
-365.82%
-191.43%
#3
-62.69%
-19.96%
20.08%
20.08%
5.29%
4.92%
390.58%
#4
8.40%
-21.37%
14.11%
14.11%
-29.95%
-28.86%
1.60%
#5
-3.33%
-2.73%
-16.28%
-16.28%
-22.94%
-22.47%
-14.81%
0.159
0.244
0.242
0.219
0.231
Average
Error
67.35%
244.06%
51.33%
51.33%
31.57%
92.37%
136.09%
120
Table H35: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
4x2
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
58.77%
61.74%
47.48%
47.48%
48.11%
47.41%
75.12%
#2
-18.54%
-10.40%
-18.34%
-18.34%
-14.59%
-16.56%
130.82%
#3
-33.75%
-27.70%
-92.95%
-92.95%
-39.92%
-82.13%
-35.89%
#4
36.85%
36.12%
31.17%
31.17%
39.73%
35.16%
37.36%
#5
-1.45%
-0.19%
-0.03%
-0.03%
-5.21%
-7.32%
-18.80%
0.159
0.244
0.242
0.219
0.231
Average
Error
29.87%
27.23%
37.99%
37.99%
29.51%
37.72%
59.60%
Table H36: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
4x2
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
59.18%
63.27%
51.32%
51.32%
53.33%
51.19%
57.67%
#2
14.06%
-4.75%
-5.53%
-5.53%
4.14%
-5.70%
29.34%
#3
-44.38%
-19.67%
-58.18%
-58.18%
-25.00%
-71.53%
-16.74%
#4
32.79%
-67.95%
32.15%
32.15%
22.51%
25.94%
49.41%
#5
-28.87%
-28.14%
-29.09%
-29.09%
-24.24%
-25.37%
-21.30%
0.159
0.244
0.242
0.219
0.231
Average
Error
35.86%
36.76%
35.25%
35.25%
25.85%
35.94%
34.89%
121
Table H37: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with probabilistic ‘and’ used in all the fuzzy rules)
5x2
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
71.64%
75.08%
72.57%
72.57%
62.59%
59.49%
85.60%
#2
-8.37%
3.65%
-10.81%
-10.81%
-28.08%
-18.82%
73.13%
0.159
0.244
Average
Error
#3
#4
#5
-15.45% 254.02% 187.77% 107.45%
-41.96% 180.65% 137.06% 87.68%
207.76% 71.75% -3.10% 73.20%
207.76% 71.75% -3.10% 73.20%
-190.90% 39.74% 33.22% 70.91%
220.39% 63.00% 11.91% 74.72%
-65.64% 302.30% 159.49% 137.23%
0.242
0.219
0.231
Table H38: ANFIS On-line Measurement Outputs (Percent Error)
(ANFIS with fuzzy ‘and’ used in all the fuzzy rules)
5x2
(ANFIS)
IMF Type
gbell
gauss
dsig
psig
trap
pi
tri
Measured
Value (mm)
Run Number
#1
76.67%
70.13%
76.60%
76.60%
53.21%
60.06%
68.11%
#2
-23.85%
4.88%
-34.34%
-34.34%
-15.61%
-37.50%
43.81%
#3
-226.78%
-152.23%
1069.63%
1069.63%
-138.18%
-111.90%
-61.78%
0.159
0.244
0.242
Average
Error
#4
#5
70.23% 5.11% 66.21%
62.19% 2.60% 44.90%
42.79% 17.49% 236.35%
42.79% 17.49% 236.35%
26.89% 12.47% 41.13%
26.30% 3.94% 36.72%
59.04% -4.11% 34.57%
0.219
0.231
122
REFERENCES
[1] Jardine, A. K. S., Lin, D., and Banjevic, D., 2006, “A Review on Machinery
Diagnostics and Prognostics Implementing Condition-Based Maintenance,”
Mechanical Systems and Signal Processing, Vol. 20, No. 7, pp. 1483–1510.
[2] Roth, J., Djurdjanovic, D., Yang, X., Mears, and L., Kurfess, T., 2010, “Quality and
Inspection of Machining Operations: Tool Condition Monitoring,” Journal of
Manufacturing Science and Engineering, Vol. 132, No. 4, pp. 1–16.
[3] Sick, B., 2002, “On-Line and Indirect Tool Wear Monitoring in turning With
Artificial Neural Networks: A Review of More Than a Decade of Research,” Mechanical
Systems and Signal Processing, Vol. 16, No. 4, pp. 487–546.
[4] Worden, K., Staszewski, W., and Hensman, J., 2011, “Natural computing for
mechanical systems research: A tutorial overview,” Mechanical Systems and Signal
Processing, Vol. 25, pp. 4–111.
[5] Teti, R., Jemielniak, K., O’Donnell G., and Dornfeld, D., 2010, “Advanced
monitoring of machining operations,” CIRP Annals - Manufacturing Technology, Vol.
59, pp. 717–739.
[6] Abellan-Nebot, J., and Subirón, F., 2010, “A review of machining monitoring systems
based on artificial intelligence process models,” International Journal of Advanced
Manufacturing Technology, Vol. 47, pp. 237–257.
[7] Liu, T. I., Kumagai, A., Wang, Y.C., Song, S. D., Fu, Z., and Lee, J., 2010, “On-line
monitoring of boring tools for control of boring operations,” Robotics and ComputerIntegrated Manufacturing, Vol. 26, pp. 230–239.
[8] Liu, T. I., Lee, J., and Gill, G. S., 2008, “On-line Monitoring and Diagnosis of
Tapping Process Using Neuro Fuzzy Systems,” Proceedings of the 3rd IEEE International
Conference on Industrial Electronics and Applications, Singapore, pp. 1–5.
[9] Samanta, B., and Nataraj, C., 2008, “Prognostics of machine condition using soft
computing,” Robotics and Computer-Integrated Manufacturing, Vol. 24, pp. 816–823.
[10] Sharma, V.S., Sharma, S. K., and Sharma, A. K., 2008, “Cutting tool wear
estimation for turning,” Journal of Intelligent Manufacturing, Vol. 19, pp. 99–108.
[11] Sharma, V. S., Sharma, S. K., and Sharma A. K., 2007, “An approach for condition
monitoring of a turning tool,” Proceedings of the Institution of Mechanical Engineers,
Part B: Journal of Engineering Manufacture, Vol. 221, pp. 635–-648.
123
[12] Rehorn, A. G., Jiang, J., and Orban, P. E., 2005, “State-of-the-Art Methods and
Results in Tool Condition Monitoring: A Review,” The International Journal of
Advanced Manufacturing Technology, Vol. 26, No. 7–8, pp. 693–710.
[13] Oraby, S. E., Al-Modhuf, A. F., and Hayhurst, D. R., 2005, “A Diagnostic
Approach for Turning Tool Based on the Dynamic Force Signals,” Journal of
Manufacturing Science and Engineering, Vol. 127, No. 3, pp. 463–475.
[14] Liang S. Y., Hecker R. L., Landers R. G., 2004, “Machining process
monitoring and control: the state-of-the-art,” Journal of Manufacturing Science and
Engineering, Vol. 126, pp.297–310
[15] Chen, K.Y., Lim, C. P., and Lai, W. K., 2005, “Application of Neural Fuzzy System
with Rule Extraction to Fault Detection and Diagnosis,” Journal of Intelligent
Manufacturing, Vol. 16, No. 6, pp. 679–691.
[16] Scheffer, C., Heyns, P. S., 2004, “An industrial tool wear monitoring system for
interrupted turning,” Mechanical Systems and Signal Processing, Vol. 18, pp. 1219–
1242.
[17] Rao, S., and Srikant, R., 2004, “Tool wear monitoring—an intelligent approach,”
Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering
Manufacture, Vol. 218, No. 8, pp. 905–912.
[18] Babuška, R., Verbruggen, H., 2003, “Neuro-fuzzy methods for nonlinear system
identification,” Annual Reviews in Control, Vol. 27, pp. 73–85.
[19] Balazinski, M., Czogata, E., Jemielniak, K., and Leski, J., 2002, “Tool Condition
Monitoring Using Artificial Intelligence Methods,” Engineering Applications of
Artificial Intelligence, Vol. 15, No. 1, pp. 73-80.
[20] Stein, J. L., and Huh, k., 2002, “Monitoring Cutting Forces in Turning: A ModelBased Approach,” ASME journal of Manufacturing Science and Engineering, Vol. 124,
pp. 26-31.
[21] Mesina, O., Langari, R., 2001, “A Neuro-Fuzzy System for Tool Condition
Monitoring in Metal Cutting,” Journal of Manufacturing Science and Engineering, Vol.
123, No. 2, pp. 312–318.
124
[22] Vora, N., Tambe, S. S., and Kulkarni, B. D. 1997, “Counterpropagation Neural
Networks for Fault Detection and Diagnosis,” Computers & Chemical Engineering, Vol.
21, No. 2, pp. 177–185.
[23] ISO 3685:1993, “Tool-life testing with single-point turning tool.”
[24] Sun, J., Hong, G. S., Wong, Y. S., Rahman, M., and Wang, Z. G., 2006, “Effective
training data selection in tool condition monitoring system,” International Journal of
Machine Tools & Manufacture, Vol. 46, pp. 218–224.
[25] Guyon, I., and Elisseeff, A., 2003, “An introduction to variable and feature
selection,” Journal of Machine Learning Research, Vol. 3, pp. 1157–1182.
[26] Cover, T. M., 1965, “Geometrical and Statistical Properties of Systems of Linear
Inequalities with Applications in Pattern Recognition,” IEEE Transaction on Computers,
Vol. 14, pp. 326–334.
[27] Whitney, A., 1971, “A direct method of non-parametric measurement selection,”
IEEE Transaction on Computers, Vol. 20, No. 9, pp. 1100–1103.
[28] HNC, Inc, 1991, HNC ExploreNet 3000, Release 2.0.
[29] The Math Works, Inc., 2011, Fuzzy Logic Toolbox: “ANFIS and the ANFIS Editor
GUI,” http://www.mathworks.com/help/toolbox/fuzzy/fp715dup12.html.