C.2 - Computational Techniques in Stat. Anal. & Exploitation of CNC Mach. Exp. Data - IGI Global, 2012

Computational
Methods for Optimizing

Manufacturing
Technology:
Models and Techniques
J. Paulo Davim
University of Aveiro, Portugal
Managing Director: Lindsay Johnston
Senior Editorial Director: Heather Probst
Book Production Manager: Sean Woznicki
Development Manager: Joel Gamon
Development Editor: Hannah Abelbeck
Acquisitions Editor: Erika Gallagher
Typesetter: Adrienne Freeland, Russell A. Spangler
Cover Design: Nick Newcomer, Lisandro Gonzalez
Published in the United States of America by

Engineering Science Reference (an imprint of IGI Global)
701 E. Chocolate Avenue
Hershey PA 17033
Tel: 717-533-8845
Fax: 717-533-8661
E-mail: [email protected]
Web site: http://www.igi-global.com
Copyright © 2012 by IGI Global. All rights reserved. No part of this publication may be reproduced, stored or distributed in
any form or by any means, electronic or mechanical, including photocopying, without written permission from the publisher.
Product or company names used in this set are for identification purposes only. Inclusion of the names of the products or
companies does not indicate a claim of ownership by IGI Global of the trademark or registered trademark.
Library of Congress Cataloging-in-Publication Data

Computational methods for optimizing manufacturing technology: models and techniques / J. Paulo Davim, editor.
p. cm.
Includes bibliographical references and index.
ISBN 978-1-4666-0128-4 (hardcover) -- ISBN 978-1-4666-0129-1 (ebook) -- ISBN 978-1-4666-0130-7 (print & perpetual
access) 1. Manufacturing processes--Mathematical models. I. Davim, J. Paulo.
TS183.C6488 2012
658.5--dc23
2011042009
British Cataloguing in Publication Data

A Cataloguing in Publication record for this book is available from the British Library.
All work contributed to this book is new, previously-unpublished material. The views expressed in this book are those of the
authors, but not necessarily of the publisher.
111
Chapter 5
Computational Techniques
in Statistical Analysis and
Exploitation of CNC Machining
Experimental Data
N. A. Fountas
School of Pedagogical & Technological Education (ASPETE), Greece
A. A. Krimpenis
N. M. Vaxevanidis
ABSTRACT
Extracting CNC machining data on- or off-line demands thorough and careful planning. Exploitation
of this data can be carried out by statistical methods, in order to obtain the most influential parameters
along with their respective level of significance. However, significance of machining parameters var-
ies according to the posed Quality Characteristics at each machining phase. In actual experiments,
measuring devices and assemblies are used, and data is recorded in computer archives. To shorten the
production time and cost, machining processes are planned on CAM software, especially when complex
part geometries, such as sculptured surfaces, are involved. Hence, planning machining experiments using
CAM software modules is an efficient approach for experimentation on the actual CNC machine tools.
Data extraction and statistical analysis methodologies are presented along with respective machining
experimental examples.
DOI: 10.4018/978-1-4666-0128-4.ch005
Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Computational Techniques in Statistical Analysis and Exploitation of CNC Machining Experimental Data
1. INTRODUCTION • The data category should be specifically

determined, i.e. whether data is a result
In today’s manufacturing environments, demands of observation or of actual experiments.
for high quality and short production time prac- Moreover, it should be clarified whether
tically drive both small/middle and large-scale data sets are “convenient” samples or are
enterprises to implement advanced production obtained by a designed sample survey. The
techniques and integrated manufacturing systems. method of data collection has crucial im-
Soft/technical computing and intelligent systems pact on results and conclusions.
are now state-of-the-art modules and key elements • It should be checked whether there is non-
for proper and reliable preparation of manufactur- response on the data. The data that cannot
ing processes. Optimization methods like Genetic be observed may be just as important as the
Algorithms, Simulated Annealing, Evolutionary observed ones.
programming, Tabu Search and so on, are very • Investigation for missing values. This is a
important elements for developing a modern and common problem that is troublesome and
integrated manufacturing environment. These time consuming to deal with.
strategies need experimental data sets and respec- • Quantitative representation and data
tive statistical analysis, in order to correspond encoding.
properly and generate accurate outputs. The main • Data normalization (see also Section 2.1.2)
thrust of this chapter is an overview of the statisti- and measurement units.
cal techniques with which it is possible to obtain, • Performance of data integrity checks and
analyse and evaluate machining experimental data estimation of errors during data entry.
used as inputs in Artificial Intelligence methods
for system and process optimization. 2.1 Experimental Procedure
Each experimental design on computer systems

2. BASICS OF DATA GATHERING should be enhanced/validated with results of
AND PROCESSING IN experiments performed on CNC machine tools.
CNC MACHINING This means that evaluation and verification of ex-
perimental designs conducted on computer-aided
Several statistical techniques are used in order to systems should be supported by corresponding
summon and evaluate the appropriate information results generated from real-life experiments.
of manufacturing processes that are controlled Actual results and data from manufacturing ex-
in terms of quality characteristics and responses. perimentation should be used as benchmarks, so
Parameter values are scouted in a pre-determined as to qualitatively assess the effectiveness of com-
value ranges that users specify. In terms of reli- putational estimations. However, this statement
ability and productivity, obtained data sets must doesn’t undermine the need of computer system
conform to predefined technological and economi- for this purpose, since they offer an efficient low
cal constraints. cost alternative.
Before one proceeds on performing experi- Different measuring devices and instruments
mental design, it is essential to understand the are used both in academic and manufacturing
philosophy of collecting data in a systematic laboratories, as well as in industrial shop-floors.
manner. Some of the basic issues, one should see This section describes some of the most com-
to, are the following (Faraway J., 2002): monly used measuring devices and assemblies that
are implemented on real tryouts. These devices
112
are presented, briefly, below, according to their 2. Profilometers and Surface Roughness Testers
specifications for measuring particular machin-
ing objectives. Surface roughness evaluation is very important
for many engineering areas such as friction and
2.1.1 Measuring Instruments wear, contact deformation, dimensional tolerances
and Assemblies and positional accuracy. Thus, surface roughness
has been the subject of experimental and theoreti-
1. Three-component Cutting Force Measuring cal investigations for many decades (Gadelmawla
Devices / Dynamometers et al., 2002). The real surface geometry resulted
from a manufacturing process is so complicated
A widely used measuring device that collects that a finite number of parameters cannot provide
machining data – cutting forces yielded during a full description. With the increase of parameters
metal cutting processes – is the 3-component used a more accurate description can be obtained
cutting force piezoelectric dynamometer. This (Petropoulos et al., 2006).
device performs very efficiently on a wide range Surface finish may be measured in two ways:
of metals and composites cutting experiments in contact and non-contact methods. Contact methods
almost every machine tool category. Influence of involve dragging a measurement stylus across the
material, speed, feed, and rake angle of tools can surface; these instruments are called profilometers.
also be investigated. Therefore, optimum cutting Profilometers and surface roughness testers are
parameters for a material can be determined. The instruments that enable data collection as far as the
apparatus is able to measure the axial force, the geometric profile and the surface roughness are
radial force and the tangential force on tools dur- concerned. The stylus (typically made of diamond)
ing cutting. The lathe tool holder is fitted with an traverses across the specimen, while a piezoelectric
adapter and can be easily mounted on lathes. In the pickup records all vertical movements. Peaks and
case of milling machines, the cutting force mea- valleys of the surface are recorded and converted
suring device is placed immediately beneath the into a value of a standardized parameter. Surface
machined part. The respective type of measuring roughness parameters are categorized, usually,
device can be equipped with strain gauges for each into three groups. These groups are defined as
force component. The holder gets connected to a amplitude, spacing and hybrid parameters. More
digital strain gauge amplifier. The strain gauges information about technical characteristics of
are read out by the amplifier which indicates the profilometers and surface roughness testers can be
forces on digital displays. found in publications by vendors of such devices.
More information and technical specifications For the correlation of operating parameters with
about cutting force measuring devices can be found the resulted surface roughness, from the extensive
in brochures and leaflets from well known vendors bibliography existed, recent works by Davim et
of measuring equipment (e.g. www.kistler.com, al., 2008, Galanis & Manolakos, 2009 and Vax-
www.tesabs.co.uk). evanidis et al., 2011, can be consulted.
Cutting force analysis together with tool
life and the resulted surface roughness are vital 2.1.2 Data Normalization and
components for evaluating machinability and for Weighing Coefficients
studying the various characteristics of a machining
process (Davim et al., 2009, Denkena & Köhler, In manufacturing, values and significance level
2010, Schulze et al., 2011). of quality characteristics are usually different,
depending upon the nature of the process. Rough
113
machining process, for example, strongly relies on A typical example of data normalization is
high productivity, thus machining time is a quality presented in the experiment presented in Section
goal that plays a vital role in the process, much 3.1.2, where the quality targets of a machining
more than others, such as surface quality. On the process are remaining volume and machining
other hand, finishing operations’ main goal is final time for a roughing operation. Since high material
product quality, so other quality goals are second- removal rates are generally desirable in roughing
ary. Values of Quality Characteristics (see Section operations, machining time is more significant
2.4) vary in order and magnitude. To that extend, than remaining volume. Hence, for roughing time
it is mentioned that machining time usually counts the weight coefficient is for example adjusted to
from a few thousand to a few hundred thousand 70% (or 0.7) and the remaining volume yields the
seconds, while surface roughness is measured in value of 30% (or 0.3) in the Quality Characteristic
a few microns and total remaining volume after function (see Section 3.1.2).
roughing may be in the order or tens of millions
cubic millimetres. In that sense, a normalization 2.2 Design of Experiments
procedure has to be carried out when more than and Orthogonal Arrays
one Quality Characteristic is put together in the
same measuring function. Summing values of The term “experiment” is defined as the systematic
different order is an inherently biased in favour process performed under controlled conditions,
of the one with the biggest magnitude. Obviously, so as to discover or illustrate an unknown effect,
the results can be quite disorienting as far as the to test and set up a hypothesis. When analysing
problem overall response is concerned. a process, experiments are often used to validate
Significance level of a quality characteristic which process inputs have a significant impact
can be determined according to user’s demands on the process result, and what the target level
by weighing coefficients. A data normalization of those inputs should be, so as to achieve the
strategy is proposed and discussed. This strategy desired result. Therefore, experimental data can
is consists of linear combination between quality be collected with many different ways. Taguchi’s
targets and their respective weight coefficients, Design of Experiments methodology with the
according to their importance in the studied prob- heavy use of orthogonal arrays may be utilized
lem. Linear combination of quality characteristics in obtaining dependable information concerning
offers no obvious bias, provided that values are variable values in taking engineering decisions
normalised beforehand. This combination can be with minimum number of experiments, thus
described as follows: minimum cost.
Let X1, X2,......Xi be the given quality charac-
teristics (as in multi-criterion optimization) and 2.2.1 Design of Experiments (DoE)
w1, w2,......wj their respective weight coefficients.
Practically, a weight coefficient’s value –if it ex- Design of Experiments (DoE) is a statistical
pressed as a percentage– represents the respective technique introduced by Fisher in England in
quality characteristic’s significance in the problem. the 1920’s to simultaneously study the effects
Significance of quality targets in the problem is of multiple variables on a problem outcome to
applied through wi values in the Quality Charac- give the best results. The methodology was first
teristic function: implemented for handling different kinds of
optimization problems. Taguchi’s standardized
QC (X1, X2,......Xi, w1, w2,......wj) = X1.w1 + X2. w2 version of DoE (commonly known as “Taguchi
+...... Xi.wj (1) method”) was introduced in the USA in the early
114
1980’s. Nowadays, it is one of the most attractive Orthogonal arrays allow designers to study a
quality design tools used by all types of engineers range of design parameters in the same time and to
in manufacturing industries. The DoE using Tagu- estimate the effects of each factor independently.
chi approach can economically satisfy the needs Hence, information about design parameters is
of problem solving and product/process design obtained with minimum time and resources. The
optimization projects in engineering domains signal-to-noise ratio is simply a quality indicator
(Roy, 2001, Raymond et al., 2009). by which experimenters can evaluate the effect
DoE approach utilizes the predetermined of changing a particular design parameter on the
parameters of a studied problem and establishes performance of the process or the product (Phadke,
an experiment according to the number of trials 1989; Roy, 2001).
(experimental runs), the factor levels and the Standard Orthogonal Arrays (L4 to L81, see
number of targets to be controlled. The major also Table 1) are available for different designed
intent is to reduce alternative scenarios without experiments.
actually removing the cause of parameter variation Appropriate selection of an orthogonal array
and achieve consistent performance by making is based on the experiment’s number of degrees
product/process insensitive to the influence of of freedom in such a way that the number of
uncontrollable factors. Two important tools used experimental runs should be greater than, or equal
in DoE methodology are orthogonal arrays and to. the number of freedom degrees. To define an
signal-to-noise (S/N) ratios (Phadke, 1989; Roy, orthogonal array, one must identify:
2001).
In terms of machining processes, the procedure • The number of investigated factors,
of DoE is conducted by determining the follow- • The levels of independent factor,
ing steps: • The occurring interactions among factors,
• The reliability of analysis, or the existence
1. Manufacturing Problem Definition (machin- of constraints in time and cost.
ing process). • The special difficulties that would be
2. Determination of machining objective (or encountered during the manufacturing
objectives). experiment.
3. Determination of the respective independent
factors and their levels. More information concerning Taguchi Or-
4. Selection of the appropriate Orthogonal thogonal Arrays can be found in corresponding
Array. literature; see for example Phadke (1989), Roy
5. Design of the manufacturing experiment. (2001) and Raymond et al., 2009.
6. Executions of the experimental runs.
7. Data analysis and interpretation. 2.3 CNC Machining
8. Run a validation manufacturing experiment. Process Parameters
2.2.2 Orthogonal Arrays (OAs) The desired quality characteristics in a machining

process (low cutting force, high surface quality,
Orthogonal Arrays are defined as matrices whose reduced machining time, low remaining volume in
columns represent the number of parameters to roughing, etc) are strongly affected by the selection
be simultaneously studied at different levels in of the process parameters, such as feedrates, speed,
different combinations of experiments, whilst the cutting depth etc. Thus, machining parameters
number of rows represents the number of runs. should not only correspond to pre-determined
115
Table 1. Standard orthogonal arrays used in DoE
Orthogonal Number of Maximum Maximum number of Columns at these levels

Array Rows number of
Factors
2 3 4 5
L4 4 3 3 - - -
L8 8 7 7 - - -
L9 9 4 - 4 - -
L12 12 11 11 - - -
L16 16 15 15 - - -
L18 18 8 1 7 - -
L25 25 6 - - - 6
L27 27 13 1 13 - -
L32 32 31 31 - - -
L36 36 23 11 12 - -
L50 50 12 1 - - 11
L54 54 26 1 25 - -
L64 64 63 63 63 - -
L81 81 40 - 40 - -
manufacturing demands, but also accommodate concerned. Table 2 indicates some of the most
the system constraints. The parameters of a ma- representative references in terms of independent
chining process are considered to be the major variables investigation with respect to machining
control variables that characterize the economical processes optimization.
and qualitative rate in terms of efficiency. However, success of machining processes does
As a result of the researches carried out so not lie only in the suitable specification of cutting
far, machining data catalogues and manufactur- parameters, but also on the appropriate selection
ing handbooks in printed or electronic form have of toolpath strategies and manufacturing param-
been developed in order to help engineers and eters that CAM systems offer in building process
programmers to select appropriate machining models. In current industrial practice, NC pro-
data for particular machining operations. These grammers and process planners take advantage
values have been obtained in controlled environ- of CAM systems due to high complexity of prod-
ments of laboratories and research units; they are ucts. Part machining is thoroughly planned on a
extremely conservative so as to protect excessive CAM system; users choose toolpath strategies,
matching errors from tool failures (tool deflection, milling parameters and their values. According
wear, breakage, etc). Hence, the recommended to these choices, CAM systems create toolpaths,
parameters in these handbooks may be feasible, perform machining simulations, and extract sta-
but not necessarily the most appropriate or optimal. tistics of created tool-paths, such as machining
Extensive research effort has been undertaken time, tool-path length, etc., as well as 3D models
in the past few years aiming to \ optimization and of machined products.
proper selection of machining conditions as far as Machining quality of parts is strongly depended
Cutting Speed, Feed Rate, Depth of Cut etc are on user’s selections, which in turn are based on
116
Table 2. Literature survey on optimized selection of independent machining variables
Independed Machining Variables

a/a Process Parameters Relative References
1 Tooling Characteristics Tansel, Ozcelik, Bao, Chen, Rincon, Yang & Yenilmez, 2006
Lin & Tai, 1999
Reddy & Rao, 2006
El-Mounayri, Kishawy & Tandon, 2002.
2 Number of passes in one direction Dong, & Vickers, 1993
Dereli, & Filiz 1999
Sönmez, Baykasoğlu, Dereli & Filiz, 1999.
3 Axial - Radial cutting depth in each pass Tansel, Ozcelik, Bao, Chen, Rincon, Yang & Yenilmez, 2006
El-Mounayri, Kishawy & Tandon, 2002
Dong, Li & Vickers, 1993
Dereli, & Filiz 1999
Sönmez, Baykasoğlu, Dereli & Filiz, 1999,
Baptista,& Simoes, 2000
Lin, 2002
Onwubolu, 2006
Shunmugam, Reddy & Narendran, 2000
Juan, Yu & Lee, 2003
Feng & Su, 2000
Quinsat & Sabourin, 2006
Sotirov, Vitanov, Ghosh, Mincoff & Harrison, 1992
Duffuaa, Shuaib & Alam, 1993
Ghani, Choudhury & Hassan, 2004
Liu & Wang. 1999
Kurdi, Schmitz, Haftka & Mann, 2004
Ozcelik, Oktem & Kurtaran, 2005.
4 Cutting Speed Lin & Tai, 1999
Liu & Wang, 1999
Kurdi, Schmitz, Haftka & Mann, 2004.
5 Feed per Revolution / Feed per tooth and Direction Tansel, Ozcelik, Bao, Chen, Rincon, Yang & Yenilmez, 2006
Lin & Tai, 1999
Reddy & Rao, 2006
Dereli & Filiz, 1999
Sönmez, Baykasoğlu, Dereli & Filiz, 1999
Baptista & Simoes, 2000
Lin, 2002
Onwubolu, 2006
Shunmugam, Bhaskara Reddy & Narendran, 2000
Feng & Su, 2000
Ozcelik, Oktem & Kurtaran, 2005
Wang, Rahman, Wong & Sun, 2005
Wang, 1998
Wang, Wong & Rahman, 2004
Cakir & Gürarda, 2000
Tolouei-Rad & Bidhend, 1997.
continued on following page
117
Table 2. Continued
Independed Machining Variables
a/a Process Parameters Relative References
6 Cutting Velocity (Vc) Reddy & Rao, 2006
Lin, 2002
Onwubolu, 2006
Liu & Wang, 1999
Wang, 1998
Tolouei-Rad & Bidhend, 1997.
personal experience of the machining process Despite the fact that all CAM systems share a
and familiarization with CAM systems. CAM common philosophy in terms of functionality prin-
systems are built so that they support even the ciples and processing, one may find differences
most perplexing cases of machining free-form between systems mainly in the representation, the
parts. Nowadays, a typical CAM module provides naming, and the definition of functions/properties.
a wide range of Machining Strategies, so as to Some of the most important and commonly used
program and simulate different cutting combina- manufacturing parameters provided in all CAM
tions with the usage of toolpath styles and process systems are noted below:
parameters according to the geometry of a part to
be machined. Common machining strategies and 1. Machining Strategy / Toolpath Style:
process parameters are provided in all CAM sys- This option enables the user to determine
tems but their characterizations, manipulations and the strategy will be used to machine a work
definitions may vary depending on the software piece. Each machining strategy offers addi-
vendor. When machining strategies and process tional parameters to define the way the tool
parameters have to be selected (especially for moves and the toolpath style.
special occasions like sculptured surface machin- 2. Stepover (Radial cutting depth): Stepover
ing), a programmer may be confused about which parameter is the Radial Distance between
strategy and parameter range should determine. two successive tool passes in X-Y plane
Since CAD/CAM systems are of great im- (3-axis machining) or in the plane that is
portance when it comes to design, planning and vertical to tool axis (5-axis machining). It
manufacturing of products, research should be can be determined either by an arithmetic
conducted to investigate their effectiveness, their value or as a ratio of cutting tool’s diameter
reliability and their robustness through industrial (engagement).
experiments. Through these experiments, it is pos- 3. Stepdown (Axial cutting depth): Stepdown
sible to perform statistical validation, verification parameter adjusts the maximum depth of cut
and selection of the most significant strategies and in the vertical direction (or tool axis direc-
their parameters that affect machining processes. tion), between successive machining planes.
118
It can be determined either by setting the that Machining Time is related to economical and
number of cutting planes, or by specifying productivity issues, so it directly affects cost and,
the distance between the planes. thus, it should be the smallest possible.
4. Machining Tolerance: The value of the
maximum allowable distance between the 2.4.2 Surface Quality
theoretical toolpath and the toolpath com-
puted by the CAM system. Surface Quality (SQ) determines the proper
5. Machining Direction: This option ad- interaction of a machined work piece with its
justs the cutting direction whether to be of environment, regardless of its working character
“Climbing” type (the front of the advanc- as a “functioning” part of an assembly or as an
ing tool cuts into the material first) or of “aesthetic” part that offers visual or tactile satisfac-
“Conventional” type (the back of the advanc- tion. Parts with poor surface quality usually wear
ing tool cuts into material first). sooner than those having fine surface finishing.
Additionally, higher friction coefficients are ob-
2.4 Quality Characteristics in served in poorly surface finished parts. Arithmetic
CNC Machining Processes average roughness (Ra) and maximum peak to
valley height (Rt) are by far the most commonly
A priori, any manufacturing process ought to used parameter in surface finish measurement and
achieve high quality products in short times. for general quality control. Despite its inherent
Machining processes are expected to have prede- limitations, it is easy to measure and offers a good
termined quality characteristics that affect surface overall description of the height characteristics
accuracy, manufacturing cost, and machining time. of a surface profile. Moreover, bearing (Abbott)
This section gives an overview of the quality char- curves and bearing ratio parameters is a useful tool
acteristics that determine the result of a machining for analyzing surface morphology and tribological
process in terms of quality and efficiency. Some performance (Petropoulos et al., 2006)
of the major quality characteristics are briefly Surface quality of a machined part is a result
described; while others are tabulated together of different combinations between deterministic
with their corresponding references; see Table 3. and stochastic machining process parameters. The
main target of a machining process – in terms of
2.4.1 Machining Time surface quality – would be to minimize surface
errors, to maintain geometrical tolerances, and
Machining Time (MT) refers to the time needed to correlate quality attributes to corresponding
for a raw stock to be transformed into a final machining process parameters.
predefined form (finished part or product) accord-
ing to the respective design specifications and its 2.4.3 Cutting Force Uniformity
blueprints. Machining Time is the overall time of
the machining process and it doesn’t involve idle Varying cutting forces dramatically affect surface
times indirectly attached to the manufacturing quality of a machined part and the machining
process, e.g. pallet changes, raw stock original process conditions. What is more, high cutting
positioning etc. However, times of all tool move- forces during machining yield high tool deflec-
ments – cutting and rapid – as well as tool changing tions and tool wears. To ease this effect, Cutting
cycles’ time are included, since they are immedi- Force Uniformity (CFU) is treated as a quality
ate results of the chosen machining strategy and target for optimization.
parameter values. Furthermore, it should be noted
119
Table 3. Literature survey of machining quality characteristics
Quality Characteristics
a/a Quality Targets Relative References
1 Minimum Machining Time (including also idle times El-Mounayri, Kishawy & Tandon, 2002
like tooling setup, rapid traverses, tool changes). Dong, Li & Vickers, 1993
Onwubolu, 2006
Wang, 1998
Tolouei-Rad & Bidhend, 1997
Tandon, El-Mounayri & Kishawy, 2002
Yang & Han, 1999.
2 Maximum Material Removal Rate (MRR) Tansel, Ozcelik, Bao, Chen, Rincon, Yang & Yenilmez, 2006
Lin, 2002
Liu & Wang. 1999
Cus & Balic, 2003.
3 Minimum Machining Cost El-Mounayri, Kishawy & Tandon, 2002
Cus & Balic, 2003.
4 Maximum Production Rate / Maximum Profit Dong, Li, & Vickers, 1993
Cus & Balic, 2003
5 Optimum Part Quality in terms of Surface Roughness Lin & Tai, 1999
Reddy & Rao, 2006
Baptista & Simoes, 2000
Lin, 2002
Cus & Balic, 2003
Agrawal, Pratihar & Choudhury, 2006
Vaxevanidis et al., 2011
6 Cutting Force Uniformity Tansel, Ozcelik, Bao, Chen, Rincon, Yang & Yenilmez, 2006
Feng & Su, 2000
7 Tool Life El-Mounayri, Kishawy & Tandon, 2002
120
Table 4. Literature survey of technological constraints
Technological Constraints
a/a Constraints Relative References
1 Maximum available motor power, maximum Cutting Dong & Vickers 1993
Force, Maximum Cutting Edge Load and Maximum Onwubolu, 2006
Tool Deviation. Shunmugam, Bhaskara & Narendran 2000
Sotirov, Vitanov, Ghosh, Mincoff & Harrison 1992
Liu & Wang 1999
Wang, 1998.
Cus & Balic, 2003.
2 Maximum available Torque, Maximum Range of Dong, Li & Vickers, 1993
Feeds and Speeds. Wang, 1998
3 Machining Parameter limits Reddy & Rao, 2006
Onwubolu, 2006.
Shunmugam, Reddy & Narendran, 2000.
Liu & Wang, 1999.
Wang, 1998
Cus & Balic, 2003
4 Cutter Geometry Tandon, El-Mounayri & Kishawy, 2002
5 Tool Wear and Life/ Generated Heat Sotirov, Vitanov, Ghosh, Mincoff & Harrison, 1992
6 Geometrical, Technological and Stability properties of Sotirov, Vitanov, Ghosh, Mincoff & Harrison, 1992
CNC Machine Tools Wang, Rahman, Wong & Sun, 2005
7 Work piece Form Errors, Surface Integrity, etc Onwubolu, 2006
Liu & Wang, 1999
Wang, 1998
2.4.4 Remaining Volume tion of the Remaining Volume lying on the target
model’s surface should be smoother, in order to
Remaining Volume (RV) after the roughing and/ reduce varying chip loads and varying cutting
or the semi-finishing stages is the uncut material forces during finishing and to finally achieve
of a roughed or semi-finished part that remains better surface accuracy.
to be removed at the finishing stage. Remaining
volume is determined by subtracting the volume of 2.4.5 Material Removal Rate
the target model from the volume of the roughed
part. Smaller Remaining Volume leads to quicker Material Removal Rate (MRR) reveals the amount
finishing operations and thus overall Machining of the material disposal from a workpiece in terms
Time is greatly reduced. Additionally, distribu- of time. A high MRR produces short machining
121
time, whilst a low MRR yields long machining the different origins. In general, the ANOVA
time. MRR is a usual quality target for optimiza- procedure embodies a group of statistical meth-
tion in roughing operations where it is desired to odologies, suitable for data analysis obtained via
rapidly dispose material, as it has been already experimental designs.
mentioned. Sample data can be ranked into two catego-
ries; observational sampling data and designed
2.5 Technological Constraints sampling data. In the first category, one can only
for Machining Optimization observe the values revealed without the ability of
intervention in the respective variables, whereas
The optimization processes should be subjected the second category allows the control of one or
to technological constraints that pre-exist due more independent variables, in order to determine
to both environmental issues and equipment the influence on the dependent variable, or the
technical properties. Such constraints have been response.
investigated in literature and are summarized in It has already been mentioned that the main
Table 4; for an overview see also Groover, 2008. target of an experiment, is to specify the effect
that one or more independent variables have on a
response or a quality characteristic. The variables
3. STATISTICAL METHODS FOR are referred to as quantitative or qualitative factors.
OBTAINING CNC MACHINING Values of each factor determined on an experiment
INFORMATION are named as “levels” (see also Section 2.2.2).
This section presents the most significant and 3.1.1 The F-Distribution
commonly used statistical methodologies to ob-
tain and evaluate machining information, which Consider Y1 and Y2 to be two random independent
is extracted by experiments carried out either on variables obtained from a x2 distribution and n1,
actual machine tools or on computer-aided soft- n2 to be their degrees of freedom respectively, the
ware modules like CAM systems. Once machining fraction W is the random variable that follows the
information has been summoned, researchers can Fisher’s F-Distribution.
benefit and utilize one of the following statistical
approaches to facilitate the proper verification, Y1 / n1
W = (2)
evaluation and analysis of the tested machining Y2 / n2
variables. Each method has both advantages and
drawbacks according to the type and the nature
of the experiment and its candidate parameters The Fisher’s distribution (F-Distribution) is
for processing. characterized by the two aforementioned freedom
degrees n1,n2, for the numerator and the denomi-
3.1 ANalysis Of VAriance (ANOVA) nator respectively. Thus is expressed as, Fn ,n .
1 2
Note here that c2 >0; hence, F>0. The Fn ,n

Analysis of Variance (ANOVA) is a statistical
1 2
distribution family is asymmetric. The density

methodology that offers the ability to divide vari-
possibility function is:
ability of a major component to its individual sub-
components, so as to reveal the specificity among
122
f (x ) = 3.1.2 One-Way ANOVA

 n + n 
G  1 2 n1 n1 −2
 2   n1  2 x 2 One-Way ANOVA is a simplified statistical
  , 0 < x < ∞,
 n   n   n  n1 +n2 methodology among other experimental designs.
G  1 G  2  2  
1 + n1x 
2
Should an experimental design study variations
 2   2  
 n2  of a single factor, the design is considered to be
completely randomized.
(3) In the following section, we proceed on dem-
onstrating an experimental design to evaluate the
where, G(n)=(n−1)!=1∙2∙3∙∙∙(n−1), n ∈ + , is the contribution of machining parameters for a rough
Gamma Function. machining operation. The experiment was mod-
It is imperative that the samples should be elled using typical CAM software (PowerMill®).
subjected to the Nominal Distribution to generate The quality targets were Remaining Volume to be
an F-Distribution. However, in occasions where machined at a finishing operation and the Ma-
the sample distributions diverge from the Nomi- chining Time. The machining strategy tested for
nal Distribution, the F-Distribution is not influ- this roughing operation is called Raster strategy
enced, if sample populations are similar. Under (Figure 1) and this is a typical strategy that allows
these circumstances, the experiment lies on the the Back & Forth tool-path style. The parameters
F-Distribution is to be robust. During experiments, contributed to the experiment were assigned to
two distributions may be compared. If this may an L27 Taguchi Orthogonal Array and the results
be the case, the fraction s12 / s22 must be calcu- were evaluated. Then the ANOVA methodology
lated. When equal variations are occurred the was performed so as to reveal the effect and the
fraction follows the law s12 / s22 = 1. In practical influence of each variable on quality character-
occasions the variations of populations are un- istics. Remaining Volume and Machining Time
known and the comparisons are conducted in after roughing were both calculated according
accordance with the sampling variations. For equations 1 and 2 respectively.
example, if s12 , s22 are the sampling variations
from two nominal distribution samples n1, n2 VR = VMm− VDm (5)
respectively, then the fraction VR follows the
TRm = Ttot − Ti (6)
F-Distribution.
where
s12 / s12
VR = (4) VR is the Remaining Volume of the roughed
s22 / s22 model in mm3; VMm is the volume of the machined
model in roughing operation in mm3; VDm is the
Under the condition of H 0 : s12 = s22 , the volume of the designed target 3D model in mm3;
TRm is the Machining Time of the workpiece mate-
s12
statistical VR = follows the distribution rial in sec (idle times are not included); Ttot is the
s22 total machining time in sec and, finally, Ti is the
F(n -1),(n -1) . idle time in sec, during the process.
1 2
The independent variables (machining pa-
rameters for Raster strategy) contributing to the
experiment are described next.
123
Figure 1. The Raster machining strategy and its tool-path style; the cutting tool follows a back & forth
movement: a) The Stepover and Stepdown machining parameters of Raster strategy. b) The Raster Angle
machining parameter
1. Stepover (Radial Cutting Depth, ae) and toolpaths, concerning machining time and
Stepdown (Axial Cutting Depth, ap) both surface quality.
measured in mm, (Figure 1a). 6. Joining Range machining parameter that
2. Thickness between the work piece and the determines the joining distance between
tool’s cutting edge. two successive Z-axis levels, or passes. Like
3. Profiling machining parameter (parameter Allowance machining parameter, Joining
that determines the cutter’s leading move- Range was treated as an ON-OFF switch
ment in terms of the given work piece profile during the experimental process.
in Z-axis heights. This particular parameter
has four conditions; No profiling, Profiling Table 5 demonstrates an example of how
before roughing pass, Profiling during machining parameters, and their respective lev-
roughing pass and Profiling after roughing els, can be organized in the columns of an L27
pass. orthogonal array. A different orthogonal array,
4. Allowance parameter, which plays the role thus a different DoE, can be created for a different
of a safety distance factor between the ma- cutting tool. Alternatively, cutting tool can be an
chined surface and the modelled surface. This independent variable and in that sense it can be
particular parameter has two conditions, ON included in the DoE. As it is presented here, the
and OFF. designed experiment consists of five factors with
5. Raster Angle machining parameter. This par- three levels and two factors with two levels. These
ticular parameter defines the angle between 2-leveled factors are placed in L27 together as one
the linear passes (XY plane) and X axis three-leveled factor after proper transformation
(Figure 1b). In this experiment the Raster of the respective Orthogonal Array. It is of great
Angle of the roughing process was adjusted importance to mention that in this experiment,
to three levels taking the values of 45°, 90°, no interactions between factors were observed,
and Auto respectively. Auto means that the as can be seen in literature (Krimpenis, Fousekis,
system automatically determines the value & Vosniakos, 2005). A stronger proof of this hy-
for this parameter, in terms of optimized pothesis is substantiated by the ANOVA results
in the aforementioned reference.
124
Table 5. Factor levels and values of machining parameters
Raster StrategyRaster Factor Levels

L27
Factors Tool Ø 25 mm Tool Ø 32 mm Tool Ø 40 mm
Col
Stepover 1 8 10 12 14 16.5 19 15 18 21
Thickness 2 0.5 1 1.5 0.5 1.0 1.5 0.5 1 1.5
Stepdown 5 2.50 3.75 4.95 2.80 3.80 5.20 3.50 4.50 5.50
Profiling 9 Before During After Before During After Before During After
Raster Angle 10 45º 90º Auto 45º 90º Auto 45º 90º Auto
Allowance-
12 On, On On, Off Off, On On, On On, Off Off, On On, On On, Off Off, On
Joining Range
For this specific example, a combination of Machining Time are presented in Krimpenis, et
the quality characteristics was used after the al., 2005. For the example of Table 5, Machining
proper specification of respective weights for each Time varies from 200 to 500 sec, while Remain-
quality target. As Machining Time is of greater ing Volume from 15,000 to 25,000 mm3 for the
importance than Remaining Volume after the studied part and for the three different cutting
roughing process for the specific DoE, a larger tools. The ANOVA results for this specific study
weight is applied. Further on, a normalization and a relatively simple sculptured surface part
procedure was conducted (see Section 2.1.2) due has shown that the most influential machining
to the incompatibility of measured units between parameter is Stepdown having about 35% influence
quality characteristics (Remaining Volume in mm3 on the QC function. Profiling and Raster Angle
and Machining Time in sec). This combination parameters do not affect QC function, since their
among targets and weights is calculated consider- significance is lower than 1%. Stepover, Thick-
ing Equation 1 (Section 2.1.2). Thus, normaliza- ness and Allowance-Joining Range parameters
tion of the quality targets is expressed with Equa- are statistically significant in the studied problem,
tion 7. affecting it with significance varying from 10 to
25%. It is noted that no interdependency exists
QC (T ,V , κ, λ) = κ ⋅ Τ + λ ⋅V (7) among the machining parameters, proving that it
is correct to consider the machining parameters
as independent variables. ANOVA also showed
where
that this approach is robust and offers results with
high integrity, since error variables are statistically
QC: Quality Characteristics function,
insignificant.
Τ: Normalized Machining Time,
V: Normalized Remaining Volume,
3.2 Principal Components
κ, λ: Weights of quality targets.
Analysis (PCA)
Here, the weights were adjusted to be κ = 0.7
Pearson and Hotelling, years ago, introduced
and λ = 0.3.
Principal Components Analysis (PCA). PCA is a
PowerMill® was used as the CAM platform
statistical methodology that allows the identifica-
for the machining simulations. The results of
tion of patterns in data and expressing these data
optimum values for the Remaining Volume and
in a way that similarities and differences can be
125
easily observed (Johnson and Wichern, 2002). where

Since localization of patterns in data is difficult –
especially for multi-dimensional data – and cannot b2m1 + b2m2 +...............+ b2mn = 1 (9)
be illustrated graphically, PCA approach is the
appropriate tool for analysis. In other words, PCA
approach is a variable reduction process. When 3.2.1 Application of PCA in a
redundancy (correlation of data) occurs within CAM Machining Experiment
variables, this methodology can be used in order
to reduce the observed variables into a smaller Since experiments in machining processes are
number of principal components that will greatly usually treated as multi-objective problems,
influence the experimental process. this methodology can be adopted in machining
Technically, a principal component is the lin- data analysis. Moreover, other similar statisti-
ear combination of optimally-weighted observed cal approaches are based on the assumption that
variables. For proper understanding of the defini- quality attributes may be uncorrelated; however,
tion, it is essential to describe the computational the reverse case may appear. Should this case be
procedure of subject scores. Below we provide a adopted, PCA is used to overcome the shortcom-
general formula for the computation of the scores ing of correlated multi-attribute optimization for
on the first component extracted in Principal machining processes (Moshat et al., 2010 and
Component Analysis: Rao, 2011).
The steps of machining optimization involved
a1 = b11 (x1) + b12 (x2) + ........ b1n (xn) (8) in Principal Components Analysis are discussed,
while presenting a CNC end-milling operation
where experiment of an aircraft structural part conducted
in typical CAM software.
• a1 is the problem’s score on Principal
Component 1 (The initial component • Collection of machining data.
extracted).
• b1n is the regression coefficient (or weight) According to the number of experimental runs,
for observed variable n, as used in creating the quality characteristics and the fundamental is-
principal component 1. sues described in the previous section, one should
• xn is the problem’s score on observed vari- proceed with gathering a robust set of data. In the
able n. presented example, data should be summoned to
evaluate optimal results for selection of Machin-
According to the above determined formula, ing Tolerance, Radial cutting depth/Stepover (ae)
the following uncorrelated linear combinations and Axial cutting depth/Stepdown (ap) so as to
can be formed: obtain the smallest possible Remaining Volume for
finishing (VR) and the smallest possible Machin-
a1 = b11 (x1) + b12 (x2) + ........ b1n (xn) ing Time (tM). The experiment is conducted by
a2 = b21 (x1) + b22 (x2) + ........ b2n (xn) choosing the appropriate machining parameters,
. tooling selections (cutting parameter selection,
. tool selection, etc), and feed-rate, as well as spindle
am = bm1 (x1) + bm2 (x2) + ........ bmn (xn) speed for the milling process. The definitions of
126
Remaining Volume and Machining Time are given X1 = {X1(1), X1(2)..............X1(k)......X1(n)}

by Equations 10 and 11. .
Xi = {Xi(1), Xi(2)..............Xi(k)......Xi(n)}
Remained Volume after Roughing: VR = VMm- .
VDm (10) .
Xm = {Xm(1), Xm(2)..............Xm(k)......Xm(n)}
Machining Time: TRm = Ttot - Ti (11) (12)
where, In the above linear expressions, Xi, represents

VR is the Remaining Volume of the roughed the ith experimental results. If X0 is the reference
model in mm3; VMm is the volume of the machined sequence (X0 = {X0(1), X0(2)....X0(k)....X0(n)}), the
model in roughing operation in mm3; VDm is the value of the elements means the optimal value of
volume of the designed target model in sec; TRm the corresponding quality characteristic. Both X0
is the Machining Time of the work piece material and Xi include n elements and X0(k), Xi(k) represent
in sec (idle times are not included); Ttot is the total the numerical value of kth element in the reference
Machining Time in sec; and finally, Ti is the idle sequence. Note that k =1,2....,n.
time in sec during the process. Further on, we proceed on the normalization
In this experiment, collection of data for both of the responses (quality characteristics).
VR and TRm is performed by assigning the ma-
chining parameters to an L9 Taguchi Orthogonal • Normalization of the responses (quality
Array with the suitable values and their levels, as characteristics).
seen in Table 1 (Section 2.2.2). This experiment
tests the performance of two different machining In order to eliminate potentially extended
strategies (Helical and Back & Forth tool path optimal values and reduce imminent large ranges
style) in CATIA V5 R18® using three different of series, a normalization procedure should be
cutting tools (D20, D25 and D32), in terms of the applied to the original experimental data. The
aforementioned quality targets. The setup process, normalization process of the responses according
the designed model and the roughed stock are to the preset requirements (smaller-the-better/SB,
illustrated in Figure 2. Larger-the-better/LB and Nominal-the-best/NB),
Results of the experiment runs after normaliza- is conducted by utilizing Equations 13-15.
tion are illustrated for Helical and Back & Forth Smaller-the-better Case (SB):
machining strategies in Tables 6 and 7, respec-
tively. min X i (k )
Xi *(k ) = (13)
X (k )
i
• Linear expression of experimental runs.
Larger-the-better Case (LB):

If the number of runs in a Taguchi OA designed
experiment is m and the number of quality charac-
Xi (k )
teristics is n, then the following linear expressions Xi *(k ) = (14)
of results can be described by the series X1,..Xi,.. max X (k )
i
Xm. This sequence of linear expressions containing

the appropriate results is:
127
Figure 2. The model of the experimental aircraft structural part (CATIA V5R18®): a) Setup process; b)
Designed 3D model; c) Roughed 3D model
Nominal-the-best Case (NB): quality characteristic. After data normalization,

the value of Xi*(k) will be between 0 and 1.
min{Xi (k ), X 0b (k )}
Xi *(k ) = (15)
max{X (k ), X 0b (k )} • Checking for correlation between two
i
quality characteristics.
where, i=1,2,....,m, whilst k=1,2,....,n. Xi*(k) is If i=1,2,......n, then the normalized series of the
the normalized data of the kth element in the ith ith quality characteristic is Ci = [X0*,(i), X1*,(i),
sequence. X0b(k) is the desired value of the kth X2*,(i),.......... Xm*,(i)]. The correlation coefficient
128
Table 6. Experimental results for helical machining strategy
RUN MACHINING TIME REMAINING VOLUME

t (sec) V (mm3)
a/a D=20 D=25 D=32 D=20 D=25 D=32
1 132194 093296 064119 657225 773343 837349
2 065874 048141 034512 681726 797522 879779
3 049097 034441 024621 716199 818958 921657
4 132099 093389 064104 684460 767677 851316
5 065854 048237 034459 7 10486 799640 895224
6 048995 034456 024602 732289 820326 943704
7 132105 093355 064014 696297 770594 872282
8 065850 048179 034482 720895 802966 913148
9 049066 034467 024592 744134 821702 964201
Table 7. Experimental results for back & forth machining strategy
RUN MACHINING TIME REMAINING VOLUME

t (sec) V (mm3)
a/a D=20 D=25 D=32 D=20 D=25 D=32
1 134663 095147 065246 669538 773348 837351
2 063211 048851 035276 693864 797526 879781
3 049343 034729 025029 716206 818962 921662
4 134635 095153 065237 684489 767682 851323
5 066811 048845 035251 710497 799645 895226
6 049321 034754 024995 732299 820335 943702
7 134624 095132 065218 696318 770589 872281
8 066802 048847 035234 720882 802960 913154
9 049308 034752 024977 744121 821697 964278
between two quality targets is calculated by Equa- The correlation can be verified by applying the
tion 16, as follows: following hypothesis:
Cov(C j ,C k ) If H0: pjk = 0, there is no correlation,

p jk = , j=1,2,......,n and k = If H0: pjk ≠ 0, there is correlation.
sC x sC
j k
1,2,........,n, j≠k. (16) After the clarification of correlation existence

between quality characteristics, the following
where, pjk is the correlation coefficient between step is to calculate the principal component score
quality characteristics j and k respectively; Cov (Step 5).
(Cj,Ck) is the covariance of quality characteristics
j and k, whilst σCj and σCk are the standard de-
viation of quality attributes j and k respectively.
129
• Calculation of the Principal Component 3.3 Regression Analysis

score.
Regression Analysis is another statistical meth-
In order to calculate the principal component odology that offers the ability to obtain useful
score, one has to conduct the following calcula- data for CNC machining operations and other
tion steps: engineering processes. Regression Analysis has
been proven very handy when it comes to cluster-
• Calculation of the eigenvalue ek and the ing distances of data points in diagrams in a way
corresponding eigenvector bk (k=1,2,....n) that their differences from a respective curve or
from the correlation matrix formed by all line are minimized (Montgomery, 2001). This
quality characteristics. method reveals and explains the relation between
• Calculation of the principal component a response or dependent variable Y and one (or
scores of the normalized reference se- more) predictor/ input/independent variables X1,
quence and comparative sequences using X2,....,Xn. If n=1, then this is said to be simple
the equation shown below: regression analysis, whereas if n>1, it is multiple
regression or multivariable regression. Further
n
details and steps of the regression analysis will
Yi (k ) = Σ Xi*( j )bkj , i=0,1,2,.....,m; k=1,2,......,n.
j =1 be demonstrated through an experimental process
(17) of a sculptured surface machining operation in a
typical CAM system.
where Yi(k) is the principal component score of the
kth element in the ith series. Xi* is the normalized 3.3.1 Application of Regression
value of the jth element in the ith sequence and bkj Analysis in CAM Machining Experiment
is the jth element of the eigenvector bk.
An test part; see Figure 3, has been prepared
• Computation of quality loss for indexes. for machining using a typical roughing strategy
The principal component having high- supported by CAD/CAM/CAE system CATIA
est accountability proportion (AP) can be V5R18®, in order to remove the main volume in
treated as the overall quality index, which roughing, followed by a special tool-path style for
is to be finally optimized. The quality loss 5-axis finishing, thus obtaining the final sculptured
Δ0,i(k) of that index (in comparison to the surface finish. Figure 3a shows the 3-axis roughing
ideal status) is computed as follows: toolpath, Figure 3b depicts the 5-axis finishing
◦◦ Δ0,i(k)=|X0* (k)- Xi*(k)|, if there is no toolpath (5-axis sweeping strategy), while Figure
significant correlation between qual- 3c depicts the designed 3D model.
ity characteristics. In machining, optimum results can be gained
◦◦ Δ0,i(k) = |Y0(k)-Yi(k)|, if there is sig- by the expression of multivariable data as:
nificant correlation between quality
characteristics. Y = G(z1, z2,.....zj,.....,zk) (18)
Conclusively, the optimal setting is then veri- where

fied by minimizing Δ0,i(k) (quality loss estimate) zj - j = 1, 2, …., k represents the process inde-
by adopting Taguchi approach. The statistical pendent variables (control inputs);
package used for running PCA was MINITAB 15®. Y - the process dependent variable (output);
130
Figure 3. The test part for sculptured surface machining (CATIA V5R18®)
G - the type of dependence relation (Premrov variables for this combinational machining ex-
et al., 2007) periment were:
In order to determine optimum G type, one
has to determine the values and variation field • Cutting Speed, Vc (m/min), was considered
of each input –both real, zj and encoded xj–, as as a combination of both roughing and
well as the experiment design type that fits best finishing Cutting Speeds with analogous
(Iliescu et al., 2007). weights
If variable zj takes minimum or maximum • Feedrate, Vf (mm/min), was considered as
real values, the xj variable’s encoded value is 1 a combination of both roughing and finish-
(low level) or 3 (high level) respectively. Each ing Feed-rates with analogous weights.
variable has three levels (minimum, maximum • Component of Cutting Depth, CDtot (mm).
and medium) within the experimental considered
region, the medium one for zj representing the As far as Vc, Vf and CDtot parameters are con-
geometrical mean of its minimum and maximum cerned, the definition of relations can be estab-
level values. Consequently, the xy medium level lished in the form of constraints, as follows (Wu
is considered as level 2. & Yao, 2008):
Machined parameters were considered together
for both roughing and finishing stages. In this VcF ≥ k1 VcR (19)
particular case, machining conditions should be
expressed as relations. Therefore, independent VfR≥ k2 VfF (20)
131
Table 8. The L9 Taguchi OA with the factors, the

CaR ≥ k3 CaF (21)
levels, and Ra results
where Runs Independed Variables QC
Vc Vf CDtot Ra
VcR, VcF are the Cutting Speeds for Roughing and
1 1 1 1 3.24
Finishing respectively, in m/min.
2 1 2 2 1.81
Vf R, VfF are the Feed rates for Roughing and Fin-
3 1 3 3 6.16
ishing respectively, in mm/min.
4 2 1 2 2.53
CaR is the component of Cutting Depth in Rough-
5 2 2 3 4.27
ing, in mm.
6 2 3 1 1.34
CaF is the component of Cutting Depth in Finish-
7 3 1 3 4.35
ing, in mm.
8 3 2 1 3.46
9 3 3 2 2.48
Note that the components of cutting depths
are calculated while taking into account both
axial and radial cutting depths for roughing and
finishing. Additionally, since the total depth of Equation 24 can also be transformed to a lin-
cut is achieved by n number of cuts (or passes), ear function of the independent variables if the
Equation 22 operates as constraint. logarithmic function is applied.
Surface Roughness can either be predicted by
CD − CaF using process models or be calculated by analytic
CaR = TOT
(22) equations. Then, values that were computed by
n the aforementioned approaches are compared to
the measured values, which were obtained by a
An L9 Taguchi Orthogonal Array has been profilometer. Finally, the regression model of the
picked up for this experiment, whose structure experiment is given by Equation 24.
is presented in Table 8. The target is to optimize
machined part quality in terms of Surface Rough- Ra= 3.23Vc-0.352. Vf0.425. CDtot0.025 (24)
ness Ra (μm). So, this is considered as Quality
Characteristic (QC). Both real and coded values The mathematical regression model of Equa-
of the variables are shown in Table 9. After con- tion 24 is obtained considering that a coefficients
ducting experiments in CATIA®, toolpath post- (see Equation 23) is generated by the regression
processing produced the respective G-codes and analysis results. Statistical software packages such
the experimental parts were actually machined in as SPSS® or MINITAB®, offer reports such the
a typical 5-axis CNC machine tool. Then, rough- one presented in Table 10. From Equation 24, it
ness was measured by a special roughness tester, is apparent that Feed Rate, Vf (mm/min), exerts
like the ones presented in Section 2.1.1. Results the strongest influence on surface roughness pa-
for Ra are tabulated in Table 10. The dependence rameter values. Plotting graphs of independent
relation considered in the regression model is and dependent variables makes variations clearly
given by Equation 23. visible, since graphical visualization can directly
stress out areas of interest.
Y = C. x1a1. x2a2. x3a3 (23)
132
3.4 Response Surface Variables ξ1, ξ2,.....ξk in Equation 26 are “natu-

Methodology (RSM) ral” values, since they are expressed in parameters’
natural units. In RSM methodology, natural vari-
Response Surface Methodology (RSM) is a ables are transformed into encoded variables x1,
combination of statistical and mathematical x2,....xk. Using the encoded variables, Equation
techniques useful for developing, improving and 26 can be written as:
optimizing processes (Montgomery, 2001). The
most extensive applications of RSM are to the η = f (x1,x2,.....xk) (27)
specific situations where several input variables
potentially influence some performance outputs Success of RSM method utilization stems from
or quality characteristic of the process. Thus, the appropriate approximation of f function by
performance output or quality characteristic is the user (Carley et al., 2004). It is preferable to
called the response. The field of response surface apply a low-order polynomial in some relatively
methodology consists of: small region of the independent variable values.
In most cases, one should use either a first-order
• The experimental strategy for exploring the or a second-order polynomial. The first-order
process space or the independent variables. model is likely to be appropriate when there is
• Empirical statistical modelling in order to interest in the approximation of the true response
develop an appropriate approximation re- surface over a relatively small region of the inde-
lationship between the yield and the pro- pendent variable space in a location where there is
cess variables. little curvature in f (Carley et al., 2004). When it
• Optimization methods for obtaining pro- comes to two independent variables, the first-order
cess variables values that produce desired model in terms of encoded variables becomes:
response values. The experiment presented
in this section is based on the second strat- η = a0 +a1x1+ a2x2 (28)
egy; this is the statistical modelling which
builds a suitable approximation model be- Only the main effects of the two variables
tween the response y and its independent x1 and x2 are included in the first-order model
variables. The approximation model is giv- that Equation 28 illustrates. If interdependency
en by Equation 25. between these variables exists, then that is added
to the model as follows:
y = f (ξ1, ξ2,.....ξk) + a (25)
where α is a term that represents other sources that Table 9. Real and coded values of the tested
yield variability, not included in the f function. variables
Usually α includes effects, such as experimental
Input Values
errors on the response, background noise, the
Variable Levels
effect of other variables, and so on. Moreover,
1 2 3
α is treated as statistical error, often assumed to Vc (m/min)
100 120 140
belong to a normal distribution with mean zero
1 2 3
and variable σ2. Then, Equation 26 is obtained. Vf (mm/min)
90 120 150
E(y) = η =E [f (ξ1, ξ2,.....ξk)] + E(a) = f (ξ1, 1 2 3

CDtot (mm)
ξ2,.....ξk) (26) 1 1.5 2
133
Table 10. The regression analysis results
Regression Analysis Results-Orthogonal Program

Model Adequacy = 0.531
R* = 0.069<1 => Adequate model
Coefficients Significance
R0 = 13046.25 >1 Significant Variable
B Coefficients A Coefficients
b0 = 1.246 a0 = 0.867
b1 = -0.467 a1 = -0.352
b2 = 0.227 a2 = 0.425
b3 = 0.023 a3 = 0.025
Y Values Errors Confidence Intervals
Experimental Runs Measured Computed ABS REL Lower Lim. Upper Lim.
1 3.745 3.739 -0.00 -0.11 3.480 4.206
2 1.537 1.529 0.01 0.62 1.546 1.846
3 6.375 6.398 -0.05 -1.20 5.624 6.479
4 2.826 2.822 0.01 0.23 2.348 2.784
5 4.268 4.270 0.01 1.30 2.498 2.710
6 1.638 1.640 0.04 0.11 3.428 3.612
7 6.269 6.267 -0.02 -0.63 3.428 3.612
8 2.683 2.670 0.06 1.80 3.428 3.612
9 3.612 3.650 -0.03 -0.92 3.428 3.612
η = a0 +a1x1+ a2x2+a12x2 (29) Most likely, this model would be useful as an

approximation to the real response surface in a
Adding the interdependency term to the first- relatively small region. The second-order model
order model introduces curvature to the response is widely used in response surface methodology
function. Curvature means that the “real” response due to any of the following reasons.
surface is so strong that, despite the included inter-
dependency, the first-order model is inadequate. • The second-order model is very flexible.
To overcome this shortcoming, second-order It can take on a wide variety of functional
models may be required. Should this be the case, a forms, so it will often work well as an ap-
second-order model including two variables may proximation to the real response surface.
be formulated (Equation 30). • It is easy to estimate the parameters in the
second-order model. The method of least
η = a0 +a1x1+ a2x2 + a11x12+ a22x22+ a12x1x2 squares can be used for this purpose.
(30) • There is considerable practical experi-
ence indicating that second-order models
134
work well in solving real response surface Independent variables’ levels were assigned
problems. to an L27 Taguchi Orthogonal Array. The initial
regression model considered for the relation
According the above assumptions, a general- among the investigated parameters is expressed
ized linear model that follows a first-order poly- in Equation 33.
nomial is given by Equation 31.
Yi = C 0 + C 1 ⋅ X1 + C 2 ⋅ X 2 + C 3 ⋅ X 3
η = a0 +a1x1+ a2x2+.......+ak xk (31) +C 12 ⋅ X1 ⋅ X 2 + C 13 ⋅ X1 ⋅ X 3 + C 23 ⋅ X 2 ⋅ X 3
+C 11 ⋅ X12 + C 22 ⋅ X 22 + C 33 ⋅ X 32
(33)
whereas a second-order model in generalized
form is:
Further on, parameter estimation is conducted
k k k by implementing the least square method. The
n = a 0 + Σ a j x j + Σ a j j x j 2 + Σ Σ aij x ij procedure was carried out in MINITAB R15® and
j =1 j =1 i < j =2
(32) the resulting table (Table 11a) contains the afore-
mentioned values, where coef column represents
the coefficients of the regression model whilst the
3.4.1 Application of RSM in a SE Coef gives the deviation yielded during the
CNC Machining Experiment coefficients calculation. T and P columns check
whether model terms are statistically important or
The design of experiments method was used in full not. S value is the model’s calculated deviation.
factorial mode for the turn machining investiga- Note that S=MS Error. The value R-Sq(R2) ensures
tion of a part made of tool steel. This analysis was validity of variable representation by the model.
based on DoE, Regression Analysis and ANOVA Operating according to the instructions provided
approaches. For the generation of the stochastic by MINITAB® user’s manual, the calculation of
modelling of turning operation, a set of tool steel R-Sq(Adj) is conducted. This term is the determina-
specimens were used. The turning operation was tion coefficient adjusted for degrees of freedom.
conducted after determining three value levels Having obtained all this information, ANOVA
of cutting speed, feed-rate and cutting depth. results can be extracted. The obtained by ANOVA
Specifically, the values of Cutting Speed were results for Ra are summarized in Table 11b.
251.3, 502.6 and 753.6 m/min. Values for feed-rate In the ANOVA table (Table 11b), DF column
were adjusted to 0.05, 0.1 and 0,15 mm/rev and represents the degrees of freedom. Columns SS
values of Cutting depth were 0.5, 1.0 and 1.5 mm or SST reveal the sum of total squares in two dif-
respectively; see also Vaxevanidis et al., (2011), ferent forms. Seq SS column represents the square
where some preliminary results are reported. sum of each factor. The latter depends on the fac-
Surface roughness of the machined part was tor placement in the model. F column is used to
undertaken as quality target (Yi). Worth mentioning disclose the significance of the terms in the regres-
that different roughness values resulted, depending sion model. F value is calculated by Equation 34.
on the measured surface roughness parameters.
The examined surface roughness parameters were Adj MS
F = (34)
the average surface roughness, Ra, and the maxi- Ms Error
mum peak-to-valley height surface roughness, Rt.
135
Table 11. Estimated A) Regression Analysis; and B) ANOVA results for Ra parameter
(a) Estimated Regression Coefficients for Ra (μm)

Term Coef. SE Coef . T P
Constant 1,639 0.9485 1.728 0.102
V (m/min) -0.006 0.0022 -2.941 0.009
F(mm/rev) 10.647 10.9011 0.977 0.342
A (mm) 0.113 1.0901 0.103 0.919
V (m/min) * V(m/min) 0.000 0.0000 3.024 0.008
F(mm/rev) * F(mm/rev) 23.867 48.3481 0.494 0.628
A (mm) * A (mm) 0.036 0.4835 0.074 0.942
V (m/min) * F(mm/rev) -0.007 0.0068 -1.048 0.309
V (m/min) * A (mm) 0.001 0.0007 0.784 0.444
F(mm/rev) * A (mm) -2.440 3.4187 -0.714 0.485
S = 0.2961 R-Sq = 79.6% R-Sq(adj) = 68.8%
(b) Analysis of Variance for Ra (μm)
Source DF Seq SS Adj SS Adj MS F P
Regression 9 5.80750 5.80750 0.645278 7.36 0.000
Linear 3 4.78904 0.90679 0.302265 3.45 0.040
Square 3 0.82363 0.82363 0.274544 3.13 0.053
Interaction 3 0.19483 0.19483 0.064943 0.74 0.542
Residual Error 17 1.49018 1.49018 0.087658

Total 26 7.29768
Hypothesis analysis of the regression model OSP U10 CNC controller. Maximum available
is conducted to ensure statistical substantiation of spindle power of the CNC machine tool is 22kW,
calculations. This is done by creating the respective whilst feed power of motors is 3kW in Z-axis and
normal probability plots of the residuals. Normal 1.5 kW in X-axis respectively. The specimens
probability plots show whether residuals of the were machined using the specified cutting condi-
estimated equation follow the Gaussian distributions according to the number of runs in the L27
tion. Another alternative procedure for validating Orthogonal Array. The cutting tools used for the
regularity of residuals is to plot the Histogram of machining operations were the DNMG 110402
residuals. In order to ensure stability among re- TP 2000. Surfaces Roughness measurements were
siduals in terms of variability, one should validate conducted using a Rank Taylor-Hobson Surtronic
whether residuals are independent to the observa- 3+ roughness tester and Talyprof software. For
tion sequence. This is carried out by generating each specimen, 15 measurements were performed
the “Plot of Residuals vs Run Order” diagram. and their averages were used for further analysis.
To test the independent nature of residuals, one The selected filter for this experiment was the
has to proceed with formulating the “Residuals Gaussian filter, whereas the cut-off length was 0.8
vs the Fitted Values” diagram. mm. The L27 orthogonal array with the independent
The experiments were conducted on an OKU- variables (Spindle Speed, Cutting Speed, Feedrate
MA LB10II CNC turning center, equipped with and Depth of Cut) used in this experiment and
136
Table 12. The L27 orthogonal array and the levels of machining parameters
a/a Specimen S (rpm) v (m/min) F (mm/r) a(mm) Ra (μm) Rmax (μm) FD

1 1,1 2000 251,3 0,15 0,5 1,782 11,1 1,336
2 1,2 2000 251,3 0,15 1 1,932 15,7 1,362
3 1,3 2000 251,3 0,15 1,5 2,018 13,4 1,376
4 2,1 2000 251,3 0,05 0,5 1,032 6,494 1,3
5 2,2 2000 251,3 0,05 1 1,04 5,266 1,276
6 2,3 2000 251,3 0,05 1,5 1,046 4,874 1,31
7 3,1 2000 251,3 0,1 1,5 1,916 8,92 1,24
8 3,2 2000 251,3 0,1 1 1,87 8,814 1,24
9 3,3 2000 251,3 0,1 0,5 1,856 8,174 1,238
10 4,1 4000 502,6 0,15 1,5 1,894 10,33 1,236
11 4,2 4000 502,6 0,15 1 1,848 8,614 1,246
12 4,3 4000 502,6 0,15 0,5 1,806 7,834 1,248
13 5,1 4000 502,6 0,1 0,5 0,774 5,392 1,316
14 5,2 4000 502,6 0,1 1 0,963 4,838 1,3
15 5,3 4000 502,6 0,1 1,5 1,002 5,38 1,298
16 6,1 4000 502,6 0,05 1 0,404 2,746 1,412
17 6,2 4000 502,6 0,05 1,5 0,49 3,458 1,378
18 6,3 4000 502,6 0,05 0,5 0,35 2,712 1,396
19 7,1 6000 753,6 0,05 1 1,006 5,642 1,62
20 7,2 6000 753,6 0,05 1,5 1,638 13,02 1,33
21 7,3 6000 753,6 0,05 0,5 0,67 4,564 1,416
22 8,1 6000 753,6 0,1 0,5 0,956 5,886 1,348
23 8,2 6000 753,6 0,1 1,5 1,036 6,412 1,312
24 8,3 6000 753,6 0,1 1 0,994 5,65 1,332
25 9,1 6000 753,6 0,15 1,5 1,652 7,578 1,28
26 9,2 6000 753,6 0,15 0,5 1,586 8,46 1,272
27 9,3 6000 753,6 0,15 1 1,614 8,788 1,276
the obtained results for the range of roughness

indicators Ra and, Rt are illustrated in Table 12).
Ra = 1, 639 − 0, 006X1 + 10, 647X 2
According to results tabulated in Table 11a,
the regression models were created for the inves- +0, 113X 3 + 23, 867X 22 + 0, 036X 32
tigated group of Surface Roughness parameters. −0, 007X1X 2 + 0, 001X1X 3 − 2, 44X 2X 3
Linear models have resulted as the following (35)
series of equations (Equations 35 and 36) illus-
trates. For the maximum peak-to-valley height Sur-
For the Average Surface Roughness (Ra): face Roughness (Rt):
137
Figure 4. Statistical analysis plots for data evaluation: a. Normal probability plot of the residuals, b.
Histogram of the residuals, c. Residuals vs data order, d. Residuals vs fitted values
Ù
R t = 9, 65 − 0, 03X1 + 37, 06X2 − 0, 42X 3
where yij is the observed values and yij the esti-
+481, 07X12 + 0, 42X 32 − 0, 15X1X 2
mated values. The residuals standardization is
−12, 23X 2 , X 3 mainly depicted in the respective normal probabil-
(36) ity diagram (Figure 4a).
Another alternative in testing standardization
where X1, X2, X3 are the variables referring to Cut- among residuals is depicted in the “Histogram of
ting Speed, Feedrate and Depth of Cut respectively. the residuals” diagram (Figure 4b). Stability of
The interactions of variables are also included in residuals variation can be examined in the “Re-
the regression models. siduals vs Data order” diagram (Figure 4c). The
Along with the generation of regression latter should not depict any correlations among
models, statistical results via plots and diagrams the residuals. If that was the case, then this would
are shown for Ra and Rt indicators. The residual lead to the conclusion that residuals are not inde-
(mean square error), ei, is the error estimation in pendent to observation data sequence. A similar
terms of the variation and it is calculated by the plot (“Residuals vs Fitted values”; see Figure 4d)
following equation: for this verification can be constructed. Having
concluded the above statistical analysis proce-
∧
eij = yij − yij (37) dures, response surfaces for the different sets of
values between variables can be generated.
Firstly, response surface plots are presented in
terms of Ra indicator for f(V, A). For this very set,
138
Figure 5. Response Surfaces for Ra = f(V, a) [α. F = 0.05 mm/r, b. F = 0,1 mm/r and c. F = 0,15 mm/r]
cutting speed V and cutting depth A are kept ment - have been illustrated in order to explain
constant, while the feed rate is adjusted according the computational parts of different statistical ap-
to predetermined value levels. Similarly, the rest proaches, as well as their efficient applications on
of the response surface plots for f(V, F) and f(F, machining examples. The most commonly used
A) are produced. The former plot was generated computational techniques and statistical methods
by consecutively changing levels of Cutting Depth like Design of Experiments (DOE), Analysis of
(A=0.5 mm, A=1.0 mm and A=1.5mm) and keep- Variance (ANOVA), Response Surface Meth-
ing the values of V and F constant. The latter was odology (RSM), Regression Analysis (RA) and
generated by changing the level values of V Principal Components Analysis (PCA), have been
(V=251.3 m/min, V=502.6 m/min and V=753.6 demonstrated and applied on machining experi-
m/min). The values for F and A were kept constant ments. The different measuring devices and as-
(see Figure 5). semblies used for collecting machining data were
According to the previously described proce- also presented. As a future perspective, the devel-
dures, as far as statistical analysis, data evaluation opment of optimization modules built with the
and response surface plots generation are con- help of programming applications is mentioned.
cerned, the process is continued by extracting the The aim is to rapidly obtain data generated from
analogous results for the Rt indicator. Since the manufacturing experiments (performed on PC or
process follows exactly the same guidelines, on actual machining equipment) and to automate
evaluation plots and the response surface diagrams the computational phases of statistical methods,
are not illustrated. so as to ease the operator’s jobs.
4. CONCLUSION AND REFERENCES

FUTURE TRENDS
Agrawal, R. K., Pratihar, D. K., & Choudhury,
In this chapter a variety of statistical methodolo- A. R. (2006). Optimization of CNC isoscallop
gies, which are used for obtaining, evaluating and free form surface machining using a genetic al-
exploiting CNC machining experimental data, has gorithm. International Journal of Machine Tools
been presented. Machining experiments - either & Manufacture, 46, 811–819. doi:10.1016/j.
on CAM systems or on actual machining equip- ijmachtools.2005.07.028
139
Baptista, R., & Antune Simoes, J. F. (2000). Three Duffuaa, S. O., Shuaib, A. N., & Alam, M. (1993).
and five axes milling of sculptured surfaces. Evaluation of optimization methods for machin-
Journal of Materials Processing Technology, 103, ing economic models. Computers & Operations
398–403. doi:10.1016/S0924-0136(99)00479-3 Research, 20(2), 227–237. doi:10.1016/0305-
0548(93)90077-V
Cakir, M. C., & Gürarda, A. (2000). Optimi-
zation of machining conditions for multi-tool Faraway, J. J. (2002). Practical regression and
milling operations. International Journal of ANOVA using R. Retrieved January 13, 2011,
Production Research, 38(15), 3537–3552. from http://www.stat.lsa.umich.edu
doi:10.1080/002075400422789
Feng, H.-Y., & Su, N. (2000). Integrated tool path
Carley, M., Kathleen, K. Y. N. & Reminga, J. and feed rate optimization for the finishing machin-
(2004). Response surface methodology. (CASOS ing of 3D plane surfaces. International Journal of
Technical Report, CMU-ISRI-04-136). Machine Tools & Manufacture, 40, 1557–1572.
doi:10.1016/S0890-6955(00)00025-0
Cus, F., & Balic, J. (2003). Optimization of cut-
ting process by GA approach. Robotics and Com- Gadelmawla, E. S., Koura, M. M., Maksoud, T.
puter-integrated Manufacturing, 19, 113–121. M. A., Elewa, I. M., & Soliman, H. H. (2002).
doi:10.1016/S0736-5845(02)00068-6 Roughness parameters. Journal of Materials
Processing Technology, 56, 1–13.
Davim, J. P., Gaitonde, V. N., & Karnik, S. R.
(2008). Investigations into the effect of cutting Galanis, N. I., & Manolakos, D. E. (2009). Sur-
conditions on surface roughness in turning of face roughness of manufactured femoral heads
free machining steel by ANN models. Journal with high speed turning. International Journal of
of Materials Processing Technology, 205, 16–23. Machining and Machinability of Materials, 5(4),
doi:10.1016/j.jmatprotec.2007.11.082 371–382. doi:10.1504/IJMMM.2009.026898
Davim, J. P., Silva, L. R., Festas, A., & Abrão, Ghani, J. A., Choudhury, I. A., & Hassan, H. H.
A. M. (2009). Machinability study on precision (2004). Application of Taguchi method in the
turning of PA66 polyamide with and without optimization of end milling parameters. Journal
glass fiber reinforcing. Materials & Design, 30, of Materials Processing Technology, 145, 84–92.
228–234. doi:10.1016/j.matdes.2008.05.003 doi:10.1016/S0924-0136(03)00865-3
Denkena, B., & Köhler, J. (2010). Consideration Groover, M. P. (2008). Automation, production
of the form of the undeformed section of cut in the systems and computer-integrated manufacturing
calculation of machining forces. Mining Science (3rd ed.). Pearson Higher Education.
and Technology, 14(4), 455–470.
Iliescu, M., Spânu, P., & Costoiu, M. (2007). Glass
Dereli, T., & Filiz, H. (1999). Optimisation of fibres reinforced polymeric composites- Statistic
process planning functions by genetic algo- models of surface roughness. Materiale Plastice,
rithms. Computers & Industrial Engineering, 36, 44(4), 365–369.
281–308. doi:10.1016/S0360-8352(99)00133-3
Johnson, R. A., & Wichern, D. W. (2002). Applied
Dong, Z., Li, H., & Vickers, G. W. (1993). Op- multivariate statistical analysis. Upper Saddle
timal rough machining of sculptured parts on a River, NJ: Prentice Hall.
CNC milling machine. Journal of Engineering for
Industry, 115, 424–431. doi:10.1115/1.2901785
140
Juan, H., Yu, S. F., & Lee, B. Y. (2003). The optimal Onwubolu, G. C. (2006). Performance-based
cutting-parameter selection of production cost in optimization of multi-pass face milling operations
HSM for SKD61 tool steels. International Journal using Tribes. International Journal of Machine
of Machine Tools & Manufacture, 43, 679–686. Tools & Manufacture, 46, 717–727. doi:10.1016/j.
doi:10.1016/S0890-6955(03)00038-5 ijmachtools.2005.07.041
Krimpenis, A., Fousekis, A., & Vosniakos, G. Ozcelik, B., Oktem, H., & Kurtaran, H. (2005).
(2005). Assessment of sculptured surface mill- Optimum surface roughness in end milling Inco-
ing strategies using design of experiments. In- nel 718 by coupling neural network model and
ternational Journal of Advanced Manufacturing genetic algorithm. International Journal of Ad-
Technology, 25, 444–453. doi:10.1007/s00170- vanced Manufacturing Technology, 27, 234–241.
003-1881-x doi:10.1007/s00170-004-2175-7
Kurdi, M. H., Schmitz, T. L., Haftka, R. T., & Petropoulos, G. P., Vaxevanidis, N. M., Pandaz-
Mann, B. P. (2004). Simultaneous optimization aras, C. N., & Antoniadis, A. A. (2006). Multi-
of removal rate and part accuracy in high-speed parameter identification and control of turned
milling. Proceedings of IMECE2004, ASME In- surface textures. International Journal of Ad-
ternational Mechanical Engineering Congress, vanced Manufacturing Technology, 29, 118–128.
November 13-19, Anaheim, CA USA. doi:10.1007/s00170-004-2483-y
Lin, J. C., & Tai, C. C. (1999). Accuracy optimi- Phadke, M. S. (1989). Quality engineering using
zation for mould surface profile milling. Interna- robust design. Englewood, NJ: Prentice Hall.
tional Journal of Advanced Manufacturing Tech-
Premrov, M., Dobrila, P., Bedenik, B., & Ŝpacapan,
nology, 15, 15–25. doi:10.1007/s001700050034
I. (2007). Slip modeling in timber-framed walls
Lin, T. R. (2002). Optimization technique for face with wood-based or fibre-plaster sheathing board.
milling stainless steel with multiple performance WSEAS Transactions on Applied and Theoretical
characteristics. International Journal of Ad- Mechanics, 2(9), 167–175.
vanced Manufacturing Technology, 19, 330–335.
Quinsat, Y., & Sabourin, L. (2006). Optimal
doi:10.1007/s001700200021
selection of machining direction for three-axis
Liu, Y., & Wang, C. (1999). A modified genetic al- milling of sculptured parts. International Jour-
gorithm based optimisation of milling parameters. nal of Advanced Manufacturing Technology, 27,
International Journal of Advanced Manufactur- 1132–1139. doi:10.1007/s00170-004-2297-y
ing Technology, 15(11), 796–799. doi:10.1007/
Rao, R. V. (2011). Advanced modeling and op-
s001700050134
timization of manufacturing processes. London,
Montgomery, D. C. (2001). Design and analysis UK: Springer.
of experiments (5th ed.). New York, NY: J Wiley
Raymond, H., Montgomery, D. C., & Anderson-
and Sons.
Cook, C. M. (2009). Response surface method-
Moshat, S., Datta, S., Bandyopadhyay, A., & ology: Process and product optimization using
Kumar-Pal, P. (2010). Optimization of CNC end designed experiment (3rd ed.). New York, NY:
milling process parameters using PCA-based Wiley-Interscience.
Taguchi method. International Journal of Engi-
neering. Science and Technology, 2(1), 92–102.
141
Reddy, N. S. K., & Rao, P. V. (2006). Selection Tansel, I. N., Ozcelik, B., Bao, W. Y., Chen, P.,
of an optimal parametric combination for achiev- Rincon, D., Yang, S. Y., & Yenilmez, A. (2006).
ing a better surface finish in dry milling using Selection of optimal cutting conditions by us-
genetic algorithms. International Journal of Ad- ing GONNS. International Journal of Machine
vanced Manufacturing Technology, 28, 463–473. Tools & Manufacture, 46, 26–35. doi:10.1016/j.
doi:10.1007/s00170-004-2381-3 ijmachtools.2005.04.012
Roy, R. K. (2001). Design of experiments using Tolouei-Rad, M., & Bidhend, I. M. (1997). On the
the Taguchi approach: 16 steps to product and optimization of machining parameters for mill-
process improvement. New York, NY: John Wiley ing operations. International Journal of Machine
& Sons. Tools & Manufacture, 37(1), 1–16. doi:10.1016/
S0890-6955(96)00044-2
Schulze, V., Becke, C., & Pabst, R. (2011). Specific
machining forces and resultant force vectors for Unal, R., & Dean, E. B. (1991). Taguchi approach
machining of reinforced plastics. CIRP Annals - to design optimization for quality and cost - An
Manufacturing Technology, 60(1), 69-72. overview. Proceedings of the International Society
of Parametric Analysts 13th Annual Conference,
Shunmugam, M. S., Reddy, S. V. B., & Narendran,
New Orleans, LA, USA.
T. T. (2000). Selection of optimal conditions in
multi-pass face-milling using a genetic algo- Vaxevanidis, N. M., Fountas, N. A., Galanis, N. I.,
rithm. International Journal of Machine Tools & Bounas, I., & Sideris, J. (2011). Multi-parametric
Manufacture, 40, 401–414. doi:10.1016/S0890- analysis of surface roughness in high-speed turn-
6955(99)00063-2 ing of a high-alloyed tool steel. Proceedings of
the 7th BalkanTrib International Conference, 5-7
Sönmez, A. İ., Baykasoğlu, A., Dereli, T., & Filiz,
October 2011, Thessaloniki, Greece. (in press)
İ. H. (1999). Dynamic optimization of multipass
milling operations via geometric programming. Wang, J. (1998). Computer-aided economic
International Journal of Machine Tools & optimization of the end-milling operations. Inter-
Manufacture, 39, 297–320. doi:10.1016/S0890- national Journal of Production Economics, 54,
6955(98)00027-3 307–320. doi:10.1016/S0925-5273(98)00008-5
Sotirov, G. R., Vitanov, V. I., Ghosh, S. K., Mincoff, Wang, Z. G., Rahman, M., Wong, Y. S., & Sun, J.
N. C., & Harrison, D. K. (1992). Multi-criteria (2005). Optimization of multi-pass milling using
optimization of cutting processes. Journal of parallel genetic algorithm and parallel genetic
Materials Processing Technology, 31, 307–315. simulated annealing. International Journal of
doi:10.1016/0924-0136(92)90030-V Machine Tools & Manufacture, 45, 1726–1734.
doi:10.1016/j.ijmachtools.2005.03.009
Taguchi, G. (1995). Quality engineering (Taguchi
Methods) for the development of electronic circuit Wang, Z. G., Wong, Y. S., & Rahman, M. (2004).
technology. IEEE Transactions on Reliability, Optimisation of multi-pass milling using genetic
44(2), 225–229. doi:10.1109/24.387375 algorithm and genetic simulated annealing. In-
ternational Journal of Advanced Manufacturing
Tandon, V., El-Mounayri, H., & Kishawy, H.
Technology, 24, 727–732. doi:10.1007/s00170-
(2002). NC end milling optimization using evolu-
003-1789-5
tionary computation. International Journal of Ad-
vanced Manufacturing Technology, 42, 595–605.
142
Wu, J., & Yao, Y. (2008). A modified ant colony CAM Systems: Systems used for the deter-
system for the selection of machining parameters. mination, the simulation and the verification of
Proceedings of the 7th International Conference machining processes and manufacturing plans.
on Grid and Cooperative Computing -GCC ‘08, CNC Machining: Machining of simple or
Shenzhen, PRC, (pp. 89-93). complex parts with the use of 3- or 5-axis CNC
turning, milling, etc machines.
Yang, D. C. H., & Han, Z. (1999). Interference
DoE: Taguchi’s Design of Experiments is a
detection and optimal tool selection in 3-axis
methodology of systematic experiment planning
NC machining of free-form surfaces. Computer
and analysis.
Aided Design, 31, 303–315. doi:10.1016/S0010-
PCA: PCA is a statistical methodology that
4485(99)00031-7
allows the identification of patterns in data, and
Yang, J. L., & Chen, J. C. (2001). A systematic ap- expressing these data in a way that similarities
proach for identifying optimum surface roughness and differences can be easily observed.
performance in end-milling operations. Journal Regression Analysis: A statistical method
of Information Technology, 17(2), 1–8. that reveals and explains the relation between
a response or dependent variable Y and one (or
more) predictor/input/independent variables X1,
X2,....,Xn.
KEY TERMS AND DEFINITIONS
RSM: RSM is a combination of statistical and
mathematical techniques useful for developing,
ANOVA: ANalysis Of VAriance is a statisti-
improving, and optimizing processes.
cal methodology that offers the ability to divide
Sculptured Surface Machining (SSM):
variability of a major component to its individual
Machining of free-form surfaces of parts, with
subcomponents, so as to reveal the specificity
the use of 3- or 5-axis CNC machining centers.
among the different origins.
143

C.2 - Computational Techniques in Stat. Anal. & Exploitation of CNC Mach. Exp. Data - IGI Global, 2012

Uploaded by

Copyright:

Available Formats

C.2 - Computational Techniques in Stat. Anal. & Exploitation of CNC Mach. Exp. Data - IGI Global, 2012

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

C.2 - Computational Techniques in Stat. Anal. & Exploitation of CNC Mach. Exp. Data - IGI Global, 2012

Uploaded by

Copyright:

Available Formats

Computational

Methods for Optimizing

Published in the United States of America by

Library of Congress Cataloging-in-Publication Data

British Cataloguing in Publication Data

1. INTRODUCTION • The data category should be specifically

Each experimental design on computer systems

2.2.2 Orthogonal Arrays (OAs) The desired quality characteristics in a machining

Table 1. Standard orthogonal arrays used in DoE

Orthogonal Number of Maximum Maximum number of Columns at these levels

Table 2. Literature survey on optimized selection of independent machining variables

Independed Machining Variables

continued on following page

Table 3. Literature survey of machining quality characteristics

Table 4. Literature survey of technological constraints

Note here that c2 >0; hence, F>0. The Fn ,n

distribution family is asymmetric. The density

f (x ) = 3.1.2 One-Way ANOVA

Table 5. Factor levels and values of machining parameters

Raster StrategyRaster Factor Levels

easily observed (Johnson and Wichern, 2002). where

Remaining Volume and Machining Time are given X1 = {X1(1), X1(2)..............X1(k)......X1(n)}

where, In the above linear expressions, Xi, represents

Larger-the-better Case (LB):

Xm. This sequence of linear expressions containing

Nominal-the-best Case (NB): quality characteristic. After data normalization,

Table 6. Experimental results for helical machining strategy

RUN MACHINING TIME REMAINING VOLUME

Table 7. Experimental results for back & forth machining strategy

RUN MACHINING TIME REMAINING VOLUME

Cov(C j ,C k ) If H0: pjk = 0, there is no correlation,

1,2,........,n, j≠k. (16) After the clarification of correlation existence

• Calculation of the Principal Component 3.3 Regression Analysis

Conclusively, the optimal setting is then veri- where

Table 8. The L9 Taguchi OA with the factors, the

3.4 Response Surface Variables ξ1, ξ2,.....ξk in Equation 26 are “natu-

E(y) = η =E [f (ξ1, ξ2,.....ξk)] + E(a) = f (ξ1, 1 2 3

Table 10. The regression analysis results

Regression Analysis Results-Orthogonal Program

η = a0 +a1x1+ a2x2+a12x2 (29) Most likely, this model would be useful as an

(a) Estimated Regression Coefficients for Ra (μm)

Residual Error 17 1.49018 1.49018 0.087658

a/a Specimen S (rpm) v (m/min) F (mm/r) a(mm) Ra (μm) Rmax (μm) FD

the obtained results for the range of roughness

4. CONCLUSION AND REFERENCES

You might also like